workaround for the joystick_get() irq problem

Move code used for all Commodore systems to new package

.idea/compiler.xml

@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="CompilerConfiguration">
+    <option name="BUILD_PROCESS_HEAP_SIZE" value="1200" />
+  </component>
+</project>

README.md

@@ -23,13 +23,15 @@ https://prog8.readthedocs.io/
 What does Prog8 provide?
 ------------------------
 
-- big reduction of source code length over raw assembly
+- reduction of source code length over raw assembly
+- fast execution speed due to compilation to native assembly code. It's possible to write certain raster interrupt 'demoscene' effects purely in Prog8.
 - 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
-- no stack frame allocations because parameters and local variables are automatically allocated statically
-- constant folding in expressions and other high-level program optimizations
+- automatic static variable allocations, automatic string and array variables and string sharing
+- subroutines with input parameters and result values
+- high-level program optimizations
+- small program boilerplate/compilersupport overhead
+- Programs can be run multiple times without reloading because of automatic variable (re)initializations.
 - conditional branches
 - floating point operations  (requires the C64 Basic ROM routines for this)
 - 'when' statement to provide a concise jump table alternative to if/elseif chains
@@ -37,9 +39,9 @@ What does Prog8 provide?
 - many built-in functions such as ``sin``, ``cos``, ``rnd``, ``abs``, ``min``, ``max``, ``sqrt``, ``msb``, ``rol``, ``ror``, ``swap``, ``sort`` and ``reverse``
 - various powerful built-in libraries to do I/O, number conversions, graphics and more  
 - 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
 - supports the sixteen 'virtual' 16-bit registers R0 .. R15 from the Commander X16, and provides them also on the C64.
+- encode strings and characters into petscii or screencodes as desired (C64/Cx16)
 
 *Rapid edit-compile-run-debug cycle:*
 

compiler/build.gradle

@@ -1,8 +1,7 @@
 plugins {
     id 'java'
     id 'application'
-    id "org.jetbrains.kotlin.jvm" version "1.4.30"
-    id 'org.jetbrains.dokka' version "0.9.18"
+    id "org.jetbrains.kotlin.jvm" version "1.4.32"
     id 'com.github.johnrengelman.shadow' version '6.1.0'
 }
 
@@ -21,7 +20,7 @@ dependencies {
     implementation project(':compilerAst')
     implementation "org.jetbrains.kotlin:kotlin-stdlib-jdk8"
     // implementation "org.jetbrains.kotlin:kotlin-reflect"
-    implementation 'org.jetbrains.kotlinx:kotlinx-cli:0.3.1'
+    implementation 'org.jetbrains.kotlinx:kotlinx-cli:0.3.2'
     // implementation 'net.razorvine:ksim65:1.8'
     // implementation "com.github.hypfvieh:dbus-java:3.2.4"
 
@@ -97,11 +96,6 @@ test {
 }
 
 
-dokka {
-    outputFormat = 'html'
-    outputDirectory = "$buildDir/kdoc"
-}
-
 task wrapper(type: Wrapper) {
     gradleVersion = '6.7'
 }

compiler/compiler.iml

@@ -1,5 +1,10 @@
 <?xml version="1.0" encoding="UTF-8"?>
 <module type="JAVA_MODULE" version="4">
+  <component name="FacetManager">
+    <facet type="Python" name="Python">
+      <configuration sdkName="Python 3.9" />
+    </facet>
+  </component>
   <component name="NewModuleRootManager" inherit-compiler-output="true">
     <exclude-output />
     <content url="file://$MODULE_DIR$">
@@ -14,5 +19,6 @@
     <orderEntry type="library" name="unittest-libs" level="project" />
     <orderEntry type="library" name="kotlinx-cli-jvm" level="project" />
     <orderEntry type="module" module-name="compilerAst" />
+    <orderEntry type="library" name="Python 3.9 interpreter library" level="application" />
   </component>
 </module>

compiler/res/prog8lib/c64/floats.asm

@@ -428,7 +428,9 @@ var_fac1_greater_f	.proc
 		cmp  #1
 		beq  +
 		lda  #0
-+		rts
+		rts
++		lda  #1
+		rts
 		.pend
 
 var_fac1_greatereq_f	.proc

compiler/res/prog8lib/c64/floats.p8

@@ -83,7 +83,7 @@ romsub $bc58 = ABS()                                        ; fac1 = ABS(fac1)
 romsub $bf71 = SQR() clobbers(A,X,Y)                        ; fac1 = SQRT(fac1)
 romsub $bf74 = SQRA() clobbers(A,X,Y)                       ; fac1 = SQRT(fac2)
 romsub $bfed = EXP() clobbers(A,X,Y)                        ; fac1 = EXP(fac1)  (e ** fac1)
-romsub $bfb4 = NEGOP() clobbers(A)                          ; switch the sign of fac1
+romsub $bfb4 = NEGOP() clobbers(A)                          ; switch the sign of fac1 (fac1 = -fac1)
 romsub $e097 = RND() clobbers(A,X,Y)                        ; fac1 = RND(fac1) float random number generator
 romsub $e264 = COS() clobbers(A,X,Y)                        ; fac1 = COS(fac1)
 romsub $e26b = SIN() clobbers(A,X,Y)                        ; fac1 = SIN(fac1)

compiler/res/prog8lib/c64/graphics.p8

@@ -2,7 +2,7 @@
 %import textio
 
 ; bitmap pixel graphics module for the C64
-; only black/white monchrome 320x200 for now
+; only black/white monochrome 320x200 for now
 ; assumes bitmap screen memory is $2000-$3fff
 
 graphics {
@@ -34,36 +34,33 @@ graphics {
     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 there are some slight errors at the first/last pixels in certain slopes...??
         if y1>y2 {
             ; make sure dy is always positive to have only 4 instead of 8 special cases
             swap(x1, x2)
             swap(y1, y2)
         }
-        word @zp dx = x2-x1 as word
-        word @zp dy = y2-y1
+        word @zp dx = (x2 as word)-x1
+        word @zp dy = (y2 as word)-y1
 
         if dx==0 {
-            vertical_line(x1, y1, abs(dy)+1 as ubyte)
+            vertical_line(x1, y1, abs(dy) as ubyte +1)
             return
         }
         if dy==0 {
             if x1>x2
                 x1=x2
-            horizontal_line(x1, y1, abs(dx)+1 as uword)
+            horizontal_line(x1, y1, abs(dx) as uword +1)
             return
         }
 
-        ; TODO rewrite the rest in optimized assembly
-
         word @zp d = 0
         ubyte positive_ix = true
         if dx < 0 {
             dx = -dx
             positive_ix = false
         }
-        dx *= 2
-        dy *= 2
+        word @zp dx2 = dx*2
+        word @zp dy2 = dy*2
         internal_plotx = x1
 
         if dx >= dy {
@@ -73,10 +70,10 @@ graphics {
                     if internal_plotx==x2
                         return
                     internal_plotx++
-                    d += dy
+                    d += dy2
                     if d > dx {
                         y1++
-                        d -= dx
+                        d -= dx2
                     }
                 }
             } else {
@@ -85,10 +82,10 @@ graphics {
                     if internal_plotx==x2
                         return
                     internal_plotx--
-                    d += dy
+                    d += dy2
                     if d > dx {
                         y1++
-                        d -= dx
+                        d -= dx2
                     }
                 }
             }
@@ -100,10 +97,10 @@ graphics {
                     if y1 == y2
                         return
                     y1++
-                    d += dx
+                    d += dx2
                     if d > dy {
                         internal_plotx++
-                        d -= dy
+                        d -= dy2
                     }
                 }
             } else {
@@ -112,10 +109,10 @@ graphics {
                     if y1 == y2
                         return
                     y1++
-                    d += dx
+                    d += dx2
                     if d > dy {
                         internal_plotx--
-                        d -= dy
+                        d -= dy2
                     }
                 }
             }

compiler/res/prog8lib/c64/syslib.p8

@@ -202,7 +202,7 @@ romsub $FFAE = UNLSN() clobbers(A)                              ; command serial
 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 $FFBA = SETLFS(ubyte logical @ A, ubyte device @ X, ubyte secondary @ 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
@@ -211,10 +211,10 @@ romsub $FFC9 = CHKOUT(ubyte logical @ X) clobbers(A,X)          ; (via 800 ($320
 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 $FFD5 = LOAD(ubyte verify @ A, uword address @ XY) -> ubyte @Pc, ubyte @ A, uword @ XY     ; (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 $FFDE = RDTIM() -> ubyte @ A, ubyte @ X, ubyte @ Y       ; read the software clock (A=lo,X=mid,Y=high)
 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
@@ -246,7 +246,7 @@ asmsub STOP2() -> ubyte @A  {
 }
 
 asmsub RDTIM16() -> uword @AY {
-    ; --  like RDTIM() but only returning the lower 16 bits for convenience
+    ; --  like RDTIM() but only returning the lower 16 bits in AY for convenience
     %asm {{
         stx  P8ZP_SCRATCH_REG
         jsr  c64.RDTIM
@@ -478,7 +478,7 @@ sys {
 
 
     asmsub  reset_system()  {
-        ; Soft-reset the system back to Basic prompt.
+        ; Soft-reset the system back to initial power-on Basic prompt.
         %asm {{
             sei
             lda  #14
@@ -489,6 +489,7 @@ sys {
 
     sub wait(uword jiffies) {
         ; --- wait approximately the given number of jiffies (1/60th seconds)
+        ;     note: the system irq handler has to be active for this to work as it depends on the system jiffy clock
         repeat jiffies {
             ubyte jiff = lsb(c64.RDTIM16())
             while jiff==lsb(c64.RDTIM16()) {
@@ -497,6 +498,29 @@ sys {
         }
     }
 
+    asmsub waitvsync() clobbers(A) {
+        ; --- busy wait till the next vsync has occurred (approximately), without depending on custom irq handling.
+        ;     note: a more accurate way to wait for vsync is to set up a vsync irq handler instead.
+        %asm {{
+-           lda  c64.RASTER
+            beq  -
+-           lda  c64.RASTER
+            bne  -
+            bit  c64.SCROLY
+            bmi  -
+            rts
+        }}
+    }
+
+    inline asmsub waitrastborder() {
+        ; --- busy wait till the raster position has reached the bottom screen border (approximately)
+        ;     note: a more accurate way to do this is by using a raster irq handler instead.
+        %asm {{
+-           bit  c64.SCROLY
+            bpl  -
+        }}
+    }
+
     asmsub memcopy(uword source @R0, uword target @R1, uword count @AY) clobbers(A,X,Y) {
         %asm {{
             ldx  cx16.r0

compiler/res/prog8lib/conv.p8

@@ -7,239 +7,213 @@ conv {
 
 ; ----- number conversions to decimal strings ----
 
-asmsub  ubyte2decimal  (ubyte value @A) -> ubyte @Y, ubyte @A, ubyte @X  {
-	; ---- A to decimal string in Y/A/X  (100s in Y, 10s in A, 1s in X)
+    str  string_out = "????????????????"       ; result buffer for the string conversion routines
+
+asmsub  str_ub0  (ubyte value @ A) clobbers(A,Y)  {
+	; ---- convert the ubyte in A in decimal string form, with left padding 0s (3 positions total)
 	%asm {{
-		ldy  #uword2decimal.ASCII_0_OFFSET
-		bne  uword2decimal.hex_try200
+            phx
+            jsr  conv.ubyte2decimal
+            sty  string_out
+            sta  string_out+1
+            stx  string_out+2
+            lda  #0
+            sta  string_out+3
+            plx
             rts
 	}}
 }
 
-asmsub  uword2decimal  (uword value @AY) -> ubyte @Y, ubyte @A, ubyte @X  {
-	;  ---- convert 16 bit uword in A/Y to decimal
-	;  output in uword2decimal.decTenThousands, decThousands, decHundreds, decTens, decOnes
-	;  (these are terminated by a zero byte so they can be easily printed)
-	;  also returns Y = 100's, A = 10's, X = 1's
-
+asmsub  str_ub  (ubyte value @ A) clobbers(A,Y)  {
+	; ---- convert the ubyte in A in decimal string form, without left padding 0s
 	%asm {{
-
-;Convert 16 bit Hex to Decimal (0-65535) Rev 2
-;By Omegamatrix    Further optimizations by tepples
-; routine from http://forums.nesdev.com/viewtopic.php?f=2&t=11341&start=15
-
-;HexToDec99
-; start in A
-; end with A = 10's, decOnes (also in X)
-
-;HexToDec255
-; start in A
-; end with Y = 100's, A = 10's, decOnes (also in X)
-
-;HexToDec999
-; start with A = high byte, Y = low byte
-; end with Y = 100's, A = 10's, decOnes (also in X)
-; requires 1 extra temp register on top of decOnes, could combine
-; these two if HexToDec65535 was eliminated...
-
-;HexToDec65535
-; start with A/Y (low/high) as 16 bit value
-; end with decTenThousand, decThousand, Y = 100's, A = 10's, decOnes (also in X)
-; (irmen: I store Y and A in decHundreds and decTens too, so all of it can be easily printed)
-
-
-ASCII_0_OFFSET 	= $30
-temp       	    = P8ZP_SCRATCH_B1	; byte in zeropage
-hexHigh      	= P8ZP_SCRATCH_W1	; byte in zeropage
-hexLow       	= P8ZP_SCRATCH_W1+1	; byte in zeropage
-
-
-HexToDec65535; SUBROUTINE
-    sty    hexHigh               ;3  @9
-    sta    hexLow                ;3  @12
+		phx
+		ldy  #0
+		sty  P8ZP_SCRATCH_B1
+		jsr  conv.ubyte2decimal
+_output_byte_digits
+                ; hundreds?
+		cpy  #'0'
+		beq  +
+		pha
 		tya
-    tax                          ;2  @14
-    lsr    a                     ;2  @16
-    lsr    a                     ;2  @18   integer divide 1024 (result 0-63)
-
-    cpx    #$A7                  ;2  @20   account for overflow of multiplying 24 from 43,000 ($A7F8) onward,
-    adc    #1                    ;2  @22   we can just round it to $A700, and the divide by 1024 is fine...
-
-    ;at this point we have a number 1-65 that we have to times by 24,
-    ;add to original sum, and Mod 1024 to get a remainder 0-999
-
-
-    sta    temp                  ;3  @25
-    asl    a                     ;2  @27
-    adc    temp                  ;3  @30  x3
-    tay                          ;2  @32
-    lsr    a                     ;2  @34
-    lsr    a                     ;2  @36
-    lsr    a                     ;2  @38
-    lsr    a                     ;2  @40
-    lsr    a                     ;2  @42
-    tax                          ;2  @44
-    tya                          ;2  @46
-    asl    a                     ;2  @48
-    asl    a                     ;2  @50
-    asl    a                     ;2  @52
-    clc                          ;2  @54
-    adc    hexLow                ;3  @57
-    sta    hexLow                ;3  @60
-    txa                          ;2  @62
-    adc    hexHigh               ;3  @65
-    sta    hexHigh               ;3  @68
-    ror    a                     ;2  @70
-    lsr    a                     ;2  @72
-    tay                          ;2  @74    integer divide 1,000 (result 0-65)
-
-    lsr    a                     ;2  @76    split the 1,000 and 10,000 digit
-    tax                          ;2  @78
-    lda    ShiftedBcdTab,x       ;4  @82
-    tax                          ;2  @84
-    rol    a                     ;2  @86
-    and    #$0F                  ;2  @88
-    ora    #ASCII_0_OFFSET
-    sta    decThousands          ;3  @91
-    txa                          ;2  @93
-    lsr    a                     ;2  @95
-    lsr    a                     ;2  @97
-    lsr    a                     ;2  @99
-    ora    #ASCII_0_OFFSET
-    sta    decTenThousands       ;3  @102
-
-    lda    hexLow                ;3  @105
-    cpy    temp                  ;3  @108
-    bmi    _doSubtract           ;2³ @110/111
-    beq    _useZero               ;2³ @112/113
-    adc    #23 + 24              ;2  @114
-_doSubtract
-    sbc    #23                   ;2  @116
-    sta    hexLow                ;3  @119
-_useZero
-    lda    hexHigh               ;3  @122
-    sbc    #0                    ;2  @124
-
-Start100s
-    and    #$03                  ;2  @126
-    tax                          ;2  @128   0,1,2,3
-    cmp    #2                    ;2  @130
-    rol    a                     ;2  @132   0,2,5,7
-    ora    #ASCII_0_OFFSET
-    tay                          ;2  @134   Y = Hundreds digit
-
-    lda    hexLow                ;3  @137
-    adc    Mod100Tab,x           ;4  @141    adding remainder of 256, 512, and 256+512 (all mod 100)
-    bcs    hex_doSub200             ;2³ @143/144
-
-hex_try200
-    cmp    #200                  ;2  @145
-    bcc    hex_try100               ;2³ @147/148
-hex_doSub200
-    iny                          ;2  @149
-    iny                          ;2  @151
-    sbc    #200                  ;2  @153
-hex_try100
-    cmp    #100                  ;2  @155
-    bcc    HexToDec99            ;2³ @157/158
-    iny                          ;2  @159
-    sbc    #100                  ;2  @161
-
-HexToDec99; SUBROUTINE
-    lsr    a                     ;2  @163
-    tax                          ;2  @165
-    lda    ShiftedBcdTab,x       ;4  @169
-    tax                          ;2  @171
-    rol    a                     ;2  @173
-    and    #$0F                  ;2  @175
-    ora    #ASCII_0_OFFSET
-    sta    decOnes               ;3  @178
-    txa                          ;2  @180
-    lsr    a                     ;2  @182
-    lsr    a                     ;2  @184
-    lsr    a                     ;2  @186
-    ora    #ASCII_0_OFFSET
-
-    ; irmen: load X with ones, and store Y and A too, for easy printing afterwards
-    sty  decHundreds
-    sta  decTens
-    ldx  decOnes
-    rts                          ;6  @192   Y=hundreds, A = tens digit, X=ones digit
-
-
-HexToDec999; SUBROUTINE
-    sty    hexLow                ;3  @9
-    jmp    Start100s             ;3  @12
-
-Mod100Tab
-    .byte 0,56,12,56+12
-
-ShiftedBcdTab
-    .byte $00,$01,$02,$03,$04,$08,$09,$0A,$0B,$0C
-    .byte $10,$11,$12,$13,$14,$18,$19,$1A,$1B,$1C
-    .byte $20,$21,$22,$23,$24,$28,$29,$2A,$2B,$2C
-    .byte $30,$31,$32,$33,$34,$38,$39,$3A,$3B,$3C
-    .byte $40,$41,$42,$43,$44,$48,$49,$4A,$4B,$4C
-
-decTenThousands   	.byte  0
-decThousands    	.byte  0
-decHundreds		.byte  0
-decTens			.byte  0
-decOnes   		.byte  0
-			.byte  0		; zero-terminate the decimal output string
-
+		ldy  P8ZP_SCRATCH_B1
+		sta  string_out,y
+		pla
+		inc  P8ZP_SCRATCH_B1
+		; tens?
++		ldy  P8ZP_SCRATCH_B1
+                cmp  #'0'
+		beq  +
+		sta  string_out,y
+		iny
++               ; ones.
+                txa
+                sta  string_out,y
+                iny
+                lda  #0
+                sta  string_out,y
+                plx
+                rts
 	}}
 }
 
-asmsub  byte2decimal  (byte value @A) -> ubyte @Y, ubyte @A, ubyte @X  {
-	; ---- A (signed byte) to decimal string in Y/A/X  (100s in Y, 10s in A, 1s in X)
-	;      note: if the number is negative, you have to deal with the '-' yourself!
+asmsub  str_b  (byte value @ A) clobbers(A,Y)  {
+	; ---- convert the byte in A in decimal string form, without left padding 0s
 	%asm {{
+            phx
+            ldy  #0
+            sty  P8ZP_SCRATCH_B1
             cmp  #0
             bpl  +
-		eor  #255
-		clc
-		adc  #1
-+		jmp  ubyte2decimal
-	}}
-}
-
-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
             pha
-		and  #$0f
-		tax
-		ldy  _hex_digits,x
+            lda  #'-'
+            sta  string_out
+            inc  P8ZP_SCRATCH_B1
             pla
-		lsr  a
-		lsr  a
-		lsr  a
-		lsr  a
-		tax
-		lda  _hex_digits,x
-		ldx  P8ZP_SCRATCH_REG
-		rts
-
-_hex_digits	.text "0123456789abcdef"	; can probably be reused for other stuff as well
++	    jsr  conv.byte2decimal
+            bra  str_ub._output_byte_digits
 	}}
 }
 
-asmsub  uword2hex  (uword value @AY) clobbers(A,Y)  {
-	; ---- convert 16 bit uword in A/Y into 4-character hexadecimal string 'uword2hex.output' (0-terminated)
+asmsub  str_ubhex  (ubyte value @ A) clobbers(A,Y)  {
+	; ---- convert the ubyte in A in hex string form
+	%asm {{
+            jsr  conv.ubyte2hex
+            sta  string_out
+            sty  string_out+1
+            lda  #0
+            sta  string_out+2
+            rts
+	}}
+}
+
+asmsub  str_ubbin  (ubyte value @ A) clobbers(A,Y)  {
+	; ---- convert the ubyte in A in binary string form
+	%asm {{
+	    sta  P8ZP_SCRATCH_B1
+	    ldy  #0
+	    sty  string_out+8
+	    ldy  #7
+-	    lsr  P8ZP_SCRATCH_B1
+            bcc  +
+            lda  #'1'
+            bne  _digit
++           lda  #'0'
+_digit      sta  string_out,y
+            dey
+	    bpl  -
+	    rts
+	}}
+}
+
+asmsub  str_uwbin  (uword value @ AY) clobbers(A,Y)  {
+	; ---- convert the uword in A/Y in binary string form
 	%asm {{
 	    sta  P8ZP_SCRATCH_REG
 	    tya
-		jsr  ubyte2hex
-		sta  output
-		sty  output+1
-		lda  P8ZP_SCRATCH_REG
-		jsr  ubyte2hex
-		sta  output+2
-		sty  output+3
+	    jsr  str_ubbin
+	    ldy  #0
+	    sty  string_out+16
+	    ldy  #7
+-	    lsr  P8ZP_SCRATCH_REG
+            bcc  +
+            lda  #'1'
+            bne  _digit
++           lda  #'0'
+_digit      sta  string_out+8,y
+            dey
+	    bpl  -
 	    rts
-output		.text  "0000", $00      ; 0-terminated output buffer (to make printing easier)
+	}}
+}
+
+asmsub  str_uwhex  (uword value @ AY) clobbers(A,Y)  {
+	; ---- convert the uword in A/Y in hexadecimal string form (4 digits)
+	%asm {{
+            pha
+            tya
+            jsr  conv.ubyte2hex
+            sta  string_out
+            sty  string_out+1
+            pla
+            jsr  conv.ubyte2hex
+            sta  string_out+2
+            sty  string_out+3
+            lda  #0
+            sta  string_out+4
+            rts
+	}}
+}
+
+asmsub  str_uw0  (uword value @ AY) clobbers(A,Y)  {
+	; ---- convert the uword in A/Y in decimal string form, with left padding 0s (5 positions total)
+	%asm {{
+	    phx
+	    jsr  conv.uword2decimal
+	    ldy  #0
+-           lda  conv.uword2decimal.decTenThousands,y
+            sta  string_out,y
+            beq  +
+            iny
+            bne  -
++           plx
+	    rts
+	}}
+}
+
+asmsub  str_uw  (uword value @ AY) clobbers(A,Y)  {
+	; ---- convert the uword in A/Y in decimal string form, without left padding 0s
+	%asm {{
+	    phx
+	    jsr  conv.uword2decimal
+	    ldx  #0
+_output_digits
+	    ldy  #0
+-           lda  conv.uword2decimal.decTenThousands,y
+            beq  _allzero
+            cmp  #'0'
+            bne  _gotdigit
+            iny
+            bne  -
+_gotdigit   sta  string_out,x
+            inx
+            iny
+            lda  conv.uword2decimal.decTenThousands,y
+            bne  _gotdigit
+_end        lda  #0
+            sta  string_out,x
+            plx
+            rts
+
+_allzero    lda  #'0'
+            sta  string_out,x
+            inx
+            bne  _end
+	}}
+}
+
+asmsub  str_w  (word value @ AY) clobbers(A,Y)  {
+	; ---- convert the (signed) word in A/Y in decimal string form, without left padding 0's
+	%asm {{
+	    cpy  #0
+	    bpl  str_uw
+	    phx
+	    pha
+	    lda  #'-'
+	    sta  string_out
+            tya
+            eor  #255
+            tay
+            pla
+            eor  #255
+            clc
+            adc  #1
+            bcc  +
+            iny
++	    jsr  conv.uword2decimal
+	    ldx  #1
+	    bne  str_uw._output_digits
 	}}
 }
 
@@ -520,4 +494,243 @@ _stop
 	}}
 }
 
+
+; ----- low level number conversions to decimal strings ----
+
+asmsub  ubyte2decimal  (ubyte value @A) -> ubyte @Y, ubyte @A, ubyte @X  {
+	; ---- A to decimal string in Y/A/X  (100s in Y, 10s in A, 1s in X)
+	%asm {{
+		ldy  #uword2decimal.ASCII_0_OFFSET
+		bne  uword2decimal.hex_try200
+		rts
+	}}
+}
+
+asmsub  uword2decimal  (uword value @AY) -> ubyte @Y, ubyte @A, ubyte @X  {
+	;  ---- convert 16 bit uword in A/Y to decimal
+	;  output in uword2decimal.decTenThousands, decThousands, decHundreds, decTens, decOnes
+	;  (these are terminated by a zero byte so they can be easily printed)
+	;  also returns Y = 100's, A = 10's, X = 1's
+
+	%asm {{
+
+;Convert 16 bit Hex to Decimal (0-65535) Rev 2
+;By Omegamatrix    Further optimizations by tepples
+; routine from http://forums.nesdev.com/viewtopic.php?f=2&t=11341&start=15
+
+;HexToDec99
+; start in A
+; end with A = 10's, decOnes (also in X)
+
+;HexToDec255
+; start in A
+; end with Y = 100's, A = 10's, decOnes (also in X)
+
+;HexToDec999
+; start with A = high byte, Y = low byte
+; end with Y = 100's, A = 10's, decOnes (also in X)
+; requires 1 extra temp register on top of decOnes, could combine
+; these two if HexToDec65535 was eliminated...
+
+;HexToDec65535
+; start with A/Y (low/high) as 16 bit value
+; end with decTenThousand, decThousand, Y = 100's, A = 10's, decOnes (also in X)
+; (irmen: I store Y and A in decHundreds and decTens too, so all of it can be easily printed)
+
+
+ASCII_0_OFFSET 	= $30
+temp       	    = P8ZP_SCRATCH_B1	; byte in zeropage
+hexHigh      	= P8ZP_SCRATCH_W1	; byte in zeropage
+hexLow       	= P8ZP_SCRATCH_W1+1	; byte in zeropage
+
+
+HexToDec65535; SUBROUTINE
+    sty    hexHigh               ;3  @9
+    sta    hexLow                ;3  @12
+    tya
+    tax                          ;2  @14
+    lsr    a                     ;2  @16
+    lsr    a                     ;2  @18   integer divide 1024 (result 0-63)
+
+    cpx    #$A7                  ;2  @20   account for overflow of multiplying 24 from 43,000 ($A7F8) onward,
+    adc    #1                    ;2  @22   we can just round it to $A700, and the divide by 1024 is fine...
+
+    ;at this point we have a number 1-65 that we have to times by 24,
+    ;add to original sum, and Mod 1024 to get a remainder 0-999
+
+
+    sta    temp                  ;3  @25
+    asl    a                     ;2  @27
+    adc    temp                  ;3  @30  x3
+    tay                          ;2  @32
+    lsr    a                     ;2  @34
+    lsr    a                     ;2  @36
+    lsr    a                     ;2  @38
+    lsr    a                     ;2  @40
+    lsr    a                     ;2  @42
+    tax                          ;2  @44
+    tya                          ;2  @46
+    asl    a                     ;2  @48
+    asl    a                     ;2  @50
+    asl    a                     ;2  @52
+    clc                          ;2  @54
+    adc    hexLow                ;3  @57
+    sta    hexLow                ;3  @60
+    txa                          ;2  @62
+    adc    hexHigh               ;3  @65
+    sta    hexHigh               ;3  @68
+    ror    a                     ;2  @70
+    lsr    a                     ;2  @72
+    tay                          ;2  @74    integer divide 1,000 (result 0-65)
+
+    lsr    a                     ;2  @76    split the 1,000 and 10,000 digit
+    tax                          ;2  @78
+    lda    ShiftedBcdTab,x       ;4  @82
+    tax                          ;2  @84
+    rol    a                     ;2  @86
+    and    #$0F                  ;2  @88
+    ora    #ASCII_0_OFFSET
+    sta    decThousands          ;3  @91
+    txa                          ;2  @93
+    lsr    a                     ;2  @95
+    lsr    a                     ;2  @97
+    lsr    a                     ;2  @99
+    ora    #ASCII_0_OFFSET
+    sta    decTenThousands       ;3  @102
+
+    lda    hexLow                ;3  @105
+    cpy    temp                  ;3  @108
+    bmi    _doSubtract           ;2³ @110/111
+    beq    _useZero               ;2³ @112/113
+    adc    #23 + 24              ;2  @114
+_doSubtract
+    sbc    #23                   ;2  @116
+    sta    hexLow                ;3  @119
+_useZero
+    lda    hexHigh               ;3  @122
+    sbc    #0                    ;2  @124
+
+Start100s
+    and    #$03                  ;2  @126
+    tax                          ;2  @128   0,1,2,3
+    cmp    #2                    ;2  @130
+    rol    a                     ;2  @132   0,2,5,7
+    ora    #ASCII_0_OFFSET
+    tay                          ;2  @134   Y = Hundreds digit
+
+    lda    hexLow                ;3  @137
+    adc    Mod100Tab,x           ;4  @141    adding remainder of 256, 512, and 256+512 (all mod 100)
+    bcs    hex_doSub200             ;2³ @143/144
+
+hex_try200
+    cmp    #200                  ;2  @145
+    bcc    hex_try100               ;2³ @147/148
+hex_doSub200
+    iny                          ;2  @149
+    iny                          ;2  @151
+    sbc    #200                  ;2  @153
+hex_try100
+    cmp    #100                  ;2  @155
+    bcc    HexToDec99            ;2³ @157/158
+    iny                          ;2  @159
+    sbc    #100                  ;2  @161
+
+HexToDec99; SUBROUTINE
+    lsr    a                     ;2  @163
+    tax                          ;2  @165
+    lda    ShiftedBcdTab,x       ;4  @169
+    tax                          ;2  @171
+    rol    a                     ;2  @173
+    and    #$0F                  ;2  @175
+    ora    #ASCII_0_OFFSET
+    sta    decOnes               ;3  @178
+    txa                          ;2  @180
+    lsr    a                     ;2  @182
+    lsr    a                     ;2  @184
+    lsr    a                     ;2  @186
+    ora    #ASCII_0_OFFSET
+
+    ; irmen: load X with ones, and store Y and A too, for easy printing afterwards
+    sty  decHundreds
+    sta  decTens
+    ldx  decOnes
+    rts                          ;6  @192   Y=hundreds, A = tens digit, X=ones digit
+
+
+HexToDec999; SUBROUTINE
+    sty    hexLow                ;3  @9
+    jmp    Start100s             ;3  @12
+
+Mod100Tab
+    .byte 0,56,12,56+12
+
+ShiftedBcdTab
+    .byte $00,$01,$02,$03,$04,$08,$09,$0A,$0B,$0C
+    .byte $10,$11,$12,$13,$14,$18,$19,$1A,$1B,$1C
+    .byte $20,$21,$22,$23,$24,$28,$29,$2A,$2B,$2C
+    .byte $30,$31,$32,$33,$34,$38,$39,$3A,$3B,$3C
+    .byte $40,$41,$42,$43,$44,$48,$49,$4A,$4B,$4C
+
+decTenThousands   	.byte  0
+decThousands    	.byte  0
+decHundreds		.byte  0
+decTens			.byte  0
+decOnes   		.byte  0
+			.byte  0		; zero-terminate the decimal output string
+
+    }}
+}
+
+asmsub  byte2decimal  (byte value @A) -> ubyte @Y, ubyte @A, ubyte @X  {
+	; ---- A (signed byte) to decimal string in Y/A/X  (100s in Y, 10s in A, 1s in X)
+	;      note: if the number is negative, you have to deal with the '-' yourself!
+	%asm {{
+		cmp  #0
+		bpl  +
+		eor  #255
+		clc
+		adc  #1
++		jmp  ubyte2decimal
+	}}
+}
+
+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
+		pha
+		and  #$0f
+		tax
+		ldy  _hex_digits,x
+		pla
+		lsr  a
+		lsr  a
+		lsr  a
+		lsr  a
+		tax
+		lda  _hex_digits,x
+		ldx  P8ZP_SCRATCH_REG
+		rts
+
+_hex_digits	.text "0123456789abcdef"	; can probably be reused for other stuff as well
+	}}
+}
+
+asmsub  uword2hex  (uword value @AY) clobbers(A,Y)  {
+	; ---- convert 16 bit uword in A/Y into 4-character hexadecimal string 'uword2hex.output' (0-terminated)
+	%asm {{
+		sta  P8ZP_SCRATCH_REG
+		tya
+		jsr  ubyte2hex
+		sta  output
+		sty  output+1
+		lda  P8ZP_SCRATCH_REG
+		jsr  ubyte2hex
+		sta  output+2
+		sty  output+3
+		rts
+output		.text  "0000", $00      ; 0-terminated output buffer (to make printing easier)
+	}}
+}
+
 }

compiler/res/prog8lib/cx16/floats.p8

@@ -8,7 +8,9 @@
 %option enable_floats
 
 floats {
-	; ---- this block contains C-64 floating point related functions ----
+	; ---- this block contains C-64 compatible floating point related functions ----
+	;      the addresses are from cx16 V39 emulator and roms! they won't work on older versions.
+
 
         const float  PI     = 3.141592653589793
         const float  TWOPI  = 6.283185307179586
@@ -43,46 +45,44 @@ 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 $fe30 = CONUPK(uword mflpt @ AY) clobbers(A,X,Y)     ; load mflpt value from memory  in A/Y into fac2
+romsub $fe33 = MUL10() clobbers(A,X,Y)                      ; fac1 *= 10
+romsub $fe36 = DIV10() clobbers(A,X,Y)                      ; fac1 /= 10 , CAUTION: result is always positive!
+romsub $fe39 = FDIV(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = mflpt in A/Y / fac1  (remainder in fac2)
+romsub $fe3c = 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
-; note: there is no FPWR() on the Cx16
-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)
+romsub $fe42 = MOVFM(uword mflpt @ AY) clobbers(A,X,Y)      ; load mflpt value from memory  in A/Y into fac1
+romsub $fe45 = MOVMF(uword mflpt @ XY) clobbers(A,X,Y)      ; store fac1 to memory  X/Y as 5-byte mflpt
+romsub $fe48 = MOVFA() clobbers(A,X)                        ; copy fac2 to fac1
+romsub $fe4b = MOVAF() clobbers(A,X)                        ; copy fac1 to fac2  (rounded)
+romsub $fe4e = MOVEF() clobbers(A,X)                        ; copy fac1 to fac2
+romsub $fe54 = SIGN() clobbers(X,Y) -> ubyte @ A            ; SIGN(fac1) to A, $ff, $0, $1 for negative, zero, positive
+romsub $fe57 = SGN() clobbers(A,X,Y)                        ; fac1 = SGN(fac1), result of SIGN (-1, 0 or 1)
+romsub $fe5a = FREADSA(byte value @ A) clobbers(A,X,Y)      ; 8 bit signed A -> float in fac1
+romsub $fe66 = ABS() clobbers(A,X,Y)                        ; fac1 = ABS(fac1)
+romsub $fe69 = 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 $fe72 = INT() clobbers(A,X,Y)                        ; INT() truncates, use FADDH first to round instead of trunc
+romsub $fe78 = FINLOG(byte value @A) clobbers (A, X, Y)           ; fac1 += signed byte in A
+romsub $fe7b = FOUT() clobbers(X) -> uword @ AY             ; fac1 -> string, address returned in AY
+romsub $fe81 = SQR() clobbers(A,X,Y)                        ; fac1 = SQRT(fac1)
+romsub $fe84 = FPWRT() clobbers(A,X,Y)                      ; fac1 = fac2 ** fac1
+romsub $fe8a = NEGOP() clobbers(A)                          ; switch the sign of fac1 (fac1 = -fac1)
+romsub $fe8d = EXP() clobbers(A,X,Y)                        ; fac1 = EXP(fac1)  (e ** fac1)
+romsub $fe96 = RND() clobbers(A,X,Y)                        ; fac1 = RND(fac1) float random number generator
+romsub $fe99 = COS() clobbers(A,X,Y)                        ; fac1 = COS(fac1)
+romsub $fe9c = SIN() clobbers(A,X,Y)                        ; fac1 = SIN(fac1)
+romsub $fe9f = TAN() clobbers(A,X,Y)                        ; fac1 = TAN(fac1)
+romsub $fea2 = 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
+        sta  _tmp
         sty  P8ZP_SCRATCH_B1
         tya
-        ldy  P8ZP_SCRATCH_W2
+        ldy  _tmp
         jsr  GIVAYF                 ; load it as signed... correct afterwards
         lda  P8ZP_SCRATCH_B1
 	    bpl  +
@@ -91,6 +91,7 @@ asmsub  GIVUAYFAY  (uword value @ AY) clobbers(A,X,Y)  {
 	    jsr  FADD
 +	    plx
         rts
+_tmp        .byte 0
 _flt65536    .byte 145,0,0,0,0       ; 65536.0
 	}}
 }
@@ -128,6 +129,14 @@ asmsub  GETADRAY  () clobbers(X) -> uword @ AY  {
 	}}
 }
 
