move community note to start page

* feat: add routine to convert 24-bit integer from A/X/Y to float in FAC1

* fix: remove duplicate definition

* fix: shift to appropriate exponent

.idea/libraries/KotlinJavaRuntime.xml

@@ -1,23 +1,23 @@
 <component name="libraryTable">
   <library name="KotlinJavaRuntime" type="repository">
-    <properties maven-id="org.jetbrains.kotlin:kotlin-stdlib-jdk8:1.9.24" />
+    <properties maven-id="org.jetbrains.kotlin:kotlin-stdlib-jdk8:2.0.20" />
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/1.9.24/kotlin-stdlib-jdk8-1.9.24.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.9.24/kotlin-stdlib-1.9.24.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/2.0.20/kotlin-stdlib-jdk8-2.0.20.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/2.0.20/kotlin-stdlib-2.0.20.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/13.0/annotations-13.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/1.9.24/kotlin-stdlib-jdk7-1.9.24.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/2.0.20/kotlin-stdlib-jdk7-2.0.20.jar!/" />
     </CLASSES>
     <JAVADOC>
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/1.9.24/kotlin-stdlib-jdk8-1.9.24-javadoc.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.9.24/kotlin-stdlib-1.9.24-javadoc.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/2.0.20/kotlin-stdlib-jdk8-2.0.20-javadoc.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/2.0.20/kotlin-stdlib-2.0.20-javadoc.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/13.0/annotations-13.0-javadoc.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/1.9.24/kotlin-stdlib-jdk7-1.9.24-javadoc.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/2.0.20/kotlin-stdlib-jdk7-2.0.20-javadoc.jar!/" />
     </JAVADOC>
     <SOURCES>
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/1.9.24/kotlin-stdlib-jdk8-1.9.24-sources.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.9.24/kotlin-stdlib-1.9.24-sources.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/2.0.20/kotlin-stdlib-jdk8-2.0.20-sources.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/2.0.20/kotlin-stdlib-2.0.20-sources.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/13.0/annotations-13.0-sources.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/1.9.24/kotlin-stdlib-jdk7-1.9.24-sources.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/2.0.20/kotlin-stdlib-jdk7-2.0.20-sources.jar!/" />
     </SOURCES>
   </library>
 </component>

.idea/libraries/antlr_antlr4.xml

@@ -1,13 +1,13 @@
 <component name="libraryTable">
   <library name="antlr.antlr4" type="repository">
-    <properties maven-id="org.antlr:antlr4:4.13.1">
+    <properties maven-id="org.antlr:antlr4:4.13.2">
       <exclude>
         <dependency maven-id="com.ibm.icu:icu4j" />
       </exclude>
     </properties>
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/org/antlr/antlr4/4.13.1/antlr4-4.13.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/antlr/antlr4-runtime/4.13.1/antlr4-runtime-4.13.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/antlr/antlr4/4.13.2/antlr4-4.13.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/antlr/antlr4-runtime/4.13.2/antlr4-runtime-4.13.2.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/antlr/antlr-runtime/3.5.3/antlr-runtime-3.5.3.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/antlr/ST4/4.3.4/ST4-4.3.4.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/abego/treelayout/org.abego.treelayout.core/1.0.3/org.abego.treelayout.core-1.0.3.jar!/" />

.idea/libraries/io_kotest_assertions_core_jvm.xml

@@ -1,17 +1,17 @@
 <component name="libraryTable">
   <library name="io.kotest.assertions.core.jvm" type="repository">
-    <properties maven-id="io.kotest:kotest-assertions-core-jvm:5.9.0" />
+    <properties maven-id="io.kotest:kotest-assertions-core-jvm:5.9.1" />
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-core-jvm/5.9.0/kotest-assertions-core-jvm-5.9.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-shared-jvm/5.9.0/kotest-assertions-shared-jvm-5.9.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-core-jvm/5.9.1/kotest-assertions-core-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-shared-jvm/5.9.1/kotest-assertions-shared-jvm-5.9.1.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/io/github/java-diff-utils/java-diff-utils/4.12/java-diff-utils-4.12.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/opentest4j/opentest4j/1.3.0/opentest4j-1.3.0.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.9.23/kotlin-stdlib-1.9.23.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/13.0/annotations-13.0.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-jdk8/1.8.0/kotlinx-coroutines-jdk8-1.8.0.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-reflect/1.9.23/kotlin-reflect-1.9.23.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-common-jvm/5.9.0/kotest-common-jvm-5.9.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-api-jvm/5.9.0/kotest-assertions-api-jvm-5.9.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-common-jvm/5.9.1/kotest-common-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-api-jvm/5.9.1/kotest-assertions-api-jvm-5.9.1.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-core-jvm/1.8.0/kotlinx-coroutines-core-jvm-1.8.0.jar!/" />
     </CLASSES>
     <JAVADOC />

.idea/libraries/io_kotest_framework_datatest.xml

@@ -0,0 +1,23 @@
+<component name="libraryTable">
+  <library name="io.kotest.framework.datatest" type="repository">
+    <properties maven-id="io.kotest:kotest-framework-datatest:5.9.1" />
+    <CLASSES>
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-datatest/5.9.1/kotest-framework-datatest-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-datatest-jvm/5.9.1/kotest-framework-datatest-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.9.23/kotlin-stdlib-1.9.23.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/13.0/annotations-13.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-core-jvm/1.8.0/kotlinx-coroutines-core-jvm-1.8.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-common-jvm/5.9.1/kotest-common-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-reflect/1.9.23/kotlin-reflect-1.9.23.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-api-jvm/5.9.1/kotest-framework-api-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-shared-jvm/5.9.1/kotest-assertions-shared-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-api-jvm/5.9.1/kotest-assertions-api-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-jdk8/1.8.0/kotlinx-coroutines-jdk8-1.8.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/github/java-diff-utils/java-diff-utils/4.12/java-diff-utils-4.12.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/opentest4j/opentest4j/1.3.0/opentest4j-1.3.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-test-jvm/1.8.0/kotlinx-coroutines-test-jvm-1.8.0.jar!/" />
+    </CLASSES>
+    <JAVADOC />
+    <SOURCES />
+  </library>
+</component>

.idea/libraries/io_kotest_runner_junit5_jvm.xml

@@ -1,14 +1,14 @@
 <component name="libraryTable">
   <library name="io.kotest.runner.junit5.jvm" type="repository">
-    <properties maven-id="io.kotest:kotest-runner-junit5-jvm:5.9.0" />
+    <properties maven-id="io.kotest:kotest-runner-junit5-jvm:5.9.1" />
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-runner-junit5-jvm/5.9.0/kotest-runner-junit5-jvm-5.9.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-api-jvm/5.9.0/kotest-framework-api-jvm-5.9.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-shared-jvm/5.9.0/kotest-assertions-shared-jvm-5.9.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-runner-junit5-jvm/5.9.1/kotest-runner-junit5-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-api-jvm/5.9.1/kotest-framework-api-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-shared-jvm/5.9.1/kotest-assertions-shared-jvm-5.9.1.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/io/github/java-diff-utils/java-diff-utils/4.12/java-diff-utils-4.12.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-test-jvm/1.8.0/kotlinx-coroutines-test-jvm-1.8.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-common-jvm/5.9.0/kotest-common-jvm-5.9.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-engine-jvm/5.9.0/kotest-framework-engine-jvm-5.9.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-common-jvm/5.9.1/kotest-common-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-engine-jvm/5.9.1/kotest-framework-engine-jvm-5.9.1.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/io/github/classgraph/classgraph/4.8.172/classgraph-4.8.172.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/com/github/ajalt/mordant/1.2.1/mordant-1.2.1.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/com/github/ajalt/colormath/1.2.0/colormath-1.2.0.jar!/" />
@@ -18,12 +18,12 @@
       <root url="jar://$MAVEN_REPOSITORY$/net/java/dev/jna/jna-platform/5.9.0/jna-platform-5.9.0.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/net/bytebuddy/byte-buddy/1.10.9/byte-buddy-1.10.9.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/net/bytebuddy/byte-buddy-agent/1.10.9/byte-buddy-agent-1.10.9.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-discovery-jvm/5.9.0/kotest-framework-discovery-jvm-5.9.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-core-jvm/5.9.0/kotest-assertions-core-jvm-5.9.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-discovery-jvm/5.9.1/kotest-framework-discovery-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-core-jvm/5.9.1/kotest-assertions-core-jvm-5.9.1.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-jdk8/1.8.0/kotlinx-coroutines-jdk8-1.8.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-api-jvm/5.9.0/kotest-assertions-api-jvm-5.9.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-extensions-jvm/5.9.0/kotest-extensions-jvm-5.9.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-concurrency-jvm/5.9.0/kotest-framework-concurrency-jvm-5.9.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-api-jvm/5.9.1/kotest-assertions-api-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-extensions-jvm/5.9.1/kotest-extensions-jvm-5.9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-concurrency-jvm/5.9.1/kotest-framework-concurrency-jvm-5.9.1.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-core-jvm/1.8.0/kotlinx-coroutines-core-jvm-1.8.0.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/23.0.0/annotations-23.0.0.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/junit/platform/junit-platform-engine/1.8.2/junit-platform-engine-1.8.2.jar!/" />

README.md

@@ -29,6 +29,8 @@ How to get it/build it
 - Or, if you want/need a bleeding edge development version, you can:
   - download a build artifact zipfile from a recent [github action build](https://github.com/irmen/prog8/actions).
   - you can also compile it yourself from source. [Instructions here](https://prog8.readthedocs.io/en/latest/compiling.html).
+    Note that if you are not using *gradle* to build it, you might have to perform some manual
+    tasks once to make it compile fully. These are explained in the linked instructions. 
 - Alternatively, you can also install the compiler as a package on some linux distros:
   - Arch (via AUR): [`prog8`](https://aur.archlinux.org/packages/prog8)
 
@@ -53,6 +55,7 @@ What does Prog8 provide?
 
 - all advantages of a higher level language over having to write assembly code manually
 - programs run very fast because compilation to native machine code
+- code often is smaller and faster than equivalent C code compiled with CC65 or even LLVM-MOS
 - modularity, symbol scoping, subroutines
 - various data types other than just bytes (16-bit words, floats, strings)
 - floating point math is supported if the target system provides floating point library routines (C64 and Cx16 both do)
@@ -66,7 +69,7 @@ What does Prog8 provide?
 - conditional branches
 - ``when`` statement to provide a concise jump table alternative to if/elseif chains
 - ``in`` expression for concise and efficient multi-value/containment check 
-- many built-in functions such as ``sin``, ``cos``, ``rnd``, ``abs``, ``min``, ``max``, ``sqrt``, ``msb``, ``rol``, ``ror``, ``swap``, ``sort`` and ``reverse``
+- several specialized built-in functions such as ``lsb``, ``msb``, ``min``, ``max``, ``rol``, ``ror``
 - 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
 - inline assembly allows you to have full control when every cycle or byte matters

codeCore/src/prog8/code/SymbolTable.kt

@@ -222,7 +222,7 @@ class StMemVar(name: String,
     init{
         require(dt!=DataType.BOOL && dt!=DataType.ARRAY_BOOL)
         if(dt in ArrayDatatypes || dt == DataType.STR)
-            require(length!=null) { "memory mapped array or string must have known length" }
+            requireNotNull(length)
     }
 }
 

codeCore/src/prog8/code/SymbolTableMaker.kt

@@ -3,7 +3,8 @@ package prog8.code
 import prog8.code.ast.*
 import prog8.code.core.*
 import prog8.code.target.VMTarget
-import java.util.*
+import kotlin.collections.ArrayDeque
+
 
 class SymbolTableMaker(private val program: PtProgram, private val options: CompilationOptions) {
     fun make(): SymbolTable {
@@ -13,8 +14,8 @@ class SymbolTableMaker(private val program: PtProgram, private val options: Comp
             st.add(StNode(it.key, StNodeType.BUILTINFUNC, PtIdentifier(it.key, it.value.returnType ?: DataType.UNDEFINED, Position.DUMMY)))
         }
 
-        val scopestack = Stack<StNode>()
-        scopestack.push(st)
+        val scopestack = ArrayDeque<StNode>()
+        scopestack.add(st)
         program.children.forEach {
             addToSt(it, scopestack)
         }
@@ -35,7 +36,7 @@ class SymbolTableMaker(private val program: PtProgram, private val options: Comp
         return st
     }
 
-    private fun addToSt(node: PtNode, scope: Stack<StNode>) {
+    private fun addToSt(node: PtNode, scope: ArrayDeque<StNode>) {
         val stNode = when(node) {
             is PtAsmSub -> {
                 val parameters = node.parameters.map { StRomSubParameter(it.first, it.second.type) }
@@ -104,7 +105,7 @@ class SymbolTableMaker(private val program: PtProgram, private val options: Comp
                     val size = (node.args[1] as PtNumber).number.toUInt()
                     val align = (node.args[2] as PtNumber).number.toUInt()
                     // don't add memory slabs in nested scope, just put them in the top level of the ST
-                    scope.firstElement().add(StMemorySlab("prog8_memoryslab_$slabname", size, align, node))
+                    scope.first().add(StMemorySlab("prog8_memoryslab_$slabname", size, align, node))
                 }
                 null
             }
@@ -112,14 +113,14 @@ class SymbolTableMaker(private val program: PtProgram, private val options: Comp
         }
 
         if(stNode!=null) {
-            scope.peek().add(stNode)
-            scope.push(stNode)
+            scope.last().add(stNode)
+            scope.add(stNode)
         }
         node.children.forEach {
             addToSt(it, scope)
         }
         if(stNode!=null)
-            scope.pop()
+            scope.removeLast()
     }
 
     private fun makeInitialArray(value: PtArray): List<StArrayElement> {

codeCore/src/prog8/code/ast/AstExpressions.kt

@@ -1,7 +1,7 @@
 package prog8.code.ast
 
 import prog8.code.core.*
-import java.util.*
+import java.util.Objects
 import kotlin.math.abs
 import kotlin.math.truncate
 
@@ -42,7 +42,18 @@ sealed class PtExpression(val type: DataType, position: Position) : PtNode(posit
                 else
                     other.left isSameAs left && other.right isSameAs right
             }
-            is PtContainmentCheck -> other is PtContainmentCheck && other.type==type && other.element isSameAs element && other.iterable isSameAs iterable
+            is PtContainmentCheck -> {
+                if(other !is PtContainmentCheck || other.type != type || !(other.needle isSameAs needle))
+                    false
+                else {
+                    if(haystackHeapVar!=null)
+                        other.haystackHeapVar!=null && other.haystackHeapVar!! isSameAs haystackHeapVar!!
+                    else if(haystackValues!=null)
+                        other.haystackValues!=null && other.haystackValues!! isSameAs haystackValues!!
+                    else
+                        false
+                }
+            }
             is PtIdentifier -> other is PtIdentifier && other.type==type && other.name==name
             is PtIrRegister -> other is PtIrRegister && other.type==type && other.register==register
             is PtMemoryByte -> other is PtMemoryByte && other.address isSameAs address
@@ -195,10 +206,17 @@ class PtBinaryExpression(val operator: String, type: DataType, position: Positio
 
 
 class PtContainmentCheck(position: Position): PtExpression(DataType.BOOL, position) {
-    val element: PtExpression
+    val needle: PtExpression
         get() = children[0] as PtExpression
-    val iterable: PtIdentifier
-        get() = children[1] as PtIdentifier
+    val haystackHeapVar: PtIdentifier?
+        get() = children[1] as? PtIdentifier
+    val haystackValues: PtArray?
+        get() = children[1] as? PtArray
+
+    companion object {
+        val MAX_SIZE_FOR_INLINE_CHECKS_BYTE = 5
+        val MAX_SIZE_FOR_INLINE_CHECKS_WORD = 4
+    }
 }
 
 
@@ -227,12 +245,6 @@ class PtMemoryByte(position: Position) : PtExpression(DataType.UBYTE, position)
 
 
 class PtBool(val value: Boolean, position: Position) : PtExpression(DataType.BOOL, position) {
-
-    companion object {
-        fun fromNumber(number: Number, position: Position): PtBool =
-            PtBool(number != 0.0, position)
-    }
-
     override fun hashCode(): Int = Objects.hash(type, value)
 
     override fun equals(other: Any?): Boolean {
@@ -267,12 +279,12 @@ class PtNumber(type: DataType, val number: Double, position: Position) : PtExpre
     override fun hashCode(): Int = Objects.hash(type, number)
 
     override fun equals(other: Any?): Boolean {
-        if(other==null || other !is PtNumber)
-            return false
+        return if(other==null || other !is PtNumber)
+            false
         else if(type!=DataType.BOOL && other.type!=DataType.BOOL)
-            return number==other.number
+            number==other.number
         else
-            return type==other.type && number==other.number
+            type==other.type && number==other.number
     }
 
     operator fun compareTo(other: PtNumber): Int = number.compareTo(other.number)
@@ -302,6 +314,7 @@ class PtRange(type: DataType, position: Position) : PtExpression(type, position)
     fun toConstantIntegerRange(): IntProgression? {
         fun makeRange(fromVal: Int, toVal: Int, stepVal: Int): IntProgression {
             return when {
+                fromVal == toVal -> fromVal .. toVal
                 fromVal <= toVal -> when {
                     stepVal <= 0 -> IntRange.EMPTY
                     stepVal == 1 -> fromVal..toVal

codeCore/src/prog8/code/ast/AstStatements.kt

@@ -5,6 +5,7 @@ import prog8.code.core.*
 
 sealed interface IPtSubroutine {
     val name: String
+    val scopedName: String
 }
 
