regenerated example disk , version 5.3

now allows taking the address of a subroutine &routine

.idea/misc.xml

@@ -16,7 +16,7 @@
       </list>
     </option>
   </component>
-  <component name="ProjectRootManager" version="2" languageLevel="JDK_1_8" default="false" project-jdk-name="Kotlin SDK" project-jdk-type="KotlinSDK">
+  <component name="ProjectRootManager" version="2" languageLevel="JDK_11" default="false" project-jdk-name="Kotlin SDK" project-jdk-type="KotlinSDK">
     <output url="file://$PROJECT_DIR$/out" />
   </component>
 </project>

README.md

@@ -12,48 +12,56 @@ Prog8 - Structured Programming Language for 8-bit 6502/65c02 microprocessors
 
 This is a structured programming language for the 8-bit 6502/6510/65c02 microprocessor from the late 1970's and 1980's
 as used in many home computers from that era. It is a medium to low level programming language,
-which aims to provide many conveniences over raw assembly code (even when using a macro assembler):
+which aims to provide many conveniences over raw assembly code (even when using a macro assembler).
 
-- reduction of source code length
+Documentation
+-------------
+Full documentation (syntax reference, how to use the language and the compiler, etc.) can be found at:
+https://prog8.readthedocs.io/
+
+
+What does Prog8 provide?
+------------------------
+
+- big reduction of source code length over raw assembly
 - modularity, symbol scoping, subroutines
 - various data types other than just bytes (16-bit words, floats, strings)
 - automatic variable allocations, automatic string and array variables and string sharing
 - subroutines with an input- and output parameter signature
-- constant folding in expressions
+- no stack frame allocations because parameters and local variables are automatically allocated statically
+- constant folding in expressions and other high-level program optimizations
 - conditional branches
-- 'when' statement to provide a concise jump table alternative to if/elseif chains
-- structs to group together sets of variables and manipulate them at once
 - floating point operations  (requires the C64 Basic ROM routines for this)
-- abstracting away low level aspects such as ZeroPage handling, program startup, explicit memory addresses
-- various code optimizations (code structure, logical and numerical expressions, unused code removal...)
-- inline assembly allows you to have full control when every cycle or byte matters
+- 'when' statement to provide a concise jump table alternative to if/elseif chains
 - many built-in functions such as ``sin``, ``cos``, ``rnd``, ``abs``, ``min``, ``max``, ``sqrt``, ``msb``, ``rol``, ``ror``, ``swap``, ``memset``, ``memcopy``, ``sort`` and ``reverse``
+- structs to group together sets of variables and manipulate them at once
+- convenience abstractions for low level aspects such as ZeroPage handling, program startup, explicit memory addresses
+- fast execution speed due to compilation to native assembly code
+- inline assembly allows you to have full control when every cycle or byte matters
 
-Rapid edit-compile-run-debug cycle:
+*Rapid edit-compile-run-debug cycle:*
 
-- use a modern PC to do the work on
-- very quick compilation times
+- use a modern PC to do the work on, use nice editors and enjoy quick compilation times
 - can automatically run the program in the Vice emulator after succesful compilation
 - breakpoints, that let the Vice emulator drop into the monitor if execution hits them
 - source code labels automatically loaded in Vice emulator so it can show them in disassembly
 
-Prog8 is mainly targeted at the Commodore-64 machine.
-Preliminary support for the [CommanderX16](https://www.commanderx16.com) is available as a second compilation target.
-Contributions to improve these or to add support for other machines are welcome!
+*Two supported compiler targets* (contributions to improve these or to add support for other machines are welcome!):
+
+- "c64": Commodore-64  (6510 CPU = almost a 6502), the main target.
+- "cx16": [CommanderX16](https://www.commanderx16.com)  (65c02 CPU) .
+- If you only use standard kernel and prog8 library routines, it is possible to compile the *exact same program* for both machines (just change the compiler target flag)!
 
 
-Documentation/manual
---------------------
-https://prog8.readthedocs.io/
 
-Required tools
---------------
+Additional required tools
+-------------------------
 
 [64tass](https://sourceforge.net/projects/tass64/) - cross assembler. Install this on your shell path.
 A recent .exe version of this tool for Windows can be obtained from my [clone](https://github.com/irmen/64tass/releases) of this project.
 For other platforms it is very easy to compile it yourself (make ; make install).
 
-A **Java runtime (jre or jdk), version 8 or newer**  is required to run a prepackaged version of the compiler.
+A **Java runtime (jre or jdk), version 11 or newer**  is required to run a prepackaged version of the compiler.
 If you want to build it from source, you'll need a Java SDK + Kotlin 1.3.x SDK (or for instance,
 IntelliJ IDEA with the Kotlin plugin).
 
@@ -67,16 +75,18 @@ Example code
 
 This code calculates prime numbers using the Sieve of Eratosthenes algorithm::
 
-    %import c64textio
+    %import textio
     %zeropage basicsafe
-
+    
     main {
-
+    
         ubyte[256] sieve
-        ubyte candidate_prime = 2
-
+        ubyte candidate_prime = 2       ; is increased in the loop
+    
         sub start() {
-            memset(sieve, 256, false)   ; clear the sieve
+            ; clear the sieve, to reset starting situation on subsequent runs
+            memset(sieve, 256, false)   
+            ; calculate primes
             txt.print("prime numbers up to 255:\n\n")
             ubyte amount=0
             repeat {
@@ -87,22 +97,23 @@ This code calculates prime numbers using the Sieve of Eratosthenes algorithm::
                 txt.print(", ")
                 amount++
             }
-            c64.CHROUT('\n')
+            txt.chrout('\n')
             txt.print("number of primes (expected 54): ")
             txt.print_ub(amount)
-            c64.CHROUT('\n')
+            txt.chrout('\n')
         }
-
+    
         sub find_next_prime() -> ubyte {
             while sieve[candidate_prime] {
                 candidate_prime++
                 if candidate_prime==0
-                    return 0        ; we wrapped; no more primes available
+                    return 0        ; we wrapped; no more primes available in the sieve
             }
+    
             ; found next one, mark the multiples and return it.
             sieve[candidate_prime] = true
             uword multiple = candidate_prime
-
+    
             while multiple < len(sieve) {
                 sieve[lsb(multiple)] = true
                 multiple += candidate_prime
@@ -112,11 +123,11 @@ This code calculates prime numbers using the Sieve of Eratosthenes algorithm::
     }
 
 
+
 when compiled an ran on a C-64 you'll get:
 
 ![c64 screen](docs/source/_static/primes_example.png)
 
-
 One of the included examples (wizzine.p8) animates a bunch of sprite balloons and looks like this:
 
 ![wizzine screen](docs/source/_static/wizzine.png)
@@ -128,3 +139,8 @@ Another example (cube3d-sprites.p8) draws the vertices of a rotating 3d cube:
 If you want to play a video game, a fully working Tetris clone is included in the examples:
 
 ![tehtriz_screen](docs/source/_static/tehtriz.png)
+
+The CommanderX16 compiler target is quite capable already too, here's a well known space ship 
+animated in 3D with hidden line removal, in the CommanderX16 emulator:
+
+![cobra3d](docs/source/_static/cobra3d.png)

compiler/build.gradle

@@ -1,11 +1,11 @@
 buildscript {
     dependencies {
-        classpath "org.jetbrains.kotlin:kotlin-gradle-plugin:1.4.0"
+        classpath "org.jetbrains.kotlin:kotlin-gradle-plugin:1.4.20"
     }
 }
 
 plugins {
-    // id "org.jetbrains.kotlin.jvm" version "1.4.0"
+    // id "org.jetbrains.kotlin.jvm" version "1.4.20"
     id 'application'
     id 'org.jetbrains.dokka' version "0.9.18"
     id 'com.github.johnrengelman.shadow' version '5.2.0'
@@ -15,8 +15,8 @@ plugins {
 apply plugin: "kotlin"
 apply plugin: "java"
 
-targetCompatibility = 1.8
-sourceCompatibility = 1.8
+targetCompatibility = 11
+sourceCompatibility = 11
 
 repositories {
     mavenLocal()
@@ -45,7 +45,7 @@ dependencies {
 
 compileKotlin {
     kotlinOptions {
-        jvmTarget = "1.8"
+        jvmTarget = "11"
         // verbose = true
         // freeCompilerArgs += "-XXLanguage:+NewInference"
     }
@@ -53,7 +53,7 @@ compileKotlin {
 
 compileTestKotlin {
     kotlinOptions {
-        jvmTarget = "1.8"
+        jvmTarget = "11"
     }
 }
 

compiler/compiler.iml

@@ -8,7 +8,7 @@
       <sourceFolder url="file://$MODULE_DIR$/test" isTestSource="true" />
       <excludeFolder url="file://$MODULE_DIR$/build" />
     </content>
-    <orderEntry type="jdk" jdkName="openjdk-11" jdkType="JavaSDK" />
+    <orderEntry type="jdk" jdkName="11" jdkType="JavaSDK" />
     <orderEntry type="sourceFolder" forTests="false" />
     <orderEntry type="library" name="KotlinJavaRuntime" level="project" />
     <orderEntry type="module" module-name="parser" />

compiler/res/prog8lib/c64/floats.asm

@@ -1,25 +1,31 @@
 ; --- low level floating point assembly routines for the C64
 
+FL_ONE_const	.byte  129     			; 1.0
+FL_ZERO_const	.byte  0,0,0,0,0		; 0.0
+
+floats_store_reg	.byte  0		; temp storage
+
 
 ub2float	.proc
 		; -- convert ubyte in SCRATCH_ZPB1 to float at address A/Y
 		;    clobbers A, Y
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		sta  P8ZP_SCRATCH_W2
 		sty  P8ZP_SCRATCH_W2+1
 		ldy  P8ZP_SCRATCH_B1
-		jsr  FREADUY
+		lda  #0
+		jsr  GIVAYF
 _fac_to_mem	ldx  P8ZP_SCRATCH_W2
 		ldy  P8ZP_SCRATCH_W2+1
 		jsr  MOVMF
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 		.pend
 
 b2float		.proc
 		; -- convert byte in SCRATCH_ZPB1 to float at address A/Y
 		;    clobbers A, Y
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		sta  P8ZP_SCRATCH_W2
 		sty  P8ZP_SCRATCH_W2+1
 		lda  P8ZP_SCRATCH_B1
@@ -29,7 +35,7 @@ b2float		.proc
 
 uw2float	.proc
 		; -- convert uword in SCRATCH_ZPWORD1 to float at address A/Y
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		sta  P8ZP_SCRATCH_W2
 		sty  P8ZP_SCRATCH_W2+1
 		lda  P8ZP_SCRATCH_W1
@@ -40,7 +46,7 @@ uw2float	.proc
 
 w2float		.proc
 		; -- convert word in SCRATCH_ZPWORD1 to float at address A/Y
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		sta  P8ZP_SCRATCH_W2
 		sty  P8ZP_SCRATCH_W2+1
 		ldy  P8ZP_SCRATCH_W1
@@ -49,13 +55,76 @@ w2float		.proc
 		jmp  ub2float._fac_to_mem
 		.pend
 
+
+cast_from_uw	.proc
+		; -- uword in A/Y into float var at (P8ZP_SCRATCH_W2)
+		stx  P8ZP_SCRATCH_REG
+		jsr  GIVUAYFAY
+		jmp  ub2float._fac_to_mem
+		.pend
+
+
+cast_from_w	.proc
+		; -- word in A/Y into float var at (P8ZP_SCRATCH_W2)
+		stx  P8ZP_SCRATCH_REG
+		jsr  GIVAYFAY
+		jmp  ub2float._fac_to_mem
+		.pend
+
+
+cast_from_ub	.proc
+		; -- ubyte in Y into float var at (P8ZP_SCRATCH_W2)
+		stx  P8ZP_SCRATCH_REG
+		jsr  FREADUY
+		jmp  ub2float._fac_to_mem
+		.pend
+
+
+cast_from_b	.proc
+		; -- byte in A into float var at (P8ZP_SCRATCH_W2)
+		stx  P8ZP_SCRATCH_REG
+		jsr  FREADSA
+		jmp  ub2float._fac_to_mem
+		.pend
+
+cast_as_uw_into_ya	.proc               ; also used for float 2 ub
+		; -- cast float at A/Y to uword into Y/A
+		jsr  MOVFM
+		jmp  cast_FAC1_as_uw_into_ya
+		.pend
+
+cast_as_w_into_ay	.proc               ; also used for float 2 b
+		; -- cast float at A/Y to word into A/Y
+		jsr  MOVFM
+		jmp  cast_FAC1_as_w_into_ay
+		.pend
+
+cast_FAC1_as_uw_into_ya	.proc               ; also used for float 2 ub
+		; -- cast fac1 to uword into Y/A
+		stx  P8ZP_SCRATCH_REG
+		jsr  GETADR     ; into Y/A
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+cast_FAC1_as_w_into_ay	.proc               ; also used for float 2 b
+		; -- cast fac1 to word into A/Y
+		stx  P8ZP_SCRATCH_REG
+		jsr  AYINT
+		ldy  $64
+		lda  $65
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+
 stack_b2float	.proc
 		; -- b2float operating on the stack
 		inx
 		lda  P8ESTACK_LO,x
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		jsr  FREADSA
-		jmp  push_fac1_as_result
+		jmp  push_fac1._internal
 		.pend
 
 stack_w2float	.proc
@@ -63,19 +132,20 @@ stack_w2float	.proc
 		inx
 		ldy  P8ESTACK_LO,x
 		lda  P8ESTACK_HI,x
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		jsr  GIVAYF
-		jmp  push_fac1_as_result
+		jmp  push_fac1._internal
 		.pend
 
 stack_ub2float	.proc
 		; -- ub2float operating on the stack
 		inx
 		lda  P8ESTACK_LO,x
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		tay
-		jsr  FREADUY
-		jmp  push_fac1_as_result
+		lda  #0
+		jsr  GIVAYF
+		jmp  push_fac1._internal
 		.pend
 
 stack_uw2float	.proc
@@ -83,16 +153,16 @@ stack_uw2float	.proc
 		inx
 		lda  P8ESTACK_LO,x
 		ldy  P8ESTACK_HI,x
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		jsr  GIVUAYFAY
-		jmp  push_fac1_as_result
+		jmp  push_fac1._internal
 		.pend
 
 stack_float2w	.proc               ; also used for float2b
 		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		jsr  AYINT
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		lda  $64
 		sta  P8ESTACK_HI,x
 		lda  $65
@@ -103,9 +173,9 @@ stack_float2w	.proc               ; also used for float2b
 
 stack_float2uw	.proc               ; also used for float2ub
 		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		jsr  GETADR
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		sta  P8ESTACK_HI,x
 		tya
 		sta  P8ESTACK_LO,x
@@ -139,22 +209,6 @@ push_float	.proc
 		rts
 		.pend
 
-func_rndf	.proc
-		; -- put a random floating point value on the stack
-		stx  P8ZP_SCRATCH_REG
-		lda  #1
-		jsr  FREADSA
-		jsr  RND		; rng into fac1
-		ldx  #<_rndf_rnum5
-		ldy  #>_rndf_rnum5
-		jsr  MOVMF	; fac1 to mem X/Y
-		ldx  P8ZP_SCRATCH_REG
-		lda  #<_rndf_rnum5
-		ldy  #>_rndf_rnum5
-		jmp  push_float
-_rndf_rnum5	.byte  0,0,0,0,0
-		.pend
-
 pop_float	.proc
 		; ---- pops mflpt5 from stack to memory A/Y
 		; (frees 3 stack positions = 6 bytes of which 1 is padding)
@@ -191,27 +245,6 @@ pop_float_fac1	.proc
 		jmp  MOVFM
 		.pend
 
-pop_float_fac2	.proc
-		; -- pops float from stack into FAC2
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  pop_float
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jmp  CONUPK
-		.pend
-
-pop_float_to_indexed_var	.proc
-		; -- pop the float on the stack, to the memory in the array at A/Y indexed by the byte on stack
-		sta  P8ZP_SCRATCH_W1
-		sty  P8ZP_SCRATCH_W1+1
-		jsr  prog8_lib.pop_index_times_5
-		jsr  prog8_lib.add_a_to_zpword
-		lda  P8ZP_SCRATCH_W1
-		ldy  P8ZP_SCRATCH_W1+1
-		jmp  pop_float
-		.pend
-
 copy_float	.proc
 		; -- copies the 5 bytes of the mflt value pointed to by SCRATCH_ZPWORD1,
 		;    into the 5 bytes pointed to by A/Y.  Clobbers A,Y.
@@ -229,15 +262,15 @@ inc_var_f	.proc
 		; -- add 1 to float pointed to by A/Y
 		sta  P8ZP_SCRATCH_W1
 		sty  P8ZP_SCRATCH_W1+1
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		jsr  MOVFM
-		lda  #<FL_FONE
-		ldy  #>FL_FONE
+		lda  #<FL_ONE_const
+		ldy  #>FL_ONE_const
 		jsr  FADD
 		ldx  P8ZP_SCRATCH_W1
 		ldy  P8ZP_SCRATCH_W1+1
 		jsr  MOVMF
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 		.pend
 
@@ -245,9 +278,9 @@ dec_var_f	.proc
 		; -- subtract 1 from float pointed to by A/Y
 		sta  P8ZP_SCRATCH_W1
 		sty  P8ZP_SCRATCH_W1+1
-		stx  P8ZP_SCRATCH_REG_X
-		lda  #<FL_FONE
-		ldy  #>FL_FONE
+		stx  P8ZP_SCRATCH_REG
+		lda  #<FL_ONE_const
+		ldy  #>FL_ONE_const
 		jsr  MOVFM
 		lda  P8ZP_SCRATCH_W1
 		ldy  P8ZP_SCRATCH_W1+1
@@ -255,7 +288,7 @@ dec_var_f	.proc
 		ldx  P8ZP_SCRATCH_W1
 		ldy  P8ZP_SCRATCH_W1+1
 		jsr  MOVMF
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 		.pend
 
@@ -277,17 +310,20 @@ pop_2_floats_f2_in_fac1	.proc
 fmath_float1	.byte 0,0,0,0,0	; storage for a mflpt5 value
 fmath_float2	.byte 0,0,0,0,0	; storage for a mflpt5 value
 
-push_fac1_as_result	.proc
-		; -- push the float in FAC1 onto the stack, and return from calculation
-		ldx  #<fmath_float1
+
+push_fac1	.proc
+		; -- push the float in FAC1 onto the stack
+		stx  P8ZP_SCRATCH_REG
+_internal	ldx  #<fmath_float1
 		ldy  #>fmath_float1
 		jsr  MOVMF
 		lda  #<fmath_float1
 		ldy  #>fmath_float1
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		jmp  push_float
 		.pend
 
+
 pow_f		.proc
 		; -- push f1 ** f2 on stack
 		lda  #<fmath_float2
@@ -296,71 +332,155 @@ pow_f		.proc
 		lda  #<fmath_float1
 		ldy  #>fmath_float1
 		jsr  pop_float
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		lda  #<fmath_float1
 		ldy  #>fmath_float1
 		jsr  CONUPK		; fac2 = float1
 		lda  #<fmath_float2
 		ldy  #>fmath_float2
 		jsr  FPWR
-		ldx  P8ZP_SCRATCH_REG_X
-		jmp  push_fac1_as_result
+		jmp  push_fac1._internal
 		.pend
 
 div_f		.proc
 		; -- push f1/f2 on stack
 		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		lda  #<fmath_float1
 		ldy  #>fmath_float1
 		jsr  FDIV
-		jmp  push_fac1_as_result
+		jmp  push_fac1._internal
 		.pend
 
 add_f		.proc
 		; -- push f1+f2 on stack
 		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		lda  #<fmath_float1
 		ldy  #>fmath_float1
 		jsr  FADD
-		jmp  push_fac1_as_result
+		jmp  push_fac1._internal
 		.pend
 
 sub_f		.proc
 		; -- push f1-f2 on stack
 		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		lda  #<fmath_float1
 		ldy  #>fmath_float1
 		jsr  FSUB
-		jmp  push_fac1_as_result
+		jmp  push_fac1._internal
 		.pend
 
 mul_f		.proc
 		; -- push f1*f2 on stack
 		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		lda  #<fmath_float1
 		ldy  #>fmath_float1
 		jsr  FMULT
-		jmp  push_fac1_as_result
+		jmp  push_fac1._internal
 		.pend
 
 neg_f		.proc
-		; -- push -flt back on stack
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  NEGOP
-		jmp  push_fac1_as_result
+		; -- toggle the sign bit on the stack
+		lda  P8ESTACK_HI+3,x
+		eor  #$80
+		sta  P8ESTACK_HI+3,x
+		rts
 		.pend
 
-abs_f		.proc
-		; -- push abs(float) on stack (as float)
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  ABS
-		jmp  push_fac1_as_result
+var_fac1_less_f	.proc
+		; -- is the float in FAC1 < the variable AY?
+		stx  P8ZP_SCRATCH_REG
+		jsr  FCOMP
+		ldx  P8ZP_SCRATCH_REG
+		cmp  #255
+		beq  +
+		lda  #0
+		rts
++		lda  #1
+		rts
+		.pend
+
+var_fac1_lesseq_f	.proc
+		; -- is the float in FAC1 <= the variable AY?
+		stx  P8ZP_SCRATCH_REG
+		jsr  FCOMP
+		ldx  P8ZP_SCRATCH_REG
+		cmp  #0
+		beq  +
+		cmp  #255
+		beq  +
+		lda  #0
+		rts
++		lda  #1
+		rts
+		.pend
+
+var_fac1_greater_f	.proc
+		; -- is the float in FAC1 > the variable AY?
+		stx  P8ZP_SCRATCH_REG
+		jsr  FCOMP
+		ldx  P8ZP_SCRATCH_REG
+		cmp  #1
+		beq  +
+		lda  #0
++		rts
+		.pend
+
+var_fac1_greatereq_f	.proc
+		; -- is the float in FAC1 >= the variable AY?
+		stx  P8ZP_SCRATCH_REG
+		jsr  FCOMP
+		ldx  P8ZP_SCRATCH_REG
+		cmp  #0
+		beq  +
+		cmp  #1
+		beq  +
+		lda  #0
+		rts
++		lda  #1
+		rts
+		.pend
+
+var_fac1_notequal_f	.proc
+		; -- are the floats numbers in FAC1 and the variable AY *not* identical?
+		stx  P8ZP_SCRATCH_REG
+		jsr  FCOMP
+		ldx  P8ZP_SCRATCH_REG
+		and  #1
+		rts
+		.pend
+
+vars_equal_f	.proc
+		; -- are the mflpt5 numbers in P8ZP_SCRATCH_W1 and AY identical?
+		sta  P8ZP_SCRATCH_W2
+		sty  P8ZP_SCRATCH_W2+1
+		ldy  #0
+		lda  (P8ZP_SCRATCH_W1),y
+		cmp  (P8ZP_SCRATCH_W2),y
+		bne  _false
+		iny
+		lda  (P8ZP_SCRATCH_W1),y
+		cmp  (P8ZP_SCRATCH_W2),y
+		bne  _false
+		iny
+		lda  (P8ZP_SCRATCH_W1),y
+		cmp  (P8ZP_SCRATCH_W2),y
+		bne  _false
+		iny
+		lda  (P8ZP_SCRATCH_W1),y
+		cmp  (P8ZP_SCRATCH_W2),y
+		bne  _false
+		iny
+		lda  (P8ZP_SCRATCH_W1),y
+		cmp  (P8ZP_SCRATCH_W2),y
+		bne  _false
+		lda  #1
+		rts
+_false		lda  #0
+		rts
 		.pend
 
 equal_f		.proc
@@ -400,6 +520,40 @@ notequal_f	.proc
 		rts
 		.pend
 
+vars_less_f	.proc
+		; -- is float in AY < float in P8ZP_SCRATCH_W2 ?
+		jsr  MOVFM
+		lda  P8ZP_SCRATCH_W2
+		ldy  P8ZP_SCRATCH_W2+1
+		stx  P8ZP_SCRATCH_REG
+		jsr  FCOMP
+		ldx  P8ZP_SCRATCH_REG
+		cmp  #255
+		bne  +
+		lda  #1
+		rts
++		lda  #0
+		rts
+		.pend
+
+vars_lesseq_f	.proc
+		; -- is float in AY <= float in P8ZP_SCRATCH_W2 ?
+		jsr  MOVFM
+		lda  P8ZP_SCRATCH_W2
+		ldy  P8ZP_SCRATCH_W2+1
+		stx  P8ZP_SCRATCH_REG
+		jsr  FCOMP
+		ldx  P8ZP_SCRATCH_REG
+		cmp  #255
+		bne  +
+-		lda  #1
+		rts
++		cmp  #0
+		beq  -
+		lda  #0
+		rts
+		.pend
+
 less_f		.proc
 		; -- is f1 < f2?
 		jsr  compare_floats
@@ -461,248 +615,33 @@ _return_true	lda  #1
 		bne  _return_result
 		.pend
 
-func_sin	.proc
-		; -- push sin(f) back onto stack
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  SIN
-		jmp  push_fac1_as_result
-		.pend
-
-func_cos	.proc
-		; -- push cos(f) back onto stack
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  COS
-		jmp  push_fac1_as_result
-		.pend
-
-func_tan	.proc
-		; -- push tan(f) back onto stack
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  TAN
-		jmp  push_fac1_as_result
-		.pend
-
-func_atan	.proc
-		; -- push atan(f) back onto stack
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  ATN
-		jmp  push_fac1_as_result
-		.pend
-
-func_ln		.proc
-		; -- push ln(f) back onto stack
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  LOG
-		jmp  push_fac1_as_result
-		.pend
-
-func_log2	.proc
-		; -- push log base 2, ln(f)/ln(2), back onto stack
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  LOG
-		jsr  MOVEF
-		lda  #<c64.FL_LOG2
-		ldy  #>c64.FL_LOG2
-		jsr  MOVFM
-		jsr  FDIVT
-		jmp  push_fac1_as_result
-		.pend
-
-func_sqrt	.proc
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  SQR
-		jmp  push_fac1_as_result
-		.pend
-
-func_rad	.proc
-		; -- convert degrees to radians (d * pi / 180)
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		lda  #<_pi_div_180
-		ldy  #>_pi_div_180
-		jsr  FMULT
-		jmp  push_fac1_as_result
-_pi_div_180	.byte 123, 14, 250, 53, 18		; pi / 180
-		.pend
-
-func_deg	.proc
-		; -- convert radians to degrees (d * (1/ pi * 180))
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		lda  #<_one_over_pi_div_180
-		ldy  #>_one_over_pi_div_180
-		jsr  FMULT
-		jmp  push_fac1_as_result
-_one_over_pi_div_180	.byte 134, 101, 46, 224, 211		; 1 / (pi * 180)
-		.pend
-
-func_round	.proc
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  FADDH
-		jsr  INT
-		jmp  push_fac1_as_result
-		.pend
-
-func_floor	.proc
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  INT
-		jmp  push_fac1_as_result
-		.pend
-
-func_ceil	.proc
-		; -- ceil: tr = int(f); if tr==f -> return  else return tr+1
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG_X
-		ldx  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  MOVMF
-		jsr  INT
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  FCOMP
-		cmp  #0
-		beq  +
-		lda  #<FL_FONE
-		ldy  #>FL_FONE
-		jsr  FADD
-+		jmp  push_fac1_as_result
-		.pend
-
-func_any_f	.proc
-		inx
-		lda  P8ESTACK_LO,x	; array size
+set_array_float_from_fac1	.proc
+		; -- set the float in FAC1 in the array (index in A, array in P8ZP_SCRATCH_W1)
 		sta  P8ZP_SCRATCH_B1
 		asl  a
 		asl  a
 		clc
-		adc  P8ZP_SCRATCH_B1	; times 5 because of float
-		jmp  prog8_lib.func_any_b._entry
-		.pend
-
-func_all_f	.proc
-		inx
-		jsr  prog8_lib.peek_address
-		lda  P8ESTACK_LO,x	; array size
-		sta  P8ZP_SCRATCH_B1
-		asl  a
-		asl  a
-		clc
-		adc  P8ZP_SCRATCH_B1	; times 5 because of float
-		tay
-		dey
--		lda  (P8ZP_SCRATCH_W1),y
-		clc
-		dey
-		adc  (P8ZP_SCRATCH_W1),y
-		dey
-		adc  (P8ZP_SCRATCH_W1),y
-		dey
-		adc  (P8ZP_SCRATCH_W1),y
-		dey
-		adc  (P8ZP_SCRATCH_W1),y
-		dey
-		cmp  #0
-		beq  +
-		cpy  #255
-        	bne  -
-		lda  #1
-		sta  P8ESTACK_LO+1,x
-		rts
-+		sta  P8ESTACK_LO+1,x
-		rts
-		.pend
-
-func_max_f	.proc
-		lda  #255
-		sta  _minmax_cmp+1
-		lda  #<_largest_neg_float
-		ldy  #>_largest_neg_float
-_minmax_entry	jsr  MOVFM
-		jsr  prog8_lib.pop_array_and_lengthmin1Y
-		stx  P8ZP_SCRATCH_REG_X
--		sty  P8ZP_SCRATCH_REG
-		lda  P8ZP_SCRATCH_W1
+		adc  P8ZP_SCRATCH_B1
 		ldy  P8ZP_SCRATCH_W1+1
-		jsr  FCOMP
-_minmax_cmp	cmp  #255			; modified
-		bne  +
-		lda  P8ZP_SCRATCH_W1
-		ldy  P8ZP_SCRATCH_W1+1
-		jsr  MOVFM
-+		lda  P8ZP_SCRATCH_W1
 		clc
-		adc  #5
-		sta  P8ZP_SCRATCH_W1
+		adc  P8ZP_SCRATCH_W1
 		bcc  +
-		inc  P8ZP_SCRATCH_W1+1
-+		ldy  P8ZP_SCRATCH_REG
-		dey
-		cpy  #255
-		bne  -
-		jmp  push_fac1_as_result
-_largest_neg_float	.byte 255,255,255,255,255		; largest negative float -1.7014118345e+38
-		.pend
-
-func_min_f	.proc
-		lda  #1
-		sta  func_max_f._minmax_cmp+1
-		lda  #<_largest_pos_float
-		ldy  #>_largest_pos_float
-		jmp  func_max_f._minmax_entry
-_largest_pos_float	.byte  255,127,255,255,255		; largest positive float
-		rts
-		.pend
-
-func_sum_f	.proc
-		lda  #<FL_ZERO
-		ldy  #>FL_ZERO
-		jsr  MOVFM
-		jsr  prog8_lib.pop_array_and_lengthmin1Y
-		stx  P8ZP_SCRATCH_REG_X
--		sty  P8ZP_SCRATCH_REG
-		lda  P8ZP_SCRATCH_W1
-		ldy  P8ZP_SCRATCH_W1+1
-		jsr  FADD
-		ldy  P8ZP_SCRATCH_REG
-		dey
-		cpy  #255
-		beq  +
-		lda  P8ZP_SCRATCH_W1
-		clc
-		adc  #5
-		sta  P8ZP_SCRATCH_W1
-		bcc  -
-		inc  P8ZP_SCRATCH_W1+1
-		bne  -
-+		jmp  push_fac1_as_result
-		.pend
-
-sign_f		.proc
-		jsr  pop_float_fac1
-		jsr  SIGN
-		sta  P8ESTACK_LO,x
-		dex
+		iny
++		stx  floats_store_reg
+		tax
+		jsr  MOVMF
+		ldx  floats_store_reg
 		rts
 		.pend
 
 
 set_0_array_float	.proc
-		; -- set a float in an array to zero (index on stack, array in SCRATCH_ZPWORD1)
-		inx
-		lda  P8ESTACK_LO,x
+		; -- set a float in an array to zero (index in A, array in P8ZP_SCRATCH_W1)
+		sta  P8ZP_SCRATCH_B1
 		asl  a
 		asl  a
 		clc
-		adc  P8ESTACK_LO,x
+		adc  P8ZP_SCRATCH_B1
 		tay
 		lda  #0
 		sta  (P8ZP_SCRATCH_W1),y
@@ -719,13 +658,12 @@ set_0_array_float	.proc
 
 
 set_array_float		.proc
-		; -- set a float in an array to a value (index on stack, float in SCRATCH_ZPWORD1, array in SCRATCH_ZPWORD2)
-		inx
-		lda  P8ESTACK_LO,x
+		; -- set a float in an array to a value (index in A, float in P8ZP_SCRATCH_W1, array in P8ZP_SCRATCH_W2)
+		sta  P8ZP_SCRATCH_B1
 		asl  a
 		asl  a
 		clc
-		adc  P8ESTACK_LO,x
+		adc  P8ZP_SCRATCH_B1
 		adc  P8ZP_SCRATCH_W2
 		ldy  P8ZP_SCRATCH_W2+1
 		bcc  +
@@ -736,16 +674,3 @@ set_array_float		.proc
 		.pend
 
 
-swap_floats	.proc
-		; -- swap floats pointed to by SCRATCH_ZPWORD1, SCRATCH_ZPWORD2
-		ldy  #4
--               lda  (P8ZP_SCRATCH_W1),y
-		pha
-		lda  (P8ZP_SCRATCH_W2),y
-		sta  (P8ZP_SCRATCH_W1),y
-		pla
-		sta  (P8ZP_SCRATCH_W2),y
-		dey
-		bpl  -
-		rts
-		.pend

compiler/res/prog8lib/c64/floats.p8

@@ -4,14 +4,14 @@
 ;
 ; indent format: TABS, size=8
 
+%target c64
 %option enable_floats
 
-
-c64flt {
+floats {
 	; ---- this block contains C-64 floating point related functions ----
 
-        const  float  PI        = 3.141592653589793
-        const  float  TWOPI	    = 6.283185307179586
+        const float  PI     = 3.141592653589793
+        const float  TWOPI  = 6.283185307179586
 
 
 ; ---- C64 basic and kernal ROM float constants and functions ----
@@ -35,13 +35,11 @@ c64flt {
 		&float  FL_TWOPI	= $e2e5  ; 2 * PI
 		&float  FL_FR4		= $e2ea  ; .25
 		; oddly enough, 0.0 isn't available in the kernel.
-        float   FL_ZERO     = 0.0    ; oddly enough 0.0 isn't available in the kernel
 
 
 ; note: fac1/2 might get clobbered even if not mentioned in the function's name.
 ; note: for subtraction and division, the left operand is in fac2, the right operand in fac1.
 
-; checked functions below:
 romsub $bba2 = MOVFM(uword mflpt @ AY) clobbers(A,Y)        ; load mflpt value from memory  in A/Y into fac1
 romsub $bba6 = FREADMEM() clobbers(A,Y)                     ; load mflpt value from memory  in $22/$23 into fac1
 romsub $ba8c = CONUPK(uword mflpt @ AY) clobbers(A,Y)       ; load mflpt value from memory  in A/Y into fac2
@@ -52,22 +50,22 @@ romsub $bc0f = MOVEF() clobbers(A,X)                        ; copy fac1 to fac2
 romsub $bbd4 = MOVMF(uword mflpt @ XY) clobbers(A,Y)        ; store fac1 to memory  X/Y as 5-byte mflpt
 
 ; fac1-> signed word in Y/A (might throw ILLEGAL QUANTITY)
-; (tip: use c64flt.FTOSWRDAY to get A/Y output; lo/hi switched to normal little endian order)
+; (tip: use floats.FTOSWRDAY to get A/Y output; lo/hi switched to normal little endian order)
 romsub $b1aa = FTOSWORDYA() clobbers(X) -> ubyte @ Y, ubyte @ A       ; note: calls AYINT.
 
 ; fac1 -> unsigned word in Y/A (might throw ILLEGAL QUANTITY) (result also in $14/15)
-; (tip: use c64flt.GETADRAY to get A/Y output; lo/hi switched to normal little endian order)
+; (tip: use floats.GETADRAY to get A/Y output; lo/hi switched to normal little endian order)
 romsub $b7f7 = GETADR() clobbers(X) -> ubyte @ Y, ubyte @ A
 
 romsub $bc9b = QINT() clobbers(A,X,Y)           ; fac1 -> 4-byte signed integer in 98-101 ($62-$65), with the MSB FIRST.
 romsub $b1bf = AYINT() clobbers(A,X,Y)          ; fac1-> signed word in 100-101 ($64-$65) MSB FIRST. (might throw ILLEGAL QUANTITY)
 
 ; GIVAYF: signed word in Y/A (note different lsb/msb order) -> float in fac1
-; (tip: use c64flt.GIVAYFAY to use A/Y input; lo/hi switched to normal order)
-; there is also c64flt.GIVUAYFAY - unsigned word in A/Y (lo/hi) to fac1
-; there is also c64flt.FREADS32  that reads from 98-101 ($62-$65) MSB FIRST
-; there is also c64flt.FREADUS32  that reads from 98-101 ($62-$65) MSB FIRST
-; there is also c64flt.FREADS24AXY  that reads signed int24 into fac1 from A/X/Y (lo/mid/hi bytes)
+; (tip: use floats.GIVAYFAY to use A/Y input; lo/hi switched to normal order)
+; there is also floats.GIVUAYFAY - unsigned word in A/Y (lo/hi) to fac1
+; there is also floats.FREADS32  that reads from 98-101 ($62-$65) MSB FIRST
+; there is also floats.FREADUS32  that reads from 98-101 ($62-$65) MSB FIRST
+; there is also floats.FREADS24AXY  that reads signed int24 into fac1 from A/X/Y (lo/mid/hi bytes)
 romsub $b391 = GIVAYF(ubyte lo @ Y, ubyte hi @ A) clobbers(A,X,Y)
 
 romsub $b3a2 = FREADUY(ubyte value @ Y) clobbers(A,X,Y)     ; 8 bit unsigned Y -> float in fac1
@@ -91,6 +89,7 @@ romsub $bb12 = FDIVT() clobbers(A,X,Y)                      ; fac1 = fac2/fac1
 romsub $bb0f = FDIV(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = mflpt in A/Y / fac1  (remainder in fac2)
 romsub $bf7b = FPWRT() clobbers(A,X,Y)                      ; fac1 = fac2 ** fac1
 romsub $bf78 = FPWR(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = fac2 ** mflpt from A/Y
+romsub $bd7e = FINLOG(byte value @A) clobbers (A, X, Y)     ; fac1 += signed byte in A
 
 romsub $aed4 = NOTOP() clobbers(A,X,Y)                      ; fac1 = NOT(fac1)
 romsub $bccc = INT() clobbers(A,X,Y)                        ; INT() truncates, use FADDH first to round instead of trunc
@@ -193,33 +192,27 @@ asmsub  GETADRAY  () clobbers(X) -> uword @ AY  {
 }
 
 sub  print_f  (float value) {
-	; ---- prints the floating point value (without a newline) using basic rom routines.
+	; ---- prints the floating point value (without a newline).
 	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
+		stx  floats_store_reg
 		lda  #<value
 		ldy  #>value
 		jsr  MOVFM		; load float into fac1
 		jsr  FOUT		; fac1 to string in A/Y
-		jsr  c64.STROUT			; print string in A/Y
-		ldx  P8ZP_SCRATCH_REG_X
+		sta  P8ZP_SCRATCH_W1
+		sty  P8ZP_SCRATCH_W1+1
+		ldy  #0
+-		lda  (P8ZP_SCRATCH_W1),y
+		beq  +
+		jsr  c64.CHROUT
+		iny
+		bne  -
++		ldx  floats_store_reg
 		rts
 	}}
 }
 
-sub  print_fln  (float value) {
-	; ---- prints the floating point value (with a newline at the end) using basic rom routines
-	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
-		lda  #<value
-		ldy  #>value
-		jsr  MOVFM		; load float into fac1
-		jsr  FPRINTLN		; print fac1 with newline
-		ldx  P8ZP_SCRATCH_REG_X
-		rts
-	}}
+%asminclude "library:c64/floats.asm", ""
+%asminclude "library:c64/floats_funcs.asm", ""
 
 }
-
-%asminclude "library:c64floats.asm", ""
-
-}  ; ------ end of block c64flt

compiler/res/prog8lib/c64/floats_funcs.asm

@@ -0,0 +1,437 @@
+; --- floating point builtin functions
+
+
+abs_f_stack	.proc
+		; -- push abs(AY) on stack
+		jsr  floats.MOVFM
+		jsr  floats.ABS
+		jmp  push_fac1
+		.pend
+
+abs_f_fac1	.proc
+		; -- FAC1 = abs(AY)
+		jsr  floats.MOVFM
+		jmp  floats.ABS
+		.pend
+
+func_atan_stack	.proc
+		jsr  func_atan_fac1
+		jmp  push_fac1
+		.pend
+
+func_atan_fac1	.proc
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		jsr  ATN
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+func_ceil_stack	.proc
+		jsr  func_ceil_fac1
+		jmp  push_fac1
+		.pend
+
+func_ceil_fac1	.proc
+		; -- ceil: tr = int(f); if tr==f -> return  else return tr+1
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		ldx  #<fmath_float1
+		ldy  #>fmath_float1
+		jsr  MOVMF
+		jsr  INT
+		lda  #<fmath_float1
+		ldy  #>fmath_float1
+		jsr  FCOMP
+		cmp  #0
+		beq  +
+		lda  #<FL_ONE_const
+		ldy  #>FL_ONE_const
+		jsr  FADD
++		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+func_floor_stack	.proc
+		jsr  func_floor_fac1
+		jmp  push_fac1
+		.pend
+
+func_floor_fac1	.proc
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		jsr  INT
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+func_round_stack	.proc
+		jsr  func_round_fac1
+		jmp  push_fac1
+		.pend
+
+func_round_fac1	.proc
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		jsr  FADDH
+		jsr  INT
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+func_sin_stack	.proc
+		jsr  func_sin_fac1
+		jmp  push_fac1
+		.pend
+
+func_sin_fac1	.proc
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		jsr  SIN
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+func_cos_stack	.proc
+		jsr  func_cos_fac1
+		jmp  push_fac1
+		.pend
+
+func_cos_fac1	.proc
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		jsr  COS
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+func_tan_stack	.proc
+		jsr  func_tan_fac1
+		jmp  push_fac1
+		.pend
+
+func_tan_fac1	.proc
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		jsr  TAN
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+func_rad_stack	.proc
+		jsr  func_rad_fac1
+		jmp  push_fac1
+		.pend
+
+func_rad_fac1	.proc
+		; -- convert degrees to radians (d * pi / 180)
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		lda  #<_pi_div_180
+		ldy  #>_pi_div_180
+		jsr  FMULT
+		ldx  P8ZP_SCRATCH_REG
+		rts
+_pi_div_180	.byte 123, 14, 250, 53, 18		; pi / 180
+		.pend
+
+func_deg_stack	.proc
+		jsr  func_deg_fac1
+		jmp  push_fac1
+		.pend
+
+func_deg_fac1	.proc
+		; -- convert radians to degrees (d * (1/ pi * 180))
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		lda  #<_one_over_pi_div_180
+		ldy  #>_one_over_pi_div_180
+		jsr  FMULT
+		ldx  P8ZP_SCRATCH_REG
+		rts
+_one_over_pi_div_180	.byte 134, 101, 46, 224, 211		; 1 / (pi * 180)
+		.pend
+
+func_ln_stack	.proc
+		jsr  func_ln_fac1
+		jmp  push_fac1
+		.pend
+
+func_ln_fac1	.proc
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		jsr  LOG
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+func_log2_stack	.proc
+		jsr  func_log2_fac1
+		jmp  push_fac1
+		.pend
+
+func_log2_fac1	.proc
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		jsr  LOG
+		jsr  MOVEF
+		lda  #<FL_LOG2
+		ldy  #>FL_LOG2
+		jsr  MOVFM
+		jsr  FDIVT
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+func_sign_f_stack	.proc
+		jsr  func_sign_f_into_A
+		sta  P8ESTACK_LO,x
+		dex
+		rts
+		.pend
+
+func_sign_f_into_A	.proc
+		jsr  MOVFM
+		jmp  SIGN
+		.pend
+
+func_sqrt_stack	.proc
+		jsr  func_sqrt_fac1
+		jmp  push_fac1
+		.pend
+
+func_sqrt_fac1	.proc
+		jsr  MOVFM
+		stx  P8ZP_SCRATCH_REG
+		jsr  SQR
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+func_rndf_stack	.proc
+		jsr  func_rndf_fac1
+		jmp  push_fac1
+		.pend
+
+func_rndf_fac1	.proc
+		stx  P8ZP_SCRATCH_REG
+		lda  #1
+		jsr  FREADSA
+		jsr  RND		; rng into fac1
+		ldx  P8ZP_SCRATCH_REG
+		rts
+		.pend
+
+func_swap_f	.proc
+		; -- swap floats pointed to by SCRATCH_ZPWORD1, SCRATCH_ZPWORD2
+		ldy  #4
+-               lda  (P8ZP_SCRATCH_W1),y
+		pha
+		lda  (P8ZP_SCRATCH_W2),y
+		sta  (P8ZP_SCRATCH_W1),y
+		pla
+		sta  (P8ZP_SCRATCH_W2),y
+		dey
+		bpl  -
+		rts
+		.pend
+
+func_reverse_f	.proc
+		; --- reverse an array of floats (array in P8ZP_SCRATCH_W1, num elements in A)
+_left_index = P8ZP_SCRATCH_W2
+_right_index = P8ZP_SCRATCH_W2+1
+_loop_count = P8ZP_SCRATCH_REG
+		pha
+		jsr  a_times_5
+		sec
+		sbc  #5
+		sta  _right_index
+		lda  #0
+		sta  _left_index
+		pla
+		lsr  a
+		sta  _loop_count
+_loop		; push the left indexed float on the stack
+		ldy  _left_index
+		lda  (P8ZP_SCRATCH_W1),y
+		pha
+		iny
+		lda  (P8ZP_SCRATCH_W1),y
+		pha
+		iny
+		lda  (P8ZP_SCRATCH_W1),y
+		pha
+		iny
+		lda  (P8ZP_SCRATCH_W1),y
+		pha
+		iny
+		lda  (P8ZP_SCRATCH_W1),y
+		pha
+		; copy right index float to left index float
+		ldy  _right_index
+		lda  (P8ZP_SCRATCH_W1),y
+		ldy  _left_index
+		sta  (P8ZP_SCRATCH_W1),y
+		inc  _left_index
+		inc  _right_index
+		ldy  _right_index
+		lda  (P8ZP_SCRATCH_W1),y
+		ldy  _left_index
+		sta  (P8ZP_SCRATCH_W1),y
+		inc  _left_index
+		inc  _right_index
+		ldy  _right_index
+		lda  (P8ZP_SCRATCH_W1),y
+		ldy  _left_index
+		sta  (P8ZP_SCRATCH_W1),y
+		inc  _left_index
+		inc  _right_index
+		ldy  _right_index
+		lda  (P8ZP_SCRATCH_W1),y
+		ldy  _left_index
+		sta  (P8ZP_SCRATCH_W1),y
+		inc  _left_index
+		inc  _right_index
+		ldy  _right_index
+		lda  (P8ZP_SCRATCH_W1),y
+		ldy  _left_index
+		sta  (P8ZP_SCRATCH_W1),y
+		; pop the float off the stack into the right index float
+		ldy  _right_index
+		pla
+		sta  (P8ZP_SCRATCH_W1),y
+		dey
+		pla
+		sta  (P8ZP_SCRATCH_W1),y
+		dey
+		pla
+		sta  (P8ZP_SCRATCH_W1),y
+		dey
+		pla
+		sta  (P8ZP_SCRATCH_W1),y
+		dey
+		pla
+		sta  (P8ZP_SCRATCH_W1),y
+		inc  _left_index
+		lda  _right_index
+		sec
+		sbc  #9
+		sta  _right_index
+		dec  _loop_count
+		bne  _loop
+		rts
+
+		.pend
+
+
+
+a_times_5	.proc
+		sta  P8ZP_SCRATCH_B1
+		asl  a
+		asl  a
+		clc
+		adc  P8ZP_SCRATCH_B1
+		rts
+		.pend
+
+func_any_f_into_A	.proc
+		jsr  a_times_5
+		jmp  prog8_lib.func_any_b_into_A
+		.pend
+
+func_all_f_into_A	.proc
+		jsr  a_times_5
+		jmp  prog8_lib.func_all_b_into_A
+		.pend
+
+func_any_f_stack	.proc
+		jsr  a_times_5
+		jmp  prog8_lib.func_any_b_stack
+		.pend
+
+func_all_f_stack	.proc
+		jsr  a_times_5
+		jmp  prog8_lib.func_all_b_stack
+		.pend
+
+func_max_f_stack	.proc
+		jsr  func_max_f_fac1
+		jmp  push_fac1
+		.pend
+
+func_max_f_fac1	.proc
+		; -- max(array) -> fac1,  array in P8ZP_SCRATCH_W1, num elts in A
+_loop_count = P8ZP_SCRATCH_REG
+		stx  floats_store_reg
+		sta  _loop_count
+		lda  #255
+		sta  _minmax_cmp+1		; modifying
+		lda  #<_largest_neg_float
+		ldy  #>_largest_neg_float
+_minmax_entry	jsr  MOVFM
+-		lda  P8ZP_SCRATCH_W1
+		ldy  P8ZP_SCRATCH_W1+1
+		jsr  FCOMP
+_minmax_cmp	cmp  #255			; modified
+		bne  +
+		lda  P8ZP_SCRATCH_W1
+		ldy  P8ZP_SCRATCH_W1+1
+		jsr  MOVFM
++		lda  P8ZP_SCRATCH_W1
+		clc
+		adc  #5
+		sta  P8ZP_SCRATCH_W1
+		bcc  +
+		inc  P8ZP_SCRATCH_W1+1
++		dec  _loop_count
+		bne  -
+		ldx  floats_store_reg
+		rts
+_largest_neg_float	.byte 255,255,255,255,255		; largest negative float -1.7014118345e+38
+		.pend
+
+func_min_f_stack	.proc
+		jsr  func_min_f_fac1
+		jmp  push_fac1
+		.pend
+
+func_min_f_fac1	.proc
+		; -- min(array) -> fac1,  array in P8ZP_SCRATCH_W1, num elts in A
+		sta  func_max_f_fac1._loop_count
+		lda  #1
+		sta  func_max_f_fac1._minmax_cmp+1
+		lda  #<_largest_pos_float
+		ldy  #>_largest_pos_float
+		jmp  func_max_f_fac1._minmax_entry
+_largest_pos_float	.byte  255,127,255,255,255		; largest positive float
+		rts
+		.pend
+
+
+func_sum_f_stack	.proc
+		jsr  func_sum_f_fac1
+		jmp  push_fac1
+		.pend
+
+func_sum_f_fac1	.proc
+		; -- sum(array) -> fac1,  array in P8ZP_SCRATCH_W1, num elts in A
+_loop_count = P8ZP_SCRATCH_REG
+		stx  floats_store_reg
+		sta  _loop_count
+		lda  #<FL_ZERO_const
+		ldy  #>FL_ZERO_const
+		jsr  MOVFM
+-		lda  P8ZP_SCRATCH_W1
+		ldy  P8ZP_SCRATCH_W1+1
+		jsr  FADD
+		lda  P8ZP_SCRATCH_W1
+		clc
+		adc  #5
+		sta  P8ZP_SCRATCH_W1
+		bcc  +
+		inc  P8ZP_SCRATCH_W1+1
++		dec  _loop_count
+		bne  -
+		ldx  floats_store_reg
+		rts
+		.pend

compiler/res/prog8lib/c64/graphics.p8

@@ -0,0 +1,249 @@
+%target c64
+%import textio
+
+; bitmap pixel graphics module for the C64
+; only black/white monchrome 320x200 for now
+; assumes bitmap screen memory is $2000-$3fff
+
+graphics {
+    const uword BITMAP_ADDRESS = $2000
+    const uword WIDTH = 320
+    const ubyte HEIGHT = 200
+
+    sub enable_bitmap_mode() {
+        ; enable bitmap screen, erase it and set colors to black/white.
+        c64.SCROLY = %00111011
+        c64.SCROLX = %00001000
+        c64.VMCSB = (c64.VMCSB & %11110000) | %00001000   ; $2000-$3fff
+        clear_screen(1, 0)
+    }
+
+    sub disable_bitmap_mode() {
+        ; enables text mode, erase the text screen, color white
+        c64.SCROLY = %00011011
+        c64.SCROLX = %00001000
+        c64.VMCSB = (c64.VMCSB & %11110000) | %00000100   ; $1000-$2fff
+        txt.fill_screen(' ', 1)
+    }
+
+    sub clear_screen(ubyte pixelcolor, ubyte bgcolor) {
+        memset(BITMAP_ADDRESS, 320*200/8, 0)
+        txt.fill_screen(pixelcolor << 4 | bgcolor, 0)
+    }
+
+    sub line(uword @zp x1, ubyte @zp y1, uword @zp x2, ubyte @zp y2) {
+        ; Bresenham algorithm.
+        ; This code special cases various quadrant loops to allow simple ++ and -- operations.
+        ; TODO rewrite this in optimized assembly
+        if y1>y2 {
+            ; make sure dy is always positive to avoid 8 instead of just 4 special cases
+            swap(x1, x2)
+            swap(y1, y2)
+        }
+        word @zp d = 0
+        ubyte positive_ix = true
+        word @zp dx = x2-x1 as word
+        word @zp dy = y2-y1
+        if dx < 0 {
+            dx = -dx
+            positive_ix = false
+        }
+        dx *= 2
+        dy *= 2
+        internal_plotx = x1
+
+        if dx >= dy {
+            if positive_ix {
+                repeat {
+                    internal_plot(y1)
+                    if internal_plotx==x2
+                        return
+                    internal_plotx++
+                    d += dy
+                    if d > dx {
+                        y1++
+                        d -= dx
+                    }
+                }
+            } else {
+                repeat {
+                    internal_plot(y1)
+                    if internal_plotx==x2
+                        return
+                    internal_plotx--
+                    d += dy
+                    if d > dx {
+                        y1++
+                        d -= dx
+                    }
+                }
+            }
+        }
+        else {
+            if positive_ix {
+                repeat {
+                    internal_plot(y1)
+                    if y1 == y2
+                        return
+                    y1++
+                    d += dx
+                    if d > dy {
+                        internal_plotx++
+                        d -= dy
+                    }
+                }
+            } else {
+                repeat {
+                    internal_plot(y1)
+                    if y1 == y2
+                        return
+                    y1++
+                    d += dx
+                    if d > dy {
+                        internal_plotx--
+                        d -= dy
+                    }
+                }
+            }
+        }
+    }
+
+    sub circle(uword xcenter, ubyte ycenter, ubyte radius) {
+        ; Midpoint algorithm
+        ubyte @zp ploty
+        ubyte @zp xx = radius
+        ubyte @zp yy = 0
+        byte @zp decisionOver2 = 1-xx as byte
+
+        while xx>=yy {
+            internal_plotx = xcenter + xx
+            ploty = ycenter + yy
+            internal_plot(ploty)
+            internal_plotx = xcenter - xx
+            internal_plot(ploty)
+            internal_plotx = xcenter + xx
+            ploty = ycenter - yy
+            internal_plot(ploty)
+            internal_plotx = xcenter - xx
+            internal_plot(ploty)
+            internal_plotx = xcenter + yy
+            ploty = ycenter + xx
+            internal_plot(ploty)
+            internal_plotx = xcenter - yy
+            internal_plot(ploty)
+            internal_plotx = xcenter + yy
+            ploty = ycenter - xx
+            internal_plot(ploty)
+            internal_plotx = xcenter - yy
+            internal_plot(ploty)
+            yy++
+            if decisionOver2<=0
+                decisionOver2 += 2*yy+1
+            else {
+                xx--
+                decisionOver2 += 2*(yy-xx)+1
+            }
+        }
+    }
+
+    sub disc(uword xcenter, ubyte ycenter, ubyte radius) {
+        ; Midpoint algorithm, filled
+        ubyte xx = radius
+        ubyte yy = 0
+        byte decisionOver2 = 1-xx as byte
+
+        while xx>=yy {
+            ubyte ycenter_plus_yy = ycenter + yy
+            ubyte ycenter_min_yy = ycenter - yy
+            ubyte ycenter_plus_xx = ycenter + xx
+            ubyte ycenter_min_xx = ycenter - xx
+
+            internal_plotx = xcenter-xx
+            repeat xx*2+1 {
+                internal_plot(ycenter_plus_yy)
+                internal_plot(ycenter_min_yy)
+                internal_plotx++
+            }
+            internal_plotx = xcenter-yy
+            repeat yy*2+1 {
+                internal_plot(ycenter_plus_xx)
+                internal_plot(ycenter_min_xx)
+                internal_plotx++
+            }
+            yy++
+            if decisionOver2<=0
+                decisionOver2 += 2*yy+1
+            else {
+                xx--
+                decisionOver2 += 2*(yy-xx)+1
+            }
+        }
+    }
+
+
+; here is the non-asm code for the plot routine below:
+;    sub plot_nonasm(uword px, ubyte py) {
+;        ubyte[] ormask = [128, 64, 32, 16, 8, 4, 2, 1]
+;        uword addr = BITMAP_ADDRESS + 320*(py>>3) + (py & 7) + (px & %0000000111111000)
+;        @(addr) |= ormask[lsb(px) & 7]
+;    }
+
+    asmsub  plot(uword plotx @XY, ubyte ploty @A) clobbers (A, X, Y) {
+        %asm {{
+            stx  internal_plotx
+            sty  internal_plotx+1
+            jmp  internal_plot
+        }}
+    }
+
+    ; for efficiency of internal algorithms here is the internal plot routine
+    ; that takes the plotx coordinate in a separate variable instead of the XY register pair:
+
+    uword internal_plotx     ; 0..319        ; separate 'parameter' for internal_plot()
+
+    asmsub  internal_plot(ubyte ploty @A) clobbers (A, X, Y) {      ; internal_plotx is 16 bits 0 to 319... doesn't fit in a register
+        %asm {{
+        tay
+        lda  internal_plotx+1
+        sta  P8ZP_SCRATCH_W2+1
+        lsr  a            ; 0
+        sta  P8ZP_SCRATCH_W2
+        lda  internal_plotx
+        pha
+        and  #7
+        tax
+
+        lda  _y_lookup_lo,y
+        clc
+        adc  P8ZP_SCRATCH_W2
+        sta  P8ZP_SCRATCH_W2
+        lda  _y_lookup_hi,y
+        adc  P8ZP_SCRATCH_W2+1
+        sta  P8ZP_SCRATCH_W2+1
+
+        pla     ; internal_plotx
+        and  #%11111000
+        tay
+        lda  (P8ZP_SCRATCH_W2),y
+        ora  _ormask,x
+        sta  (P8ZP_SCRATCH_W2),y
+        rts
+
+_ormask     .byte 128, 64, 32, 16, 8, 4, 2, 1
+
+; note: this can be even faster if we also have a 256 byte x-lookup table, but hey.
+; see http://codebase64.org/doku.php?id=base:various_techniques_to_calculate_adresses_fast_common_screen_formats_for_pixel_graphics
+; the y lookup tables encodes this formula:  BITMAP_ADDRESS + 320*(py>>3) + (py & 7)    (y from 0..199)
+; We use the 64tass syntax for range expressions to calculate this table on assembly time.
+
+_plot_y_values := $2000 + 320*(range(200)>>3) + (range(200) & 7)
+
+_y_lookup_lo    .byte  <_plot_y_values
+_y_lookup_hi    .byte  >_plot_y_values
+
+        }}
+    }
+
+}
+
+

compiler/res/prog8lib/c64/syslib.p8

@@ -5,11 +5,13 @@
 ;
 ; indent format: TABS, size=8
 
+%target c64
 
 c64 {
         &ubyte  TIME_HI         = $a0       ; software jiffy clock, hi byte
         &ubyte  TIME_MID        = $a1       ;  .. mid byte
         &ubyte  TIME_LO         = $a2       ;    .. lo byte. Updated by IRQ every 1/60 sec
+        &ubyte  STATUS          = $90       ; kernel status variable for I/O
         &ubyte  STKEY           = $91       ; various keyboard statuses (updated by IRQ)
         &ubyte  SFDX            = $cb       ; current key pressed (matrix value) (updated by IRQ)
 
@@ -202,23 +204,23 @@ romsub $FFB4 = TALK(ubyte device @ A) clobbers(A)               ; command serial
 romsub $FFB7 = READST() -> ubyte @ A                            ; read I/O status word
 romsub $FFBA = SETLFS(ubyte logical @ A, ubyte device @ X, ubyte address @ Y)   ; set logical file parameters
 romsub $FFBD = SETNAM(ubyte namelen @ A, str filename @ XY)     ; set filename parameters
-romsub $FFC0 = OPEN() clobbers(A,X,Y)                           ; (via 794 ($31A)) open a logical file
+romsub $FFC0 = OPEN() clobbers(X,Y) -> ubyte @Pc, ubyte @A      ; (via 794 ($31A)) open a logical file
 romsub $FFC3 = CLOSE(ubyte logical @ A) clobbers(A,X,Y)         ; (via 796 ($31C)) close a logical file
-romsub $FFC6 = CHKIN(ubyte logical @ X) clobbers(A,X)           ; (via 798 ($31E)) define an input channel
+romsub $FFC6 = CHKIN(ubyte logical @ X) clobbers(A,X) -> ubyte @Pc    ; (via 798 ($31E)) define an input channel
 romsub $FFC9 = CHKOUT(ubyte logical @ X) clobbers(A,X)          ; (via 800 ($320)) define an output channel
 romsub $FFCC = CLRCHN() clobbers(A,X)                           ; (via 802 ($322)) restore default devices
-romsub $FFCF = CHRIN() clobbers(Y) -> ubyte @ A                 ; (via 804 ($324)) input a character (for keyboard, read a whole line from the screen) A=byte read.
+romsub $FFCF = CHRIN() clobbers(X, Y) -> ubyte @ A   ; (via 804 ($324)) input a character (for keyboard, read a whole line from the screen) A=byte read.
 romsub $FFD2 = CHROUT(ubyte char @ A)                           ; (via 806 ($326)) output a character
 romsub $FFD5 = LOAD(ubyte verify @ A, uword address @ XY) -> ubyte @Pc, ubyte @ A, ubyte @ X, ubyte @ Y     ; (via 816 ($330)) load from device
 romsub $FFD8 = SAVE(ubyte zp_startaddr @ A, uword endaddr @ XY) -> ubyte @ Pc, ubyte @ A                    ; (via 818 ($332)) save to a device
 romsub $FFDB = SETTIM(ubyte low @ A, ubyte middle @ X, ubyte high @ Y)      ; set the software clock
 romsub $FFDE = RDTIM() -> ubyte @ A, ubyte @ X, ubyte @ Y       ; read the software clock
-romsub $FFE1 = STOP() clobbers(A,X) -> ubyte @ Pz, ubyte @ Pc   ; (via 808 ($328)) check the STOP key
-romsub $FFE4 = GETIN() clobbers(X,Y) -> ubyte @ A               ; (via 810 ($32A)) get a character
+romsub $FFE1 = STOP() clobbers(X) -> ubyte @ Pz, ubyte @ A      ; (via 808 ($328)) check the STOP key (and some others in A)
+romsub $FFE4 = GETIN() clobbers(X,Y) -> ubyte @Pc, ubyte @ A    ; (via 810 ($32A)) get a character
 romsub $FFE7 = CLALL() clobbers(A,X)                            ; (via 812 ($32C)) close all files
 romsub $FFEA = UDTIM() clobbers(A,X)                            ; update the software clock
 romsub $FFED = SCREEN() -> ubyte @ X, ubyte @ Y                 ; read number of screen rows and columns
-romsub $FFF0 = PLOT(ubyte col @ Y, ubyte row @ X, ubyte dir @ Pc) -> ubyte @ X, ubyte @ Y       ; read/set position of cursor on screen.  Use c64scr.plot for a 'safe' wrapper that preserves X.
+romsub $FFF0 = PLOT(ubyte col @ Y, ubyte row @ X, ubyte dir @ Pc) -> ubyte @ X, ubyte @ Y       ; read/set position of cursor on screen.  Use txt.plot for a 'safe' wrapper that preserves X.
 romsub $FFF3 = IOBASE() -> uword @ XY                           ; read base address of I/O devices
 
 ; ---- end of C64 ROM kernal routines ----
@@ -249,8 +251,7 @@ asmsub  init_system()  {
         sta  c64.COLOR
         lda  #0
         sta  c64.BGCOL0
-        tax
-        tay
+        jsr  disable_runstop_and_charsetswitch
         clc
         clv
         cli
@@ -258,6 +259,26 @@ asmsub  init_system()  {
     }}
 }
 
+asmsub  reset_system()  {
+    ; Soft-reset the system back to Basic prompt.
+    %asm {{
+        sei
+        lda  #14
+        sta  $01        ; bank the kernal in
+        jmp  (c64.RESET_VEC)
+    }}
+}
+
+asmsub  disable_runstop_and_charsetswitch() {
+    %asm {{
+        lda  #$80
+        sta  657    ; disable charset switching
+        lda  #239
+        sta  808    ; disable run/stop key
+        rts
+    }}
+}
+
 asmsub  set_irqvec_excl() clobbers(A)  {
 	%asm {{
 		sei
@@ -298,8 +319,6 @@ _irq_handler_init
 		sta  IRQ_SCRATCH_ZPB1
 		lda  P8ZP_SCRATCH_REG
 		sta  IRQ_SCRATCH_ZPREG
-		lda  P8ZP_SCRATCH_REG_X
-		sta  IRQ_SCRATCH_ZPREGX
 		lda  P8ZP_SCRATCH_W1
 		sta  IRQ_SCRATCH_ZPWORD1
 		lda  P8ZP_SCRATCH_W1+1
@@ -324,8 +343,6 @@ _irq_handler_end
 		sta  P8ZP_SCRATCH_B1
 		lda  IRQ_SCRATCH_ZPREG
 		sta  P8ZP_SCRATCH_REG
-		lda  IRQ_SCRATCH_ZPREGX
-		sta  P8ZP_SCRATCH_REG_X
 		lda  IRQ_SCRATCH_ZPWORD1
 		sta  P8ZP_SCRATCH_W1
 		lda  IRQ_SCRATCH_ZPWORD1+1
@@ -340,7 +357,6 @@ _irq_handler_end
 IRQ_X_REG		.byte  0
 IRQ_SCRATCH_ZPB1	.byte  0
 IRQ_SCRATCH_ZPREG	.byte  0
-IRQ_SCRATCH_ZPREGX	.byte  0
 IRQ_SCRATCH_ZPWORD1	.word  0
 IRQ_SCRATCH_ZPWORD2	.word  0
 

compiler/res/prog8lib/c64/textio.p8

@@ -4,15 +4,23 @@
 ;
 ; indent format: TABS, size=8
 
-
-%import c64lib
+%target c64
+%import syslib
 %import conv
 
 
 txt {
 
-asmsub  clear_screen (ubyte char @ A, ubyte color @ Y) clobbers(A)  {
-	; ---- clear the character screen with the given fill character and character color.
+const ubyte DEFAULT_WIDTH = 40
+const ubyte DEFAULT_HEIGHT = 25
+
+
+sub  clear_screen() {
+    clear_screenchars(' ')
+}
+
+asmsub  fill_screen (ubyte char @ A, ubyte color @ Y) clobbers(A)  {
+	; ---- fill the character screen with the given fill character and character color.
 	;      (assumes screen and color matrix are at their default addresses)
 
 	%asm {{
@@ -30,13 +38,13 @@ asmsub  clear_screenchars (ubyte char @ A) clobbers(Y)  {
 	; ---- clear the character screen with the given fill character (leaves colors)
 	;      (assumes screen matrix is at the default address)
 	%asm {{
-		ldy  #0
-_loop		sta  c64.Screen,y
-		sta  c64.Screen+$0100,y
-		sta  c64.Screen+$0200,y
-		sta  c64.Screen+$02e8,y
-		iny
-		bne  _loop
+		ldy  #250
+-		sta  c64.Screen+250*0-1,y
+		sta  c64.Screen+250*1-1,y
+		sta  c64.Screen+250*2-1,y
+		sta  c64.Screen+250*3-1,y
+		dey
+		bne  -
 		rts
         }}
 }
@@ -45,158 +53,180 @@ asmsub  clear_screencolors (ubyte color @ A) clobbers(Y)  {
 	; ---- clear the character screen colors with the given color (leaves characters).
 	;      (assumes color matrix is at the default address)
 	%asm {{
-		ldy  #0
-_loop		sta  c64.Colors,y
-		sta  c64.Colors+$0100,y
-		sta  c64.Colors+$0200,y
-		sta  c64.Colors+$02e8,y
-		iny
-		bne  _loop
+		ldy  #250
+-		sta  c64.Colors+250*0-1,y
+		sta  c64.Colors+250*1-1,y
+		sta  c64.Colors+250*2-1,y
+		sta  c64.Colors+250*3-1,y
+		dey
+		bne  -
 		rts
         }}
 }
 
-asmsub  scroll_left_full  (ubyte alsocolors @ Pc) clobbers(A, Y)  {
+sub color (ubyte txtcol) {
+    c64.COLOR = txtcol
+}
+
+sub lowercase() {
+    c64.VMCSB |= 2
+}
+
+sub uppercase() {
+    c64.VMCSB &= ~2
+}
+
+asmsub  scroll_left  (ubyte alsocolors @ Pc) clobbers(A, Y)  {
 	; ---- scroll the whole screen 1 character to the left
 	;      contents of the rightmost column are unchanged, you should clear/refill this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 
 	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
-		bcs  +
-		jmp  _scroll_screen
+		stx  P8ZP_SCRATCH_REG
+		bcc _scroll_screen
 
-+               ; scroll the color memory
++               ; scroll the screen and the color memory
 		ldx  #0
 		ldy  #38
 -
-	.for row=0, row<=24, row+=1
-		lda  c64.Colors + 40*row + 1,x
-		sta  c64.Colors + 40*row,x
-	.next
+        .for row=0, row<=24, row+=1
+            lda  c64.Screen + 40*row + 1,x
+            sta  c64.Screen + 40*row + 0,x
+            lda  c64.Colors + 40*row + 1,x
+            sta  c64.Colors + 40*row + 0,x
+        .next
+		inx
+		dey
+		bpl  -
+		rts
+
+_scroll_screen  ; scroll only the screen memory
+		ldx  #0
+		ldy  #38
+-
+        .for row=0, row<=24, row+=1
+            lda  c64.Screen + 40*row + 1,x
+            sta  c64.Screen + 40*row + 0,x
+        .next
 		inx
 		dey
 		bpl  -
 
-_scroll_screen  ; scroll the screen memory
-		ldx  #0
-		ldy  #38
--
-	.for row=0, row<=24, row+=1
-		lda  c64.Screen + 40*row + 1,x
-		sta  c64.Screen + 40*row,x
-	.next
-		inx
-		dey
-		bpl  -
-
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  scroll_right_full  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_right  (ubyte alsocolors @ Pc) clobbers(A)  {
 	; ---- scroll the whole screen 1 character to the right
 	;      contents of the leftmost column are unchanged, you should clear/refill this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
-		bcs  +
-		jmp  _scroll_screen
+		stx  P8ZP_SCRATCH_REG
+		bcc  _scroll_screen
 
-+               ; scroll the color memory
++               ; scroll the screen and the color memory
 		ldx  #38
 -
-	.for row=0, row<=24, row+=1
-		lda  c64.Colors + 40*row + 0,x
-		sta  c64.Colors + 40*row + 1,x
-	.next
+        .for row=0, row<=24, row+=1
+            lda  c64.Screen + 40*row + 0,x
+            sta  c64.Screen + 40*row + 1,x
+            lda  c64.Colors + 40*row + 0,x
+            sta  c64.Colors + 40*row + 1,x
+        .next
+		dex
+		bpl  -
+		rts
+
+_scroll_screen  ; scroll only the screen memory
+		ldx  #38
+-
+        .for row=0, row<=24, row+=1
+            lda  c64.Screen + 40*row + 0,x
+            sta  c64.Screen + 40*row + 1,x
+        .next
 		dex
 		bpl  -
 
-_scroll_screen  ; scroll the screen memory
-		ldx  #38
--
-	.for row=0, row<=24, row+=1
-		lda  c64.Screen + 40*row + 0,x
-		sta  c64.Screen + 40*row + 1,x
-	.next
-		dex
-		bpl  -
-
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  scroll_up_full  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_up  (ubyte alsocolors @ Pc) clobbers(A)  {
 	; ---- scroll the whole screen 1 character up
 	;      contents of the bottom row are unchanged, you should refill/clear this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
-		bcs  +
-		jmp  _scroll_screen
+		stx  P8ZP_SCRATCH_REG
+		bcc  _scroll_screen
 
-+               ; scroll the color memory
++               ; scroll the screen and the color memory
 		ldx #39
 -
-	.for row=1, row<=24, row+=1
-		lda  c64.Colors + 40*row,x
-		sta  c64.Colors + 40*(row-1),x
-	.next
+        .for row=1, row<=24, row+=1
+            lda  c64.Screen + 40*row,x
+            sta  c64.Screen + 40*(row-1),x
+            lda  c64.Colors + 40*row,x
+            sta  c64.Colors + 40*(row-1),x
+        .next
+		dex
+		bpl  -
+		rts
+
+_scroll_screen  ; scroll only the screen memory
+		ldx #39
+-
+        .for row=1, row<=24, row+=1
+            lda  c64.Screen + 40*row,x
+            sta  c64.Screen + 40*(row-1),x
+        .next
 		dex
 		bpl  -
 
-_scroll_screen  ; scroll the screen memory
-		ldx #39
--
-	.for row=1, row<=24, row+=1
-		lda  c64.Screen + 40*row,x
-		sta  c64.Screen + 40*(row-1),x
-	.next
-		dex
-		bpl  -
-
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  scroll_down_full  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_down  (ubyte alsocolors @ Pc) clobbers(A)  {
 	; ---- scroll the whole screen 1 character down
 	;      contents of the top row are unchanged, you should refill/clear this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
-		bcs  +
-		jmp  _scroll_screen
+		stx  P8ZP_SCRATCH_REG
+		bcc  _scroll_screen
 
-+               ; scroll the color memory
++               ; scroll the screen and the color memory
 		ldx #39
 -
-	.for row=23, row>=0, row-=1
-		lda  c64.Colors + 40*row,x
-		sta  c64.Colors + 40*(row+1),x
-	.next
+        .for row=23, row>=0, row-=1
+            lda  c64.Colors + 40*row,x
+            sta  c64.Colors + 40*(row+1),x
+            lda  c64.Screen + 40*row,x
+            sta  c64.Screen + 40*(row+1),x
+        .next
+		dex
+		bpl  -
+		rts
+
+_scroll_screen  ; scroll only the screen memory
+		ldx #39
+-
+        .for row=23, row>=0, row-=1
+            lda  c64.Screen + 40*row,x
+            sta  c64.Screen + 40*(row+1),x
+        .next
 		dex
 		bpl  -
 
-_scroll_screen  ; scroll the screen memory
-		ldx #39
--
-	.for row=23, row>=0, row-=1
-		lda  c64.Screen + 40*row,x
-		sta  c64.Screen + 40*(row+1),x
-	.next
-		dex
-		bpl  -
-
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
+romsub $FFD2 = chrout(ubyte char @ A)    ; for consistency. You can also use c64.CHROUT directly ofcourse.
+
 asmsub  print (str text @ AY) clobbers(A,Y)  {
 	; ---- print null terminated string from A/Y
 	; note: the compiler contains an optimization that will replace
@@ -218,7 +248,7 @@ asmsub  print (str text @ AY) clobbers(A,Y)  {
 asmsub  print_ub0  (ubyte value @ A) clobbers(A,Y)  {
 	; ---- print the ubyte in A in decimal form, with left padding 0s (3 positions total)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		jsr  conv.ubyte2decimal
 		pha
 		tya
@@ -227,7 +257,7 @@ asmsub  print_ub0  (ubyte value @ A) clobbers(A,Y)  {
 		jsr  c64.CHROUT
 		txa
 		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
@@ -235,7 +265,7 @@ asmsub  print_ub0  (ubyte value @ A) clobbers(A,Y)  {
 asmsub  print_ub  (ubyte value @ A) clobbers(A,Y)  {
 	; ---- print the ubyte in A in decimal form, without left padding 0s
 	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		jsr  conv.ubyte2decimal
 _print_byte_digits
 		pha
@@ -252,7 +282,7 @@ _print_byte_digits
         jsr  c64.CHROUT
 _ones   txa
 		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
@@ -260,7 +290,7 @@ _ones   txa
 asmsub  print_b  (byte value @ A) clobbers(A,Y)  {
 	; ---- print the byte in A in decimal form, without left padding 0s
 	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		pha
 		cmp  #0
 		bpl  +
@@ -268,16 +298,14 @@ asmsub  print_b  (byte value @ A) clobbers(A,Y)  {
 		jsr  c64.CHROUT
 +		pla
 		jsr  conv.byte2decimal
-		jsr  print_ub._print_byte_digits
-		ldx  P8ZP_SCRATCH_REG_X
-		rts
+		jmp  print_ub._print_byte_digits
 	}}
 }
 
 asmsub  print_ubhex  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
 	; ---- print the ubyte in A in hex form (if Carry is set, a radix prefix '$' is printed as well)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		bcc  +
 		pha
 		lda  #'$'
@@ -287,7 +315,7 @@ asmsub  print_ubhex  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
 		jsr  c64.CHROUT
 		tya
 		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
@@ -295,7 +323,7 @@ asmsub  print_ubhex  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
 asmsub  print_ubbin  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
 	; ---- print the ubyte in A in binary form (if Carry is set, a radix prefix '%' is printed as well)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		sta  P8ZP_SCRATCH_B1
 		bcc  +
 		lda  #'%'
@@ -308,7 +336,7 @@ asmsub  print_ubbin  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
 +		jsr  c64.CHROUT
 		dey
 		bne  -
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
@@ -341,7 +369,7 @@ asmsub  print_uwhex  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
 asmsub  print_uw0  (uword value @ AY) clobbers(A,Y)  {
 	; ---- print the uword in A/Y in decimal form, with left padding 0s (5 positions total)
 	%asm {{
-	    stx  P8ZP_SCRATCH_REG_X
+	    stx  P8ZP_SCRATCH_REG
 		jsr  conv.uword2decimal
 		ldy  #0
 -		lda  conv.uword2decimal.decTenThousands,y
@@ -349,7 +377,7 @@ asmsub  print_uw0  (uword value @ AY) clobbers(A,Y)  {
 		jsr  c64.CHROUT
 		iny
 		bne  -
-+		ldx  P8ZP_SCRATCH_REG_X
++		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
@@ -357,9 +385,9 @@ asmsub  print_uw0  (uword value @ AY) clobbers(A,Y)  {
 asmsub  print_uw  (uword value @ AY) clobbers(A,Y)  {
 	; ---- print the uword in A/Y in decimal form, without left padding 0s
 	%asm {{
-	    stx  P8ZP_SCRATCH_REG_X
+	    stx  P8ZP_SCRATCH_REG
 		jsr  conv.uword2decimal
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		ldy  #0
 -		lda  conv.uword2decimal.decTenThousands,y
 		beq  _allzero
@@ -422,21 +450,22 @@ asmsub  input_chars  (uword buffer @ AY) clobbers(A) -> ubyte @ Y  {
 	}}
 }
 