+asmsub  FREADUY (ubyte value @Y) {
+    ; -- 8 bit unsigned Y -> float in fac1
+    %asm {{
+        lda  #0
+        jmp  GIVAYF
+    }}
+}
+
 sub  print_f  (float value) {
 	; ---- prints the floating point value (without a newline).
 	%asm {{

compiler/res/prog8lib/cx16/gfx2.p8

@@ -36,7 +36,7 @@ gfx2 {
     sub screen_mode(ubyte mode) {
         when mode {
             1 -> {
-                ; lores monchrome
+                ; lores monochrome
                 cx16.VERA_DC_VIDEO = (cx16.VERA_DC_VIDEO & %11001111) | %00100000      ; enable only layer 1
                 cx16.VERA_DC_HSCALE = 64
                 cx16.VERA_DC_VSCALE = 64
@@ -404,7 +404,7 @@ _done
                 ; TODO also mostly usable for lores 4c?
                 void addr_mul_24_for_highres_4c(y, x)      ; 24 bits result is in r0 and r1L (highest byte)
 
-                ; TODO optimize the loop in pure assembly
+                ; TODO optimize this vertical line loop in pure assembly
                 color &= 3
                 color <<= gfx2.plot.shift4c[lsb(x) & 3]
                 ubyte mask = gfx2.plot.mask4c[lsb(x) & 3]
@@ -433,48 +433,46 @@ _done
     sub line(uword @zp x1, uword @zp y1, uword @zp x2, uword @zp y2, ubyte color) {
         ; Bresenham algorithm.
         ; This code special-cases various quadrant loops to allow simple ++ and -- operations.
-        ; TODO there are some slight errors at the first/last pixels in certain slopes...
         if y1>y2 {
             ; make sure dy is always positive to have only 4 instead of 8 special cases
             swap(x1, x2)
             swap(y1, y2)
         }
-        word @zp dx = x2-x1 as word
-        word @zp dy = y2-y1 as word
+        word @zp dx = (x2 as word)-x1
+        word @zp dy = (y2 as word)-y1
 
         if dx==0 {
-            vertical_line(x1, y1, abs(dy)+1 as uword, color)
+            vertical_line(x1, y1, abs(dy) as uword +1, color)
             return
         }
         if dy==0 {
             if x1>x2
                 x1=x2
-            horizontal_line(x1, y1, abs(dx)+1 as uword, color)
+            horizontal_line(x1, y1, abs(dx) as uword +1, color)
             return
         }
 
-        ; TODO rewrite the rest in optimized assembly (or reuse GRAPH_draw_line if we can get the FB replacement vector layer working)
         word @zp d = 0
-        ubyte positive_ix = true
+        cx16.r13 = true      ; 'positive_ix'
         if dx < 0 {
             dx = -dx
-            positive_ix = false
+            cx16.r13 = false
         }
-        dx *= 2
-        dy *= 2
+        word @zp dx2 = dx*2
+        word @zp dy2 = dy*2
         cx16.r14 = x1       ; internal plot X
 
         if dx >= dy {
-            if positive_ix {
+            if cx16.r13 {
                 repeat {
                     plot(cx16.r14, y1, color)
                     if cx16.r14==x2
                         return
                     cx16.r14++
-                    d += dy
+                    d += dy2
                     if d > dx {
                         y1++
-                        d -= dx
+                        d -= dx2
                     }
                 }
             } else {
@@ -483,25 +481,25 @@ _done
                     if cx16.r14==x2
                         return
                     cx16.r14--
-                    d += dy
+                    d += dy2
                     if d > dx {
                         y1++
-                        d -= dx
+                        d -= dx2
                     }
                 }
             }
         }
         else {
-            if positive_ix {
+            if cx16.r13 {
                 repeat {
                     plot(cx16.r14, y1, color)
                     if y1 == y2
                         return
                     y1++
-                    d += dx
+                    d += dx2
                     if d > dy {
                         cx16.r14++
-                        d -= dy
+                        d -= dy2
                     }
                 }
             } else {
@@ -510,10 +508,10 @@ _done
                     if y1 == y2
                         return
                     y1++
-                    d += dx
+                    d += dx2
                     if d > dy {
                         cx16.r14--
-                        d -= dy
+                        d -= dy2
                     }
                 }
             }

compiler/res/prog8lib/cx16/graphics.p8

@@ -4,7 +4,7 @@
 
 ; Bitmap pixel graphics module for the CommanderX16
 ; wraps the graphics functions that are in ROM.
-; only black/white monchrome 320x200 for now. (i.e. truncated at the bottom)
+; only black/white monochrome 320x200 for now. (i.e. truncated at the bottom)
 ; For full-screen 640x480 or 320x240 graphics, use the "gfx2" module instead. (but that is Cx16-specific)
 ; Note: there is no color palette manipulation here, you have to do that yourself or use the "palette" module.
 

compiler/res/prog8lib/cx16/syslib.p8

@@ -24,7 +24,7 @@ romsub $FF8D = VECTOR(uword userptr @ XY, ubyte dir @ Pc) clobbers(A,Y)     ; re
 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.   NOTE: as a Cx16 extension, also returns the number of RAM memory banks in register A !  See MEMTOP2
+romsub $FF99 = MEMTOP(uword address @ XY, ubyte dir @ Pc) -> uword @ XY     ; read/set top of memory  pointer.   NOTE: as a Cx16 extension, also returns the number of RAM memory banks in register A !  See cx16.numbanks()
 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
@@ -35,7 +35,7 @@ romsub $FFAE = UNLSN() clobbers(A)                              ; command serial
 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 $FFBA = SETLFS(ubyte logical @ A, ubyte device @ X, ubyte secondary @ 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
@@ -44,10 +44,10 @@ romsub $FFC9 = CHKOUT(ubyte logical @ X) clobbers(A,X)          ; (via 800 ($320
 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 $FFD5 = LOAD(ubyte verify @ A, uword address @ XY) -> ubyte @Pc, ubyte @ A, uword @ XY     ; (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 $FFDE = RDTIM() -> ubyte @ A, ubyte @ X, ubyte @ Y       ; read the software clock (A=lo,X=mid,Y=high)
 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
@@ -74,7 +74,7 @@ asmsub STOP2() -> ubyte @A  {
 }
 
 asmsub RDTIM16() -> uword @AY {
-    ; --  like RDTIM() but only returning the lower 16 bits for convenience
+    ; --  like RDTIM() but only returning the lower 16 bits in AY for convenience
     %asm {{
         phx
         jsr  c64.RDTIM
@@ -87,17 +87,6 @@ asmsub RDTIM16() -> uword @AY {
     }}
 }
 
-asmsub MEMTOP2() -> ubyte @A {
-    ; -- uses MEMTOP's cx16 extension to query the number of available RAM banks.
-    %asm {{
-        phx
-        sec
-        jsr  c64.MEMTOP
-        plx
-        rts
-    }}
-}
-
 }
 
 cx16 {
@@ -172,7 +161,7 @@ cx16 {
 
 ; I/O
 
-    const uword  via1   = $9f60                  ;VIA 6522 #1
+    const uword  via1   = $9f00                  ;VIA 6522 #1
     &ubyte  d1prb	= via1+0
     &ubyte  d1pra	= via1+1
     &ubyte  d1ddrb	= via1+2
@@ -190,7 +179,7 @@ cx16 {
     &ubyte  d1ier	= via1+14
     &ubyte  d1ora	= via1+15
 
-    const uword  via2   = $9f70                  ;VIA 6522 #2
+    const uword  via2   = $9f10                  ;VIA 6522 #2
     &ubyte  d2prb	= via2+0
     &ubyte  d2pra	= via2+1
     &ubyte  d2ddrb	= via2+2
@@ -208,6 +197,11 @@ cx16 {
     &ubyte  d2ier	= via2+14
     &ubyte  d2ora	= via2+15
 
+    &ubyte  ym2151adr	= $9f40
+    &ubyte  ym2151dat	= $9f41
+
+    const uword  extdev	= $9f60
+
 
 ; ---- Commander X-16 additions on top of C64 kernal routines ----
 ; spelling of the names is taken from the Commander X-16 rom sources
@@ -294,16 +288,25 @@ romsub $fecc = monitor()  clobbers(A,X,Y)
 inline asmsub rombank(ubyte rombank @A) {
     ; -- set the rom banks
     %asm {{
-        sta  $01            ; rom bank register (new)
-        sta  cx16.d1prb     ; rom bank register (old)
+        sta  $01            ; rom bank register (v39+, used to be cx16.d1prb $9f60 in v38)
     }}
 }
 
 inline asmsub rambank(ubyte rambank @A) {
     ; -- set the ram bank
     %asm {{
-        sta  $00            ; ram bank register (new)
-        sta  cx16.d1pra     ; ram bank register (old)
+        sta  $00            ; ram bank register (v39+, used to be cx16.d1pra $9f61 in v38)
+    }}
+}
+
+asmsub numbanks() -> ubyte @A {
+    ; -- uses MEMTOP's cx16 extension to query the number of available RAM banks. (each is 8 Kb)
+    %asm {{
+        phx
+        sec
+        jsr  c64.MEMTOP
+        plx
+        rts
     }}
 }
 
@@ -419,6 +422,57 @@ asmsub vpoke_xor(ubyte bank @A, uword address @R0, ubyte value @Y) clobbers (A)
     }}
 }
 
+asmsub vload(str name @R0, ubyte device @Y, ubyte bank @A, uword address @R1) -> ubyte @A {
+    ; -- like the basic command VLOAD "filename",device,bank,address
+    ;    loads a file into video memory in the given bank:address, returns success in A
+    ;    !! NOTE !! the V38 ROMs contain a bug in the LOAD code that makes the load address not work correctly,
+    ;               it works fine when loading from local filesystem
+    %asm {{
+        ; -- load a file into video ram
+        phx
+        pha
+        tya
+        tax
+        lda  #1
+        ldy  #0
+        jsr  c64.SETLFS
+        lda  cx16.r0
+        ldy  cx16.r0+1
+        jsr  prog8_lib.strlen
+        tya
+        ldx  cx16.r0
+        ldy  cx16.r0+1
+        jsr  c64.SETNAM
+        pla
+        clc
+        adc  #2
+        ldx  cx16.r1
+        ldy  cx16.r1+1
+        stz  P8ZP_SCRATCH_B1
+        jsr  c64.LOAD
+        bcs  +
+        inc  P8ZP_SCRATCH_B1
++       jsr  c64.CLRCHN
+        lda  #1
+        jsr  c64.CLOSE
+        plx
+        lda  P8ZP_SCRATCH_B1
+        rts
+    }}
+}
+
+inline asmsub joystick_get2(ubyte joynr @A) clobbers(Y) -> uword @AX  {
+    ; convenience routine to get the joystick state without requiring inline assembly that deals with the multiple return values.
+    ; Also disables interrupts to avoid the IRQ race condition mentioned here: https://github.com/commanderx16/x16-rom/issues/203
+    ; TODO once that issue is resolved, this routine can be redefined as:  romsub $ff56 = joystick_get2(ubyte joynr @A) clobbers(Y) -> uword @AX
+    %asm {{
+        sei
+        jsr  cx16.joystick_get
+        cli
+    }}
+}
+
+
 sub FB_set_pixels_from_buf(uword buffer, uword count) {
     %asm {{
             ; -- This is replacement code for the normal FB_set_pixels subroutine in ROM
@@ -460,11 +514,9 @@ asmsub  init_system()  {
     %asm {{
         sei
         cld
-        ;stz  $00
-        ;stz  $01
-        ;stz  d1prb      ; select rom bank 0 (enable kernal)
         lda  #$80
         sta  VERA_CTRL
+        stz  $01        ; select rom bank 0 (enable kernal)
         jsr  c64.IOINIT
         jsr  c64.RESTOR
         jsr  c64.CINT
@@ -664,11 +716,10 @@ sys {
 
 
     asmsub reset_system() {
-        ; Soft-reset the system back to Basic prompt.
+        ; Soft-reset the system back to initial power-on Basic prompt.
         %asm {{
             sei
-            stz  $01            ; bank the kernal in (new rom bank register)
-            stz  cx16.d1prb     ; bank the kernal in (old rom bank register)
+            stz  $01            ; bank the kernal in
             jmp  (cx16.RESET_VEC)
         }}
     }
@@ -683,6 +734,21 @@ sys {
         }
     }
 
+    asmsub waitvsync() clobbers(A, X, Y) {
+        ; --- busy wait till the next vsync has occurred (approximately), without depending on custom irq handling.
+        ;     note: system vsync irq handler has to be active for this routine to work.
+        ;     note 2: a more accurate way to wait for vsync is to set up a vsync irq handler instead.
+        %asm {{
+            jsr  c64.RDTIM
+            sta  _mod + 1
+            inc  _mod + 1
+_loop       jsr  c64.RDTIM
+_mod        cmp  #255       ; modified
+            bne  _loop
+            rts
+        }}
+    }
+
     inline asmsub memcopy(uword source @R0, uword target @R1, uword count @AY) clobbers(A,X,Y) {
         %asm {{
             sta  cx16.r2

compiler/res/prog8lib/cx16/textio.p8

@@ -420,7 +420,7 @@ _print_byte_digits
 		jsr  c64.CHROUT
 		pla
 		jsr  c64.CHROUT
-		jmp  _ones
+		bra  _ones
 +       pla
         cmp  #'0'
         beq  _ones
@@ -443,7 +443,7 @@ asmsub  print_b  (byte value @ A) clobbers(A,Y)  {
 		jsr  c64.CHROUT
 +		pla
 		jsr  conv.byte2decimal
-		jmp  print_ub._print_byte_digits
+		bra  print_ub._print_byte_digits
 	}}
 }
 
@@ -494,7 +494,7 @@ asmsub  print_uwbin  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
 		jsr  print_ubbin
 		pla
 		clc
-		jmp  print_ubbin
+		bra  print_ubbin
 	}}
 }
 
@@ -507,7 +507,7 @@ asmsub  print_uwhex  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
 		jsr  print_ubhex
 		pla
 		clc
-		jmp  print_ubhex
+		bra  print_ubhex
 	}}
 }
 
@@ -570,7 +570,7 @@ asmsub  print_w  (word value @ AY) clobbers(A,Y)  {
 		adc  #1
 		bcc  +
 		iny
-+		jmp  print_uw
++		bra  print_uw
 	}}
 }
 

compiler/res/prog8lib/diskio.p8

@@ -365,6 +365,7 @@ io_error:
         c64.SETNAM(string.length(filenameptr), filenameptr)
         c64.SETLFS(1, drivenumber, 0)
         uword end_address = address + size
+        first_byte = 0      ; result var reuse
 
         %asm {{
             lda  address
@@ -381,7 +382,6 @@ io_error:
             plp
         }}
 
-        first_byte = 0      ; result var reuse
         if_cc
             first_byte = c64.READST()==0
 

compiler/res/prog8lib/math.asm

@@ -244,8 +244,8 @@ randseed	.proc
 		.pend
 
 
-randbyte	.proc
-	; -- 8-bit pseudo random number generator into A
+fast_randbyte	.proc
+	; -- fast but bad 8-bit pseudo random number generator into A
 		lda  _seed
 		beq  _eor
 		asl  a
@@ -263,6 +263,10 @@ _seed		.byte  $3a
 
 		.pend
 
+randbyte        .proc
+	; -- 8 bit pseudo random number generator into A (by just reusing randword)
+		jmp  randword
+		.pend
 
 randword	.proc
 	; -- 16 bit pseudo random number generator into AY
@@ -1537,3 +1541,71 @@ _negative	lsr  a
 		rts
 		.pend
 
+
+square          .proc
+; -- calculate square root of signed word in AY, result in AY
+; routine by Lee Davsion, source: http://6502.org/source/integers/square.htm
+; using this routine is about twice as fast as doing a regular multiplication.
+;
+; Calculates the 16 bit unsigned integer square of the signed 16 bit integer in
+; Numberl/Numberh.  The result is always in the range 0 to 65025 and is held in
+; Squarel/Squareh
+;
+; The maximum input range is only +/-255 and no checking is done to ensure that
+; this is so.
+;
+; This routine is useful if you are trying to draw circles as for any circle
+;
+; x^2+y^2=r^2 where x and y are the co-ordinates of any point on the circle and
+; r is the circle radius
+
+numberl = P8ZP_SCRATCH_W1       ; number to square low byte
+numberh = P8ZP_SCRATCH_W1+1     ; number to square high byte
+squarel = P8ZP_SCRATCH_W2       ; square low byte
+squareh = P8ZP_SCRATCH_W2+1     ; square high byte
+tempsq = P8ZP_SCRATCH_B1        ; temp byte for intermediate result
+
+	sta  numberl
+	sty  numberh
+	stx  P8ZP_SCRATCH_REG
+
+        lda     #$00        ; clear a
+        sta     squarel     ; clear square low byte
+                            ; (no need to clear the high byte, it gets shifted out)
+        lda	numberl     ; get number low byte
+	ldx	numberh     ; get number high  byte
+	bpl	_nonneg      ; if +ve don't negate it
+                            ; else do a two's complement
+	eor	#$ff        ; invert
+        sec	            ; +1
+	adc	#$00        ; and add it
+
+_nonneg:
+	sta	tempsq      ; save abs(number)
+	ldx	#$08        ; set bit count
+
+_nextr2bit:
+	asl	squarel     ; low byte *2
+	rol	squareh     ; high byte *2+carry from low
+	asl	a           ; shift number byte
+	bcc	_nosqadd     ; don't do add if c = 0
+	tay                 ; save a
+	clc                 ; clear carry for add
+	lda	tempsq      ; get number
+	adc	squarel     ; add number^2 low byte
+	sta	squarel     ; save number^2 low byte
+	lda	#$00        ; clear a
+	adc	squareh     ; add number^2 high byte
+	sta	squareh     ; save number^2 high byte
+	tya                 ; get a back
+
+_nosqadd:
+	dex                 ; decrement bit count
+	bne	_nextr2bit   ; go do next bit
+
+	lda  squarel
+	ldy  squareh
+	ldx  P8ZP_SCRATCH_REG
+	rts
+
+		.pend

compiler/res/prog8lib/prog8_funcs.asm

@@ -387,6 +387,14 @@ func_sqrt16_into_A	.proc
 		rts
 		.pend
 
+func_fastrnd8_stack	.proc
+	; -- put a random ubyte on the estack (using fast but bad RNG)
+		jsr  math.fast_randbyte
+		sta  P8ESTACK_LO,x
+		dex
+		rts
+		.pend
+
 func_rnd_stack	.proc
 	; -- put a random ubyte on the estack
 		jsr  math.randbyte
@@ -432,6 +440,7 @@ func_min_ub_stack	.proc
 func_min_b_into_A	.proc
 		; -- min(barray) -> A.  (array in P8ZP_SCRATCH_W1, num elements in A)
 		tay
+		dey
 		lda  #127
 		sta  P8ZP_SCRATCH_B1
 -		lda  (P8ZP_SCRATCH_W1),y
@@ -548,6 +557,7 @@ func_min_w_stack	.proc
 func_max_ub_into_A	.proc
 		; -- max(ubarray) -> A  (array in P8ZP_SCRATCH_W1, num elements in A)
 		tay
+		dey
 		lda  #0
 		sta  P8ZP_SCRATCH_B1
 -		lda  (P8ZP_SCRATCH_W1),y

compiler/res/prog8lib/prog8_lib.asm

@@ -1072,3 +1072,14 @@ sign_extend_AY_byte	.proc
 		rts
 		.pend
 
+strlen          .proc
+        ; -- returns the number of bytes in the string in AY, in Y.
+		sta  P8ZP_SCRATCH_W1
+		sty  P8ZP_SCRATCH_W1+1
+		ldy  #0
+-		lda  (P8ZP_SCRATCH_W1),y
+		beq  +
+		iny
+		bne  -
++		rts
+		.pend

compiler/res/version.txt

@@ -1 +1 @@
-6.2
+6.5-BETA

compiler/src/prog8/compiler/BeforeAsmGenerationAstChanger.kt

@@ -1,6 +1,7 @@
 package prog8.compiler
 
 import prog8.ast.IFunctionCall
+import prog8.ast.INameScope
 import prog8.ast.Node
 import prog8.ast.Program
 import prog8.ast.base.*
@@ -8,27 +9,28 @@ import prog8.ast.expressions.*
 import prog8.ast.statements.*
 import prog8.ast.walk.AstWalker
 import prog8.ast.walk.IAstModification
+import prog8.ast.walk.IAstVisitor
 import prog8.compiler.target.ICompilationTarget
 
 
 internal class BeforeAsmGenerationAstChanger(val program: Program, val errors: IErrorReporter, private val compTarget: ICompilationTarget) : AstWalker() {
 
-    private val noModifications = emptyList<IAstModification>()
-
     override fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> {
         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
+            // A numeric vardecl without an initial value is initialized with zero,
+            // unless there's already an assignment below, that initializes the value.
+            // This allows you to restart the program and have the same starting values of the variables
             if(decl.allowInitializeWithZero)
             {
                 val nextAssign = decl.definingScope().nextSibling(decl) as? Assignment
-                if (nextAssign != null && nextAssign.target.isSameAs(IdentifierReference(listOf(decl.name), Position.DUMMY)))
+                if (nextAssign != null && nextAssign.target isSameAs IdentifierReference(listOf(decl.name), Position.DUMMY))
                     decl.value = null
-                else
+                else {
                     decl.value = decl.zeroElementValue()
                 }
             }
+        }
         return noModifications
     }
 
@@ -67,9 +69,11 @@ internal class BeforeAsmGenerationAstChanger(val program: Program, val errors: I
     }
 
     private val subroutineVariables = mutableListOf<Pair<String, VarDecl>>()
+    private val addedIfConditionVars = mutableSetOf<Pair<Subroutine, String>>()
 
     override fun before(subroutine: Subroutine, parent: Node): Iterable<IAstModification> {
         subroutineVariables.clear()
+        addedIfConditionVars.clear()
         return noModifications
     }
 
@@ -80,17 +84,22 @@ internal class BeforeAsmGenerationAstChanger(val program: Program, val errors: I
         val sub = scope.definingSubroutine()
         if (sub != null) {
             // 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 replacements = mutableListOf<IAstModification>()
+            val movements = mutableListOf<IAstModification.InsertFirst>()
+
+            for(decl in decls) {
+                if(decl.value!=null && decl.datatype in NumericDatatypes) {
+                    val target = AssignTarget(IdentifierReference(listOf(decl.name), decl.position), null, null, decl.position)
+                    val assign = Assignment(target, decl.value!!, decl.position)
+                    replacements.add(IAstModification.ReplaceNode(decl, assign, scope))
+                    decl.value = null
+                    decl.allowInitializeWithZero = false
+                } else {
+                    replacements.add(IAstModification.Remove(decl, scope))
                 }
-            val moveVardeclsUp = decls.map { IAstModification.InsertFirst(it, sub) }
-            return replaceVardecls + moveVardeclsUp
+                movements.add(IAstModification.InsertFirst(decl, sub))
+            }
+            return replacements + movements
         }
         return noModifications
     }
@@ -154,16 +163,6 @@ internal class BeforeAsmGenerationAstChanger(val program: Program, val errors: I
         // 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) as? Subroutine
-                if (sub != null && sub.isAsmSubroutine) {
-                    return listOf(IAstModification.ReplaceNode(typecast, typecast.expression, parent))
-                }
-            }
-        }
-
         if(sourceDt in PassByReferenceDatatypes) {
             if(typecast.type==DataType.UWORD) {
                 if(typecast.expression is IdentifierReference) {
@@ -194,9 +193,53 @@ internal class BeforeAsmGenerationAstChanger(val program: Program, val errors: I
             val booleanExpr = BinaryExpression(ifStatement.condition, "!=", NumericLiteralValue.optimalInteger(0, ifStatement.condition.position), ifStatement.condition.position)
             return listOf(IAstModification.ReplaceNode(ifStatement.condition, booleanExpr, ifStatement))
         }
+
+        if((binExpr.left as? NumericLiteralValue)?.number==0)
+            throw CompilerException("if 0==X should have been swapped to if X==0")
+
+        // split the conditional expression into separate variables if the operand(s) is not simple.
+        // DISABLED FOR NOW AS IT GENEREATES LARGER CODE IN THE SIMPLE CASES LIKE    IF X {...}  or  IF NOT X {...}
+//        val modifications = mutableListOf<IAstModification>()
+//        if(!binExpr.left.isSimple) {
+//            val sub = binExpr.definingSubroutine()!!
+//            val (variable, isNew, assignment) = addIfOperandVar(sub, "left", binExpr.left)
+//            if(isNew)
+//                modifications.add(IAstModification.InsertFirst(variable, sub))
+//            modifications.add(IAstModification.InsertBefore(ifStatement, assignment, parent as INameScope))
+//            modifications.add(IAstModification.ReplaceNode(binExpr.left, IdentifierReference(listOf(variable.name), binExpr.position), binExpr))
+//            addedIfConditionVars.add(Pair(sub, variable.name))
+//        }
+//        if(!binExpr.right.isSimple) {
+//            val sub = binExpr.definingSubroutine()!!
+//            val (variable, isNew, assignment) = addIfOperandVar(sub, "right", binExpr.right)
+//            if(isNew)
+//                modifications.add(IAstModification.InsertFirst(variable, sub))
+//            modifications.add(IAstModification.InsertBefore(ifStatement, assignment, parent as INameScope))
+//            modifications.add(IAstModification.ReplaceNode(binExpr.right, IdentifierReference(listOf(variable.name), binExpr.position), binExpr))
+//            addedIfConditionVars.add(Pair(sub, variable.name))
+//        }
+//        return modifications
         return noModifications
     }
 
+//    private fun addIfOperandVar(sub: Subroutine, side: String, operand: Expression): Triple<VarDecl, Boolean, Assignment> {
+//        val dt = operand.inferType(program).typeOrElse(DataType.STRUCT)
+//        val varname = "prog8_ifvar_${side}_${dt.name.toLowerCase()}"
+//        val tgt = AssignTarget(IdentifierReference(listOf(varname), operand.position), null, null, operand.position)
+//        val assign = Assignment(tgt, operand, operand.position)
+//        if(Pair(sub, varname) in addedIfConditionVars) {
+//            val vardecl = VarDecl(VarDeclType.VAR, dt, ZeropageWish.DONTCARE, null, varname, null, null, false, true, operand.position)
+//            return Triple(vardecl, false, assign)
+//        }
+//        val existing = sub.statements.firstOrNull { it is VarDecl && it.name == varname} as VarDecl?
+//        return if (existing == null) {
+//            val vardecl = VarDecl(VarDeclType.VAR, dt, ZeropageWish.DONTCARE, null, varname, null, null, false, true, operand.position)
+//            Triple(vardecl, true, assign)
+//        } else {
+//            Triple(existing, false, assign)
+//        }
+//    }
+
     override fun after(untilLoop: UntilLoop, parent: Node): Iterable<IAstModification> {
         val binExpr = untilLoop.condition as? BinaryExpression
         if(binExpr==null || binExpr.operator !in comparisonOperators) {
@@ -216,4 +259,93 @@ internal class BeforeAsmGenerationAstChanger(val program: Program, val errors: I
         }
         return noModifications
     }
+
+    override fun after(functionCallStatement: FunctionCallStatement, parent: Node): Iterable<IAstModification> {
+        if(functionCallStatement.target.nameInSource==listOf("cmp")) {
+            // if the datatype of the arguments of cmp() are different, cast the byte one to word.
+            val arg1 = functionCallStatement.args[0]
+            val arg2 = functionCallStatement.args[1]
+            val dt1 = arg1.inferType(program).typeOrElse(DataType.STRUCT)
+            val dt2 = arg2.inferType(program).typeOrElse(DataType.STRUCT)
+            if(dt1 in ByteDatatypes) {
+                if(dt2 in ByteDatatypes)
+                    return noModifications
+                val cast1 = TypecastExpression(arg1, if(dt1==DataType.UBYTE) DataType.UWORD else DataType.WORD, true, functionCallStatement.position)
+                return listOf(IAstModification.ReplaceNode(arg1, cast1, functionCallStatement))
+            } else {
+                if(dt2 in WordDatatypes)
+                    return noModifications
+                val cast2 = TypecastExpression(arg2, if(dt2==DataType.UBYTE) DataType.UWORD else DataType.WORD, true, functionCallStatement.position)
+                return listOf(IAstModification.ReplaceNode(arg2, cast2, functionCallStatement))
+            }
+        }
+        return noModifications
+    }
+
+    override fun after(arrayIndexedExpression: ArrayIndexedExpression, parent: Node): Iterable<IAstModification> {
+
+        val containingStatement = getContainingStatement(arrayIndexedExpression)
+        if(getComplexArrayIndexedExpressions(containingStatement).size > 1) {
+            errors.err("it's not possible to use more than one complex array indexing expression in a single statement; break it up via a temporary variable for instance", containingStatement.position)
+            return noModifications
+        }
+
+
+        val index = arrayIndexedExpression.indexer.indexExpr
+        if(index !is NumericLiteralValue && index !is IdentifierReference) {
+            // replace complex indexing expression with a temp variable to hold the computed index first
+            return getAutoIndexerVarFor(arrayIndexedExpression)
+        }
+
+        return noModifications
+    }
+
+    private fun getComplexArrayIndexedExpressions(stmt: Statement): List<ArrayIndexedExpression> {
+
+        class Searcher : IAstVisitor {
+            val complexArrayIndexedExpressions = mutableListOf<ArrayIndexedExpression>()
+            override fun visit(arrayIndexedExpression: ArrayIndexedExpression) {
+                val ix = arrayIndexedExpression.indexer.indexExpr
+                if(ix !is NumericLiteralValue && ix !is IdentifierReference)
+                    complexArrayIndexedExpressions.add(arrayIndexedExpression)
+            }
+
+            override fun visit(branchStatement: BranchStatement) {}
+
+            override fun visit(forLoop: ForLoop) {}
+
+            override fun visit(ifStatement: IfStatement) {
+                ifStatement.condition.accept(this)
+            }
+
+            override fun visit(untilLoop: UntilLoop) {
+                untilLoop.condition.accept(this)
+            }
+        }
+
+        val searcher = Searcher()
+        stmt.accept(searcher)
+        return searcher.complexArrayIndexedExpressions
+    }
+
+    private fun getContainingStatement(expression: Expression): Statement {
+        var node: Node = expression
+        while(node !is Statement)
+            node = node.parent
+
+        return node
+    }
+
+    private fun getAutoIndexerVarFor(expr: ArrayIndexedExpression): MutableList<IAstModification> {
+        val modifications = mutableListOf<IAstModification>()
+        val statement = expr.containingStatement()
+        // replace the indexer with just the variable (simply use a cx16 virtual register r9, that we HOPE is not used for other things in the expression...)
+        // assign the indexing expression to the helper variable, but only if that hasn't been done already
+        val target = AssignTarget(IdentifierReference(listOf("cx16", "r9"), expr.indexer.position), null, null, expr.indexer.position)
+        val assign = Assignment(target, expr.indexer.indexExpr, expr.indexer.position)
+        modifications.add(IAstModification.InsertBefore(statement, assign, statement.definingScope()))
+        modifications.add(IAstModification.ReplaceNode(expr.indexer.indexExpr, target.identifier!!.copy(), expr.indexer))
+        return modifications
+    }
+
 }

compiler/src/prog8/compiler/Compiler.kt

@@ -96,7 +96,7 @@ fun compileProgram(filepath: Path,
             importedFiles = imported
             processAst(programAst, errors, compilationOptions)
             if (compilationOptions.optimize)
-                optimizeAst(programAst, errors, BuiltinFunctionsFacade(BuiltinFunctions), compTarget)
+                optimizeAst(programAst, errors, BuiltinFunctionsFacade(BuiltinFunctions), compTarget, compilationOptions)
             postprocessAst(programAst, errors, compilationOptions)
 
             // printAst(programAst)
@@ -157,7 +157,7 @@ private class BuiltinFunctionsFacade(functions: Map<String, FSignature>): IBuilt
                 }
             }
             else if(func.known_returntype==null)
-                throw IllegalArgumentException("builtin function $name can't be used here because it doesn't return a value")
+                return null  // builtin function $name can't be used here because it doesn't return a value
         }
         return null
     }
@@ -214,7 +214,7 @@ private fun determineCompilationOptions(program: Program, compTarget: ICompilati
                 }
 
     if (zpType==ZeropageType.FLOATSAFE && compTarget.name == Cx16Target.name) {
-        System.err.println("Warning: Cx16 target must use zp option basicsafe instead of floatsafe")
+        System.err.println("Warning: zp option floatsafe changed to basicsafe for cx16 target")
         zpType = ZeropageType.BASICSAFE
     }
 
@@ -253,40 +253,54 @@ private fun processAst(programAst: Program, errors: IErrorReporter, compilerOpti
     errors.report()
     programAst.addTypecasts(errors)
     errors.report()
-    programAst.variousCleanups()
+    programAst.variousCleanups(programAst, errors)
+    errors.report()
     programAst.checkValid(compilerOptions, errors, compilerOptions.compTarget)
     errors.report()
     programAst.checkIdentifiers(errors, compilerOptions.compTarget)
     errors.report()
 }
 
-private fun optimizeAst(programAst: Program, errors: IErrorReporter, functions: IBuiltinFunctions, compTarget: ICompilationTarget) {
+private fun optimizeAst(programAst: Program, errors: IErrorReporter, functions: IBuiltinFunctions, compTarget: ICompilationTarget, options: CompilationOptions) {
     // optimize the parse tree
     println("Optimizing...")
+
+    val remover = UnusedCodeRemover(programAst, errors, compTarget)
+    remover.visit(programAst)
+    remover.applyModifications()
+
     while (true) {
         // keep optimizing expressions and statements until no more steps remain
         val optsDone1 = programAst.simplifyExpressions()
         val optsDone2 = programAst.splitBinaryExpressions(compTarget)
-        val optsDone3 = programAst.optimizeStatements(errors, functions, compTarget, ::loadAsmIncludeFile)
+        val optsDone3 = programAst.optimizeStatements(errors, functions, compTarget)
         programAst.constantFold(errors, compTarget) // because simplified statements and expressions can result in more constants that can be folded away
         errors.report()
         if (optsDone1 + optsDone2 + optsDone3 == 0)
             break
     }
 
-    val remover = UnusedCodeRemover(programAst, errors, compTarget, ::loadAsmIncludeFile)
-    remover.visit(programAst)
-    remover.applyModifications()
+    val inliner = SubroutineInliner(programAst, errors, options)
+    inliner.visit(programAst)
+    errors.report()
+    if(errors.noErrors()) {
+        inliner.applyModifications()
+        inliner.fixCallsToInlinedSubroutines()
+        val remover2 = UnusedCodeRemover(programAst, errors, compTarget)
+        remover2.visit(programAst)
+        remover2.applyModifications()
+    }
+
     errors.report()
 }
 
 private fun postprocessAst(programAst: Program, errors: IErrorReporter, compilerOptions: CompilationOptions) {
     programAst.addTypecasts(errors)
     errors.report()
-    programAst.variousCleanups()
+    programAst.variousCleanups(programAst, errors)
     programAst.checkValid(compilerOptions, errors, compilerOptions.compTarget)          // check if final tree is still valid
     errors.report()
-    val callGraph = CallGraph(programAst, ::loadAsmIncludeFile)
+    val callGraph = CallGraph(programAst)
     callGraph.checkRecursiveCalls(errors)
     errors.report()
     programAst.verifyFunctionArgTypes()
@@ -297,7 +311,7 @@ private fun writeAssembly(programAst: Program,
                           errors: IErrorReporter,
                           outputDir: Path,
                           compilerOptions: CompilationOptions): String {
-    // asm generation directly from the Ast,
+    // asm generation directly from the Ast
     programAst.processAstBeforeAsmGeneration(errors, compilerOptions.compTarget)
     errors.report()
 

compiler/src/prog8/compiler/ErrorReporting.kt

@@ -7,7 +7,7 @@ import prog8.parser.ParsingFailedError
 interface IErrorReporter {
     fun err(msg: String, position: Position)
     fun warn(msg: String, position: Position)
-    fun isEmpty(): Boolean
+    fun noErrors(): Boolean
     fun report()
 }
 
@@ -53,5 +53,5 @@ internal class ErrorReporter: IErrorReporter {
             throw ParsingFailedError("There are $numErrors errors and $numWarnings warnings.")
     }
 
-    override fun isEmpty() = messages.isEmpty()
+    override fun noErrors() = messages.none { it.severity==MessageSeverity.ERROR }
 }

compiler/src/prog8/compiler/astprocessing/AstChecker.kt

@@ -9,6 +9,7 @@ import prog8.ast.statements.*
 import prog8.ast.walk.IAstVisitor
 import prog8.compiler.CompilationOptions
 import prog8.compiler.IErrorReporter
+import prog8.compiler.ZeropageType
 import prog8.compiler.functions.BuiltinFunctions
 import prog8.compiler.functions.builtinFunctionReturnType
 import prog8.compiler.target.C64Target
@@ -41,6 +42,11 @@ internal class AstChecker(private val program: Program,
             }
         }
 