 class PtAsmSub(

codeCore/src/prog8/code/core/BuiltinFunctions.kt

@@ -76,14 +76,14 @@ val BuiltinFunctions: Map<String, FSignature> = mapOf(
     "ror"       to FSignature(false, listOf(FParam("item", arrayOf(DataType.UBYTE, DataType.UWORD))), null),
     "rol2"      to FSignature(false, listOf(FParam("item", arrayOf(DataType.UBYTE, DataType.UWORD))), null),
     "ror2"      to FSignature(false, listOf(FParam("item", arrayOf(DataType.UBYTE, DataType.UWORD))), null),
-    "sort"      to FSignature(false, listOf(FParam("array", ArrayDatatypes)), null),
-    "reverse"   to FSignature(false, listOf(FParam("array", ArrayDatatypes)), null),
     // cmp returns a status in the carry flag, but not a proper return value
     "cmp"       to FSignature(false, listOf(FParam("value1", IntegerDatatypes), FParam("value2", NumericDatatypes)), null),
     "prog8_lib_stringcompare"     to FSignature(true, listOf(FParam("str1", arrayOf(DataType.STR)), FParam("str2", arrayOf(DataType.STR))), DataType.BYTE),
     "prog8_lib_arraycopy"         to FSignature(false, listOf(FParam("source", ArrayDatatypes), FParam("target", ArrayDatatypes)), null),
     "prog8_lib_square_byte"       to FSignature(true, listOf(FParam("value", arrayOf(DataType.BYTE, DataType.UBYTE))), DataType.UBYTE),
     "prog8_lib_square_word"       to FSignature(true, listOf(FParam("value", arrayOf(DataType.WORD, DataType.UWORD))), DataType.UWORD),
+    "prog8_ifelse_bittest_set"    to FSignature(true, listOf(FParam("variable", ByteDatatypes), FParam("bitnumber", arrayOf(DataType.UBYTE))), DataType.BOOL),
+    "prog8_ifelse_bittest_notset" to FSignature(true, listOf(FParam("variable", ByteDatatypes), FParam("bitnumber", arrayOf(DataType.UBYTE))), DataType.BOOL),
     "abs"       to FSignature(true, listOf(FParam("value", NumericDatatypes)), null),
     "abs__byte"  to FSignature(true, listOf(FParam("value", arrayOf(DataType.BYTE))), DataType.BYTE),
     "abs__word"  to FSignature(true, listOf(FParam("value", arrayOf(DataType.WORD))), DataType.WORD),
@@ -99,8 +99,6 @@ val BuiltinFunctions: Map<String, FSignature> = mapOf(
     "divmod"           to FSignature(false, listOf(FParam("dividend", arrayOf(DataType.UBYTE, DataType.UWORD)), FParam("divisor", arrayOf(DataType.UBYTE, DataType.UWORD)), FParam("quotient", arrayOf(DataType.UBYTE, DataType.UWORD)), FParam("remainder", arrayOf(DataType.UBYTE, DataType.UWORD))), null),
     "divmod__ubyte"    to FSignature(false, listOf(FParam("dividend", arrayOf(DataType.UBYTE)), FParam("divisor", arrayOf(DataType.UBYTE)), FParam("quotient", arrayOf(DataType.UBYTE)), FParam("remainder", arrayOf(DataType.UBYTE))), null),
     "divmod__uword"    to FSignature(false, listOf(FParam("dividend", arrayOf(DataType.UWORD)), FParam("divisor", arrayOf(DataType.UWORD)), FParam("quotient", arrayOf(DataType.UWORD)), FParam("remainder", arrayOf(DataType.UWORD))), null),
-    "any"       to FSignature(true, listOf(FParam("values", ArrayDatatypes)), DataType.BOOL),
-    "all"       to FSignature(true, listOf(FParam("values", ArrayDatatypes)), DataType.BOOL),
     "lsb"       to FSignature(true, listOf(FParam("value", arrayOf(DataType.UWORD, DataType.WORD))), DataType.UBYTE),
     "msb"       to FSignature(true, listOf(FParam("value", arrayOf(DataType.UWORD, DataType.WORD))), DataType.UBYTE),
     "mkword"    to FSignature(true, listOf(FParam("msb", arrayOf(DataType.UBYTE)), FParam("lsb", arrayOf(DataType.UBYTE))), DataType.UWORD),
@@ -136,6 +134,5 @@ val BuiltinFunctions: Map<String, FSignature> = mapOf(
 val InplaceModifyingBuiltinFunctions = setOf(
     "setlsb", "setmsb",
     "rol", "ror", "rol2", "ror2",
-    "sort", "reverse",
     "divmod", "divmod__ubyte", "divmod__uword"
 )

codeCore/src/prog8/code/core/CompilationOptions.kt

@@ -27,7 +27,6 @@ class CompilationOptions(val output: OutputType,
                          var slabsHighBank: Int? = null,
                          var slabsGolden: Boolean = false,
                          var splitWordArrays: Boolean = false,
-                         var strictBool: Boolean = true,
                          var breakpointCpuInstruction: String? = null,
                          var outputDir: Path = Path(""),
                          var symbolDefs: Map<String, String> = emptyMap()

codeCore/src/prog8/code/core/Conversions.kt

@@ -1,6 +1,11 @@
 package prog8.code.core
 
 import kotlin.math.abs
+import kotlin.math.pow
+
+val powersOfTwoFloat = (0..16).map { (2.0).pow(it) }.toTypedArray()
+val negativePowersOfTwoFloat = powersOfTwoFloat.map { -it }.toTypedArray()
+val powersOfTwoInt = (0..16).map { 2.0.pow(it).toInt() }.toTypedArray()
 
 fun Number.toHex(): String {
     //  0..15 -> "0".."15"

codeCore/src/prog8/code/core/IMachineDefinition.kt

@@ -29,7 +29,6 @@ interface IMachineDefinition {
     fun convertFloatToBytes(num: Double): List<UByte>
     fun convertBytesToFloat(bytes: List<UByte>): Double
 
-    fun importLibs(compilerOptions: CompilationOptions, compilationTargetName: String): List<String>
     fun launchEmulator(selectedEmulator: Int, programNameWithPath: Path)
     fun isIOAddress(address: UInt): Boolean
 }

codeCore/src/prog8/code/core/IStringEncoding.kt

@@ -8,7 +8,8 @@ enum class Encoding(val prefix: String) {
     ISO("iso"),                 // cx16  (iso-8859-15)
     ISO5("iso5"),               // cx16  (iso-8859-5, cyrillic)
     ISO16("iso16"),             // cx16  (iso-8859-16, eastern european)
-    CP437("cp437")              // cx16  (ibm pc, codepage 437)
+    CP437("cp437"),             // cx16  (ibm pc, codepage 437)
+    KATAKANA("kata")                 // cx16  (katakana)
 }
 
 interface IStringEncoding {

codeCore/src/prog8/code/optimize/Optimizer.kt

@@ -10,8 +10,8 @@ fun optimizeIntermediateAst(program: PtProgram, options: CompilationOptions, st:
     if (!options.optimize)
         return
     while (errors.noErrors() &&
-        (optimizeCommonSubExpressions(program, errors)
-                + optimizeAssignTargets(program, st, errors)) > 0
+            (optimizeBitTest(program, options)
+            + optimizeAssignTargets(program, st, errors)) > 0
     ) {
         // keep rolling
     }
@@ -27,97 +27,6 @@ private fun walkAst(root: PtNode, act: (node: PtNode, depth: Int) -> Boolean) {
 }
 
 
-private var tempVarCounter = 0
-
-private fun optimizeCommonSubExpressions(program: PtProgram, errors: IErrorReporter): Int {
-
-    fun extractableSubExpr(expr: PtExpression): Boolean {
-        if(expr is PtArrayIndexer && expr.index.isSimple())
-            return false
-        if (expr is PtMemoryByte && expr.address.isSimple())
-            return false
-
-        val result = if(expr is PtBinaryExpression)
-            expr.type !in ByteDatatypes ||
-            !(expr.left.isSimple() && expr.right.isSimple()) ||
-            (expr.operator !in LogicalOperators && expr.operator !in BitwiseOperators)
-        else if (expr is PtArrayIndexer && expr.type !in ByteDatatypes)
-            true
-        else
-            !expr.isSimple()
-        return result
-    }
-
-    // for each Binaryexpression, recurse to find a common subexpression pair therein.
-    val commons = mutableMapOf<PtBinaryExpression, Pair<PtExpression, PtExpression>>()
-    walkAst(program) { node: PtNode, depth: Int ->
-        if(node is PtBinaryExpression) {
-            val subExpressions = mutableListOf<PtExpression>()
-            walkAst(node.left) { subNode: PtNode, subDepth: Int ->
-                if (subNode is PtExpression) {
-                    if(extractableSubExpr(subNode)) subExpressions.add(subNode)
-                    true
-                } else false
-            }
-            walkAst(node.right) { subNode: PtNode, subDepth: Int ->
-                if (subNode is PtExpression) {
-                    if(extractableSubExpr(subNode)) subExpressions.add(subNode)
-                    true
-                } else false
-            }
-
-            outer@for (first in subExpressions) {
-                for (second in subExpressions) {
-                    if (first!==second && first isSameAs second) {
-                        commons[node] = first to second
-                        break@outer     // do only 1 replacement at a time per binaryexpression
-                    }
-                }
-            }
-            false
-        } else true
-    }
-
-    // replace common subexpressions by a temp variable that is assigned only once.
-    // TODO: check for commonalities across multiple separate expressions in the current scope, not only inside a single line
-    commons.forEach { binexpr, (occurrence1, occurrence2) ->
-        val (stmtContainer, stmt) = findContainingStatements(binexpr)
-        val occurrence1idx = occurrence1.parent.children.indexOf(occurrence1)
-        val occurrence2idx = occurrence2.parent.children.indexOf(occurrence2)
-        val containerScopedName = findScopeName(stmtContainer)
-        tempVarCounter++
-        val tempvarName = "prog8_subexprvar_$tempVarCounter"
-        // TODO: some tempvars could be reused, if they are from different lines
-
-        val datatype = occurrence1.type
-        val singleReplacement1 = PtIdentifier("$containerScopedName.$tempvarName", datatype, occurrence1.position)
-        val singleReplacement2 = PtIdentifier("$containerScopedName.$tempvarName", datatype, occurrence2.position)
-        occurrence1.parent.children[occurrence1idx] = singleReplacement1
-        singleReplacement1.parent = occurrence1.parent
-        occurrence2.parent.children[occurrence2idx] = singleReplacement2
-        singleReplacement2.parent = occurrence2.parent
-
-        val tempassign = PtAssignment(binexpr.position).also { assign ->
-            assign.add(PtAssignTarget(false, binexpr.position).also { tgt->
-                tgt.add(PtIdentifier("$containerScopedName.$tempvarName", datatype, binexpr.position))
-            })
-            assign.add(occurrence1)
-            occurrence1.parent = assign
-        }
-        stmtContainer.children.add(stmtContainer.children.indexOf(stmt), tempassign)
-        tempassign.parent = stmtContainer
-
-        val tempvar = PtVariable(tempvarName, datatype, ZeropageWish.NOT_IN_ZEROPAGE, null, null, binexpr.position)
-        stmtContainer.add(0, tempvar)
-        tempvar.parent = stmtContainer
-
-        // errors.info("common subexpressions replaced by a tempvar, maybe simplify the expression manually", binexpr.position)
-    }
-
-    return commons.size
-}
-
-
 private fun optimizeAssignTargets(program: PtProgram, st: SymbolTable, errors: IErrorReporter): Int {
     var changes = 0
     walkAst(program) { node: PtNode, depth: Int ->
@@ -164,6 +73,51 @@ private fun optimizeAssignTargets(program: PtProgram, st: SymbolTable, errors: I
     return changes
 }
 
+
+private fun optimizeBitTest(program: PtProgram, options: CompilationOptions): Int {
+    if(options.compTarget.machine.cpu == CpuType.VIRTUAL)
+        return 0        // the special bittest optimization is not yet valid for the IR
+
+    var changes = 0
+    var recurse = true
+    walkAst(program) { node: PtNode, depth: Int ->
+        if(node is PtIfElse) {
+            val condition = node.condition as? PtBinaryExpression
+            if(condition!=null && (condition.operator=="==" || condition.operator=="!=")) {
+                if(condition.right.asConstInteger()==0) {
+                    val and = condition.left as? PtBinaryExpression
+                    if(and != null && and.operator=="&" && and.type == DataType.UBYTE) {
+                        val variable = and.left as? PtIdentifier
+                        val bitmask = and.right.asConstInteger()
+                        if(variable!=null && variable.type in ByteDatatypes && (bitmask==128 || bitmask==64)) {
+                            val setOrNot = if(condition.operator=="!=") "set" else "notset"
+                            val index = node.parent.children.indexOf(node)
+                            val bittestCall = PtBuiltinFunctionCall("prog8_ifelse_bittest_$setOrNot", false, true, DataType.BOOL, node.condition.position)
+                            bittestCall.add(variable)
+                            if(bitmask==128)
+                                bittestCall.add(PtNumber(DataType.UBYTE, 7.0, and.right.position))
+                            else
+                                bittestCall.add(PtNumber(DataType.UBYTE, 6.0, and.right.position))
+                            val ifElse = PtIfElse(node.position)
+                            ifElse.add(bittestCall)
+                            ifElse.add(node.ifScope)
+                            if(node.hasElse())
+                                ifElse.add(node.elseScope)
+                            node.parent.children[index] = ifElse
+                            ifElse.parent = node.parent
+                            changes++
+                            recurse = false
+                        }
+                    }
+                }
+            }
+        }
+        recurse
+    }
+    return changes
+}
+
+
 internal fun isSame(identifier: PtIdentifier, type: DataType, returnedRegister: RegisterOrPair): Boolean {
     if(returnedRegister in Cx16VirtualRegisters) {
         val regname = returnedRegister.name.lowercase()
@@ -189,24 +143,3 @@ internal fun isSame(identifier: PtIdentifier, type: DataType, returnedRegister:
     }
     return false   // there are no identifiers directly corresponding to cpu registers
 }
-
-
-internal fun findScopeName(node: PtNode): String {
-    var parent=node
-    while(parent !is PtNamedNode)
-        parent = parent.parent
-    return parent.scopedName
-}
-
-
-internal fun findContainingStatements(node: PtNode): Pair<PtNode, PtNode> {      // returns (parentstatementcontainer, childstatement)
-    var parent = node.parent
-    var child = node
-    while(true) {
-        if(parent is IPtStatementContainer) {
-            return parent to child
-        }
-        child=parent
-        parent=parent.parent
-    }
-}

codeCore/src/prog8/code/target/Encoder.kt

@@ -19,6 +19,7 @@ object Encoder: IStringEncoding {
             Encoding.ISO5 -> IsoCyrillicEncoding.encode(str)
             Encoding.ISO16 -> IsoEasternEncoding.encode(str)
             Encoding.CP437 -> Cp437Encoding.encode(str)
+            Encoding.KATAKANA -> KatakanaEncoding.encode(str)
             else -> throw InternalCompilerException("unsupported encoding $encoding")
         }
         return coded.fold(
@@ -35,6 +36,7 @@ object Encoder: IStringEncoding {
             Encoding.ISO5 -> IsoCyrillicEncoding.decode(bytes)
             Encoding.ISO16 -> IsoEasternEncoding.decode(bytes)
             Encoding.CP437 -> Cp437Encoding.decode(bytes)
+            Encoding.KATAKANA -> KatakanaEncoding.decode(bytes)
             else -> throw InternalCompilerException("unsupported encoding $encoding")
         }
         return decoded.fold(

codeCore/src/prog8/code/target/atari/AtariMachineDefinition.kt

@@ -25,13 +25,6 @@ class AtariMachineDefinition: IMachineDefinition {
     override fun convertFloatToBytes(num: Double): List<UByte> = TODO("atari float to bytes")
     override fun convertBytesToFloat(bytes: List<UByte>): Double = TODO("atari bytes to float")
 
-    override fun importLibs(compilerOptions: CompilationOptions, compilationTargetName: String): List<String> {
-        return if (compilerOptions.output == OutputType.XEX)
-            listOf("syslib")
-        else
-            emptyList()
-    }
-
     override fun launchEmulator(selectedEmulator: Int, programNameWithPath: Path) {
         val emulatorName: String
         val cmdline: List<String>

codeCore/src/prog8/code/target/c128/C128MachineDefinition.kt

@@ -36,13 +36,6 @@ class C128MachineDefinition: IMachineDefinition {
         return m5.toDouble()
     }
 
-    override fun importLibs(compilerOptions: CompilationOptions, compilationTargetName: String): List<String> {
-        return if (compilerOptions.launcher == CbmPrgLauncherType.BASIC || compilerOptions.output == OutputType.PRG)
-            listOf("syslib")
-        else
-            emptyList()
-    }
-
     override fun launchEmulator(selectedEmulator: Int, programNameWithPath: Path) {
         if(selectedEmulator!=1) {
             System.err.println("The c128 target only supports the main emulator (Vice).")

codeCore/src/prog8/code/target/c64/C64MachineDefinition.kt

@@ -37,13 +37,6 @@ class C64MachineDefinition: IMachineDefinition {
         return m5.toDouble()
     }
 
-    override fun importLibs(compilerOptions: CompilationOptions, compilationTargetName: String): List<String> {
-        return if (compilerOptions.launcher == CbmPrgLauncherType.BASIC || compilerOptions.output == OutputType.PRG)
-            listOf("syslib")
-        else
-            emptyList()
-    }
-
     override fun launchEmulator(selectedEmulator: Int, programNameWithPath: Path) {
         if(selectedEmulator!=1) {
             System.err.println("The c64 target only supports the main emulator (Vice).")

codeCore/src/prog8/code/target/cx16/CX16MachineDefinition.kt

@@ -36,13 +36,6 @@ class CX16MachineDefinition: IMachineDefinition {
         return m5.toDouble()
     }
 