-asmsub  setchr  (ubyte col @Y, ubyte row @A) clobbers(A)  {
-	; ---- set the character in SCRATCH_ZPB1 on the screen matrix at the given position
+asmsub  setchr  (ubyte col @X, ubyte row @Y, ubyte character @A) clobbers(A, Y)  {
+	; ---- sets the character in the screen matrix at the given position
 	%asm {{
-		sty  P8ZP_SCRATCH_REG
+		pha
+		tya
 		asl  a
 		tay
 		lda  _screenrows+1,y
 		sta  _mod+2
-		lda  _screenrows,y
+		txa
 		clc
-		adc  P8ZP_SCRATCH_REG
+		adc  _screenrows,y
 		sta  _mod+1
 		bcc  +
 		inc  _mod+2
-+		lda  P8ZP_SCRATCH_B1
++		pla
 _mod		sta  $ffff		; modified
 		rts
 
@@ -444,17 +473,18 @@ _screenrows	.word  $0400 + range(0, 1000, 40)
 	}}
 }
 
-asmsub  getchr  (ubyte col @Y, ubyte row @A) clobbers(Y) -> ubyte @ A {
+asmsub  getchr  (ubyte col @A, ubyte row @Y) clobbers(Y) -> ubyte @ A {
 	; ---- get the character in the screen matrix at the given location
 	%asm  {{
-		sty  P8ZP_SCRATCH_B1
+		pha
+		tya
 		asl  a
 		tay
 		lda  setchr._screenrows+1,y
 		sta  _mod+2
-		lda  setchr._screenrows,y
+		pla
 		clc
-		adc  P8ZP_SCRATCH_B1
+		adc  setchr._screenrows,y
 		sta  _mod+1
 		bcc  _mod
 		inc  _mod+2
@@ -463,21 +493,22 @@ _mod		lda  $ffff		; modified
 	}}
 }
 
-asmsub  setclr  (ubyte col @Y, ubyte row @A) clobbers(A)  {
-	; ---- set the color in SCRATCH_ZPB1 on the screen matrix at the given position
+asmsub  setclr  (ubyte col @X, ubyte row @Y, ubyte color @A) clobbers(A, Y)  {
+	; ---- set the color in A on the screen matrix at the given position
 	%asm {{
-		sty  P8ZP_SCRATCH_REG
+		pha
+		tya
 		asl  a
 		tay
 		lda  _colorrows+1,y
 		sta  _mod+2
-		lda  _colorrows,y
+		txa
 		clc
-		adc  P8ZP_SCRATCH_REG
+		adc  _colorrows,y
 		sta  _mod+1
 		bcc  +
 		inc  _mod+2
-+		lda  P8ZP_SCRATCH_B1
++		pla
 _mod		sta  $ffff		; modified
 		rts
 
@@ -485,17 +516,18 @@ _colorrows	.word  $d800 + range(0, 1000, 40)
 	}}
 }
 
-asmsub  getclr  (ubyte col @Y, ubyte row @A) clobbers(Y) -> ubyte @ A {
+asmsub  getclr  (ubyte col @A, ubyte row @Y) clobbers(Y) -> ubyte @ A {
 	; ---- get the color in the screen color matrix at the given location
 	%asm  {{
-		sty  P8ZP_SCRATCH_B1
+		pha
+		tya
 		asl  a
 		tay
 		lda  setclr._colorrows+1,y
 		sta  _mod+2
-		lda  setclr._colorrows,y
+		pla
 		clc
-		adc  P8ZP_SCRATCH_B1
+		adc  setclr._colorrows,y
 		sta  _mod+1
 		bcc  _mod
 		inc  _mod+2
@@ -504,7 +536,7 @@ _mod		lda  $ffff		; modified
 	}}
 }
 
-sub  setcc  (ubyte column, ubyte row, ubyte char, ubyte color)  {
+sub  setcc  (ubyte column, ubyte row, ubyte char, ubyte charcolor)  {
 	; ---- set char+color at the given position on the screen
 	%asm {{
 		lda  row
@@ -524,7 +556,7 @@ sub  setcc  (ubyte column, ubyte row, ubyte char, ubyte color)  {
 		inc  _colormod+2
 +		lda  char
 _charmod	sta  $ffff		; modified
-		lda  color
+		lda  charcolor
 _colormod	sta  $ffff		; modified
 		rts
 	}}
@@ -533,13 +565,31 @@ _colormod	sta  $ffff		; modified
 asmsub  plot  (ubyte col @ Y, ubyte row @ A) clobbers(A) {
 	; ---- safe wrapper around PLOT kernel routine, to save the X register.
 	%asm  {{
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		tax
 		clc
 		jsr  c64.PLOT
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
+asmsub width() clobbers(X,Y) -> ubyte @A {
+    ; -- returns the text screen width (number of columns)
+    %asm {{
+        jsr  c64.SCREEN
+        txa
+        rts
+    }}
+}
+
+asmsub height() clobbers(X, Y) -> ubyte @A {
+    ; -- returns the text screen height (number of rows)
+    %asm {{
+        jsr  c64.SCREEN
+        tya
+        rts
+    }}
+}
+
 }

compiler/res/prog8lib/conv.p8

@@ -213,7 +213,7 @@ asmsub  byte2decimal  (byte value @ A) -> ubyte @ Y, ubyte @ A, ubyte @ X  {
 asmsub  ubyte2hex  (ubyte value @ A) -> ubyte @ A, ubyte @ Y  {
 	; ---- A to hex petscii string in AY (first hex char in A, second hex char in Y)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
+		stx  P8ZP_SCRATCH_REG
 		pha
 		and  #$0f
 		tax
@@ -225,7 +225,7 @@ asmsub  ubyte2hex  (ubyte value @ A) -> ubyte @ A, ubyte @ Y  {
 		lsr  a
 		tax
 		lda  _hex_digits,x
-		ldx  P8ZP_SCRATCH_REG_X
+		ldx  P8ZP_SCRATCH_REG
 		rts
 
 _hex_digits	.text "0123456789abcdef"	; can probably be reused for other stuff as well
@@ -249,7 +249,26 @@ output	.text  "0000", $00      ; 0-terminated output buffer (to make printing ea
 	}}
 }
 
-asmsub str2uword(str string @ AY) -> uword @ AY {
+
+asmsub  str2ubyte(str string @ AY) clobbers(Y) -> ubyte @A {
+	; -- returns the unsigned byte value of the string number argument in AY
+	;    the number may NOT be preceded by a + sign and may NOT contain spaces
+	;    (any non-digit character will terminate the number string that is parsed)
+	%asm {{
+    	jmp  str2uword
+	}}
+}
+
+asmsub  str2byte(str string @ AY) clobbers(Y) -> ubyte @A {
+	; -- returns the signed byte value of the string number argument in AY
+	;    the number may be preceded by a + or - sign but may NOT contain spaces
+	;    (any non-digit character will terminate the number string that is parsed)
+	%asm {{
+    	jmp  str2word
+	}}
+}
+
+asmsub  str2uword(str string @ AY) -> uword @ AY {
 	; -- returns the unsigned word value of the string number argument in AY
 	;    the number may NOT be preceded by a + sign and may NOT contain spaces
 	;    (any non-digit character will terminate the number string that is parsed)
@@ -303,7 +322,7 @@ _result_times_10     ; (W*4 + W)*2
 	}}
 }
 
-asmsub str2word(str string @ AY) -> word @ AY {
+asmsub  str2word(str string @ AY) -> word @ AY {
 	; -- returns the signed word value of the string number argument in AY
 	;    the number may be preceded by a + or - sign but may NOT contain spaces
 	;    (any non-digit character will terminate the number string that is parsed)
@@ -358,4 +377,77 @@ _negative	.byte  0
 	}}
 }
 
+asmsub  hex2uword(str string @ AY) -> uword @AY {
+    ; -- hexadecimal string with or without '$' to uword.
+    ;    string may be in petscii or c64-screencode encoding.
+    %asm {{
+        sta  P8ZP_SCRATCH_W2
+        sty  P8ZP_SCRATCH_W2+1
+        ldy  #0
+        sty  P8ZP_SCRATCH_W1
+        sty  P8ZP_SCRATCH_W1+1
+_loop       ldy  #0
+        sty  P8ZP_SCRATCH_B1
+        lda  (P8ZP_SCRATCH_W2),y
+        beq  _stop
+        cmp  #'$'
+        beq  _skip
+        cmp  #7
+        bcc  _add_nine
+        cmp  #'9'
+        beq  _calc
+        bcs  _add_nine
+_calc       asl  P8ZP_SCRATCH_W1
+        rol  P8ZP_SCRATCH_W1+1
+        asl  P8ZP_SCRATCH_W1
+        rol  P8ZP_SCRATCH_W1+1
+        asl  P8ZP_SCRATCH_W1
+        rol  P8ZP_SCRATCH_W1+1
+        asl  P8ZP_SCRATCH_W1
+        rol  P8ZP_SCRATCH_W1+1
+        and  #$0f
+        clc
+        adc  P8ZP_SCRATCH_B1
+        ora  P8ZP_SCRATCH_W1
+        sta  P8ZP_SCRATCH_W1
+_skip       inc  P8ZP_SCRATCH_W2
+        bne  _loop
+        inc  P8ZP_SCRATCH_W2+1
+        bne  _loop
+_stop       lda  P8ZP_SCRATCH_W1
+        ldy  P8ZP_SCRATCH_W1+1
+        rts
+_add_nine   ldy  #9
+        sty  P8ZP_SCRATCH_B1
+        bne  _calc
+    }}
+}
+
+asmsub  bin2uword(str string @ AY) -> uword @AY {
+    ; -- binary string with or without '%' to uword.
+    %asm {{
+        sta  P8ZP_SCRATCH_W2
+        sty  P8ZP_SCRATCH_W2+1
+        ldy  #0
+        sty  P8ZP_SCRATCH_W1
+        sty  P8ZP_SCRATCH_W1+1
+_loop       lda  (P8ZP_SCRATCH_W2),y
+        beq  _stop
+        cmp  #'%'
+        beq  +
+        asl  P8ZP_SCRATCH_W1
+        rol  P8ZP_SCRATCH_W1+1
+        and  #1
+        ora  P8ZP_SCRATCH_W1
+        sta  P8ZP_SCRATCH_W1
++           inc  P8ZP_SCRATCH_W2
+        bne  _loop
+        inc  P8ZP_SCRATCH_W2+1
+        bne  _loop
+_stop       lda  P8ZP_SCRATCH_W1
+        ldy  P8ZP_SCRATCH_W1+1
+        rts
+    }}
+}
+
 }

compiler/res/prog8lib/cx16/c64colors.p8

@@ -0,0 +1,87 @@
+%target cx16
+
+c64colors {
+    uword[] colorpalette_dark = [   ; this is a darker palette with more contrast
+        $000,  ; 0 = black
+        $FFF,  ; 1 = white
+        $632,  ; 2 = red
+        $7AB,  ; 3 = cyan
+        $638,  ; 4 = purple
+        $584,  ; 5 = green
+        $327,  ; 6 = blue
+        $BC6,  ; 7 = yellow
+        $642,  ; 8 = orange
+        $430,  ; 9 = brown
+        $965,  ; 10 = light red
+        $444,  ; 11 = dark grey
+        $666,  ; 12 = medium grey
+        $9D8,  ; 13 = light green
+        $65B,  ; 14 = light blue
+        $999  ; 15 = light grey
+    ]
+
+    uword[] colorpalette_pepto = [  ; # this is Pepto's Commodore-64 palette  http://www.pepto.de/projects/colorvic/
+        $000,  ; 0 = black
+        $FFF,  ; 1 = white
+        $833,  ; 2 = red
+        $7cc,  ; 3 = cyan
+        $839,  ; 4 = purple
+        $5a4,  ; 5 = green
+        $229,  ; 6 = blue
+        $ef7,  ; 7 = yellow
+        $852,  ; 8 = orange
+        $530,  ; 9 = brown
+        $c67,  ; 10 = light red
+        $444,  ; 11 = dark grey
+        $777,  ; 12 = medium grey
+        $af9,  ; 13 = light green
+        $76e,  ; 14 = light blue
+        $bbb  ; 15 = light grey
+    ]
+
+    uword[] colorpalette_light = [  ; this is a lighter palette
+        $000,  ; 0 = black
+        $FFF,  ; 1 = white
+        $944,  ; 2 = red
+        $7CC,  ; 3 = cyan
+        $95A,  ; 4 = purple
+        $6A5,  ; 5 = green
+        $549,  ; 6 = blue
+        $CD8,  ; 7 = yellow
+        $963,  ; 8 = orange
+        $650,  ; 9 = brown
+        $C77,  ; 10 = light red
+        $666,  ; 11 = dark grey
+        $888,  ; 12 = medium grey
+        $AE9,  ; 13 = light green
+        $87C,  ; 14 = light blue
+        $AAA   ; 15 = light grey
+    ]
+
+    ubyte color
+
+    sub set_palette_pepto() {
+        for color in 0 to 15 {
+            uword cc = colorpalette_pepto[color]
+            cx16.vpoke(1, $fa00 + color*2, lsb(cc))     ; G, B
+            cx16.vpoke(1, $fa01 + color*2, msb(cc))     ; R
+        }
+    }
+
+    sub set_palette_light() {
+        for color in 0 to 15 {
+            uword cc = colorpalette_light[color]
+            cx16.vpoke(1, $fa00 + color*2, lsb(cc))     ; G, B
+            cx16.vpoke(1, $fa01 + color*2, msb(cc))     ; R
+        }
+    }
+
+    sub set_palette_dark() {
+        for color in 0 to 15 {
+            uword cc = colorpalette_dark[color]
+            cx16.vpoke(1, $fa00 + color*2, lsb(cc))     ; G, B
+            cx16.vpoke(1, $fa01 + color*2, msb(cc))     ; R
+        }
+    }
+
+}

compiler/res/prog8lib/cx16/floats.p8

@@ -0,0 +1,154 @@
+; Prog8 definitions for floating point handling on the CommanderX16
+;
+; Written by Irmen de Jong (irmen@razorvine.net) - license: GNU GPL 3.0
+;
+; indent format: TABS, size=8
+
+%target cx16
+%option enable_floats
+
+floats {
+	; ---- this block contains C-64 floating point related functions ----
+
+        const float  PI     = 3.141592653589793
+        const float  TWOPI  = 6.283185307179586
+
+
+; ---- ROM float functions ----
+
+		; note: the fac1 and fac2 are working registers and take 6 bytes each,
+		; floats in memory  (and rom) are stored in 5-byte MFLPT packed format.
+
+; note: fac1/2 might get clobbered even if not mentioned in the function's name.
+; note: for subtraction and division, the left operand is in fac2, the right operand in fac1.
+
+romsub $fe00 = AYINT() clobbers(A,X,Y)          ; fac1-> signed word in 100-101 ($64-$65) MSB FIRST. (might throw ILLEGAL QUANTITY)
+
+; GIVAYF: signed word in Y/A (note different lsb/msb order) -> float in fac1
+; there is also floats.GIVUAYFAY - unsigned word in A/Y (lo/hi) to fac1
+; (tip: use GIVAYFAY to use A/Y input; lo/hi switched to normal order)
+romsub $fe03 = GIVAYF(ubyte lo @ Y, ubyte hi @ A) clobbers(A,X,Y)
+
+; fac1 -> unsigned word in Y/A (might throw ILLEGAL QUANTITY) (result also in $14/15)
+; (tip: use GETADRAY to get A/Y output; lo/hi switched to normal little endian order)
+romsub $fe06 = GETADR() clobbers(X) -> ubyte @ Y, ubyte @ A
+
+romsub $fe09 = FADDH() clobbers(A,X,Y)                      ; fac1 += 0.5, for rounding- call this before INT
+romsub $fe0c = FSUB(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = mflpt from A/Y - fac1
+romsub $fe0f = FSUBT() clobbers(A,X,Y)                      ; fac1 = fac2-fac1   mind the order of the operands
+romsub $fe12 = FADD(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 += mflpt value from A/Y
+romsub $fe15 = FADDT() clobbers(A,X,Y)                      ; fac1 += fac2
+romsub $fe1b = ZEROFC() clobbers(A,X,Y)                     ; fac1 = 0
+romsub $fe1e = NORMAL() clobbers(A,X,Y)                     ; normalize fac1 (?)
+romsub $fe24 = LOG() clobbers(A,X,Y)                        ; fac1 = LN(fac1)  (natural log)
+romsub $fe27 = FMULT(uword mflpt @ AY) clobbers(A,X,Y)      ; fac1 *= mflpt value from A/Y
+romsub $fe2a = FMULTT() clobbers(A,X,Y)                     ; fac1 *= fac2
+romsub $fe33 = CONUPK(uword mflpt @ AY) clobbers(A,Y)       ; load mflpt value from memory  in A/Y into fac2
+romsub $fe36 = MUL10() clobbers(A,X,Y)                      ; fac1 *= 10
+romsub $fe3c = DIV10() clobbers(A,X,Y)                      ; fac1 /= 10 , CAUTION: result is always positive!
+romsub $fe3f = FDIV(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = mflpt in A/Y / fac1  (remainder in fac2)
+romsub $fe42 = FDIVT() clobbers(A,X,Y)                      ; fac1 = fac2/fac1  (remainder in fac2)  mind the order of the operands
+
+romsub $fe48 = MOVFM(uword mflpt @ AY) clobbers(A,Y)        ; load mflpt value from memory  in A/Y into fac1
+romsub $fe4b = MOVMF(uword mflpt @ XY) clobbers(A,Y)        ; store fac1 to memory  X/Y as 5-byte mflpt
+romsub $fe4e = MOVFA() clobbers(A,X)                        ; copy fac2 to fac1
+romsub $fe51 = MOVAF() clobbers(A,X)                        ; copy fac1 to fac2  (rounded)
+romsub $fe54 = MOVEF() clobbers(A,X)                        ; copy fac1 to fac2
+romsub $fe5a = SIGN() -> ubyte @ A                          ; SIGN(fac1) to A, $ff, $0, $1 for negative, zero, positive
+romsub $fe5d = SGN() clobbers(A,X,Y)                        ; fac1 = SGN(fac1), result of SIGN (-1, 0 or 1)
+romsub $fe60 = FREADSA(byte value @ A) clobbers(A,X,Y)      ; 8 bit signed A -> float in fac1
+romsub $fe6c = ABS()                                        ; fac1 = ABS(fac1)
+romsub $fe6f = FCOMP(uword mflpt @ AY) clobbers(X,Y) -> ubyte @ A   ; A = compare fac1 to mflpt in A/Y, 0=equal 1=fac1 is greater, 255=fac1 is less than
+romsub $fe78 = INT() clobbers(A,X,Y)                        ; INT() truncates, use FADDH first to round instead of trunc
+romsub $fe7e = FINLOG(byte value @A) clobbers (A, X, Y)           ; fac1 += signed byte in A
+romsub $fe81 = FOUT() clobbers(X) -> uword @ AY             ; fac1 -> string, address returned in AY
+romsub $fe8a = SQR() clobbers(A,X,Y)                        ; fac1 = SQRT(fac1)
+romsub $fe8d = FPWRT() clobbers(A,X,Y)                      ; fac1 = fac2 ** fac1
+romsub $fe93 = NEGOP() clobbers(A)                          ; switch the sign of fac1
+romsub $fe96 = EXP() clobbers(A,X,Y)                        ; fac1 = EXP(fac1)  (e ** fac1)
+romsub $fe9f = RND2(byte value @A) clobbers(A,X,Y)                ; fac1 = RND(A) float random number generator
+romsub $fea2 = RND() clobbers(A,X,Y)                        ; fac1 = RND(fac1) float random number generator
+romsub $fea5 = COS() clobbers(A,X,Y)                        ; fac1 = COS(fac1)
+romsub $fea8 = SIN() clobbers(A,X,Y)                        ; fac1 = SIN(fac1)
+romsub $feab = TAN() clobbers(A,X,Y)                        ; fac1 = TAN(fac1)
+romsub $feae = ATN() clobbers(A,X,Y)                        ; fac1 = ATN(fac1)
+
+
+asmsub  GIVUAYFAY  (uword value @ AY) clobbers(A,X,Y)  {
+	; ---- unsigned 16 bit word in A/Y (lo/hi) to fac1
+	%asm {{
+        phx
+        sta  P8ZP_SCRATCH_W2
+        sty  P8ZP_SCRATCH_B1
+        tya
+        ldy  P8ZP_SCRATCH_W2
+        jsr  GIVAYF                 ; load it as signed... correct afterwards
+        lda  P8ZP_SCRATCH_B1
+	    bpl  +
+	    lda  #<_flt65536
+	    ldy  #>_flt65536
+	    jsr  FADD
++	    plx
+        rts
+_flt65536    .byte 145,0,0,0,0       ; 65536.0
+	}}
+}
+
+
+asmsub  GIVAYFAY  (uword value @ AY) clobbers(A,X,Y)  {
+	; ---- signed 16 bit word in A/Y (lo/hi) to float in fac1
+	%asm {{
+		sta  P8ZP_SCRATCH_W2
+		tya
+		ldy  P8ZP_SCRATCH_W2
+		jmp  GIVAYF		; this uses the inverse order, Y/A
+	}}
+}
+
+asmsub  FTOSWRDAY  () clobbers(X) -> uword @ AY  {
+	; ---- fac1 to signed word in A/Y
+	%asm {{
+		jsr  FTOSWORDYA	; note the inverse Y/A order
+		sta  P8ZP_SCRATCH_REG
+		tya
+		ldy  P8ZP_SCRATCH_REG
+		rts
+	}}
+}
+
+asmsub  GETADRAY  () clobbers(X) -> uword @ AY  {
+	; ---- fac1 to unsigned word in A/Y
+	%asm {{
+		jsr  GETADR		; this uses the inverse order, Y/A
+		sta  P8ZP_SCRATCH_B1
+		tya
+		ldy  P8ZP_SCRATCH_B1
+		rts
+	}}
+}
+
+sub  print_f  (float value) {
+	; ---- prints the floating point value (without a newline).
+	%asm {{
+		phx
+		lda  #<value
+		ldy  #>value
+		jsr  MOVFM		; load float into fac1
+		jsr  FOUT		; fac1 to string in A/Y
+		sta  P8ZP_SCRATCH_W1
+		sty  P8ZP_SCRATCH_W1+1
+		ldy  #0
+-		lda  (P8ZP_SCRATCH_W1),y
+		beq  +
+		jsr  c64.CHROUT
+		iny
+		bne  -
++		plx
+		rts
+	}}
+}
+
+%asminclude "library:c64/floats.asm", ""
+%asminclude "library:c64/floats_funcs.asm", ""
+
+}

compiler/res/prog8lib/cx16/graphics.p8

@@ -0,0 +1,152 @@
+%target cx16
+%import syslib
+%import textio
+
+; bitmap pixel graphics module for the CommanderX16
+; wraps the graphics functions that are in ROM.
+; only black/white monchrome 320x200 for now.
+
+graphics {
+    const uword WIDTH = 320
+    const ubyte HEIGHT = 200
+
+    sub enable_bitmap_mode() {
+        ; enable bitmap screen, erase it and set colors to black/white.
+        void cx16.screen_set_mode($80)
+        cx16.r0 = 0
+        cx16.GRAPH_init()
+        clear_screen(1, 0)
+    }
+
+    sub disable_bitmap_mode() {
+        ; enables text mode, erase the text screen, color white
+        void cx16.screen_set_mode(2)
+        txt.fill_screen(' ', 1)     ; doesn't seem to fully clear the text screen after returning from gfx mode
+    }
+
+
+    sub clear_screen(ubyte pixelcolor, ubyte bgcolor) {
+        cx16.GRAPH_set_colors(pixelcolor, pixelcolor, bgcolor)
+        cx16.GRAPH_clear()
+    }
+
+    sub line(uword @zp x1, ubyte @zp y1, uword @zp x2, ubyte @zp y2) {
+        cx16.r0 = x1
+        cx16.r1 = y1
+        cx16.r2 = x2
+        cx16.r3 = y2
+        cx16.GRAPH_draw_line()
+    }
+
+    sub circle(uword xcenter, ubyte ycenter, ubyte radius) {
+        ;cx16.r0 = xcenter - radius/2
+        ;cx16.r1 = ycenter - radius/2
+        ;cx16.r2 = radius*2
+        ;cx16.r3 = radius*2
+        ;cx16.GRAPH_draw_oval(false)          ; currently this call is not implemented on cx16, does a BRK
+
+        ; Midpoint algorithm
+        ubyte @zp xx = radius
+        ubyte @zp yy = 0
+        byte @zp decisionOver2 = 1-xx as byte
+
+        while xx>=yy {
+            cx16.r0 = xcenter + xx
+            cx16.r1 = ycenter + yy
+            cx16.FB_cursor_position()
+            cx16.FB_set_pixel(1)
+            cx16.r0 = xcenter - xx
+            cx16.FB_cursor_position()
+            cx16.FB_set_pixel(1)
+            cx16.r0 = xcenter + xx
+            cx16.r1 = ycenter - yy
+            cx16.FB_cursor_position()
+            cx16.FB_set_pixel(1)
+            cx16.r0 = xcenter - xx
+            cx16.FB_cursor_position()
+            cx16.FB_set_pixel(1)
+            cx16.r0 = xcenter + yy
+            cx16.r1 = ycenter + xx
+            cx16.FB_cursor_position()
+            cx16.FB_set_pixel(1)
+            cx16.r0 = xcenter - yy
+            cx16.FB_cursor_position()
+            cx16.FB_set_pixel(1)
+            cx16.r0 = xcenter + yy
+            cx16.r1 = ycenter - xx
+            cx16.FB_cursor_position()
+            cx16.FB_set_pixel(1)
+            cx16.r0 = xcenter - yy
+            cx16.FB_cursor_position()
+            cx16.FB_set_pixel(1)
+            yy++
+            if decisionOver2<=0 {
+                decisionOver2 += 2*yy+1
+            } else {
+                xx--
+                decisionOver2 += 2*(yy-xx)+1
+            }
+        }
+    }
+
+    sub disc(uword xcenter, ubyte ycenter, ubyte radius) {
+;        cx16.r0 = xcenter - radius/2
+;        cx16.r1 = ycenter - radius/2
+;        cx16.r2 = radius*2
+;        cx16.r3 = radius*2
+;        cx16.GRAPH_draw_oval(true)          ; currently this call is not implemented on cx16, does a BRK
+
+        ubyte xx = radius
+        ubyte yy = 0
+        byte decisionOver2 = 1-xx as byte
+
+        while xx>=yy {
+            ubyte ycenter_plus_yy = ycenter + yy
+            ubyte ycenter_min_yy = ycenter - yy
+            ubyte ycenter_plus_xx = ycenter + xx
+            ubyte ycenter_min_xx = ycenter - xx
+            uword @zp plotx
+
+            cx16.r0 = xcenter-xx
+            cx16.r1 = ycenter_plus_yy
+            cx16.FB_cursor_position()
+            repeat xx*2+1
+                cx16.FB_set_pixel(1)
+
+            cx16.r0 = xcenter-xx
+            cx16.r1 = ycenter_min_yy
+            cx16.FB_cursor_position()
+            repeat xx*2+1
+                cx16.FB_set_pixel(1)
+
+            cx16.r0 = xcenter-yy
+            cx16.r1 = ycenter_plus_xx
+            cx16.FB_cursor_position()
+            repeat yy*2+1
+                cx16.FB_set_pixel(1)
+
+            cx16.r0 = xcenter-yy
+            cx16.r1 = ycenter_min_xx
+            cx16.FB_cursor_position()
+            repeat yy*2+1
+                cx16.FB_set_pixel(1)
+
+            yy++
+            if decisionOver2<=0
+                decisionOver2 += 2*yy+1
+            else {
+                xx--
+                decisionOver2 += 2*(yy-xx)+1
+            }
+        }
+    }
+
+    sub  plot(uword plotx, ubyte ploty) {
+        cx16.r0 = plotx
+        cx16.r1 = ploty
+        cx16.FB_cursor_position()
+        cx16.FB_set_pixel(1)
+    }
+}
+
+

compiler/res/prog8lib/cx16/syslib.p8

@@ -0,0 +1,326 @@
+; Prog8 definitions for the CommanderX16
+; Including memory registers, I/O registers, Basic and Kernal subroutines.
+;
+; Written by Irmen de Jong (irmen@razorvine.net) - license: GNU GPL 3.0
+;
+; indent format: TABS, size=8
+
+%target cx16
+
+
+c64 {
+
+; ---- kernal routines, these are the same as on the Commodore-64 (hence the same block name) ----
+
+; STROUT --> use txt.print
+; CLEARSCR -> use txt.clear_screen
+; HOMECRSR -> use txt.plot
+
+romsub $FF81 = CINT() clobbers(A,X,Y)                           ; (alias: SCINIT) initialize screen editor and video chip
+romsub $FF84 = IOINIT() clobbers(A, X)                          ; initialize I/O devices (CIA, SID, IRQ)
+romsub $FF87 = RAMTAS() clobbers(A,X,Y)                         ; initialize RAM, tape buffer, screen
+romsub $FF8A = RESTOR() clobbers(A,X,Y)                         ; restore default I/O vectors
+romsub $FF8D = VECTOR(uword userptr @ XY, ubyte dir @ Pc) clobbers(A,Y)     ; read/set I/O vector table
+romsub $FF90 = SETMSG(ubyte value @ A)                          ; set Kernal message control flag
+romsub $FF93 = SECOND(ubyte address @ A) clobbers(A)            ; (alias: LSTNSA) send secondary address after LISTEN
+romsub $FF96 = TKSA(ubyte address @ A) clobbers(A)              ; (alias: TALKSA) send secondary address after TALK
+romsub $FF99 = MEMTOP(uword address @ XY, ubyte dir @ Pc) -> uword @ XY     ; read/set top of memory  pointer
+romsub $FF9C = MEMBOT(uword address @ XY, ubyte dir @ Pc) -> uword @ XY     ; read/set bottom of memory  pointer
+romsub $FF9F = SCNKEY() clobbers(A,X,Y)                         ; scan the keyboard
+romsub $FFA2 = SETTMO(ubyte timeout @ A)                        ; set time-out flag for IEEE bus
+romsub $FFA5 = ACPTR() -> ubyte @ A                             ; (alias: IECIN) input byte from serial bus
+romsub $FFA8 = CIOUT(ubyte databyte @ A)                        ; (alias: IECOUT) output byte to serial bus
+romsub $FFAB = UNTLK() clobbers(A)                              ; command serial bus device to UNTALK
+romsub $FFAE = UNLSN() clobbers(A)                              ; command serial bus device to UNLISTEN
+romsub $FFB1 = LISTEN(ubyte device @ A) clobbers(A)             ; command serial bus device to LISTEN
+romsub $FFB4 = TALK(ubyte device @ A) clobbers(A)               ; command serial bus device to TALK
+romsub $FFB7 = READST() -> ubyte @ A                            ; read I/O status word
+romsub $FFBA = SETLFS(ubyte logical @ A, ubyte device @ X, ubyte address @ Y)   ; set logical file parameters
+romsub $FFBD = SETNAM(ubyte namelen @ A, str filename @ XY)     ; set filename parameters
+romsub $FFC0 = OPEN() clobbers(X,Y) -> ubyte @Pc, ubyte @A      ; (via 794 ($31A)) open a logical file
+romsub $FFC3 = CLOSE(ubyte logical @ A) clobbers(A,X,Y)         ; (via 796 ($31C)) close a logical file
+romsub $FFC6 = CHKIN(ubyte logical @ X) clobbers(A,X) -> ubyte @Pc    ; (via 798 ($31E)) define an input channel
+romsub $FFC9 = CHKOUT(ubyte logical @ X) clobbers(A,X)          ; (via 800 ($320)) define an output channel
+romsub $FFCC = CLRCHN() clobbers(A,X)                           ; (via 802 ($322)) restore default devices
+romsub $FFCF = CHRIN() clobbers(X, Y) -> ubyte @ A   ; (via 804 ($324)) input a character (for keyboard, read a whole line from the screen) A=byte read.
+romsub $FFD2 = CHROUT(ubyte char @ A)                           ; (via 806 ($326)) output a character
+romsub $FFD5 = LOAD(ubyte verify @ A, uword address @ XY) -> ubyte @Pc, ubyte @ A, ubyte @ X, ubyte @ Y     ; (via 816 ($330)) load from device
+romsub $FFD8 = SAVE(ubyte zp_startaddr @ A, uword endaddr @ XY) -> ubyte @ Pc, ubyte @ A                    ; (via 818 ($332)) save to a device
+romsub $FFDB = SETTIM(ubyte low @ A, ubyte middle @ X, ubyte high @ Y)      ; set the software clock
+romsub $FFDE = RDTIM() -> ubyte @ A, ubyte @ X, ubyte @ Y       ; read the software clock
+romsub $FFE1 = STOP() clobbers(X) -> ubyte @ Pz, ubyte @ A      ; (via 808 ($328)) check the STOP key (and some others in A)
+romsub $FFE4 = GETIN() clobbers(X,Y) -> ubyte @Pc, ubyte @ A    ; (via 810 ($32A)) get a character
+romsub $FFE7 = CLALL() clobbers(A,X)                            ; (via 812 ($32C)) close all files
+romsub $FFEA = UDTIM() clobbers(A,X)                            ; update the software clock
+romsub $FFED = SCREEN() -> ubyte @ X, ubyte @ Y                 ; read number of screen rows and columns
+romsub $FFF0 = PLOT(ubyte col @ Y, ubyte row @ X, ubyte dir @ Pc) -> ubyte @ X, ubyte @ Y       ; read/set position of cursor on screen.  Use txt.plot for a 'safe' wrapper that preserves X.
+romsub $FFF3 = IOBASE() -> uword @ XY                           ; read base address of I/O devices
+
+}
+
+cx16 {
+
+; 65c02 hardware vectors:
+    &uword  NMI_VEC         = $FFFA     ; 6502 nmi vector, determined by the kernal if banked in
+    &uword  RESET_VEC       = $FFFC     ; 6502 reset vector, determined by the kernal if banked in
+    &uword  IRQ_VEC         = $FFFE     ; 6502 interrupt vector, determined by the kernal if banked in
+
+
+; the sixteen virtual 16-bit registers
+    &uword r0  = $0002
+    &uword r1  = $0004
+    &uword r2  = $0006
+    &uword r3  = $0008
+    &uword r4  = $000a
+    &uword r5  = $000c
+    &uword r6  = $000e
+    &uword r7  = $0010
+    &uword r8  = $0012
+    &uword r9  = $0014
+    &uword r10 = $0016
+    &uword r11 = $0018
+    &uword r12 = $001a
+    &uword r13 = $001c
+    &uword r14 = $001e
+    &uword r15 = $0020
+
+; VERA registers
+
+    const uword VERA_BASE = $9F20
+    &ubyte  VERA_ADDR_L   	  = VERA_BASE + $0000
+    &ubyte  VERA_ADDR_M   	  = VERA_BASE + $0001
+    &ubyte  VERA_ADDR_H   	  = VERA_BASE + $0002
+    &ubyte  VERA_DATA0        = VERA_BASE + $0003
+    &ubyte  VERA_DATA1        = VERA_BASE + $0004
+    &ubyte  VERA_CTRL         = VERA_BASE + $0005
+    &ubyte  VERA_IEN          = VERA_BASE + $0006
+    &ubyte  VERA_ISR          = VERA_BASE + $0007
+    &ubyte  VERA_IRQ_LINE_L   = VERA_BASE + $0008
+    &ubyte  VERA_DC_VIDEO     = VERA_BASE + $0009
+    &ubyte  VERA_DC_HSCALE    = VERA_BASE + $000A
+    &ubyte  VERA_DC_VSCALE    = VERA_BASE + $000B
+    &ubyte  VERA_DC_BORDER    = VERA_BASE + $000C
+    &ubyte  VERA_DC_HSTART    = VERA_BASE + $0009
+    &ubyte  VERA_DC_HSTOP     = VERA_BASE + $000A
+    &ubyte  VERA_DC_VSTART    = VERA_BASE + $000B
+    &ubyte  VERA_DC_VSTOP     = VERA_BASE + $000C
+    &ubyte  VERA_L0_CONFIG    = VERA_BASE + $000D
+    &ubyte  VERA_L0_MAPBASE   = VERA_BASE + $000E
+    &ubyte  VERA_L0_TILEBASE  = VERA_BASE + $000F
+    &ubyte  VERA_L0_HSCROLL_L = VERA_BASE + $0010
+    &ubyte  VERA_L0_HSCROLL_H = VERA_BASE + $0011
+    &ubyte  VERA_L0_VSCROLL_L = VERA_BASE + $0012
+    &ubyte  VERA_L0_VSCROLL_H = VERA_BASE + $0013
+    &ubyte  VERA_L1_CONFIG    = VERA_BASE + $0014
+    &ubyte  VERA_L1_MAPBASE   = VERA_BASE + $0015
+    &ubyte  VERA_L1_TILEBASE  = VERA_BASE + $0016
+    &ubyte  VERA_L1_HSCROLL_L = VERA_BASE + $0017
+    &ubyte  VERA_L1_HSCROLL_H = VERA_BASE + $0018
+    &ubyte  VERA_L1_VSCROLL_L = VERA_BASE + $0019
+    &ubyte  VERA_L1_VSCROLL_H = VERA_BASE + $001A
+    &ubyte  VERA_AUDIO_CTRL   = VERA_BASE + $001B
+    &ubyte  VERA_AUDIO_RATE   = VERA_BASE + $001C
+    &ubyte  VERA_AUDIO_DATA   = VERA_BASE + $001D
+    &ubyte  VERA_SPI_DATA     = VERA_BASE + $001E
+    &ubyte  VERA_SPI_CTRL     = VERA_BASE + $001F
+; VERA_PSG_BASE     = $1F9C0
+; VERA_PALETTE_BASE = $1FA00
+; VERA_SPRITES_BASE = $1FC00
+
+; I/O
+
+    const uword  via1  = $9f60                  ;VIA 6522 #1
+    &ubyte  d1prb	= via1+0
+    &ubyte  d1pra	= via1+1
+    &ubyte  d1ddrb	= via1+2
+    &ubyte  d1ddra	= via1+3
+    &ubyte  d1t1l	= via1+4
+    &ubyte  d1t1h	= via1+5
+    &ubyte  d1t1ll	= via1+6
+    &ubyte  d1t1lh	= via1+7
+    &ubyte  d1t2l	= via1+8
+    &ubyte  d1t2h	= via1+9
+    &ubyte  d1sr	= via1+10
+    &ubyte  d1acr	= via1+11
+    &ubyte  d1pcr	= via1+12
+    &ubyte  d1ifr	= via1+13
+    &ubyte  d1ier	= via1+14
+    &ubyte  d1ora	= via1+15
+
+    const uword  via2  = $9f70                  ;VIA 6522 #2
+    &ubyte  d2prb	=via2+0
+    &ubyte  d2pra	=via2+1
+    &ubyte  d2ddrb	=via2+2
+    &ubyte  d2ddra	=via2+3
+    &ubyte  d2t1l	=via2+4
+    &ubyte  d2t1h	=via2+5
+    &ubyte  d2t1ll	=via2+6
+    &ubyte  d2t1lh	=via2+7
+    &ubyte  d2t2l	=via2+8
+    &ubyte  d2t2h	=via2+9
+    &ubyte  d2sr	=via2+10
+    &ubyte  d2acr	=via2+11
+    &ubyte  d2pcr	=via2+12
+    &ubyte  d2ifr	=via2+13
+    &ubyte  d2ier	=via2+14
+    &ubyte  d2ora	=via2+15
+
+
+; ---- Commander X-16 additions on top of C64 kernal routines ----
+; spelling of the names is taken from the Commander X-16 rom sources
+
+; supported C128 additions
+romsub $ff4a = close_all(ubyte device @A)  clobbers(A,X,Y)
+romsub $ff59 = lkupla(ubyte la @A)  clobbers(A,X,Y)
+romsub $ff5c = lkupsa(ubyte sa @Y)  clobbers(A,X,Y)
+romsub $ff5f = screen_set_mode(ubyte mode @A)  clobbers(A, X, Y) -> ubyte @Pc
+romsub $ff62 = screen_set_charset(ubyte charset @A, uword charsetptr @XY)  clobbers(A,X,Y)      ; incompatible with C128  dlchr()
+; not yet supported: romsub $ff65 = pfkey()  clobbers(A,X,Y)
+romsub $ff6e = jsrfar()
+romsub $ff74 = fetch(ubyte bank @X, ubyte index @Y)  clobbers(X) -> ubyte @A
+romsub $ff77 = stash(ubyte data @A, ubyte bank @X, ubyte index @Y)  clobbers(X)
+romsub $ff7a = cmpare(ubyte data @A, ubyte bank @X, ubyte index @Y)  clobbers(X)
+romsub $ff7d = primm()
+
+; X16 additions
+romsub $ff44 = macptr()  clobbers(A,X,Y)
+romsub $ff47 = enter_basic(ubyte cold_or_warm @Pc)  clobbers(A,X,Y)
+romsub $ff68 = mouse_config(ubyte shape @A, ubyte scale @X)  clobbers (A, X, Y)
+romsub $ff6b = mouse_get(ubyte zpdataptr @X)  clobbers(A)
+romsub $ff71 = mouse_scan()  clobbers(A, X, Y)
+romsub $ff53 = joystick_scan()  clobbers(A, X, Y)
+romsub $ff56 = joystick_get(ubyte joynr @A) -> ubyte @A, ubyte @X, ubyte @Y
+romsub $ff4d = clock_set_date_time()  clobbers(A, X, Y)      ; args: r0, r1, r2, r3L
+romsub $ff50 = clock_get_date_time()  clobbers(A, X, Y)      ; outout args: r0, r1, r2, r3L
+
+; TODO specify the correct clobbers for alle these functions below, we now assume all 3 regs are clobbered
+
+; high level graphics & fonts
+romsub $ff20 = GRAPH_init()  clobbers(A,X,Y)           ; uses vectors=r0
+romsub $ff23 = GRAPH_clear()  clobbers(A,X,Y)
+romsub $ff26 = GRAPH_set_window()  clobbers(A,X,Y)      ; uses x=r0, y=r1, width=r2, height=r3
+romsub $ff29 = GRAPH_set_colors(ubyte stroke @A, ubyte fill @X, ubyte background @Y)  clobbers (A,X,Y)
+romsub $ff2c = GRAPH_draw_line()  clobbers(A,X,Y)       ; uses x1=r0, y1=r1, x2=r2, y2=r3
+romsub $ff2f = GRAPH_draw_rect(ubyte fill @Pc)  clobbers(A,X,Y)   ; uses x=r0, y=r1, width=r2, height=r3, cornerradius=r4
+romsub $ff32 = GRAPH_move_rect()  clobbers(A,X,Y)       ; uses sx=r0, sy=r1, tx=r2, ty=r3, width=r4, height=r5
+romsub $ff35 = GRAPH_draw_oval(ubyte fill @Pc)  clobbers(A,X,Y)       ; uses x=r0, y=r1, width=r2, height=r3
+romsub $ff38 = GRAPH_draw_image()  clobbers(A,X,Y)      ; uses x=r0, y=r1, ptr=r2, width=r3, height=r4
+romsub $ff3b = GRAPH_set_font()  clobbers(A,X,Y)         ; uses ptr=r0
+romsub $ff3e = GRAPH_get_char_size(ubyte baseline @A, ubyte width @X, ubyte height_or_style @Y, ubyte is_control @Pc)  clobbers(A,X,Y)
+romsub $ff41 = GRAPH_put_char(ubyte char @A)  clobbers(A,X,Y)   ; uses x=r0, y=r1
+
+; framebuffer
+romsub $fef6 = FB_init()  clobbers(A,X,Y)
+romsub $fef9 = FB_get_info()  clobbers(X,Y) -> byte @A    ; also outputs width=r0, height=r1
+romsub $fefc = FB_set_palette(ubyte index @A, ubyte bytecount @X)  clobbers(A,X,Y)      ; also uses pointer=r0
+romsub $feff = FB_cursor_position()  clobbers(A,X,Y)    ; uses x=r0, y=r1
+romsub $ff02 = FB_cursor_next_line()  clobbers(A,X,Y)   ; uses x=r0
+romsub $ff05 = FB_get_pixel()  clobbers(X,Y) -> ubyte @A
+romsub $ff08 = FB_get_pixels()  clobbers(A,X,Y)         ; uses ptr=r0, count=r1
+romsub $ff0b = FB_set_pixel(ubyte color @A)  clobbers(A,X,Y)
+romsub $ff0e = FB_set_pixels()  clobbers(A,X,Y)         ; uses ptr=r0, count=r1
+romsub $ff11 = FB_set_8_pixels(ubyte pattern @A, ubyte color @X)  clobbers(A,X,Y)
+romsub $ff14 = FB_set_8_pixels_opaque(ubyte pattern @A, ubyte color1 @X, ubyte color2 @Y)  clobbers(A,X,Y)  ; also uses mask=r0L
+romsub $ff17 = FB_fill_pixels(ubyte color @A)  clobbers(A,X,Y)   ; also uses count=r0, step=r1
+romsub $ff1a = FB_filter_pixels()  clobbers(A,X,Y)      ; uses ptr=r0, count=r1
+romsub $ff1d = FB_move_pixels()  clobbers(A,X,Y)        ; uses sx=r0, sy=r1, tx=r2, ty=r3, count=r4
+
+; misc
+romsub $fef0 = sprite_set_image(ubyte number @A, ubyte width @X, ubyte height @Y, ubyte apply_mask @Pc)  clobbers(A,X,Y) -> ubyte @Pc  ; also uses pixels=r0, mask=r1, bpp=r2L
+romsub $fef3 = sprite_set_position(ubyte number @A)  clobbers(A,X,Y)  ; also uses x=r0 and y=r1
+romsub $fee4 = memory_fill(ubyte value @A)  clobbers(A,X,Y)      ; uses address=r0, num_bytes=r1
+romsub $fee7 = memory_copy()  clobbers(A,X,Y)                    ; uses source=r0, target=r1, num_bytes=r2
+romsub $feea = memory_crc()  clobbers(A,X,Y)                     ; uses address=r0, num_bytes=r1     result->r2
+romsub $feed = memory_decompress()  clobbers(A,X,Y)              ; uses input=r0, output=r1     result->r1
+romsub $fedb = console_init()  clobbers(A,X,Y)           ; uses x=r0, y=r1, width=r2, height=r3
+romsub $fede = console_put_char(ubyte char @A, ubyte wrapping @Pc)  clobbers(A,X,Y)
+romsub $fee1 = console_get_char()  clobbers(X,Y) -> ubyte @A
+romsub $fed8 = console_put_image()  clobbers(A,X,Y)      ; uses ptr=r0, width=r1, height=r2
+romsub $fed5 = console_set_paging_message()  clobbers(A,X,Y)     ; uses messageptr=r0
+romsub $fed2 = kbdbuf_put(ubyte key @A)  clobbers(A,X,Y)
+romsub $fecf = entropy_get() -> ubyte @A, ubyte @X, ubyte @Y
+romsub $fecc = monitor()  clobbers(A,X,Y)
+
+
+; ---- end of kernal routines ----
+
+; ---- utilities -----
+    asmsub vpeek(ubyte bank @A, uword address @XY) -> ubyte @A {
+        ; -- get a byte from VERA's video memory
+        ;    note: inefficient when reading multiple sequential bytes!
+        %asm {{
+                stz  cx16.VERA_CTRL
+                and  #1
+                sta  cx16.VERA_ADDR_H
+                sty  cx16.VERA_ADDR_M
+                stx  cx16.VERA_ADDR_L
+                lda  cx16.VERA_DATA0
+                rts
+            }}
+    }
+
+    sub vpoke(ubyte bank, uword address, ubyte value) {
+        ; -- write a single byte to VERA's video memory
+        ;    note: inefficient when writing multiple sequential bytes!
+        %asm {{
+            stz  cx16.VERA_CTRL
+            lda  bank
+            and  #1
+            sta  cx16.VERA_ADDR_H
+            lda  address
+            sta  cx16.VERA_ADDR_L
+            lda  address+1
+            sta  cx16.VERA_ADDR_M
+            lda  value
+            sta  cx16.VERA_DATA0
+            rts
+        }}
+    }
+
+
+; ---- system stuff -----
+
+asmsub init_system()  {
+    ; Initializes the machine to a sane starting state.
+    ; Called automatically by the loader program logic.
+    %asm {{
+        sei
+        cld
+        ;stz  $00
+        ;stz  $01
+        ;stz  d1prb      ; select rom bank 0
+        lda  #$80
+        sta  VERA_CTRL
+        jsr  c64.IOINIT
+        jsr  c64.RESTOR
+        jsr  c64.CINT
+        lda  #$90       ; black
+        jsr  c64.CHROUT
+        lda  #1         ; swap fg/bg
+        jsr  c64.CHROUT
+        lda  #$9e       ; yellow
+        jsr  c64.CHROUT
+        lda  #147       ; clear screen
+        jsr  c64.CHROUT
+        lda  #0
+        tax
+        tay
+        clc
+        clv
+        cli
+        rts
+    }}
+}
+
+asmsub  reset_system()  {
+    ; Soft-reset the system back to Basic prompt.
+    %asm {{
+        sei
+        lda  #14
+        sta  $01
+        stz  cx16.d1prb         ; bank the kernal in
+        jmp  (cx16.RESET_VEC)
+    }}
+}
+
+}

compiler/res/prog8lib/cx16/textio.p8

@@ -0,0 +1,695 @@
+; Prog8 definitions for the Text I/O and Screen routines for the CommanderX16
+;
+; Written by Irmen de Jong (irmen@razorvine.net) - license: GNU GPL 3.0
+;
+; indent format: TABS, size=8
+
+%target cx16
+%import syslib
+%import conv
+
+
+txt {
+
+const ubyte DEFAULT_WIDTH = 80
+const ubyte DEFAULT_HEIGHT = 60
+
+
+sub  clear_screen() {
+    clear_screenchars(' ')
+}
+
+asmsub  fill_screen (ubyte char @ A, ubyte color @ Y) clobbers(A)  {
+	; ---- fill the character screen with the given fill character and character color.
+	%asm {{
+	    sty  _ly+1
+        phx
+        pha
+        jsr  c64.SCREEN             ; get dimensions in X/Y
+        txa
+        lsr  a
+        lsr  a
+        sta  _lx+1
+        stz  cx16.VERA_CTRL
+        lda  #%00010000
+        sta  cx16.VERA_ADDR_H       ; enable auto increment by 1, bank 0.
+        stz  cx16.VERA_ADDR_L       ; start at (0,0)
+        stz  cx16.VERA_ADDR_M
+        pla
+_lx     ldx  #0                     ; modified
+        phy
+_ly     ldy  #1                     ; modified
+-       sta  cx16.VERA_DATA0
+        sty  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA0
+        sty  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA0
+        sty  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA0
+        sty  cx16.VERA_DATA0
+        dex
+        bne  -
+        ply
+        dey
+        beq  +
+        stz  cx16.VERA_ADDR_L
+        inc  cx16.VERA_ADDR_M       ; next line
+        bra  _lx
++       plx
+        rts
+        }}
+}
+
+asmsub  clear_screenchars (ubyte char @ A) clobbers(Y)  {
+	; ---- clear the character screen with the given fill character (leaves colors)
+	;      (assumes screen matrix is at the default address)
+	%asm {{
+        phx
+        pha
+        jsr  c64.SCREEN             ; get dimensions in X/Y
+        txa
+        lsr  a
+        lsr  a
+        sta  _lx+1
+        stz  cx16.VERA_CTRL
+        lda  #%00100000
+        sta  cx16.VERA_ADDR_H       ; enable auto increment by 2, bank 0.
+        stz  cx16.VERA_ADDR_L       ; start at (0,0)
+        stz  cx16.VERA_ADDR_M
+        pla
+_lx     ldx  #0                     ; modified
+-       sta  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA0
+        dex
+        bne  -
+        dey
+        beq  +
+        stz  cx16.VERA_ADDR_L
+        inc  cx16.VERA_ADDR_M       ; next line
+        bra  _lx
++       plx
+        rts
+        }}
+}
+
+asmsub  clear_screencolors (ubyte color @ A) clobbers(Y)  {
+	; ---- clear the character screen colors with the given color (leaves characters).
+	;      (assumes color matrix is at the default address)
+	%asm {{
+        phx
+        sta  _la+1
+        jsr  c64.SCREEN             ; get dimensions in X/Y
+        txa
+        lsr  a
+        lsr  a
+        sta  _lx+1
+        stz  cx16.VERA_CTRL
+        lda  #%00100000
+        sta  cx16.VERA_ADDR_H       ; enable auto increment by 2, bank 0.
+        lda  #1
+        sta  cx16.VERA_ADDR_L       ; start at (1,0)
+        stz  cx16.VERA_ADDR_M
+_lx     ldx  #0                     ; modified
+_la     lda  #0                     ; modified
+-       sta  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA0
+        dex
+        bne  -
+        dey
+        beq  +
+        lda  #1
+        sta  cx16.VERA_ADDR_L
+        inc  cx16.VERA_ADDR_M       ; next line
+        bra  _lx
++       plx
+        rts
+        }}
+}
+
+
+ubyte[16] color_to_charcode = [$90,$05,$1c,$9f,$9c,$1e,$1f,$9e,$81,$95,$96,$97,$98,$99,$9a,$9b]
+
+sub color (ubyte txtcol) {
+    txtcol &= 15
+    c64.CHROUT(color_to_charcode[txtcol])
+}
+
+sub color2 (ubyte txtcol, ubyte bgcol) {
+    txtcol &= 15
+    bgcol &= 15
+    c64.CHROUT(color_to_charcode[bgcol])
+    c64.CHROUT(1)       ; switch fg and bg colors
+    c64.CHROUT(color_to_charcode[txtcol])
+}
+
+sub lowercase() {
+    cx16.screen_set_charset(3, 0)  ; lowercase charset
+}
+
+sub uppercase() {
+    cx16.screen_set_charset(2, 0)  ; uppercase charset
+}
+
+asmsub  scroll_left() clobbers(A, Y)  {
+	; ---- scroll the whole screen 1 character to the left
+	;      contents of the rightmost column are unchanged, you should clear/refill this yourself
+	%asm {{
+	    phx
+	    jsr  c64.SCREEN
+	    dex
+	    stx  _lx+1
+        dey
+        sty  P8ZP_SCRATCH_B1    ; number of rows to scroll
+
+_nextline
+        stz  cx16.VERA_CTRL     ; data port 0: source column
+        lda  #%00010000         ; auto increment 1
+        sta  cx16.VERA_ADDR_H
+        lda  #2
+        sta  cx16.VERA_ADDR_L   ; begin in column 1
+        ldy  P8ZP_SCRATCH_B1
+        sty  cx16.VERA_ADDR_M
+        lda  #1
+        sta  cx16.VERA_CTRL     ; data port 1: destination column
+        lda  #%00010000         ; auto increment 1
+        sta  cx16.VERA_ADDR_H
+        stz  cx16.VERA_ADDR_L
+        sty  cx16.VERA_ADDR_M
+
+_lx     ldx  #0                ; modified
+-       lda  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA1    ; copy char
+        lda  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA1    ; copy color
+        dex
+        bne  -
+        dec  P8ZP_SCRATCH_B1
+        bpl  _nextline
+
+        lda  #0
+        sta  cx16.VERA_CTRL
+	    plx
+	    rts
+	}}
+}
+
+asmsub  scroll_right() clobbers(A)  {
+	; ---- scroll the whole screen 1 character to the right
+	;      contents of the leftmost column are unchanged, you should clear/refill this yourself
+	%asm {{
+	    phx
+	    jsr  c64.SCREEN
+	    dex
+	    stx  _lx+1
+	    txa
+	    asl  a
+	    dea
+	    sta  _rcol+1
+	    ina
+	    ina
+	    sta  _rcol2+1
+        dey
+        sty  P8ZP_SCRATCH_B1    ; number of rows to scroll
+
+_nextline
+        stz  cx16.VERA_CTRL     ; data port 0: source column
+        lda  #%00011000         ; auto decrement 1
+        sta  cx16.VERA_ADDR_H
+_rcol   lda  #79*2-1             ; modified
+        sta  cx16.VERA_ADDR_L   ; begin in rightmost column minus one
+        ldy  P8ZP_SCRATCH_B1
+        sty  cx16.VERA_ADDR_M
+        lda  #1
+        sta  cx16.VERA_CTRL     ; data port 1: destination column
+        lda  #%00011000         ; auto decrement 1
+        sta  cx16.VERA_ADDR_H
+_rcol2  lda  #79*2+1            ; modified
+        sta  cx16.VERA_ADDR_L
+        sty  cx16.VERA_ADDR_M
+
+_lx     ldx  #0                 ; modified
+-       lda  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA1    ; copy char
+        lda  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA1    ; copy color
+        dex
+        bne  -
+        dec  P8ZP_SCRATCH_B1
+        bpl  _nextline
+
+        lda  #0
+        sta  cx16.VERA_CTRL
+	    plx
+	    rts
+	}}
+}
+
+asmsub  scroll_up() clobbers(A, Y)  {
+	; ---- scroll the whole screen 1 character up
+	;      contents of the bottom row are unchanged, you should refill/clear this yourself
+	%asm {{
+	    phx
+	    jsr  c64.SCREEN
+	    stx  _nextline+1
+	    dey
+        sty  P8ZP_SCRATCH_B1
+        stz  cx16.VERA_CTRL         ; data port 0 is source
+        lda  #1
+        sta  cx16.VERA_ADDR_M       ; start at second line
+        stz  cx16.VERA_ADDR_L
+        lda  #%00010000
+        sta  cx16.VERA_ADDR_H       ; enable auto increment by 1, bank 0.
+
+        lda  #1
+        sta  cx16.VERA_CTRL         ; data port 1 is destination
+        stz  cx16.VERA_ADDR_M       ; start at top line
+        stz  cx16.VERA_ADDR_L
+        lda  #%00010000
+        sta  cx16.VERA_ADDR_H       ; enable auto increment by 1, bank 0.
+
+_nextline
+        ldx  #80        ; modified
+-       lda  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA1        ; copy char
+        lda  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA1        ; copy color
+        dex
+        bne  -
+        dec  P8ZP_SCRATCH_B1
+        beq  +
+        stz  cx16.VERA_CTRL         ; data port 0
+        stz  cx16.VERA_ADDR_L
+        inc  cx16.VERA_ADDR_M
+        lda  #1
+        sta  cx16.VERA_CTRL         ; data port 1
+        stz  cx16.VERA_ADDR_L
+        inc  cx16.VERA_ADDR_M
+        bra  _nextline
+
++       lda  #0
+        sta  cx16.VERA_CTRL
+	    plx
+	    rts
+	}}
+}
+
+asmsub  scroll_down() clobbers(A, Y)  {
+	; ---- scroll the whole screen 1 character down
+	;      contents of the top row are unchanged, you should refill/clear this yourself
+	%asm {{
+	    phx
+	    jsr  c64.SCREEN
+	    stx  _nextline+1
+	    dey
+        sty  P8ZP_SCRATCH_B1
+        stz  cx16.VERA_CTRL         ; data port 0 is source
+        dey
+        sty  cx16.VERA_ADDR_M       ; start at line before bottom line
+        stz  cx16.VERA_ADDR_L
+        lda  #%00010000
+        sta  cx16.VERA_ADDR_H       ; enable auto increment by 1, bank 0.
+
+        lda  #1
+        sta  cx16.VERA_CTRL         ; data port 1 is destination
+        iny
+        sty  cx16.VERA_ADDR_M       ; start at bottom line
+        stz  cx16.VERA_ADDR_L
+        lda  #%00010000
+        sta  cx16.VERA_ADDR_H       ; enable auto increment by 1, bank 0.
+
+_nextline
+        ldx  #80        ; modified
+-       lda  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA1        ; copy char
+        lda  cx16.VERA_DATA0
+        sta  cx16.VERA_DATA1        ; copy color
+        dex
+        bne  -
+        dec  P8ZP_SCRATCH_B1
+        beq  +
+        stz  cx16.VERA_CTRL         ; data port 0
+        stz  cx16.VERA_ADDR_L
+        dec  cx16.VERA_ADDR_M
+        lda  #1
+        sta  cx16.VERA_CTRL         ; data port 1
+        stz  cx16.VERA_ADDR_L
+        dec  cx16.VERA_ADDR_M
+        bra  _nextline
+
++       lda  #0
+        sta  cx16.VERA_CTRL
+	    plx
+	    rts
+	}}
+}
+
+romsub $FFD2 = chrout(ubyte char @ A)    ; for consistency. You can also use c64.CHROUT directly ofcourse.
+
+asmsub  print (str text @ AY) clobbers(A,Y)  {
+	; ---- print null terminated string from A/Y
+	; note: the compiler contains an optimization that will replace
+	;       a call to this subroutine with a string argument of just one char,
+	;       by just one call to c64.CHROUT of that single char.
+	%asm {{
+		sta  P8ZP_SCRATCH_B1
+		sty  P8ZP_SCRATCH_REG
+		ldy  #0
+-		lda  (P8ZP_SCRATCH_B1),y
+		beq  +
+		jsr  c64.CHROUT
+		iny
+		bne  -
++		rts
+	}}
+}
+
+asmsub  print_ub0  (ubyte value @ A) clobbers(A,Y)  {
+	; ---- print the ubyte in A in decimal form, with left padding 0s (3 positions total)
+	%asm {{
+		phx
+		jsr  conv.ubyte2decimal
+		pha
+		tya
+		jsr  c64.CHROUT
+		pla
+		jsr  c64.CHROUT
+		txa
+		jsr  c64.CHROUT
+		plx
+		rts
+	}}
+}
+
+asmsub  print_ub  (ubyte value @ A) clobbers(A,Y)  {
+	; ---- print the ubyte in A in decimal form, without left padding 0s
+	%asm {{
+		phx
+		jsr  conv.ubyte2decimal
+_print_byte_digits
+		pha
+		cpy  #'0'
+		beq  +
+		tya
+		jsr  c64.CHROUT
+		pla
+		jsr  c64.CHROUT
+		jmp  _ones
++       pla
+        cmp  #'0'
+        beq  _ones
+        jsr  c64.CHROUT
+_ones   txa
+		jsr  c64.CHROUT
+		plx
+		rts
+	}}
+}
+
+asmsub  print_b  (byte value @ A) clobbers(A,Y)  {
+	; ---- print the byte in A in decimal form, without left padding 0s
+	%asm {{
+		phx
+		pha
+		cmp  #0
+		bpl  +
+		lda  #'-'
+		jsr  c64.CHROUT
++		pla
+		jsr  conv.byte2decimal
+		jmp  print_ub._print_byte_digits
+	}}
+}
+
+asmsub  print_ubhex  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
+	; ---- print the ubyte in A in hex form (if Carry is set, a radix prefix '$' is printed as well)
+	%asm {{
+		phx
+		bcc  +
+		pha
+		lda  #'$'
+		jsr  c64.CHROUT
+		pla
++		jsr  conv.ubyte2hex
+		jsr  c64.CHROUT
+		tya
+		jsr  c64.CHROUT
+		plx
+		rts
+	}}
+}
+
+asmsub  print_ubbin  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
+	; ---- print the ubyte in A in binary form (if Carry is set, a radix prefix '%' is printed as well)
+	%asm {{
+		phx
+		sta  P8ZP_SCRATCH_B1
+		bcc  +
+		lda  #'%'
+		jsr  c64.CHROUT
++		ldy  #8
+-		lda  #'0'
+		asl  P8ZP_SCRATCH_B1
+		bcc  +
+		lda  #'1'
++		jsr  c64.CHROUT
+		dey
+		bne  -
+		plx
+		rts
+	}}
+}
+
+asmsub  print_uwbin  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
+	; ---- print the uword in A/Y in binary form (if Carry is set, a radix prefix '%' is printed as well)
+	%asm {{
+		pha
+		tya
+		jsr  print_ubbin
+		pla
+		clc
+		jmp  print_ubbin
+	}}
+}
+
+asmsub  print_uwhex  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
+	; ---- print the uword in A/Y in hexadecimal form (4 digits)
+	;      (if Carry is set, a radix prefix '$' is printed as well)
+	%asm {{
+		pha
+		tya
+		jsr  print_ubhex
+		pla
+		clc
+		jmp  print_ubhex
+	}}
+}
+
+asmsub  print_uw0  (uword value @ AY) clobbers(A,Y)  {
+	; ---- print the uword in A/Y in decimal form, with left padding 0s (5 positions total)
+	%asm {{
+	    phx
+		jsr  conv.uword2decimal
+		ldy  #0
+-		lda  conv.uword2decimal.decTenThousands,y
+        beq  +
+		jsr  c64.CHROUT
+		iny
+		bne  -
++		plx
+		rts
+	}}
+}
+
+asmsub  print_uw  (uword value @ AY) clobbers(A,Y)  {
+	; ---- print the uword in A/Y in decimal form, without left padding 0s
+	%asm {{
+	    phx
+		jsr  conv.uword2decimal
+		plx
+		ldy  #0
+-		lda  conv.uword2decimal.decTenThousands,y
+		beq  _allzero
+		cmp  #'0'
+		bne  _gotdigit
+		iny
+		bne  -
+
+_gotdigit
+		jsr  c64.CHROUT
+		iny
+		lda  conv.uword2decimal.decTenThousands,y
+		bne  _gotdigit
+		rts
+_allzero
+        lda  #'0'
+        jmp  c64.CHROUT
+	}}
+}
+
+asmsub  print_w  (word value @ AY) clobbers(A,Y)  {
+	; ---- print the (signed) word in A/Y in decimal form, without left padding 0's
+	%asm {{
+		cpy  #0
+		bpl  +
+		pha
+		lda  #'-'
+		jsr  c64.CHROUT
+		tya
+		eor  #255
+		tay
+		pla
+		eor  #255
+		clc
+		adc  #1
+		bcc  +
+		iny
++		jmp  print_uw
+	}}
+}
+
+asmsub  input_chars  (uword buffer @ AY) clobbers(A) -> ubyte @ Y  {
+	; ---- Input a string (max. 80 chars) from the keyboard. Returns length in Y. (string is terminated with a 0 byte as well)
+	;      It assumes the keyboard is selected as I/O channel!
+
+	%asm {{
+		sta  P8ZP_SCRATCH_W1
+		sty  P8ZP_SCRATCH_W1+1
+		ldy  #0				; char counter = 0
+-		jsr  c64.CHRIN
+		cmp  #$0d			; return (ascii 13) pressed?
+		beq  +				; yes, end.
+		sta  (P8ZP_SCRATCH_W1),y	; else store char in buffer
+		iny
+		bne  -
++		lda  #0
+		sta  (P8ZP_SCRATCH_W1),y	; finish string with 0 byte
+		rts
+
+	}}
+}
+
+asmsub  setchr  (ubyte col @X, ubyte row @Y, ubyte character @A) clobbers(A)  {
+	; ---- sets the character in the screen matrix at the given position
+	%asm {{
+		pha
+		txa
+		asl  a
+        stz  cx16.VERA_CTRL
+		stz  cx16.VERA_ADDR_H
+		sta  cx16.VERA_ADDR_L
+		sty  cx16.VERA_ADDR_M
+		pla
+        sta  cx16.VERA_DATA0
+		rts
+	}}
+}
+
+asmsub  getchr  (ubyte col @A, ubyte row @Y) -> ubyte @ A {
+	; ---- get the character in the screen matrix at the given location
+	%asm  {{
+		asl  a
+        stz  cx16.VERA_CTRL
+		stz  cx16.VERA_ADDR_H
+		sta  cx16.VERA_ADDR_L
+		sty  cx16.VERA_ADDR_M
+        lda  cx16.VERA_DATA0
+		rts
+	}}
+}
+
+asmsub  setclr  (ubyte col @X, ubyte row @Y, ubyte color @A) clobbers(A)  {
+	; ---- set the color in A on the screen matrix at the given position
+	%asm {{
+		pha
+		txa
+		asl  a
+		ina
+        stz  cx16.VERA_CTRL
+		stz  cx16.VERA_ADDR_H
+		sta  cx16.VERA_ADDR_L
+		sty  cx16.VERA_ADDR_M
+		pla
+        sta  cx16.VERA_DATA0
+		rts
+	}}
+}
+
+asmsub  getclr  (ubyte col @A, ubyte row @Y) -> ubyte @ A {
+	; ---- get the color in the screen color matrix at the given location
+	%asm  {{
+		asl  a
+		ina
+        stz  cx16.VERA_CTRL
+		stz  cx16.VERA_ADDR_H
+		sta  cx16.VERA_ADDR_L
+		sty  cx16.VERA_ADDR_M
+        lda  cx16.VERA_DATA0
+		rts
+	}}
+}
+
+sub  setcc  (ubyte column, ubyte row, ubyte char, ubyte charcolor)  {
+	; ---- set char+color at the given position on the screen
+	%asm {{
+	    phx
+		lda  column
+		asl  a
+		tax
+		ldy  row
+		lda  charcolor
+		and  #$0f
+	    sta  P8ZP_SCRATCH_B1
+        stz  cx16.VERA_CTRL
+        stz  cx16.VERA_ADDR_H
+        stx  cx16.VERA_ADDR_L
+        sty  cx16.VERA_ADDR_M
+        lda  char
+        sta  cx16.VERA_DATA0
+        inx
+        stz  cx16.VERA_ADDR_H
+        stx  cx16.VERA_ADDR_L
+        sty  cx16.VERA_ADDR_M
+        lda  cx16.VERA_DATA0
+        and  #$f0
+        ora  P8ZP_SCRATCH_B1
+        sta  cx16.VERA_DATA0
+		plx
+		rts
+    }}
+}
+
+asmsub  plot  (ubyte col @ Y, ubyte row @ A) clobbers(A) {
+	; ---- safe wrapper around PLOT kernel routine, to save the X register.
+	%asm  {{
+		phx
+		tax
+		clc
+		jsr  c64.PLOT
+		plx
+		rts
+	}}
+}
+
+asmsub width() clobbers(X,Y) -> ubyte @A {
+    ; -- returns the text screen width (number of columns)
+    %asm {{
+        jsr  c64.SCREEN
+        txa
+        rts
+    }}
+}
+
+asmsub height() clobbers(X, Y) -> ubyte @A {
+    ; -- returns the text screen height (number of rows)
+    %asm {{
+        jsr  c64.SCREEN
+        tya
+        rts
+    }}
+}
+
+}

compiler/res/prog8lib/cx16lib.p8

@@ -1,201 +0,0 @@
-; Prog8 definitions for the CommanderX16
-; Including memory registers, I/O registers, Basic and Kernal subroutines.
-;
-; Written by Irmen de Jong (irmen@razorvine.net) - license: GNU GPL 3.0
-;
-; indent format: TABS, size=8
-
-
-c64 {
-
-; ---- kernal routines, these are the same as on the Commodore-64 (hence the same block name) ----
-
-; STROUT --> use screen.print
-; CLEARSCR -> use screen.clear_screen
-; HOMECRSR -> use screen.plot
-
-romsub $FF81 = CINT() clobbers(A,X,Y)                           ; (alias: SCINIT) initialize screen editor and video chip
-romsub $FF84 = IOINIT() clobbers(A, X)                          ; initialize I/O devices (CIA, SID, IRQ)
-romsub $FF87 = RAMTAS() clobbers(A,X,Y)                         ; initialize RAM, tape buffer, screen
-romsub $FF8A = RESTOR() clobbers(A,X,Y)                         ; restore default I/O vectors
-romsub $FF8D = VECTOR(uword userptr @ XY, ubyte dir @ Pc) clobbers(A,Y)     ; read/set I/O vector table
-romsub $FF90 = SETMSG(ubyte value @ A)                          ; set Kernal message control flag
-romsub $FF93 = SECOND(ubyte address @ A) clobbers(A)            ; (alias: LSTNSA) send secondary address after LISTEN
-romsub $FF96 = TKSA(ubyte address @ A) clobbers(A)              ; (alias: TALKSA) send secondary address after TALK
-romsub $FF99 = MEMTOP(uword address @ XY, ubyte dir @ Pc) -> uword @ XY     ; read/set top of memory  pointer
-romsub $FF9C = MEMBOT(uword address @ XY, ubyte dir @ Pc) -> uword @ XY     ; read/set bottom of memory  pointer
-romsub $FF9F = SCNKEY() clobbers(A,X,Y)                         ; scan the keyboard
-romsub $FFA2 = SETTMO(ubyte timeout @ A)                        ; set time-out flag for IEEE bus
-romsub $FFA5 = ACPTR() -> ubyte @ A                             ; (alias: IECIN) input byte from serial bus
-romsub $FFA8 = CIOUT(ubyte databyte @ A)                        ; (alias: IECOUT) output byte to serial bus
-romsub $FFAB = UNTLK() clobbers(A)                              ; command serial bus device to UNTALK
-romsub $FFAE = UNLSN() clobbers(A)                              ; command serial bus device to UNLISTEN
-romsub $FFB1 = LISTEN(ubyte device @ A) clobbers(A)             ; command serial bus device to LISTEN
-romsub $FFB4 = TALK(ubyte device @ A) clobbers(A)               ; command serial bus device to TALK
-romsub $FFB7 = READST() -> ubyte @ A                            ; read I/O status word
-romsub $FFBA = SETLFS(ubyte logical @ A, ubyte device @ X, ubyte address @ Y)   ; set logical file parameters
-romsub $FFBD = SETNAM(ubyte namelen @ A, str filename @ XY)     ; set filename parameters
-romsub $FFC0 = OPEN() clobbers(A,X,Y)                           ; (via 794 ($31A)) open a logical file
-romsub $FFC3 = CLOSE(ubyte logical @ A) clobbers(A,X,Y)         ; (via 796 ($31C)) close a logical file
-romsub $FFC6 = CHKIN(ubyte logical @ X) clobbers(A,X)           ; (via 798 ($31E)) define an input channel
-romsub $FFC9 = CHKOUT(ubyte logical @ X) clobbers(A,X)          ; (via 800 ($320)) define an output channel
-romsub $FFCC = CLRCHN() clobbers(A,X)                           ; (via 802 ($322)) restore default devices
-romsub $FFCF = CHRIN() clobbers(Y) -> ubyte @ A                 ; (via 804 ($324)) input a character (for keyboard, read a whole line from the screen) A=byte read.
-romsub $FFD2 = CHROUT(ubyte char @ A)                           ; (via 806 ($326)) output a character
-romsub $FFD5 = LOAD(ubyte verify @ A, uword address @ XY) -> ubyte @Pc, ubyte @ A, ubyte @ X, ubyte @ Y     ; (via 816 ($330)) load from device
-romsub $FFD8 = SAVE(ubyte zp_startaddr @ A, uword endaddr @ XY) -> ubyte @ Pc, ubyte @ A                    ; (via 818 ($332)) save to a device
-romsub $FFDB = SETTIM(ubyte low @ A, ubyte middle @ X, ubyte high @ Y)      ; set the software clock
-romsub $FFDE = RDTIM() -> ubyte @ A, ubyte @ X, ubyte @ Y       ; read the software clock
-romsub $FFE1 = STOP() clobbers(A,X) -> ubyte @ Pz, ubyte @ Pc   ; (via 808 ($328)) check the STOP key
-romsub $FFE4 = GETIN() clobbers(X,Y) -> ubyte @ A               ; (via 810 ($32A)) get a character
-romsub $FFE7 = CLALL() clobbers(A,X)                            ; (via 812 ($32C)) close all files
-romsub $FFEA = UDTIM() clobbers(A,X)                            ; update the software clock
-romsub $FFED = SCREEN() -> ubyte @ X, ubyte @ Y                 ; read number of screen rows and columns
-romsub $FFF0 = PLOT(ubyte col @ Y, ubyte row @ X, ubyte dir @ Pc) -> ubyte @ X, ubyte @ Y       ; read/set position of cursor on screen.  Use screen.plot for a 'safe' wrapper that preserves X.
-romsub $FFF3 = IOBASE() -> uword @ XY                           ; read base address of I/O devices
-
-}
-
-cx16 {
-
-; ---- Commander X-16 additions on top of C64 kernal routines ----
-; spelling of the names is taken from the Commander X-16 rom sources
-
-; the sixteen virtual 16-bit registers
-&ubyte r0  = $02
-&ubyte r0L = $02
-&ubyte r0H = $03
-&ubyte r1  = $04
-&ubyte r1L = $04
-&ubyte r1H = $05
-&ubyte r2  = $06
-&ubyte r2L = $06
-&ubyte r2H = $07
-&ubyte r3  = $08
-&ubyte r3L = $08
-&ubyte r3H = $09
-&ubyte r4  = $0a
-&ubyte r4L = $0a
-&ubyte r4H = $0b
-&ubyte r5  = $0c
-&ubyte r5L = $0c
-&ubyte r5H = $0d
-&ubyte r6  = $0e
-&ubyte r6L = $0e
-&ubyte r6H = $0f
-&ubyte r7  = $10
-&ubyte r7L = $10
-&ubyte r7H = $11
-&ubyte r8  = $12
-&ubyte r8L = $12
-&ubyte r8H = $13
-&ubyte r9  = $14
-&ubyte r9L = $14
-&ubyte r9H = $15
-&ubyte r10 = $16
-&ubyte r10L    = $16
-&ubyte r10H    = $17
-&ubyte r11     = $18
-&ubyte r11L    = $18
-&ubyte r11H    = $19
-&ubyte r12     = $1a
-&ubyte r12L    = $1a
-&ubyte r12H    = $1b
-&ubyte r13     = $1c
-&ubyte r13L    = $1c
-&ubyte r13H    = $1d
-&ubyte r14     = $1e
-&ubyte r14L    = $1e
-&ubyte r14H    = $1f
-&ubyte r15     = $20
-&ubyte r15L    = $20
-&ubyte r15H    = $21
-
-
-; TODO subroutine args + soubroutine returnvalues + clobber registers
-
-; supported C128 additions
-romsub $ff4a = close_all()
-romsub $ff59 = lkupla()
-romsub $ff5c = lkupsa()
-romsub $ff5f = screen_set_mode()
-romsub $ff62 = screen_set_charset(ubyte charset @A, uword charsetptr @XY) clobbers(A,X,Y)      ; incompatible with C128  dlchr()
-romsub $ff65 = pfkey()
-romsub $ff6e = jsrfar()
-romsub $ff74 = fetch()
-romsub $ff77 = stash()
-romsub $ff7a = cmpare()
-romsub $ff7d = primm()
-
-; X16 additions
-romsub $ff44 = macptr()
-romsub $ff47 = enter_basic()
-romsub $ff68 = mouse_config()
-romsub $ff6b = mouse_get()
-romsub $ff71 = mouse_scan()
-romsub $ff53 = joystick_scan()
-romsub $ff56 = joystick_get()
-romsub $ff4d = clock_set_date_time()
-romsub $ff50 = clock_get_date_time()
-
-; high level graphics & fonts
-romsub $ff20 = GRAPH_init()
-romsub $ff23 = GRAPH_clear()
-romsub $ff26 = GRAPH_set_window()
-romsub $ff29 = GRAPH_set_colors()
-romsub $ff2c = GRAPH_draw_line()
-romsub $ff2f = GRAPH_draw_rect()
-romsub $ff32 = GRAPH_move_rect()
-romsub $ff35 = GRAPH_draw_oval()
-romsub $ff38 = GRAPH_draw_image()
-romsub $ff3b = GRAPH_set_font()
-romsub $ff3e = GRAPH_get_char_size()
-romsub $ff41 = GRAPH_put_char()
-
-; TODO framebuffer API not yet included, include it
-
-romsub $fef0 = sprite_set_image()
-romsub $fef3 = sprite_set_position()
-romsub $fee4 = memory_fill()
-romsub $fee7 = memory_copy()
-romsub $feea = memory_crc()
-romsub $feed = memory_decompress()
-romsub $fedb = console_init()
-romsub $fede = console_put_char()
-romsub $fee1 = console_get_char()
-romsub $fed8 = console_put_image()
-romsub $fed5 = console_set_paging_message()
-romsub $fed2 = kbdbuf_put()
-romsub $fecf = entropy_get()
-romsub $fecc = monitor()
-
-
-
-; ---- end of kernal routines ----
-
-asmsub init_system()  {
-    ; Initializes the machine to a sane starting state.
-    ; Called automatically by the loader program logic.
-    %asm {{
-        sei
-        cld
-        lda  #0
-        sta  $00
-        sta  $01
-        jsr  c64.IOINIT
-        jsr  c64.RESTOR
-        jsr  c64.CINT
-        lda  #0
-        tax
-        tay
-        clc
-        clv
-        cli
-        lda  #66
-        clc
-        jsr  console_put_char
-        rts
-    }}
-}
-
-}