+        if(compilerOptions.floats) {
+            if (compilerOptions.zeropage !in setOf(ZeropageType.FLOATSAFE, ZeropageType.BASICSAFE, ZeropageType.DONTUSE ))
+                errors.err("when floats are enabled, zero page type should be 'floatsafe' or 'basicsafe' or 'dontuse'", program.mainModule.position)
+        }
+
         super.visit(program)
     }
 
@@ -70,7 +76,7 @@ internal class AstChecker(private val program: Program,
         if(expectedReturnValues.size==1 && returnStmt.value!=null) {
             val valueDt = returnStmt.value!!.inferType(program)
             if(!valueDt.isKnown) {
-                errors.err("return value type mismatch", returnStmt.value!!.position)
+                errors.err("return value type mismatch or unknown symbol", returnStmt.value!!.position)
             } else {
                 if (expectedReturnValues[0] != valueDt.typeOrElse(DataType.STRUCT))
                     errors.err("type $valueDt of return value doesn't match subroutine's return type ${expectedReturnValues[0]}", returnStmt.value!!.position)
@@ -118,7 +124,7 @@ internal class AstChecker(private val program: Program,
                     }
                     else -> errors.err("loop variable must be numeric type", forLoop.position)
                 }
-                if(errors.isEmpty()) {
+                if(errors.noErrors()) {
                     // check loop range values
                     val range = forLoop.iterable as? RangeExpr
                     if(range!=null) {
@@ -203,13 +209,6 @@ internal class AstChecker(private val program: Program,
         if(uniqueNames.size!=subroutine.parameters.size)
             err("parameter names must be unique")
 
-        if(subroutine.inline) {
-            if (subroutine.containsDefinedVariables())
-                err("can't inline a subroutine that defines variables")
-            if (!subroutine.isAsmSubroutine && subroutine.parameters.isNotEmpty())
-                err("can't inline a non-asm subroutine that has parameters")
-        }
-
         super.visit(subroutine)
 
         // user-defined subroutines can only have zero or one return type
@@ -218,13 +217,13 @@ internal class AstChecker(private val program: Program,
             err("subroutines can only have one return value")
 
         // subroutine must contain at least one 'return' or 'goto'
-        // (or if it has an asm block, that must contain a 'rts' or 'jmp')
+        // (or if it has an asm block, that must contain a 'rts' or 'jmp' or 'bra')
         if(subroutine.statements.count { it is Return || it is Jump } == 0) {
             if (subroutine.amountOfRtsInAsm() == 0) {
                 if (subroutine.returntypes.isNotEmpty()) {
                     // for asm subroutines with an address, no statement check is possible.
                     if (subroutine.asmAddress == null && !subroutine.inline)
-                        err("non-inline subroutine has result value(s) and thus must have at least one 'return' or 'goto' in it (or 'rts' / 'jmp' in case of %asm)")
+                        err("non-inline subroutine has result value(s) and thus must have at least one 'return' or 'goto' in it (or rts/jmp/bra in case of %asm)")
                 }
             }
         }
@@ -374,7 +373,7 @@ internal class AstChecker(private val program: 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)) {
+                if(stmt.returntypes.isEmpty() || (stmt.returntypes.size == 1 && stmt.returntypes.single() isNotAssignableTo idt.typeOrElse(DataType.BYTE))) {
                     errors.err("return type mismatch", assignment.value.position)
                 }
             }
@@ -493,7 +492,7 @@ internal class AstChecker(private val program: Program,
         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?.referencesIdentifier(decl.name) == true || decl.arraysize?.indexVar?.referencesIdentifier(decl.name) == true)
+        if(decl.value?.referencesIdentifier(decl.name) == true || decl.arraysize?.indexExpr?.referencesIdentifier(decl.name) == true)
             err("recursive var declaration")
 
         // CONST can only occur on simple types (byte, word, float)
@@ -953,6 +952,20 @@ internal class AstChecker(private val program: Program,
             }
         }
 
+        // functions that don't return a value, can't be used in an expression or assignment
+        if(targetStatement is Subroutine) {
+            if(targetStatement.returntypes.isEmpty()) {
+                if(functionCall.parent is Expression || functionCall.parent is Assignment)
+                    errors.err("subroutine doesn't return a value", functionCall.position)
+            }
+        }
+        else if(targetStatement is BuiltinFunctionStatementPlaceholder) {
+            if(builtinFunctionReturnType(targetStatement.name, functionCall.args, program).isUnknown) {
+                if(functionCall.parent is Expression || functionCall.parent is Assignment)
+                    errors.err("function doesn't return a value", functionCall.position)
+            }
+        }
+
         super.visit(functionCall)
     }
 
@@ -1111,15 +1124,9 @@ internal class AstChecker(private val program: Program,
             errors.err("indexing requires a variable to act upon", arrayIndexedExpression.position)
 
         // check index value 0..255
-        val dtxNum = arrayIndexedExpression.indexer.indexNum?.inferType(program)?.typeOrElse(DataType.STRUCT)
-        if(dtxNum!=null && dtxNum != DataType.UBYTE && dtxNum != DataType.BYTE)
+        val dtxNum = arrayIndexedExpression.indexer.indexExpr.inferType(program)
+        if(!dtxNum.istype(DataType.UBYTE) && !dtxNum.istype(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)
     }

compiler/src/prog8/compiler/astprocessing/AstExtensions.kt

@@ -17,13 +17,20 @@ internal fun Program.checkValid(compilerOptions: CompilationOptions, errors: IEr
 internal fun Program.processAstBeforeAsmGeneration(errors: IErrorReporter, compTarget: ICompilationTarget) {
     val fixer = BeforeAsmGenerationAstChanger(this, errors, compTarget)
     fixer.visit(this)
-    fixer.applyModifications()
+    while(errors.noErrors() && fixer.applyModifications()>0) {
+        fixer.visit(this)
+    }
 }
 
 internal fun Program.reorderStatements(errors: IErrorReporter) {
     val reorder = StatementReorderer(this, errors)
     reorder.visit(this)
+    if(errors.noErrors()) {
         reorder.applyModifications()
+        reorder.visit(this)
+        if(errors.noErrors())
+            reorder.applyModifications()
+    }
 }
 
 internal fun Program.addTypecasts(errors: IErrorReporter) {
@@ -42,7 +49,7 @@ internal fun Program.checkIdentifiers(errors: IErrorReporter, compTarget: ICompi
     val checker2 = AstIdentifiersChecker(this, errors, compTarget)
     checker2.visit(this)
 
-    if(errors.isEmpty()) {
+    if(errors.noErrors()) {
         val transforms = AstVariousTransforms(this)
         transforms.visit(this)
         transforms.applyModifications()
@@ -56,12 +63,14 @@ internal fun Program.checkIdentifiers(errors: IErrorReporter, compTarget: ICompi
     }
 }
 
-internal fun Program.variousCleanups() {
-    val process = VariousCleanups()
+internal fun Program.variousCleanups(program: Program, errors: IErrorReporter) {
+    val process = VariousCleanups(program, errors)
     process.visit(this)
+    if(errors.noErrors())
         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,
@@ -69,7 +78,6 @@ internal fun Program.moveMainAndStartToFirst() {
 
     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))
@@ -93,5 +101,4 @@ internal fun Program.moveMainAndStartToFirst() {
         modules[0].statements.removeAll { it is Directive && it.directive == directive.directive }
         modules[0].statements.add(0, directive)
     }
-    }
 }

compiler/src/prog8/compiler/astprocessing/AstVariousTransforms.kt

@@ -14,7 +14,6 @@ import prog8.ast.walk.IAstModification
 
 
 internal class AstVariousTransforms(private val program: Program) : AstWalker() {
-    private val noModifications = emptyList<IAstModification>()
 
     override fun before(decl: VarDecl, parent: Node): Iterable<IAstModification> {
         // is it a struct variable? then define all its struct members as mangled names,
@@ -23,9 +22,7 @@ internal class AstVariousTransforms(private val program: Program) : AstWalker()
             val decls = decl.flattenStructMembers()
             decls.add(decl)
             val result = AnonymousScope(decls, decl.position)
-            return listOf(IAstModification.ReplaceNode(
-                    decl, result, parent
-            ))
+            return listOf(IAstModification.ReplaceNode(decl, result, parent))
         }
 
         return noModifications

compiler/src/prog8/compiler/astprocessing/LiteralsToAutoVars.kt

@@ -13,7 +13,6 @@ import prog8.ast.walk.IAstModification
 
 
 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) {

compiler/src/prog8/compiler/astprocessing/StatementReorderer.kt

@@ -23,7 +23,6 @@ internal class StatementReorderer(val program: Program, val errors: IErrorReport
     // - sorts the choices in when statement.
     // - insert AddressOf (&) expression where required (string params to a UWORD function param etc).
 
-    private val noModifications = emptyList<IAstModification>()
     private val directivesToMove = setOf("%output", "%launcher", "%zeropage", "%zpreserved", "%address", "%option")
 
     override fun after(module: Module, parent: Node): Iterable<IAstModification> {
@@ -89,8 +88,7 @@ internal class StatementReorderer(val program: Program, val errors: IErrorReport
         if(arrayVar!=null && arrayVar.datatype ==  DataType.UWORD) {
             // rewrite   pointervar[index]  into  @(pointervar+index)
             val indexer = arrayIndexedExpression.indexer
-            val index = (indexer.indexNum ?: indexer.indexVar)!!
-            val add = BinaryExpression(arrayIndexedExpression.arrayvar, "+", index, arrayIndexedExpression.position)
+            val add = BinaryExpression(arrayIndexedExpression.arrayvar, "+", indexer.indexExpr, arrayIndexedExpression.position)
             return if(parent is AssignTarget) {
                 // we're part of the target of an assignment, we have to actually change the assign target itself
                 val memwrite = DirectMemoryWrite(add, arrayIndexedExpression.position)
@@ -102,24 +100,8 @@ internal class StatementReorderer(val program: Program, val errors: IErrorReport
             }
         }
 
-        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
-        }
-    }
 
     override fun after(expr: BinaryExpression, parent: Node): Iterable<IAstModification> {
         // when using a simple bit shift and assigning it to a variable of a different type,
@@ -202,38 +184,6 @@ internal class StatementReorderer(val program: Program, val errors: IErrorReport
         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 a bit smarter so it can reuse indexer variables. BUT BEWARE of scoping+initialization problems then
-        // add another loop index var to be used for this expression
-        val indexerVarName = "$indexerVarPrefix${expr.indexer.hashCode()}"
-        val indexerVar = AsmGenInfo.ArrayIndexerInfo(indexerVarName, expr.indexer)
-        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))
-
-        // 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)
-        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> {
         val choices = whenStatement.choiceValues(program).sortedBy {
             it.first?.first() ?: Int.MAX_VALUE
@@ -248,52 +198,40 @@ internal class StatementReorderer(val program: Program, val errors: IErrorReport
         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) {
+                // move the vardecl (without value) to the scope of the defining subroutine and put a regular assignment in its place here.
                 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())
+                        IAstModification.InsertFirst(decl, decl.definingSubroutine() as INameScope)
                 )
             }
         }
-        }
         return noModifications
     }
 
     override fun before(assignment: Assignment, parent: Node): Iterable<IAstModification> {
         val valueType = assignment.value.inferType(program)
         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 = [ ..... ] '
+            if (assignment.value is ArrayLiteralValue) {
+                errors.err("cannot assign array literal here, use separate assignment per field", assignment.position)
             } else {
-                flattenStructAssignmentFromIdentifier(assignment)    //   'structvar1 = structvar2'
+                return copyStructValue(assignment)
             }
         }
 
         if(targetType.typeOrElse(DataType.STRUCT) in ArrayDatatypes && valueType.typeOrElse(DataType.STRUCT) in ArrayDatatypes ) {
-            assignments = if (assignment.value is ArrayLiteralValue) {
-                flattenArrayAssignmentFromArrayLiteral(assignment)    //  'arrayvar = [ ..... ] '
+            if (assignment.value is ArrayLiteralValue) {
+                errors.err("cannot assign array literal here, use separate assignment per element", assignment.position)
             } else {
-                flattenArrayAssignmentFromIdentifier(assignment)    //   'arrayvar1 = arrayvar2'
+                return copyArrayValue(assignment)
             }
         }
 
-        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
     }
 
@@ -346,114 +284,98 @@ internal class StatementReorderer(val program: Program, val errors: IErrorReport
         return noModifications
     }
 
-    private fun flattenArrayAssignmentFromArrayLiteral(assign: Assignment): List<Assignment> {
+    private fun copyArrayValue(assign: Assignment): List<IAstModification> {
         val identifier = assign.target.identifier!!
         val targetVar = identifier.targetVarDecl(program)!!
-        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)!!
+        if(targetVar.arraysize==null)
+            errors.err("array has no defined size", assign.position)
+
+        if(assign.value !is IdentifierReference) {
+            errors.err("invalid array value to assign to other array", assign.value.position)
+            return noModifications
+        }
         val sourceIdent = assign.value as IdentifierReference
         val sourceVar = sourceIdent.targetVarDecl(program)!!
         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)
+        } else {
+            if (sourceVar.arraysize!!.constIndex() != targetVar.arraysize!!.constIndex())
+                errors.err("element count mismatch", assign.position)
+            if (sourceVar.datatype != targetVar.datatype)
+                errors.err("element type mismatch", 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(!errors.noErrors())
+            return noModifications
+
+        val memcopy = FunctionCallStatement(IdentifierReference(listOf("sys", "memcopy"), assign.position),
+            mutableListOf(
+                AddressOf(sourceIdent, assign.position),
+                AddressOf(identifier, assign.position),
+                NumericLiteralValue.optimalInteger(targetVar.arraysize!!.constIndex()!!, assign.position)
+            ),
+            true,
+            assign.position
+        )
+        return listOf(IAstModification.ReplaceNode(assign, memcopy, assign.parent))
     }
 
-        if(alv==null || alv.value.size != targetVar.arraysize!!.constIndex()) {
-            errors.err("element count mismatch", position)
-            return emptyList()
-        }
-
-        // TODO use memcopy instead of individual assignments after certain amount of elements
-        // TODO what does assigning a struct var use?
-        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> {
+    private fun copyStructValue(structAssignment: Assignment): List<IAstModification> {
         val identifier = structAssignment.target.identifier!!
-        val identifierName = identifier.nameInSource.single()
-        val targetVar = identifier.targetVarDecl(program)!!
-        val struct = targetVar.struct!!
-
-        val slv = structAssignment.value as? ArrayLiteralValue
-        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
-            val mangled = mangledStructMemberName(identifierName, targetDecl.name)
-            val idref = IdentifierReference(listOf(mangled), structAssignment.position)
-            val assign = Assignment(AssignTarget(idref, null, null, structAssignment.position),
-                    sourceValue, sourceValue.position)
-            assign.linkParents(structAssignment)
-            assign
-        }
-    }
-
-    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)!!
         val struct = targetVar.struct!!
         when (structAssignment.value) {
             is IdentifierReference -> {
                 val sourceVar = (structAssignment.value as IdentifierReference).targetVarDecl(program)!!
+                val memsize = struct.memsize(program.memsizer)
                 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
+                            errors.err("struct type mismatch", structAssignment.position)
+                            return listOf()
                         }
-                        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
+                        if(struct.statements.size!=sourceStruct.statements.size) {
+                            errors.err("struct element count mismatch", structAssignment.position)
+                            return listOf()
                         }
+                        if(memsize!=sourceStruct.memsize(program.memsizer)) {
+                            errors.err("memory size mismatch", structAssignment.position)
+                            return listOf()
+                        }
+                        val memcopy = FunctionCallStatement(IdentifierReference(listOf("sys", "memcopy"), structAssignment.position),
+                            mutableListOf(
+                                AddressOf(structAssignment.value as IdentifierReference, structAssignment.position),
+                                AddressOf(identifier, structAssignment.position),
+                                NumericLiteralValue.optimalInteger(memsize, structAssignment.position)
+                            ),
+                            true,
+                            structAssignment.position
+                        )
+                        return listOf(IAstModification.ReplaceNode(structAssignment, memcopy, structAssignment.parent))
                     }
                     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
+                        val array = sourceVar.value as ArrayLiteralValue
+                        if(struct.statements.size!=array.value.size) {
+                            errors.err("struct element count mismatch", structAssignment.position)
+                            return listOf()
                         }
+                        if(memsize!=array.memsize(program.memsizer)) {
+                            errors.err("memory size mismatch", structAssignment.position)
+                            return listOf()
+                        }
+                        val memcopy = FunctionCallStatement(IdentifierReference(listOf("sys", "memcopy"), structAssignment.position),
+                            mutableListOf(
+                                AddressOf(structAssignment.value as IdentifierReference, structAssignment.position),
+                                AddressOf(identifier, structAssignment.position),
+                                NumericLiteralValue.optimalInteger(memsize, structAssignment.position)
+                            ),
+                            true,
+                            structAssignment.position
+                        )
+                        return listOf(IAstModification.ReplaceNode(structAssignment, memcopy, structAssignment.parent))
                     }
                     else -> {
                         throw FatalAstException("can only assign arrays or structs to structs")
@@ -461,7 +383,7 @@ internal class StatementReorderer(val program: Program, val errors: IErrorReport
                 }
             }
             is ArrayLiteralValue -> {
-                throw IllegalArgumentException("not going to flatten a structLv assignment here")
+                throw IllegalArgumentException("not going to do a structLv assignment here")
             }
             else -> throw FatalAstException("strange struct value")
         }

compiler/src/prog8/compiler/astprocessing/TypecastsAdder.kt

@@ -18,8 +18,6 @@ class TypecastsAdder(val program: Program, val errors: IErrorReporter) : AstWalk
      * (this includes function call arguments)
      */
 
-    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) {

compiler/src/prog8/compiler/astprocessing/VariousCleanups.kt

@@ -3,18 +3,17 @@ package prog8.compiler.astprocessing
 import prog8.ast.IFunctionCall
 import prog8.ast.INameScope
 import prog8.ast.Node
+import prog8.ast.Program
+import prog8.ast.base.FatalAstException
 import prog8.ast.base.Position
-import prog8.ast.expressions.DirectMemoryRead
-import prog8.ast.expressions.FunctionCall
-import prog8.ast.expressions.NumericLiteralValue
-import prog8.ast.expressions.TypecastExpression
+import prog8.ast.expressions.*
 import prog8.ast.statements.*
 import prog8.ast.walk.AstWalker
 import prog8.ast.walk.IAstModification
+import prog8.compiler.IErrorReporter
 
 
-internal class VariousCleanups: AstWalker() {
-    private val noModifications = emptyList<IAstModification>()
+internal class VariousCleanups(val program: Program, val errors: IErrorReporter): AstWalker() {
 
     override fun before(nopStatement: NopStatement, parent: Node): Iterable<IAstModification> {
         return listOf(IAstModification.Remove(nopStatement, parent as INameScope))
@@ -32,21 +31,12 @@ internal class VariousCleanups: AstWalker() {
             val idx = into.statements.indexOf(scope)
             if(idx>=0) {
                 into.statements.addAll(idx+1, scope.statements)
+                scope.statements.forEach { it.parent = into as Node }
                 into.statements.remove(scope)
             }
         }
     }
 
-    override fun before(typecast: TypecastExpression, parent: Node): Iterable<IAstModification> {
-        if(typecast.expression is NumericLiteralValue) {
-            val value = (typecast.expression as NumericLiteralValue).cast(typecast.type)
-            if(value.isValid)
-                return listOf(IAstModification.ReplaceNode(typecast, value.valueOrZero(), parent))
-        }
-
-        return noModifications
-    }
-
     override fun before(functionCallStatement: FunctionCallStatement, parent: Node): Iterable<IAstModification> {
         return before(functionCallStatement as IFunctionCall, parent, functionCallStatement.position)
     }
@@ -70,4 +60,62 @@ internal class VariousCleanups: AstWalker() {
         return noModifications
     }
 
+    override fun after(typecast: TypecastExpression, parent: Node): Iterable<IAstModification> {
+        if(typecast.parent!==parent)
+            throw FatalAstException("parent node mismatch at $typecast")
+
+        if(typecast.expression is NumericLiteralValue) {
+            val value = (typecast.expression as NumericLiteralValue).cast(typecast.type)
+            if(value.isValid)
+                return listOf(IAstModification.ReplaceNode(typecast, value.valueOrZero(), parent))
+        }
+
+        val sourceDt = typecast.expression.inferType(program)
+        if(sourceDt.istype(typecast.type))
+            return listOf(IAstModification.ReplaceNode(typecast, typecast.expression, parent))
+
+        return noModifications
+    }
+
+    override fun after(subroutine: Subroutine, parent: Node): Iterable<IAstModification> {
+        if(subroutine.parent!==parent)
+            throw FatalAstException("parent node mismatch at $subroutine")
+        return noModifications
+    }
+
+    override fun after(assignment: Assignment, parent: Node): Iterable<IAstModification> {
+        if(assignment.parent!==parent)
+            throw FatalAstException("parent node mismatch at $assignment")
+        return noModifications
+    }
+
+    override fun after(assignTarget: AssignTarget, parent: Node): Iterable<IAstModification> {
+        if(assignTarget.parent!==parent)
+            throw FatalAstException("parent node mismatch at $assignTarget")
+        return noModifications
+    }
+
+    override fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> {
+        if(decl.parent!==parent)
+            throw FatalAstException("parent node mismatch at $decl")
+        return noModifications
+    }
+
+    override fun after(scope: AnonymousScope, parent: Node): Iterable<IAstModification> {
+        if(scope.parent!==parent)
+            throw FatalAstException("parent node mismatch at $scope")
+        return noModifications
+    }
+
+    override fun after(returnStmt: Return, parent: Node): Iterable<IAstModification> {
+        if(returnStmt.parent!==parent)
+            throw FatalAstException("parent node mismatch at $returnStmt")
+        return noModifications
+    }
+
+    override fun after(identifier: IdentifierReference, parent: Node): Iterable<IAstModification> {
+        if(identifier.parent!==parent)
+            throw FatalAstException("parent node mismatch at $identifier")
+        return noModifications
+    }
 }

compiler/src/prog8/compiler/functions/BuiltinFunctions.kt

@@ -97,6 +97,7 @@ private val functionSignatures: List<FSignature> = listOf(
     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),
+    FSignature("cmp"         , false, listOf(FParam("value1", IntegerDatatypes), FParam("value2", NumericDatatypes)), null),
         // these few have a return value depending on the argument(s):
     FSignature("max"         , true, listOf(FParam("values", ArrayDatatypes)), null) { a, p, prg, ct -> collectionArg(a, p, prg, ::builtinMax) },    // type depends on args
     FSignature("min"         , true, listOf(FParam("values", ArrayDatatypes)), null) { a, p, prg, ct -> collectionArg(a, p, prg, ::builtinMin) },    // type depends on args
@@ -137,6 +138,7 @@ private val functionSignatures: List<FSignature> = listOf(
     FSignature("peekw"       , true, listOf(FParam("address", setOf(DataType.UWORD))), DataType.UWORD),
     FSignature("poke"        , false, listOf(FParam("address", setOf(DataType.UWORD)), FParam("value", setOf(DataType.UBYTE))), null),
     FSignature("pokew"       , false, listOf(FParam("address", setOf(DataType.UWORD)), FParam("value", setOf(DataType.UWORD))), null),
+    FSignature("fastrnd8"    , false, emptyList(), DataType.UBYTE),
     FSignature("rnd"         , false, emptyList(), DataType.UBYTE),
     FSignature("rndw"        , false, emptyList(), DataType.UWORD),
     FSignature("rndf"        , false, emptyList(), DataType.FLOAT),
@@ -320,8 +322,7 @@ private fun builtinSizeof(args: List<Expression>, position: Position, program: P
         val target = (args[0] as IdentifierReference).targetStatement(program)
                 ?: throw CannotEvaluateException("sizeof", "no target")
 
-        fun structSize(target: StructDecl) =
-                NumericLiteralValue(DataType.UBYTE, target.statements.map { memsizer.memorySize((it as VarDecl).datatype) }.sum(), position)
+        fun structSize(target: StructDecl) = NumericLiteralValue(DataType.UBYTE, target.memsize(memsizer), position)
 
         return when {
             dt.typeOrElse(DataType.STRUCT) in ArrayDatatypes -> {

compiler/src/prog8/compiler/target/ICompilationTarget.kt

@@ -7,13 +7,15 @@ import prog8.ast.base.*
 import prog8.ast.expressions.IdentifierReference
 import prog8.ast.expressions.NumericLiteralValue
 import prog8.ast.statements.AssignTarget
+import prog8.compiler.AssemblyError
 import prog8.compiler.CompilationOptions
 import prog8.compiler.IErrorReporter
 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.cbm.Petscii
+import prog8.compiler.target.cpu6502.codegen.AsmGen
 import prog8.compiler.target.cx16.CX16MachineDefinition
+import java.io.CharConversionException
 import java.nio.file.Path
 
 
@@ -70,9 +72,17 @@ internal object C64Target: ICompilationTarget {
     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)
+        try {
+            if (altEncoding) Petscii.encodeScreencode(str, true) else Petscii.encodePetscii(str, true)
+        } catch (x: CharConversionException) {
+            throw AssemblyError("There was a problem converting a string to the target machine's char encoding: ${x.message}")
+        }
     override fun decodeString(bytes: List<Short>, altEncoding: Boolean) =
-            if(altEncoding) Petscii.decodeScreencode(bytes, true) else Petscii.decodePetscii(bytes, true)
+        try {
+            if (altEncoding) Petscii.decodeScreencode(bytes, true) else Petscii.decodePetscii(bytes, true)
+        } catch (x: CharConversionException) {
+            throw AssemblyError("There was a problem decoding to a string: ${x.message}")
+        }
 
     override fun memorySize(dt: DataType): Int {
         return when(dt) {
@@ -89,9 +99,17 @@ internal object Cx16Target: ICompilationTarget {
     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)
+        try {
+            if (altEncoding) Petscii.encodeScreencode(str, true) else Petscii.encodePetscii(str, true)
+        } catch (x: CharConversionException) {
+            throw AssemblyError("There was a problem converting a string to the target machine's char encoding: ${x.message}")
+        }
     override fun decodeString(bytes: List<Short>, altEncoding: Boolean) =
-            if(altEncoding) Petscii.decodeScreencode(bytes, true) else Petscii.decodePetscii(bytes, true)
+        try {
+            if (altEncoding) Petscii.decodeScreencode(bytes, true) else Petscii.decodePetscii(bytes, true)
+        } catch (x: CharConversionException) {
+            throw AssemblyError("There was a problem decoding to a string: ${x.message}")
+        }
 
     override fun memorySize(dt: DataType): Int {
         return when(dt) {
@@ -114,5 +132,6 @@ internal fun asmGeneratorFor(
     outputDir: Path
 ): IAssemblyGenerator
 {
+    // at the moment we only have one code generation backend (for 6502 and 65c02)
     return AsmGen(program, errors, zp, options, compTarget, outputDir)
 }

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

@@ -109,6 +109,7 @@ 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(
+                            0x22, 0x23, 0x24, 0x25,
                             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,

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

@@ -1,4 +1,4 @@
-package prog8.compiler.target.c64
+package prog8.compiler.target.cbm
 
 import prog8.compiler.CompilationOptions
 import prog8.compiler.OutputType

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

@@ -1,4 +1,4 @@
-package prog8.compiler.target.c64
+package prog8.compiler.target.cbm
 
 import java.io.CharConversionException
 
@@ -1051,49 +1051,75 @@ object Petscii {
 
 
     fun encodePetscii(text: String, lowercase: Boolean = false): List<Short> {
-        val lookup = if(lowercase) encodingPetsciiLowercase else encodingPetsciiUppercase
-        return text.map {
-            val petscii = lookup[it]
-            petscii?.toShort() ?: when (it) {
+        fun encodeChar(chr: Char, lowercase: Boolean): Short {
+            val screencode = if(lowercase) encodingPetsciiLowercase[chr] else encodingPetsciiUppercase[chr]
+            return screencode?.toShort() ?: when (chr) {
                 '\u0000' -> 0.toShort()
                 in '\u8000'..'\u80ff' -> {
                     // special case: take the lower 8 bit hex value directly
-                    (it.toInt() - 0x8000).toShort()
+                    (chr.toInt() - 0x8000).toShort()
                 }
                 else -> {
                     val case = if (lowercase) "lower" else "upper"
-                    throw CharConversionException("no ${case}case Petscii character for '$it' (${it.toShort()})")
+                    throw CharConversionException("no ${case}Petscii character for '$chr' (${chr.toShort()})")
                 }
             }
         }
+
+        return text.map{
+            try {
+                encodeChar(it, lowercase)
+            } catch (x: CharConversionException) {
+                encodeChar(it, !lowercase)
+            }
+        }
     }
 
     fun decodePetscii(petscii: Iterable<Short>, lowercase: Boolean = false): String {
-        val decodeTable = if(lowercase) decodingPetsciiLowercase else decodingPetsciiUppercase
-        return petscii.map { decodeTable[it.toInt()] }.joinToString("")
+        return petscii.map {
+            val code = it.toInt()
+            try {
+                if(lowercase) decodingPetsciiLowercase[code] else decodingPetsciiUppercase[code]
+            } catch(x: CharConversionException) {
+                if(lowercase) decodingPetsciiUppercase[code] else decodingPetsciiLowercase[code]
+            }
+        }.joinToString("")
     }
 
     fun encodeScreencode(text: String, lowercase: Boolean = false): List<Short> {
-        val lookup = if(lowercase) encodingScreencodeLowercase else encodingScreencodeUppercase
-        return text.map{
-            val screencode = lookup[it]
-            screencode?.toShort() ?: when (it) {
+        fun encodeChar(chr: Char, lowercase: Boolean): Short {
+            val screencode = if(lowercase) encodingScreencodeLowercase[chr] else encodingScreencodeUppercase[chr]
+            return screencode?.toShort() ?: when (chr) {
                 '\u0000' -> 0.toShort()
                 in '\u8000'..'\u80ff' -> {
                     // special case: take the lower 8 bit hex value directly
-                    (it.toInt() - 0x8000).toShort()
+                    (chr.toInt() - 0x8000).toShort()
                 }
                 else -> {
                     val case = if (lowercase) "lower" else "upper"
-                    throw CharConversionException("no ${case}Screencode character for '$it' (${it.toShort()})")
+                    throw CharConversionException("no ${case}Screencode character for '$chr' (${chr.toShort()})")
                 }
             }
         }
+
+        return text.map{
+            try {
+                encodeChar(it, lowercase)
+            } catch (x: CharConversionException) {
+                encodeChar(it, !lowercase)
+            }
+        }
     }
 
     fun decodeScreencode(screencode: Iterable<Short>, lowercase: Boolean = false): String {
-        val decodeTable = if(lowercase) decodingScreencodeLowercase else decodingScreencodeUppercase
-        return screencode.map { decodeTable[it.toInt()] }.joinToString("")
+        return screencode.map {
+            val code = it.toInt()
+            try {
+                if (lowercase) decodingScreencodeLowercase[code] else decodingScreencodeUppercase[code]
+            } catch (x: CharConversionException) {
+                if (lowercase) decodingScreencodeUppercase[code] else decodingScreencodeLowercase[code]
+            }
+        }.joinToString("")
     }
 
     fun petscii2scr(petscii_code: Short, inverseVideo: Boolean): Short {

compiler/src/prog8/compiler/target/cpu6502/codegen/AsmGen.kt

@@ -1,4 +1,4 @@
-package prog8.compiler.target.c64.codegen
+package prog8.compiler.target.cpu6502.codegen
 
 import prog8.ast.*
 import prog8.ast.antlr.escape
@@ -9,11 +9,10 @@ import prog8.compiler.*
 import prog8.compiler.functions.BuiltinFunctions
 import prog8.compiler.functions.FSignature
 import prog8.compiler.target.*
-import prog8.compiler.target.c64.AssemblyProgram
-import prog8.compiler.target.c64.Petscii
-import prog8.compiler.target.c64.codegen.assignment.AsmAssignment
-import prog8.compiler.target.c64.codegen.assignment.AssignmentAsmGen
-import java.io.CharConversionException
+import prog8.compiler.target.cbm.AssemblyProgram
+import prog8.compiler.target.cpu6502.codegen.assignment.AsmAssignment
+import prog8.compiler.target.cpu6502.codegen.assignment.AssignmentAsmGen
+import prog8.optimizer.CallGraph
 import java.nio.file.Path
 import java.nio.file.Paths
 import java.time.LocalDate
@@ -32,6 +31,7 @@ internal class AsmGen(private val program: Program,
     // 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 callGraph = CallGraph(program)
 
     private val assemblyLines = mutableListOf<String>()
     private val globalFloatConsts = mutableMapOf<Double, String>()     // all float values in the entire program (value -> varname)
@@ -157,7 +157,16 @@ internal class AsmGen(private val program: Program,
                 pha""")
         }
 
-        out("  jmp  main.start  ; start program / force start proc to be included")
+        // make sure that on the cx16 and c64, basic rom is banked in again when we exit the program
+        when(compTarget.name) {
+            Cx16Target.name -> {
+                if(options.floats)
+                    out("  lda  #4 |  sta  $01")    // to use floats, make sure Basic rom is banked in
+                out("  jsr  main.start |  lda  #4 |  sta  $01 |  rts")
+            }
+            C64Target.name -> out("  jsr  main.start |  lda  #31 |  sta  $01 |  rts")
+            else -> jmp("main.start")
+        }
     }
 
     private fun slaballocations() {
@@ -246,15 +255,6 @@ internal class AsmGen(private val program: Program,
         } else assemblyLines.add(fragment)
     }
 
-    private fun encode(str: String, altEncoding: Boolean): List<Short> {
-        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>) {
         out("; vars allocated on zeropage")
         val variables = statements.filterIsInstance<VarDecl>().filter { it.type==VarDeclType.VAR }
@@ -293,7 +293,7 @@ internal class AsmGen(private val program: Program,
             DataType.STRUCT -> {}       // is flattened
             DataType.STR -> {
                 val str = decl.value as StringLiteralValue
-                outputStringvar(decl, encode(str.value, str.altEncoding))
+                outputStringvar(decl, compTarget.encodeString(str.value, str.altEncoding).plus(0))
             }
             DataType.ARRAY_UB -> {
                 val data = makeArrayFillDataUnsigned(decl)
@@ -390,7 +390,7 @@ internal class AsmGen(private val program: Program,
                 .filter {it.datatype == DataType.STR }
                 .map {
                     val str = it.value as StringLiteralValue
-                    it to encode(str.value, str.altEncoding)
+                    it to compTarget.encodeString(str.value, str.altEncoding).plus(0)
                 }
                 .groupBy({it.second}, {it.first})
         for((encoded, variables) in encodedstringVars) {
@@ -695,7 +695,7 @@ internal class AsmGen(private val program: Program,
             is Break -> {
                 if(loopEndLabels.isEmpty())
                     throw AssemblyError("break statement out of context  ${stmt.position}")
-                out("  jmp  ${loopEndLabels.peek()}")
+                jmp(loopEndLabels.peek())
             }
             is WhileLoop -> translate(stmt)
             is RepeatLoop -> translate(stmt)
@@ -720,7 +720,10 @@ internal class AsmGen(private val program: Program,
             return
         }
 
-        val indexName = asmVariableName(expr.indexer.indexVar!!)
+        val indexVar = expr.indexer.indexExpr as? IdentifierReference
+            ?: throw AssemblyError("array indexer should have been replaced with a temp var @ ${expr.indexer.position}")
+
+        val indexName = asmVariableName(indexVar)
         if(addOneExtra) {
             // add 1 to the result
             when(elementDt) {
@@ -818,9 +821,18 @@ internal class AsmGen(private val program: Program,
 
 
     private fun translateSubroutine(sub: Subroutine) {
+        var onlyVariables = false
+
         if(sub.inline) {
-            if(options.optimize)
-                return      // inline subroutines don't exist anymore on their own
+            if(options.optimize) {
+                if(sub.isAsmSubroutine ||callGraph.unused(sub))
+                    return
+
+                // from an inlined subroutine only the local variables are generated,
+                // all other code statements are omitted in the subroutine itself
+                // (they've been inlined at the call site, remember?)
+                onlyVariables = true
+            }
             else if(sub.amountOfRtsInAsm()==0) {
                 // make sure the NOT INLINED subroutine actually does an rts at the end
                 sub.statements.add(Return(null, Position.DUMMY))
@@ -837,7 +849,7 @@ internal class AsmGen(private val program: Program,
 
             // asmsub with most likely just an inline asm in it
             out("${sub.name}\t.proc")
-            sub.statements.forEach{ translate(it) }
+            sub.statements.forEach { translate(it) }
             out("  .pend\n")
         } else {
             // regular subroutine
@@ -861,8 +873,10 @@ internal class AsmGen(private val program: Program,
                     clc""")
             }
 
+            if(!onlyVariables) {
                 out("; statements")
-            sub.statements.forEach{ translate(it) }
+                sub.statements.forEach { translate(it) }
+            }
 
             for(removal in removals.toList()) {
                 if(removal.second==sub) {
@@ -917,6 +931,10 @@ internal class AsmGen(private val program: Program,
         checkBooleanExpression(stmt.condition)  // we require the condition to be of the form  'x <comparison> <value>'
         val booleanCondition = stmt.condition as BinaryExpression
 