-    override fun importLibs(compilerOptions: CompilationOptions, compilationTargetName: String): List<String> {
-        return if (compilerOptions.launcher == CbmPrgLauncherType.BASIC || compilerOptions.output == OutputType.PRG)
-            listOf("syslib")
-        else
-            emptyList()
-    }
-
     override fun launchEmulator(selectedEmulator: Int, programNameWithPath: Path) {
         val emulator: String
         val extraArgs: List<String>

codeCore/src/prog8/code/target/encodings/KatakanaEncoding.kt

@@ -0,0 +1,122 @@
+package prog8.code.target.encodings
+
+import com.github.michaelbull.result.Err
+import com.github.michaelbull.result.Ok
+import com.github.michaelbull.result.Result
+import java.io.CharConversionException
+import java.nio.charset.Charset
+
+
+object JapaneseCharacterConverter {
+    // adapted from https://github.com/raminduw/Japanese-Character-Converter
+
+    private val ZENKAKU_KATAKANA = charArrayOf(
+        'ァ', 'ア', 'ィ', 'イ', 'ゥ',
+        'ウ', 'ェ', 'エ', 'ォ', 'オ', 'カ', 'ガ', 'キ', 'ギ', 'ク', 'グ', 'ケ', 'ゲ',
+        'コ', 'ゴ', 'サ', 'ザ', 'シ', 'ジ', 'ス', 'ズ', 'セ', 'ゼ', 'ソ', 'ゾ', 'タ',
+        'ダ', 'チ', 'ヂ', 'ッ', 'ツ', 'ヅ', 'テ', 'デ', 'ト', 'ド', 'ナ', 'ニ', 'ヌ',
+        'ネ', 'ノ', 'ハ', 'バ', 'パ', 'ヒ', 'ビ', 'ピ', 'フ', 'ブ', 'プ', 'ヘ', 'ベ',
+        'ペ', 'ホ', 'ボ', 'ポ', 'マ', 'ミ', 'ム', 'メ', 'モ', 'ャ', 'ヤ', 'ュ', 'ユ',
+        'ョ', 'ヨ', 'ラ', 'リ', 'ル', 'レ', 'ロ', 'ヮ', 'ワ', 'ヰ', 'ヱ', 'ヲ', 'ン',
+        'ヴ', 'ヵ', 'ヶ'
+    )
+
+    private val HANKAKU_HIRAGANA = charArrayOf(
+        'ぁ', 'あ', 'ぃ', 'い', 'ぅ', 'う', 'ぇ', 'え',
+        'ぉ', 'お', 'か', 'が', 'き', 'ぎ', 'く', 'ぐ',
+        'け', 'げ', 'こ', 'ご', 'さ', 'ざ', 'し', 'じ',
+        'す', 'ず', 'せ', 'ぜ', 'そ', 'ぞ', 'た', 'だ',
+        'ち', 'ぢ', 'っ', 'つ', 'づ', 'て', 'で', 'と',
+        'ど', 'な', 'に', 'ぬ', 'ね', 'の', 'は', 'ば',
+        'ぱ', 'ひ', 'び', 'ぴ', 'ふ', 'ぶ', 'ぷ', 'へ',
+        'べ', 'ぺ', 'ほ', 'ぼ', 'ぽ', 'ま', 'み', 'む',
+        'め', 'も', 'ゃ', 'や', 'ゅ', 'ゆ', 'ょ', 'よ',
+        'ら', 'り', 'る', 'れ', 'ろ', 'ゎ', 'わ', 'ゐ',
+        'ゑ', 'を', 'ん', 'ゔ', 'ゕ', 'ゖ'
+    )
+
+    private val HANKAKU_KATAKANA = arrayOf(
+        "ｧ", "ｱ", "ｨ", "ｲ", "ｩ",
+        "ｳ", "ｪ", "ｴ", "ｫ", "ｵ", "ｶ", "ｶﾞ", "ｷ", "ｷﾞ", "ｸ", "ｸﾞ", "ｹ",
+        "ｹﾞ", "ｺ", "ｺﾞ", "ｻ", "ｻﾞ", "ｼ", "ｼﾞ", "ｽ", "ｽﾞ", "ｾ", "ｾﾞ", "ｿ",
+        "ｿﾞ", "ﾀ", "ﾀﾞ", "ﾁ", "ﾁﾞ", "ｯ", "ﾂ", "ﾂﾞ", "ﾃ", "ﾃﾞ", "ﾄ", "ﾄﾞ",
+        "ﾅ", "ﾆ", "ﾇ", "ﾈ", "ﾉ", "ﾊ", "ﾊﾞ", "ﾊﾟ", "ﾋ", "ﾋﾞ", "ﾋﾟ", "ﾌ",
+        "ﾌﾞ", "ﾌﾟ", "ﾍ", "ﾍﾞ", "ﾍﾟ", "ﾎ", "ﾎﾞ", "ﾎﾟ", "ﾏ", "ﾐ", "ﾑ", "ﾒ",
+        "ﾓ", "ｬ", "ﾔ", "ｭ", "ﾕ", "ｮ", "ﾖ", "ﾗ", "ﾘ", "ﾙ", "ﾚ", "ﾛ", "ﾜ",
+        "ﾜ", "ｲ", "ｴ", "ｦ", "ﾝ", "ｳﾞ", "ｶ", "ｹ"
+    )
+
+    private val ZENKAKU_KATAKANA_FIRST_CHAR_CODE = ZENKAKU_KATAKANA.first().code
+    private val HANKAKU_HIRAGANA_FIRST_CHAR_CODE = HANKAKU_HIRAGANA.first().code
+
+    private fun zenkakuKatakanaToHankakuKatakana(c: Char): String = if (c in ZENKAKU_KATAKANA) HANKAKU_KATAKANA[c.code - ZENKAKU_KATAKANA_FIRST_CHAR_CODE] else c.toString()
+    private fun hankakuKatakanaToZenkakuKatakana(c: Char): Char = if (c in HANKAKU_HIRAGANA) ZENKAKU_KATAKANA[c.code - HANKAKU_HIRAGANA_FIRST_CHAR_CODE] else c
+
+    fun zenkakuKatakanaToHankakuKatakana(s: String): String = buildString {
+        for (element in s) {
+            val converted = hankakuKatakanaToZenkakuKatakana(element)
+            val convertedChar = zenkakuKatakanaToHankakuKatakana(converted)
+            append(convertedChar)
+        }
+    }
+}
+
+object KatakanaEncoding {
+    val charset: Charset = Charset.forName("JIS_X0201")
+
+    fun encode(str: String): Result<List<UByte>, CharConversionException> {
+        return try {
+            val mapped = str.map { chr ->
+                when (chr) {
+
+                    '\u0000' -> 0u
+                    '\u00a0' -> 0xa0u // $a0 isn't technically a part of JIS X 0201 spec, and so we need to handle this ourselves
+
+                    '♥' -> 0xe3u
+                    '♦' -> 0xe4u
+                    '♣' -> 0xe5u
+                    '♠' -> 0xe6u
+
+                    '大' -> 0xeau
+                    '中' -> 0xebu
+                    '小' -> 0xecu
+                    '百' -> 0xedu
+                    '千' -> 0xeeu
+                    '万' -> 0xefu
+                    '♪' -> 0xf0u
+                    '土' -> 0xf1u
+                    '金' -> 0xf2u
+                    '木' -> 0xf3u
+                    '水' -> 0xf4u
+                    '火' -> 0xf5u
+                    '月' -> 0xf6u
+                    '日' -> 0xf7u
+                    '時' -> 0xf8u
+                    '分' -> 0xf9u
+                    '秒' -> 0xfau
+                    '年' -> 0xfbu
+                    '円' -> 0xfcu
+                    '人' -> 0xfdu
+                    '生' -> 0xfeu
+                    '〒' -> 0xffu
+                    in '\u8000'..'\u80ff' -> {
+                        // special case: take the lower 8 bit hex value directly
+                        (chr.code - 0x8000).toUByte()
+                    }
+                    else -> charset.encode(chr.toString())[0].toUByte()
+                }
+            }
+            Ok(mapped)
+        } catch (ce: CharConversionException) {
+            Err(ce)
+        }
+    }
+
+    fun decode(bytes: Iterable<UByte>): Result<String, CharConversionException> {
+        return try {
+            Ok(String(bytes.map { it.toByte() }.toByteArray(), charset))
+        } catch (ce: CharConversionException) {
+            Err(ce)
+        }
+    }
+}

codeCore/src/prog8/code/target/pet/PETMachineDefinition.kt

@@ -36,13 +36,6 @@ class PETMachineDefinition: IMachineDefinition {
         return m5.toDouble()
     }
 
-    override fun importLibs(compilerOptions: CompilationOptions, compilationTargetName: String): List<String> {
-        return if (compilerOptions.launcher == CbmPrgLauncherType.BASIC || compilerOptions.output == OutputType.PRG)
-            listOf("syslib")
-        else
-            emptyList()
-    }
-
     override fun launchEmulator(selectedEmulator: Int, programNameWithPath: Path) {
         if(selectedEmulator!=1) {
             System.err.println("The pet target only supports the main emulator (Vice).")

codeCore/src/prog8/code/target/virtual/VirtualMachineDefinition.kt

@@ -50,10 +50,6 @@ class VirtualMachineDefinition: IMachineDefinition {
         return Double.fromBits(b0 or b1 or b2 or b3 or b4 or b5 or b6 or b7)
     }
 
-    override fun importLibs(compilerOptions: CompilationOptions, compilationTargetName: String): List<String> {
-        return listOf("syslib")
-    }
-
     override fun launchEmulator(selectedEmulator: Int, programNameWithPath: Path) {
         println("\nStarting Virtual Machine...")
         // to not have external module dependencies in our own module, we launch the virtual machine via reflection

codeGenCpu6502/build.gradle

@@ -27,7 +27,8 @@ dependencies {
     // implementation "org.jetbrains.kotlin:kotlin-reflect"
     implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:2.0.0"
 
-    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.9.0'
+    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.9.1'
+    testImplementation 'io.kotest:kotest-framework-datatest:5.9.1'
     testImplementation 'org.junit.jupiter:junit-jupiter:5.9.2'
     testRuntimeOnly 'org.junit.platform:junit-platform-launcher'
 }

codeGenCpu6502/codeGenCpu6502.iml

@@ -14,5 +14,6 @@
     <orderEntry type="library" name="michael.bull.kotlin.result.jvm" level="project" />
     <orderEntry type="library" name="io.kotest.assertions.core.jvm" level="project" />
     <orderEntry type="library" name="io.kotest.runner.junit5.jvm" level="project" />
+    <orderEntry type="library" name="io.kotest.framework.datatest" level="project" />
   </component>
 </module>

codeGenCpu6502/src/prog8/codegen/cpu6502/AsmGen.kt

@@ -9,7 +9,7 @@ import prog8.code.ast.*
 import prog8.code.core.*
 import prog8.code.target.Cx16Target
 import prog8.codegen.cpu6502.assignment.*
-import java.util.*
+import kotlin.collections.ArrayDeque
 import kotlin.io.path.Path
 import kotlin.io.path.writeLines
 
@@ -232,7 +232,7 @@ class AsmGen6502Internal (
     private val functioncallAsmGen = FunctionCallAsmGen(program, this)
     private val programGen = ProgramAndVarsGen(program, options, errors, symbolTable, functioncallAsmGen, this, allocator, zeropage)
     private val anyExprGen = AnyExprAsmGen(this)
-    private val assignmentAsmGen = AssignmentAsmGen(program, this, options, anyExprGen, allocator)
+    private val assignmentAsmGen = AssignmentAsmGen(program, this, anyExprGen, allocator)
     private val builtinFunctionsAsmGen = BuiltinFunctionsAsmGen(program, this, assignmentAsmGen)
     private val ifElseAsmgen = IfElseAsmGen(program, symbolTable, this, allocator, assignmentAsmGen, errors)
 
@@ -374,7 +374,7 @@ class AsmGen6502Internal (
         val name = asmSymbolName(identifier.name)
 
         // see if we're inside a subroutine, if so, remove the whole prefix and just make the variable name locally scoped (64tass scopes it to the proper .proc block)
-        val subName = identifier.definingSub()?.scopedName
+        val subName = identifier.definingISub()?.scopedName
         return if (subName != null && name.length>subName.length && name.startsWith(subName) && name[subName.length] == '.')
             name.drop(subName.length + 1)
         else
@@ -385,14 +385,14 @@ class AsmGen6502Internal (
         val name = asmVariableName(identifier.name)
 
         // see if we're inside a subroutine, if so, remove the whole prefix and just make the variable name locally scoped (64tass scopes it to the proper .proc block)
-        val subName = identifier.definingSub()?.scopedName
+        val subName = identifier.definingISub()?.scopedName
         return if (subName != null && name.length>subName.length && name.startsWith(subName) && name[subName.length] == '.')
             name.drop(subName.length+1)
         else
             name
     }
 
-    fun asmVariableName(st: StNode, scope: PtSub?): String {
+    fun asmVariableName(st: StNode, scope: IPtSubroutine?): String {
         val name = asmVariableName(st.scopedName)
         if(scope==null)
             return name
@@ -772,7 +772,7 @@ class AsmGen6502Internal (
 
     private fun translate(stmt: PtRepeatLoop) {
         val endLabel = makeLabel("repeatend")
-        loopEndLabels.push(endLabel)
+        loopEndLabels.add(endLabel)
 
         when (stmt.count) {
             is PtNumber -> {
@@ -816,7 +816,7 @@ class AsmGen6502Internal (
             }
         }
 
-        loopEndLabels.pop()
+        loopEndLabels.removeLast()
     }
 
     private fun repeatWordCount(iterations: Int, stmt: PtRepeatLoop) {

codeGenCpu6502/src/prog8/codegen/cpu6502/AsmOptimizer.kt

@@ -50,6 +50,13 @@ internal fun optimizeAssembly(lines: MutableList<String>, machine: IMachineDefin
         numberOfOptimizations++
     }
 
+    mods = optimizeTSBtoRegularOr(linesByFour)
+    if(mods.isNotEmpty()) {
+        apply(mods, lines)
+        linesByFour = getLinesBy(lines, 4)
+        numberOfOptimizations++
+    }
+
     var linesByFourteen = getLinesBy(lines, 14)
     mods = optimizeSameAssignments(linesByFourteen, machine, symbolTable)
     if(mods.isNotEmpty()) {
@@ -384,6 +391,7 @@ private fun optimizeStoreLoadSame(
     for (lines in linesByFour) {
         val first = lines[1].value.trimStart()
         val second = lines[2].value.trimStart()
+        val third = lines[3].value.trimStart()
 
         // sta X + lda X,  sty X + ldy X,   stx X + ldx X  -> the second instruction can OFTEN be eliminated
         if ((first.startsWith("sta ") && second.startsWith("lda ")) ||
@@ -393,7 +401,6 @@ private fun optimizeStoreLoadSame(
                 (first.startsWith("ldy ") && second.startsWith("ldy ")) ||
                 (first.startsWith("ldx ") && second.startsWith("ldx "))
         ) {
-            val third = lines[3].value.trimStart()
             val attemptRemove =
                 if(third.isBranch()) {
                     // a branch instruction follows, we can only remove the load instruction if
@@ -508,6 +515,7 @@ private fun optimizeJsrRtsAndOtherCombinations(linesByFour: Sequence<List<Indexe
     for (lines in linesByFour) {
         val first = lines[0].value
         val second = lines[1].value
+        val third = lines[2].value
         if ((" jsr" in first || "\tjsr" in first ) && (" rts" in second || "\trts" in second)) {
             mods += Modification(lines[0].index, false, lines[0].value.replace("jsr", "jmp"))
             mods += Modification(lines[1].index, true, null)
@@ -577,6 +585,58 @@ private fun optimizeJsrRtsAndOtherCombinations(linesByFour: Sequence<List<Indexe
                 mods += Modification(lines[3].index, true, null)
             }
         }
+
+
+        fun sameLabel(branchInstr: String, jumpInstr: String, labelInstr: String): Boolean {
+            if('(' in jumpInstr) return false       // indirect jump cannot be replaced
+            val label = labelInstr.trimEnd().substringBefore(':').substringBefore(' ').substringBefore('\t')
+            val branchLabel = branchInstr.trimStart().substring(3).trim()
+            return label==branchLabel
+        }
+
+        // beq Label + jmp Addr + Label  -> bne Addr
+        if((" jmp" in second || "\tjmp " in second) && haslabel(third)) {
+            if((" beq " in first || "\tbeq " in first) && sameLabel(first, second, third)) {
+                val branch = second.replace("jmp", "bne")
+                mods.add(Modification(lines[0].index, true, null))
+                mods.add(Modification(lines[1].index, false, branch))
+            }
+            else if((" bne " in first || "\tbne " in first) && sameLabel(first, second, third)) {
+                val branch = second.replace("jmp", "beq")
+                mods.add(Modification(lines[0].index, true, null))
+                mods.add(Modification(lines[1].index, false, branch))
+            }
+            else if((" bcc " in first || "\tbcc " in first) && sameLabel(first, second, third)){
+                val branch = second.replace("jmp", "bcs")
+                mods.add(Modification(lines[0].index, true, null))
+                mods.add(Modification(lines[1].index, false, branch))
+            }
+            else if((" bcs " in first || "\tbcs " in first) && sameLabel(first, second, third)) {
+                val branch = second.replace("jmp", "bcc")
+                mods.add(Modification(lines[0].index, true, null))
+                mods.add(Modification(lines[1].index, false, branch))
+            }
+            else if((" bpl " in first || "\tbpl " in first) && sameLabel(first, second, third)) {
+                val branch = second.replace("jmp", "bmi")
+                mods.add(Modification(lines[0].index, true, null))
+                mods.add(Modification(lines[1].index, false, branch))
+            }
+            else if((" bmi " in first || "\tbmi " in first) && sameLabel(first, second, third)) {
+                val branch = second.replace("jmp", "bpl")
+                mods.add(Modification(lines[0].index, true, null))
+                mods.add(Modification(lines[1].index, false, branch))
+            }
+            else if((" bvc " in first || "\tbvc " in first) && sameLabel(first, second, third)) {
+                val branch = second.replace("jmp", "bvs")
+                mods.add(Modification(lines[0].index, true, null))
+                mods.add(Modification(lines[1].index, false, branch))
+            }
+            else if((" bvs " in first || "\tbvs " in first) && sameLabel(first, second, third)) {
+                val branch = second.replace("jmp", "bvc")
+                mods.add(Modification(lines[0].index, true, null))
+                mods.add(Modification(lines[1].index, false, branch))
+            }
+        }
     }
     return mods
 }
@@ -614,6 +674,52 @@ private fun optimizeUselessPushPopStack(linesByFour: Sequence<List<IndexedValue<
         optimize('a', lines)
         optimize('x', lines)
         optimize('y', lines)
+
+        val first = lines[1].value.trimStart()
+        val second = lines[2].value.trimStart()
+        val third = lines[3].value.trimStart()
+
+        // phy + ldy + pla -> tya + ldy
+        // phx + ldx + pla -> txa + ldx
+        // pha + lda + pla -> nop
+        if(first=="phy" && second.startsWith("ldy ") && third=="pla") {
+            mods.add(Modification(lines[3].index, true, null))
+            mods.add(Modification(lines[1].index, false, "  tya"))
+        }
+        else if(first=="phx" && second.startsWith("ldx ") && third=="pla") {
+            mods.add(Modification(lines[3].index, true, null))
+            mods.add(Modification(lines[1].index, false, "  txa"))
+        }
+        else if(first=="pha" && second.startsWith("lda ") && third=="pla") {
+            mods.add(Modification(lines[1].index, true, null))
+            mods.add(Modification(lines[2].index, true, null))
+            mods.add(Modification(lines[3].index, true, null))
+        }
+    }
+
+
+    return mods
+}
+
+
+private fun optimizeTSBtoRegularOr(linesByFour: Sequence<List<IndexedValue<String>>>): List<Modification> {
+    // Asm peephole:   lda var2 / tsb var1 / lda var1  Replace this with this to save 1 cycle:   lda var1 / ora var2 / sta var1
+    val mods = mutableListOf<Modification>()
+
+    for(lines in linesByFour) {
+        val first = lines[0].value.trimStart()
+        val second = lines[1].value.trimStart()
+        val third = lines[2].value.trimStart()
+        if(first.startsWith("lda") && second.startsWith("tsb") && third.startsWith("lda")) {
+            val operand1 = first.substring(3)
+            val operand2 = second.substring(3)
+            val operand3 = third.substring(3)
+            if(operand1!=operand2 && operand2==operand3) {
+                mods.add(Modification(lines[0].index, false, "  lda  $operand2"))
+                mods.add(Modification(lines[1].index, false, "  ora  $operand1"))
+                mods.add(Modification(lines[2].index, false, "  sta  $operand2"))
+            }
+        }
     }
     return mods
 }

codeGenCpu6502/src/prog8/codegen/cpu6502/BuiltinFunctionsAsmGen.kt

@@ -33,7 +33,6 @@ internal class BuiltinFunctionsAsmGen(private val program: PtProgram,
             "min__byte", "min__ubyte", "min__word", "min__uword" -> funcMin(fcall, resultRegister)
             "max__byte", "max__ubyte", "max__word", "max__uword" -> funcMax(fcall, resultRegister)
             "abs__byte", "abs__word", "abs__float" -> funcAbs(fcall, resultRegister, sscope)
-            "any", "all" -> funcAnyAll(fcall, resultRegister, sscope)
             "sgn" -> funcSgn(fcall, resultRegister, sscope)
             "sqrt__ubyte", "sqrt__uword", "sqrt__float" -> funcSqrt(fcall, resultRegister, sscope)
             "divmod__ubyte" -> funcDivmod(fcall)
@@ -44,8 +43,6 @@ internal class BuiltinFunctionsAsmGen(private val program: PtProgram,
             "ror2" -> funcRor2(fcall)
             "setlsb" -> funcSetLsbMsb(fcall, false)
             "setmsb" -> funcSetLsbMsb(fcall, true)
-            "sort" -> funcSort(fcall)
-            "reverse" -> funcReverse(fcall)
             "memory" -> funcMemory(fcall, discardResult, resultRegister)
             "peekw" -> funcPeekW(fcall, resultRegister)
             "peekf" -> funcPeekF(fcall, resultRegister)
@@ -68,6 +65,8 @@ internal class BuiltinFunctionsAsmGen(private val program: PtProgram,
             "cmp" -> funcCmp(fcall)
             "callfar" -> funcCallFar(fcall, resultRegister)
             "call" -> funcCall(fcall)
+            "prog8_ifelse_bittest_set" -> throw AssemblyError("prog8_ifelse_bittest_set() should have been translated as part of an ifElse statement")
+            "prog8_ifelse_bittest_notset" -> throw AssemblyError("prog8_ifelse_bittest_notset() should have been translated as part of an ifElse statement")
             "prog8_lib_stringcompare" -> funcStringCompare(fcall, resultRegister)
             "prog8_lib_square_byte" -> funcSquare(fcall, DataType.UBYTE, resultRegister)
             "prog8_lib_square_word" -> funcSquare(fcall, DataType.UWORD, resultRegister)
@@ -403,115 +402,6 @@ internal class BuiltinFunctionsAsmGen(private val program: PtProgram,
         }
     }
 