compiler/res/prog8lib/cx16logo.p8

@@ -0,0 +1,29 @@
+%import textio
+
+cx16logo {
+    sub logo_at(ubyte column, ubyte row) {
+        uword strptr
+        for strptr in logo_lines {
+            txt.plot(column, row)
+            txt.print(strptr)
+            row++
+        }
+    }
+
+    sub logo() {
+        uword strptr
+        for strptr in logo_lines
+            txt.print(strptr)
+        txt.chrout('\n')
+    }
+
+    str[] logo_lines = [
+            "\uf10d\uf11a\uf139\uf11b     \uf11a\uf13a\uf11b\n",
+            "\uf10b\uf11a▎\uf139\uf11b   \uf11a\uf13a\uf130\uf11b\n",
+            "\uf10f\uf11a▌ \uf139\uf11b \uf11a\uf13a \uf11b▌\n",
+            "\uf102 \uf132\uf11a▖\uf11b \uf11a▗\uf11b\uf132\n",
+            "\uf10e ▂\uf11a▘\uf11b \uf11a▝\uf11b▂\n",
+            "\uf104 \uf11a \uf11b\uf13a\uf11b \uf139\uf11a \uf11b\n",
+            "\uf101\uf130\uf13a   \uf139▎\uf100"
+        ]
+}

compiler/res/prog8lib/cx16textio.p8

@@ -1,240 +0,0 @@
-; Prog8 definitions for the Text I/O and Screen routines for the CommanderX16
-;
-; Written by Irmen de Jong (irmen@razorvine.net) - license: GNU GPL 3.0
-;
-; indent format: TABS, size=8
-
-
-%import cx16lib
-%import conv
-
-
-txt {
-
-asmsub  clear_screen (ubyte char @ A, ubyte color @ Y) clobbers(A)  {
-	; ---- clear the character screen with the given fill character and character color.
-	;      (assumes screen and color matrix are at their default addresses)
-
-	%asm {{
-	    brk     ; TODO
-        }}
-
-}
-
-asmsub  print (str text @ AY) clobbers(A,Y)  {
-	; ---- print null terminated string from A/Y
-	; note: the compiler contains an optimization that will replace
-	;       a call to this subroutine with a string argument of just one char,
-	;       by just one call to c64.CHROUT of that single char.
-	%asm {{
-		sta  P8ZP_SCRATCH_B1
-		sty  P8ZP_SCRATCH_REG
-		ldy  #0
--		lda  (P8ZP_SCRATCH_B1),y
-		beq  +
-		jsr  c64.CHROUT
-		iny
-		bne  -
-+		rts
-	}}
-}
-
-asmsub  print_ub0  (ubyte value @ A) clobbers(A,Y)  {
-	; ---- print the ubyte in A in decimal form, with left padding 0s (3 positions total)
-	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  conv.ubyte2decimal
-		pha
-		tya
-		jsr  c64.CHROUT
-		pla
-		jsr  c64.CHROUT
-		txa
-		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG_X
-		rts
-	}}
-}
-
-asmsub  print_ub  (ubyte value @ A) clobbers(A,Y)  {
-	; ---- print the ubyte in A in decimal form, without left padding 0s
-	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
-		jsr  conv.ubyte2decimal
-_print_byte_digits
-		pha
-		cpy  #'0'
-		beq  +
-		tya
-		jsr  c64.CHROUT
-		pla
-		jsr  c64.CHROUT
-		jmp  _ones
-+       pla
-        cmp  #'0'
-        beq  _ones
-        jsr  c64.CHROUT
-_ones   txa
-		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG_X
-		rts
-	}}
-}
-
-asmsub  print_b  (byte value @ A) clobbers(A,Y)  {
-	; ---- print the byte in A in decimal form, without left padding 0s
-	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
-		pha
-		cmp  #0
-		bpl  +
-		lda  #'-'
-		jsr  c64.CHROUT
-+		pla
-		jsr  conv.byte2decimal
-		jsr  print_ub._print_byte_digits
-		ldx  P8ZP_SCRATCH_REG_X
-		rts
-	}}
-}
-
-asmsub  print_ubhex  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
-	; ---- print the ubyte in A in hex form (if Carry is set, a radix prefix '$' is printed as well)
-	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
-		bcc  +
-		pha
-		lda  #'$'
-		jsr  c64.CHROUT
-		pla
-+		jsr  conv.ubyte2hex
-		jsr  c64.CHROUT
-		tya
-		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG_X
-		rts
-	}}
-}
-
-asmsub  print_ubbin  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
-	; ---- print the ubyte in A in binary form (if Carry is set, a radix prefix '%' is printed as well)
-	%asm {{
-		stx  P8ZP_SCRATCH_REG_X
-		sta  P8ZP_SCRATCH_B1
-		bcc  +
-		lda  #'%'
-		jsr  c64.CHROUT
-+		ldy  #8
--		lda  #'0'
-		asl  P8ZP_SCRATCH_B1
-		bcc  +
-		lda  #'1'
-+		jsr  c64.CHROUT
-		dey
-		bne  -
-		ldx  P8ZP_SCRATCH_REG_X
-		rts
-	}}
-}
-
-asmsub  print_uwbin  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
-	; ---- print the uword in A/Y in binary form (if Carry is set, a radix prefix '%' is printed as well)
-	%asm {{
-		pha
-		tya
-		jsr  print_ubbin
-		pla
-		clc
-		jmp  print_ubbin
-	}}
-}
-
-asmsub  print_uwhex  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
-	; ---- print the uword in A/Y in hexadecimal form (4 digits)
-	;      (if Carry is set, a radix prefix '$' is printed as well)
-	%asm {{
-		pha
-		tya
-		jsr  print_ubhex
-		pla
-		clc
-		jmp  print_ubhex
-	}}
-}
-
-asmsub  print_uw0  (uword value @ AY) clobbers(A,Y)  {
-	; ---- print the uword in A/Y in decimal form, with left padding 0s (5 positions total)
-	%asm {{
-	    stx  P8ZP_SCRATCH_REG_X
-		jsr  conv.uword2decimal
-		ldy  #0
--		lda  conv.uword2decimal.decTenThousands,y
-        beq  +
-		jsr  c64.CHROUT
-		iny
-		bne  -
-+		ldx  P8ZP_SCRATCH_REG_X
-		rts
-	}}
-}
-
-asmsub  print_uw  (uword value @ AY) clobbers(A,Y)  {
-	; ---- print the uword in A/Y in decimal form, without left padding 0s
-	%asm {{
-	    stx  P8ZP_SCRATCH_REG_X
-		jsr  conv.uword2decimal
-		ldx  P8ZP_SCRATCH_REG_X
-		ldy  #0
--		lda  conv.uword2decimal.decTenThousands,y
-		beq  _allzero
-		cmp  #'0'
-		bne  _gotdigit
-		iny
-		bne  -
-
-_gotdigit
-		jsr  c64.CHROUT
-		iny
-		lda  conv.uword2decimal.decTenThousands,y
-		bne  _gotdigit
-		rts
-_allzero
-        lda  #'0'
-        jmp  c64.CHROUT
-	}}
-}
-
-asmsub  print_w  (word value @ AY) clobbers(A,Y)  {
-	; ---- print the (signed) word in A/Y in decimal form, without left padding 0's
-	%asm {{
-		cpy  #0
-		bpl  +
-		pha
-		lda  #'-'
-		jsr  c64.CHROUT
-		tya
-		eor  #255
-		tay
-		pla
-		eor  #255
-		clc
-		adc  #1
-		bcc  +
-		iny
-+		jmp  print_uw
-	}}
-}
-
-asmsub  plot  (ubyte col @ Y, ubyte row @ A) clobbers(A) {
-	; ---- safe wrapper around PLOT kernel routine, to save the X register.
-	%asm  {{
-		stx  P8ZP_SCRATCH_REG_X
-		tax
-		clc
-		jsr  c64.PLOT
-		ldx  P8ZP_SCRATCH_REG_X
-		rts
-	}}
-}
-
-}

compiler/res/prog8lib/diskio.p8

@@ -0,0 +1,283 @@
+%import textio
+%import syslib
+
+; Note: this code is compatible with C64 and CX16.
+
+diskio {
+
+
+    sub directory(ubyte drivenumber) -> ubyte {
+        ; -- Shows the directory contents of disk drive 8-11 (provide as argument). Returns success flag.
+
+        c64.SETNAM(1, "$")
+        c64.SETLFS(1, drivenumber, 0)
+        void c64.OPEN()          ; open 1,8,0,"$"
+        if_cs
+            goto io_error
+        void c64.CHKIN(1)        ; use #1 as input channel
+        if_cs
+            goto io_error
+
+        repeat 4 {
+            void c64.CHRIN()     ; skip the 4 prologue bytes
+        }
+
+        ; while not key pressed / EOF encountered, read data.
+        ubyte status = c64.READST()
+        while not status {
+            ubyte low = c64.CHRIN()
+            ubyte high = c64.CHRIN()
+            txt.print_uw(mkword(high, low))
+            txt.chrout(' ')
+            ubyte @zp char
+            do {
+                char = c64.CHRIN()
+                txt.chrout(char)
+            } until char==0
+            txt.chrout('\n')
+            void c64.CHRIN()     ; skip 2 bytes
+            void c64.CHRIN()
+            status = c64.READST()
+            void c64.STOP()
+            if_nz
+                break
+        }
+
+io_error:
+        status = c64.READST()
+        c64.CLRCHN()        ; restore default i/o devices
+        c64.CLOSE(1)
+
+        if status and status != 64 {            ; 64=end of file
+            txt.print("\ni/o error, status: ")
+            txt.print_ub(status)
+            txt.chrout('\n')
+            return false
+        }
+
+        return true
+    }
+
+
+    ; internal variables for the iterative file lister
+    ubyte list_suffixmatch
+    ubyte list_in_progress = false
+    ubyte list_pattern_size
+    ubyte list_skip_disk_name
+    uword list_pattern
+    uword list_blocks
+    str   list_filename = "????????????????"
+
+    sub lf_start_list(ubyte drivenumber, uword pattern, ubyte suffixmatch) -> ubyte {
+        ; -- start an iterative file listing with optional prefix or suffix name matching.
+        lf_end_list()
+        list_pattern = pattern
+        list_suffixmatch = suffixmatch
+        list_skip_disk_name = true
+        list_in_progress = true
+        if pattern==0
+            list_pattern_size = 0
+        else
+            list_pattern_size = strlen(pattern)
+
+        c64.SETNAM(1, "$")
+        c64.SETLFS(1, drivenumber, 0)
+        void c64.OPEN()          ; open 1,8,0,"$"
+        if_cs
+            goto io_error
+        void c64.CHKIN(1)        ; use #1 as input channel
+        if_cs
+            goto io_error
+
+        repeat 4 {
+            void c64.CHRIN()     ; skip the 4 prologue bytes
+        }
+
+        if c64.READST()==0
+            return true
+
+io_error:
+        lf_end_list()
+        return false
+    }
+
+    sub lf_next_entry() -> ubyte {
+        ; -- retrieve the next entry from an iterative file listing session.
+        ;    results will be found in list_blocks and list_filename.
+        ;    if it returns false though, there are no more entries (or an error occurred).
+
+        if not list_in_progress
+            return false
+
+        while not c64.READST() {
+            uword nameptr = &list_filename
+            ubyte blocks_lsb = c64.CHRIN()
+            ubyte blocks_msb = c64.CHRIN()
+            list_blocks = mkword(blocks_msb, blocks_lsb)
+
+            ; read until the filename starts after the first "
+            while c64.CHRIN()!='\"'  {
+                if c64.READST()
+                    goto close_end
+            }
+
+            ; read the filename
+            repeat {
+                ubyte char = c64.CHRIN()
+                if_z
+                    break
+                if char=='\"'
+                    break
+                @(nameptr) = char
+                nameptr++
+            }
+
+            @(nameptr) = 0
+
+            while c64.CHRIN() {
+                ; read the rest of the entry until the end
+            }
+
+            void c64.CHRIN()     ; skip 2 bytes
+            void c64.CHRIN()
+
+            if not list_skip_disk_name {
+                if list_pattern_size {
+                    ; do filename matching
+                    if list_suffixmatch
+                        rightstr(list_filename, filename, list_pattern_size)
+                    else
+                        leftstr(list_filename, filename, list_pattern_size)
+                    if strcmp(filename, list_pattern)==0
+                        return true
+                } else
+                    return true
+            }
+            list_skip_disk_name = false
+        }
+
+close_end:
+        lf_end_list()
+        return false
+    }
+
+    sub lf_end_list() {
+        ; -- end an iterative file listing session (close channels).
+        if list_in_progress {
+            c64.CLRCHN()
+            c64.CLOSE(1)
+            list_in_progress = false
+        }
+    }
+
+
+    sub status(ubyte drivenumber) {
+        ; -- display the disk drive's current status message
+        c64.SETNAM(0, filename)
+        c64.SETLFS(15, drivenumber, 15)
+        void c64.OPEN()          ; open 15,8,15
+        if_cs
+            goto io_error
+        void c64.CHKIN(15)        ; use #15 as input channel
+        if_cs
+            goto io_error
+
+        while not c64.READST()
+            txt.chrout(c64.CHRIN())
+
+io_error:
+        c64.CLRCHN()        ; restore default i/o devices
+        c64.CLOSE(15)
+    }
+
+
+    sub save(ubyte drivenumber, uword filenameptr, uword address, uword size) -> ubyte {
+        c64.SETNAM(strlen(filenameptr), filenameptr)
+        c64.SETLFS(1, drivenumber, 0)
+        uword end_address = address + size
+
+        %asm {{
+            lda  address
+            sta  P8ZP_SCRATCH_W1
+            lda  address+1
+            sta  P8ZP_SCRATCH_W1+1
+            stx  P8ZP_SCRATCH_REG
+            lda  #<P8ZP_SCRATCH_W1
+            ldx  end_address
+            ldy  end_address+1
+            jsr  c64.SAVE
+            php
+            ldx  P8ZP_SCRATCH_REG
+            plp
+        }}
+
+        ubyte result=0
+        if_cc
+            result = c64.READST()==0
+
+        c64.CLRCHN()
+        c64.CLOSE(1)
+
+        return result
+    }
+
+    sub load(ubyte drivenumber, uword filenameptr, uword address_override) -> uword {
+        c64.SETNAM(strlen(filenameptr), filenameptr)
+        ubyte secondary = 1
+        uword end_of_load = 0
+        if address_override
+            secondary = 0
+        c64.SETLFS(1, drivenumber, secondary)
+        %asm {{
+            stx  P8ZP_SCRATCH_REG
+            lda  #0
+            ldx  address_override
+            ldy  address_override+1
+            jsr  c64.LOAD
+            bcs  +
+            stx  end_of_load
+            sty  end_of_load+1
++           ldx  P8ZP_SCRATCH_REG
+        }}
+
+        c64.CLRCHN()
+        c64.CLOSE(1)
+
+        if end_of_load
+            return end_of_load - address_override
+
+        return 0
+    }
+
+
+    str filename = "0:??????????????????????????????????????"
+
+    sub delete(ubyte drivenumber, uword filenameptr) {
+        ; -- delete a file on the drive
+        ubyte flen = strlen(filenameptr)
+        filename[0] = 's'
+        filename[1] = ':'
+        memcopy(filenameptr, &filename+2, flen+1)
+        c64.SETNAM(flen+2, filename)
+        c64.SETLFS(1, drivenumber, 15)
+        void c64.OPEN()
+        c64.CLRCHN()
+        c64.CLOSE(1)
+    }
+
+    sub rename(ubyte drivenumber, uword oldfileptr, uword newfileptr) {
+        ; -- rename a file on the drive
+        ubyte flen_old = strlen(oldfileptr)
+        ubyte flen_new = strlen(newfileptr)
+        filename[0] = 'r'
+        filename[1] = ':'
+        memcopy(newfileptr, &filename+2, flen_new)
+        filename[flen_new+2] = '='
+        memcopy(oldfileptr, &filename+3+flen_new, flen_old+1)
+        c64.SETNAM(3+flen_new+flen_old, filename)
+        c64.SETLFS(1, drivenumber, 15)
+        void c64.OPEN()
+        c64.CLRCHN()
+        c64.CLOSE(1)
+    }
+}

compiler/res/prog8lib/prog8_lib.p8

@@ -5,5 +5,6 @@
 ; indent format: TABS, size=8
 
 prog8_lib {
-	%asminclude "library:prog8lib.asm", ""
+	%asminclude "library:prog8_lib.asm", ""
+	%asminclude "library:prog8_funcs.asm", ""
 }

compiler/res/prog8lib/test_stack.p8

@@ -0,0 +1,48 @@
+%import textio
+
+test_stack {
+
+    asmsub test() {
+        %asm {{
+            stx  _saveX
+            lda  #13
+            jsr  txt.chrout
+            lda  #'-'
+            ldy  #12
+-           jsr  txt.chrout
+            dey
+            bne  -
+            lda  #13
+            jsr  txt.chrout
+            lda  #'x'
+            jsr  txt.chrout
+            lda  #'='
+            jsr  txt.chrout
+            lda  _saveX
+            jsr  txt.print_ub
+            lda  #' '
+            jsr  txt.chrout
+            lda  #'s'
+            jsr  txt.chrout
+            lda  #'p'
+            jsr  txt.chrout
+            lda  #'='
+            jsr  txt.chrout
+            tsx
+            txa
+            jsr  txt.print_ub
+            lda  #13
+            jsr  txt.chrout
+            lda  #'-'
+            ldy  #12
+-           jsr  txt.chrout
+            dey
+            bne  -
+            lda  #13
+            jsr  txt.chrout
+            ldx  _saveX
+            rts
+_saveX   .byte 0
+        }}
+    }
+}

compiler/res/version.txt

@@ -1 +1 @@
-4.0
+5.3

compiler/src/prog8/CompilerMain.kt

@@ -4,13 +4,10 @@ import kotlinx.cli.*
 import prog8.ast.base.AstException
 import prog8.compiler.CompilationResult
 import prog8.compiler.compileProgram
+import prog8.compiler.target.C64Target
+import prog8.compiler.target.Cx16Target
 import prog8.compiler.target.CompilationTarget
-import prog8.compiler.target.c64.C64MachineDefinition
-import prog8.compiler.target.c64.Petscii
-import prog8.compiler.target.c64.codegen.AsmGen
-import prog8.compiler.target.cx16.CX16MachineDefinition
 import prog8.parser.ParsingFailedError
-import java.io.IOException
 import java.nio.file.FileSystems
 import java.nio.file.Path
 import java.nio.file.StandardWatchEventKinds
@@ -36,12 +33,14 @@ fun pathFrom(stringPath: String, vararg rest: String): Path  = FileSystems.getDe
 
 private fun compileMain(args: Array<String>) {
     val cli = CommandLineInterface("prog8compiler")
-    val startEmulator by cli.flagArgument("-emu", "auto-start the Vice C-64 emulator after successful compilation")
+    val startEmulator by cli.flagArgument("-emu", "auto-start emulator after successful compilation")
     val outputDir by cli.flagValueArgument("-out", "directory", "directory for output files instead of current directory", ".")
     val dontWriteAssembly by cli.flagArgument("-noasm", "don't create assembly code")
     val dontOptimize by cli.flagArgument("-noopt", "don't perform any optimizations")
     val watchMode by cli.flagArgument("-watch", "continuous compilation mode (watches for file changes), greatly increases compilation speed")
-    val compilationTarget by cli.flagValueArgument("-target", "compilertarget", "target output of the compiler, currently 'c64' and 'cx16' available", "c64")
+    val slowCodegenWarnings by cli.flagArgument("-slowwarn", "show debug warnings about slow/problematic assembly code generation")
+    val compilationTarget by cli.flagValueArgument("-target", "compilertarget",
+            "target output of the compiler, currently '${C64Target.name}' and '${Cx16Target.name}' available", C64Target.name)
     val moduleFiles by cli.positionalArgumentsList("modules", "main module file(s) to compile", minArgs = 1)
 
     try {
@@ -50,71 +49,48 @@ private fun compileMain(args: Array<String>) {
         exitProcess(1)
     }
 
-    when(compilationTarget) {
-        "c64" -> {
-            with(CompilationTarget) {
-                name = "Commodore-64"
-                machine = C64MachineDefinition
-                encodeString = { str, altEncoding ->
-                    if(altEncoding) Petscii.encodeScreencode(str, true) else Petscii.encodePetscii(str, true)
-                }
-                decodeString = { bytes, altEncoding ->
-                    if(altEncoding) Petscii.decodeScreencode(bytes, true) else Petscii.decodePetscii(bytes, true)
-                }
-                asmGenerator = ::AsmGen
-            }
-        }
-        "cx16" -> {
-            with(CompilationTarget) {
-                name = "Commander X16"
-                machine = CX16MachineDefinition
-                encodeString = { str, altEncoding ->
-                    if(altEncoding) Petscii.encodeScreencode(str, true) else Petscii.encodePetscii(str, true)
-                }
-                decodeString = { bytes, altEncoding ->
-                    if(altEncoding) Petscii.decodeScreencode(bytes, true) else Petscii.decodePetscii(bytes, true)
-                }
-                asmGenerator = ::AsmGen
-            }
-        }
-        else -> {
-            System.err.println("invalid compilation target. Available are: c64, cx16")
-            exitProcess(1)
-        }
-    }
-
     val outputPath = pathFrom(outputDir)
     if(!outputPath.toFile().isDirectory) {
         System.err.println("Output path doesn't exist")
         exitProcess(1)
     }
 
-    if(watchMode && moduleFiles.size<=1) {
+    if(watchMode) {
         val watchservice = FileSystems.getDefault().newWatchService()
 
         while(true) {
-            val filepath = pathFrom(moduleFiles.single()).normalize()
-            println("Continuous watch mode active. Main module: $filepath")
+            println("Continuous watch mode active. Modules: $moduleFiles")
+            val results = mutableListOf<CompilationResult>()
+            for(filepathRaw in moduleFiles) {
+                val filepath = pathFrom(filepathRaw).normalize()
+                val compilationResult = compileProgram(filepath, !dontOptimize, !dontWriteAssembly, slowCodegenWarnings, compilationTarget, outputPath)
+                results.add(compilationResult)
+            }
 
-            try {
-                val compilationResult = compileProgram(filepath, !dontOptimize, !dontWriteAssembly, outputDir=outputPath)
-                println("Imported files (now watching:)")
-                for (importedFile in compilationResult.importedFiles) {
-                    print("  ")
-                    println(importedFile)
-                    importedFile.parent.register(watchservice, StandardWatchEventKinds.ENTRY_MODIFY)
-                }
-                println("[${LocalDateTime.now().withNano(0)}]  Waiting for file changes.")
+            val allImportedFiles = results.flatMap { it.importedFiles }
+            println("Imported files (now watching:)")
+            for (importedFile in allImportedFiles) {
+                print("  ")
+                println(importedFile)
+                val watchDir = importedFile.parent ?: Path.of(".")
+                watchDir.register(watchservice, StandardWatchEventKinds.ENTRY_MODIFY)
+            }
+            println("[${LocalDateTime.now().withNano(0)}]  Waiting for file changes.")
+
+            var recompile=false
+            while(!recompile) {
                 val event = watchservice.take()
-                for(changed in event.pollEvents()) {
+                for (changed in event.pollEvents()) {
                     val changedPath = changed.context() as Path
-                    println("  change detected: $changedPath")
+                    if(allImportedFiles.any { it.fileName == changedPath.fileName }) {
+                        println("  change detected: $changedPath")
+                        recompile = true
+                    }
                 }
                 event.reset()
-                println("\u001b[H\u001b[2J")      // clear the screen
-            } catch (x: Exception) {
-                throw x
             }
+
+            println("\u001b[H\u001b[2J")      // clear the screen
         }
 
     } else {
@@ -122,7 +98,7 @@ private fun compileMain(args: Array<String>) {
             val filepath = pathFrom(filepathRaw).normalize()
             val compilationResult: CompilationResult
             try {
-                compilationResult = compileProgram(filepath, !dontOptimize, !dontWriteAssembly, outputDir=outputPath)
+                compilationResult = compileProgram(filepath, !dontOptimize, !dontWriteAssembly, slowCodegenWarnings, compilationTarget, outputPath)
                 if(!compilationResult.success)
                     exitProcess(1)
             } catch (x: ParsingFailedError) {
@@ -135,7 +111,7 @@ private fun compileMain(args: Array<String>) {
                 if (compilationResult.programName.isEmpty())
                     println("\nCan't start emulator because no program was assembled.")
                 else if(startEmulator) {
-                    CompilationTarget.machine.launchEmulator(compilationResult.programName)
+                    CompilationTarget.instance.machine.launchEmulator(compilationResult.programName)
                 }
             }
         }

compiler/src/prog8/ast/AstToSourceCode.kt

@@ -53,12 +53,14 @@ class AstToSourceCode(val output: (text: String) -> Unit, val program: Program):
     }
 
     override fun visit(expr: BinaryExpression) {
+        output("(")
         expr.left.accept(this)
         if(expr.operator.any { it.isLetter() })
             output(" ${expr.operator} ")
         else
             output(expr.operator)
         expr.right.accept(this)
+        output(")")
     }
 
     override fun visit(directive: Directive) {
@@ -85,7 +87,7 @@ class AstToSourceCode(val output: (text: String) -> Unit, val program: Program):
             DataType.ARRAY_W -> "word["
             DataType.ARRAY_F -> "float["
             DataType.STRUCT -> ""       // the name of the struct is enough
-            else -> "?????2"
+            else -> "?????"
         }
     }
 
@@ -113,10 +115,14 @@ class AstToSourceCode(val output: (text: String) -> Unit, val program: Program):
             VarDeclType.CONST -> output("const ")
             VarDeclType.MEMORY -> output("&")
         }
-        output(decl.struct?.name ?: "")
+
+        if(decl.datatype==DataType.STRUCT && decl.struct!=null)
+            output(decl.struct!!.name)
+
         output(datatypeString(decl.datatype))
         if(decl.arraysize!=null) {
-            decl.arraysize!!.index.accept(this)
+            decl.arraysize!!.indexNum?.accept(this)
+            decl.arraysize!!.indexVar?.accept(this)
         }
         if(decl.isArray)
             output("]")
@@ -137,10 +143,9 @@ class AstToSourceCode(val output: (text: String) -> Unit, val program: Program):
             for(param in subroutine.parameters.zip(subroutine.asmParameterRegisters)) {
                 val reg =
                         when {
-                            param.second.stack -> "stack"
                             param.second.registerOrPair!=null -> param.second.registerOrPair.toString()
                             param.second.statusflag!=null -> param.second.statusflag.toString()
-                            else -> "?????1"
+                            else -> "?????"
                         }
                 output("${datatypeString(param.first.type)} ${param.first.name} @$reg")
                 if(param.first!==subroutine.parameters.last())
@@ -338,7 +343,7 @@ class AstToSourceCode(val output: (text: String) -> Unit, val program: Program):
         output("do ")
         untilLoop.body.accept(this)
         output(" until ")
-        untilLoop.untilCondition.accept(this)
+        untilLoop.condition.accept(this)
     }
 
     override fun visit(returnStmt: Return) {
@@ -347,9 +352,10 @@ class AstToSourceCode(val output: (text: String) -> Unit, val program: Program):
     }
 
     override fun visit(arrayIndexedExpression: ArrayIndexedExpression) {
-        arrayIndexedExpression.identifier.accept(this)
+        arrayIndexedExpression.arrayvar.accept(this)
         output("[")
-        arrayIndexedExpression.arrayspec.index.accept(this)
+        arrayIndexedExpression.indexer.indexNum?.accept(this)
+        arrayIndexedExpression.indexer.indexVar?.accept(this)
         output("]")
     }
 

compiler/src/prog8/ast/AstToplevel.kt

@@ -36,6 +36,18 @@ interface Node {
         throw FatalAstException("scope missing from $this")
     }
 
+    fun definingBlock(): Block {
+        if(this is Block)
+            return this
+        return findParentNode<Block>(this)!!
+    }
+
+    fun containingStatement(): Statement {
+        if(this is Statement)
+            return this
+        return findParentNode<Statement>(this)!!
+    }
+
     fun replaceChildNode(node: Node, replacement: Node)
 }
 
@@ -44,6 +56,7 @@ interface IFunctionCall {
     var args: MutableList<Expression>
 }
 
+
 interface INameScope {
     val name: String
     val position: Position
@@ -137,7 +150,15 @@ interface INameScope {
             }
             return null
         } else {
-            // unqualified name, find the scope the localContext is in, look in that first
+            // unqualified name
+            // special case: the do....until statement can also look INSIDE the anonymous scope
+            if(localContext.parent.parent is UntilLoop) {
+                val symbolFromInnerScope = (localContext.parent.parent as UntilLoop).body.getLabelOrVariable(scopedName[0])
+                if(symbolFromInnerScope!=null)
+                    return symbolFromInnerScope
+            }
+
+            // find the scope the localContext is in, look in that first
             var statementScope = localContext
             while(statementScope !is ParentSentinel) {
                 val localScope = statementScope.definingScope()
@@ -155,7 +176,6 @@ interface INameScope {
     }
 
     fun containsCodeOrVars() = statements.any { it !is Directive || it.directive == "%asminclude" || it.directive == "%asm"}
-    fun containsNoVars() = statements.all { it !is VarDecl }
     fun containsNoCodeNorVars() = !containsCodeOrVars()
 
     fun remove(stmt: Statement) {
@@ -195,6 +215,22 @@ interface INameScope {
         else
             null
     }
+
+    fun previousSibling(stmt: Statement): Statement? {
+        val previousIdx = statements.indexOfFirst { it===stmt } - 1
+        return if(previousIdx>=0)
+            statements[previousIdx]
+        else
+            null
+    }
+
+    fun indexOfChild(stmt: Statement): Int {
+        val idx = statements.indexOfFirst { it===stmt }
+        if(idx>=0)
+            return idx
+        else
+            throw FatalAstException("attempt to find a non-child")
+    }
 }
 
 interface IAssignable {
@@ -228,11 +264,11 @@ class Program(val name: String, val modules: MutableList<Module>): Node {
     override val position: Position = Position.DUMMY
     override var parent: Node
         get() = throw FatalAstException("program has no parent")
-        set(value) = throw FatalAstException("can't set parent of program")
+        set(_) = throw FatalAstException("can't set parent of program")
 
     override fun linkParents(parent: Node) {
         modules.forEach {
-            it.linkParents(this)
+            it.linkParents(namespace)
         }
     }
 
@@ -255,7 +291,10 @@ class Module(override val name: String,
     val importedBy = mutableListOf<Module>()
     val imports = mutableSetOf<Module>()
 
-    var loadAddress: Int = 0        // can be set with the %address directive
+    val loadAddress: Int by lazy {
+        val address = (statements.singleOrNull { it is Directive && it.directive == "%address" } as? Directive)?.args?.single()?.int ?: 0
+        address
+    }
 
     override fun linkParents(parent: Node) {
         this.parent = parent
@@ -280,7 +319,7 @@ class Module(override val name: String,
 class GlobalNamespace(val modules: List<Module>): Node, INameScope {
     override val name = "<<<global>>>"
     override val position = Position("<<<global>>>", 0, 0, 0)
-    override val statements = mutableListOf<Statement>()
+    override val statements = mutableListOf<Statement>()        // not used
     override var parent: Node = ParentSentinel
 
     override fun linkParents(parent: Node) {
@@ -312,6 +351,14 @@ class GlobalNamespace(val modules: List<Module>): Node, INameScope {
                 }
             }
         }
+
+        // special case: the do....until statement can also look INSIDE the anonymous scope
+        if(localContext.parent.parent is UntilLoop) {
+            val symbolFromInnerScope = (localContext.parent.parent as UntilLoop).body.lookup(scopedName, localContext)
+            if(symbolFromInnerScope!=null)
+                return symbolFromInnerScope
+        }
+
         // lookup something from the module.
         return when (val stmt = localContext.definingModule().lookup(scopedName, localContext)) {
             is Label, is VarDecl, is Block, is Subroutine, is StructDecl -> stmt

compiler/src/prog8/ast/antlr/Antr2Kotlin.kt

@@ -30,7 +30,7 @@ private fun ParserRuleContext.toPosition() : Position {
     val customTokensource = this.start.tokenSource as? CustomLexer
     val filename =
             when {
-                customTokensource!=null -> customTokensource.modulePath.fileName.toString()
+                customTokensource!=null -> customTokensource.modulePath.toString()
                 start.tokenSource.sourceName == IntStream.UNKNOWN_SOURCE_NAME -> "@internal@"
                 else -> File(start.inputStream.sourceName).name
             }
@@ -254,8 +254,8 @@ private fun prog8Parser.Asmsub_declContext.toAst(): AsmsubDecl {
     val clobbers = asmsub_clobbers()?.clobber()?.toAst() ?: emptySet()
     val normalParameters = params.map { SubroutineParameter(it.name, it.type, it.position) }
     val normalReturntypes = returns.map { it.type }
-    val paramRegisters = params.map { RegisterOrStatusflag(it.registerOrPair, it.statusflag, it.stack) }
-    val returnRegisters = returns.map { RegisterOrStatusflag(it.registerOrPair, it.statusflag, it.stack) }
+    val paramRegisters = params.map { RegisterOrStatusflag(it.registerOrPair, it.statusflag) }
+    val returnRegisters = returns.map { RegisterOrStatusflag(it.registerOrPair, it.statusflag) }
     return AsmsubDecl(name, normalParameters, normalReturntypes, paramRegisters, returnRegisters, clobbers)
 }
 
@@ -263,13 +263,11 @@ private class AsmSubroutineParameter(name: String,
                                      type: DataType,
                                      val registerOrPair: RegisterOrPair?,
                                      val statusflag: Statusflag?,
-                                     val stack: Boolean,
                                      position: Position) : SubroutineParameter(name, type, position)
 
 private class AsmSubroutineReturn(val type: DataType,
                                   val registerOrPair: RegisterOrPair?,
                                   val statusflag: Statusflag?,
-                                  val stack: Boolean,
                                   val position: Position)
 
 private fun prog8Parser.Asmsub_returnsContext.toAst(): List<AsmSubroutineReturn>
@@ -288,7 +286,7 @@ private fun prog8Parser.Asmsub_returnsContext.toAst(): List<AsmSubroutineReturn>
                     it.datatype().toAst(),
                     registerorpair,
                     statusregister,
-                    !it.stack?.text.isNullOrEmpty(), toPosition())
+                    toPosition())
         }
 
 private fun prog8Parser.Asmsub_paramsContext.toAst(): List<AsmSubroutineParameter>
@@ -305,8 +303,7 @@ private fun prog8Parser.Asmsub_paramsContext.toAst(): List<AsmSubroutineParamete
             else -> throw FatalAstException("invalid register or status flag '$name'")
         }
     }
-    AsmSubroutineParameter(vardecl.varname.text, datatype, registerorpair, statusregister,
-            !it.stack?.text.isNullOrEmpty(), toPosition())
+    AsmSubroutineParameter(vardecl.varname.text, datatype, registerorpair, statusregister, toPosition())
 }
 
 private fun prog8Parser.Functioncall_stmtContext.toAst(): Statement {
@@ -343,14 +340,11 @@ private fun prog8Parser.LabeldefContext.toAst(): Statement =
         Label(children[0].text, toPosition())
 
 private fun prog8Parser.SubroutineContext.toAst() : Subroutine {
+    // non-asm subroutine
+    val returntypes = sub_return_part()?.toAst() ?: emptyList()
     return Subroutine(identifier().text,
             sub_params()?.toAst() ?: emptyList(),
-            sub_return_part()?.toAst() ?: emptyList(),
-            emptyList(),
-            emptyList(),
-            emptySet(),
-            null,
-            false,
+            returntypes,
             statement_block()?.toAst() ?: mutableListOf(),
             toPosition())
 }
@@ -398,7 +392,7 @@ private fun prog8Parser.DirectiveargContext.toAst() : DirectiveArg {
 }
 
 private fun prog8Parser.IntegerliteralContext.toAst(): NumericLiteral {
-    fun makeLiteral(text: String, radix: Int, forceWord: Boolean): NumericLiteral {
+    fun makeLiteral(text: String, radix: Int): NumericLiteral {
         val integer: Int
         var datatype = DataType.UBYTE
         when (radix) {
@@ -436,14 +430,14 @@ private fun prog8Parser.IntegerliteralContext.toAst(): NumericLiteral {
             }
             else -> throw FatalAstException("invalid radix")
         }
-        return NumericLiteral(integer, if (forceWord) DataType.UWORD else datatype)
+        return NumericLiteral(integer, datatype)
     }
     val terminal: TerminalNode = children[0] as TerminalNode
     val integerPart = this.intpart.text
     return when (terminal.symbol.type) {
-        prog8Parser.DEC_INTEGER -> makeLiteral(integerPart, 10, wordsuffix()!=null)
-        prog8Parser.HEX_INTEGER -> makeLiteral(integerPart.substring(1), 16, wordsuffix()!=null)
-        prog8Parser.BIN_INTEGER -> makeLiteral(integerPart.substring(1), 2, wordsuffix()!=null)
+        prog8Parser.DEC_INTEGER -> makeLiteral(integerPart, 10)
+        prog8Parser.HEX_INTEGER -> makeLiteral(integerPart.substring(1), 16)
+        prog8Parser.BIN_INTEGER -> makeLiteral(integerPart.substring(1), 2)
         else -> throw FatalAstException(terminal.text)
     }
 }
@@ -471,8 +465,7 @@ private fun prog8Parser.ExpressionContext.toAst() : Expression {
                 litval.stringliteral()!=null -> litval.stringliteral().toAst()
                 litval.charliteral()!=null -> {
                     try {
-                        val cc=litval.charliteral()
-                        NumericLiteralValue(DataType.UBYTE, CompilationTarget.encodeString(
+                        NumericLiteralValue(DataType.UBYTE, CompilationTarget.instance.encodeString(
                                 unescape(litval.charliteral().SINGLECHAR().text, litval.toPosition()),
                                 litval.charliteral().ALT_STRING_ENCODING()!=null)[0], litval.toPosition())
                     } catch (ce: CharConversionException) {
@@ -648,7 +641,20 @@ private fun prog8Parser.VardeclContext.toAst(): VarDecl {
     )
 }
 
-internal fun escape(str: String) = str.replace("\t", "\\t").replace("\n", "\\n").replace("\r", "\\r")
+internal fun escape(str: String): String {
+    val es = str.map {
+        when(it) {
+            '\t' -> "\\t"
+            '\n' -> "\\n"
+            '\r' -> "\\r"
+            '"' -> "\\\""
+            in '\u8000'..'\u80ff' -> "\\x" + (it.toInt() - 0x8000).toString(16).padStart(2, '0')
+            in '\u0000'..'\u00ff' -> it.toString()
+            else -> "\\u" + it.toInt().toString(16).padStart(4, '0')
+        }
+    }
+    return es.joinToString("")
+}
 
 internal fun unescape(str: String, position: Position): String {
     val result = mutableListOf<Char>()
@@ -662,9 +668,15 @@ internal fun unescape(str: String, position: Position): String {
                 'n' -> '\n'
                 'r' -> '\r'
                 '"' -> '"'
+                '\'' -> '\''
                 'u' -> {
                     "${iter.nextChar()}${iter.nextChar()}${iter.nextChar()}${iter.nextChar()}".toInt(16).toChar()
                 }
+                'x' -> {
+                    // special hack 0x8000..0x80ff  will be outputted verbatim without encoding
+                    val hex = ("" + iter.nextChar() + iter.nextChar()).toInt(16)
+                    (0x8000 + hex).toChar()
+                }
                 else -> throw SyntaxError("invalid escape char in string: \\$ec", position)
             })
         } else {

compiler/src/prog8/ast/base/Base.kt

@@ -35,20 +35,25 @@ enum class DataType {
                 else -> false
             }
 
-
     infix fun isAssignableTo(targetTypes: Set<DataType>) = targetTypes.any { this isAssignableTo it }
+    infix fun isNotAssignableTo(targetType: DataType) = !this.isAssignableTo(targetType)
+    infix fun isNotAssignableTo(targetTypes: Set<DataType>) = !this.isAssignableTo(targetTypes)
 
     infix fun largerThan(other: DataType) =
-            when(this) {
-                in ByteDatatypes -> false
-                in WordDatatypes -> other in ByteDatatypes
+            when {
+                this == other -> false
+                this in ByteDatatypes -> false
+                this in WordDatatypes -> other in ByteDatatypes
+                this==STR && other==UWORD || this==UWORD && other==STR -> false
                 else -> true
             }
 
     infix fun equalsSize(other: DataType) =
-            when(this) {
-                in ByteDatatypes -> other in ByteDatatypes
-                in WordDatatypes -> other in WordDatatypes
+            when {
+                this == other -> true
+                this in ByteDatatypes -> other in ByteDatatypes
+                this in WordDatatypes -> other in WordDatatypes
+                this==STR && other==UWORD || this==UWORD && other==STR -> true
                 else -> false
             }
 
@@ -56,8 +61,8 @@ enum class DataType {
         return when(this) {
             in ByteDatatypes -> 1
             in WordDatatypes -> 2
-            FLOAT -> CompilationTarget.machine.FLOAT_MEM_SIZE
-            in PassByReferenceDatatypes -> CompilationTarget.machine.POINTER_MEM_SIZE
+            FLOAT -> CompilationTarget.instance.machine.FLOAT_MEM_SIZE
+            in PassByReferenceDatatypes -> CompilationTarget.instance.machine.POINTER_MEM_SIZE
             else -> -9999999
         }
     }
@@ -75,7 +80,9 @@ enum class RegisterOrPair {
     Y,
     AX,
     AY,
-    XY;
+    XY,
+    FAC1,
+    FAC2;
 
     companion object {
         val names by lazy { values().map { it.toString()} }
@@ -165,6 +172,7 @@ object ParentSentinel : Node {
 
 data class Position(val file: String, val line: Int, val startCol: Int, val endCol: Int) {
     override fun toString(): String = "[$file: line $line col ${startCol+1}-${endCol+1}]"
+    fun toClickableStr(): String = "$file:$line:$startCol:"
 
     companion object {
         val DUMMY = Position("<dummy>", 0, 0, 0)

compiler/src/prog8/ast/base/ErrorReporting.kt

@@ -24,7 +24,7 @@ class ErrorReporter {
                 MessageSeverity.ERROR -> System.err.print("\u001b[91m")  // bright red
                 MessageSeverity.WARNING -> System.err.print("\u001b[93m")  // bright yellow
             }
-            val msg = "${it.position} ${it.severity} ${it.message}".trim()
+            val msg = "${it.position.toClickableStr()} ${it.severity} ${it.message}".trim()
             if(msg !in alreadyReportedMessages) {
                 System.err.println(msg)
                 alreadyReportedMessages.add(msg)

compiler/src/prog8/ast/base/Errors.kt

@@ -7,11 +7,11 @@ open class FatalAstException (override var message: String) : Exception(message)
 open class AstException (override var message: String) : Exception(message)
 
 open class SyntaxError(override var message: String, val position: Position) : AstException(message) {
-    override fun toString() = "$position Syntax error: $message"
+    override fun toString() = "${position.toClickableStr()} Syntax error: $message"
 }
 
 class ExpressionError(message: String, val position: Position) : AstException(message) {
-    override fun toString() = "$position Error: $message"
+    override fun toString() = "${position.toClickableStr()} Error: $message"
 }
 
 class UndefinedSymbolError(symbol: IdentifierReference)

compiler/src/prog8/ast/base/Extensions.kt

@@ -3,6 +3,7 @@ package prog8.ast.base
 import prog8.ast.Module
 import prog8.ast.Program
 import prog8.ast.processing.*
+import prog8.ast.statements.Directive
 import prog8.compiler.CompilationOptions
 import prog8.compiler.BeforeAsmGenerationAstChanger
 
@@ -18,8 +19,8 @@ internal fun Program.processAstBeforeAsmGeneration(errors: ErrorReporter) {
     fixer.applyModifications()
 }
 
-internal fun Program.reorderStatements() {
-    val reorder = StatementReorderer(this)
+internal fun Program.reorderStatements(errors: ErrorReporter) {
+    val reorder = StatementReorderer(this, errors)
     reorder.visit(this)
     reorder.applyModifications()
 }
@@ -41,12 +42,6 @@ internal fun Module.checkImportedValid() {
     imr.applyModifications()
 }
 
-internal fun Program.checkRecursion(errors: ErrorReporter) {
-    val checker = AstRecursionChecker(namespace, errors)
-    checker.visit(this)
-    checker.processMessages(name)
-}
-
 internal fun Program.checkIdentifiers(errors: ErrorReporter) {
 
     val checker2 = AstIdentifiersChecker(this, errors)
@@ -56,6 +51,9 @@ internal fun Program.checkIdentifiers(errors: ErrorReporter) {
         val transforms = AstVariousTransforms(this)
         transforms.visit(this)
         transforms.applyModifications()
+        val lit2decl = LiteralsToAutoVars(this)
+        lit2decl.visit(this)
+        lit2decl.applyModifications()
     }
 
     if (modules.map { it.name }.toSet().size != modules.size) {
@@ -68,3 +66,37 @@ internal fun Program.variousCleanups() {
     process.visit(this)
     process.applyModifications()
 }
+
+internal fun Program.moveMainAndStartToFirst() {
+    // the module containing the program entrypoint is moved to the first in the sequence.
+    // the "main" block containing the entrypoint is moved to the top in there,
+    // and finally the entrypoint subroutine "start" itself is moved to the top in that block.
+
+    val directives = modules[0].statements.filterIsInstance<Directive>()
+    val start = this.entrypoint()
+    if(start!=null) {
+        val mod = start.definingModule()
+        val block = start.definingBlock()
+        if(!modules.remove(mod))
+            throw FatalAstException("module wrong")
+        modules.add(0, mod)
+        mod.remove(block)
+        var afterDirective = mod.statements.indexOfFirst { it !is Directive }
+        if(afterDirective<0)
+            mod.statements.add(block)
+        else
+            mod.statements.add(afterDirective, block)
+        block.remove(start)
+        afterDirective = block.statements.indexOfFirst { it !is Directive }
+        if(afterDirective<0)
+            block.statements.add(start)
+        else
+            block.statements.add(afterDirective, start)
+
+        // overwrite the directives in the module containing the entrypoint
+        for(directive in directives) {
+            modules[0].statements.removeAll { it is Directive && it.directive == directive.directive }
+            modules[0].statements.add(0, directive)
+        }
+    }
+}

compiler/src/prog8/ast/expressions/AstExpressions.kt

@@ -17,13 +17,13 @@ import kotlin.math.abs
 
 val associativeOperators = setOf("+", "*", "&", "|", "^", "or", "and", "xor", "==", "!=")
 val comparisonOperators = setOf("==", "!=", "<", ">", "<=", ">=")
-
+val augmentAssignmentOperators = setOf("+", "-", "/", "*", "**", "&", "|", "^", "<<", ">>", "%", "and", "or", "xor")
 
 sealed class Expression: Node {
     abstract fun constValue(program: Program): NumericLiteralValue?
     abstract fun accept(visitor: IAstVisitor)
     abstract fun accept(visitor: AstWalker, parent: Node)
-    abstract fun referencesIdentifiers(vararg name: String): Boolean
+    abstract fun referencesIdentifier(vararg scopedName: String): Boolean
     abstract fun inferType(program: Program): InferredTypes.InferredType
 
     infix fun isSameAs(assigntarget: AssignTarget) = assigntarget.isSameAs(this)
@@ -41,8 +41,8 @@ sealed class Expression: Node {
                         && other.left isSameAs left
                         && other.right isSameAs right)
             is ArrayIndexedExpression -> {
-                (other is ArrayIndexedExpression && other.identifier.nameInSource == identifier.nameInSource
-                        && other.arrayspec.index isSameAs arrayspec.index)
+                (other is ArrayIndexedExpression && other.arrayvar.nameInSource == arrayvar.nameInSource
+                        && other.indexer isSameAs indexer)
             }
             is DirectMemoryRead -> {
                 (other is DirectMemoryRead && other.addressExpression isSameAs addressExpression)
@@ -85,7 +85,7 @@ class PrefixExpression(val operator: String, var expression: Expression, overrid
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node)= visitor.visit(this, parent)
 
-    override fun referencesIdentifiers(vararg name: String) = expression.referencesIdentifiers(*name)
+    override fun referencesIdentifier(vararg scopedName: String) = expression.referencesIdentifier(*scopedName)
     override fun inferType(program: Program): InferredTypes.InferredType {
         val inferred = expression.inferType(program)
         return when(operator) {
@@ -142,7 +142,7 @@ class BinaryExpression(var left: Expression, var operator: String, var right: Ex
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node)= visitor.visit(this, parent)
 
-    override fun referencesIdentifiers(vararg name: String) = left.referencesIdentifiers(*name) || right.referencesIdentifiers(*name)
+    override fun referencesIdentifier(vararg scopedName: String) = left.referencesIdentifier(*scopedName) || right.referencesIdentifier(*scopedName)
     override fun inferType(program: Program): InferredTypes.InferredType {
         val leftDt = left.inferType(program)
         val rightDt = right.inferType(program)
@@ -232,20 +232,19 @@ class BinaryExpression(var left: Expression, var operator: String, var right: Ex
     }
 }
 
-class ArrayIndexedExpression(var identifier: IdentifierReference,
-                             val arrayspec: ArrayIndex,
+class ArrayIndexedExpression(var arrayvar: IdentifierReference,
+                             val indexer: ArrayIndex,
                              override val position: Position) : Expression(), IAssignable {
     override lateinit var parent: Node
     override fun linkParents(parent: Node) {
         this.parent = parent
-        identifier.linkParents(this)
-        arrayspec.linkParents(this)
+        arrayvar.linkParents(this)
+        indexer.linkParents(this)
     }
 
     override fun replaceChildNode(node: Node, replacement: Node) {
         when {
-            node===identifier -> identifier = replacement as IdentifierReference
-            node===arrayspec.index -> arrayspec.index = replacement as Expression
+            node===arrayvar -> arrayvar = replacement as IdentifierReference
             else -> throw FatalAstException("invalid replace")
         }
         replacement.parent = this
@@ -255,10 +254,10 @@ class ArrayIndexedExpression(var identifier: IdentifierReference,
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node)= visitor.visit(this, parent)
 
-    override fun referencesIdentifiers(vararg name: String) = identifier.referencesIdentifiers(*name)
+    override fun referencesIdentifier(vararg scopedName: String) = arrayvar.referencesIdentifier(*scopedName)
 
     override fun inferType(program: Program): InferredTypes.InferredType {
-        val target = identifier.targetStatement(program.namespace)
+        val target = arrayvar.targetStatement(program.namespace)
         if (target is VarDecl) {
             return when (target.datatype) {
                 DataType.STR -> InferredTypes.knownFor(DataType.UBYTE)
@@ -270,7 +269,7 @@ class ArrayIndexedExpression(var identifier: IdentifierReference,
     }
 
     override fun toString(): String {
-        return "ArrayIndexed(ident=$identifier, arraysize=$arrayspec; pos=$position)"
+        return "ArrayIndexed(ident=$arrayvar, arraysize=$indexer; pos=$position)"
     }
 }
 
@@ -291,7 +290,7 @@ class TypecastExpression(var expression: Expression, var type: DataType, val imp
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node)= visitor.visit(this, parent)
 
-    override fun referencesIdentifiers(vararg name: String) = expression.referencesIdentifiers(*name)
+    override fun referencesIdentifier(vararg scopedName: String) = expression.referencesIdentifier(*scopedName)
     override fun inferType(program: Program): InferredTypes.InferredType = InferredTypes.knownFor(type)
     override fun constValue(program: Program): NumericLiteralValue? {
         val cv = expression.constValue(program) ?: return null
@@ -322,7 +321,7 @@ data class AddressOf(var identifier: IdentifierReference, override val position:
     }
 
     override fun constValue(program: Program): NumericLiteralValue? = null
-    override fun referencesIdentifiers(vararg name: String) = false
+    override fun referencesIdentifier(vararg scopedName: String) = false
     override fun inferType(program: Program): InferredTypes.InferredType = InferredTypes.knownFor(DataType.UWORD)
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node)= visitor.visit(this, parent)
@@ -345,7 +344,7 @@ class DirectMemoryRead(var addressExpression: Expression, override val position:
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node)= visitor.visit(this, parent)
 
-    override fun referencesIdentifiers(vararg name: String) = false
+    override fun referencesIdentifier(vararg scopedName: String) = false
     override fun inferType(program: Program): InferredTypes.InferredType = InferredTypes.knownFor(DataType.UBYTE)
     override fun constValue(program: Program): NumericLiteralValue? = null
 
@@ -398,7 +397,7 @@ class NumericLiteralValue(val type: DataType,    // only numerical types allowed
         throw FatalAstException("can't replace here")
     }
 
-    override fun referencesIdentifiers(vararg name: String) = false
+    override fun referencesIdentifier(vararg scopedName: String) = false
     override fun constValue(program: Program) = this
 
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
@@ -498,7 +497,7 @@ class StringLiteralValue(val value: String,
         throw FatalAstException("can't replace here")
     }
 
-    override fun referencesIdentifiers(vararg name: String) = false
+    override fun referencesIdentifier(vararg scopedName: String) = false
     override fun constValue(program: Program): NumericLiteralValue? = null
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node)= visitor.visit(this, parent)
@@ -533,7 +532,7 @@ class ArrayLiteralValue(val type: InferredTypes.InferredType,     // inferred be
         replacement.parent = this
     }
 
-    override fun referencesIdentifiers(vararg name: String) = value.any { it.referencesIdentifiers(*name) }
+    override fun referencesIdentifier(vararg scopedName: String) = value.any { it.referencesIdentifier(*scopedName) }
     override fun constValue(program: Program): NumericLiteralValue? = null
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node)= visitor.visit(this, parent)
@@ -569,10 +568,17 @@ class ArrayLiteralValue(val type: InferredTypes.InferredType,     // inferred be
         val dts = datatypesInArray.map { it.typeOrElse(DataType.STRUCT) }
         return when {
             DataType.FLOAT in dts -> InferredTypes.InferredType.known(DataType.ARRAY_F)
+            DataType.STR in dts -> InferredTypes.InferredType.known(DataType.ARRAY_UW)
             DataType.WORD in dts -> InferredTypes.InferredType.known(DataType.ARRAY_W)
             DataType.UWORD in dts -> InferredTypes.InferredType.known(DataType.ARRAY_UW)
             DataType.BYTE in dts -> InferredTypes.InferredType.known(DataType.ARRAY_B)
             DataType.UBYTE in dts -> InferredTypes.InferredType.known(DataType.ARRAY_UB)
+            DataType.ARRAY_UW in dts ||
+                DataType.ARRAY_W in dts ||
+                DataType.ARRAY_UB in dts ||
+                DataType.ARRAY_B in dts ||
+                DataType.ARRAY_F in dts ||
+                DataType.STRUCT in dts -> InferredTypes.InferredType.known(DataType.ARRAY_UW)
             else -> InferredTypes.InferredType.unknown()
         }
     }
@@ -631,7 +637,7 @@ class RangeExpr(var from: Expression,
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node)= visitor.visit(this, parent)
 
-    override fun referencesIdentifiers(vararg name: String): Boolean  = from.referencesIdentifiers(*name) || to.referencesIdentifiers(*name)
+    override fun referencesIdentifier(vararg scopedName: String): Boolean  = from.referencesIdentifier(*scopedName) || to.referencesIdentifier(*scopedName)
     override fun inferType(program: Program): InferredTypes.InferredType {
         val fromDt=from.inferType(program)
         val toDt=to.inferType(program)
@@ -642,7 +648,14 @@ class RangeExpr(var from: Expression,
             fromDt istype DataType.STR && toDt istype DataType.STR -> InferredTypes.knownFor(DataType.STR)
             fromDt istype DataType.WORD || toDt istype DataType.WORD -> InferredTypes.knownFor(DataType.ARRAY_W)
             fromDt istype DataType.BYTE || toDt istype DataType.BYTE -> InferredTypes.knownFor(DataType.ARRAY_B)
-            else -> InferredTypes.knownFor(DataType.ARRAY_UB)
+            else -> {
+                val fdt = fromDt.typeOrElse(DataType.STRUCT)
+                val tdt = toDt.typeOrElse(DataType.STRUCT)
+                if(fdt largerThan tdt)
+                    InferredTypes.knownFor(ElementArrayTypes.getValue(fdt))
+                else
+                    InferredTypes.knownFor(ElementArrayTypes.getValue(tdt))
+            }
         }
     }
     override fun toString(): String {
@@ -664,8 +677,8 @@ class RangeExpr(var from: Expression,
         val toString = to as? StringLiteralValue
         if(fromString!=null && toString!=null ) {
             // string range -> int range over character values
-            fromVal = CompilationTarget.encodeString(fromString.value, fromString.altEncoding)[0].toInt()
-            toVal = CompilationTarget.encodeString(toString.value, fromString.altEncoding)[0].toInt()
+            fromVal = CompilationTarget.instance.encodeString(fromString.value, fromString.altEncoding)[0].toInt()
+            toVal = CompilationTarget.instance.encodeString(toString.value, fromString.altEncoding)[0].toInt()
         } else {
             val fromLv = from as? NumericLiteralValue
             val toLv = to as? NumericLiteralValue
@@ -737,7 +750,8 @@ data class IdentifierReference(val nameInSource: List<String>, override val posi
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node)= visitor.visit(this, parent)
 
-    override fun referencesIdentifiers(vararg name: String): Boolean = nameInSource.last() in name
+    override fun referencesIdentifier(vararg scopedName: String): Boolean =
+            nameInSource.size==scopedName.size && nameInSource.toTypedArray().contentEquals(scopedName)
 
     override fun inferType(program: Program): InferredTypes.InferredType {
         return when (val targetStmt = targetStatement(program.namespace)) {
@@ -759,6 +773,21 @@ data class IdentifierReference(val nameInSource: List<String>, override val posi
             else -> throw FatalAstException("requires a reference value")
         }
     }
+
+    fun firstStructVarName(namespace: INameScope): String? {
+        // take the name of the first struct member of the structvariable instead
+        // if it's just a regular variable, return null.
+        val struct = memberOfStruct(namespace) ?: return null
+        val decl = targetVarDecl(namespace)!!
+        if(decl.datatype!=DataType.STRUCT)
+            return null
+
+        val firstStructMember = struct.nameOfFirstMember()
+        // find the flattened var that belongs to this first struct member
+        val firstVarName = listOf(decl.name, firstStructMember)
+        val firstVar = definingScope().lookup(firstVarName, this) as VarDecl
+        return firstVar.name
+    }
 }
 
 class FunctionCall(override var target: IdentifierReference,
@@ -795,7 +824,7 @@ class FunctionCall(override var target: IdentifierReference,
                 val exprfunc = func.constExpressionFunc
                 if(exprfunc!=null)
                     resultValue = exprfunc(args, position, program)
-                else if(func.returntype==null)
+                else if(func.known_returntype==null)
                     throw ExpressionError("builtin function ${target.nameInSource[0]} can't be used here because it doesn't return a value", position)
             }
 
@@ -825,7 +854,7 @@ class FunctionCall(override var target: IdentifierReference,
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node)= visitor.visit(this, parent)
 
-    override fun referencesIdentifiers(vararg name: String): Boolean = target.referencesIdentifiers(*name) || args.any{it.referencesIdentifiers(*name)}
+    override fun referencesIdentifier(vararg scopedName: String): Boolean = target.referencesIdentifier(*scopedName) || args.any{it.referencesIdentifier(*scopedName)}
 
     override fun inferType(program: Program): InferredTypes.InferredType {
         val constVal = constValue(program ,false)
@@ -845,6 +874,12 @@ class FunctionCall(override var target: IdentifierReference,
                     return InferredTypes.void()     // no return value
                 if(stmt.returntypes.size==1)
                     return InferredTypes.knownFor(stmt.returntypes[0])
+
+                // multiple return values. Can occur for asmsub routines. If there is exactly one register return value, take that.
+                val numRegisterReturns = stmt.asmReturnvaluesRegisters.count { it.registerOrPair!=null }
+                if(numRegisterReturns==1)
+                    return InferredTypes.InferredType.known(DataType.UBYTE)
+
                 return InferredTypes.unknown()     // has multiple return types... so not a single resulting datatype possible
             }
             else -> return InferredTypes.unknown()

compiler/src/prog8/ast/expressions/InferredTypes.kt

@@ -35,6 +35,13 @@ object InferredTypes {
         }
 
         override fun hashCode(): Int = Objects.hash(isVoid, datatype)
+
+        infix fun isAssignableTo(targetDt: InferredType): Boolean =
+                isKnown && targetDt.isKnown && (datatype!! isAssignableTo targetDt.datatype!!)
+        infix fun isAssignableTo(targetDt: DataType): Boolean =
+                isKnown && (datatype!! isAssignableTo targetDt)
+        infix fun isNotAssignableTo(targetDt: InferredType): Boolean = !this.isAssignableTo(targetDt)
+        infix fun isNotAssignableTo(targetDt: DataType): Boolean = !this.isAssignableTo(targetDt)
     }
 
     private val unknownInstance = InferredType.unknown()

compiler/src/prog8/ast/processing/AstChecker.kt

@@ -7,7 +7,9 @@ import prog8.ast.base.*
 import prog8.ast.expressions.*
 import prog8.ast.statements.*
 import prog8.compiler.CompilationOptions
+import prog8.compiler.target.C64Target
 import prog8.compiler.target.CompilationTarget
+import prog8.compiler.target.Cx16Target
 import prog8.functions.BuiltinFunctions
 import java.io.File
 
@@ -142,6 +144,22 @@ internal class AstChecker(private val program: Program,
                     }
                     else -> errors.err("loop variable must be numeric type", forLoop.position)
                 }
+                if(errors.isEmpty()) {
+                    // check loop range values
+                    val range = forLoop.iterable as? RangeExpr
+                    if(range!=null) {
+                        val from = range.from as? NumericLiteralValue
+                        val to = range.to as? NumericLiteralValue
+                        if(from != null)
+                            checkValueTypeAndRange(loopvar.datatype, from)
+                        else if(!range.from.inferType(program).istype(loopvar.datatype))
+                            errors.err("range start value is incompatible with loop variable type", range.position)
+                        if(to != null)
+                            checkValueTypeAndRange(loopvar.datatype, to)
+                        else if(!range.to.inferType(program).istype(loopvar.datatype))
+                            errors.err("range end value is incompatible with loop variable type", range.position)
+                    }
+                }
             }
         }
 
@@ -232,19 +250,19 @@ internal class AstChecker(private val program: Program,
                         err("parameter '${param.first.name}' should be ubyte")
                 }
             }
-            for(ret in subroutine.returntypes.withIndex().zip(subroutine.asmReturnvaluesRegisters)) {
-                if(ret.second.registerOrPair in setOf(RegisterOrPair.A, RegisterOrPair.X, RegisterOrPair.Y)) {
-                    if (ret.first.value != DataType.UBYTE && ret.first.value != DataType.BYTE)
-                        err("return value #${ret.first.index + 1} should be (u)byte")
+            subroutine.returntypes.zip(subroutine.asmReturnvaluesRegisters).forEachIndexed { index, pair ->
+                if(pair.second.registerOrPair in setOf(RegisterOrPair.A, RegisterOrPair.X, RegisterOrPair.Y)) {
+                    if (pair.first != DataType.UBYTE && pair.first != DataType.BYTE)
+                        err("return value #${index + 1} should be (u)byte")
                 }
-                else if(ret.second.registerOrPair in setOf(RegisterOrPair.AX, RegisterOrPair.AY, RegisterOrPair.XY)) {
-                    if (ret.first.value != DataType.UWORD && ret.first.value != DataType.WORD
-                            && ret.first.value != DataType.STR && ret.first.value !in ArrayDatatypes && ret.first.value != DataType.FLOAT)
-                        err("return value #${ret.first.index + 1} should be (u)word/address")
+                else if(pair.second.registerOrPair in setOf(RegisterOrPair.AX, RegisterOrPair.AY, RegisterOrPair.XY)) {
+                    if (pair.first != DataType.UWORD && pair.first != DataType.WORD
+                            && pair.first != DataType.STR && pair.first !in ArrayDatatypes && pair.first != DataType.FLOAT)
+                        err("return value #${index + 1} should be (u)word/address")
                 }
-                else if(ret.second.statusflag!=null) {
-                    if (ret.first.value != DataType.UBYTE)
-                        err("return value #${ret.first.index + 1} should be ubyte")
+                else if(pair.second.statusflag!=null) {
+                    if (pair.first != DataType.UBYTE)
+                        err("return value #${index + 1} should be ubyte")
                 }
             }
 
@@ -270,6 +288,7 @@ internal class AstChecker(private val program: Program,
                             regCounts[CpuRegister.X]=regCounts.getValue(CpuRegister.X)+1
                             regCounts[CpuRegister.Y]=regCounts.getValue(CpuRegister.Y)+1
                         }
+                        RegisterOrPair.FAC1, RegisterOrPair.FAC2 -> { /* no sensible way to count this */ }
                         null ->
                             if(p.statusflag!=null)
                                 statusflagCounts[p.statusflag] = statusflagCounts.getValue(p.statusflag) + 1
@@ -311,8 +330,8 @@ internal class AstChecker(private val program: Program,
     }
 
     override fun visit(untilLoop: UntilLoop) {
-        if(untilLoop.untilCondition.inferType(program).typeOrElse(DataType.STRUCT) !in IntegerDatatypes)
-            errors.err("condition value should be an integer type", untilLoop.untilCondition.position)
+        if(untilLoop.condition.inferType(program).typeOrElse(DataType.STRUCT) !in IntegerDatatypes)
+            errors.err("condition value should be an integer type", untilLoop.condition.position)
         super.visit(untilLoop)
     }
 
@@ -323,17 +342,15 @@ internal class AstChecker(private val program: Program,
     }
 
     override fun visit(assignment: Assignment) {
-        // assigning from a functioncall COULD return multiple values (from an asm subroutine)
         if(assignment.value is FunctionCall) {
             val stmt = (assignment.value as FunctionCall).target.targetStatement(program.namespace)
-            if (stmt is Subroutine && stmt.isAsmSubroutine) {
-                if(stmt.returntypes.size>1)
-                    errors.err("It's not possible to store the multiple results of this asmsub call; you should use a small block of custom inline assembly for this.", assignment.value.position)
-                else {
-                    val idt = assignment.target.inferType(program, assignment)
-                    if(!idt.isKnown || stmt.returntypes.single()!=idt.typeOrElse(DataType.BYTE)) {
-                         errors.err("return type mismatch", assignment.value.position)
-                    }
+            if (stmt is Subroutine) {
+                val idt = assignment.target.inferType(program)
+                if(!idt.isKnown) {
+                     errors.err("return type mismatch", assignment.value.position)
+                }
+                if(stmt.returntypes.size <= 1 && stmt.returntypes.single() isNotAssignableTo idt.typeOrElse(DataType.BYTE)) {
+                    errors.err("return type mismatch", assignment.value.position)
                 }
             }
         }
@@ -353,17 +370,21 @@ internal class AstChecker(private val program: Program,
                         if (sourceVar?.struct != null) {
                             if (sourceVar.struct !== targetVar.struct)
                                 errors.err("assignment of different struct types", assignment.position)
+                        } else if(sourceVar?.isArray==true) {
+                            if((sourceVar.value as ArrayLiteralValue).value.size != targetVar.struct?.numberOfElements)
+                                errors.err("number of elements doesn't match struct definition", sourceVar.position)
                         }
                     }
                 }
             }
         }
 
-        val targetDt = assignment.target.inferType(program, assignment)
-        if(assignment.value.inferType(program) != targetDt) {
+        val targetDt = assignment.target.inferType(program)
+        val valueDt = assignment.value.inferType(program)
+        if(valueDt.isKnown && !(valueDt isAssignableTo targetDt)) {
             if(targetDt.typeOrElse(DataType.STRUCT) in IterableDatatypes)
                 errors.err("cannot assign value to string or array", assignment.value.position)
-            else
+            else if(!(valueDt.istype(DataType.STR) && targetDt.istype(DataType.UWORD)))
                 errors.err("value's type doesn't match target", assignment.value.position)
         }
 
@@ -413,19 +434,15 @@ internal class AstChecker(private val program: Program,
 
         if (assignment is Assignment) {
 
-            val targetDatatype = assignTarget.inferType(program, assignment)
+            val targetDatatype = assignTarget.inferType(program)
             if (targetDatatype.isKnown) {
                 val constVal = assignment.value.constValue(program)
                 if (constVal != null) {
                     checkValueTypeAndRange(targetDatatype.typeOrElse(DataType.BYTE), constVal)
                 } else {
                     val sourceDatatype = assignment.value.inferType(program)
-                    if (!sourceDatatype.isKnown) {
-                        if (assignment.value is FunctionCall) {
-                            val targetStmt = (assignment.value as FunctionCall).target.targetStatement(program.namespace)
-                            if (targetStmt != null)
-                                errors.err("function call doesn't return a suitable value to use in assignment", assignment.value.position)
-                        } else
+                    if (sourceDatatype.isUnknown) {
+                        if (assignment.value !is FunctionCall)
                             errors.err("assignment value is invalid or has no proper datatype", assignment.value.position)
                     } else {
                         checkAssignmentCompatible(targetDatatype.typeOrElse(DataType.BYTE), assignTarget,
@@ -438,26 +455,21 @@ internal class AstChecker(private val program: Program,
 
     override fun visit(addressOf: AddressOf) {
         val variable=addressOf.identifier.targetVarDecl(program.namespace)
-        if(variable==null)
-            errors.err("pointer-of operand must be the name of a heap variable", addressOf.position)
-        else {
-            if(variable.datatype !in ArrayDatatypes
-                    && variable.type!=VarDeclType.MEMORY
-                    && variable.datatype != DataType.STR && variable.datatype!=DataType.STRUCT)
+        if(variable!=null
+            && variable.datatype !in ArrayDatatypes
+            && variable.type!=VarDeclType.MEMORY
+            && variable.struct == null
+            && variable.datatype != DataType.STR && variable.datatype!=DataType.STRUCT)
                 errors.err("invalid pointer-of operand type", addressOf.position)
-        }
         super.visit(addressOf)
     }
 
     override fun visit(decl: VarDecl) {
-        fun err(msg: String, position: Position?=null) {
-            errors.err(msg, position ?: decl.position)
-        }
+        fun err(msg: String, position: Position?=null) = errors.err(msg, position ?: decl.position)
 
         // the initializer value can't refer to the variable itself (recursive definition)
-        if(decl.value?.referencesIdentifiers(decl.name) == true || decl.arraysize?.index?.referencesIdentifiers(decl.name) == true) {
+        if(decl.value?.referencesIdentifier(decl.name) == true || decl.arraysize?.indexVar?.referencesIdentifier(decl.name) == true)
             err("recursive var declaration")
-        }
 
         // CONST can only occur on simple types (byte, word, float)
         if(decl.type== VarDeclType.CONST) {
@@ -465,10 +477,12 @@ internal class AstChecker(private val program: Program,
                 err("const modifier can only be used on numeric types (byte, word, float)")
         }
 
-        // FLOATS
-        if(!compilerOptions.floats && decl.datatype in setOf(DataType.FLOAT, DataType.ARRAY_F) && decl.type!= VarDeclType.MEMORY) {
+        // FLOATS enabled?
+        if(!compilerOptions.floats && decl.datatype in setOf(DataType.FLOAT, DataType.ARRAY_F) && decl.type!= VarDeclType.MEMORY)
             err("floating point used, but that is not enabled via options")
-        }
+
+        if(decl.datatype == DataType.FLOAT && (decl.zeropage==ZeropageWish.REQUIRE_ZEROPAGE || decl.zeropage==ZeropageWish.PREFER_ZEROPAGE))
+            errors.warn("floating point values won't be placed in Zeropage due to size constraints", decl.position)
 
         // ARRAY without size specifier MUST have an iterable initializer value
         if(decl.isArray && decl.arraysize==null) {
@@ -537,9 +551,11 @@ internal class AstChecker(private val program: Program,
                         checkValueTypeAndRange(decl.datatype, decl.value as NumericLiteralValue)
                     }
                     else -> {
-                        err("var/const declaration needs a compile-time constant initializer value, or range, instead found: ${decl.value!!.javaClass.simpleName}")
-                        super.visit(decl)
-                        return
+                        if(decl.type==VarDeclType.CONST) {
+                            err("const declaration needs a compile-time constant initializer value, or range")
+                            super.visit(decl)
+                            return
+                        }
                     }
                 }
             }
@@ -560,13 +576,13 @@ internal class AstChecker(private val program: Program,
                     }
                 }
 
-                if(decl.value !is NumericLiteralValue) {
-                    err("value of memory var decl is not a numeric literal (it is a ${decl.value!!.javaClass.simpleName}).", decl.value?.position)
-                } else {
+                if(decl.value is NumericLiteralValue) {
                     val value = decl.value as NumericLiteralValue
                     if (value.type !in IntegerDatatypes || value.number.toInt() < 0 || value.number.toInt() > 65535) {
                         err("memory address must be valid integer 0..\$ffff", decl.value?.position)
                     }
+                } else {
+                    err("value of memory mapped variable can only be a number, perhaps you meant to use an address pointer type instead?", decl.value?.position)
                 }
             }
         }
@@ -575,8 +591,8 @@ internal class AstChecker(private val program: Program,
         if(declValue!=null && decl.type==VarDeclType.VAR) {
             if(decl.datatype==DataType.STRUCT) {
                 val valueIdt = declValue.inferType(program)
-                if(valueIdt.isUnknown)
-                    throw AstException("invalid value type")
+                if(!valueIdt.isKnown)
+                    throw AstException("unknown dt")
                 val valueDt = valueIdt.typeOrElse(DataType.STRUCT)
                 if(valueDt !in ArrayDatatypes)
                     err("initialisation of struct should be with array value", declValue.position)
@@ -585,26 +601,39 @@ internal class AstChecker(private val program: Program,
             }
         }
 
-        // array length limits
+        // array length limits and constant lenghts
         if(decl.isArray) {
-            val length = decl.arraysize!!.constIndex() ?: 1
-            when (decl.datatype) {
-                DataType.STR, DataType.ARRAY_UB, DataType.ARRAY_B -> {
-                    if(length==0 || length>256)
-                        err("string and byte array length must be 1-256")
+            val length = decl.arraysize!!.constIndex()
+            if(length==null)
+                err("array length must be a constant")
+            else {
+                when (decl.datatype) {
+                    DataType.STR, DataType.ARRAY_UB, DataType.ARRAY_B -> {
+                        if (length == 0 || length > 256)
+                            err("string and byte array length must be 1-256")
+                    }
+                    DataType.ARRAY_UW, DataType.ARRAY_W -> {
+                        if (length == 0 || length > 128)
+                            err("word array length must be 1-128")
+                    }
+                    DataType.ARRAY_F -> {
+                        if (length == 0 || length > 51)
+                            err("float array length must be 1-51")
+                    }
+                    else -> {
+                    }
                 }
-                DataType.ARRAY_UW, DataType.ARRAY_W -> {
-                    if(length==0 || length>128)
-                        err("word array length must be 1-128")
-                }
-                DataType.ARRAY_F -> {
-                    if(length==0 || length>51)
-                        err("float array length must be 1-51")
-                }
-                else -> {}
             }
         }
 