+        // DISABLED FOR NOW:
+//        if(!booleanCondition.left.isSimple || !booleanCondition.right.isSimple)
+//            throw AssemblyError("both operands for if comparison expression should have been simplified")
+
         if (stmt.elsepart.containsNoCodeNorVars()) {
             // empty else
             val endLabel = makeLabel("if_end")
@@ -930,7 +948,7 @@ internal class AsmGen(private val program: Program,
             val endLabel = makeLabel("if_end")
             expressionsAsmGen.translateComparisonExpressionWithJumpIfFalse(booleanCondition, elseLabel)
             translate(stmt.truepart)
-            out("  jmp  $endLabel")
+            jmp(endLabel)
             out(elseLabel)
             translate(stmt.elsepart)
             out(endLabel)
@@ -952,7 +970,7 @@ internal class AsmGen(private val program: Program,
                 // endless loop
                 out(repeatLabel)
                 translate(stmt.body)
-                out("  jmp  $repeatLabel")
+                jmp(repeatLabel)
                 out(endLabel)
             }
             is NumericLiteralValue -> {
@@ -976,10 +994,12 @@ internal class AsmGen(private val program: Program,
                 val name = asmVariableName(stmt.iterations as IdentifierReference)
                 when(vardecl.datatype) {
                     DataType.UBYTE, DataType.BYTE -> {
-                        repeatByteCountVar(name, repeatLabel, endLabel, stmt.body)
+                        assignVariableToRegister(name, RegisterOrPair.A)
+                        repeatByteCountInA(null, repeatLabel, endLabel, stmt.body)
                     }
                     DataType.UWORD, DataType.WORD -> {
-                        repeatWordCountVar(name, repeatLabel, endLabel, stmt.body)
+                        assignVariableToRegister(name, RegisterOrPair.AY)
+                        repeatWordCountInAY(null, repeatLabel, endLabel, stmt.body)
                     }
                     else -> throw AssemblyError("invalid loop variable datatype $vardecl")
                 }
@@ -1006,10 +1026,11 @@ internal class AsmGen(private val program: Program,
     }
 
     private fun repeatWordCountInAY(constIterations: Int?, repeatLabel: String, endLabel: String, body: AnonymousScope) {
+        // note: A/Y must have been loaded with the number of iterations!
         if(constIterations==0)
             return
-        // note: A/Y must have been loaded with the number of iterations already!
-        // TODO can be even more optimized by iterating over pages
+        // no need to explicitly test for 0 iterations as this is done in the count down logic below
+
         val counterVar = makeLabel("repeatcounter")
         out("""
                 sta  $counterVar
@@ -1018,80 +1039,48 @@ $repeatLabel    lda  $counterVar
                 bne  +
                 lda  $counterVar+1
                 beq  $endLabel
-+               lda  $counterVar
+                lda  $counterVar
                 bne  +
                 dec  $counterVar+1
 +               dec  $counterVar
 """)
         translate(body)
-        out("  jmp  $repeatLabel")
-        if(constIterations!=null && constIterations>=16 && zeropage.available() > 1) {
-            // allocate count var on ZP
-            val zpAddr = zeropage.allocate(counterVar, DataType.UWORD, body.position, errors)
-            out("""$counterVar = $zpAddr  ; auto zp UWORD""")
-        } else {
-            out("""
-$counterVar    .word  0""")
-        }
-        out(endLabel)
+        jmp(repeatLabel)
 
+        if(constIterations!=null && constIterations>=16 && zeropage.available() > 1) {
+            // allocate count var on ZP  TODO can be shared with countervars from other subroutines
+            val zpAddr = zeropage.allocate(counterVar, DataType.UWORD, body.position, errors)
+            out("$counterVar = $zpAddr  ; auto zp UWORD")
+        } else {
+            out("$counterVar    .word  0")
+        }
+
+        out(endLabel)
     }
 
     private fun repeatByteCountInA(constIterations: Int?, repeatLabel: String, endLabel: String, body: AnonymousScope) {
+        // note: A must have been loaded with the number of iterations!
         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("  beq  $endLabel   ; skip loop if zero iters")
+        val counterVar = makeLabel("repeatcounter")
         out("  sta  $counterVar")
         out(repeatLabel)
         translate(body)
         out("""
              dec  $counterVar
              bne  $repeatLabel
-             beq  $endLabel  
-$counterVar    .byte  0""")
-        out(endLabel)
+             beq  $endLabel""")
+
+        if(constIterations!=null && constIterations>=16 && zeropage.available() > 0) {
+            // allocate count var on ZP  TODO can be shared with countervars from other subroutines
+            val zpAddr = zeropage.allocate(counterVar, DataType.UBYTE, body.position, errors)
+            out("$counterVar = $zpAddr  ; auto zp UBYTE")
+        } else {
+            out("$counterVar    .byte  0")
         }
 
-    private fun repeatByteCountVar(repeatCountVar: String, repeatLabel: String, endLabel: String, body: AnonymousScope) {
-        // note: cannot use original counter variable because it should retain its original value
-        val counterVar = makeLabel("repeatcounter")
-        out("  lda  $repeatCountVar |  beq  $endLabel |  sta  $counterVar")
-        out(repeatLabel)
-        translate(body)
-        out("  dec  $counterVar |  bne  $repeatLabel")
-        // inline countervar:
-        out("""
-                beq  $endLabel  
-$counterVar    .byte  0""")
-        out(endLabel)
-    }
-
-    private fun repeatWordCountVar(repeatCountVar: String, repeatLabel: String, endLabel: String, body: AnonymousScope) {
-        // TODO can be even more optimized by iterating over pages
-        // note: cannot use original counter variable because it should retain its original value
-        val counterVar = makeLabel("repeatcounter")
-        out("""
-            lda  $repeatCountVar
-            sta  $counterVar
-            ora  $repeatCountVar+1
-            beq  $endLabel
-            lda  $repeatCountVar+1
-            sta  $counterVar+1""")
-        out(repeatLabel)
-        translate(body)
-        out("""
-            lda  $counterVar
-            bne  +
-            dec  $counterVar+1
-+           dec  $counterVar
-            lda  $counterVar
-            ora  $counterVar+1
-            bne  $repeatLabel
-            beq  $endLabel  
-$counterVar    .word  0""")
         out(endLabel)
     }
 
@@ -1104,7 +1093,7 @@ $counterVar    .word  0""")
         out(whileLabel)
         expressionsAsmGen.translateComparisonExpressionWithJumpIfFalse(booleanCondition, endLabel)
         translate(stmt.body)
-        out("  jmp  $whileLabel")
+        jmp(whileLabel)
         out(endLabel)
         loopEndLabels.pop()
     }
@@ -1138,7 +1127,7 @@ $counterVar    .word  0""")
             if(choice.values==null) {
                 // the else choice
                 translate(choice.statements)
-                out("  jmp  $endLabel")
+                jmp(endLabel)
             } else {
                 choiceBlocks.add(choiceLabel to choice.statements)
                 for (cv in choice.values!!) {
@@ -1157,11 +1146,11 @@ $counterVar    .word  0""")
                 }
             }
         }
-        out("  jmp  $endLabel")
+        jmp(endLabel)
         for(choiceBlock in choiceBlocks) {
             out(choiceBlock.first)
             translate(choiceBlock.second)
-            out("  jmp  $endLabel")
+            jmp(endLabel)
         }
         out(endLabel)
     }
@@ -1220,7 +1209,7 @@ $counterVar    .word  0""")
                 val endLabel = makeLabel("branch_end")
                 out("  $instruction  $elseLabel")
                 translate(stmt.truepart)
-                out("  jmp  $endLabel")
+                jmp(endLabel)
                 out(elseLabel)
                 translate(stmt.elsepart)
                 out(endLabel)
@@ -1277,14 +1266,12 @@ $label              nop""")
         }
     }
 
-    private fun translate(jmp: Jump) {
-        out("  jmp  ${getJumpTarget(jmp)}")
-    }
+    private fun translate(jump: Jump) = jmp(getJumpTarget(jump))
 
-    private fun getJumpTarget(jmp: Jump): String {
-        val ident = jmp.identifier
-        val label = jmp.generatedLabel
-        val addr = jmp.address
+    private fun getJumpTarget(jump: Jump): String {
+        val ident = jump.identifier
+        val label = jump.generatedLabel
+        val addr = jump.address
         return when {
             ident!=null -> {
                 val target = ident.targetStatement(program)
@@ -1300,7 +1287,7 @@ $label              nop""")
         }
     }
 
-    private fun translate(ret: Return) {
+    internal fun translate(ret: Return, withRts: Boolean=true) {
         ret.value?.let { returnvalue ->
             val sub = ret.definingSubroutine()!!
             val returnType = sub.returntypes.single()
@@ -1319,6 +1306,8 @@ $label              nop""")
                 }
             }
         }
+
+        if(withRts)
             out("  rts")
     }
 
@@ -1378,6 +1367,13 @@ $label              nop""")
         return vardecl.makeScopedName(vardecl.name) in allocatedZeropageVariables
     }
 
+    internal fun jmp(asmLabel: String) {
+        if(isTargetCpu(CpuType.CPU65c02))
+            out("  bra  $asmLabel")     // note: 64tass will convert this automatically to a jmp if the relative distance is too large
+        else
+            out("  jmp  $asmLabel")
+    }
+
     internal fun pointerViaIndexRegisterPossible(pointerOffsetExpr: Expression): Pair<Expression, Expression>? {
         if(pointerOffsetExpr is BinaryExpression && pointerOffsetExpr.operator=="+") {
             val leftDt = pointerOffsetExpr.left.inferType(program)

compiler/src/prog8/compiler/target/cpu6502/codegen/AsmOptimizer.kt

@@ -1,4 +1,4 @@
-package prog8.compiler.target.c64.codegen
+package prog8.compiler.target.cpu6502.codegen
 
 
 // note: see https://wiki.nesdev.com/w/index.php/6502_assembly_optimisations

compiler/src/prog8/compiler/target/cpu6502/codegen/BuiltinFunctionsAsmGen.kt

@@ -1,4 +1,4 @@
-package prog8.compiler.target.c64.codegen
+package prog8.compiler.target.cpu6502.codegen
 
 import prog8.ast.IFunctionCall
 import prog8.ast.Node
@@ -13,7 +13,7 @@ import prog8.ast.toHex
 import prog8.compiler.AssemblyError
 import prog8.compiler.functions.FSignature
 import prog8.compiler.target.CpuType
-import prog8.compiler.target.c64.codegen.assignment.*
+import prog8.compiler.target.cpu6502.codegen.assignment.*
 import prog8.compiler.target.subroutineFloatEvalResultVar2
 
 internal class BuiltinFunctionsAsmGen(private val program: Program, private val asmgen: AsmGen, private val assignAsmGen: AssignmentAsmGen) {
@@ -51,7 +51,7 @@ internal class BuiltinFunctionsAsmGen(private val program: Program, private val
             "ln", "log2", "sqrt", "rad",
             "deg", "round", "floor", "ceil",
             "rndf" -> funcVariousFloatFuncs(fcall, func, resultToStack, resultRegister, sscope)
-            "rnd", "rndw" -> funcRnd(func, resultToStack, resultRegister, sscope)
+            "fastrnd8", "rnd", "rndw" -> funcRnd(func, resultToStack, resultRegister, sscope)
             "sqrt16" -> funcSqrt16(fcall, func, resultToStack, resultRegister, sscope)
             "rol" -> funcRol(fcall)
             "rol2" -> funcRol2(fcall)
@@ -64,7 +64,77 @@ internal class BuiltinFunctionsAsmGen(private val program: Program, private val
             "peek" -> throw AssemblyError("peek() should have been replaced by @()")
             "pokew" -> funcPokeW(fcall)
             "poke" -> throw AssemblyError("poke() should have been replaced by @()")
-            else -> TODO("missing asmgen for builtin func ${func.name}")
+            "cmp" -> funcCmp(fcall)
+            else -> throw AssemblyError("missing asmgen for builtin func ${func.name}")
+        }
+    }
+
+    private fun funcCmp(fcall: IFunctionCall) {
+        val arg1 = fcall.args[0]
+        val arg2 = fcall.args[1]
+        val dt1 = arg1.inferType(program).typeOrElse(DataType.STRUCT)
+        val dt2 = arg2.inferType(program).typeOrElse(DataType.STRUCT)
+        if(dt1 in ByteDatatypes) {
+            if(dt2 in ByteDatatypes) {
+                when (arg2) {
+                    is IdentifierReference -> {
+                        asmgen.assignExpressionToRegister(arg1, RegisterOrPair.A)
+                        asmgen.out("  cmp  ${asmgen.asmVariableName(arg2)}")
+                    }
+                    is NumericLiteralValue -> {
+                        asmgen.assignExpressionToRegister(arg1, RegisterOrPair.A)
+                        asmgen.out("  cmp  #${arg2.number}")
+                    }
+                    is DirectMemoryRead -> {
+                        if(arg2.addressExpression is NumericLiteralValue) {
+                            asmgen.assignExpressionToRegister(arg1, RegisterOrPair.A)
+                            asmgen.out("  cmp  ${arg2.addressExpression.constValue(program)!!.number.toHex()}")
+                        } else {
+                            asmgen.assignExpressionToVariable(arg2, "P8ZP_SCRATCH_B1", DataType.UBYTE, (fcall as Node).definingSubroutine())
+                            asmgen.assignExpressionToRegister(arg1, RegisterOrPair.A)
+                            asmgen.out("  cmp  P8ZP_SCRATCH_B1")
+                        }
+                    }
+                    else -> {
+                        asmgen.assignExpressionToVariable(arg2, "P8ZP_SCRATCH_B1", DataType.UBYTE, (fcall as Node).definingSubroutine())
+                        asmgen.assignExpressionToRegister(arg1, RegisterOrPair.A)
+                        asmgen.out("  cmp  P8ZP_SCRATCH_B1")
+                    }
+                }
+            } else
+                throw AssemblyError("args for cmp() should have same dt")
+        } else {
+            // dt1 is a word
+            if(dt2 in WordDatatypes) {
+                when (arg2) {
+                    is IdentifierReference -> {
+                        asmgen.assignExpressionToRegister(arg1, RegisterOrPair.AY)
+                        asmgen.out("""
+                            cpy  ${asmgen.asmVariableName(arg2)}+1
+                            bne  +
+                            cmp  ${asmgen.asmVariableName(arg2)}
++""")
+                    }
+                    is NumericLiteralValue -> {
+                        asmgen.assignExpressionToRegister(arg1, RegisterOrPair.AY)
+                        asmgen.out("""
+                            cpy  #>${arg2.number}
+                            bne  +
+                            cmp  #<${arg2.number}
++""")
+                    }
+                    else -> {
+                        asmgen.assignExpressionToVariable(arg2, "P8ZP_SCRATCH_W1", DataType.UWORD, (fcall as Node).definingSubroutine())
+                        asmgen.assignExpressionToRegister(arg1, RegisterOrPair.AY)
+                        asmgen.out("""
+                            cpy  P8ZP_SCRATCH_W1+1
+                            bne  +
+                            cmp  P8ZP_SCRATCH_W1
++""")
+                    }
+                }
+            } else
+                throw AssemblyError("args for cmp() should have same dt")
         }
     }
 
@@ -394,8 +464,7 @@ internal class BuiltinFunctionsAsmGen(private val program: Program, private val
 
     private fun translateRolRorArrayArgs(arrayvar: IdentifierReference, indexer: ArrayIndex, operation: String, dt: Char) {
         asmgen.assignExpressionToVariable(AddressOf(arrayvar, arrayvar.position), "prog8_lib.${operation}_array_u${dt}._arg_target", DataType.UWORD, null)
-        val indexerExpr = if(indexer.indexVar!=null) indexer.indexVar!! else indexer.indexNum!!
-        asmgen.assignExpressionToVariable(indexerExpr, "prog8_lib.${operation}_array_u${dt}._arg_index", DataType.UBYTE, null)
+        asmgen.assignExpressionToVariable(indexer.indexExpr, "prog8_lib.${operation}_array_u${dt}._arg_index", DataType.UBYTE, null)
     }
 
     private fun funcVariousFloatFuncs(fcall: IFunctionCall, func: FSignature, resultToStack: Boolean, resultRegister: RegisterOrPair?, scope: Subroutine?) {
@@ -478,11 +547,11 @@ internal class BuiltinFunctionsAsmGen(private val program: Program, private val
                 }
                 DataType.ARRAY_UW -> {
                     asmgen.out("  jsr  prog8_lib.func_${function.name}_uw_into_AY")
-                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.A, scope, program, asmgen), RegisterOrPair.AY)
+                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.AY, scope, program, asmgen), RegisterOrPair.AY)
                 }
                 DataType.ARRAY_W -> {
                     asmgen.out("  jsr  prog8_lib.func_${function.name}_w_into_AY")
-                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.A, scope, program, asmgen), RegisterOrPair.AY)
+                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.AY, scope, program, asmgen), RegisterOrPair.AY)
                 }
                 DataType.ARRAY_F -> {
                     asmgen.out("  jsr  floats.func_${function.name}_f_fac1")
@@ -509,19 +578,19 @@ internal class BuiltinFunctionsAsmGen(private val program: Program, private val
             when (dt.typeOrElse(DataType.STRUCT)) {
                 DataType.ARRAY_UB, DataType.STR -> {
                     asmgen.out("  jsr  prog8_lib.func_sum_ub_into_AY")
-                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.A, scope, program, asmgen), RegisterOrPair.AY)
+                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.AY, scope, program, asmgen), RegisterOrPair.AY)
                 }
                 DataType.ARRAY_B -> {
                     asmgen.out("  jsr  prog8_lib.func_sum_b_into_AY")
-                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.A, scope, program, asmgen), RegisterOrPair.AY)
+                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.AY, scope, program, asmgen), RegisterOrPair.AY)
                 }
                 DataType.ARRAY_UW -> {
                     asmgen.out("  jsr  prog8_lib.func_sum_uw_into_AY")
-                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.A, scope, program, asmgen), RegisterOrPair.AY)
+                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.AY, scope, program, asmgen), RegisterOrPair.AY)
                 }
                 DataType.ARRAY_W -> {
                     asmgen.out("  jsr  prog8_lib.func_sum_w_into_AY")
-                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.A, scope, program, asmgen), RegisterOrPair.AY)
+                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.AY, scope, program, asmgen), RegisterOrPair.AY)
                 }
                 DataType.ARRAY_F -> {
                     asmgen.out("  jsr  floats.func_sum_f_fac1")
@@ -610,10 +679,10 @@ internal class BuiltinFunctionsAsmGen(private val program: Program, private val
                 throw AssemblyError("unknown dt")
             val elementDt = elementIDt.typeOrElse(DataType.STRUCT)
 
-            val firstNum = first.indexer.indexNum
-            val firstVar = first.indexer.indexVar
-            val secondNum = second.indexer.indexNum
-            val secondVar = second.indexer.indexVar
+            val firstNum = first.indexer.indexExpr as? NumericLiteralValue
+            val firstVar = first.indexer.indexExpr as? IdentifierReference
+            val secondNum = second.indexer.indexExpr as? NumericLiteralValue
+            val secondVar = second.indexer.indexExpr as? IdentifierReference
 
             if(firstNum!=null && secondNum!=null) {
                 swapArrayValues(elementDt, arrayVarName1, firstNum, arrayVarName2, secondNum)
@@ -924,15 +993,15 @@ internal class BuiltinFunctionsAsmGen(private val program: Program, private val
             when (dt) {
                 in ByteDatatypes -> {
                     asmgen.out("  jsr  prog8_lib.abs_b_into_A")
-                    assignAsmGen.assignRegisterByte(AsmAssignTarget.fromRegisters(resultRegister!!, scope, program, asmgen), CpuRegister.A)
+                    assignAsmGen.assignRegisterByte(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.A, scope, program, asmgen), CpuRegister.A)
                 }
                 in WordDatatypes -> {
                     asmgen.out("  jsr  prog8_lib.abs_w_into_AY")
-                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister!!, scope, program, asmgen), RegisterOrPair.AY)
+                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.AY, scope, program, asmgen), RegisterOrPair.AY)
                 }
                 DataType.FLOAT -> {
                     asmgen.out("  jsr  floats.abs_f_fac1")
-                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister!!, scope, program, asmgen), RegisterOrPair.FAC1)
+                    assignAsmGen.assignFAC1float(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.FAC1, scope, program, asmgen))
                 }
                 else -> throw AssemblyError("weird type")
             }
@@ -941,6 +1010,14 @@ internal class BuiltinFunctionsAsmGen(private val program: Program, private val
 
     private fun funcRnd(func: FSignature, resultToStack: Boolean, resultRegister: RegisterOrPair?, scope: Subroutine?) {
         when(func.name) {
+            "fastrnd8" -> {
+                if(resultToStack)
+                    asmgen.out("  jsr  prog8_lib.func_fastrnd8_stack")
+                else {
+                    asmgen.out("  jsr  math.fast_randbyte")
+                    assignAsmGen.assignRegisterByte(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.A, scope, program, asmgen), CpuRegister.A)
+                }
+            }
             "rnd" -> {
                 if(resultToStack)
                     asmgen.out("  jsr  prog8_lib.func_rnd_stack")

compiler/src/prog8/compiler/target/cpu6502/codegen/ForLoopsAsmGen.kt

@@ -1,4 +1,4 @@
-package prog8.compiler.target.c64.codegen
+package prog8.compiler.target.cpu6502.codegen
 
 import prog8.ast.Program
 import prog8.ast.base.ArrayElementTypes
@@ -57,9 +57,9 @@ internal class ForLoopsAsmGen(private val program: Program, private val asmgen:
                         lda  $varname
 $modifiedLabel          cmp  #0         ; modified 
                         beq  $endLabel
-                        $incdec  $varname
-                        jmp  $loopLabel
-$endLabel""")
+                        $incdec  $varname""")
+                    asmgen.jmp(loopLabel)
+                    asmgen.out(endLabel)
 
                 } else {
 
@@ -117,16 +117,15 @@ $modifiedLabel2 cmp  #0    ; modified
                             asmgen.out("""
 +               inc  $varname
                 bne  $loopLabel
-                inc  $varname+1
-                jmp  $loopLabel
-                            """)
+                inc  $varname+1""")
+                            asmgen.jmp(loopLabel)
                         } else {
                             asmgen.out("""
 +               lda  $varname
                 bne  +
                 dec  $varname+1
-+               dec  $varname
-                jmp  $loopLabel""")
++               dec  $varname""")
+                            asmgen.jmp(loopLabel)
                         }
                         asmgen.out(endLabel)
                     }
@@ -386,12 +385,14 @@ $loopLabel""")
                     }
                     -2 -> {
                         when (range.last) {
-                            0 -> asmgen.out("""
+                            0 -> {
+                                asmgen.out("""
                                     lda  $varname
                                     beq  $endLabel
                                     dec  $varname
-                                dec  $varname
-                                jmp  $loopLabel""")
+                                    dec  $varname""")
+                                asmgen.jmp(loopLabel)
+                            }
                             1 -> asmgen.out("""
                                     dec  $varname
                                     beq  $endLabel
@@ -413,8 +414,8 @@ $loopLabel""")
                             beq  $endLabel
                             clc
                             adc  #${range.step}
-                            sta  $varname
-                            jmp  $loopLabel""")
+                            sta  $varname""")
+                        asmgen.jmp(loopLabel)
                     }
                 }
                 asmgen.out(endLabel)
@@ -450,9 +451,9 @@ $loopLabel""")
                             sta  $varname
                             lda  $varname+1
                             adc  #>${range.step}
-                            sta  $varname+1
-                            jmp  $loopLabel
-$endLabel""")
+                            sta  $varname+1""")
+                        asmgen.jmp(loopLabel)
+                        asmgen.out(endLabel)
                     }
                 }
             }
@@ -502,9 +503,9 @@ $loopLabel""")
                 asmgen.out("""
                     lda  $varname
                     beq  $endLabel
-                    dec  $varname
-                    jmp  $loopLabel
-$endLabel""")
+                    dec  $varname""")
+                asmgen.jmp(loopLabel)
+                asmgen.out(endLabel)
             }
             1 -> {
                 asmgen.out("""
@@ -545,9 +546,9 @@ $loopLabel""")
             beq  $endLabel
 +           inc  $varname
             bne  $loopLabel
-            inc  $varname+1
-            jmp  $loopLabel
-$endLabel""")
+            inc  $varname+1""")
+        asmgen.jmp(loopLabel)
+        asmgen.out(endLabel)
         asmgen.loopEndLabels.pop()
     }
 
@@ -573,9 +574,9 @@ $loopLabel""")
 +           lda  $varname
             bne  +
             dec  $varname+1
-+           dec  $varname
-            jmp  $loopLabel
-$endLabel""")
++           dec  $varname""")
+        asmgen.jmp(loopLabel)
+        asmgen.out(endLabel)
         asmgen.loopEndLabels.pop()
     }
 

compiler/src/prog8/compiler/target/cpu6502/codegen/FunctionCallAsmGen.kt

@@ -1,4 +1,4 @@
-package prog8.compiler.target.c64.codegen
+package prog8.compiler.target.cpu6502.codegen
 
 import prog8.ast.IFunctionCall
 import prog8.ast.Node
@@ -8,10 +8,10 @@ import prog8.ast.expressions.*
 import prog8.ast.statements.*
 import prog8.compiler.AssemblyError
 import prog8.compiler.target.CpuType
-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.target.cpu6502.codegen.assignment.AsmAssignSource
+import prog8.compiler.target.cpu6502.codegen.assignment.AsmAssignTarget
+import prog8.compiler.target.cpu6502.codegen.assignment.AsmAssignment
+import prog8.compiler.target.cpu6502.codegen.assignment.TargetStorageKind
 
 
 internal class FunctionCallAsmGen(private val program: Program, private val asmgen: AsmGen) {
@@ -111,17 +111,24 @@ internal class FunctionCallAsmGen(private val program: Program, private val asmg
             }
         }
 
-        if(sub.inline && asmgen.options.optimize) {
-            if(sub.containsDefinedVariables())
-                throw AssemblyError("can't inline sub with vars")
-            if(!sub.isAsmSubroutine && sub.parameters.isNotEmpty())
-                throw AssemblyError("can't inline a non-asm subroutine with parameters")
-            asmgen.out("  \t; inlined routine follows: ${sub.name} from ${sub.position}")
-            val statements = sub.statements.filter { it !is ParameterVarDecl && it !is Directive }
-            statements.forEach { asmgen.translate(it) }
-        }
-        else {
+        if(!sub.inline || !asmgen.options.optimize) {
             asmgen.out("  jsr  $subName")
+        } else {
+            // inline the subroutine.
+            // we do this by copying the subroutine's statements at the call site.
+            // NOTE: *if* there is a return statement, it will be the only one, and the very last statement of the subroutine
+            // (this condition has been enforced by an ast check earlier)
+            asmgen.out("  \t; inlined routine follows: ${sub.name}")
+            val statements = sub.statements.filter { it !is ParameterVarDecl && it !is Directive }
+            statements.forEach {
+                if(it is Return) {
+                    asmgen.translate(it, false)     // don't use RTS for the inlined return statement
+                } else {
+                    if(!sub.inline || it !is VarDecl)
+                        asmgen.translate(it)
+                }
+            }
+            asmgen.out("  \t; inlined routine end: ${sub.name}")
         }
 
         // remember: dealing with the X register and/or dealing with return values is the responsibility of the caller
@@ -177,7 +184,7 @@ internal class FunctionCallAsmGen(private val program: Program, private val asmg
                             else
                                 asmgen.out("  lda  #0 |  sta  cx16.${argi.value.second.registerOrPair.toString().toLowerCase()}+1")
                         }
-                        in WordDatatypes ->
+                        in WordDatatypes, in IterableDatatypes ->
                             asmgen.out("""
                                 lda  P8ESTACK_LO$plusIdxStr,x
                                 sta  cx16.${argi.value.second.registerOrPair.toString().toLowerCase()}

compiler/src/prog8/compiler/target/cpu6502/codegen/PostIncrDecrAsmGen.kt

@@ -1,4 +1,4 @@
-package prog8.compiler.target.c64.codegen
+package prog8.compiler.target.cpu6502.codegen
 
 import prog8.ast.Program
 import prog8.ast.base.*
@@ -66,8 +66,9 @@ internal class PostIncrDecrAsmGen(private val program: Program, private val asmg
             targetArrayIdx!=null -> {
                 val asmArrayvarname = asmgen.asmVariableName(targetArrayIdx.arrayvar)
                 val elementDt = targetArrayIdx.inferType(program).typeOrElse(DataType.STRUCT)
-                if(targetArrayIdx.indexer.indexNum!=null) {
-                    val indexValue = targetArrayIdx.indexer.constIndex()!! * program.memsizer.memorySize(elementDt)
+                val constIndex = targetArrayIdx.indexer.constIndex()
+                if(constIndex!=null) {
+                    val indexValue = constIndex * program.memsizer.memorySize(elementDt)
                     when(elementDt) {
                         in ByteDatatypes -> asmgen.out(if (incr) "  inc  $asmArrayvarname+$indexValue" else "  dec  $asmArrayvarname+$indexValue")
                         in WordDatatypes -> {

compiler/src/prog8/compiler/target/cpu6502/codegen/assignment/AsmAssignment.kt

@@ -1,4 +1,4 @@
-package prog8.compiler.target.c64.codegen.assignment
+package prog8.compiler.target.cpu6502.codegen.assignment
 
 import prog8.ast.IMemSizer
 import prog8.ast.Program
@@ -6,7 +6,7 @@ import prog8.ast.base.*
 import prog8.ast.expressions.*
 import prog8.ast.statements.*
 import prog8.compiler.AssemblyError
-import prog8.compiler.target.c64.codegen.AsmGen
+import prog8.compiler.target.cpu6502.codegen.AsmGen
 
 
 internal enum class TargetStorageKind {
@@ -120,13 +120,7 @@ internal class AsmAssignSource(val kind: SourceStorageKind,
             asmgen.asmVariableName(array.arrayvar)
 
     companion object {
-        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(indexer: ArrayIndex, program: Program, asmgen: AsmGen): AsmAssignSource = fromAstSource(indexer.indexExpr, program, asmgen)
 
         fun fromAstSource(value: Expression, program: Program, asmgen: AsmGen): AsmAssignSource {
             val cv = value.constValue(program)

compiler/src/prog8/compiler/target/cpu6502/codegen/assignment/AssignmentAsmGen.kt

@@ -1,4 +1,4 @@
-package prog8.compiler.target.c64.codegen.assignment
+package prog8.compiler.target.cpu6502.codegen.assignment
 
 import prog8.ast.Program
 import prog8.ast.base.*
@@ -9,12 +9,13 @@ import prog8.compiler.AssemblyError
 import prog8.compiler.functions.BuiltinFunctions
 import prog8.compiler.functions.builtinFunctionReturnType
 import prog8.compiler.target.CpuType
-import prog8.compiler.target.c64.codegen.AsmGen
-import prog8.compiler.target.c64.codegen.ExpressionsAsmGen
+import prog8.compiler.target.cpu6502.codegen.AsmGen
+import prog8.compiler.target.cpu6502.codegen.ExpressionsAsmGen
 
 
 internal class AssignmentAsmGen(private val program: Program, private val asmgen: AsmGen,
-                                private val exprAsmgen: ExpressionsAsmGen) {
+                                private val exprAsmgen: ExpressionsAsmGen
+) {
 
     private val augmentableAsmGen = AugmentableAssignmentAsmGen(program, this, exprAsmgen, asmgen)
 
@@ -64,9 +65,10 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                 val value = assign.source.array!!
                 val elementDt = assign.source.datatype
                 val arrayVarName = asmgen.asmVariableName(value.arrayvar)
-                if (value.indexer.indexNum!=null) {
+                val constIndex = value.indexer.constIndex()
+                if (constIndex!=null) {
                     // constant array index value
-                    val indexValue = value.indexer.constIndex()!! * program.memsizer.memorySize(elementDt)
+                    val indexValue = constIndex * program.memsizer.memorySize(elementDt)
                     when (elementDt) {
                         in ByteDatatypes -> {
                             asmgen.out("  lda  $arrayVarName+$indexValue")
@@ -356,7 +358,7 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
 
 
         // special case optimizations
-        if(target.kind==TargetStorageKind.VARIABLE) {
+        if(target.kind== TargetStorageKind.VARIABLE) {
             if(value is IdentifierReference && valueDt != DataType.STRUCT)
                 return assignTypeCastedIdentifier(target.asmVarname, targetDt, asmgen.asmVariableName(value), valueDt)
 
@@ -429,8 +431,8 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
         }
 
         // give up, do it via eval stack
-        // TODO optimize typecasts for more special cases?
         // note: cannot use assignTypeCastedValue because that is ourselves :P
+        // TODO optimize typecasts for more special cases?
         if(this.asmgen.options.slowCodegenWarnings)
             println("warning: slow stack evaluation used for typecast: $value into $targetDt (target=${target.kind} at ${value.position}")
         asmgen.translateExpression(origTypeCastExpression)      // this performs the actual type cast in translateExpression(Typecast)
@@ -569,11 +571,7 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                     else -> throw AssemblyError("weird type")
                 }
             }
-            DataType.STR -> {
-                if (targetDt != DataType.UWORD && targetDt == DataType.STR)
-                    throw AssemblyError("cannot typecast a string into another incompatitble type")
-                TODO("assign typecasted string into target var")
-            }
+            DataType.STR -> throw AssemblyError("cannot typecast a string value")
             else -> throw AssemblyError("weird type")
         }
     }
@@ -708,11 +706,7 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                     else -> throw AssemblyError("weird type")
                 }
             }
-            DataType.STR -> {
-                if (targetDt != DataType.UWORD && targetDt == DataType.STR)
-                    throw AssemblyError("cannot typecast a string into another incompatitble type")
-                TODO("assign typecasted string into target var")
-            }
+            DataType.STR -> throw AssemblyError("cannot typecast a string value")
             else -> throw AssemblyError("weird type")
         }
     }
@@ -1073,11 +1067,11 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                     ldy  #>${target.asmVarname}
                     sta  P8ZP_SCRATCH_W1
                     sty  P8ZP_SCRATCH_W1+1""")
-                if(target.array!!.indexer.indexNum!=null) {
-                    val index = target.array.indexer.constIndex()!!
-                    asmgen.out(" lda  #$index")
+                val constIndex = target.array!!.indexer.constIndex()
+                if(constIndex!=null) {
+                    asmgen.out(" lda  #$constIndex")
                 } else {
-                    val asmvarname = asmgen.asmVariableName(target.array.indexer.indexVar!!)
+                    val asmvarname = asmgen.asmVariableName(target.array.indexer.indexExpr as IdentifierReference)
                     asmgen.out(" lda  $asmvarname")
                 }
                 asmgen.out("  jsr  floats.set_array_float_from_fac1")
@@ -1109,11 +1103,11 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                     ldy  #>${target.asmVarname}
                     sta  P8ZP_SCRATCH_W2
                     sty  P8ZP_SCRATCH_W2+1""")
-                if(target.array!!.indexer.indexNum!=null) {
-                    val index = target.array.indexer.constIndex()!!
-                    asmgen.out(" lda  #$index")
+                val constIndex = target.array!!.indexer.constIndex()
+                if(constIndex!=null) {
+                    asmgen.out(" lda  #$constIndex")
                 } else {
-                    val asmvarname = asmgen.asmVariableName(target.array.indexer.indexVar!!)
+                    val asmvarname = asmgen.asmVariableName(target.array.indexer.indexExpr as IdentifierReference)
                     asmgen.out(" lda  $asmvarname")
                 }
                 asmgen.out(" jsr  floats.set_array_float")
@@ -1156,11 +1150,11 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                     ldy  #>${target.asmVarname}
                     sta  P8ZP_SCRATCH_W2
                     sty  P8ZP_SCRATCH_W2+1""")
-                if(target.array!!.indexer.indexNum!=null) {
-                    val index = target.array.indexer.constIndex()!!
-                    asmgen.out(" lda  #$index")
+                val constIndex = target.array!!.indexer.constIndex()
+                if(constIndex!=null) {
+                    asmgen.out(" lda  #$constIndex")
                 } else {
-                    val asmvarname = asmgen.asmVariableName(target.array.indexer.indexVar!!)
+                    val asmvarname = asmgen.asmVariableName(target.array.indexer.indexExpr as IdentifierReference)
                     asmgen.out(" lda  $asmvarname")
                 }
                 asmgen.out(" jsr  floats.set_array_float")
@@ -1242,11 +1236,20 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                     """)
             }
             TargetStorageKind.ARRAY -> {
-                // TODO optimize slow stack evaluation for this case, see assignVariableUByteIntoWord
-                if(this.asmgen.options.slowCodegenWarnings)
-                    println("warning: slow stack evaluation used for sign-extend byte typecast at ${bytevar.position}")
-                asmgen.translateExpression(wordtarget.origAssign.source.expression!!)
-                assignStackValue(wordtarget)
+                if (wordtarget.constArrayIndexValue!=null) {
+                    val scaledIdx = wordtarget.constArrayIndexValue!! * 2
+                    asmgen.out("  lda  $sourceName")
+                    asmgen.signExtendAYlsb(DataType.BYTE)
+                    asmgen.out("  sta  ${wordtarget.asmVarname}+$scaledIdx |  sty  ${wordtarget.asmVarname}+$scaledIdx+1")
+                }
+                else {
+                    asmgen.saveRegisterLocal(CpuRegister.X, wordtarget.scope!!)
+                    asmgen.loadScaledArrayIndexIntoRegister(wordtarget.array!!, wordtarget.datatype, CpuRegister.X)
+                    asmgen.out("  lda  $sourceName")
+                    asmgen.signExtendAYlsb(DataType.BYTE)
+                    asmgen.out("  sta  ${wordtarget.asmVarname},x |  inx |  tya |  sta  ${wordtarget.asmVarname},x")
+                    asmgen.restoreRegisterLocal(CpuRegister.X)
+                }
             }
             TargetStorageKind.REGISTER -> {
                 when(wordtarget.register!!) {
@@ -1339,8 +1342,16 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
     }
 
     internal fun assignRegisterByte(target: AsmAssignTarget, register: CpuRegister) {
-        if(target.register !in Cx16VirtualRegisters)
+        // we make an exception in the type check for assigning something to a cx16 virtual register
+        if(target.register !in Cx16VirtualRegisters) {
+            if(target.kind==TargetStorageKind.VARIABLE) {
+                val parts = target.asmVarname.split('.')
+                if (parts.size != 2 || parts[0] != "cx16")
                     require(target.datatype in ByteDatatypes)
+            } else {
+                require(target.datatype in ByteDatatypes)
+            }
+        }
 
         when(target.kind) {
             TargetStorageKind.VARIABLE -> {
@@ -1368,7 +1379,8 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                         CpuRegister.X -> asmgen.out(" txa")
                         CpuRegister.Y -> asmgen.out(" tya")
                     }
-                    asmgen.out(" ldy  ${asmgen.asmVariableName(target.array!!.indexer.indexVar!!)} |  sta  ${target.asmVarname},y")
+                    val indexVar = target.array!!.indexer.indexExpr as IdentifierReference
+                    asmgen.out(" ldy  ${asmgen.asmVariableName(indexVar)} |  sta  ${target.asmVarname},y")
                 }
             }
             TargetStorageKind.REGISTER -> {
@@ -1776,8 +1788,9 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                         """)
                 }
                 TargetStorageKind.ARRAY -> {
-                    if (target.array!!.indexer.indexNum!=null) {
-                        val indexValue = target.array.indexer.constIndex()!! * program.memsizer.memorySize(DataType.FLOAT)
+                    val constIndex = target.array!!.indexer.constIndex()
+                    if (constIndex!=null) {
+                        val indexValue = constIndex * program.memsizer.memorySize(DataType.FLOAT)
                         if(asmgen.isTargetCpu(CpuType.CPU65c02))
                             asmgen.out("""
                                 stz  ${target.asmVarname}+$indexValue
@@ -1796,7 +1809,7 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                                 sta  ${target.asmVarname}+$indexValue+4
                             """)
                     } else {
-                        val asmvarname = asmgen.asmVariableName(target.array.indexer.indexVar!!)
+                        val asmvarname = asmgen.asmVariableName(target.array.indexer.indexExpr as IdentifierReference)
                         asmgen.out("""
                             lda  #<${target.asmVarname}
                             sta  P8ZP_SCRATCH_W1
@@ -1841,8 +1854,9 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                 }
                 TargetStorageKind.ARRAY -> {
                     val arrayVarName = target.asmVarname
-                    if (target.array!!.indexer.indexNum!=null) {
-                        val indexValue = target.array.indexer.constIndex()!! * program.memsizer.memorySize(DataType.FLOAT)
+                    val constIndex = target.array!!.indexer.constIndex()
+                    if (constIndex!=null) {
+                        val indexValue = constIndex * program.memsizer.memorySize(DataType.FLOAT)
                         asmgen.out("""
                             lda  $constFloat
                             sta  $arrayVarName+$indexValue
@@ -1856,7 +1870,7 @@ internal class AssignmentAsmGen(private val program: Program, private val asmgen
                             sta  $arrayVarName+$indexValue+4
                         """)
                     } else {
-                        val asmvarname = asmgen.asmVariableName(target.array.indexer.indexVar!!)
+                        val asmvarname = asmgen.asmVariableName(target.array.indexer.indexExpr as IdentifierReference)
                         asmgen.out("""
                             lda  #<${constFloat}
                             sta  P8ZP_SCRATCH_W1

compiler/src/prog8/compiler/target/cpu6502/codegen/assignment/AugmentableAssignmentAsmGen.kt