-    private fun funcReverse(fcall: PtBuiltinFunctionCall) {
-        val variable = fcall.args.single() as PtIdentifier
-        val symbol = asmgen.symbolTable.lookup(variable.name)
-        val (dt, numElements) = when(symbol) {
-            is StStaticVariable  -> symbol.dt to symbol.length!!
-            is StMemVar -> symbol.dt to symbol.length!!
-            else -> DataType.UNDEFINED to 0
-        }
-        val varName = asmgen.asmVariableName(variable)
-        when (dt) {
-            DataType.ARRAY_UB, DataType.ARRAY_B -> {
-                asmgen.out("""
-                    lda  #<$varName
-                    ldy  #>$varName
-                    sta  P8ZP_SCRATCH_W1
-                    sty  P8ZP_SCRATCH_W1+1
-                    lda  #$numElements
-                    jsr  prog8_lib.func_reverse_b""")
-            }
-            DataType.ARRAY_UW, DataType.ARRAY_W -> {
-                asmgen.out("""
-                    lda  #<$varName
-                    ldy  #>$varName
-                    sta  P8ZP_SCRATCH_W1
-                    sty  P8ZP_SCRATCH_W1+1
-                    lda  #$numElements
-                    jsr  prog8_lib.func_reverse_w""")
-            }
-            DataType.STR -> {
-                asmgen.out("""
-                    lda  #<$varName
-                    ldy  #>$varName
-                    sta  P8ZP_SCRATCH_W1
-                    sty  P8ZP_SCRATCH_W1+1
-                    lda  #${numElements-1}
-                    jsr  prog8_lib.func_reverse_b""")
-            }
-            DataType.ARRAY_F -> {
-                asmgen.out("""
-                    lda  #<$varName
-                    ldy  #>$varName
-                    sta  P8ZP_SCRATCH_W1
-                    sty  P8ZP_SCRATCH_W1+1
-                    lda  #$numElements
-                    jsr  floats.func_reverse_f""")
-            }
-            in SplitWordArrayTypes -> {
-                // reverse the lsb and msb arrays both, independently
-                asmgen.out("""
-                    lda  #<${varName}_lsb
-                    ldy  #>${varName}_lsb
-                    sta  P8ZP_SCRATCH_W1
-                    sty  P8ZP_SCRATCH_W1+1
-                    lda  #$numElements
-                    jsr  prog8_lib.func_reverse_b
-                    lda  #<${varName}_msb
-                    ldy  #>${varName}_msb
-                    sta  P8ZP_SCRATCH_W1
-                    sty  P8ZP_SCRATCH_W1+1
-                    lda  #$numElements
-                    jsr  prog8_lib.func_reverse_b""")
-            }
-            else -> throw AssemblyError("weird type")
-        }
-    }
-
-    private fun funcSort(fcall: PtBuiltinFunctionCall) {
-        val variable = fcall.args.single() as PtIdentifier
-        val symbol = asmgen.symbolTable.lookup(variable.name)
-        val varName = asmgen.asmVariableName(variable)
-        val (dt, numElements) = when(symbol) {
-            is StStaticVariable  -> symbol.dt to symbol.length!!
-            is StMemVar -> symbol.dt to symbol.length!!
-            else -> DataType.UNDEFINED to 0
-        }
-        when (dt) {
-            DataType.ARRAY_UB, DataType.ARRAY_B -> {
-                asmgen.out("""
-                    lda  #<$varName
-                    ldy  #>$varName
-                    sta  P8ZP_SCRATCH_W1
-                    sty  P8ZP_SCRATCH_W1+1
-                    lda  #$numElements""")
-                asmgen.out(if (dt == DataType.ARRAY_UB) "  jsr  prog8_lib.func_sort_ub" else "  jsr  prog8_lib.func_sort_b")
-            }
-            DataType.ARRAY_UW, DataType.ARRAY_W -> {
-                asmgen.out("""
-                    lda  #<$varName
-                    ldy  #>$varName
-                    sta  P8ZP_SCRATCH_W1
-                    sty  P8ZP_SCRATCH_W1+1
-                    lda  #$numElements""")
-                asmgen.out(if (dt == DataType.ARRAY_UW) "  jsr  prog8_lib.func_sort_uw" else "  jsr  prog8_lib.func_sort_w")
-            }
-            DataType.STR -> {
-                asmgen.out("""
-                    lda  #<$varName
-                    ldy  #>$varName
-                    sta  P8ZP_SCRATCH_W1
-                    sty  P8ZP_SCRATCH_W1+1
-                    lda  #${numElements-1}
-                    jsr  prog8_lib.func_sort_ub""")
-            }
-            DataType.ARRAY_F -> throw AssemblyError("sorting of floating point array is not supported")
-            in SplitWordArrayTypes -> TODO("split words sort")
-            else -> throw AssemblyError("weird type")
-        }
-    }
-
     private fun funcRor2(fcall: PtBuiltinFunctionCall) {
         val what = fcall.args.single()
         when (what.type) {
@@ -837,63 +727,6 @@ internal class BuiltinFunctionsAsmGen(private val program: PtProgram,
         assignAsmGen.assignRegisterByte(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.A, false, fcall.position, scope, asmgen), CpuRegister.A, true, true)
     }
 
-    private fun funcAnyAll(fcall: PtBuiltinFunctionCall, resultRegister: RegisterOrPair?, scope: IPtSubroutine?) {
-        val dt = fcall.args.single().type
-        val array = fcall.args[0] as PtIdentifier
-        when (dt) {
-            DataType.ARRAY_B, DataType.ARRAY_UB, DataType.STR -> {
-                outputAddressAndLengthOfArray(array)
-                asmgen.out("  jsr  prog8_lib.func_${fcall.name}_b_into_A")
-            }
-            DataType.ARRAY_UW, DataType.ARRAY_W -> {
-                outputAddressAndLengthOfArray(array)
-                asmgen.out("  jsr  prog8_lib.func_${fcall.name}_w_into_A")
-            }
-            DataType.ARRAY_F -> {
-                outputAddressAndLengthOfArray(array)
-                asmgen.out("  jsr  floats.func_${fcall.name}_f_into_A")
-            }
-            in SplitWordArrayTypes -> {
-                val numElements = (asmgen.symbolTable.lookup(array.name) as StStaticVariable).length
-                when(fcall.name) {
-                    "any" -> {
-                        // any(lsb-array) or any(msb-array)
-                        val arrayName = asmgen.asmVariableName(array)
-                        asmgen.out("""
-                            lda  #<${arrayName}_lsb
-                            ldy  #>${arrayName}_lsb
-                            sta  P8ZP_SCRATCH_W1
-                            sty  P8ZP_SCRATCH_W1+1
-                            lda  #$numElements
-                        """)
-                        asmgen.out("  jsr  prog8_lib.func_${fcall.name}_b_into_A")
-                        asmgen.out("  bne  +")      // shortcircuit
-                        asmgen.out("""
-                            pha
-                            lda  #<${arrayName}_msb
-                            ldy  #>${arrayName}_msb
-                            sta  P8ZP_SCRATCH_W1
-                            sty  P8ZP_SCRATCH_W1+1
-                            lda  #$numElements
-                        """)
-                        asmgen.out("  jsr  prog8_lib.func_${fcall.name}_b_into_A")
-                        asmgen.out("""
-                            sta  P8ZP_SCRATCH_REG
-                            pla
-                            ora  P8ZP_SCRATCH_REG
-+""")
-                    }
-                    "all" -> {
-                        TODO("split words all")
-                    }
-                    else -> throw AssemblyError("weird call")
-                }
-            }
-            else -> throw AssemblyError("weird type $dt")
-        }
-        assignAsmGen.assignRegisterByte(AsmAssignTarget.fromRegisters(resultRegister ?: RegisterOrPair.A, false, fcall.position, scope, asmgen), CpuRegister.A, dt in SignedDatatypes, true)
-    }
-
     private fun funcAbs(fcall: PtBuiltinFunctionCall, resultRegister: RegisterOrPair?, scope: IPtSubroutine?) {
         translateArguments(fcall, scope)
         val dt = fcall.args.single().type

codeGenCpu6502/src/prog8/codegen/cpu6502/ForLoopsAsmGen.kt

@@ -37,7 +37,7 @@ internal class ForLoopsAsmGen(
         val endLabel = asmgen.makeLabel("for_end")
         val modifiedLabel = asmgen.makeLabel("for_modified")
         val modifiedLabel2 = asmgen.makeLabel("for_modifiedb")
-        asmgen.loopEndLabels.push(endLabel)
+        asmgen.loopEndLabels.add(endLabel)
         val stepsize=range.step.asConstInteger()!!
 
         if(stepsize < -1) {
@@ -48,25 +48,43 @@ internal class ForLoopsAsmGen(
 
         when(iterableDt) {
             DataType.ARRAY_B, DataType.ARRAY_UB -> {
-                if (stepsize==1 || stepsize==-1) {
-
+                val varname = asmgen.asmVariableName(stmt.variable)
+                asmgen.assignExpressionToVariable(range.from, varname, ArrayToElementTypes.getValue(iterableDt))
+                if (stepsize==-1 && range.to.asConstInteger()==0) {
+                    // simple loop downto 0 step -1
+                    asmgen.out(loopLabel)
+                    asmgen.translate(stmt.statements)
+                    asmgen.out("""
+                        dec  $varname
+                        lda  $varname
+                        cmp  #255
+                        bne  $loopLabel""")
+                }
+                else if (stepsize==-1 && range.to.asConstInteger()==1) {
+                    // simple loop downto 1 step -1
+                    asmgen.out(loopLabel)
+                    asmgen.translate(stmt.statements)
+                    asmgen.out("""
+                        dec  $varname
+                        bne  $loopLabel""")
+                }
+                else if (stepsize==1 || stepsize==-1) {
                     // bytes array, step 1 or -1
 
                     val incdec = if(stepsize==1) "inc" else "dec"
                     // loop over byte range via loopvar
-                    val varname = asmgen.asmVariableName(stmt.variable)
-                    asmgen.assignExpressionToVariable(range.from, varname, ArrayToElementTypes.getValue(iterableDt))
                     asmgen.assignExpressionToRegister(range.to, RegisterOrPair.A, false)
                     // pre-check for end already reached
                     if(iterableDt==DataType.ARRAY_B) {
                         asmgen.out("  sta  $modifiedLabel+1")
-                        if(stepsize<0)
+                        if(stepsize<0) {
                             asmgen.out("""
                                 clc
                                 sbc  $varname
                                 bvc  +
                                 eor  #${'$'}80
 +                               bpl  $endLabel""")
+                        }
                         else
                             asmgen.out("""
                                 sec
@@ -75,12 +93,13 @@ internal class ForLoopsAsmGen(
                                 eor  #${'$'}80
 +                               bmi  $endLabel""")
                     } else {
-                        if(stepsize<0)
+                        if(stepsize<0) {
                             asmgen.out("""
                                 cmp  $varname
                                 beq  +
                                 bcs  $endLabel
 +""")
+                        }
                         else
                             asmgen.out("  cmp  $varname  |  bcc  $endLabel")
                         asmgen.out("  sta  $modifiedLabel+1")
@@ -100,8 +119,6 @@ $modifiedLabel          cmp  #0         ; modified
                     // bytes, step >= 2 or <= -2
 
                     // loop over byte range via loopvar
-                    val varname = asmgen.asmVariableName(stmt.variable)
-                    asmgen.assignExpressionToVariable(range.from, varname, ArrayToElementTypes.getValue(iterableDt))
                     asmgen.assignExpressionToRegister(range.to, RegisterOrPair.A, false)
                     // pre-check for end already reached
                     if(iterableDt==DataType.ARRAY_B) {
@@ -155,58 +172,85 @@ $modifiedLabel              cmp  #0     ; modified
                 }
             }
             DataType.ARRAY_W, DataType.ARRAY_UW -> {
-                when {
-
+                val varname = asmgen.asmVariableName(stmt.variable)
+                assignLoopvarWord(stmt, range)
+                if(stepsize==-1 && range.to.asConstInteger()==0) {
+                    // simple loop downto 0 step -1 (words)
+                    asmgen.out(loopLabel)
+                    asmgen.translate(stmt.statements)
+                    asmgen.out("""
+                        lda  $varname
+                        bne  ++
+                        lda  $varname+1
+                        beq  $endLabel
++                       lda  $varname
+                        bne  +
+                        dec  $varname+1
++                       dec  $varname""")
+                    asmgen.jmp(loopLabel)
+                    asmgen.out(endLabel)
+                }
+                else if (stepsize==-1 && range.to.asConstInteger()==1) {
+                    // simple loop downto 1 step -1 (words)
+                    asmgen.out(loopLabel)
+                    asmgen.translate(stmt.statements)
+                    asmgen.out("""
+                        lda  $varname
+                        cmp  #1
+                        bne  +
+                        lda  $varname+1
+                        beq  $endLabel
++                       lda  $varname
+                        bne  +
+                        dec  $varname+1
++                       dec  $varname""")
+                    asmgen.jmp(loopLabel)
+                    asmgen.out(endLabel)
+                }
+                else if (stepsize == 1 || stepsize == -1) {
                     // words, step 1 or -1
-
-                    stepsize == 1 || stepsize == -1 -> {
-                        val varname = asmgen.asmVariableName(stmt.variable)
-                        assignLoopvarWord(stmt, range)
-                        asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
-                        precheckFromToWord(iterableDt, stepsize, varname, endLabel)
-                        asmgen.out("""
-                            sty  $modifiedLabel+1
-                            sta  $modifiedLabel2+1
+                    asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
+                    precheckFromToWord(iterableDt, stepsize, varname, endLabel)
+                    asmgen.out("""
+                sty  $modifiedLabel+1
+                sta  $modifiedLabel2+1
 $loopLabel""")
-                        asmgen.translate(stmt.statements)
-                        asmgen.out("""
+                    asmgen.translate(stmt.statements)
+                    asmgen.out("""
                 lda  $varname+1
 $modifiedLabel  cmp  #0    ; modified 
                 bne  +
                 lda  $varname
 $modifiedLabel2 cmp  #0    ; modified 
                 beq  $endLabel""")
-                        if(stepsize==1) {
-                            asmgen.out("""
+                    if(stepsize==1) {
+                        asmgen.out("""
 +               inc  $varname
                 bne  $loopLabel
                 inc  $varname+1""")
-                            asmgen.jmp(loopLabel)
-                        } else {
-                            asmgen.out("""
+                        asmgen.jmp(loopLabel)
+                    } else {
+                        asmgen.out("""
 +               lda  $varname
                 bne  +
                 dec  $varname+1
 +               dec  $varname""")
-                            asmgen.jmp(loopLabel)
-                        }
-                        asmgen.out(endLabel)
+                        asmgen.jmp(loopLabel)
                     }
-                    stepsize > 0 -> {
-
-                        // (u)words, step >= 2
-                        val varname = asmgen.asmVariableName(stmt.variable)
-                        assignLoopvarWord(stmt, range)
-                        asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
-                        precheckFromToWord(iterableDt, stepsize, varname, endLabel)
-                        asmgen.out("""
-                            sty  $modifiedLabel+1
-                            sta  $modifiedLabel2+1
+                    asmgen.out(endLabel)
+                }
+                else if (stepsize > 0) {
+                    // (u)words, step >= 2
+                    asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
+                    precheckFromToWord(iterableDt, stepsize, varname, endLabel)
+                    asmgen.out("""
+                sty  $modifiedLabel+1
+                sta  $modifiedLabel2+1
 $loopLabel""")
-                        asmgen.translate(stmt.statements)
+                    asmgen.translate(stmt.statements)
 
-                        if (iterableDt == DataType.ARRAY_UW) {
-                            asmgen.out("""
+                    if (iterableDt == DataType.ARRAY_UW) {
+                        asmgen.out("""
                 lda  $varname
                 clc
                 adc  #<$stepsize
@@ -222,8 +266,8 @@ $modifiedLabel2 lda  #0     ; modified
                 bcc  $endLabel
                 bcs  $loopLabel
 $endLabel""")
-                        } else {
-                            asmgen.out("""
+                    } else {
+                        asmgen.out("""
                 lda  $varname
                 clc
                 adc  #<$stepsize
@@ -239,22 +283,20 @@ $modifiedLabel  lda  #0   ; modified
                 eor  #$80
 +               bpl  $loopLabel                
 $endLabel""")
-                        }
                     }
-                    else -> {
+                }
+                else {
 
-                        // (u)words, step <= -2
-                        val varname = asmgen.asmVariableName(stmt.variable)
-                        assignLoopvarWord(stmt, range)
-                        asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
-                        precheckFromToWord(iterableDt, stepsize, varname, endLabel)
-                        asmgen.out("""
-                            sty  $modifiedLabel+1
-                            sta  $modifiedLabel2+1
+                    // (u)words, step <= -2
+                    asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
+                    precheckFromToWord(iterableDt, stepsize, varname, endLabel)
+                    asmgen.out("""
+                sty  $modifiedLabel+1
+                sta  $modifiedLabel2+1
 $loopLabel""")
-                        asmgen.translate(stmt.statements)
+                    asmgen.translate(stmt.statements)
 