+        // string assignment is not supported in a vard
+        if(decl.datatype==DataType.STR) {
+            if(decl.value==null)
+                err("string var must be initialized with a string literal")
+            else if (decl.type==VarDeclType.VAR && decl.value !is StringLiteralValue)
+                err("string var can only be initialized with a string literal")
+        }
+
         super.visit(decl)
     }
 
@@ -685,9 +714,17 @@ internal class AstChecker(private val program: Program,
                     err("this directive may only occur in a block or at module level")
                 if(directive.args.isEmpty())
                     err("missing option directive argument(s)")
-                else if(directive.args.map{it.name in setOf("enable_floats", "force_output")}.any { !it })
+                else if(directive.args.map{it.name in setOf("enable_floats", "force_output", "no_sysinit")}.any { !it })
                     err("invalid option directive argument(s)")
             }
+            "%target" -> {
+                if(directive.parent !is Block && directive.parent !is Module)
+                    err("this directive may only occur in a block or at module level")
+                if(directive.args.size != 1)
+                    err("directive requires one argument")
+                if(directive.args.single().name !in setOf(C64Target.name, Cx16Target.name))
+                    err("invalid compilation target")
+            }
             else -> throw SyntaxError("invalid directive ${directive.directive}", directive.position)
         }
         super.visit(directive)
@@ -710,6 +747,22 @@ internal class AstChecker(private val program: Program,
             checkValueTypeAndRangeArray(array.type.typeOrElse(DataType.STRUCT), null, arrayspec, array)
         }
 
+        fun isPassByReferenceElement(e: Expression): Boolean {
+            if(e is IdentifierReference) {
+                val decl = e.targetVarDecl(program.namespace)!!
+                return decl.datatype in PassByReferenceDatatypes
+            }
+            return e is StringLiteralValue
+        }
+
+        if(array.parent is VarDecl) {
+            if (!array.value.all { it is NumericLiteralValue || it is AddressOf || isPassByReferenceElement(it) })
+                errors.err("array literal for variable initialization contains invalid types", array.position)
+        } else if(array.parent is ForLoop) {
+            if (!array.value.all { it.constValue(program) != null })
+                errors.err("array literal for iteration must contain constants. Try using a separate array variable instead?", array.position)
+        }
+
         super.visit(array)
     }
 
@@ -719,7 +772,11 @@ internal class AstChecker(private val program: Program,
     }
 
     override fun visit(expr: PrefixExpression) {
-        val dt = expr.inferType(program).typeOrElse(DataType.STRUCT)
+        val idt = expr.inferType(program)
+        if(!idt.isKnown)
+            return  // any error should be reported elsewhere
+
+        val dt = idt.typeOrElse(DataType.STRUCT)
         if(expr.operator=="-") {
             if (dt != DataType.BYTE && dt != DataType.WORD && dt != DataType.FLOAT) {
                 errors.err("can only take negative of a signed number type", expr.position)
@@ -737,6 +794,8 @@ internal class AstChecker(private val program: Program,
     }
 
     override fun visit(expr: BinaryExpression) {
+        super.visit(expr)
+
         val leftIDt = expr.left.inferType(program)
         val rightIDt = expr.right.inferType(program)
         if(!leftIDt.isKnown || !rightIDt.isKnown)
@@ -776,13 +835,12 @@ internal class AstChecker(private val program: Program,
             }
         }
 
-        if(leftDt !in NumericDatatypes)
-            errors.err("left operand is not numeric", expr.left.position)
-        if(rightDt!in NumericDatatypes)
-            errors.err("right operand is not numeric", expr.right.position)
+        if(leftDt !in NumericDatatypes && leftDt != DataType.STR)
+            errors.err("left operand is not numeric or str", expr.left.position)
+        if(rightDt!in NumericDatatypes && rightDt != DataType.STR)
+            errors.err("right operand is not numeric or str", expr.right.position)
         if(leftDt!=rightDt)
             errors.err("left and right operands aren't the same type", expr.left.position)
-        super.visit(expr)
     }
 
     override fun visit(typecast: TypecastExpression) {
@@ -842,7 +900,29 @@ internal class AstChecker(private val program: Program,
 
         val error = VerifyFunctionArgTypes.checkTypes(functionCall, functionCall.definingScope(), program)
         if(error!=null)
-            errors.err(error, functionCall.args.first().position)
+            errors.err(error, functionCall.position)
+
+        // check the functions that return multiple returnvalues.
+        val stmt = functionCall.target.targetStatement(program.namespace)
+        if (stmt is Subroutine) {
+            if (stmt.returntypes.size > 1) {
+                // Currently, it's only possible to handle ONE (or zero) return values from a subroutine.
+                // asmsub routines can have multiple return values, for instance in 2 different registers.
+                // It's not (yet) possible to handle these multiple return values because assignments
+                // are only to a single unique target at the same time.
+                //   EXCEPTION:
+                // if the asmsub returns multiple values and one of them is via a status register bit,
+                // it *is* possible to handle them by just actually assigning the register value and
+                // dealing with the status bit as just being that, the status bit after the call.
+                val (returnRegisters, returnStatusflags) = stmt.asmReturnvaluesRegisters.partition { rr -> rr.registerOrPair != null }
+                if (returnRegisters.isEmpty() || returnRegisters.size == 1) {
+                    if (returnStatusflags.any())
+                        errors.warn("this asmsub also has one or more return 'values' in one of the status flags", functionCall.position)
+                } else {
+                    errors.err("It's not possible to store the multiple result values of this asmsub call; you should use a small block of custom inline assembly for this.", functionCall.position)
+                }
+            }
+        }
 
         super.visit(functionCall)
     }
@@ -911,13 +991,10 @@ internal class AstChecker(private val program: Program,
             if(target.regXasResult())
                 errors.warn("subroutine call return value in X register is discarded and replaced by 0", position)
             if(target.isAsmSubroutine) {
-                for (arg in args.withIndex().zip(target.parameters)) {
-                    val argIDt = arg.first.value.inferType(program)
+                for (arg in args.zip(target.parameters)) {
+                    val argIDt = arg.first.inferType(program)
                     if (!argIDt.isKnown)
                         return
-                    if (target.asmParameterRegisters[arg.first.index].registerOrPair in setOf(RegisterOrPair.AX, RegisterOrPair.XY, RegisterOrPair.X)
-                            && arg.first.value !is NumericLiteralValue && arg.first.value !is IdentifierReference)
-                        errors.warn("calling a subroutine that expects X as a parameter is problematic. If you see a compiler error/crash about this later, try to change this call", position)
                 }
             }
         }
@@ -938,7 +1015,7 @@ internal class AstChecker(private val program: Program,
                 }
             }
         } else if(postIncrDecr.target.arrayindexed != null) {
-            val target = postIncrDecr.target.arrayindexed?.identifier?.targetStatement(program.namespace)
+            val target = postIncrDecr.target.arrayindexed?.arrayvar?.targetStatement(program.namespace)
             if(target==null) {
                 errors.err("undefined symbol", postIncrDecr.position)
             }
@@ -953,32 +1030,38 @@ internal class AstChecker(private val program: Program,
     }
 
     override fun visit(arrayIndexedExpression: ArrayIndexedExpression) {
-        val target = arrayIndexedExpression.identifier.targetStatement(program.namespace)
+        val target = arrayIndexedExpression.arrayvar.targetStatement(program.namespace)
         if(target is VarDecl) {
             if(target.datatype !in IterableDatatypes)
                 errors.err("indexing requires an iterable variable", arrayIndexedExpression.position)
             val arraysize = target.arraysize?.constIndex()
             if(arraysize!=null) {
                 // check out of bounds
-                val index = (arrayIndexedExpression.arrayspec.index as? NumericLiteralValue)?.number?.toInt()
+                val index = arrayIndexedExpression.indexer.constIndex()
                 if(index!=null && (index<0 || index>=arraysize))
-                    errors.err("array index out of bounds", arrayIndexedExpression.arrayspec.position)
+                    errors.err("array index out of bounds", arrayIndexedExpression.indexer.position)
             } else if(target.datatype == DataType.STR) {
                 if(target.value is StringLiteralValue) {
                     // check string lengths for non-memory mapped strings
                     val stringLen = (target.value as StringLiteralValue).value.length
-                    val index = (arrayIndexedExpression.arrayspec.index as? NumericLiteralValue)?.number?.toInt()
+                    val index = arrayIndexedExpression.indexer.constIndex()
                     if (index != null && (index < 0 || index >= stringLen))
-                        errors.err("index out of bounds", arrayIndexedExpression.arrayspec.position)
+                        errors.err("index out of bounds", arrayIndexedExpression.indexer.position)
                 }
             }
         } else
             errors.err("indexing requires a variable to act upon", arrayIndexedExpression.position)
 
         // check index value 0..255
-        val dtx = arrayIndexedExpression.arrayspec.index.inferType(program).typeOrElse(DataType.STRUCT)
-        if(dtx!= DataType.UBYTE && dtx!= DataType.BYTE)
+        val dtxNum = arrayIndexedExpression.indexer.indexNum?.inferType(program)?.typeOrElse(DataType.STRUCT)
+        if(dtxNum!=null && dtxNum != DataType.UBYTE && dtxNum != DataType.BYTE)
             errors.err("array indexing is limited to byte size 0..255", arrayIndexedExpression.position)
+        val dtxVar = arrayIndexedExpression.indexer.indexVar?.inferType(program)?.typeOrElse(DataType.STRUCT)
+        if(dtxVar!=null && dtxVar != DataType.UBYTE && dtxVar != DataType.BYTE)
+            errors.err("array indexing is limited to byte size 0..255", arrayIndexedExpression.position)
+
+        if(arrayIndexedExpression.indexer.origExpression!=null)
+            throw FatalAstException("array indexer should have been replaced with a temp var @ ${arrayIndexedExpression.indexer.position}")
 
         super.visit(arrayIndexedExpression)
     }
@@ -1069,7 +1152,7 @@ internal class AstChecker(private val program: Program,
         }
 
         if(value.type.isUnknown)
-            return err("attempt to check values of array with as yet unknown datatype")
+            return false
 
         when (targetDt) {
             DataType.STR -> return err("string value expected")
@@ -1083,10 +1166,7 @@ internal class AstChecker(private val program: Program,
                     if(arraySpecSize!=null && arraySpecSize>0) {
                         if(arraySpecSize<1 || arraySpecSize>256)
                             return err("byte array length must be 1-256")
-                        val constX = arrayspec.index.constValue(program)
-                        if(constX?.type !in IntegerDatatypes)
-                            return err("array size specifier must be constant integer value")
-                        val expectedSize = constX!!.number.toInt()
+                        val expectedSize = arrayspec.constIndex() ?: return err("array size specifier must be constant integer value")
                         if (arraySize != expectedSize)
                             return err("initializer array size mismatch (expecting $expectedSize, got $arraySize)")
                         return true
@@ -1105,10 +1185,7 @@ internal class AstChecker(private val program: Program,
                     if(arraySpecSize!=null && arraySpecSize>0) {
                         if(arraySpecSize<1 || arraySpecSize>128)
                             return err("word array length must be 1-128")
-                        val constX = arrayspec.index.constValue(program)
-                        if(constX?.type !in IntegerDatatypes)
-                            return err("array size specifier must be constant integer value")
-                        val expectedSize = constX!!.number.toInt()
+                        val expectedSize = arrayspec.constIndex() ?: return err("array size specifier must be constant integer value")
                         if (arraySize != expectedSize)
                             return err("initializer array size mismatch (expecting $expectedSize, got $arraySize)")
                         return true
@@ -1127,10 +1204,7 @@ internal class AstChecker(private val program: Program,
                     if(arraySpecSize!=null && arraySpecSize>0) {
                         if(arraySpecSize < 1 || arraySpecSize>51)
                             return err("float array length must be 1-51")
-                        val constX = arrayspec.index.constValue(program)
-                        if(constX?.type !in IntegerDatatypes)
-                            return err("array size specifier must be constant integer value")
-                        val expectedSize = constX!!.number.toInt()
+                        val expectedSize = arrayspec.constIndex() ?: return err("array size specifier must be constant integer value")
                         if (arraySize != expectedSize)
                             return err("initializer array size mismatch (expecting $expectedSize, got $arraySize)")
                     } else
@@ -1138,7 +1212,7 @@ internal class AstChecker(private val program: Program,
 
                     // check if the floating point values are all within range
                     val doubles = value.value.map {it.constValue(program)?.number!!.toDouble()}.toDoubleArray()
-                    if(doubles.any { it < CompilationTarget.machine.FLOAT_MAX_NEGATIVE || it > CompilationTarget.machine.FLOAT_MAX_POSITIVE })
+                    if(doubles.any { it < CompilationTarget.instance.machine.FLOAT_MAX_NEGATIVE || it > CompilationTarget.instance.machine.FLOAT_MAX_POSITIVE })
                         return err("floating point value overflow")
                     return true
                 }
@@ -1151,7 +1225,7 @@ internal class AstChecker(private val program: Program,
                     for(elt in value.value.zip(struct.statements)) {
                         val vardecl = elt.second as VarDecl
                         val valuetype = elt.first.inferType(program)
-                        if (!valuetype.isKnown || !(valuetype.typeOrElse(DataType.STRUCT) isAssignableTo vardecl.datatype)) {
+                        if (!valuetype.isKnown || valuetype isNotAssignableTo vardecl.datatype) {
                             errors.err("invalid struct member init value type $valuetype, expected ${vardecl.datatype}", elt.first.position)
                             return false
                         }
@@ -1283,9 +1357,7 @@ internal class AstChecker(private val program: Program,
         else if(sourceDatatype== DataType.FLOAT && targetDatatype in IntegerDatatypes)
             errors.err("cannot assign float to ${targetDatatype.name.toLowerCase()}; possible loss of precision. Suggestion: round the value or revert to integer arithmetic", position)
         else {
-            if(targetDatatype==DataType.UWORD && sourceDatatype in PassByReferenceDatatypes)
-                errors.err("cannot assign ${sourceDatatype.name.toLowerCase()} to ${targetDatatype.name.toLowerCase()}, perhaps forgot '&' ?", position)
-            else
+            if(targetDatatype!=DataType.UWORD && sourceDatatype !in PassByReferenceDatatypes)
                 errors.err("cannot assign ${sourceDatatype.name.toLowerCase()} to ${targetDatatype.name.toLowerCase()}", position)
         }
 

compiler/src/prog8/ast/processing/AstIdentifiersChecker.kt

@@ -22,7 +22,7 @@ internal class AstIdentifiersChecker(private val program: Program, private val e
     }
 
     override fun visit(block: Block) {
-        if(block.name in CompilationTarget.machine.opcodeNames)
+        if(block.name in CompilationTarget.instance.machine.opcodeNames)
             errors.err("can't use a cpu opcode name as a symbol: '${block.name}'", block.position)
 
         val existing = blocks[block.name]
@@ -34,13 +34,23 @@ internal class AstIdentifiersChecker(private val program: Program, private val e
         super.visit(block)
     }
 
+    override fun visit(directive: Directive) {
+        if(directive.directive=="%target") {
+            val compatibleTarget = directive.args.single().name
+            if (compatibleTarget != CompilationTarget.instance.name)
+                errors.err("module's compilation target ($compatibleTarget) differs from active target (${CompilationTarget.instance.name})", directive.position)
+        }
+
+        super.visit(directive)
+    }
+
     override fun visit(decl: VarDecl) {
         decl.datatypeErrors.forEach { errors.err(it.message, it.position) }
 
         if(decl.name in BuiltinFunctions)
             errors.err("builtin function cannot be redefined", decl.position)
 
-        if(decl.name in CompilationTarget.machine.opcodeNames)
+        if(decl.name in CompilationTarget.instance.machine.opcodeNames)
             errors.err("can't use a cpu opcode name as a symbol: '${decl.name}'", decl.position)
 
         if(decl.datatype==DataType.STRUCT) {
@@ -70,11 +80,16 @@ internal class AstIdentifiersChecker(private val program: Program, private val e
         if (existing != null && existing !== decl)
             nameError(decl.name, decl.position, existing)
 
+        if(decl.definingBlock().name==decl.name)
+            nameError(decl.name, decl.position, decl.definingBlock())
+        if(decl.definingSubroutine()?.name==decl.name)
+            nameError(decl.name, decl.position, decl.definingSubroutine()!!)
+
         super.visit(decl)
     }
 
     override fun visit(subroutine: Subroutine) {
-        if(subroutine.name in CompilationTarget.machine.opcodeNames) {
+        if(subroutine.name in CompilationTarget.instance.machine.opcodeNames) {
             errors.err("can't use a cpu opcode name as a symbol: '${subroutine.name}'", subroutine.position)
         } else if(subroutine.name in BuiltinFunctions) {
             // the builtin functions can't be redefined
@@ -88,14 +103,6 @@ internal class AstIdentifiersChecker(private val program: Program, private val e
             if (existing != null && existing !== subroutine)
                 nameError(subroutine.name, subroutine.position, existing)
 
-            // does the parameter redefine a variable declared elsewhere?
-            for(param in subroutine.parameters) {
-                val existingVar = subroutine.lookup(listOf(param.name), subroutine)
-                if (existingVar != null && existingVar.parent !== subroutine) {
-                    nameError(param.name, param.position, existingVar)
-                }
-            }
-
             // check that there are no local variables, labels, or other subs that redefine the subroutine's parameters
             val symbolsInSub = subroutine.allDefinedSymbols()
             val namesInSub = symbolsInSub.map{ it.first }.toSet()
@@ -119,7 +126,7 @@ internal class AstIdentifiersChecker(private val program: Program, private val e
     }
 
     override fun visit(label: Label) {
-        if(label.name in CompilationTarget.machine.opcodeNames)
+        if(label.name in CompilationTarget.instance.machine.opcodeNames)
             errors.err("can't use a cpu opcode name as a symbol: '${label.name}'", label.position)
 
         if(label.name in BuiltinFunctions) {
@@ -141,8 +148,8 @@ internal class AstIdentifiersChecker(private val program: Program, private val e
     }
 
     override fun visit(string: StringLiteralValue) {
-        if (string.value.length !in 1..255)
-            errors.err("string literal length must be between 1 and 255", string.position)
+        if (string.value.length > 255)
+            errors.err("string literal length max is 255", string.position)
 
         super.visit(string)
     }

compiler/src/prog8/ast/processing/AstRecursionChecker.kt

@@ -1,118 +0,0 @@
-package prog8.ast.processing
-
-import prog8.ast.INameScope
-import prog8.ast.base.ErrorReporter
-import prog8.ast.base.Position
-import prog8.ast.expressions.FunctionCall
-import prog8.ast.statements.FunctionCallStatement
-import prog8.ast.statements.Subroutine
-
-
-internal class AstRecursionChecker(private val namespace: INameScope,
-                                   private val errors: ErrorReporter) : IAstVisitor {
-    private val callGraph = DirectedGraph<INameScope>()
-
-    fun processMessages(modulename: String) {
-        val cycle = callGraph.checkForCycle()
-        if(cycle.isEmpty())
-            return
-        val chain = cycle.joinToString(" <-- ") { "${it.name} at ${it.position}" }
-        errors.err("Program contains recursive subroutine calls, this is not supported. Recursive chain:\n (a subroutine call in) $chain", Position.DUMMY)
-    }
-
-    override fun visit(functionCallStatement: FunctionCallStatement) {
-        val scope = functionCallStatement.definingScope()
-        val targetStatement = functionCallStatement.target.targetStatement(namespace)
-        if(targetStatement!=null) {
-            val targetScope = when (targetStatement) {
-                is Subroutine -> targetStatement
-                else -> targetStatement.definingScope()
-            }
-            callGraph.add(scope, targetScope)
-        }
-        super.visit(functionCallStatement)
-    }
-
-    override fun visit(functionCall: FunctionCall) {
-        val scope = functionCall.definingScope()
-        val targetStatement = functionCall.target.targetStatement(namespace)
-        if(targetStatement!=null) {
-            val targetScope = when (targetStatement) {
-                is Subroutine -> targetStatement
-                else -> targetStatement.definingScope()
-            }
-            callGraph.add(scope, targetScope)
-        }
-        super.visit(functionCall)
-    }
-
-    private class DirectedGraph<VT> {
-        private val graph = mutableMapOf<VT, MutableSet<VT>>()
-        private var uniqueVertices = mutableSetOf<VT>()
-        val numVertices : Int
-            get() = uniqueVertices.size
-
-        fun add(from: VT, to: VT) {
-            var targets = graph[from]
-            if(targets==null) {
-                targets = mutableSetOf()
-                graph[from] = targets
-            }
-            targets.add(to)
-            uniqueVertices.add(from)
-            uniqueVertices.add(to)
-        }
-
-        fun print() {
-            println("#vertices: $numVertices")
-            graph.forEach { (from, to) ->
-                println("$from   CALLS:")
-                to.forEach { println("   $it") }
-            }
-            val cycle = checkForCycle()
-            if(cycle.isNotEmpty()) {
-                println("CYCLIC!  $cycle")
-            }
-        }
-
-        fun checkForCycle(): MutableList<VT> {
-            val visited = uniqueVertices.associateWith { false }.toMutableMap()
-            val recStack = uniqueVertices.associateWith { false }.toMutableMap()
-            val cycle = mutableListOf<VT>()
-            for(node in uniqueVertices) {
-                if(isCyclicUntil(node, visited, recStack, cycle))
-                    return cycle
-            }
-            return mutableListOf()
-        }
-
-        private fun isCyclicUntil(node: VT,
-                                  visited: MutableMap<VT, Boolean>,
-                                  recStack: MutableMap<VT, Boolean>,
-                                  cycleNodes: MutableList<VT>): Boolean {
-
-            if(recStack[node]==true) return true
-            if(visited[node]==true) return false
-
-            // mark current node as visited and add to recursion stack
-            visited[node] = true
-            recStack[node] = true
-
-            // recurse for all neighbours
-            val neighbors = graph[node]
-            if(neighbors!=null) {
-                for (neighbour in neighbors) {
-                    if (isCyclicUntil(neighbour, visited, recStack, cycleNodes)) {
-                        cycleNodes.add(node)
-                        return true
-                    }
-                }
-            }
-
-            // pop node from recursion stack
-            recStack[node] = false
-            return false
-        }
-    }
-
-}

compiler/src/prog8/ast/processing/AstVariousTransforms.kt

@@ -10,43 +10,6 @@ import prog8.ast.statements.*
 internal class AstVariousTransforms(private val program: Program) : AstWalker() {
     private val noModifications = emptyList<IAstModification>()
 
-    override fun after(functionCallStatement: FunctionCallStatement, parent: Node): Iterable<IAstModification> {
-        if(functionCallStatement.target.nameInSource == listOf("swap")) {
-            // if x and y are both just identifiers, do not rewrite (there should be asm generation for that)
-            // otherwise:
-            // rewrite swap(x,y) as follows:
-            // - declare a temp variable of the same datatype
-            // - temp = x, x = y, y= temp
-            val first = functionCallStatement.args[0]
-            val second = functionCallStatement.args[1]
-            if(first !is IdentifierReference && second !is IdentifierReference) {
-                val dt = first.inferType(program).typeOrElse(DataType.STRUCT)
-                val tempname = "prog8_swaptmp_${functionCallStatement.hashCode()}"
-                val tempvardecl = VarDecl(VarDeclType.VAR, dt, ZeropageWish.DONTCARE, null, tempname, null, null, isArray = false, autogeneratedDontRemove = true, position = first.position)
-                val tempvar = IdentifierReference(listOf(tempname), first.position)
-                val assignTemp = Assignment(
-                        AssignTarget(tempvar, null, null, first.position),
-                        first,
-                        first.position
-                )
-                val assignFirst = Assignment(
-                        AssignTarget.fromExpr(first),
-                        second,
-                        first.position
-                )
-                val assignSecond = Assignment(
-                        AssignTarget.fromExpr(second),
-                        tempvar,
-                        first.position
-                )
-                val scope = AnonymousScope(mutableListOf(tempvardecl, assignTemp, assignFirst, assignSecond), first.position)
-                return listOf(IAstModification.ReplaceNode(functionCallStatement, scope, parent))
-            }
-        }
-
-        return noModifications
-    }
-
     override fun before(decl: VarDecl, parent: Node): Iterable<IAstModification> {
         // is it a struct variable? then define all its struct members as mangled names,
         //    and include the original decl as well.
@@ -84,78 +47,59 @@ internal class AstVariousTransforms(private val program: Program) : AstWalker()
         return noModifications
     }
 
-    override fun before(expr: BinaryExpression, parent: Node): Iterable<IAstModification> {
-        when {
-            expr.left is StringLiteralValue ->
-                return listOf(IAstModification.ReplaceNode(
-                        expr,
-                        processBinaryExprWithString(expr.left as StringLiteralValue, expr.right, expr),
-                        parent
-                ))
-            expr.right is StringLiteralValue ->
-                return listOf(IAstModification.ReplaceNode(
-                        expr,
-                        processBinaryExprWithString(expr.right as StringLiteralValue, expr.left, expr),
-                        parent
-                ))
+    override fun after(expr: BinaryExpression, parent: Node): Iterable<IAstModification> {
+        val leftStr = expr.left as? StringLiteralValue
+        val rightStr = expr.right as? StringLiteralValue
+        if(expr.operator == "+") {
+            val concatenatedString = concatString(expr)
+            if(concatenatedString!=null)
+                return listOf(IAstModification.ReplaceNode(expr, concatenatedString, parent))
         }
-
-        return noModifications
-    }
-
-    override fun after(string: StringLiteralValue, parent: Node): Iterable<IAstModification> {
-        if(string.parent !is VarDecl) {
-            // replace the literal string by a identifier reference to a new local vardecl
-            val vardecl = VarDecl.createAuto(string)
-            val identifier = IdentifierReference(listOf(vardecl.name), vardecl.position)
-            return listOf(
-                    IAstModification.ReplaceNode(string, identifier, parent),
-                    IAstModification.InsertFirst(vardecl, string.definingScope() as Node)
-            )
-        }
-        return noModifications
-    }
-
-    override fun after(array: ArrayLiteralValue, parent: Node): Iterable<IAstModification> {
-        val vardecl = array.parent as? VarDecl
-        if(vardecl!=null) {
-            // adjust the datatype of the array (to an educated guess)
-            val arrayDt = array.type
-            if(!arrayDt.istype(vardecl.datatype)) {
-                val cast = array.cast(vardecl.datatype)
-                if (cast != null && cast!=array)
-                    return listOf(IAstModification.ReplaceNode(vardecl.value!!, cast, vardecl))
+        else if(expr.operator == "*") {
+            if (leftStr!=null) {
+                val amount = expr.right.constValue(program)
+                if(amount!=null) {
+                    val string = leftStr.value.repeat(amount.number.toInt())
+                    val strval = StringLiteralValue(string, leftStr.altEncoding, expr.position)
+                    return listOf(IAstModification.ReplaceNode(expr, strval, parent))
+                }
             }
-        } else {
-            val arrayDt = array.guessDatatype(program)
-            if(arrayDt.isKnown) {
-                // this array literal is part of an expression, turn it into an identifier reference
-                val litval2 = array.cast(arrayDt.typeOrElse(DataType.STRUCT))
-                if(litval2!=null && litval2!=array) {
-                    val vardecl2 = VarDecl.createAuto(litval2)
-                    val identifier = IdentifierReference(listOf(vardecl2.name), vardecl2.position)
-                    return listOf(
-                            IAstModification.ReplaceNode(array, identifier, parent),
-                            IAstModification.InsertFirst(vardecl2, array.definingScope() as Node)
-                    )
+            else if (rightStr!=null) {
+                val amount = expr.right.constValue(program)
+                if(amount!=null) {
+                    val string = rightStr.value.repeat(amount.number.toInt())
+                    val strval = StringLiteralValue(string, rightStr.altEncoding, expr.position)
+                    return listOf(IAstModification.ReplaceNode(expr, strval, parent))
                 }
             }
         }
+
         return noModifications
     }
 
-    private fun processBinaryExprWithString(string: StringLiteralValue, operand: Expression, expr: BinaryExpression): Expression {
-        val constvalue = operand.constValue(program)
-        if(constvalue!=null) {
-            if (expr.operator == "*") {
-                // repeat a string a number of times
-                return StringLiteralValue(string.value.repeat(constvalue.number.toInt()), string.altEncoding, expr.position)
+    private fun concatString(expr: BinaryExpression): StringLiteralValue? {
+        val rightStrval = expr.right as? StringLiteralValue
+        val leftStrval = expr.left as? StringLiteralValue
+        return when {
+            expr.operator!="+" -> null
+            expr.left is BinaryExpression && rightStrval!=null -> {
+                val subStrVal = concatString(expr.left as BinaryExpression)
+                if(subStrVal==null)
+                    null
+                else
+                    StringLiteralValue("${subStrVal.value}${rightStrval.value}", subStrVal.altEncoding, rightStrval.position)
             }
+            expr.right is BinaryExpression && leftStrval!=null -> {
+                val subStrVal = concatString(expr.right as BinaryExpression)
+                if(subStrVal==null)
+                    null
+                else
+                    StringLiteralValue("${leftStrval.value}${subStrVal.value}", subStrVal.altEncoding, leftStrval.position)
+            }
+            leftStrval!=null && rightStrval!=null -> {
+                StringLiteralValue("${leftStrval.value}${rightStrval.value}", leftStrval.altEncoding, leftStrval.position)
+            }
+            else -> null
         }
-        if(expr.operator == "+" && operand is StringLiteralValue) {
-            // concatenate two strings
-            return StringLiteralValue("${string.value}${operand.value}", string.altEncoding, expr.position)
-        }
-        return expr
     }
 }

compiler/src/prog8/ast/processing/AstWalker.kt

@@ -9,79 +9,60 @@ import prog8.ast.statements.*
 interface IAstModification {
     fun perform()
 
-    class Remove(val node: Node, val parent: Node) : IAstModification {
+    class Remove(val node: Node, val parent: INameScope) : IAstModification {
         override fun perform() {
-            if(parent is INameScope) {
-                if (!parent.statements.remove(node) && parent !is GlobalNamespace)
-                    throw FatalAstException("attempt to remove non-existing node $node")
-            } else {
-                throw FatalAstException("parent of a remove modification is not an INameScope")
-            }
+            if (!parent.statements.remove(node) && parent !is GlobalNamespace)
+                throw FatalAstException("attempt to remove non-existing node $node")
         }
     }
 
-    class SetExpression(val setter: (newExpr: Expression) -> Unit, val newExpr: Expression, val parent: Node) : IAstModification {
+    class SetExpression(private val setter: (newExpr: Expression) -> Unit, private val newExpr: Expression, private val parent: Node) : IAstModification {
         override fun perform() {
             setter(newExpr)
             newExpr.linkParents(parent)
         }
     }
 
-    class InsertFirst(val stmt: Statement, val parent: Node) : IAstModification {
+    class InsertFirst(private val stmt: Statement, private val parent: INameScope) : IAstModification {
         override fun perform() {
-            if(parent is INameScope) {
-                parent.statements.add(0, stmt)
-                stmt.linkParents(parent)
-            } else {
-                throw FatalAstException("parent of an insert modification is not an INameScope")
-            }
+            parent.statements.add(0, stmt)
+            stmt.linkParents(parent as Node)
         }
     }
 
-    class InsertLast(val stmt: Statement, val parent: Node) : IAstModification {
+    class InsertLast(private val stmt: Statement, private val parent: INameScope) : IAstModification {
         override fun perform() {
-            if(parent is INameScope) {
-                parent.statements.add(stmt)
-                stmt.linkParents(parent)
-            } else {
-                throw FatalAstException("parent of an insert modification is not an INameScope")
-            }
+            parent.statements.add(stmt)
+            stmt.linkParents(parent as Node)
         }
     }
 
-    class InsertAfter(val after: Statement, val stmt: Statement, val parent: Node) : IAstModification {
+    class InsertAfter(private val after: Statement, private val stmt: Statement, private val parent: INameScope) : IAstModification {
         override fun perform() {
-            if(parent is INameScope) {
-                val idx = parent.statements.indexOfFirst { it===after } + 1
-                parent.statements.add(idx, stmt)
-                stmt.linkParents(parent)
-            } else {
-                throw FatalAstException("parent of an insert modification is not an INameScope")
-            }
+            val idx = parent.statements.indexOfFirst { it===after } + 1
+            parent.statements.add(idx, stmt)
+            stmt.linkParents(parent as Node)
         }
     }
 
-    class InsertBefore(val before: Statement, val stmt: Statement, val parent: Node) : IAstModification {
+    class InsertBefore(private val before: Statement, private val stmt: Statement, private val parent: INameScope) : IAstModification {
         override fun perform() {
-            if(parent is INameScope) {
-                val idx = parent.statements.indexOfFirst { it===before }
-                parent.statements.add(idx, stmt)
-                stmt.linkParents(parent)
-            } else {
-                throw FatalAstException("parent of an insert modification is not an INameScope")
-            }
+            val idx = parent.statements.indexOfFirst { it===before }
+            parent.statements.add(idx, stmt)
+            stmt.linkParents(parent as Node)
         }
     }
 
-    class ReplaceNode(val node: Node, val replacement: Node, val parent: Node) : IAstModification {
+    class ReplaceNode(private val node: Node, private val replacement: Node, private val parent: Node) : IAstModification {
         override fun perform() {
             parent.replaceChildNode(node, replacement)
             replacement.linkParents(parent)
         }
     }
 
-    class SwapOperands(val expr: BinaryExpression): IAstModification {
+    class SwapOperands(private val expr: BinaryExpression): IAstModification {
         override fun perform() {
+            require(expr.operator in associativeOperators)
             val tmp = expr.left
             expr.left = expr.right
             expr.right = tmp
@@ -228,6 +209,7 @@ abstract class AstWalker {
         track(before(decl, parent), decl, parent)
         decl.value?.accept(this, decl)
         decl.arraysize?.accept(this, decl)
+        decl.struct?.accept(this, decl)
         track(after(decl, parent), decl, parent)
     }
 
@@ -348,7 +330,7 @@ abstract class AstWalker {
 
     fun visit(untilLoop: UntilLoop, parent: Node) {
         track(before(untilLoop, parent), untilLoop, parent)
-        untilLoop.untilCondition.accept(this, untilLoop)
+        untilLoop.condition.accept(this, untilLoop)
         untilLoop.body.accept(this, untilLoop)
         track(after(untilLoop, parent), untilLoop, parent)
     }
@@ -361,8 +343,8 @@ abstract class AstWalker {
 
     fun visit(arrayIndexedExpression: ArrayIndexedExpression, parent: Node) {
         track(before(arrayIndexedExpression, parent), arrayIndexedExpression, parent)
-        arrayIndexedExpression.identifier.accept(this, arrayIndexedExpression)
-        arrayIndexedExpression.arrayspec.accept(this, arrayIndexedExpression)
+        arrayIndexedExpression.arrayvar.accept(this, arrayIndexedExpression)
+        arrayIndexedExpression.indexer.accept(this, arrayIndexedExpression)
         track(after(arrayIndexedExpression, parent), arrayIndexedExpression, parent)
     }
 

compiler/src/prog8/ast/processing/IAstVisitor.kt

@@ -33,6 +33,7 @@ interface IAstVisitor {
     fun visit(decl: VarDecl) {
         decl.value?.accept(this)
         decl.arraysize?.accept(this)
+        decl.struct?.accept(this)
     }
 
     fun visit(subroutine: Subroutine) {
@@ -115,7 +116,7 @@ interface IAstVisitor {
     }
 
     fun visit(untilLoop: UntilLoop) {
-        untilLoop.untilCondition.accept(this)
+        untilLoop.condition.accept(this)
         untilLoop.body.accept(this)
     }
 
@@ -124,8 +125,8 @@ interface IAstVisitor {
     }
 
     fun visit(arrayIndexedExpression: ArrayIndexedExpression) {
-        arrayIndexedExpression.identifier.accept(this)
-        arrayIndexedExpression.arrayspec.accept(this)
+        arrayIndexedExpression.arrayvar.accept(this)
+        arrayIndexedExpression.indexer.accept(this)
     }
 
     fun visit(assignTarget: AssignTarget) {

compiler/src/prog8/ast/processing/ImportedModuleDirectiveRemover.kt

@@ -1,5 +1,6 @@
 package prog8.ast.processing
 
+import prog8.ast.INameScope
 import prog8.ast.Node
 import prog8.ast.statements.Directive
 
@@ -14,7 +15,7 @@ internal class ImportedModuleDirectiveRemover: AstWalker() {
 
     override fun before(directive: Directive, parent: Node): Iterable<IAstModification> {
         if(directive.directive in moduleLevelDirectives) {
-            return listOf(IAstModification.Remove(directive, parent))
+            return listOf(IAstModification.Remove(directive, parent as INameScope))
         }
         return noModifications
     }

compiler/src/prog8/ast/processing/LiteralsToAutoVars.kt

@@ -0,0 +1,53 @@
+package prog8.ast.processing
+
+import prog8.ast.Node
+import prog8.ast.Program
+import prog8.ast.base.*
+import prog8.ast.expressions.*
+import prog8.ast.statements.*
+
+
+internal class LiteralsToAutoVars(private val program: Program) : AstWalker() {
+    private val noModifications = emptyList<IAstModification>()
+
+    override fun after(string: StringLiteralValue, parent: Node): Iterable<IAstModification> {
+        if(string.parent !is VarDecl && string.parent !is WhenChoice) {
+            // replace the literal string by a identifier reference to a new local vardecl
+            val vardecl = VarDecl.createAuto(string)
+            val identifier = IdentifierReference(listOf(vardecl.name), vardecl.position)
+            return listOf(
+                    IAstModification.ReplaceNode(string, identifier, parent),
+                    IAstModification.InsertFirst(vardecl, string.definingScope())
+            )
+        }
+        return noModifications
+    }
+
+    override fun after(array: ArrayLiteralValue, parent: Node): Iterable<IAstModification> {
+        val vardecl = array.parent as? VarDecl
+        if(vardecl!=null) {
+            // adjust the datatype of the array (to an educated guess)
+            val arrayDt = array.type
+            if(!arrayDt.istype(vardecl.datatype)) {
+                val cast = array.cast(vardecl.datatype)
+                if (cast != null && cast !== array)
+                    return listOf(IAstModification.ReplaceNode(vardecl.value!!, cast, vardecl))
+            }
+        } else {
+            val arrayDt = array.guessDatatype(program)
+            if(arrayDt.isKnown) {
+                // this array literal is part of an expression, turn it into an identifier reference
+                val litval2 = array.cast(arrayDt.typeOrElse(DataType.STRUCT))
+                if(litval2!=null) {
+                    val vardecl2 = VarDecl.createAuto(litval2)
+                    val identifier = IdentifierReference(listOf(vardecl2.name), vardecl2.position)
+                    return listOf(
+                            IAstModification.ReplaceNode(array, identifier, parent),
+                            IAstModification.InsertFirst(vardecl2, array.definingScope())
+                    )
+                }
+            }
+        }
+        return noModifications
+    }
+}

compiler/src/prog8/ast/processing/StatementReorderer.kt

@@ -6,12 +6,12 @@ import prog8.ast.expressions.*
 import prog8.ast.statements.*
 
 
-internal class StatementReorderer(val program: Program) : AstWalker() {
+internal class StatementReorderer(val program: Program, val errors: ErrorReporter) : AstWalker() {
     // Reorders the statements in a way the compiler needs.
     // - 'main' block must be the very first statement UNLESS it has an address set.
     // - library blocks are put last.
     // - blocks are ordered by address, where blocks without address are placed last.
-    // - in every scope, most directives and vardecls are moved to the top.
+    // - in every block and module, most directives and vardecls are moved to the top. (not in subroutines!)
     // - the 'start' subroutine is moved to the top.
     // - (syntax desugaring) a vardecl with a non-const initializer value is split into a regular vardecl and an assignment statement.
     // - (syntax desugaring) struct value assignment is expanded into several struct member assignments.
@@ -71,22 +71,62 @@ internal class StatementReorderer(val program: Program) : AstWalker() {
         return noModifications
     }
 
-    override fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> {
-        val declValue = decl.value
-        if(declValue!=null && decl.type== VarDeclType.VAR && decl.datatype in NumericDatatypes) {
-            val declConstValue = declValue.constValue(program)
-            if(declConstValue==null) {
-                // move the vardecl (without value) to the scope and replace this with a regular assignment
-                decl.value = null
-                val target = AssignTarget(IdentifierReference(listOf(decl.name), decl.position), null, null, decl.position)
-                val assign = Assignment(target, declValue, decl.position)
-                return listOf(
-                        IAstModification.ReplaceNode(decl, assign, parent),
-                        IAstModification.InsertFirst(decl, decl.definingScope() as Node)
-                )
+    override fun after(arrayIndexedExpression: ArrayIndexedExpression, parent: Node): Iterable<IAstModification> {
+        when (val expr2 = arrayIndexedExpression.indexer.origExpression) {
+            is NumericLiteralValue -> {
+                arrayIndexedExpression.indexer.indexNum = expr2
+                arrayIndexedExpression.indexer.origExpression = null
+                return noModifications
             }
+            is IdentifierReference -> {
+                arrayIndexedExpression.indexer.indexVar = expr2
+                arrayIndexedExpression.indexer.origExpression = null
+                return noModifications
+            }
+            is Expression -> {
+                // replace complex indexing with a temp variable
+                return getAutoIndexerVarFor(arrayIndexedExpression)
+            }
+            else -> return noModifications
         }
-        return noModifications
+    }
+
+    private fun getAutoIndexerVarFor(expr: ArrayIndexedExpression): MutableList<IAstModification> {
+        val modifications = mutableListOf<IAstModification>()
+        val subroutine = expr.definingSubroutine()!!
+        val statement = expr.containingStatement()
+        val indexerVarPrefix = "prog8_autovar_index_"
+        val repo = subroutine.asmGenInfo.usedAutoArrayIndexerForStatements
+
+        // TODO make this even smarter so that an indexerVar can be reused for a different following statement...  requires updating the partOfStatement?
+        var indexerVar = repo.firstOrNull { it.replaces isSameAs expr.indexer }
+        if(indexerVar==null) {
+            // add another loop index var to be used for this expression
+            val indexerVarName = "$indexerVarPrefix${expr.indexer.hashCode()}"
+            indexerVar = AsmGenInfo.ArrayIndexerInfo(indexerVarName, expr.indexer, statement)
+            repo.add(indexerVar)
+            // create the indexer var at block level scope
+            val vardecl = VarDecl(VarDeclType.VAR, DataType.UBYTE, ZeropageWish.PREFER_ZEROPAGE,
+                    null, indexerVarName, null, null, isArray = false, autogeneratedDontRemove = true, position = expr.position)
+            modifications.add(IAstModification.InsertFirst(vardecl, subroutine))
+        }
+        indexerVar.used++       // keep track of how many times it it used, to avoid assigning it multiple times
+
+        // replace the indexer with just the variable
+        // assign the indexing expression to the helper variable, but only if that hasn't been done already
+        val indexerExpression = expr.indexer.origExpression!!
+        val target = AssignTarget(IdentifierReference(listOf(indexerVar.name), indexerExpression.position), null, null, indexerExpression.position)
+        if(indexerVar.used==1) {
+            val assign = Assignment(target, indexerExpression, indexerExpression.position)
+            modifications.add(IAstModification.InsertBefore(statement, assign, statement.definingScope()))
+        }
+        modifications.add(IAstModification.SetExpression( {
+            expr.indexer.indexVar = it as IdentifierReference
+            expr.indexer.indexNum = null
+            expr.indexer.origExpression = null
+        }, target.identifier!!.copy(), expr.indexer))
+
+        return modifications
     }
 
     override fun after(whenStatement: WhenStatement, parent: Node): Iterable<IAstModification> {
@@ -98,23 +138,57 @@ internal class StatementReorderer(val program: Program) : AstWalker() {
         return noModifications
     }
 
+    override fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> {
+        val declValue = decl.value
+        if(declValue!=null && decl.type== VarDeclType.VAR && decl.datatype in NumericDatatypes) {
+            val declConstValue = declValue.constValue(program)
+            if(declConstValue==null) {
+                // move the vardecl (without value) to the scope and replace this with a regular assignment
+                // Unless we're dealing with a floating point variable because that will actually make things less efficient at the moment (because floats are mostly calcualated via the stack)
+                if(decl.datatype!=DataType.FLOAT) {
+                    decl.value = null
+                    decl.allowInitializeWithZero = false
+                    val target = AssignTarget(IdentifierReference(listOf(decl.name), decl.position), null, null, decl.position)
+                    val assign = Assignment(target, declValue, decl.position)
+                    return listOf(
+                            IAstModification.ReplaceNode(decl, assign, parent),
+                            IAstModification.InsertFirst(decl, decl.definingScope())
+                    )
+                }
+            }
+        }
+        return noModifications
+    }
+
     override fun before(assignment: Assignment, parent: Node): Iterable<IAstModification> {
         val valueType = assignment.value.inferType(program)
-        val targetType = assignment.target.inferType(program, assignment)
-        if(valueType.istype(DataType.STRUCT) && targetType.istype(DataType.STRUCT)) {
-            val assignments = if (assignment.value is ArrayLiteralValue) {
-                flattenStructAssignmentFromStructLiteral(assignment, program)    //  'structvar = [ ..... ] '
+        val targetType = assignment.target.inferType(program)
+        var assignments = emptyList<Assignment>()
+
+        if(targetType.istype(DataType.STRUCT) && (valueType.istype(DataType.STRUCT) || valueType.typeOrElse(DataType.STRUCT) in ArrayDatatypes )) {
+            assignments = if (assignment.value is ArrayLiteralValue) {
+                flattenStructAssignmentFromStructLiteral(assignment)    //  'structvar = [ ..... ] '
             } else {
-                flattenStructAssignmentFromIdentifier(assignment, program)    //   'structvar1 = structvar2'
+                flattenStructAssignmentFromIdentifier(assignment)    //   'structvar1 = structvar2'
             }
-            if(assignments.isNotEmpty()) {
-                val modifications = mutableListOf<IAstModification>()
-                assignments.reversed().mapTo(modifications) { IAstModification.InsertAfter(assignment, it, parent) }
-                modifications.add(IAstModification.Remove(assignment, parent))
-                return modifications
+        }
+
+        if(targetType.typeOrElse(DataType.STRUCT) in ArrayDatatypes && valueType.typeOrElse(DataType.STRUCT) in ArrayDatatypes ) {
+            assignments = if (assignment.value is ArrayLiteralValue) {
+                flattenArrayAssignmentFromArrayLiteral(assignment)    //  'arrayvar = [ ..... ] '
+            } else {
+                flattenArrayAssignmentFromIdentifier(assignment)    //   'arrayvar1 = arrayvar2'
             }
         }
 
+        if(assignments.isNotEmpty()) {
+            val modifications = mutableListOf<IAstModification>()
+            val scope = assignment.definingScope()
+            assignments.reversed().mapTo(modifications) { IAstModification.InsertAfter(assignment, it, scope) }
+            modifications.add(IAstModification.Remove(assignment, scope))
+            return modifications
+        }
+
         return noModifications
     }
 
@@ -167,15 +241,55 @@ internal class StatementReorderer(val program: Program) : AstWalker() {
         return noModifications
     }
 
-    private fun flattenStructAssignmentFromStructLiteral(structAssignment: Assignment, program: Program): List<Assignment> {
+    private fun flattenArrayAssignmentFromArrayLiteral(assign: Assignment): List<Assignment> {
+        val identifier = assign.target.identifier!!
+        val targetVar = identifier.targetVarDecl(program.namespace)!!
+        val alv = assign.value as? ArrayLiteralValue
+        return flattenArrayAssign(targetVar, alv, identifier, assign.position)
+    }
+
+    private fun flattenArrayAssignmentFromIdentifier(assign: Assignment): List<Assignment> {
+        val identifier = assign.target.identifier!!
+        val targetVar = identifier.targetVarDecl(program.namespace)!!
+        val sourceIdent = assign.value as IdentifierReference
+        val sourceVar = sourceIdent.targetVarDecl(program.namespace)!!
+        if(!sourceVar.isArray) {
+            errors.err("value must be an array",  sourceIdent.position)
+            return emptyList()
+        }
+        val alv = sourceVar.value as? ArrayLiteralValue
+        return flattenArrayAssign(targetVar, alv, identifier, assign.position)
+    }
+
+    private fun flattenArrayAssign(targetVar: VarDecl, alv: ArrayLiteralValue?, identifier: IdentifierReference, position: Position): List<Assignment> {
+        if(targetVar.arraysize==null) {
+            errors.err("array has no defined size", identifier.position)
+            return emptyList()
+        }
+
+        if(alv==null || alv.value.size != targetVar.arraysize!!.constIndex()) {
+            errors.err("element count mismatch", position)
+            return emptyList()
+        }
+
+        // TODO use a pointer loop instead of individual assignments
+        return alv.value.mapIndexed { index, value ->
+            val idx = ArrayIndexedExpression(identifier, ArrayIndex(NumericLiteralValue(DataType.UBYTE, index, position), position), position)
+            Assignment(AssignTarget(null, idx, null, position), value, value.position)
+        }
+    }
+
+    private fun flattenStructAssignmentFromStructLiteral(structAssignment: Assignment): List<Assignment> {
         val identifier = structAssignment.target.identifier!!
         val identifierName = identifier.nameInSource.single()
         val targetVar = identifier.targetVarDecl(program.namespace)!!
         val struct = targetVar.struct!!
 
         val slv = structAssignment.value as? ArrayLiteralValue
-        if(slv==null || slv.value.size != struct.numberOfElements)
-            throw FatalAstException("element count mismatch")
+        if(slv==null || slv.value.size != struct.numberOfElements) {
+            errors.err("element count mismatch", structAssignment.position)
+            return emptyList()
+        }
 
         return struct.statements.zip(slv.value).map { (targetDecl, sourceValue) ->
             targetDecl as VarDecl
@@ -188,7 +302,8 @@ internal class StatementReorderer(val program: Program) : AstWalker() {
         }
     }
 
-    private fun flattenStructAssignmentFromIdentifier(structAssignment: Assignment, program: Program): List<Assignment> {
+    private fun flattenStructAssignmentFromIdentifier(structAssignment: Assignment): List<Assignment> {
+        // TODO use memcopy beyond a certain number of elements
         val identifier = structAssignment.target.identifier!!
         val identifierName = identifier.nameInSource.single()
         val targetVar = identifier.targetVarDecl(program.namespace)!!
@@ -196,26 +311,47 @@ internal class StatementReorderer(val program: Program) : AstWalker() {
         when (structAssignment.value) {
             is IdentifierReference -> {
                 val sourceVar = (structAssignment.value as IdentifierReference).targetVarDecl(program.namespace)!!
-                if (sourceVar.struct == null)
-                    throw FatalAstException("can only assign arrays or structs to structs")
-                // struct memberwise copy
-                val sourceStruct = sourceVar.struct!!
-                if(sourceStruct!==targetVar.struct) {
-                    // structs are not the same in assignment
-                    return listOf()     // error will be printed elsewhere
-                }
-                return struct.statements.zip(sourceStruct.statements).map { member ->
-                    val targetDecl = member.first as VarDecl
-                    val sourceDecl = member.second as VarDecl
-                    if(targetDecl.name != sourceDecl.name)
-                        throw FatalAstException("struct member mismatch")
-                    val mangled = mangledStructMemberName(identifierName, targetDecl.name)
-                    val idref = IdentifierReference(listOf(mangled), structAssignment.position)
-                    val sourcemangled = mangledStructMemberName(sourceVar.name, sourceDecl.name)
-                    val sourceIdref = IdentifierReference(listOf(sourcemangled), structAssignment.position)
-                    val assign = Assignment(AssignTarget(idref, null, null, structAssignment.position), sourceIdref, member.second.position)
-                    assign.linkParents(structAssignment)
-                    assign
+                when {
+                    sourceVar.struct!=null -> {
+                        // struct memberwise copy
+                        val sourceStruct = sourceVar.struct!!
+                        if(sourceStruct!==targetVar.struct) {
+                            // structs are not the same in assignment
+                            return listOf()     // error will be printed elsewhere
+                        }
+                        if(struct.statements.size!=sourceStruct.statements.size)
+                            return listOf()     // error will be printed elsewhere
+                        return struct.statements.zip(sourceStruct.statements).map { member ->
+                            val targetDecl = member.first as VarDecl
+                            val sourceDecl = member.second as VarDecl
+                            if(targetDecl.name != sourceDecl.name)
+                                throw FatalAstException("struct member mismatch")
+                            val mangled = mangledStructMemberName(identifierName, targetDecl.name)
+                            val idref = IdentifierReference(listOf(mangled), structAssignment.position)
+                            val sourcemangled = mangledStructMemberName(sourceVar.name, sourceDecl.name)
+                            val sourceIdref = IdentifierReference(listOf(sourcemangled), structAssignment.position)
+                            val assign = Assignment(AssignTarget(idref, null, null, structAssignment.position), sourceIdref, member.second.position)
+                            assign.linkParents(structAssignment)
+                            assign
+                        }
+                    }
+                    sourceVar.isArray -> {
+                        val array = (sourceVar.value as ArrayLiteralValue).value
+                        if(struct.statements.size!=array.size)
+                            return listOf()     // error will be printed elsewhere
+                        return struct.statements.zip(array).map {
+                            val decl = it.first as VarDecl
+                            val mangled = mangledStructMemberName(identifierName, decl.name)
+                            val targetName = IdentifierReference(listOf(mangled), structAssignment.position)
+                            val target = AssignTarget(targetName, null, null, structAssignment.position)
+                            val assign = Assignment(target, it.second, structAssignment.position)
+                            assign.linkParents(structAssignment)
+                            assign
+                        }
+                    }
+                    else -> {
+                        throw FatalAstException("can only assign arrays or structs to structs")
+                    }
                 }
             }
             is ArrayLiteralValue -> {

compiler/src/prog8/ast/processing/TypecastsAdder.kt

@@ -18,6 +18,26 @@ class TypecastsAdder(val program: Program, val errors: ErrorReporter) : AstWalke
 
     private val noModifications = emptyList<IAstModification>()
 
+    override fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> {
+        val declValue = decl.value
+        if(decl.type==VarDeclType.VAR && declValue!=null && decl.struct==null) {
+            val valueDt = declValue.inferType(program)
+            if(!valueDt.istype(decl.datatype)) {
+
+                // don't add a typecast on an array initializer value
+                if(valueDt.typeOrElse(DataType.STRUCT) in IntegerDatatypes && decl.datatype in ArrayDatatypes)
+                    return noModifications
+
+                return listOf(IAstModification.ReplaceNode(
+                        declValue,
+                        TypecastExpression(declValue, decl.datatype, true, declValue.position),
+                        decl
+                ))
+            }
+        }
+        return noModifications
+    }
+
     override fun after(expr: BinaryExpression, parent: Node): Iterable<IAstModification> {
         val leftDt = expr.left.inferType(program)
         val rightDt = expr.right.inferType(program)
@@ -40,12 +60,15 @@ class TypecastsAdder(val program: Program, val errors: ErrorReporter) : AstWalke
     override fun after(assignment: Assignment, parent: Node): Iterable<IAstModification> {
         // see if a typecast is needed to convert the value's type into the proper target type
         val valueItype = assignment.value.inferType(program)
-        val targetItype = assignment.target.inferType(program, assignment)
+        val targetItype = assignment.target.inferType(program)
         if(targetItype.isKnown && valueItype.isKnown) {
             val targettype = targetItype.typeOrElse(DataType.STRUCT)
             val valuetype = valueItype.typeOrElse(DataType.STRUCT)
             if (valuetype != targettype) {
                 if (valuetype isAssignableTo targettype) {
+                    if(valuetype in IterableDatatypes && targettype==DataType.UWORD)
+                        // special case, don't typecast STR/arrays to UWORD, we support those assignments "directly"
+                        return noModifications
                     return listOf(IAstModification.ReplaceNode(
                             assignment.value,
                             TypecastExpression(assignment.value, targettype, true, assignment.value.position),
@@ -96,28 +119,30 @@ class TypecastsAdder(val program: Program, val errors: ErrorReporter) : AstWalke
 
         when(val sub = call.target.targetStatement(scope)) {
             is Subroutine -> {
-                for(arg in sub.parameters.zip(call.args.withIndex())) {
-                    val argItype = arg.second.value.inferType(program)
+                sub.parameters.zip(call.args).forEachIndexed { index, pair ->
+                    val argItype = pair.second.inferType(program)
                     if(argItype.isKnown) {
                         val argtype = argItype.typeOrElse(DataType.STRUCT)
-                        val requiredType = arg.first.type
+                        val requiredType = pair.first.type
                         if (requiredType != argtype) {
                             if (argtype isAssignableTo requiredType) {
                                 modifications += IAstModification.ReplaceNode(
-                                        call.args[arg.second.index],
-                                        TypecastExpression(arg.second.value, requiredType, true, arg.second.value.position),
+                                        call.args[index],
+                                        TypecastExpression(pair.second, requiredType, true, pair.second.position),
                                         call as Node)
                             } else if(requiredType == DataType.UWORD && argtype in PassByReferenceDatatypes) {
                                 // we allow STR/ARRAY values in place of UWORD parameters. Take their address instead.
-                                modifications += IAstModification.ReplaceNode(
-                                        call.args[arg.second.index],
-                                        AddressOf(arg.second.value as IdentifierReference, arg.second.value.position),
-                                        call as Node)
-                            } else if(arg.second.value is NumericLiteralValue) {
-                                val cast = (arg.second.value as NumericLiteralValue).cast(requiredType)
+                                if(pair.second is IdentifierReference) {
+                                    modifications += IAstModification.ReplaceNode(
+                                            call.args[index],
+                                            AddressOf(pair.second as IdentifierReference, pair.second.position),
+                                            call as Node)
+                                }
+                            } else if(pair.second is NumericLiteralValue) {
+                                val cast = (pair.second as NumericLiteralValue).cast(requiredType)
                                 if(cast.isValid)
                                     modifications += IAstModification.ReplaceNode(
-                                            call.args[arg.second.index],
+                                            call.args[index],
                                             cast.valueOrZero(),
                                             call as Node)
                             }
@@ -127,18 +152,19 @@ class TypecastsAdder(val program: Program, val errors: ErrorReporter) : AstWalke
             }
             is BuiltinFunctionStatementPlaceholder -> {
                 val func = BuiltinFunctions.getValue(sub.name)
-                for (arg in func.parameters.zip(call.args.withIndex())) {
-                    val argItype = arg.second.value.inferType(program)
+                func.parameters.zip(call.args).forEachIndexed { index, pair ->
+                    val argItype = pair.second.inferType(program)
                     if (argItype.isKnown) {
                         val argtype = argItype.typeOrElse(DataType.STRUCT)
-                        if (arg.first.possibleDatatypes.any { argtype == it })
-                            continue
-                        for (possibleType in arg.first.possibleDatatypes) {
-                            if (argtype isAssignableTo possibleType) {
-                                modifications += IAstModification.ReplaceNode(
-                                        call.args[arg.second.index],
-                                        TypecastExpression(arg.second.value, possibleType, true, arg.second.value.position),
-                                        call as Node)
+                        if (pair.first.possibleDatatypes.all { argtype != it }) {
+                            for (possibleType in pair.first.possibleDatatypes) {
+                                if (argtype isAssignableTo possibleType) {
+                                    modifications += IAstModification.ReplaceNode(
+                                            call.args[index],
+                                            TypecastExpression(pair.second, possibleType, true, pair.second.position),
+                                            call as Node)
+                                    break
+                                }
                             }
                         }
                     }

compiler/src/prog8/ast/processing/VariousCleanups.kt

@@ -12,7 +12,7 @@ internal class VariousCleanups: AstWalker() {
     private val noModifications = emptyList<IAstModification>()
 
     override fun before(nopStatement: NopStatement, parent: Node): Iterable<IAstModification> {
-        return listOf(IAstModification.Remove(nopStatement, parent))
+        return listOf(IAstModification.Remove(nopStatement, parent as INameScope))
     }
 
     override fun before(scope: AnonymousScope, parent: Node): Iterable<IAstModification> {

compiler/src/prog8/ast/processing/VerifyFunctionArgTypes.kt

@@ -4,10 +4,10 @@ import prog8.ast.IFunctionCall
 import prog8.ast.INameScope
 import prog8.ast.Program
 import prog8.ast.base.DataType
+import prog8.ast.base.FatalAstException
+import prog8.ast.expressions.Expression
 import prog8.ast.expressions.FunctionCall
-import prog8.ast.statements.BuiltinFunctionStatementPlaceholder
-import prog8.ast.statements.FunctionCallStatement
-import prog8.ast.statements.Subroutine
+import prog8.ast.statements.*
 import prog8.compiler.CompilerException
 import prog8.functions.BuiltinFunctions
 
@@ -26,31 +26,58 @@ class VerifyFunctionArgTypes(val program: Program) : IAstVisitor {
     }
 
     companion object {
+
+        private fun argTypeCompatible(argDt: DataType, paramDt: DataType): Boolean {
+            if(argDt==paramDt)
+                return true
+
+            // there are some exceptions that are considered compatible, such as STR <> UWORD
+            if(argDt==DataType.STR && paramDt==DataType.UWORD ||
+                    argDt==DataType.UWORD && paramDt==DataType.STR)
+                return true
+
+            return false
+        }
+
         fun checkTypes(call: IFunctionCall, scope: INameScope, program: Program): String? {
-            val argtypes = call.args.map { it.inferType(program).typeOrElse(DataType.STRUCT) }
+            val argITypes = call.args.map { it.inferType(program) }
+            val firstUnknownDt = argITypes.indexOfFirst { it.isUnknown }
+            if(firstUnknownDt>=0)
+                return "argument ${firstUnknownDt+1} invalid argument type"
+            val argtypes = argITypes.map { it.typeOrElse(DataType.STRUCT) }
             val target = call.target.targetStatement(scope)
             if (target is Subroutine) {
-                // asmsub types are not checked specifically at this time
                 if(call.args.size != target.parameters.size)
                     return "invalid number of arguments"
                 val paramtypes = target.parameters.map { it.type }
-                val mismatch = argtypes.zip(paramtypes).indexOfFirst { it.first != it.second}
+                val mismatch = argtypes.zip(paramtypes).indexOfFirst { !argTypeCompatible(it.first, it.second) }
                 if(mismatch>=0) {
                     val actual = argtypes[mismatch].toString()
                     val expected = paramtypes[mismatch].toString()
                     return "argument ${mismatch + 1} type mismatch, was: $actual expected: $expected"
                 }
+                if(target.isAsmSubroutine) {
+                    if(target.asmReturnvaluesRegisters.size>1) {
+                        // multiple return values will NOT work inside an expression.
+                        // they MIGHT work in a regular assignment or just a function call statement.
+                        val parent = if(call is Statement) call.parent else if(call is Expression) call.parent else null
+                        if(call !is FunctionCallStatement && parent !is Assignment && parent !is VarDecl) {
+                            return "can't use subroutine call that returns multiple return values here (try moving it into a separate assignment)"
+                        }
+                    }
+                }
             }
             else if (target is BuiltinFunctionStatementPlaceholder) {
                 val func = BuiltinFunctions.getValue(target.name)
                 if(call.args.size != func.parameters.size)
                     return "invalid number of arguments"
                 val paramtypes = func.parameters.map { it.possibleDatatypes }
-                for (x in argtypes.zip(paramtypes).withIndex()) {
-                    if (x.value.first !in x.value.second) {
-                        val actual = x.value.first.toString()
-                        val expected = x.value.second.toString()
-                        return "argument ${x.index + 1} type mismatch, was: $actual expected: $expected"
+                argtypes.zip(paramtypes).forEachIndexed { index, pair ->
+                    val anyCompatible = pair.second.any { argTypeCompatible(pair.first, it) }
+                    if (!anyCompatible) {
+                        val actual = pair.first.toString()
+                        val expected = pair.second.toString()
+                        return "argument ${index + 1} type mismatch, was: $actual expected: $expected"
                     }
                 }
             }

compiler/src/prog8/ast/statements/AstStatements.kt

@@ -5,6 +5,8 @@ import prog8.ast.base.*
 import prog8.ast.expressions.*
 import prog8.ast.processing.AstWalker
 import prog8.ast.processing.IAstVisitor
+import prog8.compiler.CompilerException
+import prog8.compiler.target.CompilationTarget
 
 
 sealed class Statement : Node {
@@ -28,12 +30,6 @@ sealed class Statement : Node {
             scope.add(name)
         return scope.joinToString(".")
     }
-
-    fun definingBlock(): Block {
-        if(this is Block)
-            return this
-        return findParentNode<Block>(this)!!
-    }
 }
 
 
@@ -48,7 +44,7 @@ class BuiltinFunctionStatementPlaceholder(val name: String, override val positio
     }
 }
 
-data class RegisterOrStatusflag(val registerOrPair: RegisterOrPair?, val statusflag: Statusflag?, val stack: Boolean)
+data class RegisterOrStatusflag(val registerOrPair: RegisterOrPair?, val statusflag: Statusflag?)
 
 class Block(override val name: String,
             val address: Int?,
@@ -175,6 +171,7 @@ open class VarDecl(val type: VarDeclType,
         private set
     var structHasBeenFlattened = false      // set later
         private set
+    var allowInitializeWithZero = true
 
     // prefix for literal values that are turned into a variable on the heap
 
@@ -189,7 +186,12 @@ open class VarDecl(val type: VarDeclType,
 
         fun createAuto(array: ArrayLiteralValue): VarDecl {
             val autoVarName = "auto_heap_value_${++autoHeapValueSequenceNumber}"
-            val declaredType = ArrayElementTypes.getValue(array.type.typeOrElse(DataType.STRUCT))
+            val arrayDt =
+                if(!array.type.isKnown)
+                    throw FatalAstException("unknown dt")
+                else
+                    array.type.typeOrElse(DataType.STRUCT)
+            val declaredType = ArrayElementTypes.getValue(arrayDt)
             val arraysize = ArrayIndex.forArray(array)
             return VarDecl(VarDeclType.VAR, declaredType, ZeropageWish.NOT_IN_ZEROPAGE, arraysize, autoVarName, null, array,
                     isArray = true, autogeneratedDontRemove = true, position = array.position)
@@ -198,7 +200,7 @@ open class VarDecl(val type: VarDeclType,
         fun defaultZero(dt: DataType, position: Position) = when(dt) {
             DataType.UBYTE -> NumericLiteralValue(DataType.UBYTE, 0,  position)
             DataType.BYTE -> NumericLiteralValue(DataType.BYTE, 0,  position)
-            DataType.UWORD -> NumericLiteralValue(DataType.UWORD, 0, position)
+            DataType.UWORD, DataType.STR -> NumericLiteralValue(DataType.UWORD, 0, position)
             DataType.WORD -> NumericLiteralValue(DataType.WORD, 0, position)
             DataType.FLOAT -> NumericLiteralValue(DataType.FLOAT, 0.0, position)
             else -> throw FatalAstException("can only determine default zero value for a numeric type")
@@ -214,8 +216,9 @@ open class VarDecl(val type: VarDeclType,
                 DataType.UWORD -> DataType.ARRAY_UW
                 DataType.WORD -> DataType.ARRAY_W
                 DataType.FLOAT -> DataType.ARRAY_F
+                DataType.STR -> DataType.ARRAY_UW       // use memory address of the string instead
                 else -> {
-                    datatypeErrors.add(SyntaxError("array can only contain bytes/words/floats", position))
+                    datatypeErrors.add(SyntaxError("array can only contain bytes/words/floats/strings(ptrs)", position))
                     DataType.ARRAY_UB
                 }
             }
@@ -232,7 +235,8 @@ open class VarDecl(val type: VarDeclType,
     }
 
     override fun replaceChildNode(node: Node, replacement: Node) {
-        require(replacement is Expression && node===value)
+        // TODO the check that node===value is too strict sometimes, but leaving it out allows for bugs to creep through ... :( Perhaps check when adding the replace if there is already a replace on the same node?
+        require(replacement is Expression)
         value = replacement
         replacement.parent = this
     }
@@ -244,12 +248,17 @@ open class VarDecl(val type: VarDeclType,
         return "VarDecl(name=$name, vartype=$type, datatype=$datatype, struct=$structName, value=$value, pos=$position)"
     }
 
-    fun zeroElementValue() = defaultZero(declaredDatatype, position)
+    fun zeroElementValue(): NumericLiteralValue {
+        if(allowInitializeWithZero)
+            return defaultZero(declaredDatatype, position)
+        else
+            throw CompilerException("attempt to get zero value for vardecl that shouldn't get it")
+    }
 
     fun flattenStructMembers(): MutableList<Statement> {
-        val result = struct!!.statements.withIndex().map {
-            val member = it.value as VarDecl
-            val initvalue = if(value!=null) (value as ArrayLiteralValue).value[it.index] else null
+        val result = struct!!.statements.mapIndexed { index, statement ->
+            val member = statement as VarDecl
+            val initvalue = if(value!=null) (value as ArrayLiteralValue).value[index] else null
             VarDecl(
                     VarDeclType.VAR,
                     member.datatype,
@@ -273,34 +282,78 @@ class ParameterVarDecl(name: String, declaredDatatype: DataType, position: Posit
     : VarDecl(VarDeclType.VAR, declaredDatatype, ZeropageWish.DONTCARE, null, name, null, null, false, true, position)
 
 
-class ArrayIndex(var index: Expression, override val position: Position) : Node {
+class ArrayIndex(var origExpression: Expression?,            // will be replaced later by either the number or the identifier
+                 override val position: Position) : Node {
+    // for code simplicity, either indexed via a constant number or via a variable (no arbitrary expressions)
     override lateinit var parent: Node
+    var indexNum: NumericLiteralValue? = origExpression as? NumericLiteralValue
+    var indexVar: IdentifierReference? = origExpression as? IdentifierReference
+
+    init {
+        if(indexNum!=null || indexVar!=null)
+            origExpression = null
+    }
 
     override fun linkParents(parent: Node) {
         this.parent = parent
-        index.linkParents(this)
+        origExpression?.linkParents(this)
+        indexNum?.linkParents(this)
+        indexVar?.linkParents(this)
     }
 
     override fun replaceChildNode(node: Node, replacement: Node) {
-        require(replacement is Expression && node===index)
-        index = replacement
-        replacement.parent = this
+        require(replacement is Expression)
+        when {
+            node===origExpression -> origExpression = replacement
+            node===indexVar -> {
+                when (replacement) {
+                    is NumericLiteralValue -> {
+                        indexVar = null
+                        indexNum = replacement
+                    }
+                    is IdentifierReference -> {
+                        indexVar = replacement
+                        indexNum = null
+                    }
+                    else -> {
+                        throw FatalAstException("invalid replace")
+                    }
+                }
+            }
+            else -> throw FatalAstException("invalid replace")
+        }
     }
 
     companion object {
         fun forArray(v: ArrayLiteralValue): ArrayIndex {
-            return ArrayIndex(NumericLiteralValue.optimalNumeric(v.value.size, v.position), v.position)
+            val indexnum = NumericLiteralValue.optimalNumeric(v.value.size, v.position)
+            return ArrayIndex(indexnum, v.position)
         }
     }
 
-    fun accept(visitor: IAstVisitor) = index.accept(visitor)
-    fun accept(visitor: AstWalker, parent: Node) = index.accept(visitor, this)
-
-    override fun toString(): String {
-        return("ArrayIndex($index, pos=$position)")
+    fun accept(visitor: IAstVisitor) {
+        origExpression?.accept(visitor)
+        indexNum?.accept(visitor)
+        indexVar?.accept(visitor)
+    }
+    fun accept(visitor: AstWalker, parent: Node) {
+        origExpression?.accept(visitor, this)
+        indexNum?.accept(visitor, this)
+        indexVar?.accept(visitor, this)
     }
 
-    fun constIndex() = (index as? NumericLiteralValue)?.number?.toInt()
+    override fun toString(): String {
+        return("ArrayIndex($indexNum, $indexVar, pos=$position)")
+    }
+
+    fun constIndex() = indexNum?.number?.toInt()
+
+    infix fun isSameAs(other: ArrayIndex): Boolean {
+        return if(indexNum!=null || indexVar!=null)
+            indexNum==other.indexNum && indexVar == other.indexVar
+        else
+            other.origExpression!=null && origExpression!! isSameAs other.origExpression!!
+    }
 }
 
 open class Assignment(var target: AssignTarget, var value: Expression, override val position: Position) : Statement() {
@@ -391,8 +444,8 @@ data class AssignTarget(var identifier: IdentifierReference?,
 
     override fun replaceChildNode(node: Node, replacement: Node) {
         when {
-            node===identifier -> identifier = replacement as IdentifierReference
-            node===arrayindexed -> arrayindexed = replacement as ArrayIndexedExpression
+            node === identifier -> identifier = replacement as IdentifierReference
+            node === arrayindexed -> arrayindexed = replacement as ArrayIndexedExpression
             else -> throw FatalAstException("invalid replace")
         }
         replacement.parent = this
@@ -401,28 +454,17 @@ data class AssignTarget(var identifier: IdentifierReference?,
     fun accept(visitor: IAstVisitor) = visitor.visit(this)
     fun accept(visitor: AstWalker, parent: Node) = visitor.visit(this, parent)
 
-    companion object {
-        fun fromExpr(expr: Expression): AssignTarget {
-            return when (expr) {
-                is IdentifierReference -> AssignTarget(expr, null, null, expr.position)
-                is ArrayIndexedExpression -> AssignTarget(null, expr, null, expr.position)
-                is DirectMemoryRead -> AssignTarget(null, null, DirectMemoryWrite(expr.addressExpression, expr.position), expr.position)
-                else -> throw FatalAstException("invalid expression object $expr")
-            }
-        }
-    }
-
-    fun inferType(program: Program, stmt: Statement): InferredTypes.InferredType {
-        if(identifier!=null) {
-            val symbol = program.namespace.lookup(identifier!!.nameInSource, stmt) ?: return InferredTypes.unknown()
+    fun inferType(program: Program): InferredTypes.InferredType {
+        if (identifier != null) {
+            val symbol = program.namespace.lookup(identifier!!.nameInSource, this) ?: return InferredTypes.unknown()
             if (symbol is VarDecl) return InferredTypes.knownFor(symbol.datatype)
         }
 
-        if(arrayindexed!=null) {
+        if (arrayindexed != null) {
             return arrayindexed!!.inferType(program)
         }
 
-        if(memoryAddress!=null)
+        if (memoryAddress != null)
             return InferredTypes.knownFor(DataType.UBYTE)
 
         return InferredTypes.unknown()
@@ -430,69 +472,93 @@ data class AssignTarget(var identifier: IdentifierReference?,
 
     fun toExpression(): Expression {
         return when {
-            identifier!=null -> identifier!!
-            arrayindexed!=null -> arrayindexed!!
-            memoryAddress!=null -> DirectMemoryRead(memoryAddress.addressExpression, memoryAddress.position)
+            identifier != null -> identifier!!
+            arrayindexed != null -> arrayindexed!!
+            memoryAddress != null -> DirectMemoryRead(memoryAddress.addressExpression, memoryAddress.position)
             else -> throw FatalAstException("invalid assignmenttarget $this")
         }
     }
 
     infix fun isSameAs(value: Expression): Boolean {
         return when {
-            this.memoryAddress!=null -> {
+            memoryAddress != null -> {
                 // if the target is a memory write, and the value is a memory read, they're the same if the address matches
-                if(value is DirectMemoryRead)
+                if (value is DirectMemoryRead)
                     this.memoryAddress.addressExpression isSameAs value.addressExpression
                 else
                     false
             }
-            this.identifier!=null -> value is IdentifierReference && value.nameInSource==identifier!!.nameInSource
-            this.arrayindexed!=null -> value is ArrayIndexedExpression &&
-                    value.identifier.nameInSource==arrayindexed!!.identifier.nameInSource &&
-                    value.arrayspec.constIndex()!=null &&
-                    arrayindexed!!.arrayspec.constIndex()!=null &&
-                    value.arrayspec.constIndex()==arrayindexed!!.arrayspec.constIndex()
+            identifier != null -> value is IdentifierReference && value.nameInSource == identifier!!.nameInSource
+            arrayindexed != null -> {
+                if(value is ArrayIndexedExpression && value.arrayvar.nameInSource == arrayindexed!!.arrayvar.nameInSource)
+                    arrayindexed!!.indexer isSameAs value.indexer
+                else
+                    false
+            }
             else -> false
         }
     }
 
     fun isSameAs(other: AssignTarget, program: Program): Boolean {
-        if(this===other)
+        if (this === other)
             return true
-        if(this.identifier!=null && other.identifier!=null)
-            return this.identifier!!.nameInSource==other.identifier!!.nameInSource
-        if(this.memoryAddress!=null && other.memoryAddress!=null) {
+        if (this.identifier != null && other.identifier != null)
+            return this.identifier!!.nameInSource == other.identifier!!.nameInSource
+        if (this.memoryAddress != null && other.memoryAddress != null) {
             val addr1 = this.memoryAddress.addressExpression.constValue(program)
             val addr2 = other.memoryAddress.addressExpression.constValue(program)
-            return addr1!=null && addr2!=null && addr1==addr2
+            return addr1 != null && addr2 != null && addr1 == addr2
         }
-        if(this.arrayindexed!=null && other.arrayindexed!=null) {
-            if(this.arrayindexed!!.identifier.nameInSource == other.arrayindexed!!.identifier.nameInSource) {
-                val x1 = this.arrayindexed!!.arrayspec.index.constValue(program)
-                val x2 = other.arrayindexed!!.arrayspec.index.constValue(program)
-                return x1!=null && x2!=null && x1==x2
+        if (this.arrayindexed != null && other.arrayindexed != null) {
+            if (this.arrayindexed!!.arrayvar.nameInSource == other.arrayindexed!!.arrayvar.nameInSource) {
+                val x1 = this.arrayindexed!!.indexer.constIndex()
+                val x2 = other.arrayindexed!!.indexer.constIndex()
+                return x1 != null && x2 != null && x1 == x2
             }
         }
         return false
     }
 