@@ -1,4 +1,4 @@
-package prog8.compiler.target.c64.codegen.assignment
+package prog8.compiler.target.cpu6502.codegen.assignment
 
 import prog8.ast.Program
 import prog8.ast.base.*
@@ -7,13 +7,14 @@ import prog8.ast.statements.Subroutine
 import prog8.ast.toHex
 import prog8.compiler.AssemblyError
 import prog8.compiler.target.CpuType
-import prog8.compiler.target.c64.codegen.AsmGen
-import prog8.compiler.target.c64.codegen.ExpressionsAsmGen
+import prog8.compiler.target.cpu6502.codegen.AsmGen
+import prog8.compiler.target.cpu6502.codegen.ExpressionsAsmGen
 
 internal class AugmentableAssignmentAsmGen(private val program: Program,
                                            private val assignmentAsmGen: AssignmentAsmGen,
                                            private val exprAsmGen: ExpressionsAsmGen,
-                                           private val asmgen: AsmGen) {
+                                           private val asmgen: AsmGen
+) {
     fun translate(assign: AsmAssignment) {
         require(assign.isAugmentable)
         require(assign.source.kind== SourceStorageKind.EXPRESSION)
@@ -159,7 +160,7 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                         when {
                             valueLv != null -> inplaceModification_byte_litval_to_variable(addr.toHex(), DataType.UBYTE, operator, valueLv.toInt())
                             ident != null -> inplaceModification_byte_variable_to_variable(addr.toHex(), DataType.UBYTE, operator, ident)
-                            // TODO more specialized code for types such as memory read etc.  -> inplaceModification_byte_memread_to_variable()
+                            memread != null -> inplaceModification_byte_memread_to_variable(addr.toHex(), DataType.UBYTE, operator, value)
                             value is TypecastExpression -> {
                                 if (tryRemoveRedundantCast(value, target, operator)) return
                                 inplaceModification_byte_value_to_variable(addr.toHex(), DataType.UBYTE, operator, value)
@@ -198,9 +199,11 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
             }
             TargetStorageKind.ARRAY -> {
                 with(target.array!!.indexer) {
+                    val indexNum = indexExpr as? NumericLiteralValue
+                    val indexVar = indexExpr as? IdentifierReference
                     when {
                         indexNum!=null -> {
-                            val targetVarName = "${target.asmVarname} + ${indexNum!!.number.toInt()*program.memsizer.memorySize(target.datatype)}"
+                            val targetVarName = "${target.asmVarname} + ${indexNum.number.toInt()*program.memsizer.memorySize(target.datatype)}"
                             when(target.datatype) {
                                 in ByteDatatypes -> {
                                     when {
@@ -267,8 +270,8 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                     }
                 }
             }
-            TargetStorageKind.REGISTER -> TODO("reg in-place modification")
-            TargetStorageKind.STACK -> TODO("stack in-place modification")
+            TargetStorageKind.REGISTER -> throw AssemblyError("missing codegen for reg in-place modification")
+            TargetStorageKind.STACK -> throw AssemblyError("missing codegen for stack in-place modification")
         }
     }
 
@@ -319,7 +322,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
             "&", "and" -> asmgen.out(" and  P8ZP_SCRATCH_B1")
             "|", "or" -> asmgen.out(" ora  P8ZP_SCRATCH_B1")
             "^", "xor" -> asmgen.out(" eor  P8ZP_SCRATCH_B1")
-            in comparisonOperators -> TODO("in-place modification for $operator")
             else -> throw AssemblyError("invalid operator for in-place modification $operator")
         }
         if(ptrOnZp)
@@ -360,7 +362,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
             "&", "and" -> asmgen.out(" and  $otherName")
             "|", "or" -> asmgen.out(" ora  $otherName")
             "^", "xor" -> asmgen.out(" eor  $otherName")
-            in comparisonOperators -> TODO("in-place modification for $operator")
             else -> throw AssemblyError("invalid operator for in-place modification $operator")
         }
         if(ptrOnZp)
@@ -463,7 +464,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                 else
                     asmgen.out("  sta  (P8ZP_SCRATCH_W1),y")
             }
-            in comparisonOperators -> TODO("in-place modification for $operator")
             else -> throw AssemblyError("invalid operator for in-place modification $operator")
         }
     }
@@ -539,7 +539,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                 asmgen.assignExpressionToRegister(value, RegisterOrPair.A)
                 asmgen.out("  eor  $name |  sta  $name")
             }
-            in comparisonOperators -> TODO("in-place modification for $operator")
             else -> throw AssemblyError("invalid operator for in-place modification $operator")
         }
     }
@@ -597,7 +596,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
             "&", "and" -> asmgen.out(" lda  $name |  and  $otherName |  sta  $name")
             "|", "or" -> asmgen.out(" lda  $name |  ora  $otherName |  sta  $name")
             "^", "xor" -> asmgen.out(" lda  $name |  eor  $otherName |  sta  $name")
-            in comparisonOperators -> TODO("in-place modification for $operator")
             else -> throw AssemblyError("invalid operator for in-place modification $operator")
         }
     }
@@ -669,7 +667,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
             "&", "and" -> asmgen.out(" lda  $name |  and  #$value |  sta  $name")
             "|", "or" -> asmgen.out(" lda  $name |  ora  #$value |  sta  $name")
             "^", "xor" -> asmgen.out(" lda  $name |  eor  #$value |  sta  $name")
-            in comparisonOperators -> TODO("in-place modification for $operator")
             else -> throw AssemblyError("invalid operator for in-place modification $operator")
         }
     }
@@ -691,8 +688,8 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                     sec
                     sbc  P8ZP_SCRATCH_B1
                     sta  $name""")
-                // TODO: tuned code for more operators
             }
+            // TODO: tuned code for more operators
             else -> {
                 inplaceModification_byte_value_to_variable(name, dt, operator, memread)
             }
@@ -722,8 +719,8 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                     bcc  +
                     dec  $name+1
 +""")
-                // TODO: tuned code for more operators
             }
+            // TODO: tuned code for more operators
             else -> {
                 inplaceModification_word_value_to_variable(name, dt, operator, memread)
             }
@@ -977,7 +974,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                     else -> asmgen.out("  lda  $name |  eor  #<$value |  sta  $name |  lda  $name+1 |  eor  #>$value |  sta  $name+1")
                 }
             }
-            in comparisonOperators -> TODO("in-place modification for $operator")
             else -> throw AssemblyError("invalid operator for in-place modification $operator")
         }
     }
@@ -1053,8 +1049,49 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                             lda  math.multiply_words.result+1
                             sta  $name+1""")
                     }
-                    "/" -> TODO("div (u)wordvar/bytevar")
-                    "%" -> TODO("(u)word remainder bytevar")
+                    "/" -> {
+                        if(dt==DataType.UWORD) {
+                            asmgen.out("""
+                                lda  $name
+                                ldy  $name+1
+                                sta  P8ZP_SCRATCH_W1
+                                sty  P8ZP_SCRATCH_W1+1
+                                lda  $otherName
+                                ldy  #0
+                                jsr  math.divmod_uw_asm
+                                sta  $name
+                                sty  $name+1
+                            """)
+                        } else {
+                            asmgen.out("""
+                                lda  $name
+                                ldy  $name+1
+                                sta  P8ZP_SCRATCH_W1
+                                sty  P8ZP_SCRATCH_W1+1
+                                lda  $otherName
+                                ldy  #0
+                                jsr  math.divmod_w_asm
+                                sta  $name
+                                sty  $name+1
+                            """)
+                        }
+                    }
+                    "%" -> {
+                        if(valueDt!=DataType.UBYTE || dt!=DataType.UWORD)
+                            throw AssemblyError("remainder of signed integers is not properly defined/implemented, use unsigned instead")
+                        asmgen.out("""
+                            lda  $name
+                            ldy  $name+1
+                            sta  P8ZP_SCRATCH_W1
+                            sty  P8ZP_SCRATCH_W1+1
+                            lda  $otherName
+                            ldy  #0
+                            jsr  math.divmod_uw_asm
+                            lda  P8ZP_SCRATCH_W2
+                            sta  $name
+                            lda  P8ZP_SCRATCH_W2+1
+                            sta  $name+1
+                        """)                    }
                     "<<" -> {
                         asmgen.out("""
                         ldy  $otherName
@@ -1099,7 +1136,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                     }
                     "|", "or" -> asmgen.out("  lda  $otherName |  ora  $name |  sta  $name")
                     "^", "xor" -> asmgen.out("  lda  $otherName |  eor  $name |  sta  $name")
-                    in comparisonOperators -> TODO("in-place modification for $operator")
                     else -> throw AssemblyError("invalid operator for in-place modification $operator")
                 }
             }
@@ -1173,7 +1209,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                     "&", "and" -> asmgen.out(" lda  $name |  and  $otherName |  sta  $name |  lda  $name+1 |  and  $otherName+1 |  sta  $name+1")
                     "|", "or" -> asmgen.out(" lda  $name |  ora  $otherName |  sta  $name |  lda  $name+1 |  ora  $otherName+1 |  sta  $name+1")
                     "^", "xor" -> asmgen.out(" lda  $name |  eor  $otherName |  sta  $name |  lda  $name+1 |  eor  $otherName+1 |  sta  $name+1")
-                    in comparisonOperators -> TODO("in-place modification for $operator")
                     else -> throw AssemblyError("invalid operator for in-place modification $operator")
                 }
             }
@@ -1364,7 +1399,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                         asmgen.assignExpressionToRegister(value, RegisterOrPair.A)
                         asmgen.out("  eor  $name |  sta  $name")
                     }
-                    in comparisonOperators -> TODO("in-place modification for $operator")
                     else -> throw AssemblyError("invalid operator for in-place modification $operator")
                 }
             }
@@ -1405,7 +1439,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                         asmgen.assignExpressionToRegister(value, RegisterOrPair.AY)
                         asmgen.out("  eor  $name |  sta  $name |  tya |  eor  $name+1 |  sta  $name+1")
                     }
-                    in comparisonOperators -> TODO("in-place modification for $operator")
                     else -> throw AssemblyError("invalid operator for in-place modification $operator")
                 }
             }
@@ -1453,7 +1486,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                     jsr  floats.FDIV
                 """)
             }
-            in comparisonOperators -> TODO("in-place float modification for $operator")
             else -> throw AssemblyError("invalid operator for in-place float modification $operator")
         }
         asmgen.out("""
@@ -1534,7 +1566,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                     jsr  floats.FDIV
                 """)
             }
-            in comparisonOperators -> TODO("in-place float modification for $operator")
             else -> throw AssemblyError("invalid operator for in-place float modification $operator")
         }
         // store Fac1 back into memory
@@ -1619,7 +1650,6 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                     jsr  floats.FDIV
                 """)
             }
-            in comparisonOperators -> TODO("in-place float modification for $operator")
             else -> throw AssemblyError("invalid operator for in-place float modification $operator")
         }
         // store Fac1 back into memory
@@ -1729,9 +1759,9 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                             }
                         }
                     }
-                    TargetStorageKind.ARRAY -> TODO("in-place not of ubyte array")
-                    TargetStorageKind.REGISTER -> TODO("reg not")
-                    TargetStorageKind.STACK -> TODO("stack not")
+                    TargetStorageKind.ARRAY -> throw AssemblyError("missing codegen for in-place not of ubyte array")
+                    TargetStorageKind.REGISTER -> throw AssemblyError("missing codegen for reg not")
+                    TargetStorageKind.STACK -> throw AssemblyError("missing codegen for stack not")
                 }
             }
             DataType.UWORD -> {
@@ -1748,9 +1778,9 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                             sta  ${target.asmVarname}+1""")
                     }
                     TargetStorageKind.MEMORY -> throw AssemblyError("no asm gen for uword-memory not")
-                    TargetStorageKind.ARRAY -> TODO("in-place not of uword array")
-                    TargetStorageKind.REGISTER -> TODO("reg not")
-                    TargetStorageKind.STACK -> TODO("stack not")
+                    TargetStorageKind.ARRAY -> throw AssemblyError("missing codegen for in-place not of uword array")
+                    TargetStorageKind.REGISTER -> throw AssemblyError("missing codegen for reg not")
+                    TargetStorageKind.STACK -> throw AssemblyError("missing codegen for stack not")
                 }
             }
             else -> throw AssemblyError("boolean-not of invalid type")
@@ -1795,9 +1825,9 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                             }
                         }
                     }
-                    TargetStorageKind.ARRAY -> TODO("in-place invert ubyte array")
-                    TargetStorageKind.REGISTER -> TODO("reg invert")
-                    TargetStorageKind.STACK -> TODO("stack invert")
+                    TargetStorageKind.ARRAY -> throw AssemblyError("missing codegen for in-place invert ubyte array")
+                    TargetStorageKind.REGISTER -> throw AssemblyError("missing codegen for reg invert")
+                    TargetStorageKind.STACK -> throw AssemblyError("missing codegen for stack invert")
                 }
             }
             DataType.UWORD -> {
@@ -1812,9 +1842,9 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                             sta  ${target.asmVarname}+1""")
                     }
                     TargetStorageKind.MEMORY -> throw AssemblyError("no asm gen for uword-memory invert")
-                    TargetStorageKind.ARRAY -> TODO("in-place invert uword array")
-                    TargetStorageKind.REGISTER -> TODO("reg invert")
-                    TargetStorageKind.STACK -> TODO("stack invert")
+                    TargetStorageKind.ARRAY -> throw AssemblyError("missing codegen for in-place invert uword array")
+                    TargetStorageKind.REGISTER -> throw AssemblyError("missing codegen for reg invert")
+                    TargetStorageKind.STACK -> throw AssemblyError("missing codegen for stack invert")
                 }
             }
             else -> throw AssemblyError("invert of invalid type")
@@ -1833,9 +1863,9 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                             sta  ${target.asmVarname}""")
                     }
                     TargetStorageKind.MEMORY -> throw AssemblyError("can't in-place negate memory ubyte")
-                    TargetStorageKind.ARRAY -> TODO("in-place negate byte array")
-                    TargetStorageKind.REGISTER -> TODO("reg negate")
-                    TargetStorageKind.STACK -> TODO("stack negate")
+                    TargetStorageKind.ARRAY -> throw AssemblyError("missing codegen for in-place negate byte array")
+                    TargetStorageKind.REGISTER -> throw AssemblyError("missing codegen for reg negate")
+                    TargetStorageKind.STACK -> throw AssemblyError("missing codegen for stack negate")
                 }
             }
             DataType.WORD -> {
@@ -1850,10 +1880,10 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                             sbc  ${target.asmVarname}+1
                             sta  ${target.asmVarname}+1""")
                     }
-                    TargetStorageKind.ARRAY -> TODO("in-place negate word array")
+                    TargetStorageKind.ARRAY -> throw AssemblyError("missing codegen for in-place negate word array")
                     TargetStorageKind.MEMORY -> throw AssemblyError("no asm gen for word memory negate")
-                    TargetStorageKind.REGISTER -> TODO("reg negate")
-                    TargetStorageKind.STACK -> TODO("stack negate")
+                    TargetStorageKind.REGISTER -> throw AssemblyError("missing codegen for reg negate")
+                    TargetStorageKind.STACK -> throw AssemblyError("missing codegen for stack negate")
                 }
             }
             DataType.FLOAT -> {
@@ -1866,8 +1896,8 @@ internal class AugmentableAssignmentAsmGen(private val program: Program,
                             sta  ${target.asmVarname}+1
                         """)
                     }
-                    TargetStorageKind.ARRAY -> TODO("in-place negate float array")
-                    TargetStorageKind.STACK -> TODO("stack float negate")
+                    TargetStorageKind.ARRAY -> throw AssemblyError("missing codegen for in-place negate float array")
+                    TargetStorageKind.STACK -> throw AssemblyError("missing codegen for stack float negate")
                     else -> throw AssemblyError("weird target kind for float")
                 }
             }

compiler/src/prog8/optimizer/BinExprSplitter.kt

@@ -1,5 +1,6 @@
 package prog8.optimizer
 
+import prog8.ast.INameScope
 import prog8.ast.Node
 import prog8.ast.Program
 import prog8.ast.expressions.*
@@ -11,7 +12,6 @@ import prog8.compiler.target.ICompilationTarget
 
 
 internal class BinExprSplitter(private val program: Program, private val compTarget: ICompilationTarget) : 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...:
@@ -57,13 +57,14 @@ X =      BinExpr                                    X   =   LeftExpr
                 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)
+                if(binExpr.right.isSimple && !assignment.isAugmentable) {
+                    val firstAssign = Assignment(assignment.target.copy(), 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))
+                        IAstModification.ReplaceNode(binExpr, augExpr, assignment),
+                        IAstModification.InsertBefore(assignment, firstAssign, assignment.parent as INameScope)
+                    )
                 }
             }
 
@@ -75,9 +76,6 @@ X =      BinExpr                                    X   =   LeftExpr
         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, program: Program) =
             if (target.identifier!=null || target.memoryAddress!=null)
                 compTarget.isInRegularRAM(target, program)

compiler/src/prog8/optimizer/CallGraph.kt

@@ -1,139 +1,72 @@
 package prog8.optimizer
 
-import prog8.ast.INameScope
 import prog8.ast.Module
 import prog8.ast.Node
 import prog8.ast.Program
 import prog8.ast.base.DataType
-import prog8.ast.base.ParentSentinel
 import prog8.ast.base.Position
+import prog8.ast.base.VarDeclType
 import prog8.ast.expressions.AddressOf
 import prog8.ast.expressions.FunctionCall
 import prog8.ast.expressions.IdentifierReference
 import prog8.ast.statements.*
 import prog8.ast.walk.IAstVisitor
 import prog8.compiler.IErrorReporter
-import java.nio.file.Path
-
-private val alwaysKeepSubroutines = setOf(
-        Pair("main", "start")
-)
-
-private val asmJumpRx = Regex("""[\-+a-zA-Z0-9_ \t]+(jmp|jsr)[ \t]+(\S+).*""", RegexOption.IGNORE_CASE)
-private val asmRefRx = Regex("""[\-+a-zA-Z0-9_ \t]+(...)[ \t]+(\S+).*""", RegexOption.IGNORE_CASE)
 
 
-class CallGraph(private val program: Program, private val asmFileLoader: (filename: String, source: Path)->String) : IAstVisitor {
+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<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
-    val usedSymbols = mutableSetOf<Statement>()
+    val imports = mutableMapOf<Module, Set<Module>>().withDefault { setOf() }
+    val importedBy = mutableMapOf<Module, Set<Module>>().withDefault { setOf() }
+    val calls = mutableMapOf<Subroutine, Set<Subroutine>>().withDefault { setOf() }
+    val calledBy = mutableMapOf<Subroutine, Set<Node>>().withDefault { setOf() }
+    private val allIdentifiersAndTargets = mutableMapOf<Pair<IdentifierReference, Position>, Statement>()
+    private val allAssemblyNodes = mutableListOf<InlineAssembly>()
 
     init {
         visit(program)
     }
 
-    fun forAllSubroutines(scope: INameScope, sub: (s: Subroutine) -> Unit) {
-        fun findSubs(scope: INameScope) {
-            scope.statements.forEach {
-                if (it is Subroutine)
-                    sub(it)
-                if (it is INameScope)
-                    findSubs(it)
-            }
-        }
-        findSubs(scope)
+    private val usedSubroutines: Set<Subroutine> by lazy {
+        calledBy.keys + program.entrypoint()
     }
 
-    override fun visit(program: Program) {
-        super.visit(program)
+    private val usedBlocks: Set<Block> by lazy {
+        val blocksFromSubroutines = usedSubroutines.map { it.definingBlock() }
+        val blocksFromLibraries = program.allBlocks().filter { it.isInLibrary }
+        val used = mutableSetOf<Block>()
 
-        program.modules.forEach {
-            it.importedBy.clear()
-            it.imports.clear()
-
-            it.importedBy.addAll(importedBy.getValue(it))
-            it.imports.addAll(imports.getValue(it))
+        allIdentifiersAndTargets.forEach {
+            if(it.key.first.definingBlock() in blocksFromSubroutines) {
+                val target = it.value.definingBlock()
+                used.add(target)
+            }
         }
 
-        val rootmodule = program.modules.first()
-        rootmodule.importedBy.add(rootmodule)       // don't discard root module
+        used + blocksFromLibraries + program.entrypoint().definingBlock()
     }
 
-    override fun visit(block: Block) {
-        if (block.definingModule().isLibraryModule) {
-            // make sure the block is not removed
-            addNodeAndParentScopes(block)
-        }
-
-        super.visit(block)
+    private val usedModules: Set<Module> by lazy {
+        usedBlocks.map { it.definingModule() }.toSet()
     }
 
     override fun visit(directive: Directive) {
         val thisModule = directive.definingModule()
         if (directive.directive == "%import") {
             val importedModule: Module = program.modules.single { it.name == directive.args[0].name }
-            imports[thisModule] = imports.getValue(thisModule).plus(importedModule)
-            importedBy[importedModule] = importedBy.getValue(importedModule).plus(thisModule)
-        } else if (directive.directive == "%asminclude") {
-            val asm = asmFileLoader(directive.args[0].str!!, thisModule.source)
-            val scope = directive.definingSubroutine()
-            if(scope!=null) {
-                scanAssemblyCode(asm, directive, scope)
-            }
+            imports[thisModule] = imports.getValue(thisModule) + importedModule
+            importedBy[importedModule] = importedBy.getValue(importedModule) + thisModule
         }
 
         super.visit(directive)
     }
 
-    override fun visit(identifier: IdentifierReference) {
-        // track symbol usage
-        val target = identifier.targetStatement(program)
-        if (target != null) {
-            addNodeAndParentScopes(target)
-        }
-        super.visit(identifier)
-    }
-
-    private fun addNodeAndParentScopes(stmt: Statement) {
-        usedSymbols.add(stmt)
-        var node: Node = stmt
-        do {
-            if (node is INameScope && node is Statement) {
-                usedSymbols.add(node)
-            }
-            node = node.parent
-        } while (node !is Module && node !is ParentSentinel)
-    }
-
-    override fun visit(subroutine: Subroutine) {
-        if (Pair(subroutine.definingScope().name, subroutine.name) in alwaysKeepSubroutines
-                || subroutine.definingModule().isLibraryModule) {
-            // make sure the entrypoint is mentioned in the used symbols
-            addNodeAndParentScopes(subroutine)
-        }
-        super.visit(subroutine)
-    }
-
-    override fun visit(decl: VarDecl) {
-        if (decl.autogeneratedDontRemove || decl.datatype==DataType.STRUCT)
-            addNodeAndParentScopes(decl)
-        else if(decl.parent is Block && decl.definingModule().isLibraryModule)
-            addNodeAndParentScopes(decl)
-
-        super.visit(decl)
-    }
-
     override fun visit(functionCall: FunctionCall) {
         val otherSub = functionCall.target.targetSubroutine(program)
         if (otherSub != null) {
             functionCall.definingSubroutine()?.let { thisSub ->
-                calls[thisSub] = calls.getValue(thisSub).plus(otherSub)
-                calledBy[otherSub] = calledBy.getValue(otherSub).plus(functionCall)
+                calls[thisSub] = calls.getValue(thisSub) + otherSub
+                calledBy[otherSub] = calledBy.getValue(otherSub) + functionCall
             }
         }
         super.visit(functionCall)
@@ -143,8 +76,8 @@ class CallGraph(private val program: Program, private val asmFileLoader: (filena
         val otherSub = functionCallStatement.target.targetSubroutine(program)
         if (otherSub != null) {
             functionCallStatement.definingSubroutine()?.let { thisSub ->
-                calls[thisSub] = calls.getValue(thisSub).plus(otherSub)
-                calledBy[otherSub] = calledBy.getValue(otherSub).plus(functionCallStatement)
+                calls[thisSub] = calls.getValue(thisSub) + otherSub
+                calledBy[otherSub] = calledBy.getValue(otherSub) + functionCallStatement
             }
         }
         super.visit(functionCallStatement)
@@ -154,8 +87,8 @@ class CallGraph(private val program: Program, private val asmFileLoader: (filena
         val otherSub = addressOf.identifier.targetSubroutine(program)
         if(otherSub!=null) {
             addressOf.definingSubroutine()?.let { thisSub ->
-                calls[thisSub] = calls.getValue(thisSub).plus(otherSub)
-                calledBy[otherSub] = calledBy.getValue(otherSub).plus(thisSub)
+                calls[thisSub] = calls.getValue(thisSub) + otherSub
+                calledBy[otherSub] = calledBy.getValue(otherSub) + thisSub
             }
         }
         super.visit(addressOf)
@@ -165,63 +98,19 @@ class CallGraph(private val program: Program, private val asmFileLoader: (filena
         val otherSub = jump.identifier?.targetSubroutine(program)
         if (otherSub != null) {
             jump.definingSubroutine()?.let { thisSub ->
-                calls[thisSub] = calls.getValue(thisSub).plus(otherSub)
-                calledBy[otherSub] = calledBy.getValue(otherSub).plus(jump)
+                calls[thisSub] = calls.getValue(thisSub) + otherSub
+                calledBy[otherSub] = calledBy.getValue(otherSub) + jump
             }
         }
         super.visit(jump)
     }
 
-    override fun visit(structDecl: StructDecl) {
-        usedSymbols.add(structDecl)
-        usedSymbols.addAll(structDecl.statements)
+    override fun visit(identifier: IdentifierReference) {
+        allIdentifiersAndTargets[Pair(identifier, identifier.position)] = identifier.targetStatement(program)!!
     }
 
     override fun visit(inlineAssembly: InlineAssembly) {
-        // parse inline asm for subroutine calls (jmp, jsr)
-        val scope = inlineAssembly.definingSubroutine()
-        scanAssemblyCode(inlineAssembly.assembly, inlineAssembly, scope)
-        super.visit(inlineAssembly)
-    }
-
-    private fun scanAssemblyCode(asm: String, context: Statement, scope: Subroutine?) {
-        asm.lines().forEach { line ->
-            val matches = asmJumpRx.matchEntire(line)
-            if (matches != null) {
-                val jumptarget = matches.groups[2]?.value
-                if (jumptarget != null && (jumptarget[0].isLetter() || jumptarget[0] == '_')) {
-                    val node = program.namespace.lookup(jumptarget.split('.'), context)
-                    if (node is Subroutine) {
-                        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) {
-                            if(scope!=null)
-                                calls[scope] = calls.getValue(scope).plus(node2)
-                            calledBy[node2] = calledBy.getValue(node2).plus(context)
-                        }
-                    }
-                }
-            } else {
-                val matches2 = asmRefRx.matchEntire(line)
-                if (matches2 != null) {
-                    val target = matches2.groups[2]?.value
-                    if (target != null && (target[0].isLetter() || target[0] == '_')) {
-                        if (target.contains('.')) {
-                            val node = program.namespace.lookup(listOf(target.substringBefore('.')), context)
-                            if (node is Subroutine) {
-                                if(scope!=null)
-                                    calls[scope] = calls.getValue(scope).plus(node)
-                                calledBy[node] = calledBy.getValue(node).plus(context)
-                            }
-                        }
-                    }
-                }
-            }
-        }
+        allAssemblyNodes.add(inlineAssembly)
     }
 
     fun checkRecursiveCalls(errors: IErrorReporter) {
@@ -274,4 +163,39 @@ class CallGraph(private val program: Program, private val asmFileLoader: (filena
         recStack[sub] = false
         return false
     }
+
+    fun unused(module: Module) = module !in usedModules
+
+    fun unused(sub: Subroutine): Boolean {
+        return sub !in usedSubroutines && !nameInAssemblyCode(sub.name)
+    }
+
+    fun unused(block: Block): Boolean {
+        return block !in usedBlocks && !nameInAssemblyCode(block.name)
+    }
+
+    fun unused(decl: VarDecl): Boolean {
+        if(decl.type!=VarDeclType.VAR || decl.datatype==DataType.STRUCT || decl.autogeneratedDontRemove || decl.parent is StructDecl)
+            return false
+
+        if(decl.definingBlock() !in usedBlocks)
+            return false
+
+        val allReferencedVardecls = allIdentifiersAndTargets.filter { it.value is VarDecl }.map { it.value }.toSet()
+        return decl !in allReferencedVardecls && !nameInAssemblyCode(decl.name)
+    }
+
+    private fun nameInAssemblyCode(name: String) = allAssemblyNodes.any { it.assembly.contains(name) }
+
+    inline fun unused(label: Label) = false   // just always output labels
+
+    fun unused(stmt: ISymbolStatement): Boolean {
+        return when(stmt) {
+            is Subroutine -> unused(stmt)
+            is Block -> unused(stmt)
+            is VarDecl -> unused(stmt)
+            is Label -> false   // just always output labels
+            else -> false
+        }
+    }
 }

compiler/src/prog8/optimizer/ConstExprEvaluator.kt

@@ -9,6 +9,7 @@ import kotlin.math.pow
 class ConstExprEvaluator {
 
     fun evaluate(left: NumericLiteralValue, operator: String, right: NumericLiteralValue): Expression {
+        try {
             return when(operator) {
                 "+" -> plus(left, right)
                 "-" -> minus(left, right)
@@ -32,6 +33,9 @@ class ConstExprEvaluator {
                 ">>" -> shiftedright(left, right)
                 else -> throw FatalAstException("const evaluation for invalid operator $operator")
             }
+        } catch (ax: FatalAstException) {
+            throw ExpressionError(ax.message, left.position)
+        }
     }
 
     private fun shiftedright(left: NumericLiteralValue, amount: NumericLiteralValue): Expression {

compiler/src/prog8/optimizer/ConstantFoldingOptimizer.kt

@@ -14,7 +14,6 @@ import kotlin.math.pow
 
 
 internal class ConstantFoldingOptimizer(private val program: Program, private val compTarget: ICompilationTarget) : AstWalker() {
-    private val noModifications = emptyList<IAstModification>()
 
     override fun before(memread: DirectMemoryRead, parent: Node): Iterable<IAstModification> {
         // @( &thing )  -->  thing

compiler/src/prog8/optimizer/ConstantIdentifierReplacer.kt

@@ -15,7 +15,6 @@ import prog8.compiler.target.ICompilationTarget
 
 // Fix up the literal value's type to match that of the vardecl
 internal class VarConstantValueTypeAdjuster(private val program: Program, private val errors: IErrorReporter) : AstWalker() {
-    private val noModifications = emptyList<IAstModification>()
 
     override fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> {
         try {
@@ -40,7 +39,6 @@ internal class VarConstantValueTypeAdjuster(private val program: Program, privat
 // 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: IErrorReporter, private val compTarget: ICompilationTarget) : 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
@@ -75,7 +73,7 @@ internal class ConstantIdentifierReplacer(private val program: Program, private
     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) {
+        if(decl.value?.referencesIdentifier(decl.name) == true || decl.arraysize?.indexExpr?.referencesIdentifier(decl.name) == true) {
             errors.err("recursive var declaration", decl.position)
             return noModifications
         }
@@ -93,19 +91,6 @@ internal class ConstantIdentifierReplacer(private val program: Program, private
                                 decl
                         ))
                     }
-                } else if(arraysize.constIndex()==null) {
-                    // see if we can calculate the size from other fields
-                    try {
-                        val cval =  arraysize.indexVar?.constValue(program) ?: arraysize.origExpression?.constValue(program)
-                        if (cval != null) {
-                            arraysize.indexVar = null
-                            arraysize.origExpression = null
-                            arraysize.indexNum = cval
-                        }
-                    } catch (x: UndefinedSymbolError) {
-                        errors.err(x.message, x.position)
-                        return noModifications
-                    }
                 }
             }
 

compiler/src/prog8/optimizer/ExpressionSimplifier.kt

@@ -25,7 +25,6 @@ import kotlin.math.pow
 internal class ExpressionSimplifier(private val program: Program) : AstWalker() {
     private val powersOfTwo = (1..16).map { (2.0).pow(it) }.toSet()
     private val negativePowersOfTwo = powersOfTwo.map { -it }.toSet()
-    private val noModifications = emptyList<IAstModification>()
 
     override fun after(typecast: TypecastExpression, parent: Node): Iterable<IAstModification> {
         val mods = mutableListOf<IAstModification>()
@@ -491,9 +490,9 @@ internal class ExpressionSimplifier(private val program: Program) : AstWalker()
                     if(!idt.isKnown)
                         throw FatalAstException("unknown dt")
                     return NumericLiteralValue(idt.typeOrElse(DataType.STRUCT), 0, expr.position)
-                } else if (cv == 2.0) {
+                } else if (cv in powersOfTwo) {
                     expr.operator = "&"
-                    expr.right = NumericLiteralValue.optimalInteger(1, expr.position)
+                    expr.right = NumericLiteralValue.optimalInteger(cv!!.toInt()-1, expr.position)
                     return null
                 }
             }

compiler/src/prog8/optimizer/Extensions.kt

@@ -4,31 +4,30 @@ import prog8.ast.IBuiltinFunctions
 import prog8.ast.Program
 import prog8.compiler.IErrorReporter
 import prog8.compiler.target.ICompilationTarget
-import java.nio.file.Path
 
 
 internal fun Program.constantFold(errors: IErrorReporter, compTarget: ICompilationTarget) {
     val valuetypefixer = VarConstantValueTypeAdjuster(this, errors)
     valuetypefixer.visit(this)
-    if(errors.isEmpty()) {
+    if(errors.noErrors()) {
         valuetypefixer.applyModifications()
 
         val replacer = ConstantIdentifierReplacer(this, errors, compTarget)
         replacer.visit(this)
-        if (errors.isEmpty()) {
+        if (errors.noErrors()) {
             replacer.applyModifications()
 
             valuetypefixer.visit(this)
-            if(errors.isEmpty()) {
+            if(errors.noErrors()) {
                 valuetypefixer.applyModifications()
 
                 val optimizer = ConstantFoldingOptimizer(this, compTarget)
                 optimizer.visit(this)
-                while (errors.isEmpty() && optimizer.applyModifications() > 0) {
+                while (errors.noErrors() && optimizer.applyModifications() > 0) {
                     optimizer.visit(this)
                 }
 