-                            asmgen.out("""
+                    asmgen.out("""
                 lda  $varname
                 sec
                 sbc  #<${stepsize.absoluteValue}
@@ -271,13 +313,12 @@ $modifiedLabel  sbc  #0    ; modified
                 eor  #$80
 +               bpl  $loopLabel                
 $endLabel""")
-                    }
                 }
             }
             else -> throw AssemblyError("range expression can only be byte or word")
         }
 
-        asmgen.loopEndLabels.pop()
+        asmgen.loopEndLabels.removeLast()
     }
 
     private fun precheckFromToWord(iterableDt: DataType, stepsize: Int, fromVar: String, endLabel: String) {
@@ -333,7 +374,7 @@ $endLabel""")
     private fun translateForOverIterableVar(stmt: PtForLoop, iterableDt: DataType, ident: PtIdentifier) {
         val loopLabel = asmgen.makeLabel("for_loop")
         val endLabel = asmgen.makeLabel("for_end")
-        asmgen.loopEndLabels.push(endLabel)
+        asmgen.loopEndLabels.add(endLabel)
         val iterableName = asmgen.asmVariableName(ident)
         val numElements = when(val symbol = asmgen.symbolTable.lookup(ident.name)) {
             is StStaticVariable -> symbol.length!!
@@ -477,17 +518,19 @@ $loopLabel          sty  $indexVar
             }
             else -> throw AssemblyError("can't iterate over $iterableDt")
         }
-        asmgen.loopEndLabels.pop()
+        asmgen.loopEndLabels.removeLast()
     }
 
     private fun translateForOverConstRange(stmt: PtForLoop, iterableDt: DataType, range: IntProgression) {
         if (range.isEmpty() || range.step==0)
             throw AssemblyError("empty range or step 0")
         if(iterableDt==DataType.ARRAY_B || iterableDt==DataType.ARRAY_UB) {
+            if(range.last==range.first) return translateForSimpleByteRangeAsc(stmt, range)
             if(range.step==1 && range.last>range.first) return translateForSimpleByteRangeAsc(stmt, range)
-            if(range.step==-1 && range.last<range.first) return translateForSimpleByteRangeDesc(stmt, range)
+            if(range.step==-1 && range.last<range.first) return translateForSimpleByteRangeDesc(stmt, range, iterableDt==DataType.ARRAY_UB)
         }
         else if(iterableDt==DataType.ARRAY_W || iterableDt==DataType.ARRAY_UW) {
+            if(range.last==range.first) return translateForSimpleWordRangeAsc(stmt, range)
             if(range.step==1 && range.last>range.first) return translateForSimpleWordRangeAsc(stmt, range)
             if(range.step==-1 && range.last<range.first) return translateForSimpleWordRangeDesc(stmt, range)
         }
@@ -495,7 +538,7 @@ $loopLabel          sty  $indexVar
         // not one of the easy cases, generate more complex code...
         val loopLabel = asmgen.makeLabel("for_loop")
         val endLabel = asmgen.makeLabel("for_end")
-        asmgen.loopEndLabels.push(endLabel)
+        asmgen.loopEndLabels.add(endLabel)
         when(iterableDt) {
             DataType.ARRAY_B, DataType.ARRAY_UB -> {
                 // loop over byte range via loopvar, step >= 2 or <= -2
@@ -507,7 +550,7 @@ $loopLabel""")
                 asmgen.translate(stmt.statements)
                 when (range.step) {
                     0, 1, -1 -> {
-                        throw AssemblyError("step 0, 1 and -1 should have been handled specifically  $stmt")
+                        throw AssemblyError("step 0, 1 and -1 should have been handled specifically  $range ${stmt.position}")
                     }
                     2 -> {
                         if(range.last==255 || range.last==254) {
@@ -601,13 +644,13 @@ $loopLabel""")
             }
             else -> throw AssemblyError("range expression can only be byte or word")
         }
-        asmgen.loopEndLabels.pop()
+        asmgen.loopEndLabels.removeLast()
     }
 
     private fun translateForSimpleByteRangeAsc(stmt: PtForLoop, range: IntProgression) {
         val loopLabel = asmgen.makeLabel("for_loop")
         val endLabel = asmgen.makeLabel("for_end")
-        asmgen.loopEndLabels.push(endLabel)
+        asmgen.loopEndLabels.add(endLabel)
         val varname = asmgen.asmVariableName(stmt.variable)
         asmgen.out("""
                 lda  #${range.first}
@@ -627,50 +670,50 @@ $endLabel""")
                 bne  $loopLabel
 $endLabel""")
         }
-        asmgen.loopEndLabels.pop()
+        asmgen.loopEndLabels.removeLast()
     }
 
-    private fun translateForSimpleByteRangeDesc(stmt: PtForLoop, range: IntProgression) {
+    private fun translateForSimpleByteRangeDesc(stmt: PtForLoop, range: IntProgression, unsigned: Boolean) {
         val loopLabel = asmgen.makeLabel("for_loop")
-        val endLabel = asmgen.makeLabel("for_end")
-        asmgen.loopEndLabels.push(endLabel)
         val varname = asmgen.asmVariableName(stmt.variable)
         asmgen.out("""
-                    lda  #${range.first}
-                    sta  $varname
+            lda  #${range.first}
+            sta  $varname
 $loopLabel""")
         asmgen.translate(stmt.statements)
         when (range.last) {
             0 -> {
-                asmgen.out("""
-                    lda  $varname
-                    beq  $endLabel
-                    dec  $varname""")
-                asmgen.jmp(loopLabel)
-                asmgen.out(endLabel)
+                if(!unsigned || range.first<=127) {
+                    asmgen.out("""
+                        dec  $varname
+                        bpl  $loopLabel""")
+                } else {
+                    asmgen.out("""
+                        dec  $varname
+                        lda  $varname
+                        cmp  #255
+                        bne  $loopLabel""")
+                }
             }
             1 -> {
                 asmgen.out("""
                     dec  $varname
-                    bne  $loopLabel
-$endLabel""")
+                    bne  $loopLabel""")
             }
             else -> {
                 asmgen.out("""
                     dec  $varname
                     lda  $varname
                     cmp  #${range.last-1}
-                    bne  $loopLabel
-$endLabel""")
+                    bne  $loopLabel""")
             }
         }
-        asmgen.loopEndLabels.pop()
     }
 
     private fun translateForSimpleWordRangeAsc(stmt: PtForLoop, range: IntProgression) {
         val loopLabel = asmgen.makeLabel("for_loop")
         val endLabel = asmgen.makeLabel("for_end")
-        asmgen.loopEndLabels.push(endLabel)
+        asmgen.loopEndLabels.add(endLabel)
         val varname = asmgen.asmVariableName(stmt.variable)
         asmgen.out("""
             lda  #<${range.first}
@@ -691,13 +734,13 @@ $loopLabel""")
             inc  $varname+1""")
         asmgen.jmp(loopLabel)
         asmgen.out(endLabel)
-        asmgen.loopEndLabels.pop()
+        asmgen.loopEndLabels.removeLast()
     }
 
     private fun translateForSimpleWordRangeDesc(stmt: PtForLoop, range: IntProgression) {
         val loopLabel = asmgen.makeLabel("for_loop")
         val endLabel = asmgen.makeLabel("for_end")
-        asmgen.loopEndLabels.push(endLabel)
+        asmgen.loopEndLabels.add(endLabel)
         val varname = asmgen.asmVariableName(stmt.variable)
         asmgen.out("""
             lda  #<${range.first}
@@ -706,20 +749,29 @@ $loopLabel""")
             sty  $varname+1
 $loopLabel""")
         asmgen.translate(stmt.statements)
+        if(range.last==0) {
+            asmgen.out("""
+                lda  $varname
+                bne  ++
+                lda  $varname+1
+                beq  $endLabel""")
+        } else {
+            asmgen.out("""
+                lda  $varname
+                cmp  #<${range.last}
+                bne  +
+                lda  $varname+1
+                cmp  #>${range.last}
+                beq  $endLabel""")
+        }
         asmgen.out("""
-            lda  $varname
-            cmp  #<${range.last}
-            bne  +
-            lda  $varname+1
-            cmp  #>${range.last}
-            beq  $endLabel
 +           lda  $varname
             bne  +
             dec  $varname+1
 +           dec  $varname""")
         asmgen.jmp(loopLabel)
         asmgen.out(endLabel)
-        asmgen.loopEndLabels.pop()
+        asmgen.loopEndLabels.removeLast()
     }
 
     private fun assignLoopvarWord(stmt: PtForLoop, range: PtRange) =

codeGenCpu6502/src/prog8/codegen/cpu6502/FunctionCallAsmGen.kt

@@ -82,7 +82,7 @@ internal class FunctionCallAsmGen(private val program: PtProgram, private val as
             is PtAddressOf -> false
             is PtIdentifier -> false
             is PtIrRegister -> false
-            is PtMemoryByte -> return usesOtherRegistersWhileEvaluating(arg.address)
+            is PtMemoryByte -> arg.address !is PtNumber && arg.address !is PtIdentifier
             is PtNumber -> false
             is PtBool -> false
             else -> true
@@ -110,9 +110,9 @@ internal class FunctionCallAsmGen(private val program: PtProgram, private val as
                         throw AssemblyError("call argument evaluation problem: can't save cpu statusregister parameter ${call.position}")
                     }
                     else {
-                        if(usedX()) asmgen.saveRegisterStack(CpuRegister.X, false)
-                        if(usedY()) asmgen.saveRegisterStack(CpuRegister.Y, false)
-                        if(usedA()) asmgen.saveRegisterStack(CpuRegister.A, false)
+                        if(usedX()) asmgen.saveRegisterStack(CpuRegister.X, usedA())
+                        if(usedY()) asmgen.saveRegisterStack(CpuRegister.Y, usedA())
+                        if(usedA()) asmgen.saveRegisterStack(CpuRegister.A, usedA())
                         val used = argumentViaRegister(sub, IndexedValue(it, param.second), arg)
                         if(usedA()) asmgen.restoreRegisterStack(CpuRegister.A, false)
                         if(usedY()) asmgen.restoreRegisterStack(CpuRegister.Y, true)
@@ -205,7 +205,7 @@ internal class FunctionCallAsmGen(private val program: PtProgram, private val as
                     if(value is PtIdentifier) {
                         val addr = PtAddressOf(Position.DUMMY)
                         addr.add(value)
-                        addr.parent = sub as PtNode
+                        addr.parent = scope as PtNode
                         AsmAssignSource.fromAstSource(addr, program, asmgen).adjustSignedUnsigned(target)
                     } else {
                         AsmAssignSource.fromAstSource(value, program, asmgen).adjustSignedUnsigned(target)
@@ -213,7 +213,7 @@ internal class FunctionCallAsmGen(private val program: PtProgram, private val as
                 } else {
                     AsmAssignSource.fromAstSource(value, program, asmgen).adjustSignedUnsigned(target)
                 }
-                asmgen.translateNormalAssignment(AsmAssignment(src, target, program.memsizer, Position.DUMMY), scope)
+                asmgen.translateNormalAssignment(AsmAssignment(src, target, program.memsizer, target.position), scope)
             }
             return RegisterOrStatusflag(register, null)
         }

codeGenCpu6502/src/prog8/codegen/cpu6502/IfElseAsmGen.kt

@@ -83,9 +83,80 @@ internal class IfElseAsmGen(private val program: PtProgram,
     }
 
     private fun fallbackTranslateForSimpleCondition(ifElse: PtIfElse) {
+        val bittest = ifElse.condition as? PtBuiltinFunctionCall
+        val jumpAfterIf = ifElse.ifScope.children.singleOrNull() as? PtJump
+
+        if(bittest!=null && bittest.name.startsWith("prog8_ifelse_bittest_")) {
+            val variable = bittest.args[0] as PtIdentifier
+            val bitnumber = (bittest.args[1] as PtNumber).number.toInt()
+            val testForBitSet = bittest.name.endsWith("_set")
+            when (bitnumber) {
+                7 -> {
+                    // test via bit + N flag
+                    asmgen.out("  bit  ${variable.name}")
+                    if(testForBitSet) {
+                        if(jumpAfterIf!=null) {
+                            val (asmLabel, indirect) = asmgen.getJumpTarget(jumpAfterIf)
+                            if(indirect)
+                                throw AssemblyError("cannot BIT to indirect label ${ifElse.position}")
+                            if(ifElse.hasElse())
+                                throw AssemblyError("didn't expect else part here ${ifElse.position}")
+                            else
+                                asmgen.out("  bmi  $asmLabel")
+                        }
+                        else
+                            translateIfElseBodies("bpl", ifElse)
+                    } else {
+                        if(jumpAfterIf!=null) {
+                            val (asmLabel, indirect) = asmgen.getJumpTarget(jumpAfterIf)
+                            if(indirect)
+                                throw AssemblyError("cannot BIT to indirect label ${ifElse.position}")
+                            if(ifElse.hasElse())
+                                throw AssemblyError("didn't expect else part here ${ifElse.position}")
+                            else
+                                asmgen.out("  bpl  $asmLabel")
+                        }
+                        else
+                            translateIfElseBodies("bmi", ifElse)
+                    }
+                    return
+                }
+                6 -> {
+                    // test via bit + V flag
+                    asmgen.out("  bit  ${variable.name}")
+                    if(testForBitSet) {
+                        if(jumpAfterIf!=null) {
+                            val (asmLabel, indirect) = asmgen.getJumpTarget(jumpAfterIf)
+                            if(indirect)
+                                throw AssemblyError("cannot BIT to indirect label ${ifElse.position}")
+                            if(ifElse.hasElse())
+                                throw AssemblyError("didn't expect else part here ${ifElse.position}")
+                            else
+                                asmgen.out("  bvs  $asmLabel")
+                        }
+                        else
+                            translateIfElseBodies("bvc", ifElse)
+                    } else {
+                        if(jumpAfterIf!=null) {
+                            val (asmLabel, indirect) = asmgen.getJumpTarget(jumpAfterIf)
+                            if(indirect)
+                                throw AssemblyError("cannot BIT to indirect label ${ifElse.position}")
+                            if(ifElse.hasElse())
+                                throw AssemblyError("didn't expect else part here ${ifElse.position}")
+                            else
+                                asmgen.out("  bvc  $asmLabel")
+                        }
+                        else
+                            translateIfElseBodies("bvs", ifElse)
+                    }
+                    return
+                }
+                else -> throw AssemblyError("prog8_ifelse_bittest can only work on bits 7 and 6")
+            }
+        }
+
         // the condition is "simple" enough to just assign its 0/1 value to a register and branch on that
         assignConditionValueToRegisterAndTest(ifElse.condition)
-        val jumpAfterIf = ifElse.ifScope.children.singleOrNull() as? PtJump
         if(jumpAfterIf!=null)
             translateJumpElseBodies("bne", "beq", jumpAfterIf, ifElse.elseScope)
         else
@@ -405,14 +476,14 @@ internal class IfElseAsmGen(private val program: PtProgram,
         val condition = stmt.condition as PtBinaryExpression
         if(signed) {
             // X>Y --> Y<X
-            asmgen.assignExpressionToRegister(condition.right, RegisterOrPair.A, signed)
+            asmgen.assignExpressionToRegister(condition.right, RegisterOrPair.A, true)
             cmpAwithByteValue(condition.left, true)
             if (jumpAfterIf != null)
                 translateJumpElseBodies("bmi", "bpl", jumpAfterIf, stmt.elseScope)
             else
                 translateIfElseBodies("bpl", stmt)
         } else {
-            asmgen.assignExpressionToRegister(condition.left, RegisterOrPair.A, signed)
+            asmgen.assignExpressionToRegister(condition.left, RegisterOrPair.A, false)
             cmpAwithByteValue(condition.right, false)
             if(jumpAfterIf!=null) {
                 val (asmLabel, indirect) = asmgen.getJumpTarget(jumpAfterIf)
@@ -848,10 +919,10 @@ _jump                       jmp  ($asmLabel)
                     }
                 } else {
                     asmgen.loadScaledArrayIndexIntoRegister(value, CpuRegister.Y)
-                    if(value.splitWords) {
-                        return compareLsbMsb("${varname}_lsb,y", "${varname}_msb,y")
+                    return if(value.splitWords) {
+                        compareLsbMsb("${varname}_lsb,y", "${varname}_msb,y")
                     } else {
-                        return compareLsbMsb("$varname,y", "$varname+1,y")
+                        compareLsbMsb("$varname,y", "$varname+1,y")
                     }
                 }
             }
@@ -985,10 +1056,10 @@ _jump                       jmp  ($asmLabel)
                     }
                 } else {
                     asmgen.loadScaledArrayIndexIntoRegister(value, CpuRegister.Y)
-                    if(value.splitWords) {
-                        return compareLsbMsb("${varname}_lsb,y", "${varname}_msb,y")
+                    return if(value.splitWords) {
+                        compareLsbMsb("${varname}_lsb,y", "${varname}_msb,y")
                     } else {
-                        return compareLsbMsb("$varname,y", "$varname+1,y")
+                        compareLsbMsb("$varname,y", "$varname+1,y")
                     }
                 }
             }

codeGenCpu6502/src/prog8/codegen/cpu6502/assignment/AssignmentAsmGen.kt

@@ -11,11 +11,12 @@ import prog8.codegen.cpu6502.VariableAllocator
 import prog8.codegen.cpu6502.returnsWhatWhere
 
 
-internal class AssignmentAsmGen(private val program: PtProgram,
-                                private val asmgen: AsmGen6502Internal,
-                                private val options: CompilationOptions,
-                                private val anyExprGen: AnyExprAsmGen,
-                                private val allocator: VariableAllocator) {
+internal class AssignmentAsmGen(
+    private val program: PtProgram,
+    private val asmgen: AsmGen6502Internal,
+    private val anyExprGen: AnyExprAsmGen,
+    private val allocator: VariableAllocator
+) {
     private val augmentableAsmGen = AugmentableAssignmentAsmGen(program, this, asmgen, allocator)
 
     fun translate(assignment: PtAssignment) {
@@ -47,6 +48,26 @@ internal class AssignmentAsmGen(private val program: PtProgram,
 
         asmgen.translate(values)
 
+        val assignmentTargets = assignment.children.dropLast(1)
+        if(sub.returns.size==assignmentTargets.size) {
+            // because we can only handle integer results right now we can just zip() it all up
+            val (statusFlagResults, registersResults) = sub.returns.zip(assignmentTargets).partition { it.first.register.statusflag!=null }
+            if (statusFlagResults.isEmpty())
+                assignRegisterResults(registersResults)
+            else if(registersResults.isEmpty())
+                assignOnlyTheStatusFlagsResults(false, statusFlagResults)
+            else
+                assignStatusFlagsAndRegistersResults(statusFlagResults, registersResults)
+        } else {
+            throw AssemblyError("number of values and targets don't match")
+        }
+    }
+
+    private fun assignStatusFlagsAndRegistersResults(
+        statusFlagResults: List<Pair<StRomSubParameter, PtNode>>,
+        registersResults: List<Pair<StRomSubParameter, PtNode>>
+    ) {
+
         fun needsToSaveA(registersResults: List<Pair<StRomSubParameter, PtNode>>): Boolean =
             if(registersResults.isEmpty())
                 false
@@ -55,124 +76,123 @@ internal class AssignmentAsmGen(private val program: PtProgram,
             else
                 true
 