-    fun isNotMemory(namespace: INameScope): Boolean {
-        if(this.memoryAddress!=null)
-            return false
-        if(this.arrayindexed!=null) {
-            val targetStmt = this.arrayindexed!!.identifier.targetVarDecl(namespace)
-            if(targetStmt!=null)
-                return targetStmt.type!= VarDeclType.MEMORY
+    fun isInRegularRAM(namespace: INameScope): Boolean {
+        when {
+            this.memoryAddress != null -> {
+                return when (this.memoryAddress.addressExpression) {
+                    is NumericLiteralValue -> {
+                        CompilationTarget.instance.machine.isRegularRAMaddress((this.memoryAddress.addressExpression as NumericLiteralValue).number.toInt())
+                    }
+                    is IdentifierReference -> {
+                        val decl = (this.memoryAddress.addressExpression as IdentifierReference).targetVarDecl(namespace)
+                        if ((decl?.type == VarDeclType.VAR || decl?.type == VarDeclType.CONST) && decl.value is NumericLiteralValue)
+                            CompilationTarget.instance.machine.isRegularRAMaddress((decl.value as NumericLiteralValue).number.toInt())
+                        else
+                            false
+                    }
+                    else -> false
+                }
+            }
+            this.arrayindexed != null -> {
+                val targetStmt = this.arrayindexed!!.arrayvar.targetVarDecl(namespace)
+                return if (targetStmt?.type == VarDeclType.MEMORY) {
+                    val addr = targetStmt.value as? NumericLiteralValue
+                    if (addr != null)
+                        CompilationTarget.instance.machine.isRegularRAMaddress(addr.number.toInt())
+                    else
+                        false
+                } else true
+            }
+            this.identifier != null -> {
+                val decl = this.identifier!!.targetVarDecl(namespace)!!
+                return if (decl.type == VarDeclType.MEMORY && decl.value is NumericLiteralValue)
+                    CompilationTarget.instance.machine.isRegularRAMaddress((decl.value as NumericLiteralValue).number.toInt())
+                else
+                    true
+            }
+            else -> return true
         }
-        if(this.identifier!=null) {
-            val targetStmt = this.identifier!!.targetVarDecl(namespace)
-            if(targetStmt!=null)
-                return targetStmt.type!= VarDeclType.MEMORY
-        }
-        return false
     }
 }
 
+
 class PostIncrDecr(var target: AssignTarget, val operator: String, override val position: Position) : Statement() {
     override lateinit var parent: Node
 
@@ -619,6 +685,22 @@ class NopStatement(override val position: Position): Statement() {
     override fun accept(visitor: AstWalker, parent: Node) = visitor.visit(this, parent)
 }
 
+
+class AsmGenInfo {
+    // This class contains various attributes that influence the assembly code generator.
+    // Conceptually it should be part of any INameScope.
+    // But because the resulting code only creates "real" scopes on a subroutine level,
+    // it's more consistent to only define these attributes on a Subroutine node.
+    var usedAutoArrayIndexerForStatements = mutableListOf<ArrayIndexerInfo>()
+    var usedRegsaveA = false
+    var usedRegsaveX = false
+    var usedRegsaveY = false
+    var usedFloatEvalResultVar1 = false
+    var usedFloatEvalResultVar2 = false
+
+    class ArrayIndexerInfo(val name: String, val replaces: ArrayIndex, val partOfStatement: Statement, var used: Int=0)
+}
+
 // the subroutine class covers both the normal user-defined subroutines,
 // and also the predefined/ROM/register-based subroutines.
 // (multiple return types can only occur for the latter type)
@@ -633,7 +715,24 @@ class Subroutine(override val name: String,
                  override var statements: MutableList<Statement>,
                  override val position: Position) : Statement(), INameScope {
 
+    constructor(name: String, parameters: List<SubroutineParameter>, returntypes: List<DataType>, statements: MutableList<Statement>, position: Position)
+            : this(name, parameters, returntypes, emptyList(), determineReturnRegisters(returntypes), emptySet(), null, false, statements, position)
+
+    companion object {
+        private fun determineReturnRegisters(returntypes: List<DataType>): List<RegisterOrStatusflag> {
+            // for non-asm subroutines, determine the return registers based on the type of the return value
+            return when(returntypes.singleOrNull()) {
+                in ByteDatatypes -> listOf(RegisterOrStatusflag(RegisterOrPair.A, null))
+                in WordDatatypes -> listOf(RegisterOrStatusflag(RegisterOrPair.AY, null))
+                DataType.FLOAT -> listOf(RegisterOrStatusflag(RegisterOrPair.AY, null))
+                null -> emptyList()
+                else -> listOf(RegisterOrStatusflag(RegisterOrPair.AY, null))
+            }
+        }
+    }
+
     override lateinit var parent: Node
+    val asmGenInfo = AsmGenInfo()
     val scopedname: String by lazy { makeScopedName(name) }
 
     override fun linkParents(parent: Node) {
@@ -657,6 +756,9 @@ class Subroutine(override val name: String,
     }
 
     fun regXasResult() = asmReturnvaluesRegisters.any { it.registerOrPair in setOf(RegisterOrPair.X, RegisterOrPair.AX, RegisterOrPair.XY) }
+    fun regXasParam() = asmParameterRegisters.any { it.registerOrPair in setOf(RegisterOrPair.X, RegisterOrPair.AX, RegisterOrPair.XY) }
+    fun shouldPreserveStatusRegisterAfterCall() = asmReturnvaluesRegisters.any { it.statusflag != null } || shouldSaveX()
+    fun shouldSaveX() = CpuRegister.X in asmClobbers || regXasResult() || regXasParam()
 
     fun amountOfRtsInAsm(): Int = statements
             .asSequence()
@@ -814,19 +916,19 @@ class RepeatLoop(var iterations: Expression?, var body: AnonymousScope, override
 }
 
 class UntilLoop(var body: AnonymousScope,
-                var untilCondition: Expression,
+                var condition: Expression,
                 override val position: Position) : Statement() {
     override lateinit var parent: Node
 
     override fun linkParents(parent: Node) {
         this.parent = parent
-        untilCondition.linkParents(this)
+        condition.linkParents(this)
         body.linkParents(this)
     }
 
     override fun replaceChildNode(node: Node, replacement: Node) {
         when {
-            node===untilCondition -> untilCondition = replacement as Expression
+            node===condition -> condition = replacement as Expression
             node===body -> body = replacement as AnonymousScope
             else -> throw FatalAstException("invalid replace")
         }
@@ -865,11 +967,11 @@ class WhenStatement(var condition: Expression,
             if(choice.values==null)
                 result.add(null to choice)
             else {
-                val values = choice.values!!.map { it.constValue(program)?.number?.toInt() }
-                if(values.contains(null))
-                    result.add(null to choice)
-                else
-                    result.add(values.filterNotNull() to choice)
+                val values = choice.values!!.map {
+                    val cv = it.constValue(program)
+                    cv?.number?.toInt() ?: it.hashCode()       // the hashcode is a nonsensical number but it avoids weird AST validation errors later
+                }
+                result.add(values to choice)
             }
         }
         return result
@@ -891,9 +993,15 @@ class WhenChoice(var values: List<Expression>?,           // if null,  this is t
     }
 
     override fun replaceChildNode(node: Node, replacement: Node) {
-        require(replacement is AnonymousScope && node===statements)
-        statements = replacement
-        replacement.parent = this
+        val choiceValues = values
+        if(replacement is AnonymousScope && node===statements) {
+            statements = replacement
+            replacement.parent = this
+        } else if(choiceValues!=null && node in choiceValues) {
+            throw FatalAstException("cannot replace choice values")
+        } else {
+            throw FatalAstException("invalid replacement")
+        }
     }
 
     override fun toString(): String {

compiler/src/prog8/compiler/BeforeAsmGenerationAstChanger.kt

@@ -1,5 +1,6 @@
 package prog8.compiler
 
+import prog8.ast.IFunctionCall
 import prog8.ast.Node
 import prog8.ast.Program
 import prog8.ast.base.*
@@ -14,9 +15,18 @@ internal class BeforeAsmGenerationAstChanger(val program: Program, val errors: E
     private val noModifications = emptyList<IAstModification>()
 
     override fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> {
-        if (decl.value == null && decl.type == VarDeclType.VAR && decl.datatype in NumericDatatypes) {
-            // a numeric vardecl without an initial value is initialized with zero.
-            decl.value = decl.zeroElementValue()
+        subroutineVariables.add(decl.name to decl)
+        if (decl.value == null && !decl.autogeneratedDontRemove && decl.type == VarDeclType.VAR && decl.datatype in NumericDatatypes) {
+            // a numeric vardecl without an initial value is initialized with zero,
+            // unless there's already an assignment below, that initializes the value
+            if(decl.allowInitializeWithZero)
+            {
+                val nextAssign = decl.definingScope().nextSibling(decl) as? Assignment
+                if (nextAssign != null && nextAssign.target.isSameAs(IdentifierReference(listOf(decl.name), Position.DUMMY)))
+                    decl.value = null
+                else
+                    decl.value = decl.zeroElementValue()
+            }
         }
         return noModifications
     }
@@ -24,52 +34,78 @@ internal class BeforeAsmGenerationAstChanger(val program: Program, val errors: E
     override fun after(assignment: Assignment, parent: Node): Iterable<IAstModification> {
         // Try to replace A = B <operator> Something  by A= B, A = A <operator> Something
         // this triggers the more efficent augmented assignment code generation more often.
+        // But it can only be done if the target variable IS NOT OCCURRING AS AN OPERAND ITSELF.
         if(!assignment.isAugmentable
                 && assignment.target.identifier != null
-                && assignment.target.isNotMemory(program.namespace)) {
+                && assignment.target.isInRegularRAM(program.namespace)) {
             val binExpr = assignment.value as? BinaryExpression
-            if(binExpr!=null && binExpr.operator !in comparisonOperators) {
-                if(binExpr.left !is BinaryExpression) {
-                    val assignLeft = Assignment(assignment.target, binExpr.left, assignment.position)
-                    return listOf(
-                            IAstModification.InsertBefore(assignment, assignLeft, parent),
-                            IAstModification.ReplaceNode(binExpr.left, assignment.target.toExpression(), binExpr))
+            if (binExpr != null && binExpr.operator !in comparisonOperators) {
+                if (binExpr.left !is BinaryExpression) {
+                    if (binExpr.right.referencesIdentifier(*assignment.target.identifier!!.nameInSource.toTypedArray())) {
+                        // the right part of the expression contains the target variable itself.
+                        // we can't 'split' it trivially because the variable will be changed halfway through.
+                        if(binExpr.operator in associativeOperators) {
+                            // A = <something-without-A>  <associativeoperator>  <otherthing-with-A>
+                            // use the other part of the expression to split.
+                            val assignRight = Assignment(assignment.target, binExpr.right, assignment.position)
+                            return listOf(
+                                    IAstModification.InsertBefore(assignment, assignRight, assignment.definingScope()),
+                                    IAstModification.ReplaceNode(binExpr.right, binExpr.left, binExpr),
+                                    IAstModification.ReplaceNode(binExpr.left, assignment.target.toExpression(), binExpr))
+                        }
+                    } else {
+                        val assignLeft = Assignment(assignment.target, binExpr.left, assignment.position)
+                        return listOf(
+                                IAstModification.InsertBefore(assignment, assignLeft, assignment.definingScope()),
+                                IAstModification.ReplaceNode(binExpr.left, assignment.target.toExpression(), binExpr))
+                    }
                 }
             }
         }
         return noModifications
     }
 
+    private val subroutineVariables = mutableListOf<Pair<String, VarDecl>>()
+
+    override fun before(subroutine: Subroutine, parent: Node): Iterable<IAstModification> {
+        subroutineVariables.clear()
+        return noModifications
+    }
+
     override fun after(scope: AnonymousScope, parent: Node): Iterable<IAstModification> {
         val decls = scope.statements.filterIsInstance<VarDecl>()
+        subroutineVariables.addAll(decls.map { it.name to it })
+
         val sub = scope.definingSubroutine()
         if (sub != null) {
-            val existingVariables = sub.statements.filterIsInstance<VarDecl>().associateBy { it.name }
-            var conflicts = false
-            decls.forEach {
-                val existing = existingVariables[it.name]
-                if (existing != null) {
-                    errors.err("variable ${it.name} already defined in subroutine ${sub.name} at ${existing.position}", it.position)
-                    conflicts = true
-                }
-            }
-            if (!conflicts) {
-                val numericVarsWithValue = decls.filter { it.value != null && it.datatype in NumericDatatypes }
-                return numericVarsWithValue.map {
-                    val initValue = it.value!!  // assume here that value has always been set by now
-                    it.value = null     // make sure no value init assignment for this vardecl will be created later (would be superfluous)
-                    val target = AssignTarget(IdentifierReference(listOf(it.name), it.position), null, null, it.position)
-                    val assign = Assignment(target, initValue, it.position)
-                    initValue.parent = assign
-                    IAstModification.InsertFirst(assign, scope)
-                } +  decls.map { IAstModification.ReplaceNode(it, NopStatement(it.position), scope) } +
-                     decls.map { IAstModification.InsertFirst(it, sub) }    // move it up to the subroutine
+            // move vardecls of the scope into the upper scope. Make sure the position remains the same!
+            val numericVarsWithValue = decls.filter { it.value != null && it.datatype in NumericDatatypes }
+            val replaceVardecls =numericVarsWithValue.map {
+                val initValue = it.value!!  // assume here that value has always been set by now
+                it.value = null     // make sure no value init assignment for this vardecl will be created later (would be superfluous)
+                val target = AssignTarget(IdentifierReference(listOf(it.name), it.position), null, null, it.position)
+                val assign = Assignment(target, initValue, it.position)
+                initValue.parent = assign
+                IAstModification.ReplaceNode(it, assign, scope)
             }
+            val moveVardeclsUp = decls.map { IAstModification.InsertFirst(it, sub) }
+            return replaceVardecls + moveVardeclsUp
         }
         return noModifications
     }
 
     override fun after(subroutine: Subroutine, parent: Node): Iterable<IAstModification> {
+        val firstDeclarations = mutableMapOf<String, VarDecl>()
+        for(decl in subroutineVariables) {
+            val existing = firstDeclarations[decl.first]
+            if(existing!=null && existing !== decl.second) {
+                errors.err("variable ${decl.first} already defined in subroutine ${subroutine.name} at ${existing.position}", decl.second.position)
+            } else {
+                firstDeclarations[decl.first] = decl.second
+            }
+        }
+
+
         // add the implicit return statement at the end (if it's not there yet), but only if it's not a kernel routine.
         // and if an assembly block doesn't contain a rts/rti, and some other situations.
         val mods = mutableListOf<IAstModification>()
@@ -91,9 +127,8 @@ internal class BeforeAsmGenerationAstChanger(val program: Program, val errors: E
                 && outerStatements[subroutineStmtIdx - 1] !is Subroutine
                 && outerStatements[subroutineStmtIdx - 1] !is Return
                 && outerScope !is Block) {
-            mods += IAstModification.InsertAfter(outerStatements[subroutineStmtIdx - 1], returnStmt, outerScope as Node)
+            mods += IAstModification.InsertAfter(outerStatements[subroutineStmtIdx - 1], returnStmt, outerScope)
         }
-
         return mods
     }
 
@@ -114,14 +149,34 @@ internal class BeforeAsmGenerationAstChanger(val program: Program, val errors: E
         // that the types of assignment values and their target are the same,
         // and that the types of both operands of a binaryexpression node are the same.
         // So, it is not easily possible to remove the typecasts that are there to make these conditions true.
+        // The only place for now where we can do this is for:
+        //    asmsub register pair parameter.
+
+        if(typecast.type in WordDatatypes) {
+            val fcall = typecast.parent as? IFunctionCall
+            if (fcall != null) {
+                val sub = fcall.target.targetStatement(program.namespace) as? Subroutine
+                if (sub != null && sub.isAsmSubroutine) {
+                    return listOf(IAstModification.ReplaceNode(typecast, typecast.expression, parent))
+                }
+            }
+        }
 
         if(sourceDt in PassByReferenceDatatypes) {
             if(typecast.type==DataType.UWORD) {
-                return listOf(IAstModification.ReplaceNode(
-                        typecast,
-                        AddressOf(typecast.expression as IdentifierReference, typecast.position),
-                        parent
-                ))
+                if(typecast.expression is IdentifierReference) {
+                    return listOf(IAstModification.ReplaceNode(
+                            typecast,
+                            AddressOf(typecast.expression as IdentifierReference, typecast.position),
+                            parent
+                    ))
+                } else if(typecast.expression is IFunctionCall) {
+                    return listOf(IAstModification.ReplaceNode(
+                            typecast,
+                            typecast.expression,
+                            parent
+                    ))
+                }
             } else {
                 errors.err("cannot cast pass-by-reference value to type ${typecast.type} (only to UWORD)", typecast.position)
             }
@@ -129,4 +184,34 @@ internal class BeforeAsmGenerationAstChanger(val program: Program, val errors: E
 
         return noModifications
     }
+
+    override fun after(ifStatement: IfStatement, parent: Node): Iterable<IAstModification> {
+        val binExpr = ifStatement.condition as? BinaryExpression
+        if(binExpr==null || binExpr.operator !in comparisonOperators) {
+            // if x  ->  if x!=0,    if x+5  ->  if x+5 != 0
+            val booleanExpr = BinaryExpression(ifStatement.condition, "!=", NumericLiteralValue.optimalInteger(0, ifStatement.condition.position), ifStatement.condition.position)
+            return listOf(IAstModification.ReplaceNode(ifStatement.condition, booleanExpr, ifStatement))
+        }
+        return noModifications
+    }
+
+    override fun after(untilLoop: UntilLoop, parent: Node): Iterable<IAstModification> {
+        val binExpr = untilLoop.condition as? BinaryExpression
+        if(binExpr==null || binExpr.operator !in comparisonOperators) {
+            // until x  ->  until x!=0,    until x+5  ->  until x+5 != 0
+            val booleanExpr = BinaryExpression(untilLoop.condition, "!=", NumericLiteralValue.optimalInteger(0, untilLoop.condition.position), untilLoop.condition.position)
+            return listOf(IAstModification.ReplaceNode(untilLoop.condition, booleanExpr, untilLoop))
+        }
+        return noModifications
+    }
+
+    override fun after(whileLoop: WhileLoop, parent: Node): Iterable<IAstModification> {
+        val binExpr = whileLoop.condition as? BinaryExpression
+        if(binExpr==null || binExpr.operator !in comparisonOperators) {
+            // while x  ->  while x!=0,    while x+5  ->  while x+5 != 0
+            val booleanExpr = BinaryExpression(whileLoop.condition, "!=", NumericLiteralValue.optimalInteger(0, whileLoop.condition.position), whileLoop.condition.position)
+            return listOf(IAstModification.ReplaceNode(whileLoop.condition, booleanExpr, whileLoop))
+        }
+        return noModifications
+    }
 }

compiler/src/prog8/compiler/Compiler.kt

@@ -27,7 +27,10 @@ data class CompilationOptions(val output: OutputType,
                               val launcher: LauncherType,
                               val zeropage: ZeropageType,
                               val zpReserved: List<IntRange>,
-                              val floats: Boolean)
+                              val floats: Boolean,
+                              val noSysInit: Boolean) {
+    var slowCodegenWarnings = false
+}
 
 
 class CompilerException(message: String?) : Exception(message)

compiler/src/prog8/compiler/Main.kt

@@ -4,7 +4,10 @@ import prog8.ast.AstToSourceCode
 import prog8.ast.Program
 import prog8.ast.base.*
 import prog8.ast.statements.Directive
+import prog8.compiler.target.C64Target
 import prog8.compiler.target.CompilationTarget
+import prog8.compiler.target.Cx16Target
+import prog8.optimizer.*
 import prog8.optimizer.UnusedCodeRemover
 import prog8.optimizer.constantFold
 import prog8.optimizer.optimizeStatements
@@ -13,6 +16,7 @@ import prog8.parser.ModuleImporter
 import prog8.parser.ParsingFailedError
 import prog8.parser.moduleName
 import java.nio.file.Path
+import kotlin.system.exitProcess
 import kotlin.system.measureTimeMillis
 
 
@@ -25,16 +29,28 @@ class CompilationResult(val success: Boolean,
 fun compileProgram(filepath: Path,
                    optimize: Boolean,
                    writeAssembly: Boolean,
+                   slowCodegenWarnings: Boolean,
+                   compilationTarget: String,
                    outputDir: Path): CompilationResult {
     var programName = ""
     lateinit var programAst: Program
     lateinit var importedFiles: List<Path>
     val errors = ErrorReporter()
 
+    when(compilationTarget) {
+        C64Target.name -> CompilationTarget.instance = C64Target
+        Cx16Target.name -> CompilationTarget.instance = Cx16Target
+        else -> {
+            System.err.println("invalid compilation target")
+            exitProcess(1)
+        }
+    }
+
     try {
         val totalTime = measureTimeMillis {
             // import main module and everything it needs
             val (ast, compilationOptions, imported) = parseImports(filepath, errors)
+            compilationOptions.slowCodegenWarnings = slowCodegenWarnings
             programAst = ast
             importedFiles = imported
             processAst(programAst, errors, compilationOptions)
@@ -78,7 +94,7 @@ fun compileProgram(filepath: Path,
 }
 
 private fun parseImports(filepath: Path, errors: ErrorReporter): Triple<Program, CompilationOptions, List<Path>> {
-    println("Parsing...")
+    println("Compiler target: ${CompilationTarget.instance.name}. Parsing...")
     val importer = ModuleImporter()
     val programAst = Program(moduleName(filepath.fileName), mutableListOf())
     importer.importModule(programAst, filepath)
@@ -90,12 +106,12 @@ private fun parseImports(filepath: Path, errors: ErrorReporter): Triple<Program,
     if (compilerOptions.launcher == LauncherType.BASIC && compilerOptions.output != OutputType.PRG)
         throw ParsingFailedError("${programAst.modules.first().position} BASIC launcher requires output type PRG.")
 
-    // depending on the mach9ine and compiler options we may have to include some libraries
-    CompilationTarget.machine.importLibs(compilerOptions, importer, programAst)
+    // depending on the machine and compiler options we may have to include some libraries
+    CompilationTarget.instance.machine.importLibs(compilerOptions, importer, programAst)
 
-    // always import prog8lib and math
+    // always import prog8_lib and math
     importer.importLibraryModule(programAst, "math")
-    importer.importLibraryModule(programAst, "prog8lib")
+    importer.importLibraryModule(programAst, "prog8_lib")
     errors.handle()
     return Triple(programAst, compilerOptions, importedFiles)
 }
@@ -106,13 +122,12 @@ private fun determineCompilationOptions(program: Program): CompilationOptions {
             as? Directive)?.args?.single()?.name?.toUpperCase()
     val launcherType = (mainModule.statements.singleOrNull { it is Directive && it.directive == "%launcher" }
             as? Directive)?.args?.single()?.name?.toUpperCase()
-    mainModule.loadAddress = (mainModule.statements.singleOrNull { it is Directive && it.directive == "%address" }
-            as? Directive)?.args?.single()?.int ?: 0
     val zpoption: String? = (mainModule.statements.singleOrNull { it is Directive && it.directive == "%zeropage" }
             as? Directive)?.args?.single()?.name?.toUpperCase()
     val allOptions = program.modules.flatMap { it.statements }.filter { it is Directive && it.directive == "%option" }.flatMap { (it as Directive).args }.toSet()
     val floatsEnabled = allOptions.any { it.name == "enable_floats" }
-    val zpType: ZeropageType =
+    val noSysInit = allOptions.any { it.name == "no_sysinit" }
+    var zpType: ZeropageType =
             if (zpoption == null)
                 if(floatsEnabled) ZeropageType.FLOATSAFE else ZeropageType.KERNALSAFE
             else
@@ -122,6 +137,12 @@ private fun determineCompilationOptions(program: Program): CompilationOptions {
                     ZeropageType.KERNALSAFE
                     // error will be printed by the astchecker
                 }
+
+    if (zpType==ZeropageType.FLOATSAFE && CompilationTarget.instance.name == Cx16Target.name) {
+        System.err.println("Warning: Cx16 target must use zp option basicsafe instead of floatsafe")
+        zpType = ZeropageType.BASICSAFE
+    }
+
     val zpReserved = mainModule.statements
             .asSequence()
             .filter { it is Directive && it.directive == "%zpreserved" }
@@ -129,21 +150,31 @@ private fun determineCompilationOptions(program: Program): CompilationOptions {
             .map { it[0].int!!..it[1].int!! }
             .toList()
 
+    if(outputType!=null && !OutputType.values().any {it.name==outputType}) {
+        System.err.println("invalid output type $outputType")
+        exitProcess(1)
+    }
+    if(launcherType!=null && !LauncherType.values().any {it.name==launcherType}) {
+        System.err.println("invalid launcher type $launcherType")
+        exitProcess(1)
+    }
+
     return CompilationOptions(
             if (outputType == null) OutputType.PRG else OutputType.valueOf(outputType),
             if (launcherType == null) LauncherType.BASIC else LauncherType.valueOf(launcherType),
-            zpType, zpReserved, floatsEnabled
+            zpType, zpReserved, floatsEnabled, noSysInit
     )
 }
 
 private fun processAst(programAst: Program, errors: ErrorReporter, compilerOptions: CompilationOptions) {
     // perform initial syntax checks and processings
-    println("Processing...")
+    println("Processing for target ${CompilationTarget.instance.name}...")
     programAst.checkIdentifiers(errors)
     errors.handle()
     programAst.constantFold(errors)
     errors.handle()
-    programAst.reorderStatements()
+    programAst.reorderStatements(errors)
+    errors.handle()
     programAst.addTypecasts(errors)
     errors.handle()
     programAst.variousCleanups()
@@ -159,14 +190,15 @@ private fun optimizeAst(programAst: Program, errors: ErrorReporter) {
     while (true) {
         // keep optimizing expressions and statements until no more steps remain
         val optsDone1 = programAst.simplifyExpressions()
-        val optsDone2 = programAst.optimizeStatements(errors)
-        programAst.constantFold(errors) // because simplified statements and expressions could give rise to more constants that can be folded away:
+        val optsDone2 = programAst.splitBinaryExpressions()
+        val optsDone3 = programAst.optimizeStatements(errors)
+        programAst.constantFold(errors) // because simplified statements and expressions can result in more constants that can be folded away
         errors.handle()
-        if (optsDone1 + optsDone2 == 0)
+        if (optsDone1 + optsDone2 + optsDone3 == 0)
             break
     }
 
-    val remover = UnusedCodeRemover(errors)
+    val remover = UnusedCodeRemover(programAst, errors)
     remover.visit(programAst)
     remover.applyModifications()
     errors.handle()
@@ -178,9 +210,11 @@ private fun postprocessAst(programAst: Program, errors: ErrorReporter, compilerO
     programAst.variousCleanups()
     programAst.checkValid(compilerOptions, errors)          // check if final tree is still valid
     errors.handle()
-    programAst.checkRecursion(errors)         // check if there are recursive subroutine calls
+    val callGraph = CallGraph(programAst)
+    callGraph.checkRecursiveCalls(errors)
     errors.handle()
     programAst.verifyFunctionArgTypes()
+    programAst.moveMainAndStartToFirst()
 }
 
 private fun writeAssembly(programAst: Program, errors: ErrorReporter, outputDir: Path,
@@ -191,11 +225,11 @@ private fun writeAssembly(programAst: Program, errors: ErrorReporter, outputDir:
 
     // printAst(programAst)
 
-    CompilationTarget.machine.initializeZeropage(compilerOptions)
-    val assembly = CompilationTarget.asmGenerator(
+    CompilationTarget.instance.machine.initializeZeropage(compilerOptions)
+    val assembly = CompilationTarget.instance.asmGenerator(
             programAst,
             errors,
-            CompilationTarget.machine.zeropage,
+            CompilationTarget.instance.machine.zeropage,
             compilerOptions,
             outputDir).compileToAssembly(optimize)
     assembly.assemble(compilerOptions)

compiler/src/prog8/compiler/Zeropage.kt

@@ -10,7 +10,6 @@ abstract class Zeropage(protected val options: CompilationOptions) {
 
     abstract val SCRATCH_B1 : Int      // temp storage for a single byte
     abstract val SCRATCH_REG : Int     // temp storage for a register
-    abstract val SCRATCH_REG_X : Int   // temp storage for register X (the evaluation stack pointer)
     abstract val SCRATCH_W1 : Int      // temp storage 1 for a word  $fb+$fc
     abstract val SCRATCH_W2 : Int      // temp storage 2 for a word  $fb+$fc
 
@@ -65,18 +64,10 @@ abstract class Zeropage(protected val options: CompilationOptions) {
 
     private fun makeAllocation(address: Int, size: Int, datatype: DataType, name: String?): Int {
         free.removeAll(address until address+size)
-        allocations[address] = Pair(name ?: "<unnamed>", datatype)
+        allocations[address] = (name ?: "<unnamed>") to datatype
         return address
     }
 
     private fun loneByte(address: Int) = address in free && address-1 !in free && address+1 !in free
     private fun sequentialFree(address: Int, size: Int) = free.containsAll((address until address+size).toList())
-
-    enum class ExitProgramStrategy {
-        CLEAN_EXIT,
-        SYSTEM_RESET
-    }
-
-    abstract val exitProgramStrategy: ExitProgramStrategy
-
 }

compiler/src/prog8/compiler/target/CompilationTarget.kt

@@ -4,15 +4,44 @@ import prog8.ast.Program
 import prog8.ast.base.ErrorReporter
 import prog8.compiler.CompilationOptions
 import prog8.compiler.Zeropage
+import prog8.compiler.target.c64.C64MachineDefinition
+import prog8.compiler.target.c64.Petscii
+import prog8.compiler.target.c64.codegen.AsmGen
+import prog8.compiler.target.cx16.CX16MachineDefinition
 import java.nio.file.Path
 
 
 internal interface CompilationTarget {
+    val name: String
+    val machine: IMachineDefinition
+    fun encodeString(str: String, altEncoding: Boolean): List<Short>
+    fun decodeString(bytes: List<Short>, altEncoding: Boolean): String
+    fun asmGenerator(program: Program, errors: ErrorReporter, zp: Zeropage, options: CompilationOptions, path: Path): IAssemblyGenerator
+
     companion object {
-        lateinit var name: String
-        lateinit var machine: IMachineDefinition
-        lateinit var encodeString: (str: String, altEncoding: Boolean) -> List<Short>
-        lateinit var decodeString: (bytes: List<Short>, altEncoding: Boolean) -> String
-        lateinit var asmGenerator: (Program, ErrorReporter, Zeropage, CompilationOptions, Path) -> IAssemblyGenerator
+        lateinit var instance: CompilationTarget
     }
 }
+
+
+internal object C64Target: CompilationTarget {
+    override val name = "c64"
+    override val machine = C64MachineDefinition
+    override fun encodeString(str: String, altEncoding: Boolean) =
+            if(altEncoding) Petscii.encodeScreencode(str, true) else Petscii.encodePetscii(str, true)
+    override fun decodeString(bytes: List<Short>, altEncoding: Boolean) =
+            if(altEncoding) Petscii.decodeScreencode(bytes, true) else Petscii.decodePetscii(bytes, true)
+    override fun asmGenerator(program: Program, errors: ErrorReporter, zp: Zeropage, options: CompilationOptions, path: Path) =
+            AsmGen(program, errors, zp, options, path)
+}
+
+internal object Cx16Target: CompilationTarget {
+    override val name = "cx16"
+    override val machine = CX16MachineDefinition
+    override fun encodeString(str: String, altEncoding: Boolean) =
+            if(altEncoding) Petscii.encodeScreencode(str, true) else Petscii.encodePetscii(str, true)
+    override fun decodeString(bytes: List<Short>, altEncoding: Boolean) =
+            if(altEncoding) Petscii.decodeScreencode(bytes, true) else Petscii.decodePetscii(bytes, true)
+    override fun asmGenerator(program: Program, errors: ErrorReporter, zp: Zeropage, options: CompilationOptions, path: Path) =
+            AsmGen(program, errors, zp, options, path)
+}

compiler/src/prog8/compiler/target/IAssemblyGenerator.kt

@@ -7,6 +7,8 @@ internal interface IAssemblyGenerator {
 }
 
 internal const val generatedLabelPrefix = "_prog8_label_"
+internal const val subroutineFloatEvalResultVar1 = "_prog8_float_eval_result1"
+internal const val subroutineFloatEvalResultVar2 = "_prog8_float_eval_result2"
 
 internal interface IAssemblyProgram {
     val name: String

compiler/src/prog8/compiler/target/IMachineDefinition.kt

@@ -11,6 +11,11 @@ internal interface IMachineFloat {
     fun makeFloatFillAsm(): String
 }
 
+internal enum class CpuType {
+    CPU6502,
+    CPU65c02
+}
+
 internal interface IMachineDefinition {
     val FLOAT_MAX_NEGATIVE: Double
     val FLOAT_MAX_POSITIVE: Double
@@ -23,12 +28,12 @@ internal interface IMachineDefinition {
 
     val opcodeNames: Set<String>
     var zeropage: Zeropage
-    val initSystemProcname: String
-    val cpu: String
+    val cpu: CpuType
 
     fun initializeZeropage(compilerOptions: CompilationOptions)
     fun getFloat(num: Number): IMachineFloat
     fun getFloatRomConst(number: Double): String?
     fun importLibs(compilerOptions: CompilationOptions, importer: ModuleImporter, program: Program)
     fun launchEmulator(programName: String)
+    fun isRegularRAMaddress(address: Int): Boolean
 }

compiler/src/prog8/compiler/target/c64/AssemblyProgram.kt

@@ -2,6 +2,7 @@ package prog8.compiler.target.c64
 
 import prog8.compiler.CompilationOptions
 import prog8.compiler.OutputType
+import prog8.compiler.target.CompilationTarget
 import prog8.compiler.target.IAssemblyProgram
 import prog8.compiler.target.generatedLabelPrefix
 import java.nio.file.Path
@@ -22,12 +23,12 @@ class AssemblyProgram(override val name: String, outputDir: Path) : IAssemblyPro
         val outFile = when (options.output) {
             OutputType.PRG -> {
                 command.add("--cbm-prg")
-                println("\nCreating prg.")
+                println("\nCreating prg for target ${CompilationTarget.instance.name}.")
                 prgFile
             }
             OutputType.RAW -> {
                 command.add("--nostart")
-                println("\nCreating raw binary.")
+                println("\nCreating raw binary for target ${CompilationTarget.instance.name}.")
                 binFile
             }
         }

compiler/src/prog8/compiler/target/c64/C64MachineDefinition.kt

@@ -2,6 +2,7 @@ package prog8.compiler.target.c64
 
 import prog8.ast.Program
 import prog8.compiler.*
+import prog8.compiler.target.CpuType
 import prog8.compiler.target.IMachineDefinition
 import prog8.compiler.target.IMachineFloat
 import prog8.parser.ModuleImporter
@@ -12,7 +13,7 @@ import kotlin.math.pow
 
 internal object C64MachineDefinition: IMachineDefinition {
 
-    override val cpu = "6502"
+    override val cpu = CpuType.CPU6502
 
     // 5-byte cbm MFLPT format limitations:
     override val FLOAT_MAX_POSITIVE = 1.7014118345e+38         // bytes: 255,127,255,255,255
@@ -28,7 +29,6 @@ internal object C64MachineDefinition: IMachineDefinition {
     override val ESTACK_HI = 0xcf00        //  $ce00-$ceff inclusive
 
     override lateinit var zeropage: Zeropage
-    override val initSystemProcname = "c64.init_system"
 
     override fun getFloat(num: Number) = Mflpt5.fromNumber(num)
 
@@ -37,28 +37,29 @@ internal object C64MachineDefinition: IMachineDefinition {
         val mflpt5 = Mflpt5.fromNumber(number)
         val floatbytes = shortArrayOf(mflpt5.b0, mflpt5.b1, mflpt5.b2, mflpt5.b3, mflpt5.b4)
         when {
-            floatbytes.contentEquals(shortArrayOf(0x00, 0x00, 0x00, 0x00, 0x00)) -> return "c64flt.FL_ZERO"
-            floatbytes.contentEquals(shortArrayOf(0x82, 0x49, 0x0f, 0xda, 0xa1)) -> return "c64flt.FL_PIVAL"
-            floatbytes.contentEquals(shortArrayOf(0x90, 0x80, 0x00, 0x00, 0x00)) -> return "c64flt.FL_N32768"
-            floatbytes.contentEquals(shortArrayOf(0x81, 0x00, 0x00, 0x00, 0x00)) -> return "c64flt.FL_FONE"
-            floatbytes.contentEquals(shortArrayOf(0x80, 0x35, 0x04, 0xf3, 0x34)) -> return "c64flt.FL_SQRHLF"
-            floatbytes.contentEquals(shortArrayOf(0x81, 0x35, 0x04, 0xf3, 0x34)) -> return "c64flt.FL_SQRTWO"
-            floatbytes.contentEquals(shortArrayOf(0x80, 0x80, 0x00, 0x00, 0x00)) -> return "c64flt.FL_NEGHLF"
-            floatbytes.contentEquals(shortArrayOf(0x80, 0x31, 0x72, 0x17, 0xf8)) -> return "c64flt.FL_LOG2"
-            floatbytes.contentEquals(shortArrayOf(0x84, 0x20, 0x00, 0x00, 0x00)) -> return "c64flt.FL_TENC"
-            floatbytes.contentEquals(shortArrayOf(0x9e, 0x6e, 0x6b, 0x28, 0x00)) -> return "c64flt.FL_NZMIL"
-            floatbytes.contentEquals(shortArrayOf(0x80, 0x00, 0x00, 0x00, 0x00)) -> return "c64flt.FL_FHALF"
-            floatbytes.contentEquals(shortArrayOf(0x81, 0x38, 0xaa, 0x3b, 0x29)) -> return "c64flt.FL_LOGEB2"
-            floatbytes.contentEquals(shortArrayOf(0x81, 0x49, 0x0f, 0xda, 0xa2)) -> return "c64flt.FL_PIHALF"
-            floatbytes.contentEquals(shortArrayOf(0x83, 0x49, 0x0f, 0xda, 0xa2)) -> return "c64flt.FL_TWOPI"
-            floatbytes.contentEquals(shortArrayOf(0x7f, 0x00, 0x00, 0x00, 0x00)) -> return "c64flt.FL_FR4"
+            floatbytes.contentEquals(shortArrayOf(0x00, 0x00, 0x00, 0x00, 0x00)) -> return "floats.FL_ZERO_const"        // not a ROM const
+            floatbytes.contentEquals(shortArrayOf(0x81, 0x00, 0x00, 0x00, 0x00)) -> return "floats.FL_ONE_const"        // not a ROM const
+            floatbytes.contentEquals(shortArrayOf(0x82, 0x49, 0x0f, 0xda, 0xa1)) -> return "floats.FL_PIVAL"
+            floatbytes.contentEquals(shortArrayOf(0x90, 0x80, 0x00, 0x00, 0x00)) -> return "floats.FL_N32768"
+            floatbytes.contentEquals(shortArrayOf(0x81, 0x00, 0x00, 0x00, 0x00)) -> return "floats.FL_FONE"
+            floatbytes.contentEquals(shortArrayOf(0x80, 0x35, 0x04, 0xf3, 0x34)) -> return "floats.FL_SQRHLF"
+            floatbytes.contentEquals(shortArrayOf(0x81, 0x35, 0x04, 0xf3, 0x34)) -> return "floats.FL_SQRTWO"
+            floatbytes.contentEquals(shortArrayOf(0x80, 0x80, 0x00, 0x00, 0x00)) -> return "floats.FL_NEGHLF"
+            floatbytes.contentEquals(shortArrayOf(0x80, 0x31, 0x72, 0x17, 0xf8)) -> return "floats.FL_LOG2"
+            floatbytes.contentEquals(shortArrayOf(0x84, 0x20, 0x00, 0x00, 0x00)) -> return "floats.FL_TENC"
+            floatbytes.contentEquals(shortArrayOf(0x9e, 0x6e, 0x6b, 0x28, 0x00)) -> return "floats.FL_NZMIL"
+            floatbytes.contentEquals(shortArrayOf(0x80, 0x00, 0x00, 0x00, 0x00)) -> return "floats.FL_FHALF"
+            floatbytes.contentEquals(shortArrayOf(0x81, 0x38, 0xaa, 0x3b, 0x29)) -> return "floats.FL_LOGEB2"
+            floatbytes.contentEquals(shortArrayOf(0x81, 0x49, 0x0f, 0xda, 0xa2)) -> return "floats.FL_PIHALF"
+            floatbytes.contentEquals(shortArrayOf(0x83, 0x49, 0x0f, 0xda, 0xa2)) -> return "floats.FL_TWOPI"
+            floatbytes.contentEquals(shortArrayOf(0x7f, 0x00, 0x00, 0x00, 0x00)) -> return "floats.FL_FR4"
             else -> {
                 // attempt to correct for a few rounding issues
                 when (number.toBigDecimal().setScale(10, RoundingMode.HALF_DOWN).toDouble()) {
-                    3.1415926536 -> return "c64flt.FL_PIVAL"
-                    1.4142135624 -> return "c64flt.FL_SQRTWO"
-                    0.7071067812 -> return "c64flt.FL_SQRHLF"
-                    0.6931471806 -> return "c64flt.FL_LOG2"
+                    3.1415926536 -> return "floats.FL_PIVAL"
+                    1.4142135624 -> return "floats.FL_SQRTWO"
+                    0.7071067812 -> return "floats.FL_SQRHLF"
+                    0.6931471806 -> return "floats.FL_LOG2"
                     else -> {}
                 }
             }
@@ -68,7 +69,7 @@ internal object C64MachineDefinition: IMachineDefinition {
 
     override fun importLibs(compilerOptions: CompilationOptions, importer: ModuleImporter, program: Program) {
         if (compilerOptions.launcher == LauncherType.BASIC || compilerOptions.output == OutputType.PRG)
-            importer.importLibraryModule(program, "c64lib")
+            importer.importLibraryModule(program, "syslib")
     }
 
     override fun launchEmulator(programName: String) {
@@ -88,6 +89,8 @@ internal object C64MachineDefinition: IMachineDefinition {
         }
     }
 
+    override fun isRegularRAMaddress(address: Int): Boolean = address<0xa000 || address in 0xc000..0xcfff
+
     override fun initializeZeropage(compilerOptions: CompilationOptions) {
         zeropage = C64Zeropage(compilerOptions)
     }
@@ -107,18 +110,11 @@ internal object C64MachineDefinition: IMachineDefinition {
     internal class C64Zeropage(options: CompilationOptions) : Zeropage(options) {
 
         override val SCRATCH_B1 = 0x02      // temp storage for a single byte
-        override val SCRATCH_REG = 0x03     // temp storage for a register
-        override val SCRATCH_REG_X = 0xfa   // temp storage for register X (the evaluation stack pointer)
+        override val SCRATCH_REG = 0x03     // temp storage for a register, must be B1+1
         override val SCRATCH_W1 = 0xfb      // temp storage 1 for a word  $fb+$fc
         override val SCRATCH_W2 = 0xfd      // temp storage 2 for a word  $fb+$fc
 
 
-        override val exitProgramStrategy: ExitProgramStrategy = when (options.zeropage) {
-            ZeropageType.BASICSAFE, ZeropageType.DONTUSE -> ExitProgramStrategy.CLEAN_EXIT
-            ZeropageType.FLOATSAFE, ZeropageType.KERNALSAFE, ZeropageType.FULL -> ExitProgramStrategy.SYSTEM_RESET
-        }
-
-
         init {
             if (options.floats && options.zeropage !in setOf(ZeropageType.FLOATSAFE, ZeropageType.BASICSAFE, ZeropageType.DONTUSE ))
                 throw CompilerException("when floats are enabled, zero page type should be 'floatsafe' or 'basicsafe' or 'dontuse'")
@@ -126,12 +122,14 @@ internal object C64MachineDefinition: IMachineDefinition {
             if (options.zeropage == ZeropageType.FULL) {
                 free.addAll(0x04..0xf9)
                 free.add(0xff)
-                free.removeAll(listOf(SCRATCH_B1, SCRATCH_REG, SCRATCH_REG_X, SCRATCH_W1, SCRATCH_W1 + 1, SCRATCH_W2, SCRATCH_W2 + 1))
+                free.removeAll(listOf(SCRATCH_B1, SCRATCH_REG, SCRATCH_W1, SCRATCH_W1 + 1, SCRATCH_W2, SCRATCH_W2 + 1))
                 free.removeAll(listOf(0xa0, 0xa1, 0xa2, 0x91, 0xc0, 0xc5, 0xcb, 0xf5, 0xf6))        // these are updated by IRQ
             } else {
                 if (options.zeropage == ZeropageType.KERNALSAFE || options.zeropage == ZeropageType.FLOATSAFE) {
-                    free.addAll(listOf(0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10, 0x11,
-                            0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, 0x20, 0x21,
+                    free.addAll(listOf(
+                            0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10, 0x11,
+                            0x16, 0x17, 0x18, 0x19, 0x1a,
+                            0x1b, 0x1c, 0x1d, 0x1e, 0x1f, 0x20, 0x21,
                             0x22, 0x23, 0x24, 0x25,
                             0x39, 0x3a, 0x3b, 0x3c, 0x3d, 0x3e, 0x3f, 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46,
                             0x47, 0x48, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f, 0x51, 0x52, 0x53,
@@ -148,20 +146,20 @@ internal object C64MachineDefinition: IMachineDefinition {
                 if (options.zeropage == ZeropageType.FLOATSAFE) {
                     // remove the zero page locations used for floating point operations from the free list
                     free.removeAll(listOf(
-                            0x12, 0x26, 0x27, 0x28, 0x29, 0x2a,
+                            0x10, 0x11, 0x12, 0x26, 0x27, 0x28, 0x29, 0x2a,
                             0x57, 0x58, 0x59, 0x5a, 0x5b, 0x5c, 0x5d, 0x5e, 0x5f, 0x60,
                             0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68,
                             0x69, 0x6a, 0x6b, 0x6c, 0x6d, 0x6e, 0x6f, 0x70, 0x71, 0x72,
-                            0x8b, 0x8c, 0x8d, 0x8e, 0x8f, 0xf
+                            0x8b, 0x8c, 0x8d, 0x8e, 0x8f, 0xff
                     ))
                 }
 
                 if(options.zeropage!=ZeropageType.DONTUSE) {
-                    // add the other free Zp addresses,
-                    // these are valid for the C-64 (when no RS232 I/O is performed) but to keep BASIC running fully:
+                    // add the free Zp addresses
+                    // these are valid for the C-64 but allow BASIC to keep running fully *as long as you don't use tape I/O*
                     free.addAll(listOf(0x04, 0x05, 0x06, 0x0a, 0x0e,
-                            0x94, 0x95, 0xa7, 0xa8, 0xa9, 0xaa,
-                            0xb5, 0xb6, 0xf7, 0xf8, 0xf9))
+                            0x92, 0x96, 0x9b, 0x9c, 0x9e, 0x9f, 0xa5, 0xa6,
+                            0xb0, 0xb1, 0xbe, 0xbf, 0xf9))
                 } else {
                     // don't use the zeropage at all
                     free.clear()
@@ -169,7 +167,6 @@ internal object C64MachineDefinition: IMachineDefinition {
             }
             require(SCRATCH_B1 !in free)
             require(SCRATCH_REG !in free)
-            require(SCRATCH_REG_X !in free)
             require(SCRATCH_W1 !in free)
             require(SCRATCH_W2 !in free)
 

compiler/src/prog8/compiler/target/c64/Petscii.kt

@@ -1054,11 +1054,16 @@ object Petscii {
         val lookup = if(lowercase) encodingPetsciiLowercase else encodingPetsciiUppercase
         return text.map {
             val petscii = lookup[it]
-            petscii?.toShort() ?: if(it=='\u0000')
-                0.toShort()
-            else {
-                val case = if (lowercase) "lower" else "upper"
-                throw CharConversionException("no ${case}case Petscii character for '$it' (${it.toShort()})")
+            petscii?.toShort() ?: when (it) {
+                '\u0000' -> 0.toShort()
+                in '\u8000'..'\u80ff' -> {
+                    // special case: take the lower 8 bit hex value directly
+                    (it.toInt() - 0x8000).toShort()
+                }
+                else -> {
+                    val case = if (lowercase) "lower" else "upper"
+                    throw CharConversionException("no ${case}case Petscii character for '$it' (${it.toShort()})")
+                }
             }
         }
     }
@@ -1072,11 +1077,16 @@ object Petscii {
         val lookup = if(lowercase) encodingScreencodeLowercase else encodingScreencodeUppercase
         return text.map{
             val screencode = lookup[it]
-            screencode?.toShort() ?: if(it=='\u0000')
-                0.toShort()
-            else {
-                val case = if (lowercase) "lower" else "upper"
-                throw CharConversionException("no ${case}Screencode character for '$it' (${it.toShort()})")
+            screencode?.toShort() ?: when (it) {
+                '\u0000' -> 0.toShort()
+                in '\u8000'..'\u80ff' -> {
+                    // special case: take the lower 8 bit hex value directly
+                    (it.toInt() - 0x8000).toShort()
+                }
+                else -> {
+                    val case = if (lowercase) "lower" else "upper"
+                    throw CharConversionException("no ${case}Screencode character for '$it' (${it.toShort()})")
+                }
             }
         }
     }

compiler/src/prog8/compiler/target/c64/codegen/AsmGen.kt

@@ -8,19 +8,19 @@ import prog8.ast.base.*
 import prog8.ast.expressions.*
 import prog8.ast.statements.*
 import prog8.compiler.*
+import prog8.compiler.target.*
 import prog8.compiler.target.CompilationTarget
+import prog8.compiler.target.CpuType
 import prog8.compiler.target.IAssemblyGenerator
 import prog8.compiler.target.IAssemblyProgram
 import prog8.compiler.target.c64.AssemblyProgram
 import prog8.compiler.target.c64.Petscii
-import prog8.compiler.target.c64.codegen.assignment.AsmAssignSource
-import prog8.compiler.target.c64.codegen.assignment.AsmAssignTarget
 import prog8.compiler.target.c64.codegen.assignment.AsmAssignment
 import prog8.compiler.target.c64.codegen.assignment.AssignmentAsmGen
-import prog8.compiler.target.c64.codegen.assignment.TargetStorageKind
 import prog8.compiler.target.generatedLabelPrefix
 import prog8.functions.BuiltinFunctions
 import prog8.functions.FSignature
+import java.io.CharConversionException
 import java.nio.file.Path
 import java.time.LocalDate
 import java.time.LocalDateTime
@@ -34,6 +34,10 @@ internal class AsmGen(private val program: Program,
                       val options: CompilationOptions,
                       private val outputDir: Path): IAssemblyGenerator {
 
+    // for expressions and augmented assignments:
+    val optimizedByteMultiplications = setOf(3,5,6,7,9,10,11,12,13,14,15,20,25,40,50,80,100)
+    val optimizedWordMultiplications = setOf(3,5,6,7,9,10,12,15,20,25,40,50,80,100,320)
+
     private val assemblyLines = mutableListOf<String>()
     private val globalFloatConsts = mutableMapOf<Double, String>()     // all float values in the entire program (value -> varname)
     private val allocatedZeropageVariables = mutableMapOf<String, Pair<Int, DataType>>()
@@ -42,10 +46,10 @@ internal class AsmGen(private val program: Program,
     private val forloopsAsmGen = ForLoopsAsmGen(program, this)
     private val postincrdecrAsmGen = PostIncrDecrAsmGen(program, this)
     private val functioncallAsmGen = FunctionCallAsmGen(program, this)
-    private val assignmentAsmGen = AssignmentAsmGen(program, this)
     private val expressionsAsmGen = ExpressionsAsmGen(program, this)
+    private val assignmentAsmGen = AssignmentAsmGen(program, this, expressionsAsmGen)
     internal val loopEndLabels = ArrayDeque<String>()
-    internal val blockLevelVarInits = mutableMapOf<Block, MutableSet<VarDecl>>()
+    private val blockLevelVarInits = mutableMapOf<Block, MutableSet<VarDecl>>()
 
     override fun compileToAssembly(optimize: Boolean): IAssemblyProgram {
         assemblyLines.clear()
@@ -61,24 +65,34 @@ internal class AsmGen(private val program: Program,
             block2asm(b)
         footer()
 
-        if(optimize) {
-            var optimizationsDone = 1
-            while (optimizationsDone > 0) {
-                optimizationsDone = optimizeAssembly(assemblyLines)
-            }
-        }
-
         val outputFile = outputDir.resolve("${program.name}.asm").toFile()
         outputFile.printWriter().use {
             for (line in assemblyLines) { it.println(line) }
         }
 
+        if(optimize) {
+            assemblyLines.clear()
+            assemblyLines.addAll(outputFile.readLines())
+            var optimizationsDone = 1
+            while (optimizationsDone > 0) {
+                optimizationsDone = optimizeAssembly(assemblyLines)
+            }
+            outputFile.printWriter().use {
+                for (line in assemblyLines) { it.println(line) }
+            }
+        }
+
         return AssemblyProgram(program.name, outputDir)
     }
 
+
     private fun header() {
         val ourName = this.javaClass.name
-        val cpu = CompilationTarget.machine.cpu
+        val cpu = when(CompilationTarget.instance.machine.cpu) {
+            CpuType.CPU6502 -> "6502"
+            CpuType.CPU65c02 -> "65c02"
+            else -> "unsupported"
+        }
 
         out("; $cpu assembly code for '${program.name}'")
         out("; generated by $ourName on ${LocalDateTime.now().withNano(0)}")
@@ -89,18 +103,16 @@ internal class AsmGen(private val program: Program,
         program.actualLoadAddress = program.definedLoadAddress
         if (program.actualLoadAddress == 0)   // fix load address
             program.actualLoadAddress = if (options.launcher == LauncherType.BASIC)
-                CompilationTarget.machine.BASIC_LOAD_ADDRESS else CompilationTarget.machine.RAW_LOAD_ADDRESS
+                CompilationTarget.instance.machine.BASIC_LOAD_ADDRESS else CompilationTarget.instance.machine.RAW_LOAD_ADDRESS
 
         // the global prog8 variables needed
-        val zp = CompilationTarget.machine.zeropage
-        val initproc = CompilationTarget.machine.initSystemProcname
+        val zp = CompilationTarget.instance.machine.zeropage
         out("P8ZP_SCRATCH_B1 = ${zp.SCRATCH_B1}")
         out("P8ZP_SCRATCH_REG = ${zp.SCRATCH_REG}")
-        out("P8ZP_SCRATCH_REG_X = ${zp.SCRATCH_REG_X}")
         out("P8ZP_SCRATCH_W1 = ${zp.SCRATCH_W1}    ; word")
         out("P8ZP_SCRATCH_W2 = ${zp.SCRATCH_W2}    ; word")
-        out("P8ESTACK_LO = ${CompilationTarget.machine.ESTACK_LO.toHex()}")
-        out("P8ESTACK_HI = ${CompilationTarget.machine.ESTACK_HI.toHex()}")
+        out("P8ESTACK_LO = ${CompilationTarget.instance.machine.ESTACK_LO.toHex()}")
+        out("P8ESTACK_HI = ${CompilationTarget.instance.machine.ESTACK_HI.toHex()}")
 
         when {
             options.launcher == LauncherType.BASIC -> {
@@ -113,18 +125,14 @@ internal class AsmGen(private val program: Program,
                 out("  .null  $9e, format(' %d ', _prog8_entrypoint), $3a, $8f, ' prog8 by idj'")
                 out("+\t.word  0")
                 out("_prog8_entrypoint\t; assembly code starts here\n")
-                out("  tsx")
-                out("  stx  prog8_lib.orig_stackpointer")
-                if(!initproc.isNullOrEmpty())
-                    out("  jsr  $initproc")
+                if(!options.noSysInit)
+                    out("  jsr  ${CompilationTarget.instance.name}.init_system")
             }
             options.output == OutputType.PRG -> {
                 out("; ---- program without basic sys call ----")
                 out("* = ${program.actualLoadAddress.toHex()}\n")
-                out("  tsx")
-                out("  stx  prog8_lib.orig_stackpointer")
-                if(!initproc.isNullOrEmpty())
-                    out("  jsr  $initproc")
+                if(!options.noSysInit)
+                    out("  jsr  ${CompilationTarget.instance.name}.init_system")
             }
             options.output == OutputType.RAW -> {
                 out("; ---- raw assembler program ----")
@@ -132,40 +140,23 @@ internal class AsmGen(private val program: Program,
             }
         }
 
-        if (zeropage.exitProgramStrategy != Zeropage.ExitProgramStrategy.CLEAN_EXIT) {
-            // disable shift-commodore charset switching and run/stop key
-            out("  lda  #$80")
-            out("  lda  #$80")
-            out("  sta  657\t; disable charset switching")
-            out("  lda  #239")
-            out("  sta  808\t; disable run/stop key")
+        if(options.zeropage !in setOf(ZeropageType.BASICSAFE, ZeropageType.DONTUSE)) {
+            out("""
+                ; zeropage is clobbered so we need to reset the machine at exit
+                lda  #>${CompilationTarget.instance.name}.reset_system
+                pha
+                lda  #<${CompilationTarget.instance.name}.reset_system
+                pha""")
         }
 
-        out("  ldx  #\$ff\t; init estack pointer")
-
-        out("  ; initialize the variables in each block that has globals")
-        program.allBlocks().forEach {
-            if(it.statements.filterIsInstance<VarDecl>().any { vd->vd.value!=null && vd.type==VarDeclType.VAR && vd.datatype in NumericDatatypes})
-                out("  jsr  ${it.name}.prog8_init_vars")
-        }
-
-        out("  clc")
-        when (zeropage.exitProgramStrategy) {
-            Zeropage.ExitProgramStrategy.CLEAN_EXIT -> {
-                out("  jmp  main.start\t; jump to program entrypoint")
-            }
-            Zeropage.ExitProgramStrategy.SYSTEM_RESET -> {
-                out("  jsr  main.start\t; call program entrypoint")
-                out("  jmp  (c64.RESET_VEC)\t; cold reset")
-            }
-        }
+        out("  jmp  main.start  ; start program / force start proc to be included")
     }
 
     private fun footer() {
         // the global list of all floating point constants for the whole program
         out("; global float constants")
         for (flt in globalFloatConsts) {
-            val floatFill = CompilationTarget.machine.getFloat(flt.key).makeFloatFillAsm()
+            val floatFill = CompilationTarget.instance.machine.getFloat(flt.key).makeFloatFillAsm()
             val floatvalue = flt.key
             out("${flt.value}\t.byte  $floatFill  ; float $floatvalue")
         }
@@ -207,13 +198,8 @@ internal class AsmGen(private val program: Program,
     }
 
     private fun assignInitialValueToVar(decl: VarDecl, variableName: List<String>) {
-        val variable = IdentifierReference(variableName, decl.position)
-        variable.linkParents(decl.parent)
-        val asgn = AsmAssignment(
-                AsmAssignSource.fromAstSource(decl.value!!, program),
-                AsmAssignTarget(TargetStorageKind.VARIABLE, program, this, decl.datatype, variable = variable),
-                false, decl.position)
-        assignmentAsmGen.translateNormalAssignment(asgn)
+        val asmName = asmVariableName(variableName)
+        assignmentAsmGen.assignExpressionToVariable(decl.value!!, asmName, decl.datatype, decl.definingSubroutine())
     }
 
     private var generatedLabelSequenceNumber: Int = 0
@@ -240,8 +226,12 @@ internal class AsmGen(private val program: Program,
     }
 
     private fun encode(str: String, altEncoding: Boolean): List<Short> {
-         val bytes = if(altEncoding) Petscii.encodeScreencode(str, true) else Petscii.encodePetscii(str, true)
-         return bytes.plus(0)
+        try {
+            val bytes = if (altEncoding) Petscii.encodeScreencode(str, true) else Petscii.encodePetscii(str, true)
+            return bytes.plus(0)
+        } catch(x: CharConversionException) {
+            throw AssemblyError("There was a problem converting a string to the target machine's char encoding: ${x.message}")
+        }
     }
 
     private fun zeropagevars2asm(statements: List<Statement>) {
@@ -262,7 +252,7 @@ internal class AsmGen(private val program: Program,
                         errors.handle()
                         out("${variable.name} = $address\t; auto zp ${variable.datatype}")
                         // make sure we add the var to the set of zpvars for this block
-                        allocatedZeropageVariables[fullName] = Pair(address, variable.datatype)
+                        allocatedZeropageVariables[fullName] = address to variable.datatype
                     } catch (x: ZeropageDepletedError) {
                         // leave it as it is.
                     }
@@ -335,7 +325,7 @@ internal class AsmGen(private val program: Program,
                         }
                 val floatFills = array.map {
                     val number = (it as NumericLiteralValue).number
-                    CompilationTarget.machine.getFloat(number).makeFloatFillAsm()
+                    CompilationTarget.instance.machine.getFloat(number).makeFloatFillAsm()
                 }
                 out(name)
                 for (f in array.zip(floatFills))
@@ -348,7 +338,10 @@ internal class AsmGen(private val program: Program,
         out("\n; memdefs and kernel subroutines")
         val memvars = statements.filterIsInstance<VarDecl>().filter { it.type==VarDeclType.MEMORY || it.type==VarDeclType.CONST }
         for(m in memvars) {
-            out("  ${m.name} = ${(m.value as NumericLiteralValue).number.toHex()}")
+            if(m.value is NumericLiteralValue)
+                out("  ${m.name} = ${(m.value as NumericLiteralValue).number.toHex()}")
+            else
+                out("  ${m.name} = ${asmVariableName((m.value as AddressOf).identifier)}")
         }
         val asmSubs = statements.filterIsInstance<Subroutine>().filter { it.isAsmSubroutine }
         for(sub in asmSubs) {
@@ -416,10 +409,17 @@ internal class AsmGen(private val program: Program,
                     "$"+number.toString(16).padStart(2, '0')
                 }
             DataType.ARRAY_UW -> array.map {
-                if(it is NumericLiteralValue) {
-                    "$" + it.number.toInt().toString(16).padStart(4, '0')
-                } else {
-                    (it as AddressOf).identifier.nameInSource.joinToString(".")
+                when (it) {
+                    is NumericLiteralValue -> {
+                        "$" + it.number.toInt().toString(16).padStart(4, '0')
+                    }
+                    is AddressOf -> {
+                        it.identifier.firstStructVarName(program.namespace) ?: asmSymbolName(it.identifier)
+                    }
+                    is IdentifierReference -> {
+                        it.firstStructVarName(program.namespace) ?: asmSymbolName(it)
+                    }
+                    else -> throw AssemblyError("weird array elt dt")
                 }
             }
             else -> throw AssemblyError("invalid arraysize type")
@@ -469,7 +469,7 @@ internal class AsmGen(private val program: Program,
     }
 
     internal fun getFloatAsmConst(number: Double): String {
-        var asmName = CompilationTarget.machine.getFloatRomConst(number)
+        var asmName = CompilationTarget.instance.machine.getFloatRomConst(number)
         if(asmName.isNullOrEmpty()) {
             // no ROM float const for this value, create our own
             asmName = globalFloatConsts[number]
@@ -499,8 +499,14 @@ internal class AsmGen(private val program: Program,
         }
     }
 
+    internal fun asmSymbolName(name: String) = fixNameSymbols(name)
+    internal fun asmVariableName(name: String) = fixNameSymbols(name)
+    internal fun asmSymbolName(name: Iterable<String>) = fixNameSymbols(name.joinToString("."))
+    internal fun asmVariableName(name: Iterable<String>) = fixNameSymbols(name.joinToString("."))
+
+
     internal fun loadByteFromPointerIntoA(pointervar: IdentifierReference): Pair<Boolean, String> {
-        // returns if the pointer is already on the ZP itself or not (in which case SCRATCH_W1 is used as intermediary)
+        // returns if the pointer is already on the ZP itself or not (in the latter case SCRATCH_W1 is used as intermediary)
         val sourceName = asmVariableName(pointervar)
         val vardecl = pointervar.targetVarDecl(program.namespace)!!
         val scopedName = vardecl.makeScopedName(vardecl.name)
@@ -541,24 +547,70 @@ internal class AsmGen(private val program: Program,
         }
     }
 
-    internal fun fixNameSymbols(name: String) = name.replace("<", "prog8_").replace(">", "")     // take care of the autogenerated invalid (anon) label names
+    private  fun fixNameSymbols(name: String) = name.replace("<", "prog8_").replace(">", "")     // take care of the autogenerated invalid (anon) label names
 
-    internal fun saveRegister(register: CpuRegister) {
-        when(register) {
-            CpuRegister.A -> out("  pha")
-            CpuRegister.X -> out("  txa | pha")
-            CpuRegister.Y -> out("  tya | pha")
+    internal fun saveRegister(register: CpuRegister, dontUseStack: Boolean, scope: Subroutine) {
+        if(dontUseStack) {
+            when (register) {
+                CpuRegister.A -> {
+                    out("  sta  _prog8_regsaveA")
+                    scope.asmGenInfo.usedRegsaveA = true
+                }
+                CpuRegister.X -> {
+                    out("  stx  _prog8_regsaveX")
+                    scope.asmGenInfo.usedRegsaveX = true
+                }
+                CpuRegister.Y -> {
+                    out("  sty  _prog8_regsaveY")
+                    scope.asmGenInfo.usedRegsaveY = true
+                }
+            }
+
+        } else {
+            when (register) {
+                CpuRegister.A -> out("  pha")
+                CpuRegister.X -> {
+                    if (CompilationTarget.instance.machine.cpu == CpuType.CPU65c02) out("  phx")
+                    else {
+                        out("  stx  _prog8_regsaveX")
+                        scope.asmGenInfo.usedRegsaveX = true
+                    }
+                }
+                CpuRegister.Y -> {
+                    if (CompilationTarget.instance.machine.cpu == CpuType.CPU65c02) out("  phy")
+                    else {
+                        out("  sty  _prog8_regsaveY")
+                        scope.asmGenInfo.usedRegsaveY = true
+                    }
+                }
+            }
         }
     }
 
-    internal fun restoreRegister(register: CpuRegister) {
-        when(register) {
-            CpuRegister.A -> out("  pla")
-            CpuRegister.X -> out("  pla | tax")
-            CpuRegister.Y -> out("  pla | tay")
+    internal fun restoreRegister(register: CpuRegister, dontUseStack: Boolean) {
+        if(dontUseStack) {
+            when (register) {
+                CpuRegister.A -> out("  sta  _prog8_regsaveA")
+                CpuRegister.X -> out("  ldx  _prog8_regsaveX")
+                CpuRegister.Y -> out("  ldy  _prog8_regsaveY")
+            }
+
+        } else {
+            when (register) {
+                CpuRegister.A -> out("  pla")
+                CpuRegister.X -> {
+                    if (CompilationTarget.instance.machine.cpu == CpuType.CPU65c02) out("  plx")
+                    else out("  ldx  _prog8_regsaveX")
+                }
+                CpuRegister.Y -> {
+                    if (CompilationTarget.instance.machine.cpu == CpuType.CPU65c02) out("  ply")
+                    else out("  ldy  _prog8_regsaveY")
+                }
+            }
         }
     }
 
+
     internal fun translate(stmt: Statement) {
         outputSourceLine(stmt)
         when(stmt) {
@@ -572,19 +624,10 @@ internal class AsmGen(private val program: Program,
             is FunctionCallStatement -> {
                 val functionName = stmt.target.nameInSource.last()
                 val builtinFunc = BuiltinFunctions[functionName]
-                if(builtinFunc!=null) {
+                if (builtinFunc != null) {
                     builtinFunctionsAsmGen.translateFunctioncallStatement(stmt, builtinFunc)
                 } else {
-                    functioncallAsmGen.translateFunctionCall(stmt)
-                    // discard any results from the stack:
-                    val sub = stmt.target.targetSubroutine(program.namespace)!!
-                    val returns = sub.returntypes.zip(sub.asmReturnvaluesRegisters)
-                    for((t, reg) in returns) {
-                        if(reg.stack) {
-                            if (t in IntegerDatatypes || t in PassByReferenceDatatypes) out("  inx")
-                            else if (t == DataType.FLOAT) out("  inx |  inx |  inx")
-                        }
-                    }
+                    functioncallAsmGen.translateFunctionCallStatement(stmt)
                 }
             }
             is Assignment -> assignmentAsmGen.translate(stmt)
@@ -615,101 +658,75 @@ internal class AsmGen(private val program: Program,
                                                   register: CpuRegister,
                                                   addOneExtra: Boolean=false) {
         val reg = register.toString().toLowerCase()
-        val index = expr.arrayspec.index
-        if(index is NumericLiteralValue) {
-            val indexValue = index.number.toInt() * elementDt.memorySize() + if(addOneExtra) 1 else 0
+        val indexnum = expr.indexer.constIndex()
+        if(indexnum!=null) {
+            val indexValue = indexnum * elementDt.memorySize() + if(addOneExtra) 1 else 0
             out("  ld$reg  #$indexValue")
             return
         }
 
+        val indexName = asmVariableName(expr.indexer.indexVar!!)
         if(addOneExtra) {
             // add 1 to the result
-            if (index is IdentifierReference) {
-                val indexName = asmVariableName(index)
-                when(elementDt) {
-                    in ByteDatatypes -> {
-                        out("  ldy  $indexName |  iny")
-                        when(register) {
-                            CpuRegister.A -> out(" tya")
-                            CpuRegister.X -> out(" tyx")
-                            CpuRegister.Y -> {}
-                        }
+            when(elementDt) {
+                in ByteDatatypes -> {
+                    out("  ldy  $indexName |  iny")
+                    when(register) {
+                        CpuRegister.A -> out(" tya")
+                        CpuRegister.X -> out(" tyx")
+                        CpuRegister.Y -> {}
                     }
-                    in WordDatatypes -> {
-                        out("  lda  $indexName |  sec |  rol a")
-                        when(register) {
-                            CpuRegister.A -> {}
-                            CpuRegister.X -> out(" tax")
-                            CpuRegister.Y -> out(" tay")
-                        }
-                    }
-                    DataType.FLOAT -> {
-                        require(DataType.FLOAT.memorySize()==5)
-                        out("""
-                                    lda  $indexName
-                                    asl  a
-                                    asl  a
-                                    sec
-                                    adc  $indexName""")
-                        when(register) {
-                            CpuRegister.A -> {}
-                            CpuRegister.X -> out(" tax")
-                            CpuRegister.Y -> out(" tay")
-                        }
-                    }
-                    else -> throw AssemblyError("weird dt")
                 }
-            }
-            else {
-                expressionsAsmGen.translateExpression(index)
-                out("""
-                    inc  P8ESTACK_LO,x
-                    bne  +
-                    inc  P8ESTACK_HI,x
-+""")
-                when(register) {
-                    CpuRegister.A -> out("  inx |  lda  P8ESTACK_LO,x")
-                    CpuRegister.X -> throw AssemblyError("can't use X here")
-                    CpuRegister.Y -> out("  inx |  ldy  P8ESTACK_LO,x")
+                in WordDatatypes -> {
+                    out("  lda  $indexName |  sec |  rol a")
+                    when(register) {
+                        CpuRegister.A -> {}
+                        CpuRegister.X -> out(" tax")
+                        CpuRegister.Y -> out(" tay")
+                    }
                 }
+                DataType.FLOAT -> {
+                    require(DataType.FLOAT.memorySize()==5)
+                    out("""
+                                lda  $indexName
+                                asl  a
+                                asl  a
+                                sec
+                                adc  $indexName""")
+                    when(register) {
+                        CpuRegister.A -> {}
+                        CpuRegister.X -> out(" tax")
+                        CpuRegister.Y -> out(" tay")
+                    }
+                }
+                else -> throw AssemblyError("weird dt")
             }
         } else {
-            if (index is IdentifierReference) {
-                val indexName = asmVariableName(index)
-                when(elementDt) {
-                    in ByteDatatypes -> out("  ld$reg  $indexName")
-                    in WordDatatypes -> {
-                        out("  lda  $indexName |  asl a")
-                        when(register) {
-                            CpuRegister.A -> {}
-                            CpuRegister.X -> out(" tax")
-                            CpuRegister.Y -> out(" tay")
-                        }
+            when(elementDt) {
+                in ByteDatatypes -> out("  ld$reg  $indexName")
+                in WordDatatypes -> {
+                    out("  lda  $indexName |  asl a")
+                    when(register) {
+                        CpuRegister.A -> {}
+                        CpuRegister.X -> out(" tax")
+                        CpuRegister.Y -> out(" tay")
                     }
-                    DataType.FLOAT -> {
-                        require(DataType.FLOAT.memorySize()==5)
-                        out("""
-                                    lda  $indexName
-                                    asl  a
-                                    asl  a
-                                    clc
-                                    adc  $indexName""")
-                        when(register) {
-                            CpuRegister.A -> {}
-                            CpuRegister.X -> out(" tax")
-                            CpuRegister.Y -> out(" tay")
-                        }
+                }
+                DataType.FLOAT -> {
+                    require(DataType.FLOAT.memorySize()==5)
+                    out("""
+                                lda  $indexName
+                                asl  a
+                                asl  a
+                                clc
+                                adc  $indexName""")
+                    when(register) {
+                        CpuRegister.A -> {}
+                        CpuRegister.X -> out(" tax")
+                        CpuRegister.Y -> out(" tay")
                     }
-                    else -> throw AssemblyError("weird dt")
-                }
-            }
-            else {
-                expressionsAsmGen.translateExpression(index)
-                when(register) {
-                    CpuRegister.A -> out("  inx |  lda  P8ESTACK_LO,x")
-                    CpuRegister.X -> throw AssemblyError("can't use X here")
-                    CpuRegister.Y -> out("  inx |  ldy  P8ESTACK_LO,x")
                 }
+                else -> throw AssemblyError("weird dt")
             }
         }
     }
@@ -717,15 +734,28 @@ internal class AsmGen(private val program: Program,
     internal fun translateExpression(expression: Expression) =
             expressionsAsmGen.translateExpression(expression)
 
-    internal fun translateFunctioncallExpression(functionCall: FunctionCall, signature: FSignature) =
-            builtinFunctionsAsmGen.translateFunctioncallExpression(functionCall, signature)
+    internal fun translateExpression(indexer: ArrayIndex) =
+            expressionsAsmGen.translateExpression(indexer)
 
-    internal fun translateFunctionCall(functionCall: FunctionCall) =
-            functioncallAsmGen.translateFunctionCall(functionCall)
+    internal fun translateBuiltinFunctionCallExpression(functionCall: FunctionCall, signature: FSignature, resultToStack: Boolean) =
+            builtinFunctionsAsmGen.translateFunctioncallExpression(functionCall, signature, resultToStack)
+
+    internal fun translateFunctionCall(functionCall: FunctionCall, preserveStatusRegisterAfterCall: Boolean) =
+            functioncallAsmGen.translateFunctionCall(functionCall, preserveStatusRegisterAfterCall)
 
     internal fun translateNormalAssignment(assign: AsmAssignment) =
             assignmentAsmGen.translateNormalAssignment(assign)
 
+    internal fun assignExpressionToRegister(expr: Expression, register: RegisterOrPair) =
+            assignmentAsmGen.assignExpressionToRegister(expr, register)
+
+    internal fun assignExpressionToVariable(expr: Expression, asmVarName: String, dt: DataType, scope: Subroutine?) =
+            assignmentAsmGen.assignExpressionToVariable(expr, asmVarName, dt, scope)
+
+    internal fun assignVariableToRegister(asmVarName: String, register: RegisterOrPair) =
+            assignmentAsmGen.assignVariableToRegister(asmVarName, register)
+
+
     private fun translateSubroutine(sub: Subroutine) {
         out("")
         outputSourceLine(sub)
@@ -743,9 +773,37 @@ internal class AsmGen(private val program: Program,
             out("${sub.name}\t.proc")
             zeropagevars2asm(sub.statements)
             memdefs2asm(sub.statements)
+
+            // the main.start subroutine is the program's entrypoint and should perform some initialization logic
+            if(sub.name=="start" && sub.definingBlock().name=="main") {
+                out("; program startup initialization")
+                out("  cld")
+                program.allBlocks().forEach {
+                    if(it.statements.filterIsInstance<VarDecl>().any { vd->vd.value!=null && vd.type==VarDeclType.VAR && vd.datatype in NumericDatatypes})
+                        out("  jsr  ${it.name}.prog8_init_vars")
+                }
+                out("""
+                    tsx
+                    stx  prog8_lib.orig_stackpointer    ; required for func_exit                    
+                    ldx  #255       ; init estack ptr
+                    clv
+                    clc""")
+            }
+
             out("; statements")
             sub.statements.forEach{ translate(it) }
             out("; variables")
+            out("; register saves")
+            if(sub.asmGenInfo.usedRegsaveA)
+                out("_prog8_regsaveA     .byte  0")
+            if(sub.asmGenInfo.usedRegsaveX)
+                out("_prog8_regsaveX     .byte  0")
+            if(sub.asmGenInfo.usedRegsaveY)
+                out("_prog8_regsaveY     .byte  0")
+            if(sub.asmGenInfo.usedFloatEvalResultVar1)
+                out("$subroutineFloatEvalResultVar1    .byte 0,0,0,0,0")
+            if(sub.asmGenInfo.usedFloatEvalResultVar2)
+                out("$subroutineFloatEvalResultVar2    .byte 0,0,0,0,0")
             vardecls2asm(sub.statements)
             out("  .pend\n")
         }
@@ -756,12 +814,12 @@ internal class AsmGen(private val program: Program,
                 when (condition) {
                     BranchCondition.CS -> "bcc"
                     BranchCondition.CC -> "bcs"
-                    BranchCondition.EQ, BranchCondition.Z -> "beq"
-                    BranchCondition.NE, BranchCondition.NZ -> "bne"
+                    BranchCondition.EQ, BranchCondition.Z -> "bne"
+                    BranchCondition.NE, BranchCondition.NZ -> "beq"
                     BranchCondition.VS -> "bvc"
                     BranchCondition.VC -> "bvs"
-                    BranchCondition.MI, BranchCondition.NEG -> "bmi"
-                    BranchCondition.PL, BranchCondition.POS -> "bpl"
+                    BranchCondition.MI, BranchCondition.NEG -> "bpl"
+                    BranchCondition.PL, BranchCondition.POS -> "bmi"
                 }
             } else {
                 when (condition) {
@@ -777,25 +835,32 @@ internal class AsmGen(private val program: Program,
             }
 
     private fun translate(stmt: IfStatement) {
-        expressionsAsmGen.translateExpression(stmt.condition)
-        translateTestStack(stmt.condition.inferType(program).typeOrElse(DataType.STRUCT))
-        val elseLabel = makeLabel("if_else")
-        val endLabel = makeLabel("if_end")
-        out("  beq  $elseLabel")
-        translate(stmt.truepart)
-        out("  jmp  $endLabel")
-        out(elseLabel)
-        translate(stmt.elsepart)
-        out(endLabel)
+        checkBooleanExpression(stmt.condition)  // we require the condition to be of the form  'x <comparison> <value>'
+        val booleanCondition = stmt.condition as BinaryExpression
+
+        if (stmt.elsepart.containsNoCodeNorVars()) {
+            // empty else
+            val endLabel = makeLabel("if_end")
+            expressionsAsmGen.translateComparisonExpressionWithJumpIfFalse(booleanCondition, endLabel)
+            translate(stmt.truepart)
+            out(endLabel)
+        }
+        else {
+            // both true and else parts
+            val elseLabel = makeLabel("if_else")
+            val endLabel = makeLabel("if_end")
+            expressionsAsmGen.translateComparisonExpressionWithJumpIfFalse(booleanCondition, elseLabel)
+            translate(stmt.truepart)
+            out("  jmp  $endLabel")
+            out(elseLabel)
+            translate(stmt.elsepart)
+            out(endLabel)
+        }
     }
 