-                if (errors.isEmpty()) {
+                if (errors.noErrors()) {
                     replacer.visit(this)
                     replacer.applyModifications()
                 }
@@ -36,16 +35,15 @@ internal fun Program.constantFold(errors: IErrorReporter, compTarget: ICompilati
         }
     }
 
-    if(errors.isEmpty())
+    if(errors.noErrors())
         modules.forEach { it.linkParents(namespace) }   // re-link in final configuration
 }
 
 
 internal fun Program.optimizeStatements(errors: IErrorReporter,
                                         functions: IBuiltinFunctions,
-                                        compTarget: ICompilationTarget,
-                                        asmFileLoader: (filename: String, source: Path)->String): Int {
-    val optimizer = StatementOptimizer(this, errors, functions, compTarget, asmFileLoader)
+                                        compTarget: ICompilationTarget): Int {
+    val optimizer = StatementOptimizer(this, errors, functions, compTarget)
     optimizer.visit(this)
     val optimizationCount = optimizer.applyModifications()
 

compiler/src/prog8/optimizer/StatementOptimizer.kt

@@ -12,67 +12,25 @@ import prog8.ast.walk.IAstModification
 import prog8.ast.walk.IAstVisitor
 import prog8.compiler.IErrorReporter
 import prog8.compiler.target.ICompilationTarget
-import java.nio.file.Path
 import kotlin.math.floor
 
+internal const val retvarName = "prog8_retval"
+
 
 internal class StatementOptimizer(private val program: Program,
                                   private val errors: IErrorReporter,
                                   private val functions: IBuiltinFunctions,
-                                  private val compTarget: ICompilationTarget,
-                                  asmFileLoader: (filename: String, source: Path)->String
-) : AstWalker() {
+                                  private val compTarget: ICompilationTarget) : AstWalker() {
 
-    private val noModifications = emptyList<IAstModification>()
-    private val callgraph = CallGraph(program, asmFileLoader)
+    private val subsThatNeedReturnVariable = mutableSetOf<Triple<INameScope, DataType, Position>>()
 
-    override fun after(block: Block, parent: Node): Iterable<IAstModification> {
-        if("force_output" !in block.options()) {
-            if (block.containsNoCodeNorVars()) {
-                if(block.name != program.internedStringsModuleName)
-                    errors.warn("removing empty block '${block.name}'", block.position)
-                return listOf(IAstModification.Remove(block, parent as INameScope))
-            }
-
-            if (block !in callgraph.usedSymbols) {
-                errors.warn("removing unused block '${block.name}'", block.position)
-                return listOf(IAstModification.Remove(block, parent as INameScope))
-            }
-        }
-        return noModifications
-    }
 
     override fun after(subroutine: Subroutine, parent: Node): Iterable<IAstModification> {
-        val forceOutput = "force_output" in subroutine.definingBlock().options()
-        if(subroutine.asmAddress==null && !forceOutput) {
-            if(subroutine.containsNoCodeNorVars() && !subroutine.inline) {
-                errors.warn("removing empty subroutine '${subroutine.name}'", subroutine.position)
-                val removals = callgraph.calledBy.getValue(subroutine).map {
-                    IAstModification.Remove(it, it.definingScope())
-                }.toMutableList()
-                removals += IAstModification.Remove(subroutine, subroutine.definingScope())
-                return removals
+        for(returnvar in subsThatNeedReturnVariable) {
+            val decl = VarDecl(VarDeclType.VAR, returnvar.second, ZeropageWish.DONTCARE, null, retvarName, null, null, false, true, returnvar.third)
+            returnvar.first.statements.add(0, decl)
         }
-        }
-
-        if(subroutine !in callgraph.usedSymbols && !forceOutput) {
-            if(!subroutine.isAsmSubroutine)
-                errors.warn("removing unused subroutine '${subroutine.name}'", subroutine.position)
-            return listOf(IAstModification.Remove(subroutine, subroutine.definingScope()))
-        }
-
-        return noModifications
-    }
-
-    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, decl.definingScope()))
-        }
-
+        subsThatNeedReturnVariable.clear()
         return noModifications
     }
 
@@ -393,7 +351,7 @@ internal class StatementOptimizer(private val program: Program,
                                 // 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(rightCv.toInt()) {
-                                    incs.statements.add(PostIncrDecr(assignment.target, "++", assignment.position))
+                                    incs.statements.add(PostIncrDecr(assignment.target.copy(), "++", assignment.position))
                                 }
                                 return listOf(IAstModification.ReplaceNode(assignment, incs, parent))
                             }
@@ -407,7 +365,7 @@ internal class StatementOptimizer(private val program: Program,
                                 // 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(rightCv.toInt()) {
-                                    decs.statements.add(PostIncrDecr(assignment.target, "--", assignment.position))
+                                    decs.statements.add(PostIncrDecr(assignment.target.copy(), "--", assignment.position))
                                 }
                                 return listOf(IAstModification.ReplaceNode(assignment, decs, parent))
                             }
@@ -435,21 +393,17 @@ internal class StatementOptimizer(private val program: Program,
     }
 
     override fun after(returnStmt: Return, parent: Node): Iterable<IAstModification> {
-        fun returnViaIntermediary(value: Expression): Iterable<IAstModification>? {
-            val returnDt = returnStmt.definingSubroutine()!!.returntypes.single()
+        fun returnViaIntermediaryVar(value: Expression): Iterable<IAstModification>? {
+            val subr = returnStmt.definingSubroutine()!!
+            val returnDt = subr.returntypes.single()
             if (returnDt in IntegerDatatypes) {
-                // first assign to intermediary, then return that register
-                val returnValueIntermediary =
-                    when(returnDt) {
-                        DataType.UBYTE -> IdentifierReference(listOf("prog8_lib", "retval_interm_ub"), returnStmt.position)
-                        DataType.BYTE -> IdentifierReference(listOf("prog8_lib", "retval_interm_b"), returnStmt.position)
-                        DataType.UWORD -> IdentifierReference(listOf("prog8_lib", "retval_interm_uw"), returnStmt.position)
-                        DataType.WORD -> IdentifierReference(listOf("prog8_lib", "retval_interm_w"), returnStmt.position)
-                        else -> throw FatalAstException("weird return dt")
-                    }
-                val tgt = AssignTarget(returnValueIntermediary, null, null, returnStmt.position)
+                // first assign to intermediary variable, then return that
+                subsThatNeedReturnVariable.add(Triple(subr, returnDt, returnStmt.position))
+                val returnValueIntermediary1 = IdentifierReference(listOf(retvarName), returnStmt.position)
+                val returnValueIntermediary2 = IdentifierReference(listOf(retvarName), returnStmt.position)
+                val tgt = AssignTarget(returnValueIntermediary1, null, null, returnStmt.position)
                 val assign = Assignment(tgt, value, returnStmt.position)
-                val returnReplacement = Return(returnValueIntermediary, returnStmt.position)
+                val returnReplacement = Return(returnValueIntermediary2, returnStmt.position)
                 return listOf(
                     IAstModification.InsertBefore(returnStmt, assign, parent as INameScope),
                     IAstModification.ReplaceNode(returnStmt, returnReplacement, parent)
@@ -460,12 +414,12 @@ internal class StatementOptimizer(private val program: Program,
 
         when(returnStmt.value) {
             is PrefixExpression -> {
-                val mod = returnViaIntermediary(returnStmt.value!!)
+                val mod = returnViaIntermediaryVar(returnStmt.value!!)
                 if(mod!=null)
                     return mod
             }
             is BinaryExpression -> {
-                val mod = returnViaIntermediary(returnStmt.value!!)
+                val mod = returnViaIntermediaryVar(returnStmt.value!!)
                 if(mod!=null)
                     return mod
             }

compiler/src/prog8/optimizer/SubroutineInliner.kt

@@ -0,0 +1,96 @@
+package prog8.optimizer
+
+import prog8.ast.IFunctionCall
+import prog8.ast.Node
+import prog8.ast.Program
+import prog8.ast.base.Position
+import prog8.ast.expressions.FunctionCall
+import prog8.ast.expressions.IdentifierReference
+import prog8.ast.statements.*
+import prog8.ast.walk.AstWalker
+import prog8.ast.walk.IAstModification
+import prog8.compiler.CompilationOptions
+import prog8.compiler.IErrorReporter
+
+
+internal class SubroutineInliner(private val program: Program, val errors: IErrorReporter, private val compilerOptions: CompilationOptions): AstWalker() {
+    private var callsToInlinedSubroutines = mutableListOf<Pair<IFunctionCall, Node>>()
+
+    fun fixCallsToInlinedSubroutines() {
+        for((call, parent) in callsToInlinedSubroutines) {
+            val sub = call.target.targetSubroutine(program)!!
+            val intermediateReturnValueVar = sub.statements.filterIsInstance<VarDecl>().singleOrNull { it.name.endsWith(retvarName) }
+            if(intermediateReturnValueVar!=null) {
+                val scope = parent.definingScope()
+                if(!scope.statements.filterIsInstance<VarDecl>().any { it.name==intermediateReturnValueVar.name}) {
+                    val decl = intermediateReturnValueVar.copy()
+                    scope.statements.add(0, decl)
+                    decl.linkParents(scope as Node)
+                }
+            }
+        }
+    }
+
+    override fun after(subroutine: Subroutine, parent: Node): Iterable<IAstModification> {
+        return if(compilerOptions.optimize && subroutine.inline && !subroutine.isAsmSubroutine)
+            annotateInlinedSubroutineIdentifiers(subroutine)
+        else
+            noModifications
+    }
+
+    override fun after(functionCallStatement: FunctionCallStatement, parent: Node): Iterable<IAstModification> {
+        return after(functionCallStatement as IFunctionCall, parent, functionCallStatement.position)
+    }
+
+    override fun after(functionCall: FunctionCall, parent: Node): Iterable<IAstModification> {
+        return after(functionCall as IFunctionCall, parent, functionCall.position)
+    }
+
+    private fun after(functionCall: IFunctionCall, parent: Node, position: Position): Iterable<IAstModification> {
+        val sub = functionCall.target.targetSubroutine(program)
+        if(sub != null && compilerOptions.optimize && sub.inline && !sub.isAsmSubroutine)
+            callsToInlinedSubroutines.add(Pair(functionCall, parent))
+
+        return noModifications
+    }
+
+    private fun annotateInlinedSubroutineIdentifiers(sub: Subroutine): List<IAstModification> {
+        // this adds name prefixes to the identifiers used in the subroutine,
+        // so that the statements can be inlined (=copied) in the call site and still reference
+        // the correct symbols as seen from the scope of the subroutine.
+
+        class Annotator: AstWalker() {
+            var numReturns=0
+
+            override fun before(identifier: IdentifierReference, parent: Node): Iterable<IAstModification> {
+                val stmt = identifier.targetStatement(program)!!
+                if(stmt is BuiltinFunctionStatementPlaceholder)
+                    return noModifications
+
+                val prefixed = stmt.makeScopedName(identifier.nameInSource.last()).split('.')
+                val withPrefix = IdentifierReference(prefixed, identifier.position)
+                return listOf(IAstModification.ReplaceNode(identifier, withPrefix, parent))
+            }
+
+            override fun before(returnStmt: Return, parent: Node): Iterable<IAstModification> {
+                numReturns++
+                if(parent !== sub || sub.indexOfChild(returnStmt)<sub.statements.size-1)
+                    errors.err("return statement must be the very last statement in the inlined subroutine", sub.position)
+                return noModifications
+            }
+
+            fun theModifications(): List<IAstModification> {
+                return this.modifications.map { it.first }.toList()
+            }
+        }
+
+        val annotator = Annotator()
+        sub.accept(annotator, sub.parent)
+        if(annotator.numReturns>1) {
+            errors.err("inlined subroutine can only have one return statement", sub.position)
+            return noModifications
+        }
+        return annotator.theModifications()
+    }
+
+}

compiler/src/prog8/optimizer/UnusedCodeRemover.kt

@@ -1,8 +1,10 @@
 package prog8.optimizer
 
 import prog8.ast.INameScope
+import prog8.ast.Module
 import prog8.ast.Node
 import prog8.ast.Program
+import prog8.ast.base.VarDeclType
 import prog8.ast.expressions.BinaryExpression
 import prog8.ast.expressions.FunctionCall
 import prog8.ast.expressions.PrefixExpression
@@ -12,43 +14,20 @@ import prog8.ast.walk.AstWalker
 import prog8.ast.walk.IAstModification
 import prog8.compiler.IErrorReporter
 import prog8.compiler.target.ICompilationTarget
-import java.nio.file.Path
 
 
 internal class UnusedCodeRemover(private val program: Program,
                                  private val errors: IErrorReporter,
-                                 private val compTarget: ICompilationTarget,
-                                 private val asmFileLoader: (filename: String, source: Path)->String): AstWalker() {
+                                 private val compTarget: ICompilationTarget): AstWalker() {
 
-    override fun before(program: Program, parent: Node): Iterable<IAstModification> {
-        val callgraph = CallGraph(program, asmFileLoader)
-        val removals = mutableListOf<IAstModification>()
+    private val callgraph = CallGraph(program)
 
-        // remove all subroutines that aren't called, or are empty
-        val entrypoint = program.entrypoint()
-        program.modules.forEach {
-            callgraph.forAllSubroutines(it) { sub ->
-                val forceOutput = "force_output" in sub.definingBlock().options()
-                if (sub !== entrypoint && !forceOutput && !sub.isAsmSubroutine && (callgraph.calledBy[sub].isNullOrEmpty() || sub.containsNoCodeNorVars())) {
-                    removals.add(IAstModification.Remove(sub, sub.definingScope()))
+    override fun before(module: Module, parent: Node): Iterable<IAstModification> {
+        return if (!module.isLibraryModule && (module.containsNoCodeNorVars() || callgraph.unused(module)))
+            listOf(IAstModification.Remove(module, module.definingScope()))
+        else
+            noModifications
     }
-            }
-        }
-
-        program.modules.flatMap { it.statements }.filterIsInstance<Block>().forEach { block ->
-            if (block.containsNoCodeNorVars() && "force_output" !in block.options())
-                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.definingScope()))
-        }
-
-        return removals
-    }
-
 
     override fun before(breakStmt: Break, parent: Node): Iterable<IAstModification> {
         reportUnreachable(breakStmt, parent as INameScope)
@@ -84,15 +63,58 @@ internal class UnusedCodeRemover(private val program: Program,
     }
 
     override fun after(block: Block, parent: Node): Iterable<IAstModification> {
+        if("force_output" !in block.options()) {
+            if (block.containsNoCodeNorVars()) {
+                if(block.name != program.internedStringsModuleName)
+                    errors.warn("removing unused block '${block.name}'", block.position)
+                return listOf(IAstModification.Remove(block, parent as INameScope))
+            }
+            if(callgraph.unused(block)) {
+                errors.warn("removing unused block '${block.name}'", block.position)
+                return listOf(IAstModification.Remove(block, parent as INameScope))
+            }
+        }
+
         val removeDoubleAssignments = deduplicateAssignments(block.statements)
         return removeDoubleAssignments.map { IAstModification.Remove(it, block) }
     }
 
     override fun after(subroutine: Subroutine, parent: Node): Iterable<IAstModification> {
+        val forceOutput = "force_output" in subroutine.definingBlock().options()
+        if (subroutine !== program.entrypoint() && !forceOutput && !subroutine.inline && !subroutine.isAsmSubroutine) {
+            if(callgraph.unused(subroutine)) {
+                if(!subroutine.definingModule().isLibraryModule)
+                    errors.warn("removing unused subroutine '${subroutine.name}'", subroutine.position)
+                return listOf(IAstModification.Remove(subroutine, subroutine.definingScope()))
+            }
+            if(subroutine.containsNoCodeNorVars()) {
+                if(!subroutine.definingModule().isLibraryModule)
+                    errors.warn("removing empty subroutine '${subroutine.name}'", subroutine.position)
+                val removals = mutableListOf(IAstModification.Remove(subroutine, subroutine.definingScope()))
+                callgraph.calledBy[subroutine]?.let {
+                    for(node in it)
+                        removals.add(IAstModification.Remove(node, node.definingScope()))
+                }
+                return removals
+            }
+        }
+
         val removeDoubleAssignments = deduplicateAssignments(subroutine.statements)
         return removeDoubleAssignments.map { IAstModification.Remove(it, subroutine) }
     }
 
+    override fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> {
+        val forceOutput = "force_output" in decl.definingBlock().options()
+        if(!forceOutput && !decl.autogeneratedDontRemove && callgraph.unused(decl)) {
+            if(decl.type == VarDeclType.VAR && !decl.definingBlock().isInLibrary)
+                errors.warn("removing unused variable '${decl.name}'", decl.position)
+
+            return listOf(IAstModification.Remove(decl, decl.definingScope()))
+        }
+
+        return noModifications
+    }
+
     private fun deduplicateAssignments(statements: List<Statement>): List<Assignment> {
         // removes 'duplicate' assignments that assign the same target directly after another
         val linesToRemove = mutableListOf<Assignment>()

compiler/test/UnitTests.kt

@@ -19,7 +19,7 @@ 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.cbm.Petscii
 import prog8.compiler.target.cx16.CX16MachineDefinition.CX16Zeropage
 import java.io.CharConversionException
 import java.nio.file.Path
@@ -191,7 +191,7 @@ class TestC64Zeropage {
         val zp1 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.BASICSAFE, emptyList(), true, false, C64Target))
         assertEquals(18, zp1.available())
         val zp2 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FLOATSAFE, emptyList(), false, false, C64Target))
-        assertEquals(89, zp2.available())
+        assertEquals(85, zp2.available())
         val zp3 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.KERNALSAFE, emptyList(), false, false, C64Target))
         assertEquals(125, zp3.available())
         val zp4 = C64Zeropage(CompilationOptions(OutputType.RAW, LauncherType.NONE, ZeropageType.FULL, emptyList(), false, false, C64Target))
@@ -348,8 +348,8 @@ class TestPetscii {
                 listOf<Short>(72, 69, 76, 76, 79, 32, 0xd7, 0xcf, 0xd2, 0xcc, 0xc4, 32, 49, 50, 51, 32, 64, 33, 0x5c)))
         assertThat(Petscii.encodePetscii("\uf11a", true), equalTo(listOf<Short>(0x12)))   // reverse vid
         assertThat(Petscii.encodePetscii("✓", true), equalTo(listOf<Short>(0xfa)))
-        assertFailsWith<CharConversionException> { Petscii.encodePetscii("π", true) }
-        assertFailsWith<CharConversionException> { Petscii.encodePetscii("♥", true) }
+        assertThat("expect lowercase error fallback", Petscii.encodePetscii("π", true), equalTo(listOf<Short>(255)))
+        assertThat("expect lowercase error fallback", Petscii.encodePetscii("♥", true), equalTo(listOf<Short>(0xd3)))
 
         assertThat(Petscii.decodePetscii(listOf(72, 0xd7, 0x5c, 0xfa, 0x12), true), equalTo("hW£✓\uF11A"))
         assertFailsWith<ArrayIndexOutOfBoundsException> { Petscii.decodePetscii(listOf(-1), true) }
@@ -363,7 +363,7 @@ class TestPetscii {
         assertThat(Petscii.encodePetscii("\uf11a"), equalTo(listOf<Short>(0x12)))   // reverse vid
         assertThat(Petscii.encodePetscii("♥"), equalTo(listOf<Short>(0xd3)))
         assertThat(Petscii.encodePetscii("π"), equalTo(listOf<Short>(0xff)))
-        assertFailsWith<CharConversionException> { Petscii.encodePetscii("✓") }
+        assertThat("expecting fallback", Petscii.encodePetscii("✓"), equalTo(listOf<Short>(250)))
 
         assertThat(Petscii.decodePetscii(listOf(72, 0x5c, 0xd3, 0xff)), equalTo("H£♥π"))
         assertFailsWith<ArrayIndexOutOfBoundsException> { Petscii.decodePetscii(listOf(-1)) }
@@ -376,8 +376,8 @@ class TestPetscii {
                 listOf<Short>(0x08, 0x05, 0x0c, 0x0c, 0x0f, 0x20, 0x57, 0x4f, 0x52, 0x4c, 0x44, 0x20, 0x31, 0x32, 0x33, 0x20, 0x00, 0x21, 0x1c)
         ))
         assertThat(Petscii.encodeScreencode("✓", true), equalTo(listOf<Short>(0x7a)))
-        assertFailsWith<CharConversionException> { Petscii.encodeScreencode("♥", true) }
-        assertFailsWith<CharConversionException> { Petscii.encodeScreencode("π", true) }
+        assertThat("expect fallback", Petscii.encodeScreencode("♥", true), equalTo(listOf<Short>(83)))
+        assertThat("expect fallback", Petscii.encodeScreencode("π", true), equalTo(listOf<Short>(94)))
 
         assertThat(Petscii.decodeScreencode(listOf(0x08, 0x57, 0x1c, 0x7a), true), equalTo("hW£✓"))
         assertFailsWith<ArrayIndexOutOfBoundsException> { Petscii.decodeScreencode(listOf(-1), true) }
@@ -390,8 +390,9 @@ class TestPetscii {
                 listOf<Short>(0x17, 0x0f, 0x12, 0x0c, 0x04, 0x20, 0x31, 0x32, 0x33, 0x20, 0x00, 0x21, 0x1c)))
         assertThat(Petscii.encodeScreencode("♥"), equalTo(listOf<Short>(0x53)))
         assertThat(Petscii.encodeScreencode("π"), equalTo(listOf<Short>(0x5e)))
-        assertFailsWith<CharConversionException> { Petscii.encodeScreencode("✓") }
-        assertFailsWith<CharConversionException> { Petscii.encodeScreencode("hello") }
+        assertThat(Petscii.encodeScreencode("HELLO"), equalTo(listOf<Short>(8, 5, 12, 12, 15)))
+        assertThat("expecting fallback", Petscii.encodeScreencode("hello"), equalTo(listOf<Short>(8, 5, 12, 12, 15)))
+        assertThat("expecting fallback", Petscii.encodeScreencode("✓"), equalTo(listOf<Short>(122)))
 
         assertThat(Petscii.decodeScreencode(listOf(0x17, 0x1c, 0x53, 0x5e)), equalTo("W£♥π"))
         assertFailsWith<ArrayIndexOutOfBoundsException> { Petscii.decodeScreencode(listOf(-1)) }

compiler/test/arithmetic/Makefile

@@ -0,0 +1,13 @@
+.PHONY: all clean test
+
+all: test
+
+clean:
+	rm -f *.prg *.asm *.vice-*
+
+test: clean
+	p8compile -target cx16 *.p8 >/dev/null
+	for program in *.prg; do \
+		echo "RUNNING:" $$program ; \
+		x16emu -run -prg $$program >/dev/null ; \
+	done

compiler/test/comparisons/Makefile

@@ -0,0 +1,18 @@
+.PHONY: all clean test
+
+all: test
+
+clean:
+	rm -f *.prg *.asm *.vice-* test_*.p8
+
+test: clean generate test_prgs
+
+generate:
+	python make_tests.py
+	p8compile -noopt -target cx16 *.p8 >/dev/null
+
+test_prgs:
+	for program in *.prg; do \
+		echo "RUNNING:" $$program ; \
+		x16emu -run -prg $$program >/dev/null ; \
+	done

compiler/test/comparisons/make_tests.py

@@ -0,0 +1,508 @@
+# generates various Prog8 files with a huge amount of number comparion tests,
+# for all supported datatypes and all comparison operators.
+
+import sys
+
+index = 0
+
+
+def minmaxvalues(dt):
+    if dt == "ubyte":
+        return 0, 255
+    elif dt == "uword":
+        return 0, 65535
+    elif dt == "byte":
+        return -128, 127
+    elif dt == "word":
+        return -32768, 32767
+    elif dt == "float":
+        return -99999999, 99999999
+    else:
+        raise ValueError(dt)
+
+
+def gen_test(dt, comparison, left, right, expected):
+    global index
+    etxt = f"{left} {comparison} {right}"
+    if eval(etxt) != expected:
+        raise ValueError("invalid comparison: "+etxt+" for "+dt)
+    if expected:
+        stmt_ok = lambda ix: "num_successes++"
+        stmt_else = lambda ix: f"error({ix})"
+    else:
+        stmt_ok = lambda ix: f"error({ix})"
+        stmt_else = lambda ix: "num_successes++"
+
+    def c(number):
+        if dt not in ("byte", "ubyte"):
+            return f"({number} as {dt})"
+        return str(number)
+
+    print(
+f"""        left = {c(left)}
+        right = {c(right)}
+"""
+    )
+
+    # const <op> const
+    index += 1
+    print(
+f"""        ; test #{index}
+        if {c(left)} {comparison} {c(right)} {{
+            {stmt_ok(index)}
+        }} else {{
+            {stmt_else(index)}
+        }}
+""")
+    # const <op> var
+    index += 1
+    print(
+f"""        ; test #{index}
+        if {c(left)} {comparison} right {{
+            {stmt_ok(index)}
+        }} else {{
+            {stmt_else(index)}
+        }}
+""")
+    # const <op> expr
+    index += 1
+    print(
+f"""        ; test #{index}
+        if {c(left)} {comparison} right+zero {{
+            {stmt_ok(index)}
+        }} else {{
+            {stmt_else(index)}
+        }}
+""")
+    # var <op> const
+    index += 1
+    print(
+f"""        ; test #{index}
+        if left {comparison} {c(right)} {{
+            {stmt_ok(index)}
+        }} else {{
+            {stmt_else(index)}
+        }}
+""")
+    # var <op> var
+    index += 1
+    print(
+f"""        ; test #{index}
+        if left {comparison} right {{
+            {stmt_ok(index)}
+        }} else {{
+            {stmt_else(index)}
+        }}
+""")
+    # var <op> expr
+    index += 1
+    print(
+f"""        ; test #{index}
+        if left {comparison} right+zero {{
+            {stmt_ok(index)}
+        }} else {{
+            {stmt_else(index)}
+        }}
+""")
+    # expr <op> const
+    index += 1
+    print(
+f"""        ; test #{index}
+        if left+zero {comparison} {c(right)} {{
+            {stmt_ok(index)}
+        }} else {{
+            {stmt_else(index)}
+        }}
+""")
+    # expr <op> var
+    index += 1
+    print(
+f"""        ; test #{index}
+        if left+zero {comparison} right {{
+            {stmt_ok(index)}
+        }} else {{
+            {stmt_else(index)}
+        }}
+""")
+    # expr <op> expr
+    index += 1
+    print(
+f"""        ; test #{index}
+        if left+zero {comparison} right+zero {{
+            {stmt_ok(index)}
+        }} else {{
+            {stmt_else(index)}
+        }}
+""")
+
+
+def gen_comp_header(dt, operator):
+    print("        ; tests: ", dt, operator)
+    print("        comparison = \""+operator+"\"")
+    print("        txt.print(datatype)")
+    print("        txt.spc()")
+    print("        txt.print(comparison)")
+    print("        txt.nl()")
+
+
+def gen_comp_equal(dt):
+    minval, maxval = minmaxvalues(dt)
+    gen_comp_header(dt, "==")
+    gen_test(dt, "==", 0, 0, True)
+    gen_test(dt, "==", 0, 1, False)
+    gen_test(dt, "==", 100, 100, True)
+    gen_test(dt, "==", 100, 101, False)
+    if maxval >= 200:
+        gen_test(dt, "==", 200, 200, True)
+        gen_test(dt, "==", 200, 201, False)
+    if maxval >= 9999:
+        gen_test(dt, "==", 9999, 9999, True)
+        gen_test(dt, "==", 9999, 10000, False)
+        gen_test(dt, "==", 0x5000, 0x5000, True)
+        gen_test(dt, "==", 0x5000, 0x5001, False)
+        gen_test(dt, "==", 0x5000, 0x4fff, False)
+    if maxval >= 30000:
+        gen_test(dt, "==", 30000, 30000, True)
+        gen_test(dt, "==", 30000, 30001, False)
+    if maxval >= 40000:
+        gen_test(dt, "==", 0xf000, 0xf000, True)
+        gen_test(dt, "==", 0xf000, 0xf001, False)
+        gen_test(dt, "==", 0xf000, 0xffff, False)
+    if minval < 0:
+        gen_test(dt, "==", 0, -1, False)
+        gen_test(dt, "==", -100, -100, True)
+    if minval < -200:
+        gen_test(dt, "==", -200, -200, True)
+        gen_test(dt, "==", -200, -201, False)
+    if minval < -9999:
+        gen_test(dt, "==", -0x5000, -0x5000, True)
+        gen_test(dt, "==", -0x5000, -0x5001, False)
+        gen_test(dt, "==", -0x5000, -0x4fff, False)
+        gen_test(dt, "==", -9999, -9999, True)
+        gen_test(dt, "==", -9999, -10000, False)
+    gen_test(dt, "==", minval, minval, True)
+    gen_test(dt, "==", minval, minval+1, False)
+    gen_test(dt, "==", maxval, maxval, True)
+    gen_test(dt, "==", maxval, maxval-1, False)
+
+
+def gen_comp_notequal(dt):
+    minval, maxval = minmaxvalues(dt)
+    gen_comp_header(dt, "!=")
+    gen_test(dt, "!=", 0, 0, False)
+    gen_test(dt, "!=", 0, 1, True)
+    gen_test(dt, "!=", 100, 100, False)
+    gen_test(dt, "!=", 100, 101, True)
+    if maxval >= 200:
+        gen_test(dt, "!=", 200, 200, False)
+        gen_test(dt, "!=", 200, 201, True)
+    if maxval >= 9999:
+        gen_test(dt, "!=", 9999, 9999, False)
+        gen_test(dt, "!=", 9999, 10000, True)
+        gen_test(dt, "!=", 0x5000, 0x5000, False)
+        gen_test(dt, "!=", 0x5000, 0x5001, True)
+        gen_test(dt, "!=", 0x5000, 0x4fff, True)
+    if maxval >= 30000:
+        gen_test(dt, "!=", 30000, 30000, False)
+        gen_test(dt, "!=", 30000, 30001, True)
+    if maxval >= 40000:
+        gen_test(dt, "!=", 0xf000, 0xf000, False)
+        gen_test(dt, "!=", 0xf000, 0xf001, True)
+        gen_test(dt, "!=", 0xf000, 0xffff, True)
+    if minval < 0:
+        gen_test(dt, "!=", 0, -1, True)
+        gen_test(dt, "!=", -100, -100, False)
+    if minval < -200:
+        gen_test(dt, "!=", -200, -200, False)
+        gen_test(dt, "!=", -200, -201, True)
+    if minval < -9999:
+        gen_test(dt, "!=", -0x5000, -0x5000, False)
+        gen_test(dt, "!=", -0x5000, -0x5001, True)
+        gen_test(dt, "!=", -0x5000, -0x4fff, True)
+        gen_test(dt, "!=", -9999, -9999, False)
+        gen_test(dt, "!=", -9999, -10000, True)
+    gen_test(dt, "!=", minval, minval, False)
+    gen_test(dt, "!=", minval, minval+1, True)
+    gen_test(dt, "!=", maxval, maxval, False)
+    gen_test(dt, "!=", maxval, maxval-1, True)
+
+
+def gen_comp_less(dt):
+    minval, maxval = minmaxvalues(dt)
+    gen_comp_header(dt, "<")
+    gen_test(dt, "<", 0, 0, False)
+    gen_test(dt, "<", 0, 1, True)
+    gen_test(dt, "<", 100, 100, False)
+    gen_test(dt, "<", 100, 101, True)
+    gen_test(dt, "<", 100, 99, False)
+    if maxval >= 200:
+        gen_test(dt, "<", 200, 200, False)
+        gen_test(dt, "<", 200, 201, True)
+        gen_test(dt, "<", 200, 199, False)
+    if maxval >= 9999:
+        gen_test(dt, "<", 9999, 9999, False)
+        gen_test(dt, "<", 9999, 10000, True)
+        gen_test(dt, "<", 9999, 9998, False)
+        gen_test(dt, "<", 0x5000, 0x5000, False)
+        gen_test(dt, "<", 0x5000, 0x5001, True)
+        gen_test(dt, "<", 0x5000, 0x4fff, False)
+    if maxval >= 30000:
+        gen_test(dt, "<", 30000, 30000, False)
+        gen_test(dt, "<", 30000, 30001, True)
+        gen_test(dt, "<", 30000, 29999, False)
+    if maxval >= 40000:
+        gen_test(dt, "<", 0xf000, 0xf000, False)
+        gen_test(dt, "<", 0xf000, 0xf001, True)
+        gen_test(dt, "<", 0xf000, 0xefff, False)
+    if minval < 0:
+        gen_test(dt, "<", 0, -1, False)
+        gen_test(dt, "<", -100, -100, False)
+        gen_test(dt, "<", -100, -101, False)
+        gen_test(dt, "<", -100, -99, True)
+    if minval < -200:
+        gen_test(dt, "<", -200, -200, False)
+        gen_test(dt, "<", -200, -201, False)
+        gen_test(dt, "<", -200, -199, True)
+    if minval < -9999:
+        gen_test(dt, "<", -0x5000, -0x5000, False)
+        gen_test(dt, "<", -0x5000, -0x5001, False)
+        gen_test(dt, "<", -0x5000, -0x4fff, True)
+        gen_test(dt, "<", -9999, -9999, False)
+        gen_test(dt, "<", -9999, -10000, False)
+        gen_test(dt, "<", -9999, -9998, True)
+
+
+def gen_comp_greater(dt):
+    minval, maxval = minmaxvalues(dt)
+    gen_comp_header(dt, ">")
+    gen_test(dt, ">", 0, 0, False)
+    gen_test(dt, ">", 0, 1, False)
+    gen_test(dt, ">", 100, 100, False)
+    gen_test(dt, ">", 100, 101, False)
+    gen_test(dt, ">", 100, 99, True)
+    if maxval >= 200:
+        gen_test(dt, ">", 200, 200, False)
+        gen_test(dt, ">", 200, 201, False)
+        gen_test(dt, ">", 200, 199, True)
+    if maxval >= 9999:
+        gen_test(dt, ">", 9999, 9999, False)
+        gen_test(dt, ">", 9999, 10000, False)
+        gen_test(dt, ">", 9999, 9998, True)
+        gen_test(dt, ">", 0x5000, 0x5000, False)
+        gen_test(dt, ">", 0x5000, 0x5001, False)
+        gen_test(dt, ">", 0x5000, 0x4fff, True)
+    if maxval >= 30000:
+        gen_test(dt, ">", 30000, 30000, False)
+        gen_test(dt, ">", 30000, 30001, False)
+        gen_test(dt, ">", 30000, 29999, True)
+    if maxval >= 40000:
+        gen_test(dt, ">", 0xf000, 0xf000, False)
+        gen_test(dt, ">", 0xf000, 0xf001, False)
+        gen_test(dt, ">", 0xf000, 0xefff, True)
+    if minval < 0:
+        gen_test(dt, ">", 0, -1, True)
+        gen_test(dt, ">", -100, -100, False)
+        gen_test(dt, ">", -100, -101, True)
+        gen_test(dt, ">", -100, -99, False)
+    if minval < -200:
+        gen_test(dt, ">", -200, -200, False)
+        gen_test(dt, ">", -200, -201, True)
+        gen_test(dt, ">", -200, -199, False)
+    if minval < -9999:
+        gen_test(dt, ">", -0x5000, -0x5000, False)
+        gen_test(dt, ">", -0x5000, -0x5001, True)
+        gen_test(dt, ">", -0x5000, -0x4fff, False)
+        gen_test(dt, ">", -9999, -9999, False)
+        gen_test(dt, ">", -9999, -10000, True)
+        gen_test(dt, ">", -9999, -9998, False)
+
+
+def gen_comp_lessequal(dt):
+    minval, maxval = minmaxvalues(dt)
+    gen_comp_header(dt, "<=")
+    gen_test(dt, "<=", 0, 0, True)
+    gen_test(dt, "<=", 0, 1, True)
+    gen_test(dt, "<=", 100, 100, True)
+    gen_test(dt, "<=", 100, 101, True)
+    gen_test(dt, "<=", 100, 99, False)
+    if maxval >= 200:
+        gen_test(dt, "<=", 200, 200, True)
+        gen_test(dt, "<=", 200, 201, True)
+        gen_test(dt, "<=", 200, 199, False)
+    if maxval >= 9999:
+        gen_test(dt, "<=", 9999, 9999, True)
+        gen_test(dt, "<=", 9999, 10000, True)
+        gen_test(dt, "<=", 9999, 9998, False)
+        gen_test(dt, "<=", 0x5000, 0x5000, True)
+        gen_test(dt, "<=", 0x5000, 0x5001, True)
+        gen_test(dt, "<=", 0x5000, 0x4fff, False)
+    if maxval >= 30000:
+        gen_test(dt, "<=", 30000, 30000, True)
+        gen_test(dt, "<=", 30000, 30001, True)
+        gen_test(dt, "<=", 30000, 29999, False)
+    if maxval >= 40000:
+        gen_test(dt, "<=", 0xf000, 0xf000, True)
+        gen_test(dt, "<=", 0xf000, 0xf001, True)
+        gen_test(dt, "<=", 0xf000, 0xefff, False)
+    if minval < 0:
+        gen_test(dt, "<=", 0, -1, False)
+        gen_test(dt, "<=", -100, -100, True)
+        gen_test(dt, "<=", -100, -101, False)
+        gen_test(dt, "<=", -100, -99, True)
+    if minval < -200:
+        gen_test(dt, "<=", -200, -200, True)
+        gen_test(dt, "<=", -200, -201, False)
+        gen_test(dt, "<=", -200, -199, True)
+    if minval < -9999:
+        gen_test(dt, "<=", -0x5000, -0x5000, True)
+        gen_test(dt, "<=", -0x5000, -0x5001, False)
+        gen_test(dt, "<=", -0x5000, -0x4fff, True)
+        gen_test(dt, "<=", -9999, -9999, True)
+        gen_test(dt, "<=", -9999, -10000, False)
+        gen_test(dt, "<=", -9999, -9998, True)
+
+
+def gen_comp_greaterequal(dt):
+    minval, maxval = minmaxvalues(dt)
+    gen_comp_header(dt, ">=")
+    gen_test(dt, ">=", 0, 0, True)
+    gen_test(dt, ">=", 0, 1, False)
+    gen_test(dt, ">=", 100, 100, True)
+    gen_test(dt, ">=", 100, 101, False)
+    gen_test(dt, ">=", 100, 99, True)
+    if maxval >= 200:
+        gen_test(dt, ">=", 200, 200, True)
+        gen_test(dt, ">=", 200, 201, False)
+        gen_test(dt, ">=", 200, 199, True)
+    if maxval >= 9999:
+        gen_test(dt, ">=", 9999, 9999, True)
+        gen_test(dt, ">=", 9999, 10000, False)
+        gen_test(dt, ">=", 9999, 9998, True)
+        gen_test(dt, ">=", 0x5000, 0x5000, True)
+        gen_test(dt, ">=", 0x5000, 0x5001, False)
+        gen_test(dt, ">=", 0x5000, 0x4fff, True)
+    if maxval >= 30000:
+        gen_test(dt, ">=", 30000, 30000, True)
+        gen_test(dt, ">=", 30000, 30001, False)
+        gen_test(dt, ">=", 30000, 29999, True)
+    if maxval >= 40000:
+        gen_test(dt, ">=", 0xf000, 0xf000, True)
+        gen_test(dt, ">=", 0xf000, 0xf001, False)
+        gen_test(dt, ">=", 0xf000, 0xefff, True)
+    if minval < 0:
+        gen_test(dt, ">=", 0, -1, True)
+        gen_test(dt, ">=", -100, -100, True)
+        gen_test(dt, ">=", -100, -101, True)
+        gen_test(dt, ">=", -100, -99, False)
+    if minval < -200:
+        gen_test(dt, ">=", -200, -200, True)
+        gen_test(dt, ">=", -200, -201, True)
+        gen_test(dt, ">=", -200, -199, False)
+    if minval < -9999:
+        gen_test(dt, ">=", -0x5000, -0x5000, True)
+        gen_test(dt, ">=", -0x5000, -0x5001, True)
+        gen_test(dt, ">=", -0x5000, -0x4fff, False)
+        gen_test(dt, ">=", -9999, -9999, True)
+        gen_test(dt, ">=", -9999, -10000, True)
+        gen_test(dt, ">=", -9999, -9998, False)
+
+
+def generate_test_routine_equalsnotequals(dt):
+    print(f"""
+    sub test_comparisons() {{
+        {dt}  left
+        {dt}  right
+        {dt}  zero = 0
+""")
+    gen_comp_equal(dt)
+    gen_comp_notequal(dt)
+    print("    }")
+
+
+def generate_test_routine_lessgreater(dt):
+    print(f"""
+    sub test_comparisons() {{
+        {dt}  left
+        {dt}  right
+        {dt}  zero = 0
+""")
+    gen_comp_less(dt)
+    gen_comp_greater(dt)
+    print("    }")
+
+
+def generate_test_routine_lessequalsgreaterequals(dt):
+    print(f"""
+    sub test_comparisons() {{
+        {dt}  left
+        {dt}  right
+        {dt}  zero = 0
+""")
+    gen_comp_lessequal(dt)
+    gen_comp_greaterequal(dt)
+    print("    }")
+
+
+def generate(dt, operators):
+    global index
+    index = 0
+    print(f"""
+%import textio
+%import floats
+%import test_stack
+%zeropage basicsafe
+
+main {{
+    uword num_errors = 0
+    uword num_successes = 0
+    str datatype = "{dt}"
+    uword comparison
+
+    sub start() {{
+        test_comparisons()
+        print_results()
+        test_stack.test()
+    }}
+
+    sub error(uword index) {{
+        txt.print(" ! error in test ")
+        txt.print_uw(index)
+        txt.chrout(' ')
+        txt.print(datatype)
+        txt.chrout(' ')
+        txt.print(comparison)
+        txt.nl()
+        num_errors++
+    }}
+""")
+
+    if operators=="eq":
+        generate_test_routine_equalsnotequals(dt)
+    elif operators=="lt":
+        generate_test_routine_lessgreater(dt)
+    elif operators=="lteq":
+        generate_test_routine_lessequalsgreaterequals(dt)
+    else:
+        raise ValueError(operators)
+
+    print(f"""
+    sub print_results() {{
+        txt.nl()
+        txt.print("total {index}: ")
+        txt.print_uw(num_successes)
+        txt.print(" good, ")
+        txt.print_uw(num_errors)
+        txt.print(" errors!\\n")
+    }}
+}}
+""")
+
+
+if __name__ == '__main__':
+    for dt in ["ubyte", "uword", "byte", "word", "float"]:
+        sys.stdout = open(f"test_{dt}_eq.p8", "wt")
+        generate(dt, "eq")
+        sys.stdout = open(f"test_{dt}_lt.p8", "wt")
+        generate(dt, "lt")
+        sys.stdout = open(f"test_{dt}_lteq.p8", "wt")
+        generate(dt, "lteq")