-        fun assignCarryFlagResult(target: PtAssignTarget, saveA: Boolean) {
-            if(saveA) asmgen.out("  pha")
-            asmgen.out("  lda  #0  |  rol  a")
-            val tgt = AsmAssignTarget.fromAstAssignment(target, target.definingISub(), asmgen)
-            assignRegisterByte(tgt, CpuRegister.A, false, false)
-            if(saveA) asmgen.out("  pla")
+        if(registersResults.all {
+                val tgt = it.second as PtAssignTarget
+                tgt.void || tgt.identifier!=null})
+        {
+            // all other results are just stored into identifiers directly so first handle those
+            // (simple store instructions that don't modify the carry flag)
+            assignRegisterResults(registersResults)
+            assignOnlyTheStatusFlagsResults(false, statusFlagResults)
+        } else {
+            val saveA = needsToSaveA(registersResults)
+            assignOnlyTheStatusFlagsResults(saveA, statusFlagResults)
+            assignRegisterResults(registersResults)
+        }
+    }
+
+    private fun assignOnlyTheStatusFlagsResults(saveA: Boolean, statusFlagResults: List<Pair<StRomSubParameter, PtNode>>) {
+        // assigning flags to their variables targets requires load-instructions that destroy flags
+        // so if there's more than 1, we need to save and restore the flags
+        val saveFlags = statusFlagResults.size>1
+
+        fun hasFlag(statusFlagResults: List<Pair<StRomSubParameter, PtNode>>, flag: Statusflag): PtAssignTarget? {
+            for ((returns, target) in statusFlagResults) {
+                if(returns.register.statusflag!! == flag)
+                    return target as PtAssignTarget
+            }
+            return null
         }
 
-        fun assignZeroFlagResult(target: PtAssignTarget, saveA: Boolean) {
-            if(saveA) asmgen.out("  pha")
-            asmgen.out("""
+        val targetCarry = hasFlag(statusFlagResults, Statusflag.Pc)
+        val targetZero = hasFlag(statusFlagResults, Statusflag.Pz)
+        val targetNeg = hasFlag(statusFlagResults, Statusflag.Pn)
+        val targetOverflow = hasFlag(statusFlagResults, Statusflag.Pv)
+
+        if(saveA) asmgen.out("  pha")
+        if(targetZero!=null && !targetZero.void)
+            assignZeroFlagResult(targetZero, saveFlags)
+        if(targetNeg!=null && !targetNeg.void)
+            assignNegativeFlagResult(targetNeg, saveFlags)
+        if(targetCarry!=null && !targetCarry.void)
+            assignCarryFlagResult(targetCarry)
+        if(targetOverflow!=null && !targetOverflow.void)
+            assignOverflowFlagResult(targetOverflow)
+        if(saveA) asmgen.out("  pla")
+    }
+
+    private fun assignRegisterResults(registersResults: List<Pair<StRomSubParameter, PtNode>>) {
+        registersResults.forEach { (returns, target) ->
+            target as PtAssignTarget
+            if(!target.void) {
+                val targetIdent = target.identifier
+                val targetMem = target.memory
+                if(targetIdent!=null || targetMem!=null) {
+                    val tgt = AsmAssignTarget.fromAstAssignment(target, target.definingISub(), asmgen)
+                    when(returns.type) {
+                        in ByteDatatypesWithBoolean -> {
+                            if(returns.register.registerOrPair in Cx16VirtualRegisters) {
+                                assignVirtualRegister(tgt, returns.register.registerOrPair!!)
+                            } else {
+                                assignRegisterByte(tgt, returns.register.registerOrPair!!.asCpuRegister(), false, false)
+                            }
+                        }
+                        in WordDatatypes -> {
+                            assignRegisterpairWord(tgt, returns.register.registerOrPair!!)
+                        }
+                        else -> throw AssemblyError("weird dt")
+                    }
+                }
+                else TODO("array target for multi-value assignment")        // Not done yet due to result register clobbering complexity
+            }
+        }
+    }
+
+    private fun assignCarryFlagResult(target: PtAssignTarget) {
+        // overflow is not clobbered so no need to save/restore it
+        asmgen.out("  lda  #0  |  rol  a")
+        val tgt = AsmAssignTarget.fromAstAssignment(target, target.definingISub(), asmgen)
+        assignRegisterByte(tgt, CpuRegister.A, false, false)
+    }
+
+    private fun assignZeroFlagResult(target: PtAssignTarget, saveFlags: Boolean) {
+        if(saveFlags) asmgen.out("  php")
+        asmgen.out("""
                 beq  +
                 lda  #0
                 beq  ++
 +               lda  #1
 +""")
-            val tgt = AsmAssignTarget.fromAstAssignment(target, target.definingISub(), asmgen)
-            assignRegisterByte(tgt, CpuRegister.A, false, false)
-            if(saveA) asmgen.out("  pla")
-        }
+        val tgt = AsmAssignTarget.fromAstAssignment(target, target.definingISub(), asmgen)
+        assignRegisterByte(tgt, CpuRegister.A, false, false)
+        if(saveFlags) asmgen.out("  plp")
+    }
 
-        fun assignNegativeFlagResult(target: PtAssignTarget, saveA: Boolean) {
-            if(saveA) asmgen.out("  pha")
-            asmgen.out("""
+    private fun assignNegativeFlagResult(target: PtAssignTarget, saveFlags: Boolean) {
+        if(saveFlags) asmgen.out("  php")
+        asmgen.out("""
                 bmi  +
                 lda  #0
                 beq  ++
 +               lda  #1
 +""")
-            val tgt = AsmAssignTarget.fromAstAssignment(target, target.definingISub(), asmgen)
-            assignRegisterByte(tgt, CpuRegister.A, false, false)
-            if(saveA) asmgen.out("  pla")
-        }
+        val tgt = AsmAssignTarget.fromAstAssignment(target, target.definingISub(), asmgen)
+        assignRegisterByte(tgt, CpuRegister.A, false, false)
+        if(saveFlags) asmgen.out("  plp")
+    }
 
-        fun assignOverflowFlagResult(target: PtAssignTarget, saveA: Boolean) {
-            if(saveA) asmgen.out("  pha")
-            asmgen.out("""
+    private fun assignOverflowFlagResult(target: PtAssignTarget) {
+        // overflow is not clobbered so no need to save/restore it
+        asmgen.out("""
                 bvs  +
                 lda  #0
                 beq  ++
 +               lda  #1
 +""")
-            val tgt = AsmAssignTarget.fromAstAssignment(target, target.definingISub(), asmgen)
-            assignRegisterByte(tgt, CpuRegister.A, false, false)
-            if(saveA) asmgen.out("  pla")
-        }
-
-        fun assignRegisterResults(registersResults: List<Pair<StRomSubParameter, PtNode>>) {
-            registersResults.forEach { (returns, target) ->
-                target as PtAssignTarget
-                if(!target.void) {
-                    val targetIdent = target.identifier
-                    val targetMem = target.memory
-                    if(targetIdent!=null || targetMem!=null) {
-                        val tgt = AsmAssignTarget.fromAstAssignment(target, target.definingISub(), asmgen)
-                        when(returns.type) {
-                            in ByteDatatypesWithBoolean -> {
-                                if(returns.register.registerOrPair in Cx16VirtualRegisters) {
-                                    assignVirtualRegister(tgt, returns.register.registerOrPair!!)
-                                } else {
-                                    assignRegisterByte(tgt, returns.register.registerOrPair!!.asCpuRegister(), false, false)
-                                }
-                            }
-                            in WordDatatypes -> {
-                                assignRegisterpairWord(tgt, returns.register.registerOrPair!!)
-                            }
-                            else -> throw AssemblyError("weird dt")
-                        }
-                    }
-                    else TODO("array target for multi-value assignment")        // Not done yet due to result register clobbering complexity
-                }
-            }
-        }
-
-        val assignmentTargets = assignment.children.dropLast(1)
-        if(sub.returns.size==assignmentTargets.size) {
-            // because we can only handle integer results right now we can just zip() it all up
-            val (statusFlagResults, registersResults) = sub.returns.zip(assignmentTargets).partition { it.first.register.statusflag!=null }
-            if(statusFlagResults.isNotEmpty()) {
-                if(statusFlagResults.size>1)
-                    TODO("handle multiple status flag results")
-                val (returns, target) = statusFlagResults.single()
-                target as PtAssignTarget
-                if(target.void) {
-                    // forget about the Carry status flag, only assign the normal return values
-                    assignRegisterResults(registersResults)
-                    return
-                }
-                if(registersResults.all {
-                    val tgt = it.second as PtAssignTarget
-                    tgt.void || tgt.identifier!=null})
-                {
-                    // all other results are just stored into identifiers directly so first handle those
-                    // (simple store instructions that don't modify the carry flag)
-                    assignRegisterResults(registersResults)
-                    return when(returns.register.statusflag!!) {
-                        Statusflag.Pc -> assignCarryFlagResult(target, false)
-                        Statusflag.Pz -> assignZeroFlagResult(target, false)
-                        Statusflag.Pv -> assignOverflowFlagResult(target, false)
-                        Statusflag.Pn -> assignNegativeFlagResult(target, false)
-                    }
-                }
-
-                val saveA = needsToSaveA(registersResults)
-                when(returns.register.statusflag!!) {
-                    Statusflag.Pc -> assignCarryFlagResult(target, saveA)
-                    Statusflag.Pz -> assignZeroFlagResult(target, saveA)
-                    Statusflag.Pv -> assignOverflowFlagResult(target, saveA)
-                    Statusflag.Pn -> assignNegativeFlagResult(target, saveA)
-                }
-            }
-            assignRegisterResults(registersResults)
-        } else {
-            throw AssemblyError("number of values and targets don't match")
-        }
+        val tgt = AsmAssignTarget.fromAstAssignment(target, target.definingISub(), asmgen)
+        assignRegisterByte(tgt, CpuRegister.A, false, false)
     }
 
-
     fun translateNormalAssignment(assign: AsmAssignment, scope: IPtSubroutine?) {
         when(assign.source.kind) {
             SourceStorageKind.LITERALBOOLEAN -> {
@@ -201,7 +221,7 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                 val variable = assign.source.asmVarname
                 when (assign.target.datatype) {
                     DataType.BOOL -> {
-                        if (assign.source.datatype == DataType.BOOL || !options.strictBool) assignVariableByte(assign.target, variable)
+                        if (assign.source.datatype == DataType.BOOL) assignVariableByte(assign.target, variable)
                         else throw AssemblyError("assigning non-bool variable to boolean, should have been typecasted")
                     }
                     DataType.UBYTE, DataType.BYTE -> assignVariableByte(assign.target, variable)
@@ -284,12 +304,6 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                 }
             }
             SourceStorageKind.MEMORY -> {
-                fun assignViaExprEval(expression: PtExpression) {
-                    assignExpressionToVariable(expression, "P8ZP_SCRATCH_W2", DataType.UWORD)
-                    asmgen.loadAFromZpPointerVar("P8ZP_SCRATCH_W2", false)
-                    assignRegisterByte(assign.target, CpuRegister.A, false, true)
-                }
-
                 val value = assign.source.memory!!
                 when (value.address) {
                     is PtNumber -> {
@@ -304,10 +318,10 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                         if(asmgen.tryOptimizedPointerAccessWithA(addrExpr, false)) {
                             assignRegisterByte(assign.target, CpuRegister.A, false, true)
                         } else {
-                            assignViaExprEval(value.address)
+                            assignByteFromAddressExpression(value.address, assign.target)
                         }
                     }
-                    else -> assignViaExprEval(value.address)
+                    else -> assignByteFromAddressExpression(value.address, assign.target)
                 }
             }
             SourceStorageKind.EXPRESSION -> {
@@ -319,6 +333,106 @@ internal class AssignmentAsmGen(private val program: PtProgram,
         }
     }
 
+    private fun assignByteFromAddressExpression(address: PtExpression, target: AsmAssignTarget) {
+        if(address is PtBinaryExpression) {
+            if(address.operator=="+" && address.right.type==DataType.UWORD) {
+                if (address.left is PtIdentifier) {
+                    // use (zp),Y instead of explicitly calculating the full zp pointer value
+                    val pointer = (address.left as PtIdentifier).name
+                    when(val index=address.right) {
+                        is PtIdentifier -> {
+                            val indexName = index.name
+                            asmgen.out("""
+                                lda  $pointer
+                                sta  P8ZP_SCRATCH_W2
+                                lda  $pointer+1
+                                clc
+                                adc  $indexName+1
+                                sta  P8ZP_SCRATCH_W2+1
+                                ldy  $indexName
+                                lda  (P8ZP_SCRATCH_W2),y""")
+                            assignRegisterByte(target, CpuRegister.A, false, true)
+                            return
+                        }
+                        is PtNumber -> {
+                            val indexValue = index.number.toInt().toString()
+                            asmgen.out("""
+                                lda  $pointer
+                                sta  P8ZP_SCRATCH_W2
+                                lda  $pointer+1
+                                clc
+                                adc  #>$indexValue
+                                sta  P8ZP_SCRATCH_W2+1
+                                ldy  #<$indexValue
+                                lda  (P8ZP_SCRATCH_W2),y""")
+                            assignRegisterByte(target, CpuRegister.A, false, true)
+                            return
+                        }
+                        else -> {}
+                    }
+                }
+            }
+//          else if(address.operator=="-") {
+//              // does this ever occur? we could optimize it too, but it seems like a pathological case
+//          }
+        }
+        assignExpressionToVariable(address, "P8ZP_SCRATCH_W2", DataType.UWORD)
+        asmgen.loadAFromZpPointerVar("P8ZP_SCRATCH_W2", false)
+        assignRegisterByte(target, CpuRegister.A, false, true)
+    }
+
+    private fun storeByteInAToAddressExpression(address: PtExpression, saveA: Boolean) {
+        if(address is PtBinaryExpression) {
+            if(address.operator=="+") {
+                if (address.left is PtIdentifier && address.right.type==DataType.UWORD) {
+                    // use (zp),Y instead of explicitly calculating the full zp pointer value
+                    val pointer = (address.left as PtIdentifier).name
+                    when(val index=address.right) {
+                        is PtIdentifier -> {
+                            val indexName = index.name
+                            asmgen.out("""
+                                tax
+                                lda  $pointer
+                                sta  P8ZP_SCRATCH_W2
+                                lda  $pointer+1
+                                clc
+                                adc  $indexName+1
+                                sta  P8ZP_SCRATCH_W2+1
+                                ldy  $indexName
+                                txa
+                                sta  (P8ZP_SCRATCH_W2),y""")
+                            return
+                        }
+                        is PtNumber -> {
+                            val indexValue = index.number.toInt().toString()
+                            asmgen.out("""
+                                tax
+                                lda  $pointer
+                                sta  P8ZP_SCRATCH_W2
+                                lda  $pointer+1
+                                clc
+                                adc  #>$indexValue
+                                sta  P8ZP_SCRATCH_W2+1
+                                ldy  #<$indexValue
+                                txa
+                                sta  (P8ZP_SCRATCH_W2),y""")
+                            return
+                        }
+                        else -> {}
+                    }
+                }
+            }
+//          else if(address.operator=="-") {
+//              // does this ever occur? we could optimize it too, but it seems like a pathological case
+//          }
+        }
+        if(saveA) asmgen.out("  pha")
+        assignExpressionToVariable(address, "P8ZP_SCRATCH_W2", DataType.UWORD)
+        if(saveA) asmgen.out("  pla")
+        asmgen.storeAIntoZpPointerVar("P8ZP_SCRATCH_W2", false)
+    }
+
+
     private fun assignExpression(assign: AsmAssignment, scope: IPtSubroutine?) {
         when(val value = assign.source.expression!!) {
             is PtAddressOf -> {
@@ -703,8 +817,8 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                 RegisterOrPair.Y -> "y"
                 else -> TODO("comparison to word register")
             }
-            val assignTrue = PtInlineAssembly("  ld${reg}  #1", false, assign.target.position)
-            val assignFalse = PtInlineAssembly("  ld${reg}  #0", false, assign.target.position)
+            val assignTrue = PtInlineAssembly("\tld${reg}  #1", false, assign.target.position)
+            val assignFalse = PtInlineAssembly("\tld${reg}  #0", false, assign.target.position)
             ifPart.add(assignTrue)
             elsePart.add(assignFalse)
             val ifelse = PtIfElse(assign.position)
@@ -754,7 +868,7 @@ internal class AssignmentAsmGen(private val program: PtProgram,
   sta  ${assign.target.asmVarname}+1""", false, assign.target.position)
                         }
                     } else {
-                        assignTrue = PtInlineAssembly("  lda  #1\n  sta  ${assign.target.asmVarname}", false, assign.target.position)
+                        assignTrue = PtInlineAssembly("\tlda  #1\n  sta  ${assign.target.asmVarname}", false, assign.target.position)
                     }
                 }
                 TargetStorageKind.MEMORY -> {
@@ -839,6 +953,7 @@ internal class AssignmentAsmGen(private val program: PtProgram,
     }
 