-    private fun translateTestStack(dataType: DataType) {
-        when(dataType) {
-            in ByteDatatypes -> out("  inx |  lda  P8ESTACK_LO,x")
-            in WordDatatypes -> out("  inx |  lda  P8ESTACK_LO,x |  ora  P8ESTACK_HI,x")
-            DataType.FLOAT -> throw AssemblyError("conditional value should be an integer (boolean)")
-            else -> throw AssemblyError("non-numerical dt")
-        }
+    private fun checkBooleanExpression(condition: Expression) {
+        if(condition !is BinaryExpression || condition.operator !in comparisonOperators)
+            throw AssemblyError("expected boolean expression $condition")
     }
 
     private fun translate(stmt: RepeatLoop) {
@@ -843,15 +908,16 @@ internal class AsmGen(private val program: Program,
                 }
             }
             else -> {
-                translateExpression(stmt.iterations!!)
-                val dt = stmt.iterations!!.inferType(program).typeOrElse(DataType.STRUCT)
-                when (dt) {
+                val dt = stmt.iterations!!.inferType(program)
+                if(!dt.isKnown)
+                    throw AssemblyError("unknown dt")
+                when (dt.typeOrElse(DataType.STRUCT)) {
                     in ByteDatatypes -> {
-                        out("  inx |  lda  P8ESTACK_LO,x")
+                        assignExpressionToRegister(stmt.iterations!!, RegisterOrPair.A)
                         repeatByteCountInA(null, repeatLabel, endLabel, stmt.body)
                     }
                     in WordDatatypes -> {
-                        out("  inx |  lda  P8ESTACK_LO,x |  ldy  P8ESTACK_HI,x")
+                        assignExpressionToRegister(stmt.iterations!!, RegisterOrPair.AY)
                         repeatWordCountInAY(null, repeatLabel, endLabel, stmt.body)
                     }
                     else -> throw AssemblyError("invalid loop expression datatype $dt")
@@ -863,6 +929,8 @@ internal class AsmGen(private val program: Program,
     }
 
     private fun repeatWordCountInAY(constIterations: Int?, repeatLabel: String, endLabel: String, body: AnonymousScope) {
+        if(constIterations==0)
+            return
         // note: A/Y must have been loaded with the number of iterations already!
         val counterVar = makeLabel("repeatcounter")
         out("""
@@ -892,8 +960,12 @@ $counterVar    .word  0""")
     }
 
     private fun repeatByteCountInA(constIterations: Int?, repeatLabel: String, endLabel: String, body: AnonymousScope) {
+        if(constIterations==0)
+            return
         // note: A must have been loaded with the number of iterations already!
         val counterVar = makeLabel("repeatcounter")
+        if(constIterations==null)
+            out("  beq  $endLabel")
         out("""
             sta  $counterVar
 $repeatLabel""")
@@ -914,25 +986,13 @@ $counterVar    .byte  0""")
     }
 
     private fun translate(stmt: WhileLoop) {
+        checkBooleanExpression(stmt.condition)  // we require the condition to be of the form  'x <comparison> <value>'
+        val booleanCondition = stmt.condition as BinaryExpression
         val whileLabel = makeLabel("while")
         val endLabel = makeLabel("whileend")
         loopEndLabels.push(endLabel)
         out(whileLabel)
-        expressionsAsmGen.translateExpression(stmt.condition)
-        val conditionDt = stmt.condition.inferType(program)
-        if(!conditionDt.isKnown)
-            throw AssemblyError("unknown condition dt")
-        if(conditionDt.typeOrElse(DataType.BYTE) in ByteDatatypes) {
-            out("  inx |  lda  P8ESTACK_LO,x  |  beq  $endLabel")
-        } else {
-            out("""
-                inx
-                lda  P8ESTACK_LO,x
-                bne  +
-                lda  P8ESTACK_HI,x
-                beq  $endLabel
-+  """)
-        }
+        expressionsAsmGen.translateComparisonExpressionWithJumpIfFalse(booleanCondition, endLabel)
         translate(stmt.body)
         out("  jmp  $whileLabel")
         out(endLabel)
@@ -940,41 +1000,29 @@ $counterVar    .byte  0""")
     }
 
     private fun translate(stmt: UntilLoop) {
+        checkBooleanExpression(stmt.condition)  // we require the condition to be of the form  'x <comparison> <value>'
+        val booleanCondition = stmt.condition as BinaryExpression
         val repeatLabel = makeLabel("repeat")
         val endLabel = makeLabel("repeatend")
         loopEndLabels.push(endLabel)
         out(repeatLabel)
         translate(stmt.body)
-        expressionsAsmGen.translateExpression(stmt.untilCondition)
-        val conditionDt = stmt.untilCondition.inferType(program)
-        if(!conditionDt.isKnown)
-            throw AssemblyError("unknown condition dt")
-        if(conditionDt.typeOrElse(DataType.BYTE) in ByteDatatypes) {
-            out("  inx |  lda  P8ESTACK_LO,x  |  beq  $repeatLabel")
-        } else {
-            out("""
-                inx
-                lda  P8ESTACK_LO,x
-                bne  +
-                lda  P8ESTACK_HI,x
-                beq  $repeatLabel
-+ """)
-        }
+        expressionsAsmGen.translateComparisonExpressionWithJumpIfFalse(booleanCondition, repeatLabel)
         out(endLabel)
         loopEndLabels.pop()
     }
 
     private fun translate(stmt: WhenStatement) {
-        expressionsAsmGen.translateExpression(stmt.condition)
         val endLabel = makeLabel("choice_end")
         val choiceBlocks = mutableListOf<Pair<String, AnonymousScope>>()
         val conditionDt = stmt.condition.inferType(program)
         if(!conditionDt.isKnown)
             throw AssemblyError("unknown condition dt")
         if(conditionDt.typeOrElse(DataType.BYTE) in ByteDatatypes)
-            out("  inx |  lda  P8ESTACK_LO,x")
+            assignExpressionToRegister(stmt.condition, RegisterOrPair.A)
         else
-            out("  inx |  lda  P8ESTACK_LO,x |  ldy  P8ESTACK_HI,x")
+            assignExpressionToRegister(stmt.condition, RegisterOrPair.AY)
+
         for(choice in stmt.choices) {
             val choiceLabel = makeLabel("choice")
             if(choice.values==null) {
@@ -982,7 +1030,7 @@ $counterVar    .byte  0""")
                 translate(choice.statements)
                 out("  jmp  $endLabel")
             } else {
-                choiceBlocks.add(Pair(choiceLabel, choice.statements))
+                choiceBlocks.add(choiceLabel to choice.statements)
                 for (cv in choice.values!!) {
                     val value = (cv as NumericLiteralValue).number.toInt()
                     if(conditionDt.typeOrElse(DataType.BYTE) in ByteDatatypes) {
@@ -999,6 +1047,7 @@ $counterVar    .byte  0""")
                 }
             }
         }
+        out("  jmp  $endLabel")
         for(choiceBlock in choiceBlocks) {
             out(choiceBlock.first)
             translate(choiceBlock.second)
@@ -1022,26 +1071,48 @@ $counterVar    .byte  0""")
 
         val jump = stmt.truepart.statements.first() as? Jump
         if(jump!=null) {
-            // branch with only a jump
+            // branch with only a jump (goto)
             val instruction = branchInstruction(stmt.condition, false)
             out("  $instruction  ${getJumpTarget(jump)}")
             translate(stmt.elsepart)
         } else {
+            val truePartIsJustBreak = stmt.truepart.statements.firstOrNull() is Break
+            val elsePartIsJustBreak = stmt.elsepart.statements.firstOrNull() is Break
             if(stmt.elsepart.containsNoCodeNorVars()) {
+                if(truePartIsJustBreak) {
+                    // branch with just a break (jump out of loop)
+                    val instruction = branchInstruction(stmt.condition, false)
+                    val loopEndLabel = loopEndLabels.peek()
+                    out("  $instruction  $loopEndLabel")
+                } else {
+                    val instruction = branchInstruction(stmt.condition, true)
+                    val elseLabel = makeLabel("branch_else")
+                    out("  $instruction  $elseLabel")
+                    translate(stmt.truepart)
+                    out(elseLabel)
+                }
+            }
+            else if(truePartIsJustBreak) {
+                // branch with just a break (jump out of loop)
+                val instruction = branchInstruction(stmt.condition, false)
+                val loopEndLabel = loopEndLabels.peek()
+                out("  $instruction  $loopEndLabel")
+                translate(stmt.elsepart)
+            } else if(elsePartIsJustBreak) {
+                // branch with just a break (jump out of loop) but true/false inverted
+                val instruction = branchInstruction(stmt.condition, true)
+                val loopEndLabel = loopEndLabels.peek()
+                out("  $instruction  $loopEndLabel")
+                translate(stmt.truepart)
+            } else {
                 val instruction = branchInstruction(stmt.condition, true)
                 val elseLabel = makeLabel("branch_else")
+                val endLabel = makeLabel("branch_end")
                 out("  $instruction  $elseLabel")
                 translate(stmt.truepart)
-                out(elseLabel)
-            } else {
-                val instruction = branchInstruction(stmt.condition, false)
-                val trueLabel = makeLabel("branch_true")
-                val endLabel = makeLabel("branch_end")
-                out("  $instruction  $trueLabel")
-                translate(stmt.elsepart)
                 out("  jmp  $endLabel")
-                out(trueLabel)
-                translate(stmt.truepart)
+                out(elseLabel)
+                translate(stmt.elsepart)
                 out(endLabel)
             }
         }
@@ -1087,7 +1158,9 @@ $counterVar    .byte  0""")
             "%breakpoint" -> {
                 val label = "_prog8_breakpoint_${breakpointLabels.size+1}"
                 breakpointLabels.add(label)
-                out("$label\tnop")
+                out("""
+                    nop
+$label              nop""")
             }
         }
     }
@@ -1113,7 +1186,28 @@ $counterVar    .byte  0""")
     }
 
     private fun translate(ret: Return) {
-        ret.value?.let { expressionsAsmGen.translateExpression(it) }
+        ret.value?.let { returnvalue ->
+            val sub = ret.definingSubroutine()!!
+            val returnType = sub.returntypes.single()
+            val returnReg = sub.asmReturnvaluesRegisters.single()
+            if(returnReg.registerOrPair==null)
+                throw AssemblyError("normal subroutines can't return value in status register directly")
+
+            when (returnType) {
+                in IntegerDatatypes -> {
+                    assignmentAsmGen.assignExpressionToRegister(returnvalue, returnReg.registerOrPair)
+                }
+                DataType.FLOAT -> {
+                    // return the float value via FAC1
+                    assignExpressionToRegister(returnvalue, RegisterOrPair.FAC1)
+                }
+                else -> {
+                    // all else take its address and assign that also to AY register pair
+                    val addrofValue = AddressOf(returnvalue as IdentifierReference, returnvalue.position)
+                    assignmentAsmGen.assignExpressionToRegister(addrofValue, returnReg.registerOrPair)
+                }
+            }
+        }
         out("  rts")
     }
 
@@ -1121,4 +1215,48 @@ $counterVar    .byte  0""")
         val assembly = asm.assembly.trimEnd().trimStart('\n')
         assemblyLines.add(assembly)
     }
+
+    internal fun signExtendAYlsb(valueDt: DataType) {
+        // sign extend signed byte in A to full word in AY
+        when(valueDt) {
+            DataType.UBYTE -> out("  ldy  #0")
+            DataType.BYTE -> out("  jsr  prog8_lib.sign_extend_AY_byte")
+            else -> throw AssemblyError("need byte type")
+        }
+    }
+
+    internal fun signExtendStackLsb(valueDt: DataType) {
+        // sign extend signed byte on stack to signed word on stack
+        when(valueDt) {
+            DataType.UBYTE -> {
+                if(CompilationTarget.instance.machine.cpu == CpuType.CPU65c02)
+                    out("  stz  P8ESTACK_HI+1,x")
+                else
+                    out("  lda  #0 |  sta  P8ESTACK_HI+1,x")
+            }
+            DataType.BYTE -> out("  jsr  prog8_lib.sign_extend_stack_byte")
+            else -> throw AssemblyError("need byte type")
+        }
+    }
+
+    internal fun signExtendVariableLsb(asmvar: String, valueDt: DataType) {
+        // sign extend signed byte in a var to a full word in that variable
+        when(valueDt) {
+            DataType.UBYTE -> {
+                if(CompilationTarget.instance.machine.cpu == CpuType.CPU65c02)
+                    out("  stz  $asmvar+1")
+                else
+                    out("  lda  #0 |  sta  $asmvar+1")
+            }
+            DataType.BYTE -> {
+                out("""
+                    lda  $asmvar
+                    ora  #$7f
+                    bmi  +
+                    lda  #0
++                   sta  $asmvar+1""")
+            }
+            else -> throw AssemblyError("need byte type")
+        }
+    }
 }

compiler/src/prog8/compiler/target/c64/codegen/AsmOptimizer.kt

@@ -75,11 +75,11 @@ private fun getLinesBy(lines: MutableList<String>, windowSize: Int) =
 
 private fun optimizeCmpSequence(linesByFour: List<List<IndexedValue<String>>>): List<Modification> {
     // the when statement (on bytes) generates a sequence of:
-    //	 lda $ce01,x
-    //	 cmp #$20
+    //	 lda  $ce01,x
+    //	 cmp  #$20
     //	 beq  check_prog8_s72choice_32
-    //	 lda $ce01,x
-    //	 cmp #$21
+    //	 lda  $ce01,x
+    //	 cmp  #$21
     //	 beq  check_prog8_s73choice_33
     // the repeated lda can be removed
     val mods = mutableListOf<Modification>()
@@ -151,7 +151,8 @@ private fun optimizeSameAssignments(linesByFourteen: List<List<IndexedValue<Stri
         }
 
         if(first.startsWith("lda") && second.startsWith("ldy") && third.startsWith("sta") && fourth.startsWith("sty") &&
-                fifth.startsWith("lda") && sixth.startsWith("ldy") && seventh.startsWith("jsr  c64flt.copy_float")) {
+                fifth.startsWith("lda") && sixth.startsWith("ldy") &&
+                (seventh.startsWith("jsr  floats.copy_float") || seventh.startsWith("jsr  cx16flt.copy_float"))) {
 
             val nineth = pair[8].value.trimStart()
             val tenth = pair[9].value.trimStart()
@@ -161,7 +162,8 @@ private fun optimizeSameAssignments(linesByFourteen: List<List<IndexedValue<Stri
             val fourteenth = pair[13].value.trimStart()
 
             if(eighth.startsWith("lda") && nineth.startsWith("ldy") && tenth.startsWith("sta") && eleventh.startsWith("sty") &&
-                    twelveth.startsWith("lda") && thirteenth.startsWith("ldy") && fourteenth.startsWith("jsr  c64flt.copy_float")) {
+                    twelveth.startsWith("lda") && thirteenth.startsWith("ldy") &&
+                    (fourteenth.startsWith("jsr  floats.copy_float") || fourteenth.startsWith("jsr  cx16flt.copy_float"))) {
 
                 if(first.substring(4) == eighth.substring(4) && second.substring(4)==nineth.substring(4)) {
                     // identical float init
@@ -178,6 +180,7 @@ private fun optimizeSameAssignments(linesByFourteen: List<List<IndexedValue<Stri
 
 private fun optimizeStoreLoadSame(linesByFour: List<List<IndexedValue<String>>>): List<Modification> {
     // sta X + lda X,  sty X + ldy X,   stx X + ldx X  -> the second instruction can be eliminated
+    // TODO this is not true if X is not a regular RAM memory address (but instead mapped I/O or ROM)
     val mods = mutableListOf<Modification>()
     for (pair in linesByFour) {
         val first = pair[0].value.trimStart()
@@ -193,8 +196,8 @@ private fun optimizeStoreLoadSame(linesByFour: List<List<IndexedValue<String>>>)
                 (first.startsWith("sty ") && second.startsWith("ldy ")) ||
                 (first.startsWith("stx ") && second.startsWith("ldx "))
         ) {
-            val firstLoc = first.substring(4)
-            val secondLoc = second.substring(4)
+            val firstLoc = first.substring(4).trimStart()
+            val secondLoc = second.substring(4).trimStart()
             if (firstLoc == secondLoc) {
                 mods.add(Modification(pair[1].index, true, null))
             }

compiler/src/prog8/compiler/target/c64/codegen/ForLoopsAsmGen.kt

@@ -1,15 +1,13 @@
 package prog8.compiler.target.c64.codegen
 
 import prog8.ast.Program
+import prog8.ast.base.ArrayElementTypes
 import prog8.ast.base.DataType
+import prog8.ast.base.RegisterOrPair
 import prog8.ast.expressions.IdentifierReference
 import prog8.ast.expressions.RangeExpr
 import prog8.ast.statements.ForLoop
 import prog8.compiler.AssemblyError
-import prog8.compiler.target.c64.codegen.assignment.AsmAssignSource
-import prog8.compiler.target.c64.codegen.assignment.AsmAssignTarget
-import prog8.compiler.target.c64.codegen.assignment.AsmAssignment
-import prog8.compiler.target.c64.codegen.assignment.TargetStorageKind
 import prog8.compiler.toHex
 import kotlin.math.absoluteValue
 
@@ -18,7 +16,7 @@ internal class ForLoopsAsmGen(private val program: Program, private val asmgen:
     internal fun translate(stmt: ForLoop) {
         val iterableDt = stmt.iterable.inferType(program)
         if(!iterableDt.isKnown)
-            throw AssemblyError("can't determine iterable dt")
+            throw AssemblyError("unknown dt")
         when(stmt.iterable) {
             is RangeExpr -> {
                 val range = (stmt.iterable as RangeExpr).toConstantIntegerRange()
@@ -46,28 +44,22 @@ internal class ForLoopsAsmGen(private val program: Program, private val asmgen:
             DataType.ARRAY_B, DataType.ARRAY_UB -> {
                 if (stepsize==1 || stepsize==-1) {
 
-                    // bytes, step 1 or -1
+                    // bytes array, step 1 or -1
 
                     val incdec = if(stepsize==1) "inc" else "dec"
                     // loop over byte range via loopvar
                     val varname = asmgen.asmVariableName(stmt.loopVar)
-                    asmgen.translateExpression(range.to)
-                    asmgen.translateExpression(range.from)
+                    asmgen.assignExpressionToVariable(range.from, varname, ArrayElementTypes.getValue(iterableDt), null)
+                    asmgen.assignExpressionToVariable(range.to, "$modifiedLabel+1", ArrayElementTypes.getValue(iterableDt), null)
+                    asmgen.out(loopLabel)
+                    asmgen.translate(stmt.body)
                     asmgen.out("""
-                inx
-                lda  P8ESTACK_LO,x
-                sta  $varname
-                lda  P8ESTACK_LO+1,x
-                sta  $modifiedLabel+1
-$loopLabel""")
-                        asmgen.translate(stmt.body)
-                        asmgen.out("""
-                lda  $varname
-$modifiedLabel  cmp  #0         ; modified 
-                beq  $endLabel
-                $incdec  $varname
-                jmp  $loopLabel
-$endLabel       inx""")
+                        lda  $varname
+$modifiedLabel          cmp  #0         ; modified 
+                        beq  $endLabel
+                        $incdec  $varname
+                        jmp  $loopLabel
+$endLabel""")
 
                 } else {
 
@@ -75,36 +67,29 @@ $endLabel       inx""")
 
                     // loop over byte range via loopvar
                     val varname = asmgen.asmVariableName(stmt.loopVar)
-                    asmgen.translateExpression(range.to)
-                    asmgen.translateExpression(range.from)
-                    asmgen.out("""
-                inx
-                lda  P8ESTACK_LO,x
-                sta  $varname
-                lda  P8ESTACK_LO+1,x
-                sta  $modifiedLabel+1
-$loopLabel""")
+                    asmgen.assignExpressionToVariable(range.from, varname, ArrayElementTypes.getValue(iterableDt), null)
+                    asmgen.assignExpressionToVariable(range.to, "$modifiedLabel+1", ArrayElementTypes.getValue(iterableDt), null)
+                    asmgen.out(loopLabel)
                     asmgen.translate(stmt.body)
-                    asmgen.out("""
-                lda  $varname""")
                     if(stepsize>0) {
                         asmgen.out("""
-                clc
-                adc  #$stepsize
-                sta  $varname
-$modifiedLabel  cmp  #0    ; modified
-                bcc  $loopLabel
-                beq  $loopLabel""")
+                            lda  $varname
+                            clc
+                            adc  #$stepsize
+                            sta  $varname
+$modifiedLabel              cmp  #0    ; modified
+                            bmi  $loopLabel
+                            beq  $loopLabel""")
                     } else {
                         asmgen.out("""
-                sec
-                sbc  #${stepsize.absoluteValue}
-                sta  $varname
-$modifiedLabel  cmp  #0     ; modified 
-                bcs  $loopLabel""")
+                            lda  $varname
+                            sec
+                            sbc  #${stepsize.absoluteValue}
+                            sta  $varname
+$modifiedLabel              cmp  #0     ; modified
+                            bpl  $loopLabel""")
                     }
-                    asmgen.out("""
-$endLabel       inx""")
+                    asmgen.out(endLabel)
                 }
             }
             DataType.ARRAY_W, DataType.ARRAY_UW -> {
@@ -113,13 +98,11 @@ $endLabel       inx""")
                     // words, step 1 or -1
 
                     stepsize == 1 || stepsize == -1 -> {
-                        asmgen.translateExpression(range.to)
-                        assignLoopvar(stmt, range)
                         val varname = asmgen.asmVariableName(stmt.loopVar)
+                        assignLoopvar(stmt, range)
+                        asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
                         asmgen.out("""
-                            lda  P8ESTACK_HI+1,x
-                            sta  $modifiedLabel+1
-                            lda  P8ESTACK_LO+1,x
+                            sty  $modifiedLabel+1
                             sta  $modifiedLabel2+1
 $loopLabel""")
                         asmgen.translate(stmt.body)
@@ -146,21 +129,17 @@ $modifiedLabel2 cmp  #0    ; modified
                 jmp  $loopLabel""")
                         }
                         asmgen.out(endLabel)
-                        asmgen.out(" inx")
                     }
                     stepsize > 0 -> {
 
                         // (u)words, step >= 2
-                        asmgen.translateExpression(range.to)
-                        asmgen.out("""
-                            lda  P8ESTACK_HI+1,x
-                            sta  $modifiedLabel+1
-                            lda  P8ESTACK_LO+1,x
-                            sta  $modifiedLabel2+1
-                        """)
-                        assignLoopvar(stmt, range)
                         val varname = asmgen.asmVariableName(stmt.loopVar)
-                        asmgen.out(loopLabel)
+                        assignLoopvar(stmt, range)
+                        asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
+                        asmgen.out("""
+                            sty  $modifiedLabel+1
+                            sta  $modifiedLabel2+1
+$loopLabel""")
                         asmgen.translate(stmt.body)
 
                         if (iterableDt == DataType.ARRAY_UW) {
@@ -179,7 +158,7 @@ $modifiedLabel2 lda  #0     ; modified
                 cmp  $varname
                 bcc  $endLabel
                 bcs  $loopLabel
-$endLabel       inx""")
+$endLabel""")
                         } else {
                             asmgen.out("""
                 lda  $varname
@@ -196,22 +175,19 @@ $modifiedLabel  lda  #0   ; modified
                 bvc  +
                 eor  #$80
 +               bpl  $loopLabel                
-$endLabel       inx""")
+$endLabel""")
                         }
                     }
                     else -> {
 
                         // (u)words, step <= -2
-                        asmgen.translateExpression(range.to)
-                        asmgen.out("""
-                            lda  P8ESTACK_HI+1,x
-                            sta  $modifiedLabel+1
-                            lda  P8ESTACK_LO+1,x
-                            sta  $modifiedLabel2+1
-                        """)
-                        assignLoopvar(stmt, range)
                         val varname = asmgen.asmVariableName(stmt.loopVar)
-                        asmgen.out(loopLabel)
+                        assignLoopvar(stmt, range)
+                        asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
+                        asmgen.out("""
+                            sty  $modifiedLabel+1
+                            sta  $modifiedLabel2+1
+$loopLabel""")
                         asmgen.translate(stmt.body)
 
                         if(iterableDt==DataType.ARRAY_UW) {
@@ -229,7 +205,7 @@ $modifiedLabel  cmp  #0    ; modified
                 lda  $varname
 $modifiedLabel2 cmp  #0    ; modified 
                 bcs  $loopLabel
-$endLabel       inx""")
+$endLabel""")
                         } else {
                             asmgen.out("""
                 lda  $varname
@@ -247,7 +223,7 @@ $modifiedLabel  sbc  #0    ; modified
                 bvc  +
                 eor  #$80
 +               bpl  $loopLabel                
-$endLabel       inx""")
+$endLabel""")
                         }
                     }
                 }
@@ -258,13 +234,6 @@ $endLabel       inx""")
         asmgen.loopEndLabels.pop()
     }
 
-    private fun assignLoopvar(stmt: ForLoop, range: RangeExpr) {
-        val target = AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, stmt.loopVarDt(program).typeOrElse(DataType.STRUCT), variable=stmt.loopVar)
-        val src = AsmAssignSource.fromAstSource(range.from, program).adjustDataTypeToTarget(target)
-        val assign = AsmAssignment(src, target, false, range.position)
-        asmgen.translateNormalAssignment(assign)
-    }
-
     private fun translateForOverIterableVar(stmt: ForLoop, iterableDt: DataType, ident: IdentifierReference) {
         val loopLabel = asmgen.makeLabel("for_loop")
         val endLabel = asmgen.makeLabel("for_end")
@@ -615,4 +584,7 @@ $loopLabel""")
 $endLabel""")
         asmgen.loopEndLabels.pop()
     }
+
+    private fun assignLoopvar(stmt: ForLoop, range: RangeExpr) =
+        asmgen.assignExpressionToVariable(range.from, asmgen.asmVariableName(stmt.loopVar), stmt.loopVarDt(program).typeOrElse(DataType.STRUCT), stmt.definingSubroutine())
 }

compiler/src/prog8/compiler/target/c64/codegen/FunctionCallAsmGen.kt

@@ -1,9 +1,11 @@
 package prog8.compiler.target.c64.codegen
 
 import prog8.ast.IFunctionCall
+import prog8.ast.Node
 import prog8.ast.Program
 import prog8.ast.base.*
 import prog8.ast.expressions.*
+import prog8.ast.statements.RegisterOrStatusflag
 import prog8.ast.statements.Subroutine
 import prog8.ast.statements.SubroutineParameter
 import prog8.compiler.AssemblyError
@@ -12,13 +14,23 @@ import prog8.compiler.target.c64.codegen.assignment.*
 
 internal class FunctionCallAsmGen(private val program: Program, private val asmgen: AsmGen) {
 
-    internal fun translateFunctionCall(stmt: IFunctionCall) {
+    internal fun translateFunctionCallStatement(stmt: IFunctionCall) {
+        val sub = stmt.target.targetSubroutine(program.namespace)!!
+        val preserveStatusRegisterAfterCall = sub.shouldPreserveStatusRegisterAfterCall()
+        translateFunctionCall(stmt, preserveStatusRegisterAfterCall)
+        // functioncalls no longer return results on the stack, so simply ignore the results in the registers
+        if(preserveStatusRegisterAfterCall)
+            asmgen.out("  plp\t; restore status flags from call")
+    }
+
+
+    internal fun translateFunctionCall(stmt: IFunctionCall, preserveStatusRegisterAfterCall: Boolean) {
         // output the code to setup the parameters and perform the actual call
         // does NOT output the code to deal with the result values!
         val sub = stmt.target.targetSubroutine(program.namespace) ?: throw AssemblyError("undefined subroutine ${stmt.target}")
-        val saveX = CpuRegister.X in sub.asmClobbers || sub.regXasResult()
+        val saveX = sub.shouldSaveX()
         if(saveX)
-            asmgen.out("  stx  P8ZP_SCRATCH_REG_X")        // we only save X for now (required! is the eval stack pointer), screw A and Y...
+            asmgen.saveRegister(CpuRegister.X, preserveStatusRegisterAfterCall, (stmt as Node).definingSubroutine()!!)
 
         val subName = asmgen.asmSymbolName(stmt.target)
         if(stmt.args.isNotEmpty()) {
@@ -31,7 +43,7 @@ internal class FunctionCallAsmGen(private val program: Program, private val asmg
                 // via registers
                 if(sub.parameters.size==1) {
                     // just a single parameter, no risk of clobbering registers
-                    argumentViaRegister(sub, sub.parameters.withIndex().single(), stmt.args[0])
+                    argumentViaRegister(sub, IndexedValue(0, sub.parameters.single()), stmt.args[0])
                 } else {
                     // multiple register arguments, risk of register clobbering.
                     // evaluate arguments onto the stack, and load the registers from the evaluated values on the stack.
@@ -56,65 +68,111 @@ internal class FunctionCallAsmGen(private val program: Program, private val asmg
         }
         asmgen.out("  jsr  $subName")
 
+        if(preserveStatusRegisterAfterCall) {
+            asmgen.out("  php\t; save status flags from call")
+            // note: the containing statement (such as the FunctionCallStatement or the Assignment or the Expression)
+            //       must take care of popping this value again at the end!
+        }
+
         if(saveX)
-            asmgen.out("  ldx  P8ZP_SCRATCH_REG_X")        // restore X again
+            asmgen.restoreRegister(CpuRegister.X, preserveStatusRegisterAfterCall)
     }
 
     private fun registerArgsViaStackEvaluation(stmt: IFunctionCall, sub: Subroutine) {
         // this is called when one or more of the arguments are 'complex' and
         // cannot be assigned to a register easily or risk clobbering other registers.
+
+        if(sub.parameters.isEmpty())
+            return
+
+        // 1. load all arguments reversed onto the stack: first arg goes last (is on top).
         for (arg in stmt.args.reversed())
             asmgen.translateExpression(arg)
-        for (regparam in sub.asmParameterRegisters) {
+
+        var argForCarry: IndexedValue<Pair<Expression, RegisterOrStatusflag>>? = null
+        var argForXregister: IndexedValue<Pair<Expression, RegisterOrStatusflag>>? = null
+        var argForAregister: IndexedValue<Pair<Expression, RegisterOrStatusflag>>? = null
+
+        asmgen.out("  inx")     // align estack pointer
+
+        for(argi in stmt.args.zip(sub.asmParameterRegisters).withIndex()) {
             when {
-                regparam.statusflag==Statusflag.Pc -> {
-                    asmgen.out("""
-                        inx
-                        pha
-                        lda  P8ESTACK_LO,x
-                        beq  +
-                        sec  
-                        bcs  ++
-+                       clc
-+                       pla""")
+                argi.value.second.statusflag == Statusflag.Pc -> {
+                    require(argForCarry == null)
+                    argForCarry = argi
                 }
-                regparam.statusflag!=null -> {
-                    throw AssemblyError("can only use Carry as status flag parameter")
+                argi.value.second.statusflag != null -> throw AssemblyError("can only use Carry as status flag parameter")
+                argi.value.second.registerOrPair in setOf(RegisterOrPair.X, RegisterOrPair.AX, RegisterOrPair.XY) -> {
+                    require(argForXregister==null)
+                    argForXregister = argi
                 }
-                regparam.registerOrPair!=null -> {
-                    val tgt = AsmAssignTarget.fromRegisters(regparam.registerOrPair, program, asmgen)
-                    val source = AsmAssignSource(SourceStorageKind.STACK, program, tgt.datatype)
-                    val asgn = AsmAssignment(source, tgt, false, Position.DUMMY)
-                    asmgen.translateNormalAssignment(asgn)
+                argi.value.second.registerOrPair in setOf(RegisterOrPair.A, RegisterOrPair.AY) -> {
+                    require(argForAregister == null)
+                    argForAregister = argi
                 }
-                else -> {}
+                argi.value.second.registerOrPair == RegisterOrPair.Y -> {
+                    asmgen.out("  ldy  P8ESTACK_LO+${argi.index},x")
+                }
+                else -> throw AssemblyError("weird argument")
             }
         }
+
+        if(argForCarry!=null) {
+            asmgen.out("""
+                lda  P8ESTACK_LO+${argForCarry.index},x
+                beq  +
+                sec
+                bcs  ++
++               clc
++               php""")             // push the status flags
+        }
+
+        if(argForAregister!=null) {
+            when(argForAregister.value.second.registerOrPair) {
+                RegisterOrPair.A -> asmgen.out("  lda  P8ESTACK_LO+${argForAregister.index},x")
+                RegisterOrPair.AY -> asmgen.out("  lda  P8ESTACK_LO+${argForAregister.index},x |  ldy  P8ESTACK_HI+${argForAregister.index},x")
+                else -> throw AssemblyError("weird arg")
+            }
+        }
+
+        if(argForXregister!=null) {
+
+            if(argForAregister!=null)
+                asmgen.out("  pha")
+            when(argForXregister.value.second.registerOrPair) {
+                RegisterOrPair.X -> asmgen.out("  lda  P8ESTACK_LO+${argForXregister.index},x |  tax")
+                RegisterOrPair.AX -> asmgen.out("  ldy  P8ESTACK_LO+${argForXregister.index},x |  lda  P8ESTACK_HI+${argForXregister.index},x |  tax |  tya")
+                RegisterOrPair.XY -> asmgen.out("  ldy  P8ESTACK_HI+${argForXregister.index},x |  lda  P8ESTACK_LO+${argForXregister.index},x |  tax")
+                else -> throw AssemblyError("weird arg")
+            }
+            if(argForAregister!=null)
+                asmgen.out("  pla")
+        } else {
+            repeat(sub.parameters.size - 1) { asmgen.out("  inx") }       // unwind stack
+        }
+
+        if(argForCarry!=null)
+            asmgen.out("  plp")       // set the carry flag back to correct value
     }
 
     private fun argumentViaVariable(sub: Subroutine, parameter: IndexedValue<SubroutineParameter>, value: Expression) {
         // pass parameter via a regular variable (not via registers)
         val valueIDt = value.inferType(program)
         if(!valueIDt.isKnown)
-            throw AssemblyError("arg type unknown")
+            throw AssemblyError("unknown dt")
         val valueDt = valueIDt.typeOrElse(DataType.STRUCT)
         if(!isArgumentTypeCompatible(valueDt, parameter.value.type))
             throw AssemblyError("argument type incompatible")
 
-        val scopedParamVar = (sub.scopedname+"."+parameter.value.name).split(".")
-        val identifier = IdentifierReference(scopedParamVar, sub.position)
-        identifier.linkParents(value.parent)
-        val tgt = AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, parameter.value.type, variable = identifier)
-        val source = AsmAssignSource.fromAstSource(value, program).adjustDataTypeToTarget(tgt)
-        val asgn = AsmAssignment(source, tgt, false, Position.DUMMY)
-        asmgen.translateNormalAssignment(asgn)
+        val varName = asmgen.asmVariableName(sub.scopedname+"."+parameter.value.name)
+        asmgen.assignExpressionToVariable(value, varName, parameter.value.type, sub)
     }
 
     private fun argumentViaRegister(sub: Subroutine, parameter: IndexedValue<SubroutineParameter>, value: Expression) {
         // pass argument via a register parameter
         val valueIDt = value.inferType(program)
         if(!valueIDt.isKnown)
-            throw AssemblyError("arg type unknown")
+            throw AssemblyError("unknown dt")
         val valueDt = valueIDt.typeOrElse(DataType.STRUCT)
         if(!isArgumentTypeCompatible(valueDt, parameter.value.type))
             throw AssemblyError("argument type incompatible")
@@ -122,14 +180,15 @@ internal class FunctionCallAsmGen(private val program: Program, private val asmg
         val paramRegister = sub.asmParameterRegisters[parameter.index]
         val statusflag = paramRegister.statusflag
         val register = paramRegister.registerOrPair
-        val stack = paramRegister.stack
+        val requiredDt = parameter.value.type
+        if(requiredDt!=valueDt) {
+            if(valueDt largerThan requiredDt)
+                throw AssemblyError("can only convert byte values to word param types")
+        }
         when {
-            stack -> {
-                // push arg onto the stack
-                // note: argument order is reversed (first argument will be deepest on the stack)
-                asmgen.translateExpression(value)
-            }
             statusflag!=null -> {
+                if(requiredDt!=valueDt)
+                    throw AssemblyError("for statusflag, byte value is required")
                 if (statusflag == Statusflag.Pc) {
                     // this param needs to be set last, right before the jsr
                     // for now, this is already enforced on the subroutine definition by the Ast Checker
@@ -151,17 +210,13 @@ internal class FunctionCallAsmGen(private val program: Program, private val asmg
 """)
                         }
                         else -> {
-                            asmgen.translateExpression(value)
+                            asmgen.assignExpressionToRegister(value, RegisterOrPair.A)
                             asmgen.out("""
-            inx
-            pha
-            lda  P8ESTACK_LO,x
-            beq  +
-            sec  
-            bcs  ++
-+           clc
-+           pla
-""")
+                                beq  +
+                                sec
+                                bcs  ++
++                               clc
++""")
                         }
                     }
                 }
@@ -169,15 +224,28 @@ internal class FunctionCallAsmGen(private val program: Program, private val asmg
             }
             else -> {
                 // via register or register pair
-                val target = AsmAssignTarget.fromRegisters(register!!, program, asmgen)
-                val src = if(valueDt in PassByReferenceDatatypes) {
-                    val addr = AddressOf(value as IdentifierReference, Position.DUMMY)
-                    AsmAssignSource.fromAstSource(addr, program).adjustDataTypeToTarget(target)
-                } else {
-                    AsmAssignSource.fromAstSource(value, program).adjustDataTypeToTarget(target)
+                register!!
+                if(requiredDt largerThan valueDt) {
+                    // we need to sign extend the source, do this via temporary word variable
+                    val scratchVar = asmgen.asmVariableName("P8ZP_SCRATCH_W1")
+                    asmgen.assignExpressionToVariable(value, scratchVar, DataType.UBYTE, sub)
+                    asmgen.signExtendVariableLsb(scratchVar, valueDt)
+                    asmgen.assignVariableToRegister(scratchVar, register)
+                }
+                else {
+                    val target = AsmAssignTarget.fromRegisters(register, sub, program, asmgen)
+                    val src = if(valueDt in PassByReferenceDatatypes) {
+                        if(value is IdentifierReference) {
+                            val addr = AddressOf(value, Position.DUMMY)
+                            AsmAssignSource.fromAstSource(addr, program, asmgen).adjustSignedUnsigned(target)
+                        } else {
+                            AsmAssignSource.fromAstSource(value, program, asmgen).adjustSignedUnsigned(target)
+                        }
+                    } else {
+                        AsmAssignSource.fromAstSource(value, program, asmgen).adjustSignedUnsigned(target)
+                    }
+                    asmgen.translateNormalAssignment(AsmAssignment(src, target, false, Position.DUMMY))
                 }
-
-                asmgen.translateNormalAssignment(AsmAssignment(src, target, false, Position.DUMMY))
             }
         }
     }

compiler/src/prog8/compiler/target/c64/codegen/PostIncrDecrAsmGen.kt

@@ -15,10 +15,11 @@ internal class PostIncrDecrAsmGen(private val program: Program, private val asmg
         val targetIdent = stmt.target.identifier
         val targetMemory = stmt.target.memoryAddress
         val targetArrayIdx = stmt.target.arrayindexed
+        val scope = stmt.definingSubroutine()
         when {
             targetIdent!=null -> {
                 val what = asmgen.asmVariableName(targetIdent)
-                when (stmt.target.inferType(program, stmt).typeOrElse(DataType.STRUCT)) {
+                when (stmt.target.inferType(program).typeOrElse(DataType.STRUCT)) {
                     in ByteDatatypes -> asmgen.out(if (incr) "  inc  $what" else "  dec  $what")
                     in WordDatatypes -> {
                         if(incr)
@@ -33,7 +34,7 @@ internal class PostIncrDecrAsmGen(private val program: Program, private val asmg
                     }
                     DataType.FLOAT -> {
                         asmgen.out("  lda  #<$what |  ldy  #>$what")
-                        asmgen.out(if(incr) "  jsr  c64flt.inc_var_f" else "  jsr  c64flt.dec_var_f")
+                        asmgen.out(if(incr) "  jsr  floats.inc_var_f" else "  jsr  floats.dec_var_f")
                     }
                     else -> throw AssemblyError("need numeric type")
                 }
@@ -53,14 +54,8 @@ internal class PostIncrDecrAsmGen(private val program: Program, private val asmg
                             asmgen.out("+\tdec  ${'$'}ffff\t; modified")
                     }
                     else -> {
-                        asmgen.translateExpression(addressExpr)
-                        asmgen.out("""
-                            inx
-                            lda  P8ESTACK_LO,x
-                            sta  (+) + 1
-                            lda  P8ESTACK_HI,x
-                            sta  (+) + 2
-                        """)
+                        asmgen.assignExpressionToRegister(addressExpr, RegisterOrPair.AY)
+                        asmgen.out("  sta  (+) + 1 |  sty  (+) + 2")
                         if(incr)
                             asmgen.out("+\tinc  ${'$'}ffff\t; modified")
                         else
@@ -69,66 +64,64 @@ internal class PostIncrDecrAsmGen(private val program: Program, private val asmg
                 }
             }
             targetArrayIdx!=null -> {
-                val index = targetArrayIdx.arrayspec.index
-                val asmArrayvarname = asmgen.asmVariableName(targetArrayIdx.identifier)
+                val asmArrayvarname = asmgen.asmVariableName(targetArrayIdx.arrayvar)
                 val elementDt = targetArrayIdx.inferType(program).typeOrElse(DataType.STRUCT)
-                when(index) {
-                    is NumericLiteralValue -> {
-                        val indexValue = index.number.toInt() * elementDt.memorySize()
-                        when(elementDt) {
-                            in ByteDatatypes -> asmgen.out(if (incr) "  inc  $asmArrayvarname+$indexValue" else "  dec  $asmArrayvarname+$indexValue")
-                            in WordDatatypes -> {
-                                if(incr)
-                                    asmgen.out(" inc  $asmArrayvarname+$indexValue |  bne  + |  inc  $asmArrayvarname+$indexValue+1 |+")
-                                else
-                                    asmgen.out("""
-        lda  $asmArrayvarname+$indexValue
-        bne  +
-        dec  $asmArrayvarname+$indexValue+1
+                if(targetArrayIdx.indexer.indexNum!=null) {
+                    val indexValue = targetArrayIdx.indexer.constIndex()!! * elementDt.memorySize()
+                    when(elementDt) {
+                        in ByteDatatypes -> asmgen.out(if (incr) "  inc  $asmArrayvarname+$indexValue" else "  dec  $asmArrayvarname+$indexValue")
+                        in WordDatatypes -> {
+                            if(incr)
+                                asmgen.out(" inc  $asmArrayvarname+$indexValue |  bne  + |  inc  $asmArrayvarname+$indexValue+1 |+")
+                            else
+                                asmgen.out("""
+    lda  $asmArrayvarname+$indexValue
+    bne  +
+    dec  $asmArrayvarname+$indexValue+1
 +       dec  $asmArrayvarname+$indexValue 
 """)
-                            }
-                            DataType.FLOAT -> {
-                                asmgen.out("  lda  #<$asmArrayvarname+$indexValue |  ldy  #>$asmArrayvarname+$indexValue")
-                                asmgen.out(if(incr) "  jsr  c64flt.inc_var_f" else "  jsr  c64flt.dec_var_f")
-                            }
-                            else -> throw AssemblyError("need numeric type")
                         }
-                    }
-                    else -> {
-                        asmgen.loadScaledArrayIndexIntoRegister(targetArrayIdx, elementDt, CpuRegister.A)
-                        asmgen.out("  stx  P8ZP_SCRATCH_REG_X |  tax")
-                        when(elementDt) {
-                            in ByteDatatypes -> {
-                                asmgen.out(if(incr) "  inc  $asmArrayvarname,x" else "  dec  $asmArrayvarname,x")
-                            }
-                            in WordDatatypes -> {
-                                if(incr)
-                                    asmgen.out(" inc  $asmArrayvarname,x |  bne  + |  inc  $asmArrayvarname+1,x |+")
-                                else
-                                    asmgen.out("""
-        lda  $asmArrayvarname,x
-        bne  +
-        dec  $asmArrayvarname+1,x
-+       dec  $asmArrayvarname 
-""")
-                            }
-                            DataType.FLOAT -> {
-                                asmgen.out("""
-                        ldy  #>$asmArrayvarname
-                        clc
-                        adc  #<$asmArrayvarname
-                        bcc  +
-                        iny
-+                       jsr  c64flt.inc_var_f""")
-                            }
-                            else -> throw AssemblyError("weird array elt dt")
+                        DataType.FLOAT -> {
+                            asmgen.out("  lda  #<$asmArrayvarname+$indexValue |  ldy  #>$asmArrayvarname+$indexValue")
+                            asmgen.out(if(incr) "  jsr  floats.inc_var_f" else "  jsr  floats.dec_var_f")
                         }
-                        asmgen.out("  ldx  P8ZP_SCRATCH_REG_X")
+                        else -> throw AssemblyError("need numeric type")
                     }
                 }
+                else
+                {
+                    asmgen.loadScaledArrayIndexIntoRegister(targetArrayIdx, elementDt, CpuRegister.A)
+                    asmgen.saveRegister(CpuRegister.X, false, scope!!)
+                    asmgen.out("  tax")
+                    when(elementDt) {
+                        in ByteDatatypes -> {
+                            asmgen.out(if(incr) "  inc  $asmArrayvarname,x" else "  dec  $asmArrayvarname,x")
+                        }
+                        in WordDatatypes -> {
+                            if(incr)
+                                asmgen.out(" inc  $asmArrayvarname,x |  bne  + |  inc  $asmArrayvarname+1,x |+")
+                            else
+                                asmgen.out("""
+    lda  $asmArrayvarname,x
+    bne  +
+    dec  $asmArrayvarname+1,x
++       dec  $asmArrayvarname 
+""")
+                        }
+                        DataType.FLOAT -> {
+                            asmgen.out("""
+                    ldy  #>$asmArrayvarname
+                    clc
+                    adc  #<$asmArrayvarname
+                    bcc  +
+                    iny
++                       jsr  floats.inc_var_f""")
+                        }
+                        else -> throw AssemblyError("weird array elt dt")
+                    }
+                    asmgen.restoreRegister(CpuRegister.X, false)
+                }
             }
-            else -> throw AssemblyError("weird target type ${stmt.target}")
         }
     }
 }

compiler/src/prog8/compiler/target/c64/codegen/assignment/AsmAssignment.kt

@@ -3,9 +3,7 @@ package prog8.compiler.target.c64.codegen.assignment
 import prog8.ast.Program
 import prog8.ast.base.*
 import prog8.ast.expressions.*
-import prog8.ast.statements.AssignTarget
-import prog8.ast.statements.Assignment
-import prog8.ast.statements.DirectMemoryWrite
+import prog8.ast.statements.*
 import prog8.compiler.AssemblyError
 import prog8.compiler.target.c64.codegen.AsmGen
 
@@ -29,10 +27,11 @@ internal enum class SourceStorageKind {
 }
 
 internal class AsmAssignTarget(val kind: TargetStorageKind,
-                               program: Program,
-                               asmgen: AsmGen,
+                               private val program: Program,
+                               private val asmgen: AsmGen,
                                val datatype: DataType,
-                               val variable: IdentifierReference? = null,
+                               val scope: Subroutine?,
+                               private val variableAsmName: String? = null,
                                val array: ArrayIndexedExpression? = null,
                                val memory: DirectMemoryWrite? = null,
                                val register: RegisterOrPair? = null,
@@ -40,51 +39,53 @@ internal class AsmAssignTarget(val kind: TargetStorageKind,
                                )
 {
     val constMemoryAddress by lazy { memory?.addressExpression?.constValue(program)?.number?.toInt() ?: 0}
-    val constArrayIndexValue by lazy { array?.arrayspec?.constIndex() }
-    val vardecl by lazy { variable?.targetVarDecl(program.namespace)!! }
-    val asmVarname by lazy {
-        if(variable!=null)
-            asmgen.asmVariableName(variable)
+    val constArrayIndexValue by lazy { array?.indexer?.constIndex() }
+    val asmVarname: String
+        get() = if(array==null)
+            variableAsmName!!
         else
-            asmgen.asmVariableName(array!!.identifier)
-    }
+            asmgen.asmVariableName(array.arrayvar)
 
     lateinit var origAssign: AsmAssignment
 
     init {
-        if(variable!=null && vardecl.type == VarDeclType.CONST)
-            throw AssemblyError("can't assign to a constant")
-        if(register!=null && datatype !in IntegerDatatypes)
-            throw AssemblyError("register must be integer type")
+        if(register!=null && datatype !in NumericDatatypes)
+            throw AssemblyError("register must be integer or float type")
     }
 
     companion object {
         fun fromAstAssignment(assign: Assignment, program: Program, asmgen: AsmGen): AsmAssignTarget = with(assign.target) {
-            val dt = inferType(program, assign).typeOrElse(DataType.STRUCT)
+            val idt = inferType(program)
+            if(!idt.isKnown)
+                throw AssemblyError("unknown dt")
+            val dt = idt.typeOrElse(DataType.STRUCT)
             when {
-                identifier != null -> AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, dt, variable=identifier, origAstTarget =  this)
-                arrayindexed != null -> AsmAssignTarget(TargetStorageKind.ARRAY, program, asmgen, dt, array = arrayindexed, origAstTarget =  this)
-                memoryAddress != null -> AsmAssignTarget(TargetStorageKind.MEMORY, program, asmgen, dt, memory =  memoryAddress, origAstTarget =  this)
+                identifier != null -> AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, dt, assign.definingSubroutine(), variableAsmName = asmgen.asmVariableName(identifier!!), origAstTarget =  this)
+                arrayindexed != null -> AsmAssignTarget(TargetStorageKind.ARRAY, program, asmgen, dt, assign.definingSubroutine(), array = arrayindexed, origAstTarget =  this)
+                memoryAddress != null -> AsmAssignTarget(TargetStorageKind.MEMORY, program, asmgen, dt, assign.definingSubroutine(), memory =  memoryAddress, origAstTarget =  this)
                 else -> throw AssemblyError("weird target")
             }
         }
 
-        fun fromRegisters(registers: RegisterOrPair, program: Program, asmgen: AsmGen): AsmAssignTarget =
+        fun fromRegisters(registers: RegisterOrPair, scope: Subroutine?, program: Program, asmgen: AsmGen): AsmAssignTarget =
                 when(registers) {
                     RegisterOrPair.A,
                     RegisterOrPair.X,
-                    RegisterOrPair.Y -> AsmAssignTarget(TargetStorageKind.REGISTER, program, asmgen, DataType.UBYTE, register = registers)
+                    RegisterOrPair.Y -> AsmAssignTarget(TargetStorageKind.REGISTER, program, asmgen, DataType.UBYTE, scope, register = registers)
                     RegisterOrPair.AX,
                     RegisterOrPair.AY,
-                    RegisterOrPair.XY -> AsmAssignTarget(TargetStorageKind.REGISTER, program, asmgen, DataType.UWORD, register = registers)
+                    RegisterOrPair.XY -> AsmAssignTarget(TargetStorageKind.REGISTER, program, asmgen, DataType.UWORD, scope, register = registers)
+                    RegisterOrPair.FAC1,
+                    RegisterOrPair.FAC2 -> AsmAssignTarget(TargetStorageKind.REGISTER, program, asmgen, DataType.FLOAT, scope, register = registers)
                 }
     }
 }
 
 internal class AsmAssignSource(val kind: SourceStorageKind,
                                private val program: Program,
+                               private val asmgen: AsmGen,
                                val datatype: DataType,
-                               val variable: IdentifierReference? = null,
+                               private val variableAsmName: String? = null,
                                val array: ArrayIndexedExpression? = null,
                                val memory: DirectMemoryRead? = null,
                                val register: CpuRegister? = null,
@@ -93,53 +94,78 @@ internal class AsmAssignSource(val kind: SourceStorageKind,
 )
 {
     val constMemoryAddress by lazy { memory?.addressExpression?.constValue(program)?.number?.toInt() ?: 0}
-    val constArrayIndexValue by lazy { array?.arrayspec?.constIndex() }
-    val vardecl by lazy { variable?.targetVarDecl(program.namespace)!! }
+    val constArrayIndexValue by lazy { array?.indexer?.constIndex() }
+
+    val asmVarname: String
+        get() = if(array==null)
+            variableAsmName!!
+        else
+            asmgen.asmVariableName(array.arrayvar)
 
     companion object {
-        fun fromAstSource(value: Expression, program: Program): AsmAssignSource {
+        fun fromAstSource(indexer: ArrayIndex, program: Program, asmgen: AsmGen): AsmAssignSource {
+            return when {
+                indexer.indexNum!=null -> fromAstSource(indexer.indexNum!!, program, asmgen)
+                indexer.indexVar!=null -> fromAstSource(indexer.indexVar!!, program, asmgen)
+                else -> throw AssemblyError("weird indexer")
+            }
+        }
+
+        fun fromAstSource(value: Expression, program: Program, asmgen: AsmGen): AsmAssignSource {
             val cv = value.constValue(program)
             if(cv!=null)
-                return AsmAssignSource(SourceStorageKind.LITERALNUMBER, program, cv.type, number = cv)
+                return AsmAssignSource(SourceStorageKind.LITERALNUMBER, program, asmgen, cv.type, number = cv)
 
             return when(value) {
-                is NumericLiteralValue -> AsmAssignSource(SourceStorageKind.LITERALNUMBER, program, value.type, number = cv)
+                is NumericLiteralValue -> AsmAssignSource(SourceStorageKind.LITERALNUMBER, program, asmgen, value.type, number = cv)
                 is StringLiteralValue -> throw AssemblyError("string literal value should not occur anymore for asm generation")
                 is ArrayLiteralValue -> throw AssemblyError("array literal value should not occur anymore for asm generation")
                 is IdentifierReference -> {
                     val dt = value.inferType(program).typeOrElse(DataType.STRUCT)
-                    AsmAssignSource(SourceStorageKind.VARIABLE, program, dt, variable = value)
+                    AsmAssignSource(SourceStorageKind.VARIABLE, program, asmgen, dt, variableAsmName = asmgen.asmVariableName(value))
                 }
                 is DirectMemoryRead -> {
-                    AsmAssignSource(SourceStorageKind.MEMORY, program, DataType.UBYTE, memory = value)
+                    AsmAssignSource(SourceStorageKind.MEMORY, program, asmgen, DataType.UBYTE, memory = value)
                 }
                 is ArrayIndexedExpression -> {
                     val dt = value.inferType(program).typeOrElse(DataType.STRUCT)
-                    AsmAssignSource(SourceStorageKind.ARRAY, program, dt, array = value)
+                    AsmAssignSource(SourceStorageKind.ARRAY, program, asmgen, dt, array = value)
+                }
+                is FunctionCall -> {
+                    when (val sub = value.target.targetStatement(program.namespace)) {
+                        is Subroutine -> {
+                            val returnType = sub.returntypes.zip(sub.asmReturnvaluesRegisters).firstOrNull { rr -> rr.second.registerOrPair != null }?.first
+                                    ?: throw AssemblyError("can't translate zero return values in assignment")
+
+                            AsmAssignSource(SourceStorageKind.EXPRESSION, program, asmgen, returnType, expression = value)
+                        }
+                        is BuiltinFunctionStatementPlaceholder -> {
+                            val returnType = value.inferType(program)
+                            if(!returnType.isKnown)
+                                throw AssemblyError("unknown dt")
+                            AsmAssignSource(SourceStorageKind.EXPRESSION, program, asmgen, returnType.typeOrElse(DataType.STRUCT), expression = value)
+                        }
+                        else -> {
+                            throw AssemblyError("weird call")
+                        }
+                    }
                 }
                 else -> {
-                    val dt = value.inferType(program).typeOrElse(DataType.STRUCT)
-                    AsmAssignSource(SourceStorageKind.EXPRESSION, program, dt, expression = value)
+                    val dt = value.inferType(program)
+                    if(!dt.isKnown)
+                        throw AssemblyError("unknown dt")
+                    AsmAssignSource(SourceStorageKind.EXPRESSION, program, asmgen, dt.typeOrElse(DataType.STRUCT), expression = value)
                 }
             }
         }
     }
 
-    fun getAstValue(): Expression = when(kind) {
-        SourceStorageKind.LITERALNUMBER -> number!!
-        SourceStorageKind.VARIABLE -> variable!!
-        SourceStorageKind.ARRAY -> array!!
-        SourceStorageKind.MEMORY -> memory!!
-        SourceStorageKind.EXPRESSION -> expression!!
-        SourceStorageKind.REGISTER -> throw AssemblyError("cannot get a register source as Ast node")
-        SourceStorageKind.STACK -> throw AssemblyError("cannot get a stack source as Ast node")
-    }
-
-    fun withAdjustedDt(newType: DataType) =
-            AsmAssignSource(kind, program, newType, variable, array, memory, register, number, expression)
-
-    fun adjustDataTypeToTarget(target: AsmAssignTarget): AsmAssignSource {
+    fun adjustSignedUnsigned(target: AsmAssignTarget): AsmAssignSource {
         // allow some signed/unsigned relaxations
+
+        fun withAdjustedDt(newType: DataType) =
+                AsmAssignSource(kind, program, asmgen, newType, variableAsmName, array, memory, register, number, expression)
+
         if(target.datatype!=datatype) {
             if(target.datatype in ByteDatatypes && datatype in ByteDatatypes) {
                 return withAdjustedDt(target.datatype)
@@ -159,6 +185,10 @@ internal class AsmAssignment(val source: AsmAssignSource,
                              val position: Position) {
 
     init {
-        require(source.datatype==target.datatype) {"source and target datatype must be identical"}
+        if(target.register !in setOf(RegisterOrPair.XY, RegisterOrPair.AX, RegisterOrPair.AY))
+            require(source.datatype != DataType.STRUCT) { "must not be placeholder datatype" }
+            require(source.datatype.memorySize() <= target.datatype.memorySize()) {
+                "source storage size must be less or equal to target datatype storage size"
+            }
     }
 }

compiler/src/prog8/compiler/target/cx16/CX16MachineDefinition.kt

@@ -2,6 +2,7 @@ package prog8.compiler.target.cx16
 
 import prog8.ast.Program
 import prog8.compiler.*
+import prog8.compiler.target.CpuType
 import prog8.compiler.target.IMachineDefinition
 import prog8.compiler.target.c64.C64MachineDefinition
 import prog8.parser.ModuleImporter
@@ -9,7 +10,7 @@ import java.io.IOException
 
 internal object CX16MachineDefinition: IMachineDefinition {
 
-    override val cpu = "65c02"
+    override val cpu = CpuType.CPU65c02
 
     // 5-byte cbm MFLPT format limitations:
     override val FLOAT_MAX_POSITIVE = 1.7014118345e+38         // bytes: 255,127,255,255,255
@@ -25,14 +26,13 @@ internal object CX16MachineDefinition: IMachineDefinition {
     override val ESTACK_HI = 0x0500        //  $0500-$05ff inclusive
 
     override lateinit var zeropage: Zeropage
-    override val initSystemProcname = "cx16.init_system"
 
     override fun getFloat(num: Number) = C64MachineDefinition.Mflpt5.fromNumber(num)
 
-    override fun getFloatRomConst(number: Double): String? = null       // TODO Does Cx16 have ROM float locations?
+    override fun getFloatRomConst(number: Double): String? = null       // Cx16 has no pulblic ROM float locations
     override fun importLibs(compilerOptions: CompilationOptions, importer: ModuleImporter, program: Program) {
         if (compilerOptions.launcher == LauncherType.BASIC || compilerOptions.output == OutputType.PRG)
-            importer.importLibraryModule(program, "cx16lib")
+            importer.importLibraryModule(program, "syslib")
     }
 
     override fun launchEmulator(programName: String) {
@@ -52,39 +52,34 @@ internal object CX16MachineDefinition: IMachineDefinition {
         }
     }
 
+    override fun isRegularRAMaddress(address: Int): Boolean = address < 0x9f00 || address in 0xa000..0xbfff
+
     override fun initializeZeropage(compilerOptions: CompilationOptions) {
         zeropage = CX16Zeropage(compilerOptions)
     }
 
     // 6502 opcodes (including aliases and illegal opcodes), these cannot be used as variable or label names
-    // TODO add 65C02 opcodes
-    override val opcodeNames = setOf("adc", "ahx", "alr", "anc", "and", "ane", "arr", "asl", "asr", "axs", "bcc", "bcs",
+    override val opcodeNames = setOf("adc", "and", "asl", "bcc", "bcs",
             "beq", "bge", "bit", "blt", "bmi", "bne", "bpl", "brk", "bvc", "bvs", "clc",
-            "cld", "cli", "clv", "cmp", "cpx", "cpy", "dcm", "dcp", "dec", "dex", "dey",
+            "cld", "cli", "clv", "cmp", "cpx", "cpy", "dec", "dex", "dey",
             "eor", "gcc", "gcs", "geq", "gge", "glt", "gmi", "gne", "gpl", "gvc", "gvs",
-            "inc", "ins", "inx", "iny", "isb", "isc", "jam", "jmp", "jsr", "lae", "las",
-            "lax", "lda", "lds", "ldx", "ldy", "lsr", "lxa", "nop", "ora", "pha", "php",
-            "pla", "plp", "rla", "rol", "ror", "rra", "rti", "rts", "sax", "sbc", "sbx",
-            "sec", "sed", "sei", "sha", "shl", "shr", "shs", "shx", "shy", "slo", "sre",
-            "sta", "stx", "sty", "tas", "tax", "tay", "tsx", "txa", "txs", "tya", "xaa")
+            "inc", "inx", "iny", "jmp", "jsr",
+            "lda", "ldx", "ldy", "lsr", "nop", "ora", "pha", "php",
+            "pla", "plp", "rol", "ror", "rti", "rts", "sbc",
+            "sec", "sed", "sei",
+            "sta", "stx", "sty", "tax", "tay", "tsx", "txa", "txs", "tya",
+            "bra", "phx", "phy", "plx", "ply", "stz", "trb", "tsb", "bbr", "bbs",
+            "rmb", "smb", "stp", "wai")
 
 
     internal class CX16Zeropage(options: CompilationOptions) : Zeropage(options) {
 
         override val SCRATCH_B1 = 0x79      // temp storage for a single byte
-        override val SCRATCH_REG = 0x7a     // temp storage for a register
-        override val SCRATCH_REG_X = 0x7b   // temp storage for register X (the evaluation stack pointer)
+        override val SCRATCH_REG = 0x7a     // temp storage for a register, must be B1+1
         override val SCRATCH_W1 = 0x7c      // temp storage 1 for a word  $7c+$7d
         override val SCRATCH_W2 = 0x7e      // temp storage 2 for a word  $7e+$7f
 
 
-        override val exitProgramStrategy: ExitProgramStrategy = when (options.zeropage) {
-            ZeropageType.BASICSAFE, ZeropageType.DONTUSE -> ExitProgramStrategy.CLEAN_EXIT
-            ZeropageType.KERNALSAFE, ZeropageType.FULL -> ExitProgramStrategy.SYSTEM_RESET
-            else -> ExitProgramStrategy.SYSTEM_RESET
-        }
-
-
         init {
             if (options.floats && options.zeropage !in setOf(ZeropageType.BASICSAFE, ZeropageType.DONTUSE ))
                 throw CompilerException("when floats are enabled, zero page type should be 'basicsafe' or 'dontuse'")
@@ -94,16 +89,16 @@ internal object CX16MachineDefinition: IMachineDefinition {
             when (options.zeropage) {
                 ZeropageType.FULL -> {
                     free.addAll(0x22..0xff)
-                    free.removeAll(listOf(SCRATCH_B1, SCRATCH_REG, SCRATCH_REG_X, SCRATCH_W1, SCRATCH_W1 + 1, SCRATCH_W2, SCRATCH_W2 + 1))
+                    free.removeAll(listOf(SCRATCH_B1, SCRATCH_REG, SCRATCH_W1, SCRATCH_W1 + 1, SCRATCH_W2, SCRATCH_W2 + 1))
                 }
                 ZeropageType.KERNALSAFE -> {
                     free.addAll(0x22..0x7f)
                     free.addAll(0xa9..0xff)
-                    free.removeAll(listOf(SCRATCH_B1, SCRATCH_REG, SCRATCH_REG_X, SCRATCH_W1, SCRATCH_W1 + 1, SCRATCH_W2, SCRATCH_W2 + 1))
+                    free.removeAll(listOf(SCRATCH_B1, SCRATCH_REG, SCRATCH_W1, SCRATCH_W1 + 1, SCRATCH_W2, SCRATCH_W2 + 1))
                 }
                 ZeropageType.BASICSAFE -> {
                     free.addAll(0x22..0x7f)
-                    free.removeAll(listOf(SCRATCH_B1, SCRATCH_REG, SCRATCH_REG_X, SCRATCH_W1, SCRATCH_W1 + 1, SCRATCH_W2, SCRATCH_W2 + 1))
+                    free.removeAll(listOf(SCRATCH_B1, SCRATCH_REG, SCRATCH_W1, SCRATCH_W1 + 1, SCRATCH_W2, SCRATCH_W2 + 1))
                 }
                 ZeropageType.DONTUSE -> {
                     free.clear() // don't use zeropage at all
@@ -113,7 +108,6 @@ internal object CX16MachineDefinition: IMachineDefinition {
 
             require(SCRATCH_B1 !in free)
             require(SCRATCH_REG !in free)
-            require(SCRATCH_REG_X !in free)
             require(SCRATCH_W1 !in free)
             require(SCRATCH_W2 !in free)
 

compiler/src/prog8/functions/BuiltinFunctions.kt

@@ -15,95 +15,165 @@ class FParam(val name: String, val possibleDatatypes: Set<DataType>)
 typealias ConstExpressionCaller = (args: List<Expression>, position: Position, program: Program) -> NumericLiteralValue
 
 
-class FSignature(val pure: Boolean,      // does it have side effects?
+class ReturnConvention(val dt: DataType, val reg: RegisterOrPair?, val floatFac1: Boolean)
+class ParamConvention(val dt: DataType, val reg: RegisterOrPair?, val variable: Boolean)
+class CallConvention(val params: List<ParamConvention>, val returns: ReturnConvention) {
+    override fun toString(): String {
+        val paramConvs =  params.mapIndexed { index, it ->
+            when {
+                it.reg!=null -> "$index:${it.reg}"
+                it.variable -> "$index:variable"
+                else -> "$index:???"
+            }
+        }
+        val returnConv =
+                when {
+                    returns.reg!=null -> returns.reg.toString()
+                    returns.floatFac1 -> "floatFAC1"
+                    else -> "<no returnvalue>"
+                }
+        return "CallConvention[" + paramConvs.joinToString() + " ; returns: $returnConv]"
+    }
+}
+
+
+class FSignature(val name: String,
+                 val pure: Boolean,      // does it have side effects?
                  val parameters: List<FParam>,
-                 val returntype: DataType?,
-                 val constExpressionFunc: ConstExpressionCaller? = null)
+                 val known_returntype: DataType?,     // specify return type if fixed, otherwise null if it depends on the arguments
+                 val constExpressionFunc: ConstExpressionCaller? = null) {
 
+    fun callConvention(actualParamTypes: List<DataType>): CallConvention {
+        val returns = when(known_returntype) {
+            DataType.UBYTE, DataType.BYTE -> ReturnConvention(known_returntype, RegisterOrPair.A, false)
+            DataType.UWORD, DataType.WORD -> ReturnConvention(known_returntype, RegisterOrPair.AY, false)
+            DataType.FLOAT -> ReturnConvention(known_returntype, null, true)
+            in PassByReferenceDatatypes -> ReturnConvention(known_returntype!!, RegisterOrPair.AY, false)
+            else -> {
+                val paramType = actualParamTypes.first()
+                if(pure)
+                    // return type depends on arg type
+                    when(paramType) {
+                        DataType.UBYTE, DataType.BYTE -> ReturnConvention(paramType, RegisterOrPair.A, false)
+                        DataType.UWORD, DataType.WORD -> ReturnConvention(paramType, RegisterOrPair.AY, false)
+                        DataType.FLOAT -> ReturnConvention(paramType, null, true)
+                        in PassByReferenceDatatypes -> ReturnConvention(paramType, RegisterOrPair.AY, false)
+                        else -> ReturnConvention(paramType, null, false)
+                    }
+                else {
+                    ReturnConvention(paramType, null, false)
+                }
+            }
+        }
+        return when {
+            actualParamTypes.isEmpty() -> CallConvention(emptyList(), returns)
+            actualParamTypes.size==1 -> {
+                // one parameter? via register/registerpair
+                val paramConv = when(val paramType = actualParamTypes[0]) {
+                    DataType.UBYTE, DataType.BYTE -> ParamConvention(paramType, RegisterOrPair.A, false)
+                    DataType.UWORD, DataType.WORD -> ParamConvention(paramType, RegisterOrPair.AY, false)
+                    DataType.FLOAT -> ParamConvention(paramType, RegisterOrPair.AY, false)
+                    in PassByReferenceDatatypes -> ParamConvention(paramType, RegisterOrPair.AY, false)
+                    else -> ParamConvention(paramType, null, false)
+                }
+                CallConvention(listOf(paramConv), returns)
+            }
+            else -> {
+                // multiple parameters? via variables
+                val paramConvs = actualParamTypes.map { ParamConvention(it, null, true) }
+                CallConvention(paramConvs, returns)
+            }
+        }
+    }
+}
 
-val BuiltinFunctions = mapOf(
+private val functionSignatures: List<FSignature> = listOf(
         // this set of function have no return value and operate in-place:
-    "rol"         to FSignature(false, listOf(FParam("item", setOf(DataType.UBYTE, DataType.UWORD))), null),
-    "ror"         to FSignature(false, listOf(FParam("item", setOf(DataType.UBYTE, DataType.UWORD))), null),
-    "rol2"        to FSignature(false, listOf(FParam("item", setOf(DataType.UBYTE, DataType.UWORD))), null),
-    "ror2"        to FSignature(false, listOf(FParam("item", setOf(DataType.UBYTE, DataType.UWORD))), null),
-    "sort"        to FSignature(false, listOf(FParam("array", ArrayDatatypes)), null),
-    "reverse"     to FSignature(false, listOf(FParam("array", ArrayDatatypes)), null),
+    FSignature("rol"         , false, listOf(FParam("item", setOf(DataType.UBYTE, DataType.UWORD))), null),
+    FSignature("ror"         , false, listOf(FParam("item", setOf(DataType.UBYTE, DataType.UWORD))), null),
+    FSignature("rol2"        , false, listOf(FParam("item", setOf(DataType.UBYTE, DataType.UWORD))), null),
+    FSignature("ror2"        , false, listOf(FParam("item", setOf(DataType.UBYTE, DataType.UWORD))), null),
+    FSignature("sort"        , false, listOf(FParam("array", ArrayDatatypes)), null),
+    FSignature("reverse"     , false, listOf(FParam("array", ArrayDatatypes)), null),
         // these few have a return value depending on the argument(s):
-    "max"         to FSignature(true, listOf(FParam("values", ArrayDatatypes)), null) { a, p, prg -> collectionArg(a, p, prg, ::builtinMax) },    // type depends on args
-    "min"         to FSignature(true, listOf(FParam("values", ArrayDatatypes)), null) { a, p, prg -> collectionArg(a, p, prg, ::builtinMin) },    // type depends on args
-    "sum"         to FSignature(true, listOf(FParam("values", ArrayDatatypes)), null) { a, p, prg -> collectionArg(a, p, prg, ::builtinSum) },      // type depends on args
-    "abs"         to FSignature(true, listOf(FParam("value", NumericDatatypes)), null, ::builtinAbs),      // type depends on argument
-    "len"         to FSignature(true, listOf(FParam("values", IterableDatatypes)), null, ::builtinLen),    // type is UBYTE or UWORD depending on actual length
-    "sizeof"      to FSignature(true, listOf(FParam("object", DataType.values().toSet())), DataType.UBYTE, ::builtinSizeof),
+    FSignature("max"         , true, listOf(FParam("values", ArrayDatatypes)), null) { a, p, prg -> collectionArg(a, p, prg, ::builtinMax) },    // type depends on args
+    FSignature("min"         , true, listOf(FParam("values", ArrayDatatypes)), null) { a, p, prg -> collectionArg(a, p, prg, ::builtinMin) },    // type depends on args
+    FSignature("sum"         , true, listOf(FParam("values", ArrayDatatypes)), null) { a, p, prg -> collectionArg(a, p, prg, ::builtinSum) },      // type depends on args
+    FSignature("abs"         , true, listOf(FParam("value", NumericDatatypes)), null, ::builtinAbs),      // type depends on argument
+    FSignature("len"         , true, listOf(FParam("values", IterableDatatypes)), null, ::builtinLen),    // type is UBYTE or UWORD depending on actual length
+    FSignature("sizeof"      , true, listOf(FParam("object", DataType.values().toSet())), DataType.UBYTE, ::builtinSizeof),
         // normal functions follow:
-    "sgn"         to FSignature(true, listOf(FParam("value", NumericDatatypes)), DataType.BYTE, ::builtinSgn ),
-    "sin"         to FSignature(true, listOf(FParam("rads", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::sin) },
-    "sin8"        to FSignature(true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.BYTE, ::builtinSin8 ),
-    "sin8u"       to FSignature(true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.UBYTE, ::builtinSin8u ),
-    "sin16"       to FSignature(true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.WORD, ::builtinSin16 ),
-    "sin16u"      to FSignature(true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.UWORD, ::builtinSin16u ),
-    "cos"         to FSignature(true, listOf(FParam("rads", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::cos) },
-    "cos8"        to FSignature(true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.BYTE, ::builtinCos8 ),
-    "cos8u"       to FSignature(true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.UBYTE, ::builtinCos8u ),
-    "cos16"       to FSignature(true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.WORD, ::builtinCos16 ),
-    "cos16u"      to FSignature(true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.UWORD, ::builtinCos16u ),
-    "tan"         to FSignature(true, listOf(FParam("rads", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::tan) },
-    "atan"        to FSignature(true, listOf(FParam("rads", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::atan) },
-    "ln"          to FSignature(true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::log) },
-    "log2"        to FSignature(true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, ::log2) },
-    "sqrt16"      to FSignature(true, listOf(FParam("value", setOf(DataType.UWORD))), DataType.UBYTE) { a, p, prg -> oneIntArgOutputInt(a, p, prg) { sqrt(it.toDouble()).toInt() } },
-    "sqrt"        to FSignature(true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::sqrt) },
-    "rad"         to FSignature(true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::toRadians) },
-    "deg"         to FSignature(true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::toDegrees) },
-    "round"       to FSignature(true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArgOutputWord(a, p, prg, Math::round) },
-    "floor"       to FSignature(true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArgOutputWord(a, p, prg, Math::floor) },
-    "ceil"        to FSignature(true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArgOutputWord(a, p, prg, Math::ceil) },
-    "any"         to FSignature(true, listOf(FParam("values", ArrayDatatypes)), DataType.UBYTE) { a, p, prg -> collectionArg(a, p, prg, ::builtinAny) },
-    "all"         to FSignature(true, listOf(FParam("values", ArrayDatatypes)), DataType.UBYTE) { a, p, prg -> collectionArg(a, p, prg, ::builtinAll) },
-    "lsb"         to FSignature(true, listOf(FParam("value", setOf(DataType.UWORD, DataType.WORD))), DataType.UBYTE) { a, p, prg -> oneIntArgOutputInt(a, p, prg) { x: Int -> x and 255 }},
-    "msb"         to FSignature(true, listOf(FParam("value", setOf(DataType.UWORD, DataType.WORD))), DataType.UBYTE) { a, p, prg -> oneIntArgOutputInt(a, p, prg) { x: Int -> x ushr 8 and 255}},
-    "mkword"      to FSignature(true, listOf(FParam("msb", setOf(DataType.UBYTE)), FParam("lsb", setOf(DataType.UBYTE))), DataType.UWORD, ::builtinMkword),
-    "rnd"         to FSignature(true, emptyList(), DataType.UBYTE),
-    "rndw"        to FSignature(true, emptyList(), DataType.UWORD),
-    "rndf"        to FSignature(true, emptyList(), DataType.FLOAT),
-    "exit"        to FSignature(false, listOf(FParam("returnvalue", setOf(DataType.UBYTE))), null),
-    "rsave"       to FSignature(false, emptyList(), null),
-    "rrestore"    to FSignature(false, emptyList(), null),
-    "set_carry"   to FSignature(false, emptyList(), null),
-    "clear_carry" to FSignature(false, emptyList(), null),
-    "set_irqd"    to FSignature(false, emptyList(), null),
-    "clear_irqd"  to FSignature(false, emptyList(), null),
-    "read_flags"  to FSignature(false, emptyList(), DataType.UBYTE),
-    "swap"        to FSignature(false, listOf(FParam("first", NumericDatatypes), FParam("second", NumericDatatypes)), null),
-    "memcopy"     to FSignature(false, listOf(
+    FSignature("sgn"         , true, listOf(FParam("value", NumericDatatypes)), DataType.BYTE, ::builtinSgn ),
+    FSignature("sin"         , true, listOf(FParam("rads", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::sin) },
+    FSignature("sin8"        , true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.BYTE, ::builtinSin8 ),
+    FSignature("sin8u"       , true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.UBYTE, ::builtinSin8u ),
+    FSignature("sin16"       , true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.WORD, ::builtinSin16 ),
+    FSignature("sin16u"      , true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.UWORD, ::builtinSin16u ),
+    FSignature("cos"         , true, listOf(FParam("rads", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::cos) },
+    FSignature("cos8"        , true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.BYTE, ::builtinCos8 ),
+    FSignature("cos8u"       , true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.UBYTE, ::builtinCos8u ),
+    FSignature("cos16"       , true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.WORD, ::builtinCos16 ),
+    FSignature("cos16u"      , true, listOf(FParam("angle8", setOf(DataType.UBYTE))), DataType.UWORD, ::builtinCos16u ),
+    FSignature("tan"         , true, listOf(FParam("rads", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::tan) },
+    FSignature("atan"        , true, listOf(FParam("rads", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::atan) },
+    FSignature("ln"          , true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::log) },
+    FSignature("log2"        , true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, ::log2) },
+    FSignature("sqrt16"      , true, listOf(FParam("value", setOf(DataType.UWORD))), DataType.UBYTE) { a, p, prg -> oneIntArgOutputInt(a, p, prg) { sqrt(it.toDouble()).toInt() } },
+    FSignature("sqrt"        , true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::sqrt) },
+    FSignature("rad"         , true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::toRadians) },
+    FSignature("deg"         , true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArg(a, p, prg, Math::toDegrees) },
+    FSignature("round"       , true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArgOutputWord(a, p, prg, Math::round) },
+    FSignature("floor"       , true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArgOutputWord(a, p, prg, Math::floor) },
+    FSignature("ceil"        , true, listOf(FParam("value", setOf(DataType.FLOAT))), DataType.FLOAT) { a, p, prg -> oneDoubleArgOutputWord(a, p, prg, Math::ceil) },
+    FSignature("any"         , true, listOf(FParam("values", ArrayDatatypes)), DataType.UBYTE) { a, p, prg -> collectionArg(a, p, prg, ::builtinAny) },
+    FSignature("all"         , true, listOf(FParam("values", ArrayDatatypes)), DataType.UBYTE) { a, p, prg -> collectionArg(a, p, prg, ::builtinAll) },
+    FSignature("lsb"         , true, listOf(FParam("value", setOf(DataType.UWORD, DataType.WORD))), DataType.UBYTE) { a, p, prg -> oneIntArgOutputInt(a, p, prg) { x: Int -> x and 255 }},
+    FSignature("msb"         , true, listOf(FParam("value", setOf(DataType.UWORD, DataType.WORD))), DataType.UBYTE) { a, p, prg -> oneIntArgOutputInt(a, p, prg) { x: Int -> x ushr 8 and 255}},
+    FSignature("mkword"      , true, listOf(FParam("msb", setOf(DataType.UBYTE)), FParam("lsb", setOf(DataType.UBYTE))), DataType.UWORD, ::builtinMkword),
+    FSignature("rnd"         , false, emptyList(), DataType.UBYTE),
+    FSignature("rndw"        , false, emptyList(), DataType.UWORD),
+    FSignature("rndf"        , false, emptyList(), DataType.FLOAT),
+    FSignature("exit"        , false, listOf(FParam("returnvalue", setOf(DataType.UBYTE))), null),
+    FSignature("rsave"       , false, emptyList(), null),
+    FSignature("rrestore"    , false, emptyList(), null),
+    FSignature("set_carry"   , false, emptyList(), null),
+    FSignature("clear_carry" , false, emptyList(), null),
+    FSignature("set_irqd"    , false, emptyList(), null),
+    FSignature("clear_irqd"  , false, emptyList(), null),
+    FSignature("read_flags"  , false, emptyList(), DataType.UBYTE),
+    FSignature("swap"        , false, listOf(FParam("first", NumericDatatypes), FParam("second", NumericDatatypes)), null),
+    FSignature("memcopy"     , false, listOf(
                             FParam("from", IterableDatatypes + DataType.UWORD),
                             FParam("to", IterableDatatypes + DataType.UWORD),
-                            FParam("numbytes", setOf(DataType.UBYTE))), null),
-    "memset"      to FSignature(false, listOf(
+                            FParam("numbytes", setOf(DataType.UBYTE, DataType.UWORD))), null),
+    FSignature("memset"      , false, listOf(
                             FParam("address", IterableDatatypes + DataType.UWORD),
                             FParam("numbytes", setOf(DataType.UWORD)),
                             FParam("bytevalue", ByteDatatypes)), null),
-    "memsetw"     to FSignature(false, listOf(
+    FSignature("memsetw"     , false, listOf(
                             FParam("address", IterableDatatypes + DataType.UWORD),
                             FParam("numwords", setOf(DataType.UWORD)),
                             FParam("wordvalue", setOf(DataType.UWORD, DataType.WORD))), null),
-    "strlen"      to FSignature(true, listOf(FParam("string", setOf(DataType.STR))), DataType.UBYTE, ::builtinStrlen),
-    "substr"      to FSignature(false, listOf(
+    FSignature("strlen"      , true, listOf(FParam("string", setOf(DataType.STR))), DataType.UBYTE, ::builtinStrlen),
+    FSignature("substr"      , false, listOf(
             FParam("source", IterableDatatypes + DataType.UWORD),
             FParam("target", IterableDatatypes + DataType.UWORD),
             FParam("start", setOf(DataType.UBYTE)),
             FParam("length", setOf(DataType.UBYTE))), null),
-    "leftstr"      to FSignature(false, listOf(
+    FSignature("leftstr"     , false, listOf(
             FParam("source", IterableDatatypes + DataType.UWORD),
             FParam("target", IterableDatatypes + DataType.UWORD),
             FParam("length", setOf(DataType.UBYTE))), null),
-    "rightstr"      to FSignature(false, listOf(
+    FSignature("rightstr"    , false, listOf(
             FParam("source", IterableDatatypes + DataType.UWORD),
             FParam("target", IterableDatatypes + DataType.UWORD),
-            FParam("length", setOf(DataType.UBYTE))), null)
+            FParam("length", setOf(DataType.UBYTE))), null),
+    FSignature("strcmp"      , false, listOf(FParam("s1", IterableDatatypes + DataType.UWORD), FParam("s2", IterableDatatypes + DataType.UWORD)), DataType.BYTE, null)
 )
 