compiler/test/performance/perf10.p8

@@ -396,14 +396,18 @@ galaxy10 {
     sub init(ubyte galaxy10num) {
         number = 1
         planet10.number = 255
-        seed = [base0, base1, base2]
+        seed[0] = base0
+        seed[1] = base1
+        seed[2] = base2
         repeat galaxy10num-1 {
             nextgalaxy10()
         }
     }
 
     sub nextgalaxy10() {
-        seed = [twist(seed[0]), twist(seed[1]), twist(seed[2])]
+        seed[0] = twist(seed[0])
+        seed[1] = twist(seed[1])
+        seed[2] = twist(seed[2])
         number++
         if number==9
             number = 1
@@ -576,8 +580,10 @@ galaxy10 {
             planet10.species_look = (seed2_msb ^ msb(seed[1])) & 7   ;bits 0-2 of (w0_hi EOR w1_hi)
             planet10.species_kind = (planet10.species_look + (seed2_msb & 3)) & 7      ;Add bits 0-1 of w2_hi to A from previous step, and take bits 0-2 of the result
         }
-
-        planet10.goatsoup_seed = [lsb(seed[1]), msb(seed[1]), lsb(seed[2]), seed2_msb]
+        planet10.goatsoup_seed[0] = lsb(seed[1])
+        planet10.goatsoup_seed[1] = msb(seed[1])
+        planet10.goatsoup_seed[2] = lsb(seed[2])
+        planet10.goatsoup_seed[3] = seed2_msb
     }
 
     sub tweakseed() {

compiler/test/performance/perf2.p8

@@ -396,14 +396,18 @@ galaxy2 {
     sub init(ubyte galaxy2num) {
         number = 1
         planet2.number = 255
-        seed = [base0, base1, base2]
+        seed[0] = base0
+        seed[1] = base1
+        seed[2] = base2
         repeat galaxy2num-1 {
             nextgalaxy2()
         }
     }
 
     sub nextgalaxy2() {
-        seed = [twist(seed[0]), twist(seed[1]), twist(seed[2])]
+        seed[0] = twist(seed[0])
+        seed[1] = twist(seed[1])
+        seed[2] = twist(seed[2])
         number++
         if number==9
             number = 1
@@ -577,7 +581,10 @@ galaxy2 {
             planet2.species_kind = (planet2.species_look + (seed2_msb & 3)) & 7      ;Add bits 0-1 of w2_hi to A from previous step, and take bits 0-2 of the result
         }
 
-        planet2.goatsoup_seed = [lsb(seed[1]), msb(seed[1]), lsb(seed[2]), seed2_msb]
+        planet2.goatsoup_seed[0] = lsb(seed[1])
+        planet2.goatsoup_seed[1] = msb(seed[1])
+        planet2.goatsoup_seed[2] = lsb(seed[2])
+        planet2.goatsoup_seed[3] = seed2_msb
     }
 
     sub tweakseed() {

compiler/test/performance/perf3.p8

@@ -396,14 +396,18 @@ galaxy3 {
     sub init(ubyte galaxy3num) {
         number = 1
         planet3.number = 255
-        seed = [base0, base1, base2]
+        seed[0] = base0
+        seed[1] = base1
+        seed[2] = base2
         repeat galaxy3num-1 {
             nextgalaxy3()
         }
     }
 
     sub nextgalaxy3() {
-        seed = [twist(seed[0]), twist(seed[1]), twist(seed[2])]
+        seed[0] = twist(seed[0])
+        seed[1] = twist(seed[1])
+        seed[2] = twist(seed[2])
         number++
         if number==9
             number = 1
@@ -577,7 +581,10 @@ galaxy3 {
             planet3.species_kind = (planet3.species_look + (seed2_msb & 3)) & 7      ;Add bits 0-1 of w2_hi to A from previous step, and take bits 0-2 of the result
         }
 
-        planet3.goatsoup_seed = [lsb(seed[1]), msb(seed[1]), lsb(seed[2]), seed2_msb]
+        planet3.goatsoup_seed[0] = lsb(seed[1])
+        planet3.goatsoup_seed[1] = msb(seed[1])
+        planet3.goatsoup_seed[2] = lsb(seed[2])
+        planet3.goatsoup_seed[3] = seed2_msb
     }
 
     sub tweakseed() {

compiler/test/performance/perf4.p8

@@ -396,14 +396,18 @@ galaxy4 {
     sub init(ubyte galaxy4num) {
         number = 1
         planet4.number = 255
-        seed = [base0, base1, base2]
+        seed[0] = base0
+        seed[1] = base1
+        seed[2] = base2
         repeat galaxy4num-1 {
             nextgalaxy4()
         }
     }
 
     sub nextgalaxy4() {
-        seed = [twist(seed[0]), twist(seed[1]), twist(seed[2])]
+        seed[0] = twist(seed[0])
+        seed[1] = twist(seed[1])
+        seed[2] = twist(seed[2])
         number++
         if number==9
             number = 1
@@ -577,7 +581,10 @@ galaxy4 {
             planet4.species_kind = (planet4.species_look + (seed2_msb & 3)) & 7      ;Add bits 0-1 of w2_hi to A from previous step, and take bits 0-2 of the result
         }
 
-        planet4.goatsoup_seed = [lsb(seed[1]), msb(seed[1]), lsb(seed[2]), seed2_msb]
+        planet4.goatsoup_seed[0] = lsb(seed[1])
+        planet4.goatsoup_seed[1] = msb(seed[1])
+        planet4.goatsoup_seed[2] = lsb(seed[2])
+        planet4.goatsoup_seed[3] = seed2_msb
     }
 
     sub tweakseed() {

compiler/test/performance/perf5.p8

@@ -396,14 +396,18 @@ galaxy5 {
     sub init(ubyte galaxy5num) {
         number = 1
         planet5.number = 255
-        seed = [base0, base1, base2]
+        seed[0] = base0
+        seed[1] = base1
+        seed[2] = base2
         repeat galaxy5num-1 {
             nextgalaxy5()
         }
     }
 
     sub nextgalaxy5() {
-        seed = [twist(seed[0]), twist(seed[1]), twist(seed[2])]
+        seed[0] = twist(seed[0])
+        seed[1] = twist(seed[1])
+        seed[2] = twist(seed[2])
         number++
         if number==9
             number = 1
@@ -577,7 +581,10 @@ galaxy5 {
             planet5.species_kind = (planet5.species_look + (seed2_msb & 3)) & 7      ;Add bits 0-1 of w2_hi to A from previous step, and take bits 0-2 of the result
         }
 
-        planet5.goatsoup_seed = [lsb(seed[1]), msb(seed[1]), lsb(seed[2]), seed2_msb]
+        planet5.goatsoup_seed[0] = lsb(seed[1])
+        planet5.goatsoup_seed[1] = msb(seed[1])
+        planet5.goatsoup_seed[2] = lsb(seed[2])
+        planet5.goatsoup_seed[3] = seed2_msb
     }
 
     sub tweakseed() {

compiler/test/performance/perf6.p8

@@ -396,14 +396,18 @@ galaxy6 {
     sub init(ubyte galaxy6num) {
         number = 1
         planet6.number = 255
-        seed = [base0, base1, base2]
+        seed[0] = base0
+        seed[1] = base1
+        seed[2] = base2
         repeat galaxy6num-1 {
             nextgalaxy6()
         }
     }
 
     sub nextgalaxy6() {
-        seed = [twist(seed[0]), twist(seed[1]), twist(seed[2])]
+        seed[0] = twist(seed[0])
+        seed[1] = twist(seed[1])
+        seed[2] = twist(seed[2])
         number++
         if number==9
             number = 1
@@ -577,7 +581,10 @@ galaxy6 {
             planet6.species_kind = (planet6.species_look + (seed2_msb & 3)) & 7      ;Add bits 0-1 of w2_hi to A from previous step, and take bits 0-2 of the result
         }
 
-        planet6.goatsoup_seed = [lsb(seed[1]), msb(seed[1]), lsb(seed[2]), seed2_msb]
+        planet6.goatsoup_seed[0] = lsb(seed[1])
+        planet6.goatsoup_seed[1] = msb(seed[1])
+        planet6.goatsoup_seed[2] = lsb(seed[2])
+        planet6.goatsoup_seed[3] = seed2_msb
     }
 
     sub tweakseed() {

compiler/test/performance/perf7.p8

@@ -396,14 +396,18 @@ galaxy7 {
     sub init(ubyte galaxy7num) {
         number = 1
         planet7.number = 255
-        seed = [base0, base1, base2]
+        seed[0] = base0
+        seed[1] = base1
+        seed[2] = base2
         repeat galaxy7num-1 {
             nextgalaxy7()
         }
     }
 
     sub nextgalaxy7() {
-        seed = [twist(seed[0]), twist(seed[1]), twist(seed[2])]
+        seed[0] = twist(seed[0])
+        seed[1] = twist(seed[1])
+        seed[2] = twist(seed[2])
         number++
         if number==9
             number = 1
@@ -577,7 +581,10 @@ galaxy7 {
             planet7.species_kind = (planet7.species_look + (seed2_msb & 3)) & 7      ;Add bits 0-1 of w2_hi to A from previous step, and take bits 0-2 of the result
         }
 
-        planet7.goatsoup_seed = [lsb(seed[1]), msb(seed[1]), lsb(seed[2]), seed2_msb]
+        planet7.goatsoup_seed[0] = lsb(seed[1])
+        planet7.goatsoup_seed[1] = msb(seed[1])
+        planet7.goatsoup_seed[2] = lsb(seed[2])
+        planet7.goatsoup_seed[3] = seed2_msb
     }
 
     sub tweakseed() {

compiler/test/performance/perf8.p8

@@ -396,14 +396,18 @@ galaxy8 {
     sub init(ubyte galaxy8num) {
         number = 1
         planet8.number = 255
-        seed = [base0, base1, base2]
+        seed[0] = base0
+        seed[1] = base1
+        seed[2] = base2
         repeat galaxy8num-1 {
             nextgalaxy8()
         }
     }
 
     sub nextgalaxy8() {
-        seed = [twist(seed[0]), twist(seed[1]), twist(seed[2])]
+        seed[0] = twist(seed[0])
+        seed[1] = twist(seed[1])
+        seed[2] = twist(seed[2])
         number++
         if number==9
             number = 1
@@ -577,7 +581,10 @@ galaxy8 {
             planet8.species_kind = (planet8.species_look + (seed2_msb & 3)) & 7      ;Add bits 0-1 of w2_hi to A from previous step, and take bits 0-2 of the result
         }
 
-        planet8.goatsoup_seed = [lsb(seed[1]), msb(seed[1]), lsb(seed[2]), seed2_msb]
+        planet8.goatsoup_seed[0] = lsb(seed[1])
+        planet8.goatsoup_seed[1] = msb(seed[1])
+        planet8.goatsoup_seed[2] = lsb(seed[2])
+        planet8.goatsoup_seed[3] = seed2_msb
     }
 
     sub tweakseed() {

compiler/test/performance/perf9.p8

@@ -396,14 +396,18 @@ galaxy9 {
     sub init(ubyte galaxy9num) {
         number = 1
         planet9.number = 255
-        seed = [base0, base1, base2]
+        seed[0] = base0
+        seed[1] = base1
+        seed[2] = base2
         repeat galaxy9num-1 {
             nextgalaxy9()
         }
     }
 
     sub nextgalaxy9() {
-        seed = [twist(seed[0]), twist(seed[1]), twist(seed[2])]
+        seed[0] = twist(seed[0])
+        seed[1] = twist(seed[1])
+        seed[2] = twist(seed[2])
         number++
         if number==9
             number = 1
@@ -576,8 +580,10 @@ galaxy9 {
             planet9.species_look = (seed2_msb ^ msb(seed[1])) & 7   ;bits 0-2 of (w0_hi EOR w1_hi)
             planet9.species_kind = (planet9.species_look + (seed2_msb & 3)) & 7      ;Add bits 0-1 of w2_hi to A from previous step, and take bits 0-2 of the result
         }
-
-        planet9.goatsoup_seed = [lsb(seed[1]), msb(seed[1]), lsb(seed[2]), seed2_msb]
+        planet9.goatsoup_seed[0] = lsb(seed[1])
+        planet9.goatsoup_seed[1] = msb(seed[1])
+        planet9.goatsoup_seed[2] = lsb(seed[2])
+        planet9.goatsoup_seed[3] = seed2_msb
     }
 
     sub tweakseed() {

compiler/test/performance/perftest.p8

@@ -406,14 +406,18 @@ galaxy {
     sub init(ubyte galaxynum) {
         number = 1
         planet.number = 255
-        seed = [base0, base1, base2]
+        seed[0] = base0
+        seed[1] = base1
+        seed[2] = base2
         repeat galaxynum-1 {
             nextgalaxy()
         }
     }
 
     sub nextgalaxy() {
-        seed = [twist(seed[0]), twist(seed[1]), twist(seed[2])]
+        seed[0] = twist(seed[0])
+        seed[1] = twist(seed[1])
+        seed[2] = twist(seed[2])
         number++
         if number==9
             number = 1
@@ -587,7 +591,10 @@ galaxy {
             planet.species_kind = (planet.species_look + (seed2_msb & 3)) & 7      ;Add bits 0-1 of w2_hi to A from previous step, and take bits 0-2 of the result
         }
 
-        planet.goatsoup_seed = [lsb(seed[1]), msb(seed[1]), lsb(seed[2]), seed2_msb]
+        planet.goatsoup_seed[0] = lsb(seed[1])
+        planet.goatsoup_seed[1] = msb(seed[1])
+        planet.goatsoup_seed[2] = lsb(seed[2])
+        planet.goatsoup_seed[3] = seed2_msb
     }
 
     sub tweakseed() {

compilerAst/build.gradle

@@ -1,7 +1,7 @@
 plugins {
     id 'antlr'
     id 'java'
-    id "org.jetbrains.kotlin.jvm" version "1.4.30"
+    id "org.jetbrains.kotlin.jvm" version "1.4.32"
 }
 
 targetCompatibility = 11

compilerAst/src/prog8/ast/AstToSourceCode.kt

@@ -121,8 +121,7 @@ class AstToSourceCode(val output: (text: String) -> Unit, val program: Program):
 
         output(datatypeString(decl.datatype))
         if(decl.arraysize!=null) {
-            decl.arraysize!!.indexNum?.accept(this)
-            decl.arraysize!!.indexVar?.accept(this)
+            decl.arraysize!!.indexExpr.accept(this)
         }
         if(decl.isArray)
             output("]")
@@ -138,8 +137,11 @@ class AstToSourceCode(val output: (text: String) -> Unit, val program: Program):
 
     override fun visit(subroutine: Subroutine) {
         output("\n")
+        outputi("")
+        if(subroutine.inline)
+            output("inline ")
         if(subroutine.isAsmSubroutine) {
-            outputi("asmsub ${subroutine.name} (")
+            output("asmsub ${subroutine.name} (")
             for(param in subroutine.parameters.zip(subroutine.asmParameterRegisters)) {
                 val reg =
                         when {
@@ -153,7 +155,7 @@ class AstToSourceCode(val output: (text: String) -> Unit, val program: Program):
             }
         }
         else {
-            outputi("sub ${subroutine.name} (")
+            output("sub ${subroutine.name} (")
             for(param in subroutine.parameters) {
                 output("${datatypeString(param.type)} ${param.name}")
                 if(param!==subroutine.parameters.last())
@@ -365,8 +367,7 @@ class AstToSourceCode(val output: (text: String) -> Unit, val program: Program):
     override fun visit(arrayIndexedExpression: ArrayIndexedExpression) {
         arrayIndexedExpression.arrayvar.accept(this)
         output("[")
-        arrayIndexedExpression.indexer.indexNum?.accept(this)
-        arrayIndexedExpression.indexer.indexVar?.accept(this)
+        arrayIndexedExpression.indexer.indexExpr.accept(this)
         output("]")
     }
 

compilerAst/src/prog8/ast/AstToplevel.kt

@@ -178,7 +178,6 @@ interface INameScope {
         }
     }
 
-    fun containsDefinedVariables() = statements.any { it is VarDecl && (it !is ParameterVarDecl) }
     fun containsCodeOrVars() = statements.any { it !is Directive || it.directive == "%asminclude" || it.directive == "%asm"}
     fun containsNoCodeNorVars() = !containsCodeOrVars()
 
@@ -274,24 +273,22 @@ class Program(val name: String,
     init {
         // insert a container module for all interned strings later
         if(modules.firstOrNull()?.name != internedStringsModuleName) {
-            val internedStringsModule = Module(internedStringsModuleName, mutableListOf(), Position.DUMMY, false, Path.of(""))
+            val internedStringsModule = Module(internedStringsModuleName, mutableListOf(), Position.DUMMY, true, Path.of(""))
             modules.add(0, internedStringsModule)
-            val block = Block(internedStringsModuleName, null, mutableListOf(), false, Position.DUMMY)
+            val block = Block(internedStringsModuleName, null, mutableListOf(), true, Position.DUMMY)
             internedStringsModule.statements.add(block)
             internedStringsModule.linkParents(this)
             internedStringsModule.program = this
         }
     }
 
-    fun entrypoint(): Subroutine? {
+    fun entrypoint(): Subroutine {
         val mainBlocks = allBlocks().filter { it.name=="main" }
         if(mainBlocks.size > 1)
             throw FatalAstException("more than one 'main' block")
-        return if(mainBlocks.isEmpty()) {
-            null
-        } else {
-            mainBlocks[0].subScope("start") as Subroutine?
-        }
+        if(mainBlocks.isEmpty())
+            throw FatalAstException("no 'main' block")
+        return mainBlocks[0].subScope("start") as Subroutine
     }
 
     fun internString(string: StringLiteralValue): List<String> {
@@ -339,8 +336,6 @@ class Module(override val name: String,
 
     override lateinit var parent: Node
     lateinit var program: Program
-    val importedBy = mutableListOf<Module>()
-    val imports = mutableSetOf<Module>()
 
     val loadAddress: Int by lazy {
         val address = (statements.singleOrNull { it is Directive && it.directive == "%address" } as? Directive)?.args?.single()?.int ?: 0

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

@@ -60,7 +60,13 @@ enum class DataType {
 enum class CpuRegister {
     A,
     X,
-    Y
+    Y;
+
+    fun asRegisterOrPair(): RegisterOrPair = when(this) {
+        A -> RegisterOrPair.A
+        X -> RegisterOrPair.X
+        Y -> RegisterOrPair.Y
+    }
 }
 
 enum class RegisterOrPair {

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

@@ -6,7 +6,7 @@ import prog8.ast.base.*
 import prog8.ast.statements.*
 import prog8.ast.walk.AstWalker
 import prog8.ast.walk.IAstVisitor
-import java.util.*
+import java.util.Objects
 import kotlin.math.abs
 
 
@@ -22,6 +22,7 @@ sealed class Expression: Node {
     abstract fun accept(visitor: AstWalker, parent: Node)
     abstract fun referencesIdentifier(vararg scopedName: String): Boolean
     abstract fun inferType(program: Program): InferredTypes.InferredType
+    abstract val isSimple: Boolean
 
     infix fun isSameAs(assigntarget: AssignTarget) = assigntarget.isSameAs(this)
 
@@ -105,6 +106,8 @@ class PrefixExpression(val operator: String, var expression: Expression, overrid
         }
     }
 
+    override val isSimple = false
+
     override fun toString(): String {
         return "Prefix($operator $expression)"
     }
@@ -133,6 +136,8 @@ class BinaryExpression(var left: Expression, var operator: String, var right: Ex
         return "[$left $operator $right]"
     }
 
+    override val isSimple = false
+
     // binary expression should actually have been optimized away into a single value, before const value was requested...
     override fun constValue(program: Program): NumericLiteralValue? = null
 
@@ -242,6 +247,8 @@ class ArrayIndexedExpression(var arrayvar: IdentifierReference,
         indexer.linkParents(this)
     }
 
+    override val isSimple = indexer.indexExpr is NumericLiteralValue || indexer.indexExpr is IdentifierReference
+
     override fun replaceChildNode(node: Node, replacement: Node) {
         when {
             node===arrayvar -> arrayvar = replacement as IdentifierReference
@@ -271,6 +278,8 @@ class ArrayIndexedExpression(var arrayvar: IdentifierReference,
     override fun toString(): String {
         return "ArrayIndexed(ident=$arrayvar, arraysize=$indexer; pos=$position)"
     }
+
+    fun copy() = ArrayIndexedExpression(arrayvar.copy(), indexer.copy(), position)
 }
 
 class TypecastExpression(var expression: Expression, var type: DataType, val implicit: Boolean, override val position: Position) : Expression() {
@@ -281,6 +290,8 @@ class TypecastExpression(var expression: Expression, var type: DataType, val imp
         expression.linkParents(this)
     }
 
+    override val isSimple = false
+
     override fun replaceChildNode(node: Node, replacement: Node) {
         require(replacement is Expression && node===expression)
         expression = replacement
@@ -314,6 +325,8 @@ data class AddressOf(var identifier: IdentifierReference, override val position:
         identifier.parent=this
     }
 
+    override val isSimple = true
+
     override fun replaceChildNode(node: Node, replacement: Node) {
         require(replacement is IdentifierReference && node===identifier)
         identifier = replacement
@@ -335,6 +348,8 @@ class DirectMemoryRead(var addressExpression: Expression, override val position:
         this.addressExpression.linkParents(this)
     }
 
+    override val isSimple = true
+
     override fun replaceChildNode(node: Node, replacement: Node) {
         require(replacement is Expression && node===addressExpression)
         addressExpression = replacement
@@ -351,6 +366,8 @@ class DirectMemoryRead(var addressExpression: Expression, override val position:
     override fun toString(): String {
         return "DirectMemoryRead($addressExpression)"
     }
+
+    fun copy() =  DirectMemoryRead(addressExpression, position)
 }
 
 class NumericLiteralValue(val type: DataType,    // only numerical types allowed
@@ -358,6 +375,8 @@ class NumericLiteralValue(val type: DataType,    // only numerical types allowed
                           override val position: Position) : Expression() {
     override lateinit var parent: Node
 
+    override val isSimple = true
+
     companion object {
         fun fromBoolean(bool: Boolean, position: Position) =
                 NumericLiteralValue(DataType.UBYTE, if (bool) 1 else 0, position)
@@ -489,6 +508,8 @@ class StringLiteralValue(val value: String,
         this.parent = parent
     }
 
+    override val isSimple = true
+
     override fun replaceChildNode(node: Node, replacement: Node) {
         throw FatalAstException("can't replace here")
     }
@@ -519,6 +540,8 @@ class ArrayLiteralValue(val type: InferredTypes.InferredType,     // inferred be
         value.forEach {it.linkParents(this)}
     }
 
+    override val isSimple = true
+
     override fun replaceChildNode(node: Node, replacement: Node) {
         require(replacement is Expression)
         val idx = value.indexOfFirst { it===node }
@@ -542,6 +565,14 @@ class ArrayLiteralValue(val type: InferredTypes.InferredType,     // inferred be
         return type==other.type && value.contentEquals(other.value)
     }
 
+    fun memsize(memsizer: IMemSizer): Int {
+        if(type.isKnown) {
+            val eltType = ArrayElementTypes.getValue(type.typeOrElse(DataType.STRUCT))
+            return memsizer.memorySize(eltType) * value.size
+        }
+        else throw IllegalArgumentException("array datatype is not yet known")
+    }
+
     fun guessDatatype(program: Program): InferredTypes.InferredType {
         // Educated guess of the desired array literal's datatype.
         // If it's inside a for loop, assume the data type of the loop variable is what we want.
@@ -617,6 +648,8 @@ class RangeExpr(var from: Expression,
         step.linkParents(this)
     }
 
+    override val isSimple = true
+
     override fun replaceChildNode(node: Node, replacement: Node) {
         require(replacement is Expression)
         when {
@@ -703,20 +736,21 @@ internal fun makeRange(fromVal: Int, toVal: Int, stepVal: Int): IntProgression {
     }
 }
 
-data class IdentifierReference(val nameInSource: List<String>, override val position: Position) : Expression(),
-    IAssignable {
+data class IdentifierReference(val nameInSource: List<String>, override val position: Position) : Expression(), IAssignable {
     override lateinit var parent: Node
 
+    override val isSimple = true
+
     fun targetStatement(program: Program) =
         if(nameInSource.size==1 && nameInSource[0] in program.builtinFunctions.names)
-            BuiltinFunctionStatementPlaceholder(nameInSource[0], position)
+            BuiltinFunctionStatementPlaceholder(nameInSource[0], position, parent)
         else
             program.namespace.lookup(nameInSource, this)
 
     fun targetVarDecl(program: Program): VarDecl? = targetStatement(program) as? VarDecl
     fun targetSubroutine(program: Program): Subroutine? = targetStatement(program) as? Subroutine
 
-    override fun equals(other: Any?) = other is IdentifierReference && other.nameInSource==nameInSource
+    override fun equals(other: Any?) = other is IdentifierReference && other.nameInSource==nameInSource     // NOTE: only compare by the name, not the position!
     override fun hashCode() = nameInSource.hashCode()
 
     override fun linkParents(parent: Node) {
@@ -786,6 +820,8 @@ class FunctionCall(override var target: IdentifierReference,
         args.forEach { it.linkParents(this) }
     }
 
+    override val isSimple = target.nameInSource.size==1 && (target.nameInSource[0] in setOf("msb", "lsb", "peek", "peekw"))
+
     override fun replaceChildNode(node: Node, replacement: Node) {
         if(node===target)
             target=replacement as IdentifierReference

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

@@ -7,6 +7,10 @@ import prog8.ast.walk.AstWalker
 import prog8.ast.walk.IAstVisitor
 
 
+interface ISymbolStatement {
+    val name: String
+}
+
 sealed class Statement : Node {
     abstract fun accept(visitor: IAstVisitor)
     abstract fun accept(visitor: AstWalker, parent: Node)
@@ -31,8 +35,7 @@ sealed class Statement : Node {
 }
 
 
-class BuiltinFunctionStatementPlaceholder(val name: String, override val position: Position) : Statement() {
-    override var parent: Node = ParentSentinel
+class BuiltinFunctionStatementPlaceholder(val name: String, override val position: Position, override var parent: Node) : Statement() {
     override fun linkParents(parent: Node) {}
     override fun accept(visitor: IAstVisitor) = throw FatalAstException("should not iterate over this node")
     override fun accept(visitor: AstWalker, parent: Node) = throw FatalAstException("should not iterate over this node")
@@ -48,7 +51,7 @@ class Block(override val name: String,
             val address: Int?,
             override var statements: MutableList<Statement>,
             val isInLibrary: Boolean,
-            override val position: Position) : Statement(), INameScope {
+            override val position: Position) : Statement(), INameScope, ISymbolStatement {
     override lateinit var parent: Node
 
     override fun linkParents(parent: Node) {
@@ -95,7 +98,7 @@ data class DirectiveArg(val str: String?, val name: String?, val int: Int?, over
     override fun replaceChildNode(node: Node, replacement: Node) = throw FatalAstException("can't replace here")
 }
 
-data class Label(val name: String, override val position: Position) : Statement() {
+data class Label(override val name: String, override val position: Position) : Statement(), ISymbolStatement {
     override lateinit var parent: Node
 
     override fun linkParents(parent: Node) {
@@ -153,17 +156,16 @@ enum class ZeropageWish {
     NOT_IN_ZEROPAGE
 }
 
-
 open class VarDecl(val type: VarDeclType,
                    private val declaredDatatype: DataType,
                    val zeropage: ZeropageWish,
                    var arraysize: ArrayIndex?,
-                   val name: String,
+                   override val name: String,
                    private val structName: String?,
                    var value: Expression?,
                    val isArray: Boolean,
                    val autogeneratedDontRemove: Boolean,
-                   override val position: Position) : Statement() {
+                   override val position: Position) : Statement(), ISymbolStatement {
     override lateinit var parent: Node
     var struct: StructDecl? = null        // set later (because at parse time, we only know the name)
         private set
@@ -221,14 +223,18 @@ open class VarDecl(val type: VarDeclType,
         value?.linkParents(this)
         if(structName!=null) {
             val structStmt = definingScope().lookup(listOf(structName), this)
-            if(structStmt!=null)
-                struct = definingScope().lookup(listOf(structName), this) as StructDecl
+            if(structStmt!=null) {
+                val node = definingScope().lookup(listOf(structName), this)
+                if(node is StructDecl)
+                    struct = node
+                else
+                    datatypeErrors.add(SyntaxError("invalid datatype declaration", position))
+            }
         }
     }
 
     override fun replaceChildNode(node: Node, replacement: Node) {
-        // 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)
+        require(replacement is Expression && (value==null || node===value))
         value = replacement
         replacement.parent = this
     }
@@ -250,7 +256,14 @@ open class VarDecl(val type: VarDeclType,
     fun flattenStructMembers(): MutableList<Statement> {
         val result = struct!!.statements.mapIndexed { index, statement ->
             val member = statement as VarDecl
-            val initvalue = if(value!=null) (value as ArrayLiteralValue).value[index] else null
+            val initvalueOrig = if(value!=null) (value as ArrayLiteralValue).value[index] else null
+            val initvalue = when(initvalueOrig) {
+                is AddressOf -> initvalueOrig.copy()
+                is ArrayIndexedExpression -> initvalueOrig.copy()
+                is DirectMemoryRead -> initvalueOrig.copy()
+                is IdentifierReference -> initvalueOrig.copy()
+                else -> initvalueOrig
+            }
             VarDecl(
                     VarDeclType.VAR,
                     member.datatype,
@@ -267,53 +280,32 @@ open class VarDecl(val type: VarDeclType,
         structHasBeenFlattened = true
         return result
     }
+
+    fun copy(): VarDecl {
+        val c = VarDecl(type, declaredDatatype, zeropage, arraysize, name, structName, value, isArray, autogeneratedDontRemove, position)
+        c.allowInitializeWithZero = this.allowInitializeWithZero
+        c.structHasBeenFlattened = this.structHasBeenFlattened
+        return c
+    }
 }
 
 // a vardecl used only for subroutine parameters
 class ParameterVarDecl(name: String, declaredDatatype: DataType, position: Position)
     : VarDecl(VarDeclType.VAR, declaredDatatype, ZeropageWish.DONTCARE, null, name, null, null, false, true, position)
 
-
-class ArrayIndex(var origExpression: Expression?,            // will be replaced later by either the number or the identifier
+class ArrayIndex(var indexExpr: Expression,
                  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
-        origExpression?.linkParents(this)
-        indexNum?.linkParents(this)
-        indexVar?.linkParents(this)
+        indexExpr.linkParents(this)
     }
 
     override fun replaceChildNode(node: Node, replacement: Node) {
         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")
-        }
+        if (node===indexExpr) indexExpr = replacement
+        else throw FatalAstException("invalid replace")
     }
 
     companion object {
@@ -323,29 +315,17 @@ class ArrayIndex(var origExpression: Expression?,            // will be replaced
         }
     }
 