     private fun optimizedDivideExpr(expr: PtBinaryExpression, target: AsmAssignTarget): Boolean {
+        // replacing division by shifting is done in an optimizer step.
         when(expr.type) {
             DataType.UBYTE -> {
                 assignExpressionToRegister(expr.left, RegisterOrPair.A, false)
@@ -1010,7 +1125,10 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                             }
                         } else {
                             if (signed && shifts > 0) {
-                                asmgen.out("  ldy  #$shifts |  jsr  math.lsr_byte_A")
+                                if(shifts==1)
+                                    asmgen.out("  cmp  #$80 |  ror  a")
+                                else
+                                    asmgen.out("  ldy  #$shifts |  jsr  math.lsr_byte_A")
                             } else {
                                 repeat(shifts) {
                                     asmgen.out("  lsr  a")
@@ -1044,7 +1162,18 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                             }
                         } else {
                             if(signed && shifts>0) {
-                                asmgen.out("  ldx  #$shifts |  jsr  math.lsr_word_AY")
+                                if(shifts==1) {
+                                    asmgen.out("""
+                                        pha
+                                        tya
+                                        cmp  #$80
+                                        ror  a
+                                        tay
+                                        pla
+                                        ror  a""")
+                                }
+                                else
+                                    asmgen.out("  ldx  #$shifts |  jsr  math.lsr_word_AY")
                             } else {
                                 if(shifts>0) {
                                     asmgen.out("  sty  P8ZP_SCRATCH_B1")
@@ -1108,10 +1237,17 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                 }
                 is PtNumber -> {
                     assignExpressionToRegister(left, RegisterOrPair.A, dt==DataType.BYTE)
-                    if(expr.operator=="+")
-                        asmgen.out("  clc |  adc  #${right.number.toHex()}")
-                    else
-                        asmgen.out("  sec |  sbc  #${right.number.toHex()}")
+                    if(right.number==1.0 && asmgen.isTargetCpu(CpuType.CPU65c02)) {
+                        if (expr.operator == "+")
+                            asmgen.out("  inc  a")
+                        else
+                            asmgen.out("  dec  a")
+                    } else {
+                        if (expr.operator == "+")
+                            asmgen.out("  clc |  adc  #${right.number.toHex()}")
+                        else
+                            asmgen.out("  sec |  sbc  #${right.number.toHex()}")
+                    }
                     assignRegisterByte(target, CpuRegister.A, dt in SignedDatatypes, true)
                     return true
                 }
@@ -1246,8 +1382,39 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                 }
                 is PtNumber -> {
                     assignExpressionToRegister(left, RegisterOrPair.AY, dt==DataType.WORD)
-                    if(expr.operator=="+") {
-                        asmgen.out("""
+                    if(right.number==1.0 && asmgen.isTargetCpu(CpuType.CPU65c02)) {
+                        if(expr.operator=="+") {
+                            asmgen.out("""
+                                inc  a
+                                bne  +
+                                iny
++""")
+                        } else {
+                            asmgen.out("""
+                                dec  a
+                                cmp  #255
+                                bne  +
+                                dey
++""")
+                        }
+                    } else if(dt!=DataType.WORD && right.number.toInt() in 0..255) {
+                        if(expr.operator=="+") {
+                            asmgen.out("""
+                                clc
+                                adc  #${right.number.toHex()}
+                                bcc  +
+                                iny
++""")                   } else if(expr.operator=="-") {
+                            asmgen.out("""
+                                sec
+                                sbc  #${right.number.toHex()}
+                                bcs  +
+                                dey
++""")
+                        }
+                    } else {
+                        if(expr.operator=="+") {
+                            asmgen.out("""
                                 clc
                                 adc  #<${right.number.toHex()}
                                 tax
@@ -1255,8 +1422,8 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                                 adc  #>${right.number.toHex()}
                                 tay
                                 txa""")
-                    } else if(expr.operator=="-") {
-                        asmgen.out("""
+                        } else if(expr.operator=="-") {
+                            asmgen.out("""
                                 sec
                                 sbc  #<${right.number.toHex()}
                                 tax
@@ -1264,6 +1431,7 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                                 sbc  #>${right.number.toHex()}
                                 tay
                                 txa""")
+                        }
                     }
                     assignRegisterpairWord(target, RegisterOrPair.AY)
                     return true
@@ -1335,11 +1503,16 @@ internal class AssignmentAsmGen(private val program: PtProgram,
         // special optimization for  bytevalue +/- pointervar[y]  (actually: bytevalue +/-  @(address) )
         val address = mem.address as? PtBinaryExpression
         if(address is PtBinaryExpression) {
+            val constOffset = address.right.asConstInteger()
+            // check that the offset is actually a byte (so that it fits in a single register)
+            if(constOffset==null && address.right.type !in ByteDatatypes)
+                return false
+            if(constOffset!=null && constOffset !in -128..255)
+                return false
             val ptrVar = address.left as? PtIdentifier
             if(ptrVar!=null && asmgen.isZpVar(ptrVar)) {
                 assignExpressionToRegister(value, RegisterOrPair.A, false)
                 val pointername = asmgen.asmVariableName(ptrVar)
-                val constOffset = address.right.asConstInteger()
                 if (constOffset != null && constOffset < 256) {
                     // we have value + @(zpptr + 255), or value - @(zpptr+255)
                     asmgen.out("  ldy  #$constOffset")
@@ -1657,9 +1830,56 @@ internal class AssignmentAsmGen(private val program: PtProgram,
     }
 
     private fun containmentCheckIntoA(containment: PtContainmentCheck) {
-        val elementDt = containment.element.type
-        val symbol = asmgen.symbolTable.lookup(containment.iterable.name)!!
-        val symbolName = asmgen.asmVariableName(symbol, containment.definingSub())
+        val elementDt = containment.needle.type
+
+        if(containment.haystackValues!=null) {
+            val haystack = containment.haystackValues!!.children.map {
+                if(it is PtBool) it.asInt()
+                else (it as PtNumber).number.toInt()
+            }
+            when(elementDt) {
+                in ByteDatatypesWithBoolean -> {
+                    require(haystack.size in 0..PtContainmentCheck.MAX_SIZE_FOR_INLINE_CHECKS_BYTE)
+                    assignExpressionToRegister(containment.needle, RegisterOrPair.A, elementDt == DataType.BYTE)
+                    for(number in haystack) {
+                        asmgen.out("""
+                            cmp  #$number
+                            beq  +""")
+                    }
+                    asmgen.out("""
+                        lda  #0
+                        beq  ++
++                       lda  #1
++""")
+                }
+                in WordDatatypes -> {
+                    require(haystack.size in 0..PtContainmentCheck.MAX_SIZE_FOR_INLINE_CHECKS_WORD)
+                    assignExpressionToRegister(containment.needle, RegisterOrPair.AY, elementDt == DataType.WORD)
+                    val gottemLabel = asmgen.makeLabel("gottem")
+                    val endLabel = asmgen.makeLabel("end")
+                    for(number in haystack) {
+                        asmgen.out("""
+                            cmp  #<$number
+                            bne  +
+                            cpy  #>$number
+                            beq  $gottemLabel
++                       """)
+                    }
+                    asmgen.out("""
+                        lda  #0
+                        beq  $endLabel
+$gottemLabel            lda  #1
+$endLabel""")
+                }
+                DataType.FLOAT -> throw AssemblyError("containmentchecks for floats should always be done on an array variable with subroutine")
+                else -> throw AssemblyError("weird dt $elementDt")
+            }
+
+            return
+        }
+
+        val symbol = asmgen.symbolTable.lookup(containment.haystackHeapVar!!.name)!!
+        val symbolName = asmgen.asmVariableName(symbol, containment.definingISub())
         val (dt, numElements) = when(symbol) {
             is StStaticVariable  -> symbol.dt to symbol.length!!
             is StMemVar -> symbol.dt to symbol.length!!
@@ -1667,7 +1887,7 @@ internal class AssignmentAsmGen(private val program: PtProgram,
         }
         when(dt) {
             DataType.STR -> {
-                assignExpressionToRegister(containment.element, RegisterOrPair.A, elementDt == DataType.BYTE)
+                assignExpressionToRegister(containment.needle, RegisterOrPair.A, elementDt == DataType.BYTE)
                 asmgen.out("  pha")     // need to keep the scratch var safe so we have to do it in this order
                 assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, DataType.UWORD, containment.definingISub(), containment.position,"P8ZP_SCRATCH_W1"), symbolName, null, null)
                 asmgen.out("  pla")
@@ -1675,13 +1895,13 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                 asmgen.out("  jsr  prog8_lib.containment_bytearray")
             }
             DataType.ARRAY_F -> {
-                assignExpressionToRegister(containment.element, RegisterOrPair.FAC1, true)
+                assignExpressionToRegister(containment.needle, RegisterOrPair.FAC1, true)
                 assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, DataType.UWORD, containment.definingISub(), containment.position, "P8ZP_SCRATCH_W1"), symbolName, null, null)
                 asmgen.out("  ldy  #$numElements")
                 asmgen.out("  jsr  floats.containment_floatarray")
             }
-            DataType.ARRAY_B, DataType.ARRAY_UB -> {
-                assignExpressionToRegister(containment.element, RegisterOrPair.A, elementDt == DataType.BYTE)
+            DataType.ARRAY_B, DataType.ARRAY_UB, DataType.ARRAY_BOOL -> {
+                assignExpressionToRegister(containment.needle, RegisterOrPair.A, elementDt == DataType.BYTE)
                 asmgen.out("  pha")     // need to keep the scratch var safe so we have to do it in this order
                 assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, DataType.UWORD, containment.definingISub(), containment.position, "P8ZP_SCRATCH_W1"), symbolName, null, null)
                 asmgen.out("  pla")
@@ -1689,7 +1909,7 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                 asmgen.out("  jsr  prog8_lib.containment_bytearray")
             }
             DataType.ARRAY_W, DataType.ARRAY_UW -> {
-                assignExpressionToVariable(containment.element, "P8ZP_SCRATCH_W1", elementDt)
+                assignExpressionToVariable(containment.needle, "P8ZP_SCRATCH_W1", elementDt)
                 assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, DataType.UWORD, containment.definingISub(), containment.position, "P8ZP_SCRATCH_W2"), symbolName, null, null)
                 asmgen.out("  ldy  #$numElements")
                 asmgen.out("  jsr  prog8_lib.containment_wordarray")
@@ -3143,11 +3363,20 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                     }
                     else {
                         if (regs !in Cx16VirtualRegisters) {
-                            when (regs) {
-                                RegisterOrPair.AX -> asmgen.out("  pha |  txa |  pha")
-                                RegisterOrPair.AY -> asmgen.out("  pha |  tya |  pha")
-                                RegisterOrPair.XY -> asmgen.out("  txa |  pha |  tya |  pha")
-                                else -> throw AssemblyError("expected reg pair")
+                            if (asmgen.isTargetCpu(CpuType.CPU65c02)) {
+                                when (regs) {
+                                    RegisterOrPair.AX -> asmgen.out("  pha |  phx")
+                                    RegisterOrPair.AY -> asmgen.out("  pha |  phy")
+                                    RegisterOrPair.XY -> asmgen.out("  phx |  phy")
+                                    else -> throw AssemblyError("expected reg pair")
+                                }
+                            } else {
+                                when (regs) {
+                                    RegisterOrPair.AX -> asmgen.out("  pha |  txa |  pha")
+                                    RegisterOrPair.AY -> asmgen.out("  pha |  tya |  pha")
+                                    RegisterOrPair.XY -> asmgen.out("  txa |  pha |  tya |  pha")
+                                    else -> throw AssemblyError("expected reg pair")
+                                }
                             }
                             asmgen.loadScaledArrayIndexIntoRegister(target.array, CpuRegister.Y)
                             asmgen.out("""
@@ -3186,11 +3415,20 @@ internal class AssignmentAsmGen(private val program: PtProgram,
                     }
                     else {
                         if (regs !in Cx16VirtualRegisters) {
-                            when (regs) {
-                                RegisterOrPair.AX -> asmgen.out("  pha |  txa |  pha")
-                                RegisterOrPair.AY -> asmgen.out("  pha |  tya |  pha")
-                                RegisterOrPair.XY -> asmgen.out("  txa |  pha |  tya |  pha")
-                                else -> throw AssemblyError("expected reg pair")
+                            if (asmgen.isTargetCpu(CpuType.CPU65c02)) {
+                                when (regs) {
+                                    RegisterOrPair.AX -> asmgen.out("  pha |  phx")
+                                    RegisterOrPair.AY -> asmgen.out("  pha |  phy")
+                                    RegisterOrPair.XY -> asmgen.out("  phx |  phy")
+                                    else -> throw AssemblyError("expected reg pair")
+                                }
+                            } else {
+                                when (regs) {
+                                    RegisterOrPair.AX -> asmgen.out("  pha |  txa |  pha")
+                                    RegisterOrPair.AY -> asmgen.out("  pha |  tya |  pha")
+                                    RegisterOrPair.XY -> asmgen.out("  txa |  pha |  tya |  pha")
+                                    else -> throw AssemblyError("expected reg pair")
+                                }
                             }
                             asmgen.loadScaledArrayIndexIntoRegister(target.array, CpuRegister.Y)
                             asmgen.out("""
@@ -3689,17 +3927,8 @@ internal class AssignmentAsmGen(private val program: PtProgram,
 
         fun storeViaExprEval() {
             when(addressExpr) {
-                is PtNumber, is PtIdentifier -> {
-                    assignExpressionToVariable(addressExpr, "P8ZP_SCRATCH_W2", DataType.UWORD)
-                    asmgen.storeAIntoZpPointerVar("P8ZP_SCRATCH_W2", false)
-                }
-                else -> {
-                    // same as above but we need to save the A register
-                    asmgen.out("  pha")
-                    assignExpressionToVariable(addressExpr, "P8ZP_SCRATCH_W2", DataType.UWORD)
-                    asmgen.out("  pla")
-                    asmgen.storeAIntoZpPointerVar("P8ZP_SCRATCH_W2", false)
-                }
+                is PtNumber, is PtIdentifier -> storeByteInAToAddressExpression(addressExpr, false)
+                else -> storeByteInAToAddressExpression(addressExpr, true)
             }
         }
 

codeGenCpu6502/src/prog8/codegen/cpu6502/assignment/AugmentableAssignmentAsmGen.kt

@@ -225,7 +225,7 @@ internal class AugmentableAssignmentAsmGen(private val program: PtProgram,
                             SourceStorageKind.LITERALNUMBER -> inplacemodificationSplitWordWithLiteralval(target.asmVarname, target.datatype, index, operator, value.number!!.number.toInt())
                             else -> {
                                 // TODO: more optimized code for VARIABLE, REGISTER, MEMORY, ARRAY, EXPRESSION in the case of split-word arrays
-                                val scope = target.origAstTarget?.definingSub()
+                                val scope = target.origAstTarget?.definingISub()
                                 val regTarget = AsmAssignTarget.fromRegisters(RegisterOrPair.R0, false, target.position, scope, asmgen)
                                 val assignToReg = AsmAssignment(value, regTarget, program.memsizer, target.position)
                                 assignmentAsmGen.translateNormalAssignment(assignToReg, scope)
@@ -1086,6 +1086,21 @@ $shortcutLabel:""")
             }
         }
 
+        if(asmgen.isTargetCpu(CpuType.CPU65c02)) {
+            if(operator=="&" && value is PtPrefix && value.operator=="~") {
+                // M &= ~A  -->  use TRB 65c02 instruction for that
+                asmgen.assignExpressionToRegister(value.value, RegisterOrPair.A, dt in SignedDatatypes)
+                asmgen.out("  trb  $name")
+                return
+            }
+            else if(operator=="|") {
+                // M |= A --> use TSB 65c02 instruction for that
+                asmgen.assignExpressionToRegister(value, RegisterOrPair.A, dt in SignedDatatypes)
+                asmgen.out("  tsb  $name")
+                return
+            }
+        }
+
         // normal evaluation
         asmgen.assignExpressionToRegister(value, RegisterOrPair.A, dt in SignedDatatypes)
         inplacemodificationRegisterAwithVariableWithSwappedOperands(operator, name, dt in SignedDatatypes)
@@ -1094,6 +1109,15 @@ $shortcutLabel:""")
 
     private fun inplacemodificationByteVariableWithVariable(name: String, dt: DataType, operator: String, otherName: String) {
         // note: no logical and/or shortcut here, not worth it due to simple right operand
+
+        if(asmgen.isTargetCpu(CpuType.CPU65c02)) {
+            if(operator=="|") {
+                // M |= A  -->  use TSB 65c02 instruction for that
+                asmgen.out("  lda  $otherName |  tsb  $name")
+                return
+            }
+        }
+
         asmgen.out("  lda  $name")
         inplacemodificationRegisterAwithVariable(operator, otherName, dt in SignedDatatypes)
         asmgen.out("  sta  $name")
@@ -1459,6 +1483,7 @@ $shortcutLabel:""")
                     asmgen.out("  lda  $name |  ldy  #$value |  jsr  math.multiply_bytes |  sta  $name")
             }
             "/" -> {
+                // replacing division by shifting is done in an optimizer step.
                 if (dt == DataType.UBYTE)
                     asmgen.out("  lda  $name |  ldy  #$value |  jsr  math.divmod_ub_asm |  sty  $name")
                 else
@@ -1642,7 +1667,7 @@ $shortcutLabel:""")
     }
 
     private fun immediateOrInplace(name: String, value: Int) {
-        if(asmgen.isTargetCpu(CpuType.CPU65c02) && ((value and (value-1))==0)) {
+        if(asmgen.isTargetCpu(CpuType.CPU65c02)) {
             asmgen.out(" lda  #$value |  tsb  $name")       // set bit
         } else  {
             asmgen.out(" lda  $name |  ora  #$value |  sta  $name")
@@ -1807,33 +1832,36 @@ $shortcutLabel:""")
                 }
             }
             "/" -> {
-                if(value==0)
+                // replacing division by shifting is done in an optimizer step.
+                if(value==0) {
                     throw AssemblyError("division by zero")
-                if(dt==DataType.WORD) {
-                    asmgen.out("""
-                        lda  $lsb
-                        ldy  $msb
-                        sta  P8ZP_SCRATCH_W1
-                        sty  P8ZP_SCRATCH_W1+1
-                        lda  #<$value
-                        ldy  #>$value
-                        jsr  math.divmod_w_asm
-                        sta  $lsb
-                        sty  $msb
-                    """)
-                }
-                else {
-                    asmgen.out("""
-                        lda  $lsb
-                        ldy  $msb
-                        sta  P8ZP_SCRATCH_W1
-                        sty  P8ZP_SCRATCH_W1+1
-                        lda  #<$value
-                        ldy  #>$value
-                        jsr  math.divmod_uw_asm
-                        sta  $lsb
-                        sty  $msb
-                    """)
+                } else {
+                    if(dt==DataType.WORD) {
+                        asmgen.out("""
+                            lda  $lsb
+                            ldy  $msb
+                            sta  P8ZP_SCRATCH_W1
+                            sty  P8ZP_SCRATCH_W1+1
+                            lda  #<$value
+                            ldy  #>$value
+                            jsr  math.divmod_w_asm
+                            sta  $lsb
+                            sty  $msb
+                        """)
+                    }
+                    else {
+                        asmgen.out("""
+                            lda  $lsb
+                            ldy  $msb
+                            sta  P8ZP_SCRATCH_W1
+                            sty  P8ZP_SCRATCH_W1+1
+                            lda  #<$value
+                            ldy  #>$value
+                            jsr  math.divmod_uw_asm
+                            sta  $lsb
+                            sty  $msb
+                        """)
+                    }
                 }
             }
             "%" -> {

codeGenIntermediate/build.gradle

@@ -29,8 +29,9 @@ dependencies {
     // implementation "org.jetbrains.kotlin:kotlin-reflect"
     implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:2.0.0"
 
-    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.9.0'
+    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.9.1'
     testImplementation 'org.junit.jupiter:junit-jupiter:5.9.2'
+    testImplementation 'io.kotest:kotest-framework-datatest:5.9.1'
     testRuntimeOnly 'org.junit.platform:junit-platform-launcher'
 