+val BuiltinFunctions = functionSignatures.associateBy { it.name }
+
+
 fun builtinMax(array: List<Number>): Number = array.maxByOrNull { it.toDouble() }!!
 
 fun builtinMin(array: List<Number>): Number = array.minByOrNull { it.toDouble() }!!
@@ -143,17 +213,17 @@ fun builtinFunctionReturnType(function: String, args: List<Expression>, program:
     }
 
     val func = BuiltinFunctions.getValue(function)
-    if(func.returntype!=null)
-        return InferredTypes.knownFor(func.returntype)
-    // function has return values, but the return type depends on the arguments
+    if(func.known_returntype!=null)
+        return InferredTypes.knownFor(func.known_returntype)
 
+    // function has return values, but the return type depends on the arguments
     return when (function) {
         "abs" -> {
             val dt = args.single().inferType(program)
-            if(dt.typeOrElse(DataType.STRUCT) in NumericDatatypes)
-                return dt
+            return if(dt.typeOrElse(DataType.STRUCT) in NumericDatatypes)
+                dt
             else
-                throw FatalAstException("weird datatype passed to abs $dt")
+                InferredTypes.InferredType.unknown()
         }
         "max", "min" -> {
             when(val dt = datatypeFromIterableArg(args.single())) {
@@ -285,9 +355,9 @@ private fun builtinStrlen(args: List<Expression>, position: Position, program: P
         return NumericLiteralValue.optimalInteger(argument.value.length, argument.position)
     val vardecl = (argument as IdentifierReference).targetVarDecl(program.namespace)
     if(vardecl!=null) {
-        if(vardecl.datatype!=DataType.STR)
+        if(vardecl.datatype!=DataType.STR && vardecl.datatype!=DataType.UWORD)
             throw SyntaxError("strlen must have string argument", position)
-        if(vardecl.autogeneratedDontRemove) {
+        if(vardecl.autogeneratedDontRemove && vardecl.value!=null) {
             return NumericLiteralValue.optimalInteger((vardecl.value as StringLiteralValue).value.length, argument.position)
         }
     }

compiler/src/prog8/optimizer/BinExprSplitter.kt

@@ -0,0 +1,87 @@
+package prog8.optimizer
+
+import prog8.ast.INameScope
+import prog8.ast.Node
+import prog8.ast.Program
+import prog8.ast.expressions.*
+import prog8.ast.processing.AstWalker
+import prog8.ast.processing.IAstModification
+import prog8.ast.statements.AssignTarget
+import prog8.ast.statements.Assignment
+
+
+internal class BinExprSplitter(private val program: Program) : AstWalker() {
+    private val noModifications = emptyList<IAstModification>()
+
+//    override fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> {
+// TODO somehow if we do this, the resulting code for some programs (cube3d.p8) gets hundreds of bytes larger...:
+//        if(decl.type==VarDeclType.VAR ) {
+//            val binExpr = decl.value as? BinaryExpression
+//            if (binExpr != null && binExpr.operator in augmentAssignmentOperators) {
+//                // split into a vardecl with just the left expression, and an aug. assignment with the right expression.
+//                val augExpr = BinaryExpression(IdentifierReference(listOf(decl.name), decl.position), binExpr.operator, binExpr.right, binExpr.position)
+//                val target = AssignTarget(IdentifierReference(listOf(decl.name), decl.position), null, null, decl.position)
+//                val assign = Assignment(target, augExpr, binExpr.position)
+//                println("SPLIT VARDECL $decl")
+//                return listOf(
+//                        IAstModification.SetExpression({ decl.value = it }, binExpr.left, decl),
+//                        IAstModification.InsertAfter(decl, assign, parent)
+//                )
+//            }
+//        }
+//        return noModifications
+//    }
+
+    override fun after(assignment: Assignment, parent: Node): Iterable<IAstModification> {
+
+        val binExpr = assignment.value as? BinaryExpression
+        if (binExpr != null) {
+/*
+
+Reduce the complexity of a (binary) expression that has to be evaluated on the eval stack,
+by attempting to splitting it up into individual simple steps.
+We only consider a binary expression *one* level deep (so the operands must not be a combined expression)
+
+
+X =      BinExpr                                    X   =   LeftExpr
+        <operator>                                     followed by
+          /   \             IF 'X' not used         X   =   BinExpr
+         /     \             IN expression ==>             <operator>
+        /       \                                           /   \
+    LeftExpr.  RightExpr.                                  /     \
+                                                          X     RightExpr.
+
+
+ */
+            if(binExpr.operator in augmentAssignmentOperators && isSimpleTarget(assignment.target, program.namespace)) {
+                if(assignment.target isSameAs binExpr.left || assignment.target isSameAs binExpr.right)
+                    return noModifications
+
+                if(isSimpleExpression(binExpr.right) && !assignment.isAugmentable) {
+                    val firstAssign = Assignment(assignment.target, binExpr.left, binExpr.left.position)
+                    val targetExpr = assignment.target.toExpression()
+                    val augExpr = BinaryExpression(targetExpr, binExpr.operator, binExpr.right, binExpr.right.position)
+                    return listOf(
+                            IAstModification.InsertBefore(assignment, firstAssign, assignment.definingScope()),
+                            IAstModification.ReplaceNode(assignment.value, augExpr, assignment))
+                }
+            }
+
+            // TODO further unraveling of binary expression trees into flat statements.
+            // however this should probably be done in a more generic way to also service
+            // the expressiontrees that are not used in an assignment statement...
+        }
+
+        return noModifications
+    }
+
+    private fun isSimpleExpression(expr: Expression) =
+            expr is IdentifierReference || expr is NumericLiteralValue || expr is AddressOf || expr is DirectMemoryRead || expr is StringLiteralValue || expr is ArrayLiteralValue || expr is RangeExpr
+
+    private fun isSimpleTarget(target: AssignTarget, namespace: INameScope) =
+            if (target.identifier!=null || target.memoryAddress!=null)
+                target.isInRegularRAM(namespace)
+            else
+                false
+
+}

compiler/src/prog8/optimizer/CallGraph.kt

@@ -1,11 +1,10 @@
 package prog8.optimizer
 
-import prog8.ast.INameScope
-import prog8.ast.Module
-import prog8.ast.Node
-import prog8.ast.Program
+import prog8.ast.*
 import prog8.ast.base.DataType
+import prog8.ast.base.ErrorReporter
 import prog8.ast.base.ParentSentinel
+import prog8.ast.base.Position
 import prog8.ast.expressions.FunctionCall
 import prog8.ast.expressions.IdentifierReference
 import prog8.ast.processing.IAstVisitor
@@ -25,7 +24,7 @@ class CallGraph(private val program: Program) : IAstVisitor {
 
     val imports = mutableMapOf<Module, List<Module>>().withDefault { mutableListOf() }
     val importedBy = mutableMapOf<Module, List<Module>>().withDefault { mutableListOf() }
-    val calls = mutableMapOf<INameScope, List<Subroutine>>().withDefault { mutableListOf() }
+    val calls = mutableMapOf<Subroutine, List<Subroutine>>().withDefault { mutableListOf() }
     val calledBy = mutableMapOf<Subroutine, List<Node>>().withDefault { mutableListOf() }
 
     // TODO  add dataflow graph: what statements use what variables - can be used to eliminate unused vars
@@ -79,8 +78,10 @@ class CallGraph(private val program: Program) : IAstVisitor {
             importedBy[importedModule] = importedBy.getValue(importedModule).plus(thisModule)
         } else if (directive.directive == "%asminclude") {
             val asm = loadAsmIncludeFile(directive.args[0].str!!, thisModule.source)
-            val scope = directive.definingScope()
-            scanAssemblyCode(asm, directive, scope)
+            val scope = directive.definingSubroutine()
+            if(scope!=null) {
+                scanAssemblyCode(asm, directive, scope)
+            }
         }
 
         super.visit(directive)
@@ -167,12 +168,12 @@ class CallGraph(private val program: Program) : IAstVisitor {
 
     override fun visit(inlineAssembly: InlineAssembly) {
         // parse inline asm for subroutine calls (jmp, jsr)
-        val scope = inlineAssembly.definingScope()
+        val scope = inlineAssembly.definingSubroutine()
         scanAssemblyCode(inlineAssembly.assembly, inlineAssembly, scope)
         super.visit(inlineAssembly)
     }
 
-    private fun scanAssemblyCode(asm: String, context: Statement, scope: INameScope) {
+    private fun scanAssemblyCode(asm: String, context: Statement, scope: Subroutine?) {
         asm.lines().forEach { line ->
             val matches = asmJumpRx.matchEntire(line)
             if (matches != null) {
@@ -180,13 +181,15 @@ class CallGraph(private val program: Program) : IAstVisitor {
                 if (jumptarget != null && (jumptarget[0].isLetter() || jumptarget[0] == '_')) {
                     val node = program.namespace.lookup(jumptarget.split('.'), context)
                     if (node is Subroutine) {
-                        calls[scope] = calls.getValue(scope).plus(node)
+                        if(scope!=null)
+                            calls[scope] = calls.getValue(scope).plus(node)
                         calledBy[node] = calledBy.getValue(node).plus(context)
                     } else if (jumptarget.contains('.')) {
                         // maybe only the first part already refers to a subroutine
                         val node2 = program.namespace.lookup(listOf(jumptarget.substringBefore('.')), context)
                         if (node2 is Subroutine) {
-                            calls[scope] = calls.getValue(scope).plus(node2)
+                            if(scope!=null)
+                                calls[scope] = calls.getValue(scope).plus(node2)
                             calledBy[node2] = calledBy.getValue(node2).plus(context)
                         }
                     }
@@ -199,7 +202,8 @@ class CallGraph(private val program: Program) : IAstVisitor {
                         if (target.contains('.')) {
                             val node = program.namespace.lookup(listOf(target.substringBefore('.')), context)
                             if (node is Subroutine) {
-                                calls[scope] = calls.getValue(scope).plus(node)
+                                if(scope!=null)
+                                    calls[scope] = calls.getValue(scope).plus(node)
                                 calledBy[node] = calledBy.getValue(node).plus(context)
                             }
                         }
@@ -208,4 +212,55 @@ class CallGraph(private val program: Program) : IAstVisitor {
             }
         }
     }
+
+    fun checkRecursiveCalls(errors: ErrorReporter) {
+        val cycles = recursionCycles()
+        if(cycles.any()) {
+            errors.warn("Program contains recursive subroutine calls. These only works in very specific limited scenarios!", Position.DUMMY)
+            val printed = mutableSetOf<Subroutine>()
+            for(chain in cycles) {
+                if(chain[0] !in printed) {
+                    val chainStr = chain.joinToString(" <-- ") { "${it.name} at ${it.position}" }
+                    errors.warn("Cycle in (a subroutine call in) $chainStr", Position.DUMMY)
+                    printed.add(chain[0])
+                }
+            }
+        }
+    }
+
+    private fun recursionCycles(): List<List<Subroutine>> {
+        val chains = mutableListOf<MutableList<Subroutine>>()
+        for(caller in calls.keys) {
+            val visited = calls.keys.associateWith { false }.toMutableMap()
+            val recStack = calls.keys.associateWith { false }.toMutableMap()
+            val chain = mutableListOf<Subroutine>()
+            if(hasCycle(caller, visited, recStack, chain))
+                chains.add(chain)
+        }
+        return chains
+    }
+
+    private fun hasCycle(sub: Subroutine, visited: MutableMap<Subroutine, Boolean>, recStack: MutableMap<Subroutine, Boolean>, chain: MutableList<Subroutine>): Boolean {
+        // mark current node as visited and add to recursion stack
+        if(recStack[sub]==true)
+            return true
+        if(visited[sub]==true)
+            return false
+
+        // mark visited and add to recursion stack
+        visited[sub] = true
+        recStack[sub] = true
+
+        // recurse for all neighbours
+        for(called in calls.getValue(sub)) {
+            if(hasCycle(called, visited, recStack, chain)) {
+                chain.add(called)
+                return true
+            }
+        }
+
+        // pop from recursion stack
+        recStack[sub] = false
+        return false
+    }
 }

compiler/src/prog8/optimizer/ConstantFoldingOptimizer.kt

@@ -7,178 +7,6 @@ import prog8.ast.expressions.*
 import prog8.ast.processing.AstWalker
 import prog8.ast.processing.IAstModification
 import prog8.ast.statements.*
-import prog8.compiler.target.CompilationTarget
-
-
-// First thing to do is replace all constant identifiers with their actual value,
-// and the array var initializer values and sizes.
-// This is needed because further constant optimizations depend on those.
-internal class ConstantIdentifierReplacer(private val program: Program, private val errors: ErrorReporter) : AstWalker() {
-    private val noModifications = emptyList<IAstModification>()
-
-    override fun after(identifier: IdentifierReference, parent: Node): Iterable<IAstModification> {
-        // replace identifiers that refer to const value, with the value itself
-        // if it's a simple type and if it's not a left hand side variable
-        if(identifier.parent is AssignTarget)
-            return noModifications
-        var forloop = identifier.parent as? ForLoop
-        if(forloop==null)
-            forloop = identifier.parent.parent as? ForLoop
-        if(forloop!=null && identifier===forloop.loopVar)
-            return noModifications
-
-        val cval = identifier.constValue(program) ?: return noModifications
-        return when (cval.type) {
-            in NumericDatatypes -> listOf(IAstModification.ReplaceNode(identifier, NumericLiteralValue(cval.type, cval.number, identifier.position), identifier.parent))
-            in PassByReferenceDatatypes -> throw FatalAstException("pass-by-reference type should not be considered a constant")
-            else -> noModifications
-        }
-    }
-
-    override fun before(decl: VarDecl, parent: Node): Iterable<IAstModification> {
-        // the initializer value can't refer to the variable itself (recursive definition)
-        // TODO: use call graph for this?
-        if(decl.value?.referencesIdentifiers(decl.name) == true || decl.arraysize?.index?.referencesIdentifiers(decl.name) == true) {
-            errors.err("recursive var declaration", decl.position)
-            return noModifications
-        }
-
-        if(decl.type==VarDeclType.CONST || decl.type==VarDeclType.VAR) {
-            if(decl.isArray){
-                if(decl.arraysize==null) {
-                    // for arrays that have no size specifier (or a non-constant one) attempt to deduce the size
-                    val arrayval = decl.value as? ArrayLiteralValue
-                    if(arrayval!=null) {
-                        return listOf(IAstModification.SetExpression(
-                                { decl.arraysize = ArrayIndex(it, decl.position) },
-                                NumericLiteralValue.optimalInteger(arrayval.value.size, decl.position),
-                                decl
-                        ))
-                    }
-                }
-                else if(decl.arraysize?.constIndex()==null) {
-                    val size = decl.arraysize!!.index.constValue(program)
-                    if(size!=null) {
-                        return listOf(IAstModification.SetExpression(
-                                { decl.arraysize = ArrayIndex(it, decl.position) },
-                                size, decl
-                        ))
-                    }
-                }
-            }
-
-            when(decl.datatype) {
-                DataType.FLOAT -> {
-                    // vardecl: for scalar float vars, promote constant integer initialization values to floats
-                    val litval = decl.value as? NumericLiteralValue
-                    if (litval!=null && litval.type in IntegerDatatypes) {
-                        val newValue = NumericLiteralValue(DataType.FLOAT, litval.number.toDouble(), litval.position)
-                        return listOf(IAstModification.ReplaceNode(decl.value!!, newValue, decl))
-                    }
-                }
-                DataType.ARRAY_UB, DataType.ARRAY_B, DataType.ARRAY_UW, DataType.ARRAY_W -> {
-                    val numericLv = decl.value as? NumericLiteralValue
-                    val rangeExpr = decl.value as? RangeExpr
-                    if(rangeExpr!=null) {
-                        // convert the initializer range expression to an actual array
-                        val declArraySize = decl.arraysize?.constIndex()
-                        if(declArraySize!=null && declArraySize!=rangeExpr.size())
-                            errors.err("range expression size doesn't match declared array size", decl.value?.position!!)
-                        val constRange = rangeExpr.toConstantIntegerRange()
-                        if(constRange!=null) {
-                            val eltType = rangeExpr.inferType(program).typeOrElse(DataType.UBYTE)
-                            val newValue = if(eltType in ByteDatatypes) {
-                                ArrayLiteralValue(InferredTypes.InferredType.known(decl.datatype),
-                                        constRange.map { NumericLiteralValue(eltType, it.toShort(), decl.value!!.position) }.toTypedArray(),
-                                        position = decl.value!!.position)
-                            } else {
-                                ArrayLiteralValue(InferredTypes.InferredType.known(decl.datatype),
-                                        constRange.map { NumericLiteralValue(eltType, it, decl.value!!.position) }.toTypedArray(),
-                                        position = decl.value!!.position)
-                            }
-                            return listOf(IAstModification.ReplaceNode(decl.value!!, newValue, decl))
-                        }
-                    }
-                    if(numericLv!=null && numericLv.type==DataType.FLOAT)
-                        errors.err("arraysize requires only integers here", numericLv.position)
-                    val size = decl.arraysize?.constIndex() ?: return noModifications
-                    if (rangeExpr==null && numericLv!=null) {
-                        // arraysize initializer is empty or a single int, and we know the size; create the arraysize.
-                        val fillvalue = numericLv.number.toInt()
-                        when(decl.datatype){
-                            DataType.ARRAY_UB -> {
-                                if(fillvalue !in 0..255)
-                                    errors.err("ubyte value overflow", numericLv.position)
-                            }
-                            DataType.ARRAY_B -> {
-                                if(fillvalue !in -128..127)
-                                    errors.err("byte value overflow", numericLv.position)
-                            }
-                            DataType.ARRAY_UW -> {
-                                if(fillvalue !in 0..65535)
-                                    errors.err("uword value overflow", numericLv.position)
-                            }
-                            DataType.ARRAY_W -> {
-                                if(fillvalue !in -32768..32767)
-                                    errors.err("word value overflow", numericLv.position)
-                            }
-                            else -> {}
-                        }
-                        // create the array itself, filled with the fillvalue.
-                        val array = Array(size) {fillvalue}.map { NumericLiteralValue(ArrayElementTypes.getValue(decl.datatype), it, numericLv.position) }.toTypedArray<Expression>()
-                        val refValue = ArrayLiteralValue(InferredTypes.InferredType.known(decl.datatype), array, position = numericLv.position)
-                        return listOf(IAstModification.ReplaceNode(decl.value!!, refValue, decl))
-                    }
-                }
-                DataType.ARRAY_F  -> {
-                    val size = decl.arraysize?.constIndex() ?: return noModifications
-                    val litval = decl.value as? NumericLiteralValue
-                    val rangeExpr = decl.value as? RangeExpr
-                    if(rangeExpr!=null) {
-                        // convert the initializer range expression to an actual array of floats
-                        val declArraySize = decl.arraysize?.constIndex()
-                        if(declArraySize!=null && declArraySize!=rangeExpr.size())
-                            errors.err("range expression size doesn't match declared array size", decl.value?.position!!)
-                        val constRange = rangeExpr.toConstantIntegerRange()
-                        if(constRange!=null) {
-                            val newValue = ArrayLiteralValue(InferredTypes.InferredType.known(DataType.ARRAY_F),
-                                        constRange.map { NumericLiteralValue(DataType.FLOAT, it.toDouble(), decl.value!!.position) }.toTypedArray(),
-                                        position = decl.value!!.position)
-                            return listOf(IAstModification.ReplaceNode(decl.value!!, newValue, decl))
-                        }
-                    }
-                    if(rangeExpr==null && litval!=null) {
-                        // arraysize initializer is a single int, and we know the size.
-                        val fillvalue = litval.number.toDouble()
-                        if (fillvalue < CompilationTarget.machine.FLOAT_MAX_NEGATIVE || fillvalue > CompilationTarget.machine.FLOAT_MAX_POSITIVE)
-                            errors.err("float value overflow", litval.position)
-                        else {
-                            // create the array itself, filled with the fillvalue.
-                            val array = Array(size) {fillvalue}.map { NumericLiteralValue(DataType.FLOAT, it, litval.position) }.toTypedArray<Expression>()
-                            val refValue = ArrayLiteralValue(InferredTypes.InferredType.known(DataType.ARRAY_F), array, position = litval.position)
-                            return listOf(IAstModification.ReplaceNode(decl.value!!, refValue, decl))
-                        }
-                    }
-                }
-                else -> {
-                    // nothing to do for this type
-                    // this includes strings and structs
-                }
-            }
-        }
-
-        val declValue = decl.value
-        if(declValue!=null && decl.type==VarDeclType.VAR
-                && declValue is NumericLiteralValue && !declValue.inferType(program).istype(decl.datatype)) {
-            // cast the numeric literal to the appropriate datatype of the variable
-            val cast = declValue.cast(decl.datatype)
-            if(cast.isValid)
-                return listOf(IAstModification.ReplaceNode(decl.value!!, cast.valueOrZero(), decl))
-        }
-
-        return noModifications
-    }
-}
 
 
 internal class ConstantFoldingOptimizer(private val program: Program) : AstWalker() {
@@ -216,9 +44,24 @@ internal class ConstantFoldingOptimizer(private val program: Program) : AstWalke
                     else -> throw ExpressionError("can only take negative of int or float", subexpr.position)
                 }
                 "~" -> when (subexpr.type) {
-                    in IntegerDatatypes -> {
+                    DataType.BYTE -> {
                         listOf(IAstModification.ReplaceNode(expr,
-                                NumericLiteralValue.optimalInteger(subexpr.number.toInt().inv(), subexpr.position),
+                                NumericLiteralValue(DataType.BYTE, subexpr.number.toInt().inv(), subexpr.position),
+                                parent))
+                    }
+                    DataType.UBYTE -> {
+                        listOf(IAstModification.ReplaceNode(expr,
+                                NumericLiteralValue(DataType.UBYTE, subexpr.number.toInt().inv() and 255, subexpr.position),
+                                parent))
+                    }
+                    DataType.WORD -> {
+                        listOf(IAstModification.ReplaceNode(expr,
+                                NumericLiteralValue(DataType.WORD, subexpr.number.toInt().inv(), subexpr.position),
+                                parent))
+                    }
+                    DataType.UWORD -> {
+                        listOf(IAstModification.ReplaceNode(expr,
+                                NumericLiteralValue(DataType.UWORD, subexpr.number.toInt().inv() and 65535, subexpr.position),
                                 parent))
                     }
                     else -> throw ExpressionError("can only take bitwise inversion of int", subexpr.position)
@@ -318,11 +161,11 @@ internal class ConstantFoldingOptimizer(private val program: Program) : AstWalke
     }
 
     override fun after(forLoop: ForLoop, parent: Node): Iterable<IAstModification> {
-        fun adjustRangeDt(rangeFrom: NumericLiteralValue, targetDt: DataType, rangeTo: NumericLiteralValue, stepLiteral: NumericLiteralValue?, range: RangeExpr): RangeExpr {
+        fun adjustRangeDt(rangeFrom: NumericLiteralValue, targetDt: DataType, rangeTo: NumericLiteralValue, stepLiteral: NumericLiteralValue?, range: RangeExpr): RangeExpr? {
             val fromCast = rangeFrom.cast(targetDt)
             val toCast = rangeTo.cast(targetDt)
             if(!fromCast.isValid || !toCast.isValid)
-                return range
+                return null
 
             val newStep =
                 if(stepLiteral!=null) {
@@ -351,28 +194,32 @@ internal class ConstantFoldingOptimizer(private val program: Program) : AstWalke
                 if(rangeFrom.type!= DataType.UBYTE) {
                     // attempt to translate the iterable into ubyte values
                     val newIter = adjustRangeDt(rangeFrom, loopvar.datatype, rangeTo, stepLiteral, iterableRange)
-                    return listOf(IAstModification.ReplaceNode(forLoop.iterable, newIter, forLoop))
+                    if(newIter!=null)
+                        return listOf(IAstModification.ReplaceNode(forLoop.iterable, newIter, forLoop))
                 }
             }
             DataType.BYTE -> {
                 if(rangeFrom.type!= DataType.BYTE) {
                     // attempt to translate the iterable into byte values
                     val newIter = adjustRangeDt(rangeFrom, loopvar.datatype, rangeTo, stepLiteral, iterableRange)
-                    return listOf(IAstModification.ReplaceNode(forLoop.iterable, newIter, forLoop))
+                    if(newIter!=null)
+                        return listOf(IAstModification.ReplaceNode(forLoop.iterable, newIter, forLoop))
                 }
             }
             DataType.UWORD -> {
                 if(rangeFrom.type!= DataType.UWORD) {
                     // attempt to translate the iterable into uword values
                     val newIter = adjustRangeDt(rangeFrom, loopvar.datatype, rangeTo, stepLiteral, iterableRange)
-                    return listOf(IAstModification.ReplaceNode(forLoop.iterable, newIter, forLoop))
+                    if(newIter!=null)
+                        return listOf(IAstModification.ReplaceNode(forLoop.iterable, newIter, forLoop))
                 }
             }
             DataType.WORD -> {
                 if(rangeFrom.type!= DataType.WORD) {
                     // attempt to translate the iterable into word values
                     val newIter = adjustRangeDt(rangeFrom, loopvar.datatype, rangeTo, stepLiteral, iterableRange)
-                    return listOf(IAstModification.ReplaceNode(forLoop.iterable, newIter, forLoop))
+                    if(newIter!=null)
+                        return listOf(IAstModification.ReplaceNode(forLoop.iterable, newIter, forLoop))
                 }
             }
             else -> throw FatalAstException("invalid loopvar datatype $loopvar")

compiler/src/prog8/optimizer/ConstantIdentifierReplacer.kt

@@ -0,0 +1,192 @@
+package prog8.optimizer
+
+import prog8.ast.Node
+import prog8.ast.Program
+import prog8.ast.base.*
+import prog8.ast.expressions.*
+import prog8.ast.processing.AstWalker
+import prog8.ast.processing.IAstModification
+import prog8.ast.statements.ArrayIndex
+import prog8.ast.statements.AssignTarget
+import prog8.ast.statements.ForLoop
+import prog8.ast.statements.VarDecl
+import prog8.compiler.target.CompilationTarget
+
+// Fix up the literal value's type to match that of the vardecl
+internal class VarConstantValueTypeAdjuster(private val program: Program) : AstWalker() {
+    private val noModifications = emptyList<IAstModification>()
+
+    override fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> {
+        val declConstValue = decl.value?.constValue(program)
+        if(declConstValue!=null && (decl.type==VarDeclType.VAR || decl.type==VarDeclType.CONST)
+                && !declConstValue.inferType(program).istype(decl.datatype)) {
+            // cast the numeric literal to the appropriate datatype of the variable
+            val cast = declConstValue.cast(decl.datatype)
+            if(cast.isValid)
+                return listOf(IAstModification.ReplaceNode(decl.value!!, cast.valueOrZero(), decl))
+        }
+        return noModifications
+    }
+}
+
+
+// Replace all constant identifiers with their actual value,
+// and the array var initializer values and sizes.
+// This is needed because further constant optimizations depend on those.
+internal class ConstantIdentifierReplacer(private val program: Program, private val errors: ErrorReporter) : AstWalker() {
+    private val noModifications = emptyList<IAstModification>()
+
+    override fun after(identifier: IdentifierReference, parent: Node): Iterable<IAstModification> {
+        // replace identifiers that refer to const value, with the value itself
+        // if it's a simple type and if it's not a left hand side variable
+        if(identifier.parent is AssignTarget)
+            return noModifications
+        var forloop = identifier.parent as? ForLoop
+        if(forloop==null)
+            forloop = identifier.parent.parent as? ForLoop
+        if(forloop!=null && identifier===forloop.loopVar)
+            return noModifications
+
+        val cval = identifier.constValue(program) ?: return noModifications
+        return when (cval.type) {
+            in NumericDatatypes -> listOf(IAstModification.ReplaceNode(identifier, NumericLiteralValue(cval.type, cval.number, identifier.position), identifier.parent))
+            in PassByReferenceDatatypes -> throw FatalAstException("pass-by-reference type should not be considered a constant")
+            else -> noModifications
+        }
+    }
+
+    override fun before(decl: VarDecl, parent: Node): Iterable<IAstModification> {
+        // the initializer value can't refer to the variable itself (recursive definition)
+        // TODO: use call graph for this?
+        if(decl.value?.referencesIdentifier(decl.name) == true || decl.arraysize?.indexVar?.referencesIdentifier(decl.name) == true) {
+            errors.err("recursive var declaration", decl.position)
+            return noModifications
+        }
+
+        if(decl.type== VarDeclType.CONST || decl.type== VarDeclType.VAR) {
+            if(decl.isArray){
+                val arraysize = decl.arraysize
+                if(arraysize==null) {
+                    // for arrays that have no size specifier attempt to deduce the size
+                    val arrayval = decl.value as? ArrayLiteralValue
+                    if(arrayval!=null) {
+                        return listOf(IAstModification.SetExpression(
+                                { decl.arraysize = ArrayIndex(it, decl.position) },
+                                NumericLiteralValue.optimalInteger(arrayval.value.size, decl.position),
+                                decl
+                        ))
+                    }
+                } else if(arraysize.constIndex()==null) {
+                    // see if we can calculate the size from other fields
+                    val cval = arraysize.indexVar?.constValue(program) ?: arraysize.origExpression?.constValue(program)
+                    if(cval!=null) {
+                        arraysize.indexVar = null
+                        arraysize.origExpression = null
+                        arraysize.indexNum = cval
+                    }
+                }
+            }
+
+            when(decl.datatype) {
+                DataType.FLOAT -> {
+                    // vardecl: for scalar float vars, promote constant integer initialization values to floats
+                    val litval = decl.value as? NumericLiteralValue
+                    if (litval!=null && litval.type in IntegerDatatypes) {
+                        val newValue = NumericLiteralValue(DataType.FLOAT, litval.number.toDouble(), litval.position)
+                        return listOf(IAstModification.ReplaceNode(decl.value!!, newValue, decl))
+                    }
+                }
+                DataType.ARRAY_UB, DataType.ARRAY_B, DataType.ARRAY_UW, DataType.ARRAY_W -> {
+                    val numericLv = decl.value as? NumericLiteralValue
+                    val rangeExpr = decl.value as? RangeExpr
+                    if(rangeExpr!=null) {
+                        // convert the initializer range expression to an actual array
+                        val declArraySize = decl.arraysize?.constIndex()
+                        if(declArraySize!=null && declArraySize!=rangeExpr.size())
+                            errors.err("range expression size doesn't match declared array size", decl.value?.position!!)
+                        val constRange = rangeExpr.toConstantIntegerRange()
+                        if(constRange!=null) {
+                            val eltType = rangeExpr.inferType(program).typeOrElse(DataType.UBYTE)
+                            val newValue = if(eltType in ByteDatatypes) {
+                                ArrayLiteralValue(InferredTypes.InferredType.known(decl.datatype),
+                                        constRange.map { NumericLiteralValue(eltType, it.toShort(), decl.value!!.position) }.toTypedArray(),
+                                        position = decl.value!!.position)
+                            } else {
+                                ArrayLiteralValue(InferredTypes.InferredType.known(decl.datatype),
+                                        constRange.map { NumericLiteralValue(eltType, it, decl.value!!.position) }.toTypedArray(),
+                                        position = decl.value!!.position)
+                            }
+                            return listOf(IAstModification.ReplaceNode(decl.value!!, newValue, decl))
+                        }
+                    }
+                    if(numericLv!=null && numericLv.type== DataType.FLOAT)
+                        errors.err("arraysize requires only integers here", numericLv.position)
+                    val size = decl.arraysize?.constIndex() ?: return noModifications
+                    if (rangeExpr==null && numericLv!=null) {
+                        // arraysize initializer is empty or a single int, and we know the size; create the arraysize.
+                        val fillvalue = numericLv.number.toInt()
+                        when(decl.datatype){
+                            DataType.ARRAY_UB -> {
+                                if(fillvalue !in 0..255)
+                                    errors.err("ubyte value overflow", numericLv.position)
+                            }
+                            DataType.ARRAY_B -> {
+                                if(fillvalue !in -128..127)
+                                    errors.err("byte value overflow", numericLv.position)
+                            }
+                            DataType.ARRAY_UW -> {
+                                if(fillvalue !in 0..65535)
+                                    errors.err("uword value overflow", numericLv.position)
+                            }
+                            DataType.ARRAY_W -> {
+                                if(fillvalue !in -32768..32767)
+                                    errors.err("word value overflow", numericLv.position)
+                            }
+                            else -> {}
+                        }
+                        // create the array itself, filled with the fillvalue.
+                        val array = Array(size) {fillvalue}.map { NumericLiteralValue(ArrayElementTypes.getValue(decl.datatype), it, numericLv.position) }.toTypedArray<Expression>()
+                        val refValue = ArrayLiteralValue(InferredTypes.InferredType.known(decl.datatype), array, position = numericLv.position)
+                        return listOf(IAstModification.ReplaceNode(decl.value!!, refValue, decl))
+                    }
+                }
+                DataType.ARRAY_F -> {
+                    val size = decl.arraysize?.constIndex() ?: return noModifications
+                    val litval = decl.value as? NumericLiteralValue
+                    val rangeExpr = decl.value as? RangeExpr
+                    if(rangeExpr!=null) {
+                        // convert the initializer range expression to an actual array of floats
+                        val declArraySize = decl.arraysize?.constIndex()
+                        if(declArraySize!=null && declArraySize!=rangeExpr.size())
+                            errors.err("range expression size doesn't match declared array size", decl.value?.position!!)
+                        val constRange = rangeExpr.toConstantIntegerRange()
+                        if(constRange!=null) {
+                            val newValue = ArrayLiteralValue(InferredTypes.InferredType.known(DataType.ARRAY_F),
+                                    constRange.map { NumericLiteralValue(DataType.FLOAT, it.toDouble(), decl.value!!.position) }.toTypedArray(),
+                                    position = decl.value!!.position)
+                            return listOf(IAstModification.ReplaceNode(decl.value!!, newValue, decl))
+                        }
+                    }
+                    if(rangeExpr==null && litval!=null) {
+                        // arraysize initializer is a single int, and we know the size.
+                        val fillvalue = litval.number.toDouble()
+                        if (fillvalue < CompilationTarget.instance.machine.FLOAT_MAX_NEGATIVE || fillvalue > CompilationTarget.instance.machine.FLOAT_MAX_POSITIVE)
+                            errors.err("float value overflow", litval.position)
+                        else {
+                            // create the array itself, filled with the fillvalue.
+                            val array = Array(size) {fillvalue}.map { NumericLiteralValue(DataType.FLOAT, it, litval.position) }.toTypedArray<Expression>()
+                            val refValue = ArrayLiteralValue(InferredTypes.InferredType.known(DataType.ARRAY_F), array, position = litval.position)
+                            return listOf(IAstModification.ReplaceNode(decl.value!!, refValue, decl))
+                        }
+                    }
+                }
+                else -> {
+                    // nothing to do for this type
+                    // this includes strings and structs
+                }
+            }
+        }
+
+        return noModifications
+    }
+}

compiler/src/prog8/optimizer/ExpressionSimplifier.kt

@@ -175,28 +175,6 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
                 // unsigned >= 0 --> true
                 return listOf(IAstModification.ReplaceNode(expr, NumericLiteralValue.fromBoolean(true, expr.position), parent))
             }
-            when(leftDt) {
-                DataType.BYTE -> {
-                    // signed >=0   --> signed ^ $80
-                    return listOf(IAstModification.ReplaceNode(
-                            expr,
-                            BinaryExpression(expr.left, "^", NumericLiteralValue.optimalInteger(0x80, expr.position), expr.position),
-                            parent
-                    ))
-                }
-                DataType.WORD -> {
-                    // signedw >=0   --> msb(signedw) ^ $80
-                    return listOf(IAstModification.ReplaceNode(
-                            expr,
-                            BinaryExpression(FunctionCall(IdentifierReference(listOf("msb"), expr.position),
-                                    mutableListOf(expr.left),
-                                    expr.position
-                            ), "^", NumericLiteralValue.optimalInteger(0x80, expr.position), expr.position),
-                            parent
-                    ))
-                }
-                else -> {}
-            }
         }
 
         if(expr.operator == "<" && rightVal?.number == 0) {
@@ -297,6 +275,49 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
         return noModifications
     }
 
+    override fun after(functionCall: FunctionCall, parent: Node): Iterable<IAstModification> {
+        if(functionCall.target.nameInSource == listOf("lsb")) {
+            val arg = functionCall.args[0]
+            if(arg is TypecastExpression) {
+                val valueDt = arg.expression.inferType(program)
+                if (valueDt.istype(DataType.BYTE) || valueDt.istype(DataType.UBYTE)) {
+                    // useless lsb() of byte value that was casted to word
+                    return listOf(IAstModification.ReplaceNode(functionCall, arg.expression, parent))
+                }
+            } else {
+                val argDt = arg.inferType(program)
+                if (argDt.istype(DataType.BYTE) || argDt.istype(DataType.UBYTE)) {
+                    // useless lsb() of byte value
+                    return listOf(IAstModification.ReplaceNode(functionCall, arg, parent))
+                }
+            }
+        }
+        else if(functionCall.target.nameInSource == listOf("msb")) {
+            val arg = functionCall.args[0]
+            if(arg is TypecastExpression) {
+                val valueDt = arg.expression.inferType(program)
+                if (valueDt.istype(DataType.BYTE) || valueDt.istype(DataType.UBYTE)) {
+                    // useless msb() of byte value that was casted to word, replace with 0
+                    return listOf(IAstModification.ReplaceNode(
+                            functionCall,
+                            NumericLiteralValue(valueDt.typeOrElse(DataType.UBYTE), 0, arg.expression.position),
+                            parent))
+                }
+            } else {
+                val argDt = arg.inferType(program)
+                if (argDt.istype(DataType.BYTE) || argDt.istype(DataType.UBYTE)) {
+                    // useless msb() of byte value, replace with 0
+                    return listOf(IAstModification.ReplaceNode(
+                            functionCall,
+                            NumericLiteralValue(argDt.typeOrElse(DataType.UBYTE), 0, arg.position),
+                            parent))
+                }
+            }
+        }
+
+        return noModifications
+    }
+
     private fun determineY(x: Expression, subBinExpr: BinaryExpression): Expression? {
         return when {
             subBinExpr.left isSameAs x -> subBinExpr.right
@@ -330,6 +351,13 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
         }
         // no need to check for left val constant (because of associativity)
 
+        val rnum = rightVal?.number?.toDouble()
+        if(rnum!=null && rnum<0.0) {
+            expr.operator = "-"
+            expr.right = NumericLiteralValue(rightVal.type, -rnum, rightVal.position)
+            return expr
+        }
+
         return null
     }
 
@@ -344,12 +372,16 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
 
         if (rightVal != null) {
             // right value is a constant, see if we can optimize
-            val rightConst: NumericLiteralValue = rightVal
-            when (rightConst.number.toDouble()) {
-                0.0 -> {
-                    // left
-                    return expr.left
-                }
+            val rnum = rightVal.number.toDouble()
+            if (rnum == 0.0) {
+                // left
+                return expr.left
+            }
+
+            if(rnum<0.0) {
+                expr.operator = "+"
+                expr.right = NumericLiteralValue(rightVal.type, -rnum, rightVal.position)
+                return expr
             }
         }
         if (leftVal != null) {
@@ -362,6 +394,7 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
             }
         }
 
+
         return null
     }
 
@@ -444,7 +477,10 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
         when (expr.operator) {
             "%" -> {
                 if (cv == 1.0) {
-                    return NumericLiteralValue(expr.inferType(program).typeOrElse(DataType.STRUCT), 0, expr.position)
+                    val idt = expr.inferType(program)
+                    if(!idt.isKnown)
+                        throw FatalAstException("unknown dt")
+                    return NumericLiteralValue(idt.typeOrElse(DataType.STRUCT), 0, expr.position)
                 } else if (cv == 2.0) {
                     expr.operator = "&"
                     expr.right = NumericLiteralValue.optimalInteger(1, expr.position)
@@ -573,8 +609,10 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
         if (amount == 0) {
             return expr.left
         }
-        val targetDt = expr.left.inferType(program).typeOrElse(DataType.STRUCT)
-        when (targetDt) {
+        val targetIDt = expr.left.inferType(program)
+        if(!targetIDt.isKnown)
+            throw FatalAstException("unknown dt")
+        when (val targetDt = targetIDt.typeOrElse(DataType.STRUCT)) {
             DataType.UBYTE, DataType.BYTE -> {
                 if (amount >= 8) {
                     return NumericLiteralValue(targetDt, 0, expr.position)
@@ -584,7 +622,7 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
                 if (amount >= 16) {
                     return NumericLiteralValue(targetDt, 0, expr.position)
                 } else if (amount >= 8) {
-                    val lsb = TypecastExpression(expr.left, DataType.UBYTE, true, expr.position)
+                    val lsb = FunctionCall(IdentifierReference(listOf("lsb"), expr.position), mutableListOf(expr.left), expr.position)
                     if (amount == 8) {
                         return FunctionCall(IdentifierReference(listOf("mkword"), expr.position), mutableListOf(lsb, NumericLiteralValue.optimalInteger(0, expr.position)), expr.position)
                     }
@@ -606,7 +644,10 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
         if (amount == 0) {
             return expr.left
         }
-        when (expr.left.inferType(program).typeOrElse(DataType.STRUCT)) {
+        val idt = expr.left.inferType(program)
+        if(!idt.isKnown)
+            throw FatalAstException("unknown dt")
+        when (idt.typeOrElse(DataType.STRUCT)) {
             DataType.UBYTE -> {
                 if (amount >= 8) {
                     return NumericLiteralValue.optimalInteger(0, expr.position)
@@ -623,8 +664,9 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
                     return NumericLiteralValue.optimalInteger(0, expr.position)
                 } else if (amount >= 8) {
                     val msb = FunctionCall(IdentifierReference(listOf("msb"), expr.position), mutableListOf(expr.left), expr.position)
-                    if (amount == 8)
-                        return msb
+                    if (amount == 8) {
+                        return TypecastExpression(msb, DataType.UWORD, true, expr.position)
+                    }
                     return BinaryExpression(msb, ">>", NumericLiteralValue.optimalInteger(amount - 8, expr.position), expr.position)
                 }
             }
@@ -632,14 +674,6 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
                 if (amount > 16) {
                     expr.right = NumericLiteralValue.optimalInteger(16, expr.right.position)
                     return null
-                } else if (amount >= 8) {
-                    val msbAsByte = TypecastExpression(
-                            FunctionCall(IdentifierReference(listOf("msb"), expr.position), mutableListOf(expr.left), expr.position),
-                            DataType.BYTE,
-                            true, expr.position)
-                    if (amount == 8)
-                        return msbAsByte
-                    return BinaryExpression(msbAsByte, ">>", NumericLiteralValue.optimalInteger(amount - 8, expr.position), expr.position)
                 }
             }
             else -> {

compiler/src/prog8/optimizer/Extensions.kt

@@ -5,20 +5,31 @@ import prog8.ast.base.ErrorReporter
 
 
 internal fun Program.constantFold(errors: ErrorReporter) {
-    val replacer = ConstantIdentifierReplacer(this, errors)
-    replacer.visit(this)
+    val valuetypefixer = VarConstantValueTypeAdjuster(this)
+    valuetypefixer.visit(this)
     if(errors.isEmpty()) {
-        replacer.applyModifications()
+        valuetypefixer.applyModifications()
 
-        val optimizer = ConstantFoldingOptimizer(this)
-        optimizer.visit(this)
-        while (errors.isEmpty() && optimizer.applyModifications() > 0) {
-            optimizer.visit(this)
-        }
-
-        if(errors.isEmpty()) {
-            replacer.visit(this)
+        val replacer = ConstantIdentifierReplacer(this, errors)
+        replacer.visit(this)
+        if (errors.isEmpty()) {
             replacer.applyModifications()
+
+            valuetypefixer.visit(this)
+            if(errors.isEmpty()) {
+                valuetypefixer.applyModifications()
+
+                val optimizer = ConstantFoldingOptimizer(this)
+                optimizer.visit(this)
+                while (errors.isEmpty() && optimizer.applyModifications() > 0) {
+                    optimizer.visit(this)
+                }
+
+                if (errors.isEmpty()) {
+                    replacer.visit(this)
+                    replacer.applyModifications()
+                }
+            }
         }
     }
 
@@ -42,3 +53,9 @@ internal fun Program.simplifyExpressions() : Int {
     opti.visit(this)
     return opti.applyModifications()
 }
+
+internal fun Program.splitBinaryExpressions() : Int {
+    val opti = BinExprSplitter(this)
+    opti.visit(this)
+    return opti.applyModifications()
+}

compiler/src/prog8/optimizer/StatementOptimizer.kt

@@ -25,12 +25,12 @@ internal class StatementOptimizer(private val program: Program,
         if("force_output" !in block.options()) {
             if (block.containsNoCodeNorVars()) {
                 errors.warn("removing empty block '${block.name}'", block.position)
-                return listOf(IAstModification.Remove(block, parent))
+                return listOf(IAstModification.Remove(block, parent as INameScope))
             }
 
             if (block !in callgraph.usedSymbols) {
                 errors.warn("removing unused block '${block.name}'", block.position)
-                return listOf(IAstModification.Remove(block, parent))
+                return listOf(IAstModification.Remove(block, parent as INameScope))
             }
         }
         return noModifications
@@ -42,38 +42,28 @@ internal class StatementOptimizer(private val program: Program,
             if(subroutine.containsNoCodeNorVars()) {
                 errors.warn("removing empty subroutine '${subroutine.name}'", subroutine.position)
                 val removals = callgraph.calledBy.getValue(subroutine).map {
-                    IAstModification.Remove(it, it.parent)
+                    IAstModification.Remove(it, it.definingScope())
                 }.toMutableList()
-                removals += IAstModification.Remove(subroutine, parent)
+                removals += IAstModification.Remove(subroutine, subroutine.definingScope())
                 return removals
             }
         }
 
-        val linesToRemove = deduplicateAssignments(subroutine.statements)
-        if(linesToRemove.isNotEmpty()) {
-            linesToRemove.reversed().forEach{subroutine.statements.removeAt(it)}
-        }
-
         if(subroutine !in callgraph.usedSymbols && !forceOutput) {
             errors.warn("removing unused subroutine '${subroutine.name}'", subroutine.position)
-            return listOf(IAstModification.Remove(subroutine, parent))
+            return listOf(IAstModification.Remove(subroutine, subroutine.definingScope()))
         }
 
         return noModifications
     }
 
-    override fun after(scope: AnonymousScope, parent: Node): Iterable<IAstModification> {
-        val linesToRemove = deduplicateAssignments(scope.statements)
-        return linesToRemove.reversed().map { IAstModification.Remove(scope.statements[it], scope) }
-    }
-
     override fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> {
         val forceOutput = "force_output" in decl.definingBlock().options()
         if(decl !in callgraph.usedSymbols && !forceOutput) {
             if(decl.type == VarDeclType.VAR)
                 errors.warn("removing unused variable '${decl.name}'", decl.position)
 
-            return listOf(IAstModification.Remove(decl, parent))
+            return listOf(IAstModification.Remove(decl, decl.definingScope()))
         }
 
         return noModifications
@@ -84,7 +74,7 @@ internal class StatementOptimizer(private val program: Program,
             val functionName = functionCallStatement.target.nameInSource[0]
             if (functionName in pureBuiltinFunctions) {
                 errors.warn("statement has no effect (function return value is discarded)", functionCallStatement.position)
-                return listOf(IAstModification.Remove(functionCallStatement, parent))
+                return listOf(IAstModification.Remove(functionCallStatement, functionCallStatement.definingScope()))
             }
         }
 
@@ -104,7 +94,7 @@ internal class StatementOptimizer(private val program: Program,
                 if(string!=null) {
                     val pos = functionCallStatement.position
                     if (string.value.length == 1) {
-                        val firstCharEncoded = CompilationTarget.encodeString(string.value, string.altEncoding)[0]
+                        val firstCharEncoded = CompilationTarget.instance.encodeString(string.value, string.altEncoding)[0]
                         val chrout = FunctionCallStatement(
                                 IdentifierReference(listOf("c64", "CHROUT"), pos),
                                 mutableListOf(NumericLiteralValue(DataType.UBYTE, firstCharEncoded.toInt(), pos)),
@@ -112,7 +102,7 @@ internal class StatementOptimizer(private val program: Program,
                         )
                         return listOf(IAstModification.ReplaceNode(functionCallStatement, chrout, parent))
                     } else if (string.value.length == 2) {
-                        val firstTwoCharsEncoded = CompilationTarget.encodeString(string.value.take(2), string.altEncoding)
+                        val firstTwoCharsEncoded = CompilationTarget.instance.encodeString(string.value.take(2), string.altEncoding)
                         val chrout1 = FunctionCallStatement(
                                 IdentifierReference(listOf("c64", "CHROUT"), pos),
                                 mutableListOf(NumericLiteralValue(DataType.UBYTE, firstTwoCharsEncoded[0].toInt(), pos)),
@@ -137,7 +127,7 @@ internal class StatementOptimizer(private val program: Program,
         if(subroutine!=null) {
             val first = subroutine.statements.asSequence().filterNot { it is VarDecl || it is Directive }.firstOrNull()
             if(first is Return)
-                return listOf(IAstModification.Remove(functionCallStatement, parent))
+                return listOf(IAstModification.Remove(functionCallStatement, functionCallStatement.definingScope()))
         }
 
         return noModifications
@@ -160,7 +150,7 @@ internal class StatementOptimizer(private val program: Program,
     override fun after(ifStatement: IfStatement, parent: Node): Iterable<IAstModification> {
         // remove empty if statements
         if(ifStatement.truepart.containsNoCodeNorVars() && ifStatement.elsepart.containsNoCodeNorVars())
-            return listOf(IAstModification.Remove(ifStatement, parent))
+            return listOf(IAstModification.Remove(ifStatement, ifStatement.definingScope()))
 
         // empty true part? switch with the else part
         if(ifStatement.truepart.containsNoCodeNorVars() && ifStatement.elsepart.containsCodeOrVars()) {
@@ -193,12 +183,12 @@ internal class StatementOptimizer(private val program: Program,
     override fun after(forLoop: ForLoop, parent: Node): Iterable<IAstModification> {
         if(forLoop.body.containsNoCodeNorVars()) {
             errors.warn("removing empty for loop", forLoop.position)
-            return listOf(IAstModification.Remove(forLoop, parent))
+            return listOf(IAstModification.Remove(forLoop, forLoop.definingScope()))
         } else if(forLoop.body.statements.size==1) {
             val loopvar = forLoop.body.statements[0] as? VarDecl
             if(loopvar!=null && loopvar.name==forLoop.loopVar.nameInSource.singleOrNull()) {
                 // remove empty for loop (only loopvar decl in it)
-                return listOf(IAstModification.Remove(forLoop, parent))
+                return listOf(IAstModification.Remove(forLoop, forLoop.definingScope()))
             }
         }
 
@@ -220,7 +210,7 @@ internal class StatementOptimizer(private val program: Program,
                 val size = sv.value.length
                 if(size==1) {
                     // loop over string of length 1 -> just assign the single character
-                    val character = CompilationTarget.encodeString(sv.value, sv.altEncoding)[0]
+                    val character = CompilationTarget.instance.encodeString(sv.value, sv.altEncoding)[0]
                     val byte = NumericLiteralValue(DataType.UBYTE, character, iterable.position)
                     val scope = AnonymousScope(mutableListOf(), forLoop.position)
                     scope.statements.add(Assignment(AssignTarget(forLoop.loopVar, null, null, forLoop.position), byte, forLoop.position))
@@ -249,11 +239,11 @@ internal class StatementOptimizer(private val program: Program,
     }
 
     override fun before(untilLoop: UntilLoop, parent: Node): Iterable<IAstModification> {
-        val constvalue = untilLoop.untilCondition.constValue(program)
+        val constvalue = untilLoop.condition.constValue(program)
         if(constvalue!=null) {
             if(constvalue.asBooleanValue) {
                 // always true -> keep only the statement block (if there are no break statements)
-                errors.warn("condition is always true", untilLoop.untilCondition.position)
+                errors.warn("condition is always true", untilLoop.condition.position)
                 if(!hasBreak(untilLoop.body))
                     return listOf(IAstModification.ReplaceNode(untilLoop, untilLoop.body, parent))
             } else {
@@ -275,7 +265,7 @@ internal class StatementOptimizer(private val program: Program,
             } else {
                 // always false -> remove the while statement altogether
                 errors.warn("condition is always false", whileLoop.condition.position)
-                listOf(IAstModification.Remove(whileLoop, parent))
+                listOf(IAstModification.Remove(whileLoop, whileLoop.definingScope()))
             }
         }
         return noModifications
@@ -286,12 +276,12 @@ internal class StatementOptimizer(private val program: Program,
         if(iter!=null) {
             if(repeatLoop.body.containsNoCodeNorVars()) {
                 errors.warn("empty loop removed", repeatLoop.position)
-                return listOf(IAstModification.Remove(repeatLoop, parent))
+                return listOf(IAstModification.Remove(repeatLoop, repeatLoop.definingScope()))
             }
             val iterations = iter.constValue(program)?.number?.toInt()
             if (iterations == 0) {
                 errors.warn("iterations is always 0, removed loop", iter.position)
-                return listOf(IAstModification.Remove(repeatLoop, parent))
+                return listOf(IAstModification.Remove(repeatLoop, repeatLoop.definingScope()))
             }
             if (iterations == 1) {
                 errors.warn("iterations is always 1", iter.position)
@@ -318,7 +308,7 @@ internal class StatementOptimizer(private val program: Program,
         val scope = jump.definingScope()
         val label = jump.identifier?.targetStatement(scope)
         if(label!=null && scope.statements.indexOf(label) == scope.statements.indexOf(jump)+1)
-            return listOf(IAstModification.Remove(jump, parent))
+            return listOf(IAstModification.Remove(jump, jump.definingScope()))
 
         return noModifications
     }
@@ -351,7 +341,7 @@ internal class StatementOptimizer(private val program: Program,
                                 )
                                 return listOf(
                                         IAstModification.ReplaceNode(binExpr, expr2, binExpr.parent),
-                                        IAstModification.InsertAfter(assignment, addConstant, parent))
+                                        IAstModification.InsertAfter(assignment, addConstant, assignment.definingScope()))
                             } else if (op2 == "-") {
                                 // A = A +/- B - N
                                 val expr2 = BinaryExpression(binExpr.left, binExpr.operator, rExpr.left, binExpr.position)
@@ -362,7 +352,7 @@ internal class StatementOptimizer(private val program: Program,
                                 )
                                 return listOf(
                                         IAstModification.ReplaceNode(binExpr, expr2, binExpr.parent),
-                                        IAstModification.InsertAfter(assignment, subConstant, parent))
+                                        IAstModification.InsertAfter(assignment, subConstant, assignment.definingScope()))
                             }
                         }
                     }
@@ -384,10 +374,10 @@ internal class StatementOptimizer(private val program: Program,
     override fun after(assignment: Assignment, parent: Node): Iterable<IAstModification> {
         if(assignment.target isSameAs assignment.value) {
             // remove assignment to self
-            return listOf(IAstModification.Remove(assignment, parent))
+            return listOf(IAstModification.Remove(assignment, assignment.definingScope()))
         }
 
-        val targetIDt = assignment.target.inferType(program, assignment)
+        val targetIDt = assignment.target.inferType(program)
         if(!targetIDt.isKnown)
             throw FatalAstException("can't infer type of assignment target")
 
@@ -395,21 +385,21 @@ internal class StatementOptimizer(private val program: Program,
         val targetDt = targetIDt.typeOrElse(DataType.STRUCT)
         val bexpr=assignment.value as? BinaryExpression
         if(bexpr!=null) {
-            val cv = bexpr.right.constValue(program)?.number?.toDouble()
-            if (cv != null && assignment.target isSameAs bexpr.left) {
+            val rightCv = bexpr.right.constValue(program)?.number?.toDouble()
+            if (rightCv != null && assignment.target isSameAs bexpr.left) {
                 // assignments of the form:  X = X <operator> <expr>
                 // remove assignments that have no effect (such as X=X+0)
                 // optimize/rewrite some other expressions
                 val vardeclDt = (assignment.target.identifier?.targetVarDecl(program.namespace))?.type
                 when (bexpr.operator) {
                     "+" -> {
-                        if (cv == 0.0) {
-                            return listOf(IAstModification.Remove(assignment, parent))
-                        } else if (targetDt in IntegerDatatypes && floor(cv) == cv) {
-                            if (vardeclDt != VarDeclType.MEMORY && cv in 1.0..4.0) {
+                        if (rightCv == 0.0) {
+                            return listOf(IAstModification.Remove(assignment, assignment.definingScope()))
+                        } else if (targetDt in IntegerDatatypes && floor(rightCv) == rightCv) {
+                            if (vardeclDt != VarDeclType.MEMORY && rightCv in 1.0..4.0) {
                                 // replace by several INCs if it's not a memory address (inc on a memory mapped register doesn't work very well)
                                 val incs = AnonymousScope(mutableListOf(), assignment.position)
-                                repeat(cv.toInt()) {
+                                repeat(rightCv.toInt()) {
                                     incs.statements.add(PostIncrDecr(assignment.target, "++", assignment.position))
                                 }
                                 return listOf(IAstModification.ReplaceNode(assignment, incs, parent))
@@ -417,62 +407,38 @@ internal class StatementOptimizer(private val program: Program,
                         }
                     }
                     "-" -> {
-                        if (cv == 0.0) {
-                            return listOf(IAstModification.Remove(assignment, parent))
-                        } else if (targetDt in IntegerDatatypes && floor(cv) == cv) {
-                            if (vardeclDt != VarDeclType.MEMORY && cv in 1.0..4.0) {
+                        if (rightCv == 0.0) {
+                            return listOf(IAstModification.Remove(assignment, assignment.definingScope()))
+                        } else if (targetDt in IntegerDatatypes && floor(rightCv) == rightCv) {
+                            if (vardeclDt != VarDeclType.MEMORY && rightCv in 1.0..4.0) {
                                 // replace by several DECs if it's not a memory address (dec on a memory mapped register doesn't work very well)
                                 val decs = AnonymousScope(mutableListOf(), assignment.position)
-                                repeat(cv.toInt()) {
+                                repeat(rightCv.toInt()) {
                                     decs.statements.add(PostIncrDecr(assignment.target, "--", assignment.position))
                                 }
                                 return listOf(IAstModification.ReplaceNode(assignment, decs, parent))
                             }
                         }
                     }
-                    "*" -> if (cv == 1.0) return listOf(IAstModification.Remove(assignment, parent))
-                    "/" -> if (cv == 1.0) return listOf(IAstModification.Remove(assignment, parent))
-                    "**" -> if (cv == 1.0) return listOf(IAstModification.Remove(assignment, parent))
-                    "|" -> if (cv == 0.0) return listOf(IAstModification.Remove(assignment, parent))
-                    "^" -> if (cv == 0.0) return listOf(IAstModification.Remove(assignment, parent))
+                    "*" -> if (rightCv == 1.0) return listOf(IAstModification.Remove(assignment, assignment.definingScope()))
+                    "/" -> if (rightCv == 1.0) return listOf(IAstModification.Remove(assignment, assignment.definingScope()))
+                    "**" -> if (rightCv == 1.0) return listOf(IAstModification.Remove(assignment, assignment.definingScope()))
+                    "|" -> if (rightCv == 0.0) return listOf(IAstModification.Remove(assignment, assignment.definingScope()))
+                    "^" -> if (rightCv == 0.0) return listOf(IAstModification.Remove(assignment, assignment.definingScope()))
                     "<<" -> {
-                        if (cv == 0.0)
-                            return listOf(IAstModification.Remove(assignment, parent))
+                        if (rightCv == 0.0)
+                            return listOf(IAstModification.Remove(assignment, assignment.definingScope()))
                     }
                     ">>" -> {
-                        if (cv == 0.0)
-                            return listOf(IAstModification.Remove(assignment, parent))
+                        if (rightCv == 0.0)
+                            return listOf(IAstModification.Remove(assignment, assignment.definingScope()))
                     }
                 }
 
             }
         }
-        return noModifications
-    }
 
-    private fun deduplicateAssignments(statements: List<Statement>): MutableList<Int> {
-        // removes 'duplicate' assignments that assign the isSameAs target
-        val linesToRemove = mutableListOf<Int>()
-        var previousAssignmentLine: Int? = null
-        for (i in statements.indices) {
-            val stmt = statements[i] as? Assignment
-            if (stmt != null && stmt.value is NumericLiteralValue) {
-                if (previousAssignmentLine == null) {
-                    previousAssignmentLine = i
-                    continue
-                } else {
-                    val prev = statements[previousAssignmentLine] as Assignment
-                    if (prev.target.isSameAs(stmt.target, program)) {
-                        // get rid of the previous assignment, if the target is not MEMORY
-                        if (prev.target.isNotMemory(program.namespace))
-                            linesToRemove.add(previousAssignmentLine)
-                    }
-                    previousAssignmentLine = i
-                }
-            } else
-                previousAssignmentLine = null
-        }
-        return linesToRemove
+        return noModifications
     }
 
     private fun hasBreak(scope: INameScope): Boolean {

compiler/src/prog8/optimizer/UnusedCodeRemover.kt

@@ -8,7 +8,8 @@ import prog8.ast.processing.AstWalker
 import prog8.ast.processing.IAstModification
 import prog8.ast.statements.*
 
-internal class UnusedCodeRemover(private val errors: ErrorReporter): AstWalker() {
+
+internal class UnusedCodeRemover(private val program: Program, private val errors: ErrorReporter): AstWalker() {
 
     override fun before(program: Program, parent: Node): Iterable<IAstModification> {
         val callgraph = CallGraph(program)
@@ -19,20 +20,20 @@ internal class UnusedCodeRemover(private val errors: ErrorReporter): AstWalker()
         program.modules.forEach {
             callgraph.forAllSubroutines(it) { sub ->
                 if (sub !== entrypoint && !sub.isAsmSubroutine && (callgraph.calledBy[sub].isNullOrEmpty() || sub.containsNoCodeNorVars())) {
-                    removals.add(IAstModification.Remove(sub, sub.definingScope() as Node))
+                    removals.add(IAstModification.Remove(sub, sub.definingScope()))
                 }
             }
         }
 
         program.modules.flatMap { it.statements }.filterIsInstance<Block>().forEach { block ->
             if (block.containsNoCodeNorVars() && "force_output" !in block.options())
-                removals.add(IAstModification.Remove(block, block.definingScope() as Node))
+                removals.add(IAstModification.Remove(block, block.definingScope()))
         }
 
         // remove modules that are not imported, or are empty (unless it's a library modules)
         program.modules.forEach {
             if (!it.isLibraryModule && (it.importedBy.isEmpty() || it.containsNoCodeNorVars()))
-                removals.add(IAstModification.Remove(it, it.parent))
+                removals.add(IAstModification.Remove(it, it.definingScope()))
         }
 
         return removals
@@ -66,4 +67,37 @@ internal class UnusedCodeRemover(private val errors: ErrorReporter): AstWalker()
             else -> errors.warn("unreachable code", next.position)
         }
     }
+
+    override fun after(scope: AnonymousScope, parent: Node): Iterable<IAstModification> {
+        val removeDoubleAssignments = deduplicateAssignments(scope.statements)
+        return removeDoubleAssignments.map { IAstModification.Remove(it, scope) }
+    }
+
+    override fun after(block: Block, parent: Node): Iterable<IAstModification> {
+        val removeDoubleAssignments = deduplicateAssignments(block.statements)
+        return removeDoubleAssignments.map { IAstModification.Remove(it, block) }
+    }
+
+    override fun after(subroutine: Subroutine, parent: Node): Iterable<IAstModification> {
+        val removeDoubleAssignments = deduplicateAssignments(subroutine.statements)
+        return removeDoubleAssignments.map { IAstModification.Remove(it, subroutine) }
+    }
+
+    private fun deduplicateAssignments(statements: List<Statement>): List<Assignment> {
+        // removes 'duplicate' assignments that assign the same target directly after another
+        val linesToRemove = mutableListOf<Assignment>()
+
+        for (stmtPairs in statements.windowed(2, step = 1)) {
+            val assign1 = stmtPairs[0] as? Assignment
+            val assign2 = stmtPairs[1] as? Assignment
+            if (assign1 != null && assign2 != null && !assign2.isAugmentable) {
+                if (assign1.target.isSameAs(assign2.target, program) && assign1.target.isInRegularRAM(program.namespace))  {
+                    if(assign2.target.identifier==null || !assign2.value.referencesIdentifier(*(assign2.target.identifier!!.nameInSource.toTypedArray())))
+                        linesToRemove.add(assign1)
+                }
+            }
+        }
+
+        return linesToRemove
+    }
 }

compiler/src/prog8/parser/ModuleParsing.kt

@@ -9,6 +9,7 @@ import prog8.ast.base.SyntaxError
 import prog8.ast.base.checkImportedValid
 import prog8.ast.statements.Directive
 import prog8.ast.statements.DirectiveArg
+import prog8.compiler.target.CompilationTarget
 import prog8.pathFrom
 import java.io.InputStream
 import java.nio.file.Files
@@ -19,14 +20,6 @@ import java.nio.file.Paths
 internal class ParsingFailedError(override var message: String) : Exception(message)
 
 
-private class LexerErrorListener: BaseErrorListener() {
-    var  numberOfErrors: Int = 0
-    override fun syntaxError(p0: Recognizer<*, *>?, p1: Any?, p2: Int, p3: Int, p4: String?, p5: RecognitionException?) {
-        numberOfErrors++
-    }
-}
-
-
 internal class CustomLexer(val modulePath: Path, input: CharStream?) : prog8Lexer(input)
 
 
@@ -60,13 +53,28 @@ internal class ModuleImporter {
         return executeImportDirective(program, import, Paths.get(""))
     }
 
+    private class MyErrorListener: ConsoleErrorListener() {
+        var  numberOfErrors: Int = 0
+        override fun syntaxError(recognizer: Recognizer<*, *>?, offendingSymbol: Any?, line: Int, charPositionInLine: Int, msg: String, e: RecognitionException?) {
+            numberOfErrors++
+            when (recognizer) {
+                is CustomLexer -> System.err.println("${recognizer.modulePath}:$line:$charPositionInLine: $msg")
+                is prog8Parser -> System.err.println("${recognizer.inputStream.sourceName}:$line:$charPositionInLine: $msg")
+                else -> System.err.println("$line:$charPositionInLine $msg")
+            }
+        }
+    }
+
     private fun importModule(program: Program, stream: CharStream, modulePath: Path, isLibrary: Boolean): Module {
         val moduleName = moduleName(modulePath.fileName)
         val lexer = CustomLexer(modulePath, stream)
-        val lexerErrors = LexerErrorListener()
+        lexer.removeErrorListeners()
+        val lexerErrors = MyErrorListener()
         lexer.addErrorListener(lexerErrors)
         val tokens = CommentHandlingTokenStream(lexer)
         val parser = prog8Parser(tokens)
+        parser.removeErrorListeners()
+        parser.addErrorListener(MyErrorListener())
         val parseTree = parser.module()
         val numberOfErrors = parser.numberOfSyntaxErrors + lexerErrors.numberOfErrors
         if(numberOfErrors > 0)
@@ -84,7 +92,7 @@ internal class ModuleImporter {
         // accept additional imports
         val lines = moduleAst.statements.toMutableList()
         lines.asSequence()
-                .mapIndexed { i, it -> Pair(i, it) }
+                .mapIndexed { i, it -> i to it }
                 .filter { (it.second as? Directive)?.directive == "%import" }
                 .forEach { executeImportDirective(program, it.second as Directive, modulePath) }
 
@@ -123,13 +131,14 @@ internal class ModuleImporter {
         if(existing!=null)
             return null
 
-        val resource = tryGetEmbeddedResource("$moduleName.p8")
+        val rsc = tryGetEmbeddedResource("$moduleName.p8")
         val importedModule =
-                if(resource!=null) {
+                if(rsc!=null) {
                     // load the module from the embedded resource
+                    val (resource, resourcePath) = rsc
                     resource.use {
                         println("importing '$moduleName' (library)")
-                        importModule(program, CharStreams.fromStream(it), Paths.get("@embedded@/$moduleName"), true)
+                        importModule(program, CharStreams.fromStream(it), Paths.get("@embedded@/$resourcePath"), true)
                     }
                 } else {
                     val modulePath = discoverImportedModuleFile(moduleName, source, import.position)
@@ -140,7 +149,18 @@ internal class ModuleImporter {
         return importedModule
     }
 
-    private fun tryGetEmbeddedResource(name: String): InputStream? {
-        return object{}.javaClass.getResourceAsStream("/prog8lib/$name")
+    private fun tryGetEmbeddedResource(name: String): Pair<InputStream, String>? {
+        val target = CompilationTarget.instance.name
+        val targetSpecificPath = "/prog8lib/$target/$name"
+        val targetSpecificResource = object{}.javaClass.getResourceAsStream(targetSpecificPath)
+        if(targetSpecificResource!=null)
+            return Pair(targetSpecificResource, targetSpecificPath)
+
+        val generalPath = "/prog8lib/$name"
+        val generalResource = object{}.javaClass.getResourceAsStream(generalPath)
+        if(generalResource!=null)
+            return Pair(generalResource, generalPath)
+
+        return null
     }
 }

compiler/test/UnitTests.kt

@@ -5,17 +5,18 @@ import org.hamcrest.Matchers.closeTo
 import org.hamcrest.Matchers.equalTo
 import org.junit.jupiter.api.Test
 import org.junit.jupiter.api.TestInstance
-import prog8.ast.base.DataType
-import prog8.ast.base.ErrorReporter
-import prog8.ast.base.Position
-import prog8.ast.expressions.NumericLiteralValue
-import prog8.ast.expressions.StringLiteralValue
+import prog8.ast.base.*
+import prog8.ast.expressions.*
+import prog8.ast.statements.*
 import prog8.compiler.*
+import prog8.compiler.target.C64Target
+import prog8.compiler.target.CompilationTarget
 import prog8.compiler.target.c64.C64MachineDefinition.C64Zeropage
 import prog8.compiler.target.c64.C64MachineDefinition.FLOAT_MAX_NEGATIVE
 import prog8.compiler.target.c64.C64MachineDefinition.FLOAT_MAX_POSITIVE
 import prog8.compiler.target.c64.C64MachineDefinition.Mflpt5
 import prog8.compiler.target.c64.Petscii
+import prog8.compiler.target.cx16.CX16MachineDefinition
 import java.io.CharConversionException
 import kotlin.test.*
 
@@ -129,7 +130,7 @@ class TestC64Zeropage {
 
     @Test
     fun testNames() {
-        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), false))
+        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), false, false))
 
         zp.allocate("", DataType.UBYTE, null, errors)
         zp.allocate("", DataType.UBYTE, null, errors)
@@ -142,37 +143,37 @@ class TestC64Zeropage {
 
     @Test
     fun testZpFloatEnable() {
-        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), false))
+        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), false, false))
         assertFailsWith<CompilerException> {
             zp.allocate("", DataType.FLOAT, null, errors)
         }
-        val zp2 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.DONTUSE, emptyList(), true))
+        val zp2 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.DONTUSE, emptyList(), true, false))
         assertFailsWith<CompilerException> {
             zp2.allocate("", DataType.FLOAT, null, errors)
         }
-        val zp3 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FLOATSAFE, emptyList(), true))
+        val zp3 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FLOATSAFE, emptyList(), true, false))
         zp3.allocate("", DataType.FLOAT, null, errors)
     }
 
     @Test
     fun testZpModesWithFloats() {
-        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), false))
-        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.KERNALSAFE, emptyList(), false))
-        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), false))
-        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FLOATSAFE, emptyList(), false))
-        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), true))
-        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FLOATSAFE, emptyList(), true))
+        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), false, false))
+        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.KERNALSAFE, emptyList(), false, false))
+        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), false, false))
+        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FLOATSAFE, emptyList(), false, false))
+        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), true, false))
+        C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FLOATSAFE, emptyList(), true, false))
         assertFailsWith<CompilerException> {
-            C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), true))
+            C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), true, false))
         }
         assertFailsWith<CompilerException> {
-            C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.KERNALSAFE, emptyList(), true))
+            C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.KERNALSAFE, emptyList(), true, false))
         }
     }
 
     @Test
     fun testZpDontuse() {
-        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.DONTUSE, emptyList(), false))
+        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.DONTUSE, emptyList(), false, false))
         println(zp.free)
         assertEquals(0, zp.available())
         assertFailsWith<CompilerException> {
@@ -182,19 +183,19 @@ class TestC64Zeropage {
 