-    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 accept(visitor: IAstVisitor) = indexExpr.accept(visitor)
+    fun accept(visitor: AstWalker, parent: Node)  = indexExpr.accept(visitor, this)
 
     override fun toString(): String {
-        return("ArrayIndex($indexNum, $indexVar, pos=$position)")
+        return("ArrayIndex($indexExpr, pos=$position)")
     }
 
-    fun constIndex() = indexNum?.number?.toInt()
+    fun constIndex() = (indexExpr as? NumericLiteralValue)?.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!!
-    }
+    infix fun isSameAs(other: ArrayIndex): Boolean = indexExpr isSameAs other.indexExpr
+    fun copy() = ArrayIndex(indexExpr, position)
 }
 
 open class Assignment(var target: AssignTarget, var value: Expression, override val position: Position) : Statement() {
@@ -463,9 +443,10 @@ data class AssignTarget(var identifier: IdentifierReference?,
     }
 
     fun toExpression(): Expression {
+        // return a copy of the assignment target but as a source expression.
         return when {
-            identifier != null -> identifier!!
-            arrayindexed != null -> arrayindexed!!
+            identifier != null -> identifier!!.copy()
+            arrayindexed != null -> arrayindexed!!.copy()
             memoryAddress != null -> DirectMemoryRead(memoryAddress.addressExpression, memoryAddress.position)
             else -> throw FatalAstException("invalid assignmenttarget $this")
         }
@@ -510,8 +491,9 @@ data class AssignTarget(var identifier: IdentifierReference?,
         }
         return false
     }
-}
 
+    fun copy() = AssignTarget(identifier?.copy(), arrayindexed?.copy(), memoryAddress?.copy(), position)
+}
 
 class PostIncrDecr(var target: AssignTarget, val operator: String, override val position: Position) : Statement() {
     override lateinit var parent: Node
@@ -639,20 +621,16 @@ 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)
 }
 
 // the subroutine class covers both the normal user-defined subroutines,
@@ -668,7 +646,7 @@ class Subroutine(override val name: String,
                  val isAsmSubroutine: Boolean,
                  val inline: Boolean,
                  override var statements: MutableList<Statement>,
-                 override val position: Position) : Statement(), INameScope {
+                 override val position: Position) : Statement(), INameScope, ISymbolStatement {
 
     constructor(name: String, parameters: List<SubroutineParameter>, returntypes: List<DataType>, statements: MutableList<Statement>, inline: Boolean, position: Position)
             : this(name, parameters, returntypes, emptyList(), determineReturnRegisters(returntypes), emptySet(), null, false, inline, statements, position)
@@ -728,10 +706,9 @@ class Subroutine(override val name: String,
             .asSequence()
             .filter { it is InlineAssembly }
             .map { (it as InlineAssembly).assembly }
-            .count { " rti" in it || "\trti" in it || " rts" in it || "\trts" in it || " jmp" in it || "\tjmp" in it }
+            .count { " rti" in it || "\trti" in it || " rts" in it || "\trts" in it || " jmp" in it || "\tjmp" in it || " bra" in it || "\tbra" in it}
 }
 
-
 open class SubroutineParameter(val name: String,
                                val type: DataType,
                                override val position: Position) : Node {
@@ -978,7 +955,6 @@ class WhenChoice(var values: MutableList<Expression>?,           // if null,  th
     fun accept(visitor: AstWalker, parent: Node) = visitor.visit(this, parent)
 }
 
-
 class StructDecl(override val name: String,
                  override var statements: MutableList<Statement>,      // actually, only vardecls here
                  override val position: Position): Statement(), INameScope {
@@ -1000,6 +976,9 @@ class StructDecl(override val name: String,
     val numberOfElements: Int
         get() = this.statements.size
 
+    fun memsize(memsizer: IMemSizer) =
+        statements.map { memsizer.memorySize((it as VarDecl).datatype) }.sum()
+
     override fun accept(visitor: IAstVisitor) = visitor.visit(this)
     override fun accept(visitor: AstWalker, parent: Node) = visitor.visit(this, parent)
 
@@ -1026,4 +1005,5 @@ class DirectMemoryWrite(var addressExpression: Expression, override val position
 
     fun accept(visitor: IAstVisitor) = visitor.visit(this)
     fun accept(visitor: AstWalker, parent: Node) = visitor.visit(this, parent)
+    fun copy() = DirectMemoryWrite(addressExpression, position)
 }

compilerAst/src/prog8/ast/walk/AstWalker.kt

@@ -56,7 +56,7 @@ interface IAstModification {
         }
     }
 
-    class ReplaceNode(private val node: Node, private val replacement: Node, private val parent: Node) :
+    class ReplaceNode(val node: Node, private val replacement: Node, private val parent: Node) :
         IAstModification {
         override fun perform() {
             parent.replaceChildNode(node, replacement)
@@ -76,94 +76,117 @@ interface IAstModification {
 
 
 abstract class AstWalker {
-    open fun before(addressOf: AddressOf, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(array: ArrayLiteralValue, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(arrayIndexedExpression: ArrayIndexedExpression, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(assignTarget: AssignTarget, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(assignment: Assignment, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(block: Block, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(branchStatement: BranchStatement, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(breakStmt: Break, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(decl: VarDecl, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(directive: Directive, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(expr: BinaryExpression, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(expr: PrefixExpression, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(forLoop: ForLoop, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(repeatLoop: RepeatLoop, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(functionCall: FunctionCall, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(functionCallStatement: FunctionCallStatement, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(identifier: IdentifierReference, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(ifStatement: IfStatement, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(inlineAssembly: InlineAssembly, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(jump: Jump, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(label: Label, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(memread: DirectMemoryRead, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(memwrite: DirectMemoryWrite, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(module: Module, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(nopStatement: NopStatement, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(numLiteral: NumericLiteralValue, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(postIncrDecr: PostIncrDecr, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(program: Program, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(range: RangeExpr, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(untilLoop: UntilLoop, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(returnStmt: Return, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(scope: AnonymousScope, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(string: StringLiteralValue, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(structDecl: StructDecl, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(subroutine: Subroutine, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(typecast: TypecastExpression, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(whenChoice: WhenChoice, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(whenStatement: WhenStatement, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun before(whileLoop: WhileLoop, parent: Node): Iterable<IAstModification> = emptyList()
+    protected val noModifications = emptyList<IAstModification>()
 
-    open fun after(addressOf: AddressOf, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(array: ArrayLiteralValue, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(arrayIndexedExpression: ArrayIndexedExpression, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(assignTarget: AssignTarget, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(assignment: Assignment, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(block: Block, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(branchStatement: BranchStatement, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(breakStmt: Break, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(builtinFunctionStatementPlaceholder: BuiltinFunctionStatementPlaceholder, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(directive: Directive, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(expr: BinaryExpression, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(expr: PrefixExpression, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(forLoop: ForLoop, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(repeatLoop: RepeatLoop, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(functionCall: FunctionCall, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(functionCallStatement: FunctionCallStatement, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(identifier: IdentifierReference, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(ifStatement: IfStatement, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(inlineAssembly: InlineAssembly, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(jump: Jump, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(label: Label, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(memread: DirectMemoryRead, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(memwrite: DirectMemoryWrite, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(module: Module, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(nopStatement: NopStatement, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(numLiteral: NumericLiteralValue, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(postIncrDecr: PostIncrDecr, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(program: Program, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(range: RangeExpr, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(untilLoop: UntilLoop, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(returnStmt: Return, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(scope: AnonymousScope, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(string: StringLiteralValue, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(structDecl: StructDecl, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(subroutine: Subroutine, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(typecast: TypecastExpression, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(whenChoice: WhenChoice, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(whenStatement: WhenStatement, parent: Node): Iterable<IAstModification> = emptyList()
-    open fun after(whileLoop: WhileLoop, parent: Node): Iterable<IAstModification> = emptyList()
+    open fun before(addressOf: AddressOf, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(array: ArrayLiteralValue, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(arrayIndexedExpression: ArrayIndexedExpression, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(assignTarget: AssignTarget, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(assignment: Assignment, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(block: Block, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(branchStatement: BranchStatement, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(breakStmt: Break, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(decl: VarDecl, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(directive: Directive, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(expr: BinaryExpression, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(expr: PrefixExpression, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(forLoop: ForLoop, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(repeatLoop: RepeatLoop, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(functionCall: FunctionCall, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(functionCallStatement: FunctionCallStatement, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(identifier: IdentifierReference, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(ifStatement: IfStatement, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(inlineAssembly: InlineAssembly, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(jump: Jump, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(label: Label, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(memread: DirectMemoryRead, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(memwrite: DirectMemoryWrite, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(module: Module, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(nopStatement: NopStatement, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(numLiteral: NumericLiteralValue, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(postIncrDecr: PostIncrDecr, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(program: Program, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(range: RangeExpr, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(untilLoop: UntilLoop, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(returnStmt: Return, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(scope: AnonymousScope, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(string: StringLiteralValue, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(structDecl: StructDecl, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(subroutine: Subroutine, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(typecast: TypecastExpression, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(whenChoice: WhenChoice, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(whenStatement: WhenStatement, parent: Node): Iterable<IAstModification> = noModifications
+    open fun before(whileLoop: WhileLoop, parent: Node): Iterable<IAstModification> = noModifications
 
-    private val modifications = mutableListOf<Triple<IAstModification, Node, Node>>()
+    open fun after(addressOf: AddressOf, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(array: ArrayLiteralValue, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(arrayIndexedExpression: ArrayIndexedExpression, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(assignTarget: AssignTarget, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(assignment: Assignment, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(block: Block, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(branchStatement: BranchStatement, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(breakStmt: Break, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(builtinFunctionStatementPlaceholder: BuiltinFunctionStatementPlaceholder, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(decl: VarDecl, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(directive: Directive, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(expr: BinaryExpression, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(expr: PrefixExpression, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(forLoop: ForLoop, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(repeatLoop: RepeatLoop, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(functionCall: FunctionCall, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(functionCallStatement: FunctionCallStatement, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(identifier: IdentifierReference, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(ifStatement: IfStatement, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(inlineAssembly: InlineAssembly, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(jump: Jump, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(label: Label, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(memread: DirectMemoryRead, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(memwrite: DirectMemoryWrite, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(module: Module, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(nopStatement: NopStatement, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(numLiteral: NumericLiteralValue, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(postIncrDecr: PostIncrDecr, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(program: Program, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(range: RangeExpr, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(untilLoop: UntilLoop, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(returnStmt: Return, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(scope: AnonymousScope, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(string: StringLiteralValue, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(structDecl: StructDecl, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(subroutine: Subroutine, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(typecast: TypecastExpression, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(whenChoice: WhenChoice, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(whenStatement: WhenStatement, parent: Node): Iterable<IAstModification> = noModifications
+    open fun after(whileLoop: WhileLoop, parent: Node): Iterable<IAstModification> = noModifications
+
+    protected val modifications = mutableListOf<Triple<IAstModification, Node, Node>>()
 
     private fun track(mods: Iterable<IAstModification>, node: Node, parent: Node) {
-        for (it in mods) modifications += Triple(it, node, parent)
+        for (it in mods) {
+//            if(it is IAstModification.ReplaceNode) {
+//                val replaceKey = Pair(it.node, it.node.position)
+//                if(replaceKey in modificationsReplacedNodes)
+//                    throw FatalAstException("there already is a node replacement for $replaceKey - optimizer can't deal with multiple replacements for same node yet. Split the ast modification?")
+//                else
+//                    modificationsReplacedNodes.add(replaceKey)
+//            }
+            modifications += Triple(it, node, parent)
+        }
     }
 
     fun applyModifications(): Int {
+        // check if there are double removes, keep only the last one
+        val removals = modifications.filter { it.first is IAstModification.Remove }
+        if(removals.size>0) {
+            val doubles = removals.groupBy { (it.first as IAstModification.Remove).node }.filter { it.value.size>1 }
+            doubles.forEach {
+                for(doubleRemove in it.value.dropLast(1)) {
+                    if(!modifications.removeIf { mod-> mod.first === doubleRemove.first })
+                        throw FatalAstException("ast remove problem")
+                }
+            }
+        }
+
         modifications.forEach {
             it.first.perform()
         }

compilerAst/src/prog8/parser/ModuleParsing.kt

@@ -41,8 +41,8 @@ class ModuleImporter {
         if(!Files.isReadable(filePath))
             throw ParsingFailedError("No such file: $filePath")
 
-        val input = CharStreams.fromPath(filePath)
-        return importModule(program, input, filePath, false, encoder, compilationTargetName)
+        val content = filePath.toFile().readText().replace("\r\n", "\n")
+        return importModule(program, CharStreams.fromString(content), filePath, false, encoder, compilationTargetName)
     }
 
     fun importLibraryModule(program: Program, name: String,
@@ -62,6 +62,8 @@ class ModuleImporter {
                 is prog8Parser -> System.err.println("${recognizer.inputStream.sourceName}:$line:$charPositionInLine: $msg")
                 else -> System.err.println("$line:$charPositionInLine $msg")
             }
+            if(numberOfErrors>=5)
+                throw ParsingFailedError("There are too many parse errors. Stopping.")
         }
     }
 
@@ -120,7 +122,8 @@ class ModuleImporter {
                     val (resource, resourcePath) = rsc
                     resource.use {
                         println("importing '$moduleName' (library)")
-                        importModule(program, CharStreams.fromStream(it), Paths.get("@embedded@/$resourcePath"),
+                        val content = it.reader().readText().replace("\r\n", "\n")
+                        importModule(program, CharStreams.fromString(content), Paths.get("@embedded@/$resourcePath"),
                             true, encoder, compilationTargetName)
                     }
                 } else {

dbusCompilerService/build.gradle

@@ -2,7 +2,7 @@
 plugins {
     id 'java'
     id 'application'
-    id "org.jetbrains.kotlin.jvm" version "1.4.30"
+    id "org.jetbrains.kotlin.jvm" version "1.4.32"
     id 'com.github.johnrengelman.shadow' version '6.1.0'
 }
 
@@ -18,7 +18,7 @@ repositories {
 dependencies {
     implementation "org.jetbrains.kotlin:kotlin-stdlib-jdk8"
     // implementation "org.jetbrains.kotlin:kotlin-reflect"
-    implementation 'org.jetbrains.kotlinx:kotlinx-cli:0.3.1'
+    implementation 'org.jetbrains.kotlinx:kotlinx-cli:0.3.2'
     implementation "com.github.hypfvieh:dbus-java:3.2.4"
     implementation "org.slf4j:slf4j-simple:1.7.30"
 

docs/source/index.rst

@@ -42,20 +42,23 @@ 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
+  It generates a machine code program runnable on actual 8-bit 6502 hardware.
+- Fast execution speed due to compilation to native assembly code. It's possible to write certain raster interrupt 'demoscene' effects purely in Prog8.
+- Provides 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
+- Subroutines with parameters and return values
+- complex nested expressions are possible
+- Variables are 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``, ``sort`` and ``reverse``
+- Programs can be run multiple times without reloading because of automatic variable (re)initializations.
 - Supports the sixteen 'virtual' 16-bit registers R0 .. R15 from the Commander X16, also on the C64.
 - If you only use standard kernal and prog8 library routines, it is possible to compile the *exact same program* for both machines (just change the compiler target flag)!
 
@@ -145,6 +148,8 @@ For MacOS you can use the Homebrew system to install a recent version of OpenJDK
 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>`_.
+Make sure you use cx16 emulator and roms **V39 or newer**! Starting from version 6.5, prog8 targets that system version.
+Your program may work on V38 but that will only be by luck.
 
 .. important::
     **Building the compiler itself:** (*Only needed if you have not downloaded a pre-built 'fat-jar'*)
@@ -161,11 +166,11 @@ If you're targeting the CommanderX16 instead, there's the `x16emu <https://githu
     :maxdepth: 2
     :caption: Contents of this manual:
 
-    targetsystem.rst
     building.rst
     programming.rst
     syntaxreference.rst
     libraries.rst
+    targetsystem.rst
     technical.rst
     todo.rst
 

docs/source/libraries.rst

@@ -99,13 +99,32 @@ sys (part of syslib)
     Returns the last address of the program in memory + 1.
     Can be used to load dynamic data after the program, instead of hardcoding something.
 
+``wait(uword jiffies)``
+    wait approximately the given number of jiffies (1/60th seconds)
+    note: the system irq handler has to be active for this to work as it depends on the system jiffy clock
+
+``waitvsync()``
+    busy wait till the next vsync has occurred (approximately), without depending on custom irq handling.
+    can be used to avoid screen flicker/tearing when updating screen contents.
+    note: a more accurate way to wait for vsync is to set up a vsync irq handler instead.
+    note for cx16: the system irq handler has to be active for this to work (this is not required on c64)
+
+``waitrastborder()`` (c64 target only)
+    busy wait till the raster position has reached the bottom screen border (approximately)
+    can be used to avoid screen flicker/tearing when updating screen contents.
+    note: a more accurate way to do this is by using a raster irq handler instead.
+
+``reset_system()``
+    Soft-reset the system back to initial power-on Basic prompt.
+    (called automatically by Prog8 when the main subroutine returns and the program is not using basicsafe zeropage option)
+
 
 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
+- strings in decimal, hex and binary format into numbers (bytes, words)
 
 
 textio (txt.*)

docs/source/programming.rst

@@ -233,6 +233,10 @@ to worry about this yourself)
 
 The largest 5-byte MFLPT float that can be stored is: **1.7014118345e+38**   (negative: **-1.7014118345e+38**)
 
+.. note::
+    On the Commander X16, to use floating point operations, ROM bank 4 has to be enabled (BASIC).
+    Importing the ``floats`` library will do this for you if needed.
+
 
 Arrays
 ^^^^^^
@@ -451,6 +455,14 @@ Breaking out of a loop prematurely is possible with the ``break`` statement.
     after the loop without first assigning a new value to it!
     (this is an optimization issue to avoid having to deal with mostly useless post-loop logic to adjust the loop variable's value)
 
+.. warning::
+    For efficiency reasons, it is assumed that the ending value of the for loop is actually >= the starting value
+    (or <= if the step is negative).  This means that for loops in prog8 behave differently than in other
+    languages if this is *not* the case! A for loop from ubyte 10 to ubyte 2, for example, will iterate through
+    all values 10, 11, 12, 13, .... 254, 255, 0  (wrapped), 1, 2. In other languages the entire loop will
+    be skipped in such cases. But prog8 omits the overhead of an extra loop range check and/or branch for every for loop
+    by assuming the normal ranges.
+
 
 Conditional Execution
 ---------------------
@@ -642,7 +654,7 @@ Subroutines can be defined in a Block, but also nested inside another subroutine
 With ``asmsub`` you can define a low-level subroutine that is implemented in inline assembly and takes any parameters
 in registers directly.
 
-Trivial subroutines can be tagged as inline to tell the compiler to copy their code
+Trivial subroutines can be tagged as ``inline`` to tell the compiler to copy their code
 in-place to the locations where the subroutine is called, rather than inserting an actual call and return to the
 subroutine. This may increase code size significantly and can only be used in limited scenarios, so YMMV.
 
@@ -739,9 +751,11 @@ sin16(x)
 
 sqrt16(w)
     16 bit unsigned integer Square root. Result is unsigned byte.
+    To do the reverse, squaring an integer, just write ``x*x``.
 
 sqrt(x)
     Floating point Square root.
+    To do the reverse, squaring a floating point number, just write ``x*x`` or ``x**2``.
 
 tan(x)
     Tangent.
@@ -757,11 +771,12 @@ 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).
+    Number of values in the array value x, or the number of characters in a string (excluding the 0-byte).
     Note: this can be different from the number of *bytes* in memory if the datatype isn't a byte. See sizeof().
     Note: lengths of strings and arrays are determined at compile-time! If your program modifies the actual
-    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)
+    length of the string during execution, the value of len(s) may no longer be correct!
+    (use the ``string.length`` routine if you want to dynamically determine the length by counting to the
+    first 0-byte)
 
 max(x)
     Maximum of the values in the array value x
@@ -788,6 +803,12 @@ sort(array)
 
 Miscellaneous
 ^^^^^^^^^^^^^
+
+cmp(x,y)
+    Compare the integer value x to integer value y. Doesn't return a value or boolean result, only sets the processor's status bits!
+    You can use a conditional jumps (``if_cc`` etcetera) to act on this.
+    Normally you should just use a comparison expression (``x < y``)
+
 lsb(x)
     Get the least significant byte of the word x. Equivalent to the cast "x as ubyte".
 
@@ -824,6 +845,9 @@ rndw()
 rndf()
     returns a pseudo-random float between 0.0 and 1.0
 
+fastrnd8()
+    returns a pseudo-random byte from 0..255 (using a fast but not very good rng)
+
 rol(x)
     Rotate the bits in x (byte or word) one position to the left.
     This uses the CPU's rotate semantics: bit 0 will be set to the current value of the Carry flag,

docs/source/syntaxreference.rst

@@ -129,7 +129,7 @@ Directives
       take care of that yourself. The program will just start running from whatever state the machine is in when the
       program was launched.
     - ``force_output`` (in a block) 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).
+      Can be useful to make sure some data is generated that would otherwise be discarded because the compiler thinks it's not referenced (such as sprite data)
     - ``align_word`` (in a block) will make the assembler align the start address of this block on a word boundary in memory (so, an even memory address).
     - ``align_page`` (in a block) will make the assembler align the start address of this block on a page boundary in memory (so, the LSB of the address is 0).
 
@@ -550,7 +550,7 @@ and can have nothing following it. The close curly brace must be on its own line
 The parameters is a (possibly empty) comma separated list of "<datatype> <parametername>" pairs specifying the input parameters.
 The return type has to be specified if the subroutine returns a value.
 The ``inline`` keyword makes their code copied in-place to the locations where the subroutine is called,
-rather than having an actual call and return to the subroutine. This is meant for trivial subroutines only
+rather than having an actual call and return to the subroutine. This is meant for very small subroutines only
 as it can increase code size significantly.
 
 
@@ -592,7 +592,7 @@ flag such as Carry (Pc).
     Asmsubs can also be tagged as ``inline asmsub`` to make trivial pieces of assembly inserted
     directly instead of a call to them. Note that it is literal copy-paste of code that is done,
     so make sure the assembly is actually written to behave like such - which probably means you
-    don't want a ``rts`` or ``jmp`` in it!
+    don't want a ``rts`` or ``jmp`` or ``bra`` in it!
 
 
 .. note::

docs/source/technical.rst

@@ -1,6 +1,6 @@
-===============
-Technical stuff
-===============
+=================
+Technical details
+=================
 
 All variables are static in memory
 ----------------------------------
@@ -83,3 +83,37 @@ Some builtin functions have a fully custom implementation.
 The compiler will warn about routines that are called and that return a value, if you're not
 doing something with that returnvalue. This can be on purpuse if you're simply not interested in it.
 Use the ``void`` keyword in front of the subroutine call to get rid of the warning in that case.
+
+
+The 6502 CPU's X-register: off-limits
+-------------------------------------
+
+Prog8 uses the cpu's X-register as a pointer in its internal expression evaluation stack.
+When only writing code in Prog8, this is taken care of behind the scenes for you by the compiler.
+However when you are including or linking with assembly routines or kernal/ROM calls that *do*
+use the X register (either clobbering it internally, or using it as a parameter, or return value register),
+those calls will destroy Prog8's stack pointer and this will result in invalid calculations.
+
+You should avoid using the X register in your assembly code, or take preparations.
+If you make sure that the value of the X register is preserved before calling a routine
+that uses it, and restored when the routine is done, you'll be ok.
+
+Routines that return a value in the X register can be called from Prog8 but the return value is
+inaccessible unless you write a short piece of inline assembly code to deal with it yourself, such as::
+
+    ubyte returnvalue
+
+    %asm {{
+        stx  P8ZP_SCRATCH_REG       ; use 'phx/plx' if using 65c02 cpu
+        ldx  #10
+        jsr  routine_using_x
+        stx  returnvalue
+        ldx  P8ZP_SCRATCH_REG
+    }}
+    ; now use 'returnvalue' variable
+
+Prog8 also provides some help to deal with this:
+
+- you should use a ``clobbers(X)`` specification for asmsub routines that modify the X register; the compiler will preserve it for you automatically when such a routine is called
+- the ``sys.rsave()`` and ``sys.rrestore()`` routines can preserve and restore *all* registers (but this is very slow and overkill if you only need to save X)
+

docs/source/todo.rst

@@ -2,31 +2,29 @@
 TODO
 ====
 
-- optimize assigning array and struct variables (multi-element assings -> memcopy)
 - 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)
-- optimize swap of two memread values with index, using the same pointer expression/variable, like swap(@(ptr+1), @(ptr+2))
-- optimize several inner loops in gfx2
-- try to fix the bresenham line routines in graphics and gfx2 (sometimes they're a pixel 'off')
-- add modes 2 and 3 to gfx2 (lowres 4 color and 16 color)
+- c64: make the graphics.BITMAP_ADDRESS configurable (VIC banking)
+- get rid of all other TODO's in the code ;-)
+
+
+Low prio
+^^^^^^^^
+- optimize several inner loops in gfx2 even further?
+- add modes 2 and 3 to gfx2 (lowres 4 color and 16 color)?
 - add a flood fill routine to gfx2?
 - add a f_seek() routine for the Cx16 that uses its seek dos api?
-- refactor the asmgen into their own submodule?
-- refactor the compiler optimizers into their own submodule?
-- optimizer: detect variables that are written but never read - mark those as unused too and remove them, such as uword unused = memory("unused222", 20) - also remove the memory slab allocation
-- add a compiler option to not remove unused subroutines. this allows for building library programs
+- refactor the asmgen into own submodule
+- refactor the compiler optimizers into own submodule
+- add a compiler option to not remove unused subroutines. this allows for building library programs. But this won't work with 64tass's .proc ...
 - make it possible to use cpu opcodes such as 'nop' as variable names by prefixing all asm vars with something such as ``v_``
-- option to load the built-in library files from a directory instead of the embedded ones (for easier library development/debugging)
-- c64: make the graphics.BITMAP_ADDRESS configurable (VIC banking)
-- 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.
 
+- find a way to optimize if-statement codegen so that "if var & %10000" doesn't use stack & subroutine call, but also that the simple case "if X {...}" remains fast
+- optimizer: detect variables that are written but never read - mark those as unused too and remove them, such as uword unused = memory("unused222", 20) - also remove the memory slab allocation
 - further optimize assignment codegeneration, such as the following:
 - rewrite expression code generator to not use eval stack but a fixed number of predetermined value 'variables' (1 per nesting level?)
 - binexpr splitting (beware self-referencing expressions and asm code ballooning though)

examples/balloonflight.p8

@@ -42,7 +42,7 @@ main {
                 active_height--
                 upwards = false
             } else {
-                target_height = 8 + rnd() % 16
+                target_height = 8 + fastrnd8() % 16
                 if upwards
                     mountain = 233
                 else
@@ -57,7 +57,7 @@ main {
             txt.scroll_left(true)
 
             ; float the balloon
-            if rnd() & %10000
+            if fastrnd8() & %10000
                 c64.SPXY[1] ++
             else
                 c64.SPXY[1] --
@@ -71,10 +71,10 @@ main {
                 txt.setcc(39, yy, 160, 8)        ; draw mountain
             }
 
-            yy = rnd()
+            yy = fastrnd8()
             if yy > 100 {
                 ; draw a star
-                txt.setcc(39, yy % (active_height-1), '.', rnd())
+                txt.setcc(39, yy % (active_height-1), '.', fastrnd8())
             }
 
             if yy > 200 {
@@ -85,7 +85,7 @@ main {
                     tree = 88
                 else if yy & %00100000 != 0
                     tree = 65
-                if rnd() > 130
+                if fastrnd8() > 130
                     treecolor = 13
                 txt.setcc(39, active_height, tree, treecolor)
             }

examples/balls.p8

@@ -24,12 +24,12 @@ main {
         ; Setup Starting Ball Positions
         ubyte lp
         for lp in 0 to ballCount-1 {
-            BX[lp] = rnd() % txt.DEFAULT_WIDTH
-            BY[lp] = rnd() % txt.DEFAULT_HEIGHT
-            BC[lp] = rnd() & 15
-            DX[lp] = rnd() & 1
-            DY[lp] = rnd() & 1
-            void rnd()
+            BX[lp] = fastrnd8() % txt.DEFAULT_WIDTH
+            BY[lp] = fastrnd8() % txt.DEFAULT_HEIGHT
+            BC[lp] = fastrnd8() & 15
+            DX[lp] = fastrnd8() & 1
+            DY[lp] = fastrnd8() & 1
+            void fastrnd8()
         }
 
         ; start clock

examples/cmp/comparison_ifs_byte.p8

@@ -1,110 +0,0 @@
-%import textio
-%zeropage basicsafe
-
-; Note: this program is compatible with C64 and CX16.
-
-main {
-
-    sub start()  {
-
-        byte v1
-        byte v2
-
-        v1 = 100
-        v2 = 127
-        if v1==v2
-            txt.print("error in 100==127!\n")
-        else
-            txt.print("ok: 100 not == 127\n")
-
-        if v1!=v2
-            txt.print("ok: 100 != 127\n")
-        else
-            txt.print("error in 100!=127!\n")
-
-        if v1<v2
-            txt.print("ok: 100 < 127\n")
-        else
-            txt.print("error in 100<127!\n")
-
-        if v1<=v2
-            txt.print("ok: 100 <= 127\n")
-        else
-            txt.print("error in 100<=127!\n")
-
-        if v1>v2
-            txt.print("error in 100>127!\n")
-        else
-            txt.print("ok: 100 is not >127\n")
-
-        if v1>=v2
-            txt.print("error in 100>=127!\n")
-        else
-            txt.print("ok: 100 is not >=127\n")
-
-
-        v1 = 125
-        v2 = 22
-        if v1==v2
-            txt.print("error in 125==22!\n")
-        else
-            txt.print("ok: 125 not == 22\n")
-
-        if v1!=v2
-            txt.print("ok: 125 != 22\n")
-        else
-            txt.print("error in 125!=22!\n")
-
-        if v1<v2
-            txt.print("error in 125<22!\n")
-        else
-            txt.print("ok: 125 is not < 22\n")
-
-        if v1<=v2
-            txt.print("error in 125<=22!\n")
-        else
-            txt.print("ok: 125 is not <= 22\n")
-
-        if v1>v2
-            txt.print("ok: 125 > 22\n")
-        else
-            txt.print("error in 125>22!\n")
-
-        if v1>=v2
-            txt.print("ok: 125 >= 22\n")
-        else
-            txt.print("error in 125>=22!\n")
-
-        v1 = 22
-        v2 = 22
-        if v1==v2
-            txt.print("ok: 22 == 22\n")
-        else
-            txt.print("error in 22==22!\n")
-
-        if v1!=v2
-            txt.print("error in 22!=22!\n")
-        else
-            txt.print("ok: 22 is not != 22\n")
-
-        if v1<v2
-            txt.print("error in 22<22!\n")
-        else
-            txt.print("ok: 22 is not < 22\n")
-
-        if v1<=v2
-            txt.print("ok: 22 <= 22\n")
-        else
-            txt.print("error in 22<=22!\n")
-
-        if v1>v2
-            txt.print("error in 22>22!\n")
-        else
-            txt.print("ok: 22 is not > 22\n")
-
-        if v1>=v2
-            txt.print("ok: 22 >= 22\n")
-        else
-            txt.print("error in 22>=22!\n")
-    }
-}

examples/cmp/comparison_ifs_float.p8

@@ -1,111 +0,0 @@
-%import textio
-%import floats
-%zeropage basicsafe
-
-; Note: this program is compatible with C64 and CX16.
-
-main {
-
-    sub start()  {
-
-        float v1
-        float v2
-
-        v1 = 1.11
-        v2 = 699.99
-        if v1==v2
-            txt.print("error in 1.11==699.99!\n")
-        else
-            txt.print("ok: 1.11 not == 699.99\n")
-
-        if v1!=v2
-            txt.print("ok: 1.11 != 699.99\n")
-        else
-            txt.print("error in 1.11!=699.99!\n")
-
-        if v1<v2
-            txt.print("ok: 1.11 < 699.99\n")
-        else
-            txt.print("error in 1.11<699.99!\n")
-
-        if v1<=v2
-            txt.print("ok: 1.11 <= 699.99\n")
-        else
-            txt.print("error in 1.11<=699.99!\n")
-
-        if v1>v2
-            txt.print("error in 1.11>699.99!\n")
-        else
-            txt.print("ok: 1.11 is not >699.99\n")
-
-        if v1>=v2
-            txt.print("error in 1.11>=699.99!\n")
-        else
-            txt.print("ok: 1.11 is not >=699.99\n")
-
-
-        v1 = 555.5
-        v2 = -22.2
-        if v1==v2
-            txt.print("error in 555.5==-22.2!\n")
-        else
-            txt.print("ok: 555.5 not == -22.2\n")
-
-        if v1!=v2
-            txt.print("ok: 555.5 != -22.2\n")
-        else
-            txt.print("error in 555.5!=-22.2!\n")
-
-        if v1<v2
-            txt.print("error in 555.5<-22.2!\n")
-        else
-            txt.print("ok: 555.5 is not < -22.2\n")
-
-        if v1<=v2
-            txt.print("error in 555.5<=-22.2!\n")
-        else
-            txt.print("ok: 555.5 is not <= -22.2\n")
-
-        if v1>v2
-            txt.print("ok: 555.5 > -22.2\n")
-        else
-            txt.print("error in 555.5>-22.2!\n")
-
-        if v1>=v2
-            txt.print("ok: 555.5 >= -22.2\n")
-        else
-            txt.print("error in 555.5>=-22.2!\n")
-
-        v1 = -22.2
-        v2 = -22.2
-        if v1==v2
-            txt.print("ok: -22.2 == -22.2\n")
-        else
-            txt.print("error in -22.2==-22.2!\n")
-
-        if v1!=v2
-            txt.print("error in -22.2!=-22.2!\n")
-        else
-            txt.print("ok: -22.2 is not != -22.2\n")
-
-        if v1<v2
-            txt.print("error in -22.2<-22.2!\n")
-        else
-            txt.print("ok: -22.2 is not < -22.2\n")
-
-        if v1<=v2
-            txt.print("ok: -22.2 <= -22.2\n")
-        else
-            txt.print("error in -22.2<=-22.2!\n")
-
-        if v1>v2
-            txt.print("error in -22.2>-22.2!\n")
-        else
-            txt.print("ok: -22.2 is not > -22.2\n")
-
-        if v1>=v2
-            txt.print("ok: -22.2 >= -22.2\n")
-        else
-            txt.print("error in -22.2>=-22.2!\n")
-    }
-}

examples/cmp/comparison_ifs_ubyte.p8

@@ -1,111 +0,0 @@
-%import textio
-%zeropage basicsafe
-
-
-; Note: this program is compatible with C64 and CX16.
-
-main {
-
-    sub start()  {
-
-        ubyte v1
-        ubyte v2
-
-        v1 = 100
-        v2 = 200
-        if v1==v2
-            txt.print("error in 100==200!\n")
-        else
-            txt.print("ok: 100 not == 200\n")
-
-        if v1!=v2
-            txt.print("ok: 100 != 200\n")
-        else
-            txt.print("error in 100!=200!\n")
-
-        if v1<v2
-            txt.print("ok: 100 < 200\n")
-        else
-            txt.print("error in 100<200!\n")
-
-        if v1<=v2
-            txt.print("ok: 100 <= 200\n")
-        else
-            txt.print("error in 100<=200!\n")
-
-        if v1>v2
-            txt.print("error in 100>200!\n")
-        else
-            txt.print("ok: 100 is not >200\n")
-
-        if v1>=v2
-            txt.print("error in 100>=200!\n")
-        else
-            txt.print("ok: 100 is not >=200\n")
-
-
-        v1 = 155
-        v2 = 22
-        if v1==v2
-            txt.print("error in 155==22!\n")
-        else
-            txt.print("ok: 155 not == 22\n")
-
-        if v1!=v2
-            txt.print("ok: 155 != 22\n")
-        else
-            txt.print("error in 155!=22!\n")
-
-        if v1<v2
-            txt.print("error in 155<22!\n")
-        else
-            txt.print("ok: 155 is not < 22\n")
-
-        if v1<=v2
-            txt.print("error in 155<=22!\n")
-        else
-            txt.print("ok: 155 is not <= 22\n")
-
-        if v1>v2
-            txt.print("ok: 155 > 22\n")
-        else
-            txt.print("error in 155>22!\n")
-
-        if v1>=v2
-            txt.print("ok: 155 >= 22\n")
-        else
-            txt.print("error in 155>=22!\n")
-
-        v1 = 22
-        v2 = 22
-        if v1==v2
-            txt.print("ok: 22 == 22\n")
-        else
-            txt.print("error in 22==22!\n")
-
-        if v1!=v2
-            txt.print("error in 22!=22!\n")
-        else
-            txt.print("ok: 22 is not != 22\n")
-
-        if v1<v2
-            txt.print("error in 22<22!\n")
-        else
-            txt.print("ok: 22 is not < 22\n")
-
-        if v1<=v2
-            txt.print("ok: 22 <= 22\n")
-        else
-            txt.print("error in 22<=22!\n")
-
-        if v1>v2
-            txt.print("error in 22>22!\n")
-        else
-            txt.print("ok: 22 is not > 22\n")
-
-        if v1>=v2
-            txt.print("ok: 22 >= 22\n")
-        else
-            txt.print("error in 22>=22!\n")
-    }
-}