     @Test
     fun testFreeSpaces() {
-        val zp1 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), true))
-        assertEquals(16, zp1.available())
-        val zp2 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FLOATSAFE, emptyList(), false))
-        assertEquals(91, zp2.available())
-        val zp3 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.KERNALSAFE, emptyList(), false))
+        val zp1 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), true, false))
+        assertEquals(18, zp1.available())
+        val zp2 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FLOATSAFE, emptyList(), false, false))
+        assertEquals(89, zp2.available())
+        val zp3 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.KERNALSAFE, emptyList(), false, false))
         assertEquals(125, zp3.available())
-        val zp4 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), false))
+        val zp4 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), false, false))
         assertEquals(238, zp4.available())
     }
 
     @Test
     fun testReservedSpace() {
-        val zp1 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), false))
+        val zp1 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), false, false))
         assertEquals(238, zp1.available())
         assertTrue(50 in zp1.free)
         assertTrue(100 in zp1.free)
@@ -203,7 +204,7 @@ class TestC64Zeropage {
         assertTrue(200 in zp1.free)
         assertTrue(255 in zp1.free)
         assertTrue(199 in zp1.free)
-        val zp2 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, listOf(50 .. 100, 200..255), false))
+        val zp2 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, listOf(50 .. 100, 200..255), false, false))
         assertEquals(139, zp2.available())
         assertFalse(50 in zp2.free)
         assertFalse(100 in zp2.free)
@@ -216,8 +217,8 @@ class TestC64Zeropage {
 
     @Test
     fun testBasicsafeAllocation() {
-        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), true))
-        assertEquals(16, zp.available())
+        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), true, false))
+        assertEquals(18, zp.available())
 
         assertFailsWith<ZeropageDepletedError> {
             // in regular zp there aren't 5 sequential bytes free
@@ -239,7 +240,7 @@ class TestC64Zeropage {
 
     @Test
     fun testFullAllocation() {
-        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), false))
+        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), false, false))
         assertEquals(238, zp.available())
         val loc = zp.allocate("", DataType.UWORD, null, errors)
         assertTrue(loc > 3)
@@ -269,20 +270,38 @@ class TestC64Zeropage {
 
     @Test
     fun testEfficientAllocation() {
-        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(),  true))
-        assertEquals(16, zp.available())
+        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(),  true, false))
+        assertEquals(18, zp.available())
         assertEquals(0x04, zp.allocate("", DataType.WORD, null, errors))
         assertEquals(0x06, zp.allocate("", DataType.UBYTE, null, errors))
         assertEquals(0x0a, zp.allocate("", DataType.UBYTE, null, errors))
-        assertEquals(0x94, zp.allocate("", DataType.UWORD, null, errors))
-        assertEquals(0xa7, zp.allocate("", DataType.UWORD, null, errors))
-        assertEquals(0xa9, zp.allocate("", DataType.UWORD, null, errors))
-        assertEquals(0xb5, zp.allocate("", DataType.UWORD, null, errors))
-        assertEquals(0xf7, zp.allocate("", DataType.UWORD, null, errors))
+        assertEquals(0x9b, zp.allocate("", DataType.UWORD, null, errors))
+        assertEquals(0x9e, zp.allocate("", DataType.UWORD, null, errors))
+        assertEquals(0xa5, zp.allocate("", DataType.UWORD, null, errors))
+        assertEquals(0xb0, zp.allocate("", DataType.UWORD, null, errors))
+        assertEquals(0xbe, zp.allocate("", DataType.UWORD, null, errors))
         assertEquals(0x0e, zp.allocate("", DataType.UBYTE, null, errors))
+        assertEquals(0x92, zp.allocate("", DataType.UBYTE, null, errors))
+        assertEquals(0x96, zp.allocate("", DataType.UBYTE, null, errors))
         assertEquals(0xf9, zp.allocate("", DataType.UBYTE, null, errors))
         assertEquals(0, zp.available())
     }
+
+    @Test
+    fun testReservedLocations() {
+        val zp = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), false, false))
+        assertEquals(zp.SCRATCH_REG, zp.SCRATCH_B1+1, "zp _B1 and _REG must be next to each other to create a word")
+    }
+}
+
+
+@TestInstance(TestInstance.Lifecycle.PER_CLASS)
+class TestCx16Zeropage {
+    @Test
+    fun testReservedLocations() {
+        val zp = CX16MachineDefinition.CX16Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), false, false))
+        assertEquals(zp.SCRATCH_REG, zp.SCRATCH_B1+1, "zp _B1 and _REG must be next to each other to create a word")
+    }
 }
 
 
@@ -379,3 +398,169 @@ class TestPetscii {
         assertFalse(abc!=abc)
     }
 }
+
+
+class TestMemory {
+    @Test
+    fun testInValidRamC64_memory_addresses() {
+        CompilationTarget.instance = C64Target
+
+        var memexpr = NumericLiteralValue.optimalInteger(0x0000, Position.DUMMY)
+        var target = AssignTarget(null, null, DirectMemoryWrite(memexpr, Position.DUMMY), Position.DUMMY)
+        var scope = AnonymousScope(mutableListOf(), Position.DUMMY)
+        assertTrue(target.isInRegularRAM(scope))
+
+        memexpr = NumericLiteralValue.optimalInteger(0x1000, Position.DUMMY)
+        target = AssignTarget(null, null, DirectMemoryWrite(memexpr, Position.DUMMY), Position.DUMMY)
+        scope = AnonymousScope(mutableListOf(), Position.DUMMY)
+        assertTrue(target.isInRegularRAM(scope))
+
+        memexpr = NumericLiteralValue.optimalInteger(0x9fff, Position.DUMMY)
+        target = AssignTarget(null, null, DirectMemoryWrite(memexpr, Position.DUMMY), Position.DUMMY)
+        scope = AnonymousScope(mutableListOf(), Position.DUMMY)
+        assertTrue(target.isInRegularRAM(scope))
+
+        memexpr = NumericLiteralValue.optimalInteger(0xc000, Position.DUMMY)
+        target = AssignTarget(null, null, DirectMemoryWrite(memexpr, Position.DUMMY), Position.DUMMY)
+        scope = AnonymousScope(mutableListOf(), Position.DUMMY)
+        assertTrue(target.isInRegularRAM(scope))
+
+        memexpr = NumericLiteralValue.optimalInteger(0xcfff, Position.DUMMY)
+        target = AssignTarget(null, null, DirectMemoryWrite(memexpr, Position.DUMMY), Position.DUMMY)
+        scope = AnonymousScope(mutableListOf(), Position.DUMMY)
+        assertTrue(target.isInRegularRAM(scope))
+    }
+
+    @Test
+    fun testNotInValidRamC64_memory_addresses() {
+        CompilationTarget.instance = C64Target
+
+        var memexpr = NumericLiteralValue.optimalInteger(0xa000, Position.DUMMY)
+        var target = AssignTarget(null, null, DirectMemoryWrite(memexpr, Position.DUMMY), Position.DUMMY)
+        var scope = AnonymousScope(mutableListOf(), Position.DUMMY)
+        assertFalse(target.isInRegularRAM(scope))
+
+        memexpr = NumericLiteralValue.optimalInteger(0xafff, Position.DUMMY)
+        target = AssignTarget(null, null, DirectMemoryWrite(memexpr, Position.DUMMY), Position.DUMMY)
+        scope = AnonymousScope(mutableListOf(), Position.DUMMY)
+        assertFalse(target.isInRegularRAM(scope))
+
+        memexpr = NumericLiteralValue.optimalInteger(0xd000, Position.DUMMY)
+        target = AssignTarget(null, null, DirectMemoryWrite(memexpr, Position.DUMMY), Position.DUMMY)
+        scope = AnonymousScope(mutableListOf(), Position.DUMMY)
+        assertFalse(target.isInRegularRAM(scope))
+
+        memexpr = NumericLiteralValue.optimalInteger(0xffff, Position.DUMMY)
+        target = AssignTarget(null, null, DirectMemoryWrite(memexpr, Position.DUMMY), Position.DUMMY)
+        scope = AnonymousScope(mutableListOf(), Position.DUMMY)
+        assertFalse(target.isInRegularRAM(scope))
+    }
+
+    @Test
+    fun testInValidRamC64_memory_identifiers() {
+        CompilationTarget.instance = C64Target
+        var target = createTestProgramForMemoryRefViaVar(0x1000, VarDeclType.VAR)
+        assertTrue(target.isInRegularRAM(target.definingScope()))
+        target = createTestProgramForMemoryRefViaVar(0xd020, VarDeclType.VAR)
+        assertFalse(target.isInRegularRAM(target.definingScope()))
+        target = createTestProgramForMemoryRefViaVar(0x1000, VarDeclType.CONST)
+        assertTrue(target.isInRegularRAM(target.definingScope()))
+        target = createTestProgramForMemoryRefViaVar(0xd020, VarDeclType.CONST)
+        assertFalse(target.isInRegularRAM(target.definingScope()))
+        target = createTestProgramForMemoryRefViaVar(0x1000, VarDeclType.MEMORY)
+        assertFalse(target.isInRegularRAM(target.definingScope()))
+    }
+
+    @Test
+    private fun createTestProgramForMemoryRefViaVar(address: Int, vartype: VarDeclType): AssignTarget {
+        val decl = VarDecl(vartype, DataType.BYTE, ZeropageWish.DONTCARE, null, "address", null, NumericLiteralValue.optimalInteger(address, Position.DUMMY), false, false, Position.DUMMY)
+        val memexpr = IdentifierReference(listOf("address"), Position.DUMMY)
+        val target = AssignTarget(null, null, DirectMemoryWrite(memexpr, Position.DUMMY), Position.DUMMY)
+        val assignment = Assignment(target, NumericLiteralValue.optimalInteger(0, Position.DUMMY), Position.DUMMY)
+        val subroutine = Subroutine("test", emptyList(), emptyList(), emptyList(), emptyList(), emptySet(), null, false, mutableListOf(decl, assignment), Position.DUMMY)
+        subroutine.linkParents(ParentSentinel)
+        return target
+    }
+
+    @Test
+    fun testInValidRamC64_memory_expression() {
+        CompilationTarget.instance = C64Target
+        val memexpr = PrefixExpression("+", NumericLiteralValue.optimalInteger(0x1000, Position.DUMMY), Position.DUMMY)
+        val target = AssignTarget(null, null, DirectMemoryWrite(memexpr, Position.DUMMY), Position.DUMMY)
+        val scope = AnonymousScope(mutableListOf(), Position.DUMMY)
+        assertFalse(target.isInRegularRAM(scope))
+    }
+
+    @Test
+    fun testInValidRamC64_variable() {
+        CompilationTarget.instance = C64Target
+        val decl = VarDecl(VarDeclType.VAR, DataType.BYTE, ZeropageWish.DONTCARE, null, "address", null, null, false, false, Position.DUMMY)
+        val target = AssignTarget(IdentifierReference(listOf("address"), Position.DUMMY), null, null, Position.DUMMY)
+        val assignment = Assignment(target, NumericLiteralValue.optimalInteger(0, Position.DUMMY), Position.DUMMY)
+        val subroutine = Subroutine("test", emptyList(), emptyList(), emptyList(), emptyList(), emptySet(), null, false, mutableListOf(decl, assignment), Position.DUMMY)
+        subroutine.linkParents(ParentSentinel)
+        assertTrue(target.isInRegularRAM(target.definingScope()))
+    }
+
+    @Test
+    fun testInValidRamC64_memmap_variable() {
+        CompilationTarget.instance = C64Target
+        val address = 0x1000
+        val decl = VarDecl(VarDeclType.MEMORY, DataType.UBYTE, ZeropageWish.DONTCARE, null, "address", null, NumericLiteralValue.optimalInteger(address, Position.DUMMY), false, false, Position.DUMMY)
+        val target = AssignTarget(IdentifierReference(listOf("address"), Position.DUMMY), null, null, Position.DUMMY)
+        val assignment = Assignment(target, NumericLiteralValue.optimalInteger(0, Position.DUMMY), Position.DUMMY)
+        val subroutine = Subroutine("test", emptyList(), emptyList(), emptyList(), emptyList(), emptySet(), null, false, mutableListOf(decl, assignment), Position.DUMMY)
+        subroutine.linkParents(ParentSentinel)
+        assertTrue(target.isInRegularRAM(target.definingScope()))
+    }
+
+    @Test
+    fun testNotInValidRamC64_memmap_variable() {
+        CompilationTarget.instance = C64Target
+        val address = 0xd020
+        val decl = VarDecl(VarDeclType.MEMORY, DataType.UBYTE, ZeropageWish.DONTCARE, null, "address", null, NumericLiteralValue.optimalInteger(address, Position.DUMMY), false, false, Position.DUMMY)
+        val target = AssignTarget(IdentifierReference(listOf("address"), Position.DUMMY), null, null, Position.DUMMY)
+        val assignment = Assignment(target, NumericLiteralValue.optimalInteger(0, Position.DUMMY), Position.DUMMY)
+        val subroutine = Subroutine("test", emptyList(), emptyList(), emptyList(), emptyList(), emptySet(), null, false, mutableListOf(decl, assignment), Position.DUMMY)
+        subroutine.linkParents(ParentSentinel)
+        assertFalse(target.isInRegularRAM(target.definingScope()))
+    }
+
+    @Test
+    fun testInValidRamC64_array() {
+        CompilationTarget.instance = C64Target
+        val decl = VarDecl(VarDeclType.VAR, DataType.ARRAY_UB, ZeropageWish.DONTCARE, null, "address", null, null, false, false, Position.DUMMY)
+        val arrayindexed = ArrayIndexedExpression(IdentifierReference(listOf("address"), Position.DUMMY), ArrayIndex(NumericLiteralValue.optimalInteger(1, Position.DUMMY), Position.DUMMY), Position.DUMMY)
+        val target = AssignTarget(null, arrayindexed, null, Position.DUMMY)
+        val assignment = Assignment(target, NumericLiteralValue.optimalInteger(0, Position.DUMMY), Position.DUMMY)
+        val subroutine = Subroutine("test", emptyList(), emptyList(), emptyList(), emptyList(), emptySet(), null, false, mutableListOf(decl, assignment), Position.DUMMY)
+        subroutine.linkParents(ParentSentinel)
+        assertTrue(target.isInRegularRAM(target.definingScope()))
+    }
+
+    @Test
+    fun testInValidRamC64_array_memmapped() {
+        CompilationTarget.instance = C64Target
+        val address = 0x1000
+        val decl = VarDecl(VarDeclType.MEMORY, DataType.ARRAY_UB, ZeropageWish.DONTCARE, null, "address", null, NumericLiteralValue.optimalInteger(address, Position.DUMMY), false, false, Position.DUMMY)
+        val arrayindexed = ArrayIndexedExpression(IdentifierReference(listOf("address"), Position.DUMMY), ArrayIndex(NumericLiteralValue.optimalInteger(1, Position.DUMMY), Position.DUMMY), Position.DUMMY)
+        val target = AssignTarget(null, arrayindexed, null, Position.DUMMY)
+        val assignment = Assignment(target, NumericLiteralValue.optimalInteger(0, Position.DUMMY), Position.DUMMY)
+        val subroutine = Subroutine("test", emptyList(), emptyList(), emptyList(), emptyList(), emptySet(), null, false, mutableListOf(decl, assignment), Position.DUMMY)
+        subroutine.linkParents(ParentSentinel)
+        assertTrue(target.isInRegularRAM(target.definingScope()))
+    }
+
+    @Test
+    fun testNotValidRamC64_array_memmapped() {
+        CompilationTarget.instance = C64Target
+        val address = 0xe000
+        val decl = VarDecl(VarDeclType.MEMORY, DataType.ARRAY_UB, ZeropageWish.DONTCARE, null, "address", null, NumericLiteralValue.optimalInteger(address, Position.DUMMY), false, false, Position.DUMMY)
+        val arrayindexed = ArrayIndexedExpression(IdentifierReference(listOf("address"), Position.DUMMY), ArrayIndex(NumericLiteralValue.optimalInteger(1, Position.DUMMY), Position.DUMMY), Position.DUMMY)
+        val target = AssignTarget(null, arrayindexed, null, Position.DUMMY)
+        val assignment = Assignment(target, NumericLiteralValue.optimalInteger(0, Position.DUMMY), Position.DUMMY)
+        val subroutine = Subroutine("test", emptyList(), emptyList(), emptyList(), emptyList(), emptySet(), null, false, mutableListOf(decl, assignment), Position.DUMMY)
+        subroutine.linkParents(ParentSentinel)
+        assertFalse(target.isInRegularRAM(target.definingScope()))
+    }
+}

docs/source/building.rst

@@ -41,7 +41,7 @@ of that build task, you can start the compiler with:
 
 (You should probably make an alias...)
 
-.. note::
+.. hint::
     Development and testing is done on Linux, but the compiler should run on most
     operating systems. If you do have trouble building or running
     the compiler on another operating system, please let me know!
@@ -83,7 +83,7 @@ For normal use the compiler is invoked with the command:
 By default, assembly code is generated and written to ``sourcefile.asm``.
 It is then (automatically) fed to the `64tass <https://sourceforge.net/projects/tass64/>`_ cross assembler tool
 that assembles it into the final program.
-If you use the option to let the compiler auto-start a C-64 emulator, it will do so after
+If you use the option to let the compiler auto-start an emulator, it will do so after
 a successful compilation. This will load your program and the symbol and breakpoint lists
 (for the machine code monitor) into the emulator.
 
@@ -93,6 +93,8 @@ Almost instant compilation times (less than a second) can be achieved when using
 Start the compiler with the ``-watch`` argument to enable this.
 It will compile your program and then instead of exiting, it waits for any changes in the module source files.
 As soon as a change happens, the program gets compiled again.
+It is possible to use the watch mode with multiple modules as well, but it will
+recompile everything in that list even if only of the files got updated.
 
 Other options
 ^^^^^^^^^^^^^
@@ -149,10 +151,10 @@ If your running program hits one of the breakpoints, Vice will halt execution an
 Troubleshooting
 ---------------
 
-Getting an assembler error about undefined symbols such as ``not defined 'c64flt'``?
-This happens when your program uses floating point values, and you forgot to import ``c64flt`` library.
+Getting an assembler error about undefined symbols such as ``not defined 'floats'``?
+This happens when your program uses floating point values, and you forgot to import ``floats`` library.
 If you use floating points, the compiler needs routines from that library.
-Fix it by adding an ``%import c64flt``.
+Fix it by adding an ``%import floats``.
 
 
 Examples

docs/source/index.rst

@@ -12,6 +12,7 @@ What is Prog8?
 
 This is an experimental compiled programming language targeting the 8-bit
 `6502 <https://en.wikipedia.org/wiki/MOS_Technology_6502>`_ /
+`65c02 <https://en.wikipedia.org/wiki/MOS_Technology_65C02>`_ /
 `6510 <https://en.wikipedia.org/wiki/MOS_Technology_6510>`_ microprocessor.
 This CPU is from the late 1970's and early 1980's and was used in many home computers from that era,
 such as the `Commodore-64 <https://en.wikipedia.org/wiki/Commodore_64>`_.
@@ -37,21 +38,43 @@ This software is licensed under the GNU GPL 3.0, see https://www.gnu.org/license
     :alt: Fully playable tetris clone
 
 
-Code examples
--------------
+Language features
+-----------------
+
+- It is a cross-compiler running on modern machines (Linux, MacOS, Windows, ...)
+  The generated output is a machine code program runnable on actual 8-bit 6502 hardware.
+- Provide a very convenient edit/compile/run cycle by being able to directly launch
+  the compiled program in an emulator and provide debugging information to this emulator.
+- Based on simple and familiar imperative structured programming (it looks like a mix of C and Python)
+- Modular programming and scoping via modules, code blocks, and subroutines.
+- Provide high level programming constructs but at the same time stay close to the metal;
+  still able to directly use memory addresses and ROM subroutines,
+  and inline assembly to have full control when every register, cycle or byte matters
+- Arbitrary number of subroutine parameters, Complex nested expressions are possible
+- No stack frame allocations because parameters and local variables are automatically allocated statically
+- Nested subroutines can access variables from outer scopes to avoids the overhead to pass everything via parameters
+- Variable data types include signed and unsigned bytes and words, arrays, strings and floats.
+- High-level code optimizations, such as const-folding, expression and statement simplifications/rewriting.
+- Many built-in functions, such as ``sin``, ``cos``, ``rnd``, ``abs``, ``min``, ``max``, ``sqrt``, ``msb``, ``rol``, ``ror``, ``swap``, ``memset``, ``memcopy``, ``substr``, ``sort`` and ``reverse`` (and others)
+- If you only use standard kernel and prog8 library routines, it is possible to compile the *exact same program* for both machines (just change the compiler target flag)!
+
+
+Code example
+------------
 
 This code calculates prime numbers using the Sieve of Eratosthenes algorithm::
 
-    %import c64textio
+    %import textio
     %zeropage basicsafe
 
     main {
-
         ubyte[256] sieve
-        ubyte candidate_prime = 2
+        ubyte candidate_prime = 2       ; is increased in the loop
 
         sub start() {
-            memset(sieve, 256, false)   ; clear the sieve
+            ; clear the sieve, to reset starting situation on subsequent runs
+            memset(sieve, 256, false)
+            ; calculate primes
             txt.print("prime numbers up to 255:\n\n")
             ubyte amount=0
             repeat {
@@ -62,18 +85,19 @@ This code calculates prime numbers using the Sieve of Eratosthenes algorithm::
                 txt.print(", ")
                 amount++
             }
-            c64.CHROUT('\n')
+            txt.chrout('\n')
             txt.print("number of primes (expected 54): ")
             txt.print_ub(amount)
-            c64.CHROUT('\n')
+            txt.chrout('\n')
         }
 
         sub find_next_prime() -> ubyte {
             while sieve[candidate_prime] {
                 candidate_prime++
                 if candidate_prime==0
-                    return 0        ; we wrapped; no more primes available
+                    return 0        ; we wrapped; no more primes available in the sieve
             }
+
             ; found next one, mark the multiples and return it.
             sieve[candidate_prime] = true
             uword multiple = candidate_prime
@@ -87,82 +111,19 @@ This code calculates prime numbers using the Sieve of Eratosthenes algorithm::
     }
 
 
-
 when compiled an ran on a C-64 you get this:
 
 .. image:: _static/primes_example.png
     :align: center
     :alt: result when run on C-64
 
+when the exact same program is compiled for the Commander X16 target, and run on the emulator, you get this:
 
-// TODO fix code example
-The following programs shows a use of the high level ``struct`` type::
+.. image:: _static/primes_cx16.png
+    :align: center
+    :alt: result when run on CX16 emulator
 
 
-    %import c64textio
-    %zeropage basicsafe
-
-    main {
-
-        struct Color {
-            ubyte red
-            ubyte green
-            ubyte blue
-        }
-
-        sub start() {
-
-            Color purple = [255, 0, 255]
-
-            Color other
-
-            other = purple
-            other.red /= 2
-            other.green = 10 + other.green / 2
-            other.blue = 99
-
-            txt.print_ub(other.red)
-            c64.CHROUT(',')
-            txt.print_ub(other.green)
-            c64.CHROUT(',')
-            txt.print_ub(other.blue)
-            c64.CHROUT('\n')
-        }
-    }
-
-
-when compiled and ran, it prints  ``127,10,99`` on the screen.
-
-
-Design principles and features
-------------------------------
-
-- It is a cross-compiler running on modern machines (Linux, MacOS, Windows, ...)
-  The generated output is a machine code program runnable on actual 8-bit 6502 hardware.
-- Based on simple and familiar imperative structured programming (it looks like a mix of C and Python)
-- 'One statement per line' code, resulting in clear readable programs.
-- Modular programming and scoping via modules, code blocks, and subroutines.
-- Provide high level programming constructs but at the same time stay close to the metal;
-  still able to directly use memory addresses and ROM subroutines,
-  and inline assembly to have full control when every register, cycle or byte matters
-- Arbitrary number of subroutine parameters
-- Complex nested expressions are possible
-- Nested subroutines can access variables from outer scopes to avoids the overhead to pass everything via parameters
-- Values are typed. Available data types include signed and unsigned bytes and words, arrays, strings and floats.
-- No dynamic memory allocation or sizing! All variables stay fixed size as determined at compile time.
-- Provide various quality of life language features and library subroutines specifically for the target platform.
-- Provide a very convenient edit/compile/run cycle by being able to directly launch
-  the compiled program in an emulator and provide debugging information to this emulator.
-- Arbitrary control flow jumps and branches are possible,
-  and will usually translate directly into the appropriate single 6502 jump/branch instruction.
-- There are no complicated built-in error handling or overflow checks, you'll have to take care
-  of this yourself if required. This keeps the language and code simple and efficient.
-- The compiler tries to optimize the program and generated code a bit, but hand-tuning of the
-  performance or space-critical parts will likely still be required. This is supported by
-  the ability to easily write embedded assembly code directly in the program source code.
-- There are many built-in functions, such as ``sin``, ``cos``, ``rnd``, ``abs``, ``min``, ``max``, ``sqrt``, ``msb``, ``rol``, ``ror``, ``swap``, ``memset``, ``memcopy``, ``substr``, ``sort`` and ``reverse`` (and others)
-- Assembling the generated code into a program wil be done by an external cross-assembler tool.
-
 
 .. _requirements:
 
@@ -172,13 +133,17 @@ Required tools
 `64tass <https://sourceforge.net/projects/tass64/>`_ - cross assembler. Install this on your shell path.
 It's very easy to compile yourself.
 A recent precompiled .exe for Windows can be obtained from my `clone <https://github.com/irmen/64tass/releases>`_ of this project.
+*You need at least version 1.55.2257 of this assembler to correctly use the breakpoints feature.*
+It's possible to use older versions, but it is very likely that the automatic Vice breakpoints won't work with them.
 
-A **Java runtime (jre or jdk), version 8 or newer**  is required to run the prog8 compiler itself.
-If you're scared of Oracle's licensing terms, most Linux distributions ship OpenJDK instead.
-Fnd for Windows it's possible to get that as well. Check out `AdoptOpenJDK <https://adoptopenjdk.net/>`_ .
+A **Java runtime (jre or jdk), version 11 or newer**  is required to run the prog8 compiler itself.
+If you're scared of Oracle's licensing terms, most Linux distributions ship OpenJDK in their packages repository instead.
+For Windows it's possible to get that as well; check out `AdoptOpenJDK <https://adoptopenjdk.net/>`_ .
+For MacOS you can use the Homebrew system to install a recent version of OpenJDK.
 
-Finally: a **C-64 emulator** (or a real C-64 ofcourse) can be nice to test and run your programs on.
-The compiler assumes the presence of the `Vice emulator <http://vice-emu.sourceforge.net/>`_.
+Finally: an **emulator** (or a real machine ofcourse) to test and run your programs on.
+In C64 mode, thhe compiler assumes the presence of the `Vice emulator <http://vice-emu.sourceforge.net/>`_.
+If you're targeting the CommanderX16 instead, there's the `x16emu <https://github.com/commanderx16/x16-emulator>`_.
 
 .. important::
     **Building the compiler itself:** (*Only needed if you have not downloaded a pre-built 'fat-jar'*)
@@ -199,6 +164,7 @@ The compiler assumes the presence of the `Vice emulator <http://vice-emu.sourcef
     building.rst
     programming.rst
     syntaxreference.rst
+    libraries.rst
     todo.rst
 
 

docs/source/libraries.rst

@@ -0,0 +1,104 @@
+************************
+Compiler library modules
+************************
+
+The compiler provides several "built-in" library modules with useful subroutine and variables.
+
+Some of these may be specific for a certain compilation target, or work slightly different,
+but some effort is put into making them available across compilation targets.
+
+This means that as long as your program is only using the subroutines from these
+libraries and not using hardware- and/or system dependent code, and isn't hardcoding certain
+assumptions like the screen size, the exact same source program can
+be compiled for multiple different target platforms. Many of the example programs that come
+with Prog8 are written like this.
+
+You can ``%import`` and use these modules explicitly, but the compiler may also import one or more
+of these library modules automatically as required.
+
+.. caution::
+    The resulting compiled binary program *only works on the target machine it was compiled for*.
+    You must recompile the program for every target you want to run it on.
+
+
+
+syslib
+------
+The "system library" for your target machine. It contains many system-specific definitions such
+as ROM/kernal subroutine definitions, memory location constants, and utility subroutines.
+
+Many of these definitions overlap for the C64 and Commander X16 targets so it is still possible
+to write programs that work on both targets without modifications.
+
+conv
+----
+Routines to convert strings to numbers or vice versa.
+
+- numbers to strings, in various formats (binary, hex, decimal)
+- strings in decimal, hex and binary format into numbers
+
+
+textio (txt.*)
+--------------
+This will probably be the most used library module. It contains a whole lot of routines
+dealing with text-based input and output (to the screen). Such as
+
+- printing strings and numbers
+- reading text input from the user via the keyboard
+- filling or clearing the screen and colors
+- scrolling the text on the screen
+- placing individual characters on the screen
+
+
+diskio
+------
+Provides several routines that deal with disk drive I/O, such as:
+
+- list files on disk, optionally filtering by prefix or suffix
+- show disk directory as-is
+- display disk drive status
+- load and save data from and to the disk
+- delete and rename files on the disk
+
+
+floats
+------
+Provides definitions for the ROM/kernel subroutines and utility routines dealing with floating
+point variables.  This includes ``print_f``, the routine used to print floating point numbers.
+
+
+graphics
+--------
+High-res monochrome bitmap graphics routines:
+
+- clearing the screen
+- drawing lines
+- drawing circles and discs (filled circles)
+- plotting individual pixels
+
+
+math
+----
+Low level math routines. You should not normally have to bother with this directly.
+The compiler needs it to implement most of the math operations in your programs.
+
+
+cx16logo
+--------
+A 'fun' module that contains the Commander X16 logo and that allows you
+to print it anywhere on the screen.
+
+
+c64colors
+---------
+Available for the CommanderX16 target, a module that contains a few better
+color palettes for how the colors of the VIC-II looked on the Commodore-64.
+There are subroutines to activate one of the several palettes of your liking.
+The Commander X16's default colors for this (the first 16 colors) are too saturated
+and are quite different than how a C-64 looked.
+
+
+prog8_lib
+---------
+Low level language support. You should not normally have to bother with this directly.
+The compiler needs it for verious built-in system routines.

docs/source/programming.rst

@@ -226,7 +226,7 @@ This is because routines in the C-64 BASIC and KERNAL ROMs are used for that.
 So floating point operations will only work if the C-64 BASIC ROM (and KERNAL ROM)
 are banked in.
 
-Also your code needs to import the ``c64flt`` library to enable floating point support
+Also your code needs to import the ``floats`` library to enable floating point support
 in the compiler, and to gain access to the floating point routines.
 (this library contains the directive to enable floating points, you don't have
 to worry about this yourself)
@@ -236,12 +236,15 @@ The largest 5-byte MFLPT float that can be stored is: **1.7014118345e+38**   (ne
 
 Arrays
 ^^^^^^
-Array types are also supported. They can be made of bytes, words or floats::
+Array types are also supported. They can be made of bytes, words or floats, strings, and other arrays
+(although the usefulness of the latter is very limited for now)::
 
     byte[10]  array                   ; array of 10 bytes, initially set to 0
     byte[]  array = [1, 2, 3, 4]      ; initialize the array, size taken from value
     byte[99] array = 255              ; initialize array with 99 times 255 [255, 255, 255, 255, ...]
     byte[] array = 100 to 199         ; initialize array with [100, 101, ..., 198, 199]
+    str[] names = ["ally", "pete"]    ; array of string pointers/addresses (equivalent to uword)
+    uword[] others = [names, array]   ; array of pointers/addresses to other arrays
 
     value = array[3]            ; the fourth value in the array (index is 0-based)
     char = string[4]            ; the fifth character (=byte) in the string
@@ -257,6 +260,12 @@ Note that the various keywords for the data type and variable type (``byte``, ``
 can't be used as *identifiers* elsewhere. You can't make a variable, block or subroutine with the name ``byte``
 for instance.
 
+
+It's possible to assign a new array to another array, this will overwrite all elements in the original
+array with those in the value array. The number and types of elements have to match.
+For large arrays this is a slow operation because every element is copied over. It should probably be avoided.
+
+
 **Arrays at a specific memory location:**
 Using the memory-mapped syntax it is possible to define an array to be located at a specific memory location.
 For instance to reference the first 5 rows of the Commodore 64's screen matrix as an array, you can define::
@@ -284,7 +293,7 @@ This @-prefix can also be used for character byte values.
 You can concatenate two string literals using '+' (not very useful though) or repeat
 a string literal a given number of times using '*'. You can also assign a new string
 value to another string. No bounds check is done so be sure the destination string is
-large enough to contain the new value::
+large enough to contain the new value (it is overwritten in memory)::
 
     str string1 = "first part" + "second part"
     str string2 = "hello!" * 10
@@ -293,13 +302,28 @@ large enough to contain the new value::
     string1 = "new value"
 
 
+There are several 'escape sequences' to help you put special characters into strings, such
+as newlines, quote characters themselves, and so on. The ones used most often are
+``\\``, ``\"``, ``\n``, ``\r``.  For a detailed description of all of them and what they mean,
+read the syntax reference on strings.
+
+
+.. hint::
+    Strings/arrays and uwords (=memory address) can often be interchanged.
+    An array of strings is actually an array of uwords where every element is the memory
+    address of the string. You can pass a memory address to assembly functions
+    that require a string as an argument.
+    For regular assignments you still need to use an explicit ``&`` (address-of) to take
+    the address of the string or array.
+
 .. caution::
-    It's probably best to avoid changing strings after they've been created. This
-    includes changing certain letters by index, or by assigning a new value, or by
+    It's probably best to avoid changing the contents in strings and treat them as static.
+    This includes changing certain letters by index, or by assigning a new value, or by
     modifying the string via other means for example ``substr`` function and its cousins.
-    This is because if your program exits and is restarted (without loading it again),
-    it will then start working with the changed strings instead of the original ones!
-    The same is true for arrays.
+    This is because the changes persist in memory. If your program exits and is restarted
+    (without reloading it from disk), it will then start working with the modified strings
+    instead of the original ones!
+    The same is true for arrays! So be careful to (re)initialize them if needed.
 
 
 Structs
@@ -327,7 +351,7 @@ and then create a variable with it::
         ubyte blue
     }
 
-    Color rgb = {255,122,0}     ; note the curly braces here instead of brackets
+    Color rgb = [255,122,0]     ; note that struct initializer value is same as an array
     Color another               ; the init value is optional, like arrays
 
     another = rgb           ; assign all of the values of rgb to another
@@ -352,13 +376,6 @@ address you specified, and setting the varible will directly modify that memory
 	&word  SCREENCOLORS = $d020             ; a 16-bit word at the addres $d020-$d021
 
 
-.. note::
-    Directly accessing random memory locations is not yet supported without the
-    intermediate step of declaring a memory-mapped variable for the memory location.
-    The advantages of this however, is that it's clearer what the memory location
-    stands for, and the compiler also knows the data type.
-
-
 Converting types into other types
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
@@ -578,25 +595,24 @@ within parentheses will be evaluated first. So ``(4 + 8) * 2`` is 24 and not 20,
 and ``(true or false) and false`` is false instead of true.
 
 .. attention::
-    **calculations keep their datatype:**
+    **calculations keep their datatype even if the target variable is larger:**
     When you do calculations on a BYTE type, the result will remain a BYTE.
     When you do calculations on a WORD type, the result will remain a WORD.
     For instance::
 
         byte b = 44
         word w = b*55   ; the result will be 116! (even though the target variable is a word)
-        w *= 999        ; the result will be -15188  (the multiplication stays within a word)
+        w *= 999        ; the result will be -15188  (the multiplication stays within a word, but overflows)
 
-    The compiler will NOT give a warning about this! It's doing this for
+    *The compiler does NOT warn about this!* It's doing this for
     performance reasons - so you won't get sudden 16 bit (or even float)
     calculations where you needed only simple fast byte arithmetic.
     If you do need the extended resulting value, cast at least one of the
-    operands of an operator to the larger datatype. For example::
+    operands explicitly to the larger datatype. For example::
 
         byte b = 44
-        word w = b*55.w     ; the result will be 2420
-        w = (b as word)*55  ; same result
-
+        w = (b as word)*55
+        w = b*(55 as word)
 
 
 
@@ -637,23 +653,20 @@ There's a set of predefined functions in the language. These are fixed and can't
 You can use them in expressions and the compiler will evaluate them at compile-time if possible.
 
 
-sin(x)
-	Sine.  (floating point version)
+Math
+^^^^
+
+abs(x)
+    Absolute value.
+
+atan(x)
+    Arctangent.
+
+ceil(x)
+    Rounds the floating point up to an integer towards positive infinity.
 
 cos(x)
-	Cosine.  (floating point version)
-
-sin8u(x)
-    Fast 8-bit ubyte sine of angle 0..255, result is in range 0..255
-
-sin8(x)
-    Fast 8-bit byte sine of angle 0..255, result is in range -127..127
-
-sin16u(x)
-    Fast 16-bit uword sine of angle 0..255, result is in range 0..65535
-
-sin16(x)
-    Fast 16-bit word sine of angle 0..255, result is in range -32767..32767
+    Cosine.  (floating point version)
 
 cos8u(x)
     Fast 8-bit ubyte cosine of angle 0..255, result is in range 0..255
@@ -667,14 +680,11 @@ cos16u(x)
 cos16(x)
    Fast 16-bit word cosine of angle 0..255, result is in range -32767..32767
 
-abs(x)
-    Absolute value.
+deg(x)
+    Radians to degrees.
 
-tan(x)
-    Tangent.
-
-atan(x)
-    Arctangent.
+floor (x)
+    Rounds the floating point down to an integer towards minus infinity.
 
 ln(x)
     Natural logarithm (base e).
@@ -682,45 +692,48 @@ ln(x)
 log2(x)
     Base 2 logarithm.
 
+rad(x)
+    Degrees to radians.
+
+round(x)
+    Rounds the floating point to the closest integer.
+
+sin(x)
+    Sine.  (floating point version)
+
+sgn(x)
+    Get the sign of the value. Result is -1, 0 or 1 (negative, zero, positive).
+
+sin8u(x)
+    Fast 8-bit ubyte sine of angle 0..255, result is in range 0..255
+
+sin8(x)
+    Fast 8-bit byte sine of angle 0..255, result is in range -127..127
+
+sin16u(x)
+    Fast 16-bit uword sine of angle 0..255, result is in range 0..65535
+
+sin16(x)
+    Fast 16-bit word sine of angle 0..255, result is in range -32767..32767
+
 sqrt16(w)
     16 bit unsigned integer Square root. Result is unsigned byte.
 
 sqrt(x)
     Floating point Square root.
 
-round(x)
-    Rounds the floating point to the closest integer.
+tan(x)
+    Tangent.
 
-floor (x)
-    Rounds the floating point down to an integer towards minus infinity.
 
-ceil(x)
-    Rounds the floating point up to an integer towards positive infinity.
+Array operations
+^^^^^^^^^^^^^^^^
 
-rad(x)
-    Degrees to radians.
+any(x)
+    1 ('true') if any of the values in the array value x is 'true' (not zero), else 0 ('false')
 
-deg(x)
-    Radians to degrees.
-
-max(x)
-    Maximum of the values in the array value x
-
-min(x)
-    Minimum of the values in the array value x
-
-sum(x)
-    Sum of the values in the array value x
-
-sort(array)
-    Sort the array in ascending order (in-place)
-    Note: sorting a floating-point array is not supported right now, as a general sorting routine for this will
-    be extremely slow. Either build one yourself or find another solution that doesn't require sorting
-    floating point values.
-
-reverse(array)
-    Reverse the values in the array (in-place). Supports all data types including floats.
-    Can be used after sort() to sort an array in descending order.
+all(x)
+    1 ('true') if all of the values in the array value x are 'true' (not zero), else 0 ('false')
 
 len(x)
     Number of values in the array value x, or the number of characters in a string (excluding the size or 0-byte).
@@ -729,15 +742,83 @@ len(x)
     length of the string during execution, the value of len(string) may no longer be correct!
     (use strlen function if you want to dynamically determine the length)
 
-sizeof(name)
-    Number of bytes that the object 'name' occupies in memory. This is a constant determined by the data type of
-    the object. For instance, for a variable of type uword, the sizeof is 2.
-    For an 10 element array of floats, it is 50 (on the C-64, where a float is 5 bytes).
-    Note: usually you will be interested in the number of elements in an array, use len() for that.
+max(x)
+    Maximum of the values in the array value x
+
+min(x)
+    Minimum of the values in the array value x
+
+reverse(array)
+    Reverse the values in the array (in-place).
+    Can be used after sort() to sort an array in descending order.
+
+sum(x)
+    Sum of the values in the array value x
+
+sort(array)
+    Sort the array in ascending order (in-place)
+    Supported are arrays of bytes or word values.
+    Sorting a floating-point array is not supported right now, as a general sorting routine for this will
+    be extremely slow. Either build one yourself or find another solution that doesn't require sorting.
+    Finally, note that sorting an array with strings in it will not do what you might think;
+    it considers the array as just an array of integer words and sorts the string *pointers* accordingly.
+    Sorting strings alphabetically has to be programmed yourself if you need it.
+
+
+Strings and memory blocks
+^^^^^^^^^^^^^^^^^^^^^^^^^
+memcopy(from, to, numbytes)
+    Efficiently copy a number of bytes from a memory location to another.
+    NOTE: 'to' must NOT overlap with 'from', unless it is *before* 'from'.
+    Because this function imposes some overhead to handle the parameters,
+    it is only faster if the number of bytes is larger than a certain threshold.
+    Compare the generated code to see if it was beneficial or not.
+    The most efficient will often be to write a specialized copy routine in assembly yourself!
+
+memset(address, numbytes, bytevalue)
+    Efficiently set a part of memory to the given (u)byte value.
+    But the most efficient will always be to write a specialized fill routine in assembly yourself!
+    Note that for clearing the screen, very fast specialized subroutines are
+    available in the ``textio`` and ``graphics`` library modules.
+
+memsetw(address, numwords, wordvalue)
+    Efficiently set a part of memory to the given (u)word value.
+    But the most efficient will always be to write a specialized fill routine in assembly yourself!
+
+leftstr(source, target, length)
+    Copies the left side of the source string of the given length to target string.
+    It is assumed the target string buffer is large enough to contain the result.
+    Also, you have to make sure yourself that length is smaller or equal to the length of the source string.
+    Modifies in-place, doesn't return a value (so can't be used in an expression).
+
+rightstr(source, target, length)
+    Copies the right side of the source string of the given length to target string.
+    It is assumed the target string buffer is large enough to contain the result.
+    Also, you have to make sure yourself that length is smaller or equal to the length of the source string.
+    Modifies in-place, doesn't return a value (so can't be used in an expression).
 
 strlen(str)
     Number of bytes in the string. This value is determined during runtime and counts upto
     the first terminating 0 byte in the string, regardless of the size of the string during compilation time.
+    Don't confuse this with ``len`` and ``sizeof``
+
+strcmp(string1, string2)
+    Returns -1, 0 or 1 depeding on wether string1 sorts before, equal or after string2.
+    Note that you can also directly compare strings and string values with eachother
+    using ``==``, ``<`` etcetera (it will use strcmp for you under water automatically).
+
+substr(source, target, start, length)
+    Copies a segment from the source string, starting at the given index,
+    and of the given length to target string.
+    It is assumed the target string buffer is large enough to contain the result.
+    Also, you have to make sure yourself that start and length are within bounds of the strings.
+    Modifies in-place, doesn't return a value (so can't be used in an expression).
+
+Miscellaneous
+^^^^^^^^^^^^^
+exit(returncode)
+    Immediately stops the program and exits it, with the returncode in the A register.
+    Note: custom interrupt handlers remain active unless manually cleared first!
 
 lsb(x)
     Get the least significant byte of the word x. Equivalent to the cast "x as ubyte".
@@ -745,19 +826,10 @@ lsb(x)
 msb(x)
     Get the most significant byte of the word x.
 
-sgn(x)
-    Get the sign of the value. Result is -1, 0 or 1 (negative, zero, positive).
-
 mkword(msb, lsb)
     Efficiently create a word value from two bytes (the msb and the lsb). Avoids multiplication and shifting.
     So mkword($80, $22) results in $8022.
 
-any(x)
-    1 ('true') if any of the values in the array value x is 'true' (not zero), else 0 ('false')
-
-all(x)
-	1 ('true') if all of the values in the array value x are 'true' (not zero), else 0 ('false')
-
 rnd()
     returns a pseudo-random byte from 0..255
 
@@ -791,51 +863,6 @@ ror2(x)
     It uses some extra logic to not consider the carry flag as extra rotation bit.
     Modifies in-place, doesn't return a value (so can't be used in an expression).
 
-memcopy(from, to, numbytes)
-    Efficiently copy a number of bytes (1 - 256) from a memory location to another.
-    NOTE: 'to' must NOT overlap with 'from', unless it is *before* 'from'.
-    Because this function imposes some overhead to handle the parameters,
-    it is only faster if the number of bytes is larger than a certain threshold.
-    Compare the generated code to see if it was beneficial or not.
-    The most efficient will always be to write a specialized copy routine in assembly yourself!
-
-memset(address, numbytes, bytevalue)
-    Efficiently set a part of memory to the given (u)byte value.
-    But the most efficient will always be to write a specialized fill routine in assembly yourself!
-    Note that for clearing the character screen, very fast specialized subroutines are
-    available in the ``screen`` block (part of the ``c64textio`` or ``cx16textio`` modules)
-
-memsetw(address, numwords, wordvalue)
-    Efficiently set a part of memory to the given (u)word value.
-    But the most efficient will always be to write a specialized fill routine in assembly yourself!
-
-leftstr(source, target, length)
-    Copies the left side of the source string of the given length to target string.
-    It is assumed the target string buffer is large enough to contain the result.
-    Modifies in-place, doesn't return a value (so can't be used in an expression).
-
-rightstr(source, target, length)
-    Copies the right side of the source string of the given length to target string.
-    It is assumed the target string buffer is large enough to contain the result.
-    Modifies in-place, doesn't return a value (so can't be used in an expression).
-
-substr(source, target, start, length)
-    Copies a segment from the source string, starting at the given index,
-    and of the given length to target string.
-    It is assumed the target string buffer is large enough to contain the result.
-    Modifies in-place, doesn't return a value (so can't be used in an expression).
-
-swap(x, y)
-    Swap the values of numerical variables (or memory locations) x and y in a fast way.
-
-set_carry()  /  clear_carry()
-    Set (or clear) the CPU status register Carry flag. No result value.
-    (translated into ``SEC`` or ``CLC`` cpu instruction)
-
-set_irqd()  /  clear_irqd()
-    Set (or clear) the CPU status register Interrupt Disable flag. No result value.
-    (translated into ``SEI`` or ``CLI`` cpu instruction)
-
 rsave()
     Saves the CPU registers and the status flags.
     You can now more or less 'safely' use the registers directly, until you
@@ -850,10 +877,22 @@ rrestore()
 read_flags()
     Returns the current value of the CPU status register.
 
-exit(returncode)
-    Immediately stops the program and exits it, with the returncode in the A register.
-    Note: custom interrupt handlers remain active unless manually cleared first!
+sizeof(name)
+    Number of bytes that the object 'name' occupies in memory. This is a constant determined by the data type of
+    the object. For instance, for a variable of type uword, the sizeof is 2.
+    For an 10 element array of floats, it is 50 (on the C-64, where a float is 5 bytes).
+    Note: usually you will be interested in the number of elements in an array, use len() for that.
 
+set_carry()  /  clear_carry()
+    Set (or clear) the CPU status register Carry flag. No result value.
+    (translated into ``SEC`` or ``CLC`` cpu instruction)
+
+set_irqd()  /  clear_irqd()
+    Set (or clear) the CPU status register Interrupt Disable flag. No result value.
+    (translated into ``SEI`` or ``CLI`` cpu instruction)
+
+swap(x, y)
+    Swap the values of numerical variables (or memory locations) x and y in a fast way.
 
 
 Library routines

docs/source/syntaxreference.rst

@@ -33,6 +33,13 @@ This makes it easier to understand and relate the generated code. Examples::
 Directives
 -----------
 
+.. data:: %target <target>
+
+    Level: module.
+    Global setting, specifies that this module can only work for the given compiler target.
+    If compiled with a different target, compilation is aborted with an error message.
+
+
 .. data:: %output <type>
 
 	Level: module.
@@ -60,7 +67,8 @@ Directives
     - style ``kernalsafe`` -- use the part of the ZP that is 'free' or only used by BASIC routines,
       and don't change anything else.  This allows full use of KERNAL ROM routines (but not BASIC routines),
       including default IRQs during normal system operation.
-      When the program exits, a system reset is performed (because BASIC will be in a corrupt state).
+      It's not possible to return cleanly to BASIC when the program exits. The only choice is
+      to perform a system reset. (A ``system_reset`` subroutine is available in the syslib to help you do this)
     - style ``floatsafe`` -- like the previous one but also reserves the addresses that
       are required to perform floating point operations (from the BASIC kernel). No clean exit is possible.
     - style ``basicsafe`` -- the most restricted mode; only use the handful 'free' addresses in the ZP, and don't
@@ -70,10 +78,11 @@ Directives
     - style ``full`` -- claim the whole ZP for variables for the program, overwriting everything,
       except the few addresses mentioned above that are used by the system's IRQ routine.
       Even though the default IRQ routine is still active, it is impossible to use most BASIC and KERNAL ROM routines.
-      This includes many floating point operations and several utility routines that do I/O, such as ``print_string``.
-      As with ``kernalsafe``, it is not possible to cleanly exit the program, other than to reset the machine.
+      This includes many floating point operations and several utility routines that do I/O, such as ``print``.
       This option makes programs smaller and faster because even more variables can
       be stored in the ZP (which allows for more efficient assembly code).
+      It's not possible to return cleanly to BASIC when the program exits. The only choice is
+      to perform a system reset. (A ``system_reset`` subroutine is available in the syslib to help you do this)
     - style ``dontuse`` -- don't use *any* location in the zeropage.
 
     Also read :ref:`zeropage`.
@@ -110,33 +119,39 @@ Directives
 
 	Level: module, block.
 	Sets special compiler options.
-	For a module option, only the ``enable_floats`` option is recognised, which will tell the compiler
-	to deal with floating point numbers (by using various subroutines from the Commodore-64 kernal).
-	Otherwise, floating point support is not enabled.
-	When used in a block with the ``force_output`` option, it will force the block to be outputted
-	in the final program. Can be useful to make sure some
-	data is generated that would otherwise be discarded because it's not referenced (such as sprite data).
+
+    - For a module option, there is ``enable_floats``, which will tell the compiler
+      to deal with floating point numbers (by using various subroutines from the Commodore-64 kernal).
+      Otherwise, floating point support is not enabled. Normally you don't have to use this yourself as
+      importing the ``floats`` library is required anyway and that will enable it for you automatically.
+    - There's also ``no_sysinit`` which cause the resulting program to *not* include
+      the system re-initialization logic of clearing the screen, resetting I/O config etc. You'll have to
+      take care of that yourself. The program will just start running from whatever state the machine is in when the
+      program was launched.
+    - When used in a block with the ``force_output`` option, it will force the block to be outputted
+      in the final program. Can be useful to make sure some
+      data is generated that would otherwise be discarded because it's not referenced (such as sprite data).
 
 
 .. data:: %asmbinary "<filename>" [, <offset>[, <length>]]
 
-	Level: block.
-        This directive can only be used inside a block.
-        The assembler will include the file as binary bytes at this point, prog8 will not process this at all.
-        The optional offset and length can be used to select a particular piece of the file.
-        The file is located relative to the current working directory!
+    Level: block.
+    This directive can only be used inside a block.
+    The assembler will include the file as binary bytes at this point, prog8 will not process this at all.
+    The optional offset and length can be used to select a particular piece of the file.
+    The file is located relative to the current working directory!
 
 .. data:: %asminclude "<filename>", "scopelabel"
 
-	Level: block.
-        This directive can only be used inside a block.
-        The assembler will include the file as raw assembly source text at this point,
-        prog8 will not process this at all, with one exception: the labels.
-        The scopelabel argument will be used as a prefix to access the labels from the included source code,
-        otherwise you would risk symbol redefinitions or duplications.
-        If you know what you are doing you can leave it as an empty string to not have a scope prefix.
-        The compiler first looks for the file relative to the same directory as the module containing this statement is in,
-        if the file can't be found there it is searched relative to the current directory.
+    Level: block.
+    This directive can only be used inside a block.
+    The assembler will include the file as raw assembly source text at this point,
+    prog8 will not process this at all, with one exception: the labels.
+    The scopelabel argument will be used as a prefix to access the labels from the included source code,
+    otherwise you would risk symbol redefinitions or duplications.
+    If you know what you are doing you can leave it as an empty string to not have a scope prefix.
+    The compiler first looks for the file relative to the same directory as the module containing this statement is in,
+    if the file can't be found there it is searched relative to the current directory.
 
 .. data:: %breakpoint
 
@@ -267,6 +282,7 @@ type identifier  type                     storage size       example var declara
 ``word[]``       signed word array        depends on value   ``word[] myvar = [1, 2, 3, 4]``
 ``uword[]``      unsigned word array      depends on value   ``uword[] myvar = [1, 2, 3, 4]``
 ``float[]``      floating-point array     depends on value   ``float[] myvar = [1.1, 2.2, 3.3, 4.4]``
+``str[]``        array with string ptrs   2*x bytes + strs   ``str[] names = ["ally", "pete"]``
 ``str``          string (petscii)         varies             ``str myvar = "hello."``
                                                              implicitly terminated by a 0-byte
 ===============  =======================  =================  =========================================
@@ -289,7 +305,8 @@ of something with an operand starting with 1 or 0, you'll have to add a space in
 - When an integer value ranges from 256..65535 the compiler sees it as a ``uword``.  For -32768..32767 it's a ``word``.
 - When a hex number has 3 or 4 digits, for example ``$0004``, it is seen as a ``word`` otherwise as a ``byte``.
 - When a binary number has 9 to 16 digits, for example ``%1100110011``, it is seen as a ``word`` otherwise as a ``byte``.
-- You can force a byte value into a word value by adding the ``.w`` datatype suffix to the number: ``$2a.w`` is equivalent to ``$002a``.
+- If the number fits in a byte but you really require it as a word value, you'll have to explicitly cast it: ``60 as uword``
+  or you can use the full word hexadecimal notation ``$003c``.
 
 
 Data type conversion
@@ -383,7 +400,25 @@ After defining a struct you can use the name of the struct as a data type to dec
 
 Struct variables can be assigned a struct literal value (also in their declaration as initial value)::
 
-    Color rgb = {255, 100, 0}          ; curly braces instead of brackets
+    Color rgb = [255, 100, 0]          ; note that the value is an array
+
+
+String
+^^^^^^
+
+``"hello"``   is a string translated into the default character encoding (PETSCII)
+
+``@"hello"``  is a string translated into the alternate character encoding (Screencodes/pokes)
+
+There are several escape sequences available to put special characters into your string value:
+
+- ``\\`` - the backslash itself, has to be escaped because it is the escape symbol by itself
+- ``\n`` - newline character (move cursor down and to beginning of next line)
+- ``\r`` - carriage return character (more or less the same as newline if printing to the screen)
+- ``\"`` - quote character (otherwise it would terminate the string)
+- ``\'`` - apostrophe character (has to be escaped in character literals, is okay inside a string)
+- ``\uHHHH`` - a unicode codepoint \u0000 - \uffff (16-bit hexadecimal)
+- ``\xHH`` - 8-bit hex value that will be copied verbatim *without encoding*
 
 
 Operators
@@ -470,6 +505,8 @@ takes no parameters.  If the subroutine returns a value, usually you assign it t
 If you're not interested in the return value, prefix the function call with the ``void`` keyword.
 Otherwise the compiler will warn you about discarding the result of the call.
 
+Multiple return values
+^^^^^^^^^^^^^^^^^^^^^^
 Normal subroutines can only return zero or one return values.
 However, the special ``asmsub`` routines (implemented in assembly code) or ``romsub`` routines
 (referencing a routine in kernel ROM) can return more than one return value.
@@ -477,9 +514,16 @@ For example a status in the carry bit and a number in A, or a 16-bit value in A/
 It is not possible to process the results of a call to these kind of routines
 directly from the language, because only single value assignments are possible.
 You can still call the subroutine and not store the results.
-But if you want to do something with the values it returns, you'll have to write
-a small block of custom inline assembly that does the call and stores the values
-appropriately. Don't forget to save/restore the registers if required.
+
+**There is an exception:** if there's just one return value in a register, and one or more others that are returned
+as bits in the status register (such as the Carry bit), the compiler allows you to call the subroutine.
+It will then store the result value in a variable if required, and *keep the status register untouched
+after the call* so you can use a conditional branch statement for that.
+Note that this makes no sense inside an expression, so the compiler will still give an error for that.
+
+If there really are multiple return values (other than a combined 16 bit return value in 2 registers),
+you'll have to write a small block of custom inline assembly that does the call and stores the values
+appropriately. Don't forget to save/restore any registers that are modified.
 
 
 Subroutine definitions
@@ -514,18 +558,20 @@ and returning stuff in several registers as well. The ``clobbers`` clause is use
 what CPU registers are clobbered by the call instead of being unchanged or returning a meaningful result value.
 
 
-Subroutines that are implemented purely in assembly code and which have an assembly calling convention (i.e.
-the parameters are strictly passed via cpu registers), are defined like this::
+User subroutines in the program source code that are implemented purely in assembly and which have an assembly calling convention (i.e.
+the parameters are strictly passed via cpu registers), are defined with ``asmsub`` like this::
 
-    asmsub  FREADS32() clobbers(A,X,Y)  {
+    asmsub  clear_screenchars (ubyte char @ A) clobbers(Y)  {
         %asm {{
-            lda  $62
-            eor  #$ff
-            asl  a
-            lda  #0
-            ldx  #$a0
-            jmp  $bc4f
-        }}
+            ldy  #0
+    _loop   sta  c64.Screen,y
+            sta  c64.Screen+$0100,y
+            sta  c64.Screen+$0200,y
+            sta  c64.Screen+$02e8,y
+            iny
+            bne  _loop
+            rts
+            }}
     }
 
 the statement body of such a subroutine should consist of just an inline assembly block.

docs/source/targetsystem.rst

@@ -9,10 +9,16 @@ Prog8 targets the following hardware:
 - optional use of memory mapped I/O registers
 - optional use of system ROM routines
 
-The main target machine is the well-known Commodore-64, which is an example of this.
-Another (preliminary) supported target machine is the `CommanderX16 <https://www.commanderx16.com/>`_ .
+Currently there are two machines that are supported as compiler target (selectable via the ``-target`` compiler argument):
 
-This chapter explains the relevant system details of such machines.
+- 'c64': the well-known Commodore-64, premium support
+- 'cx16': the `CommanderX16 <https://www.commanderx16.com/>`_ a project from the 8-Bit Guy. Support for this is still experimental.
+
+This chapter explains the relevant system details of these machines.
+
+.. hint::
+    If you only use standard kernel and prog8 library routines,
+    it is possible to compile the *exact same program* for both machines (just change the compiler target flag)!
 
 
 Memory Model

docs/source/todo.rst

@@ -2,30 +2,28 @@
 TODO
 ====
 
-- optimize assignment codegeneration
-- get rid of all TODO's ;-)
+- see if we can group some errors together for instance the (now single) errors about unidentified symbols
+- Cx16 target: support full-screen 640x480 and 320x240 graphics? That requires our own custom graphics routines though to draw lines.
+- hoist all variable declarations up to the subroutine scope *before* even the constant folding takes place (to avoid undefined symbol errors when referring to a variable from another nested scope in the subroutine)
 - make it possible to use cpu opcodes such as 'nop' as variable names by prefixing all asm vars with something such as '_'
 - option to load the built-in library files from a directory instead of the embedded ones (for easier library development/debugging)
-- aliases for imported symbols for example perhaps '%alias print = c64scr.print' ?
-- investigate support for 8bitguy's Commander X16 platform https://www.commanderx16.com  and https://github.com/commanderx16/x16-docs
-- see if we can group some errors together for instance the (now single) errors about unidentified symbols
-
+- use VIC banking to move up the graphics bitmap memory location. Move it to $e000 under the kernal rom?
+- some support for recursive subroutines?
+    - via %option recursive?: allocate all params and local vars on estack, don't allow nested subroutines, can begin by first not allowing any local variables just fixing the parameters
+    - Or via a special recursive call operation that copies the current values of all local vars (including arguments) to the stack, replaces the arguments, jsr subroutine, and after returning copy the stack back to the local variables
+- get rid of all other TODO's in the code ;-)
 
 More optimizations
 ^^^^^^^^^^^^^^^^^^
 
 Add more compiler optimizations to the existing ones.
 
-- more targeted optimizations for assigment asm code, such as the following:
-- subroutine calling convention? like: 1 byte arg -> pass in A, 2 bytes -> pass in A+Y, return value likewise.
-- remove unreachable code after an exit(), return or goto
-- add a compiler option to not include variable initialization code (useful if the program is expected to run only once, such as a game)
-  the program will then rely solely on the values as they are in memory at the time of program startup.
-- Also some library routines and code patterns could perhaps be optimized further
-- can the parameter passing to subroutines be optimized to avoid copying?
+- further optimize assignment codegeneration, such as the following:
+- binexpr splitting (beware self-referencing expressions and asm code ballooning though)
+- detect var->var argument passing to subroutines and avoid the second variable and copying of the value
 - more optimizations on the language AST level
 - more optimizations on the final assembly source level
-- note: abandoned subroutine inlining because of problems referencing non-local stuff. Can't move everything around.
+- note: subroutine inlining is abandoned because of problems referencing non-local stuff. Can't move everything around.
 
 
 Eval stack redesign? (lot of work)

examples/arithmetic/aggregates.p8

@@ -1,5 +1,5 @@
-%import c64flt
-%import c64textio
+%import floats
+%import textio
 %zeropage basicsafe
 
 main {

examples/arithmetic/bitshift.p8

@@ -1,4 +1,4 @@
-%import c64textio
+%import textio
 %zeropage basicsafe
 
 

examples/arithmetic/builtins.p8

@@ -0,0 +1,836 @@
+%import textio
+%import floats
+%import syslib
+%import test_stack
+%zeropage basicsafe
+
+main {
+
+    sub start() {
+        rotations()
+        strings()
+        integers()
+        floatingpoint()
+
+        test_stack.test()
+    }
+
+    sub rotations() {
+        ubyte[] ubarr = [%11000111]
+        uword[] uwarr = [%1100111110101010]
+
+        repeat(10) {
+            txt.chrout('\n')
+        }
+
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        rol(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        rol(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        rol(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        rol(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        uwarr[0] = %1100111110101010
+        ror(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        ror(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        ror(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        ror(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        ror(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        clear_carry()
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        rol2(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        rol2(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        rol2(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        rol2(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        ror2(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        ror2(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        ror2(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        ror2(uwarr[0])
+        txt.print_uwbin(uwarr[0], true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        rol(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        rol(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        rol(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        rol(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        set_carry()
+        ror(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        ror(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        ror(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        ror(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        set_carry()
+        ror(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        rol2(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        rol2(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        rol2(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        rol2(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        ror2(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        ror2(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        ror2(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        ror2(ubarr[0])
+        txt.print_ubbin(ubarr[0], true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        &ubyte  membyte = $c000
+        uword addr = $c000
+
+        @(addr) = %10110101
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        set_carry()
+        rol(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        set_carry()
+        rol(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        set_carry()
+        rol(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        set_carry()
+        rol(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        @(addr) = %10110101
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        set_carry()
+        ror(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        set_carry()
+        ror(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        set_carry()
+        ror(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        set_carry()
+        ror(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        @(addr) = %10110101
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        rol2(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        rol2(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        rol2(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        rol2(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        @(addr) = %10110101
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        ror2(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        ror2(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        ror2(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        ror2(@(addr))
+        txt.print_ubbin(@(addr), true)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        test_stack.test()
+
+    }
+
+    sub strings() {
+        const uword ADDR = $8400
+        const uword ADDR2 = $8000
+
+        memset(ADDR2, 40*25, '*')
+        memset(ADDR2, 40, '1')
+        memset(ADDR2+24*40, 39, '2')
+        memsetw(ADDR2, 40*25/2, $3132)
+        memsetw(ADDR2, 20, $4142)
+        memsetw(ADDR2+24*40, 19, $4241)
+        memcopy(ADDR2, ADDR, 200)
+
+        str result = "?" *10
+        str s1 = "irmen"
+        str s2 = "hello"
+        str dots = "....."
+
+        ubyte ub
+        byte bb
+        ubyte zero=0
+
+        bb = strcmp(s1, s2)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        bb = strcmp(s2, s1)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        txt.print_ub(s1==s2)
+        txt.chrout('\n')
+        txt.print_ub(s1<s2)
+        txt.chrout('\n')
+        txt.print_ub(s1>s2)
+        txt.chrout('\n')
+        bb = zero+strcmp(s1,s2)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+        bb = zero+strcmp(s2,s1)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+
+        ub = strlen(s1)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+strlen(s1)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        leftstr(s1, result, 3)
+        txt.print(result)
+        txt.chrout('\n')
+        leftstr(s1, result, len(s1))
+        txt.print(result)
+        txt.chrout('\n')
+        txt.chrout('\n')
+
+        result = "x"*8
+        rightstr(s2, result, 3)
+        txt.print(result)
+        txt.chrout('\n')
+        rightstr(s2, result, len(s1))
+        txt.print(result)
+        txt.chrout('\n')
+
+        result = "y"*10
+        substr(s2, result, 1, 3)
+        txt.print(result)
+        txt.chrout('\n')
+
+        test_stack.test()
+
+    }
+
+    sub integers() {
+        ubyte[]  ubarr = [1,2,3,4,5,0,4,3,2,1, 255, 255, 255]
+        byte[]  barr = [1,2,3,4,5,-4,0,-3,2,1, -128, -128, -127]
+        uword[]  uwarr = [100,200,300,400,0,500,400,300,200,100]
+        word[] warr = [100,200,300,400,500,0,-400,-300,200,100,-99, -4096]
+
+        ubyte zero=0
+        ubyte ub
+        ubyte ub2
+        byte bb
+        uword uw
+        word ww
+
+        repeat(20) {
+            txt.chrout('\n')
+        }
+
+        ub = read_flags()
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+read_flags()*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+
+        ub = rnd()
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+rnd()*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        uw = rndw()
+        txt.print_uw(uw)
+        txt.chrout('\n')
+        uw = zero+rndw()*1+zero
+        txt.print_uw(uw)
+        txt.chrout('\n')
+
+
+        uw = 50000
+        ub = sqrt16(uw)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+sqrt16(uw)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        bb = -100
+        bb = sgn(bb)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        bb = -100
+        bb = zero+sgn(bb)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+
+        ub = 100
+        bb = sgn(ub)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        ub = 100
+        bb = zero+sgn(ub)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+
+        ww = -1000
+        bb = sgn(ww)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        bb = zero+sgn(ww)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+
+        uw = 1000
+        bb = sgn(uw)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        bb = zero+sgn(uw)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+
+        ub = 0
+        uw = sin16u(ub)
+        txt.print_uw(uw)
+        txt.chrout('\n')
+        uw = zero+sin16u(ub)*1+zero
+        txt.print_uw(uw)
+        txt.chrout('\n')
+
+        ub = 0
+        uw = cos16u(ub)
+        txt.print_uw(uw)
+        txt.chrout('\n')
+        uw = zero+cos16u(ub)*1+zero
+        txt.print_uw(uw)
+        txt.chrout('\n')
+
+        ub = 0
+        ww = sin16(ub)
+        txt.print_w(ww)
+        txt.chrout('\n')
+        ww = zero+sin16(ub)*1+zero
+        txt.print_w(ww)
+        txt.chrout('\n')
+
+        ub = 0
+        ww = cos16(ub)
+        txt.print_w(ww)
+        txt.chrout('\n')
+        uw = 0
+        ww = zero+cos16(ub)*1+zero
+        txt.print_w(ww)
+        txt.chrout('\n')
+
+        ub2 = 0
+        ub = sin8u(ub2)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+sin8u(ub2)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        ub2 = 0
+        ub = cos8u(ub2)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+cos8u(ub2)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        ub2 = 0
+        bb = sin8(ub2)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        bb = zero+sin8(ub2)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+
+        ub2 = 0
+        bb = cos8(ub2)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        bb = zero+cos8(ub2)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+
+        bb = -100
+        bb = abs(bb)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        bb = -100
+        bb = zero+abs(bb)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+
+        ww = -1000
+        ww = abs(ww)
+        txt.print_w(ww)
+        txt.chrout('\n')
+        ww = -1000
+        ww = zero+abs(ww)*1+zero
+        txt.print_w(ww)
+        txt.chrout('\n')
+
+        ub = min(ubarr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+min(ubarr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        bb = min(barr)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        bb = zero+min(barr)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+
+        uw = min(uwarr)
+        txt.print_uw(uw)
+        txt.chrout('\n')
+        uw = zero+min(uwarr)*1+zero
+        txt.print_uw(uw)
+        txt.chrout('\n')
+
+        ww = min(warr)
+        txt.print_w(ww)
+        txt.chrout('\n')
+        ww = zero+min(warr)*1+zero
+        txt.print_w(ww)
+        txt.chrout('\n')
+
+        ub = max(ubarr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+max(ubarr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        bb = max(barr)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        bb = zero+max(barr)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+
+        uw = max(uwarr)
+        txt.print_uw(uw)
+        txt.chrout('\n')
+        uw = zero+max(uwarr)*1+zero
+        txt.print_uw(uw)
+        txt.chrout('\n')
+
+        ww = max(warr)
+        txt.print_w(ww)
+        txt.chrout('\n')
+        ww = zero+max(warr)*1+zero
+        txt.print_w(ww)
+        txt.chrout('\n')
+
+        ub = any(ubarr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+any(ubarr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        ub = any(barr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+any(barr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        ub = any(uwarr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+any(uwarr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        ub = any(warr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+any(warr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        ub = all(ubarr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+all(ubarr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        ub = all(barr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+all(barr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        ub = all(uwarr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+all(uwarr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        ub = all(warr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+all(warr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+
+        uw = sum(ubarr)
+        txt.print_uw(uw)
+        txt.chrout('\n')
+        uw = zero+sum(ubarr)*1+zero
+        txt.print_uw(uw)
+        txt.chrout('\n')
+
+        ww = sum(barr)
+        txt.print_w(ww)
+        txt.chrout('\n')
+        ww = zero+sum(barr)*1+zero
+        txt.print_w(ww)
+        txt.chrout('\n')
+
+        uw = sum(uwarr)
+        txt.print_uw(uw)
+        txt.chrout('\n')
+        uw = zero+sum(uwarr)*1+zero
+        txt.print_uw(uw)
+        txt.chrout('\n')
+
+        ww = sum(warr)
+        txt.print_w(ww)
+        txt.chrout('\n')
+        ww = zero+sum(warr)*1+zero
+        txt.print_w(ww)
+        txt.chrout('\n')
+
+        sort(ubarr)
+        sort(barr)
+        sort(uwarr)
+        sort(warr)
+        reverse(ubarr)
+        reverse(barr)
+        reverse(uwarr)
+        reverse(warr)
+
+        test_stack.test()
+    }
+
+    sub floatingpoint() {
+        ubyte[]  barr = [1,2,3,4,5,0,4,3,2,1]
+        float[] flarr = [1.1, 2.2, 3.3, 0.0, -9.9, 5.5, 4.4]
+
+        ubyte zero=0
+        ubyte ub
+        byte bb
+        uword uw
+        float fl
+        float fzero=0.0
+
+        fl = -9.9
+        fl = abs(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = -9.9
+        fl = fzero+abs(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = 9.9
+        fl = atan(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = 9.9
+        fl = fzero+atan(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = -9.9
+        fl = ceil(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = -9.9
+        fl = fzero+ceil(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = -9.9
+        fl = cos(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = -9.9
+        fl = fzero+cos(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = -9.9
+        fl = sin(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = -9.9
+        fl = fzero+sin(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = 9.9
+        fl = tan(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = 9.9
+        fl = fzero+tan(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = 3.1415927
+        fl = deg(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = 3.1415927
+        fl = fzero+deg(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = 90
+        fl = rad(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = 90
+        fl = fzero+rad(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = -9.9
+        fl = floor(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = -9.9
+        fl = fzero+floor(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = 3.1415927
+        fl = ln(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = 3.1415927
+        fl = fzero+ln(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = 3.1415927
+        fl = log2(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = 3.1415927
+        fl = fzero+log2(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = -9.9
+        fl = round(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = -9.9
+        fl = fzero+round(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = -9.9
+        bb = sgn(fl)
+        txt.print_b(bb)
+        txt.chrout('\n')
+        fl = -9.9
+        bb = zero+sgn(fl)*1+zero
+        txt.print_b(bb)
+        txt.chrout('\n')
+
+        fl = 3.1415927
+        fl = sqrt(fl)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = 3.1415927
+        fl = fzero+sqrt(fl)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        fl = rndf()
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = fzero+rndf()*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        swap(fl, fzero)
+        swap(fzero, fl)
+
+        ub = any(flarr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+any(flarr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = all(flarr)
+        txt.print_ub(ub)
+        txt.chrout('\n')
+        ub = zero+all(flarr)*1+zero
+        txt.print_ub(ub)
+        txt.chrout('\n')
+
+        reverse(flarr)
+        for ub in 0 to len(flarr)-1 {
+            floats.print_f(flarr[ub])
+            txt.chrout(',')
+        }
+        txt.chrout('\n')
+        fl = max(flarr)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = fzero+max(flarr)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = min(flarr)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = fzero+min(flarr)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = sum(flarr)
+        floats.print_f(fl)
+        txt.chrout('\n')
+        fl = fzero+sum(flarr)*1.0+fzero
+        floats.print_f(fl)
+        txt.chrout('\n')
+
+        test_stack.test()
+    }
+}

examples/arithmetic/div.p8

@@ -1,9 +1,7 @@
-%import c64flt
-%import c64textio
+%import floats
+%import textio
 %zeropage basicsafe
 
-; TODO implement signed byte/word DIV asm generation, fix unsigned DIV asm generation (for in-place)
-
 main {
 
     sub start() {
@@ -11,17 +9,17 @@ main {
         div_ubyte(100, 6, 16)
         div_ubyte(255, 2, 127)
 
-        ;div_byte(0, 1, 0)      ; TODO implement
-        ;div_byte(100, -6, -16) ; TODO implement
-        ;div_byte(127, -2, -63) ; TODO implement
+        div_byte(0, 1, 0)
+        div_byte(100, -6, -16)
+        div_byte(127, -2, -63)
 
         div_uword(0,1,0)
         div_uword(40000,500,80)
         div_uword(43211,2,21605)
 
-        ;div_word(0,1,0)            ; TODO implement
-        ;div_word(-20000,500,-40)   ; TODO implement
-        ;div_word(-2222,2,-1111)    ; TODO implement
+        div_word(0,1,0)
+        div_word(-20000,500,-40)
+        div_word(-2222,2,-1111)
 
         div_float(0,1,0)
         div_float(999.9,111.0,9.008108108108107)
@@ -95,11 +93,11 @@ main {
             txt.print("err! ")
 
         txt.print("float ")
-        c64flt.print_f(a1)
+        floats.print_f(a1)
         txt.print(" / ")
-        c64flt.print_f(a2)
+        floats.print_f(a2)
         txt.print(" = ")
-        c64flt.print_f(r)
+        floats.print_f(r)
         c64.CHROUT('\n')
     }
 }

examples/arithmetic/minus.p8

@@ -1,5 +1,6 @@
-%import c64flt
-%import c64textio
+%import floats
+%import textio
+%import test_stack
 %zeropage basicsafe
 
 main {
@@ -31,6 +32,8 @@ main {
         minus_float(0,0,0)
         minus_float(2.5,1.5,1.0)
         minus_float(-1.5,3.5,-5.0)
+
+        test_stack.test()
     }
 
     sub minus_ubyte(ubyte a1, ubyte a2, ubyte c) {
@@ -97,15 +100,16 @@ main {
         float r = a1-a2
         if abs(r-c)<0.00001
             txt.print(" ok  ")
-        else
+        else {
             txt.print("err! ")
+        }
 
-        txt.print("float ")
-        c64flt.print_f(a1)
+        txt.print("  float ")
+        floats.print_f(a1)
         txt.print(" - ")
-        c64flt.print_f(a2)
+        floats.print_f(a2)
         txt.print(" = ")
-        c64flt.print_f(r)
+        floats.print_f(r)
         c64.CHROUT('\n')
     }
 }

examples/arithmetic/mult.p8

@@ -1,5 +1,5 @@
-%import c64flt
-%import c64textio
+%import floats
+%import textio
 %zeropage basicsafe
 
 main {
@@ -95,11 +95,11 @@ main {
             txt.print("err! ")
 
         txt.print("float ")
-        c64flt.print_f(a1)
+        floats.print_f(a1)
         txt.print(" * ")
-        c64flt.print_f(a2)
+        floats.print_f(a2)
         txt.print(" = ")
-        c64flt.print_f(r)
+        floats.print_f(r)
         c64.CHROUT('\n')
     }
 }

examples/arithmetic/plus.p8

@@ -1,5 +1,5 @@
-%import c64flt
-%import c64textio
+%import floats
+%import textio
 %zeropage basicsafe
 
 main {
@@ -99,11 +99,11 @@ main {
             txt.print("err! ")
 
         txt.print("float ")
-        c64flt.print_f(a1)
+        floats.print_f(a1)
         txt.print(" + ")
-        c64flt.print_f(a2)
+        floats.print_f(a2)
         txt.print(" = ")
-        c64flt.print_f(r)
+        floats.print_f(r)
         c64.CHROUT('\n')
     }
 }

examples/arithmetic/postincrdecr.p8

@@ -1,5 +1,5 @@
-%import c64flt
-%import c64textio
+%import floats
+%import textio
 %zeropage basicsafe
 
 main {
@@ -132,9 +132,9 @@ main {
         else
             txt.print("err! ")
         txt.print(" float ")
-        c64flt.print_f(value)
+        floats.print_f(value)
         c64.CHROUT(',')
-        c64flt.print_f(expected)
+        floats.print_f(expected)
         c64.CHROUT('\n')
     }
 }

examples/arithmetic/remainder.p8

@@ -1,4 +1,4 @@
-%import c64textio
+%import textio
 %zeropage basicsafe
 
 main {