 }

codeGenIntermediate/codeGenIntermediate.iml

@@ -14,6 +14,7 @@
     <orderEntry type="module" module-name="intermediate" />
     <orderEntry type="library" name="io.kotest.assertions.core.jvm" level="project" />
     <orderEntry type="library" name="io.kotest.runner.junit5.jvm" level="project" />
+    <orderEntry type="library" name="io.kotest.framework.datatest" level="project" />
     <orderEntry type="library" name="michael.bull.kotlin.result.jvm" level="project" />
   </component>
 </module>

codeGenIntermediate/src/prog8/codegen/intermediate/AssignmentGen.kt

@@ -57,10 +57,7 @@ internal class AssignmentGen(private val codeGen: IRCodeGen, private val express
             RegisterOrPair.XY -> IRInstruction(Opcode.LOADHXY, IRDataType.WORD, reg1=regNum)
             in Cx16VirtualRegisters -> IRInstruction(Opcode.LOADM, IRDataType.WORD, reg1=regNum, labelSymbol = "cx16.${returns.register.registerOrPair.toString().lowercase()}")
             null -> {
-                when(returns.register.statusflag) {
-                    Statusflag.Pc -> IRInstruction(Opcode.LOADHA, IRDataType.BYTE, reg1=regNum)
-                    else -> throw AssemblyError("weird statusflag as returnvalue")
-                }
+                TODO("assign CPU status flag ${returns.register.statusflag!!}")
             }
             else -> throw AssemblyError("cannot load register")
         }
@@ -1390,9 +1387,9 @@ internal class AssignmentGen(private val codeGen: IRCodeGen, private val express
                 val valueReg = codeGen.registers.nextFree()
                 result += IRCodeChunk(null, null).also {
                     it += IRInstruction(Opcode.LOADM, vmDt, reg1=valueReg, labelSymbol = array.variable.name, symbolOffset = constIndex*eltSize)
-                    when(comparisonOperator) {
-                        "==" -> it += IRInstruction(Opcode.SZ, vmDt, reg1=cmpResultReg, reg2=valueReg)
-                        "!=" -> it += IRInstruction(Opcode.SNZ, vmDt, reg1=cmpResultReg, reg2=valueReg)
+                    it += when(comparisonOperator) {
+                        "==" -> IRInstruction(Opcode.SZ, vmDt, reg1=cmpResultReg, reg2=valueReg)
+                        "!=" -> IRInstruction(Opcode.SNZ, vmDt, reg1=cmpResultReg, reg2=valueReg)
                         "<" -> return null  // TODO("array <0 inplace")) // TODO?
                         "<=" -> return null  // TODO("array <=0 inplace")) // TODO?
                         ">" -> return null  // TODO("array >0 inplace")) // TODO?
@@ -1413,9 +1410,9 @@ internal class AssignmentGen(private val codeGen: IRCodeGen, private val express
                 addToResult(result, indexTr, indexTr.resultReg, -1)
                 result += IRCodeChunk(null, null).also {
                     it += IRInstruction(Opcode.LOADX, vmDt, reg1=valueReg, reg2=indexTr.resultReg, labelSymbol = array.variable.name)
-                    when(comparisonOperator) {
-                        "==" -> it += IRInstruction(Opcode.SZ, vmDt, reg1=cmpResultReg, reg2=valueReg)
-                        "!=" -> it += IRInstruction(Opcode.SNZ, vmDt, reg1=cmpResultReg, reg2=valueReg)
+                    it += when(comparisonOperator) {
+                        "==" -> IRInstruction(Opcode.SZ, vmDt, reg1=cmpResultReg, reg2=valueReg)
+                        "!=" -> IRInstruction(Opcode.SNZ, vmDt, reg1=cmpResultReg, reg2=valueReg)
                         "<" -> return null  // TODO("array <0 inplace")) // TODO?
                         "<=" -> return null  // TODO("array <=0 inplace")) // TODO?
                         ">" -> return null  // TODO("array >0 inplace")) // TODO?

codeGenIntermediate/src/prog8/codegen/intermediate/BuiltinFuncGen.kt

@@ -9,8 +9,6 @@ internal class BuiltinFuncGen(private val codeGen: IRCodeGen, private val exprGe
 
     fun translate(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         return when(call.name) {
-            "any" -> funcAny(call)
-            "all" -> funcAll(call)
             "abs__byte", "abs__word", "abs__float" -> funcAbs(call)
             "cmp" -> funcCmp(call)
             "sgn" -> funcSgn(call)
@@ -34,14 +32,14 @@ internal class BuiltinFuncGen(private val codeGen: IRCodeGen, private val exprGe
             "clamp__byte", "clamp__ubyte", "clamp__word", "clamp__uword" -> funcClamp(call)
             "min__byte", "min__ubyte", "min__word", "min__uword" -> funcMin(call)
             "max__byte", "max__ubyte", "max__word", "max__uword" -> funcMax(call)
-            "sort" -> funcSort(call)
-            "reverse" -> funcReverse(call)
             "setlsb" -> funcSetLsbMsb(call, false)
             "setmsb" -> funcSetLsbMsb(call, true)
             "rol" -> funcRolRor(call)
             "ror" -> funcRolRor(call)
             "rol2" -> funcRolRor(call)
             "ror2" -> funcRolRor(call)
+            "prog8_ifelse_bittest_set" -> throw AssemblyError("prog8_ifelse_bittest_set() should have been translated as part of an ifElse statement")
+            "prog8_ifelse_bittest_notset" -> throw AssemblyError("prog8_ifelse_bittest_notset() should have been translated as part of an ifElse statement")
             "prog8_lib_stringcompare" -> funcStringCompare(call)
             "prog8_lib_square_byte" -> funcSquare(call, IRDataType.BYTE)
             "prog8_lib_square_word" -> funcSquare(call, IRDataType.WORD)
@@ -132,10 +130,10 @@ internal class BuiltinFuncGen(private val codeGen: IRCodeGen, private val exprGe
         val addressTr = exprGen.translateExpression(call.args[0])
         addToResult(result, addressTr, addressTr.resultReg, -1)
         addInstr(result, IRInstruction(Opcode.CALLI, reg1 = addressTr.resultReg), null)
-        if(call.void)
-            return ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
+        return if(call.void)
+            ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
         else
-            return ExpressionCodeResult(result, IRDataType.WORD, codeGen.registers.nextFree(), -1)      // TODO actually the result is returned in CPU registers AY...
+            ExpressionCodeResult(result, IRDataType.WORD, codeGen.registers.nextFree(), -1)      // TODO actually the result is returned in CPU registers AY...
     }
 
     private fun funcCallfar(call: PtBuiltinFunctionCall): ExpressionCodeResult {
@@ -204,79 +202,6 @@ internal class BuiltinFuncGen(private val codeGen: IRCodeGen, private val exprGe
         return ExpressionCodeResult(result, dt, leftTr.resultReg, -1)
     }
 
-    private fun funcAny(call: PtBuiltinFunctionCall): ExpressionCodeResult {
-        val arrayName = call.args[0] as PtIdentifier
-        val arrayLength = codeGen.symbolTable.getLength(arrayName.name)
-        val result = mutableListOf<IRCodeChunkBase>()
-        val lengthReg = codeGen.registers.nextFree()
-
-        if(arrayName.type in SplitWordArrayTypes) {
-            // any(lsb-array) or any(msb-array)
-            addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
-            val trLsb = exprGen.translateExpression(PtIdentifier(arrayName.name+"_lsb", DataType.ARRAY_UB, call.position))
-            addToResult(result, trLsb, trLsb.resultReg, -1)
-            addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = arrayLength), null)
-            result += codeGen.makeSyscall(IMSyscall.ANY_BYTE, listOf(IRDataType.WORD to trLsb.resultReg, IRDataType.BYTE to lengthReg), IRDataType.BYTE to trLsb.resultReg)
-            val shortcircuitLabel = codeGen.createLabelName()
-            result += IRCodeChunk(null, null).also {
-                it += IRInstruction(Opcode.CMPI, IRDataType.BYTE, reg1 = trLsb.resultReg, immediate = 0)
-                it += IRInstruction(Opcode.BSTNE, labelSymbol = shortcircuitLabel)
-                it += IRInstruction(Opcode.PREPARECALL, immediate = 2)
-            }
-            val trMsb = exprGen.translateExpression(PtIdentifier(arrayName.name+"_msb", DataType.ARRAY_UB, call.position))
-            addToResult(result, trMsb, trMsb.resultReg, -1)
-            addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = arrayLength), null)
-            result += codeGen.makeSyscall(IMSyscall.ANY_BYTE, listOf(IRDataType.WORD to trMsb.resultReg, IRDataType.BYTE to lengthReg), IRDataType.BYTE to trMsb.resultReg)
-            addInstr(result, IRInstruction(Opcode.ORR, IRDataType.BYTE, reg1=trLsb.resultReg, reg2=trMsb.resultReg), null)
-            result += IRCodeChunk(shortcircuitLabel, null)
-            return ExpressionCodeResult(result, IRDataType.BYTE, trLsb.resultReg, -1)
-        }
-
-        val syscall =
-            when (arrayName.type) {
-                DataType.ARRAY_UB,
-                DataType.ARRAY_B -> IMSyscall.ANY_BYTE
-                DataType.ARRAY_UW,
-                DataType.ARRAY_W -> IMSyscall.ANY_WORD
-                DataType.ARRAY_F -> IMSyscall.ANY_FLOAT
-                else -> throw IllegalArgumentException("weird type")
-            }
-        addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
-        val tr = exprGen.translateExpression(arrayName)
-        addToResult(result, tr, tr.resultReg, -1)
-        addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = arrayLength!! and 255), null)
-        result += codeGen.makeSyscall(syscall, listOf(IRDataType.WORD to tr.resultReg, IRDataType.BYTE to lengthReg), IRDataType.BYTE to tr.resultReg)
-        return ExpressionCodeResult(result, IRDataType.BYTE, tr.resultReg, -1)
-    }
-
-    private fun funcAll(call: PtBuiltinFunctionCall): ExpressionCodeResult {
-        val arrayName = call.args[0] as PtIdentifier
-        val arrayLength = codeGen.symbolTable.getLength(arrayName.name)
-
-        if(arrayName.type in SplitWordArrayTypes) {
-            // this is a bit complicated to calculate.... have to check all recombined (lsb,msb) words for $0000
-            TODO("all(split words $arrayName)")
-        }
-
-        val syscall =
-            when(arrayName.type) {
-                DataType.ARRAY_UB,
-                DataType.ARRAY_B -> IMSyscall.ALL_BYTE
-                DataType.ARRAY_UW,
-                DataType.ARRAY_W -> IMSyscall.ALL_WORD
-                DataType.ARRAY_F -> IMSyscall.ALL_FLOAT
-                else -> throw IllegalArgumentException("weird type")
-            }
-        val result = mutableListOf<IRCodeChunkBase>()
-        addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
-        val tr = exprGen.translateExpression(arrayName)
-        addToResult(result, tr, tr.resultReg, -1)
-        val lengthReg = codeGen.registers.nextFree()
-        addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = arrayLength!! and 255), null)
-        result += codeGen.makeSyscall(syscall, listOf(IRDataType.WORD to tr.resultReg, IRDataType.BYTE to lengthReg), IRDataType.BYTE to tr.resultReg)
-        return ExpressionCodeResult(result, IRDataType.BYTE, tr.resultReg, -1)
-    }
-
     private fun funcAbs(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val sourceDt = call.args.single().type
         val result = mutableListOf<IRCodeChunkBase>()
@@ -369,65 +294,6 @@ internal class BuiltinFuncGen(private val codeGen: IRCodeGen, private val exprGe
         }
     }
 
-    private fun funcReverse(call: PtBuiltinFunctionCall): ExpressionCodeResult {
-        val arrayName = call.args[0] as PtIdentifier
-        val arrayLength = codeGen.symbolTable.getLength(arrayName.name)
-        val lengthReg = codeGen.registers.nextFree()
-        val result = mutableListOf<IRCodeChunkBase>()
-
-        if(arrayName.type in SplitWordArrayTypes) {
-            // reverse the lsb and msb arrays both, independently
-            addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
-            val trLsb = exprGen.translateExpression(PtIdentifier(arrayName.name+"_lsb", DataType.ARRAY_UB, call.position))
-            addToResult(result, trLsb, trLsb.resultReg, -1)
-            addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = if(arrayName.type==DataType.STR) arrayLength!!-1 else arrayLength), null)
-            result += codeGen.makeSyscall(IMSyscall.REVERSE_BYTES, listOf(IRDataType.WORD to trLsb.resultReg, IRDataType.BYTE to lengthReg), null)
-            val trMsb = exprGen.translateExpression(PtIdentifier(arrayName.name+"_msb", DataType.ARRAY_UB, call.position))
-            addToResult(result, trMsb, trMsb.resultReg, -1)
-            addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = if(arrayName.type==DataType.STR) arrayLength!!-1 else arrayLength), null)
-            result += codeGen.makeSyscall(IMSyscall.REVERSE_BYTES, listOf(IRDataType.WORD to trMsb.resultReg, IRDataType.BYTE to lengthReg), null)
-            return ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
-        }
-
-        val syscall =
-            when(arrayName.type) {
-                DataType.ARRAY_UB, DataType.ARRAY_B, DataType.STR -> IMSyscall.REVERSE_BYTES
-                DataType.ARRAY_UW, DataType.ARRAY_W -> IMSyscall.REVERSE_WORDS
-                DataType.ARRAY_F -> IMSyscall.REVERSE_FLOATS
-                else -> throw IllegalArgumentException("weird type to reverse")
-            }
-        addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
-        val tr = exprGen.translateExpression(arrayName)
-        addToResult(result, tr, tr.resultReg, -1)
-        addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = if(arrayName.type==DataType.STR) arrayLength!!-1 else arrayLength), null)
-        result += codeGen.makeSyscall(syscall, listOf(IRDataType.WORD to tr.resultReg, IRDataType.BYTE to lengthReg), null)
-        return ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
-    }
-
-    private fun funcSort(call: PtBuiltinFunctionCall): ExpressionCodeResult {
-        val arrayName = call.args[0] as PtIdentifier
-        val arrayLength = codeGen.symbolTable.getLength(arrayName.name)
-        val syscall =
-            when(arrayName.type) {
-                DataType.ARRAY_UB -> IMSyscall.SORT_UBYTE
-                DataType.ARRAY_B -> IMSyscall.SORT_BYTE
-                DataType.ARRAY_UW -> IMSyscall.SORT_UWORD
-                DataType.ARRAY_W -> IMSyscall.SORT_WORD
-                DataType.STR -> IMSyscall.SORT_UBYTE
-                DataType.ARRAY_F -> throw IllegalArgumentException("sorting a floating point array is not supported")
-                in SplitWordArrayTypes -> TODO("split word sort")
-                else -> throw IllegalArgumentException("weird type to sort")
-            }
-        val result = mutableListOf<IRCodeChunkBase>()
-        addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
-        val tr = exprGen.translateExpression(arrayName)
-        addToResult(result, tr, tr.resultReg, -1)
-        val lengthReg = codeGen.registers.nextFree()
-        addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = if(arrayName.type==DataType.STR) arrayLength!!-1 else arrayLength), null)
-        result += codeGen.makeSyscall(syscall, listOf(IRDataType.WORD to tr.resultReg, IRDataType.BYTE to lengthReg), null)
-        return ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
-    }
-
     private fun funcMkword(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
         val resultReg = codeGen.registers.nextFree()
@@ -714,6 +580,7 @@ internal class BuiltinFuncGen(private val codeGen: IRCodeGen, private val exprGe
         val arr = (arg as? PtArrayIndexer)
         val index = arr?.index?.asConstInteger()
         if(arr!=null && index!=null) {
+            if(arr.splitWords)  TODO("IR rol/ror on split words array")
             val variable = arr.variable.name
             val itemsize = codeGen.program.memsizer.memorySize(arr.type)
             addInstr(result, IRInstruction(opcodeMemAndReg.first, vmDt, labelSymbol = variable, symbolOffset = index*itemsize), null)

codeGenIntermediate/src/prog8/codegen/intermediate/ExpressionGen.kt

@@ -1,5 +1,6 @@
 package prog8.codegen.intermediate
 
+import prog8.code.StNode
 import prog8.code.StRomSub
 import prog8.code.StSub
 import prog8.code.ast.*
@@ -167,55 +168,93 @@ internal class ExpressionGen(private val codeGen: IRCodeGen) {
     }
 
     private fun translate(check: PtContainmentCheck): ExpressionCodeResult {
+        val elementDt = check.needle.type
         val result = mutableListOf<IRCodeChunkBase>()
-        when(check.iterable.type) {
+
+        if(check.haystackValues!=null) {
+            val haystack = check.haystackValues!!.children.map {
+                if(it is PtBool) it.asInt()
+                else (it as PtNumber).number.toInt()
+            }
+            when(elementDt) {
+                in IntegerDatatypesWithBoolean -> {
+                    if (elementDt in ByteDatatypesWithBoolean) require(haystack.size in 0..PtContainmentCheck.MAX_SIZE_FOR_INLINE_CHECKS_BYTE)
+                    if (elementDt in WordDatatypes) require(haystack.size in 0..PtContainmentCheck.MAX_SIZE_FOR_INLINE_CHECKS_WORD)
+                    val gottemLabel = codeGen.createLabelName()
+                    val endLabel = codeGen.createLabelName()
+                    val elementTr = translateExpression(check.needle)
+                    addToResult(result, elementTr, elementTr.resultReg, -1)
+                    result += IRCodeChunk(null, null).also {
+                        for(value in haystack){
+                            it += IRInstruction(Opcode.CMPI, irType(elementDt), elementTr.resultReg, immediate = value)
+                            it += IRInstruction(Opcode.BSTEQ, labelSymbol = gottemLabel)
+                        }
+                        it += IRInstruction(Opcode.LOAD, IRDataType.BYTE, elementTr.resultReg, immediate = 0)
+                        it += IRInstruction(Opcode.JUMP, labelSymbol = endLabel)
+                    }
+                    addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, elementTr.resultReg, immediate = 1), gottemLabel)
+                    result += IRCodeChunk(endLabel, null)
+                    return ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
+
+                }
+                DataType.FLOAT -> throw AssemblyError("containmentchecks for floats should always be done on an array variable with subroutine")
+                else -> throw AssemblyError("weird dt $elementDt")
+            }
+        }
+
+        val haystackVar = check.haystackHeapVar!!
+        when(haystackVar.type) {
             DataType.STR -> {
                 addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
-                val elementTr = translateExpression(check.element)
+                val elementTr = translateExpression(check.needle)
                 addToResult(result, elementTr, elementTr.resultReg, -1)
-                val iterableTr = translateExpression(check.iterable)
+                val iterableTr = translateExpression(haystackVar)
                 addToResult(result, iterableTr, iterableTr.resultReg, -1)
                 result += codeGen.makeSyscall(IMSyscall.STRING_CONTAINS, listOf(IRDataType.BYTE to elementTr.resultReg, IRDataType.WORD to iterableTr.resultReg), IRDataType.BYTE to elementTr.resultReg)
+                addInstr(result, IRInstruction(Opcode.CMPI, IRDataType.BYTE, reg1=elementTr.resultReg, immediate = 0), null)
                 return ExpressionCodeResult(result, IRDataType.BYTE, elementTr.resultReg, -1)
             }
             DataType.ARRAY_UB, DataType.ARRAY_B -> {
                 addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 3), null)
-                val elementTr = translateExpression(check.element)
+                val elementTr = translateExpression(check.needle)
                 addToResult(result, elementTr, elementTr.resultReg, -1)
-                val iterableTr = translateExpression(check.iterable)
+                val iterableTr = translateExpression(haystackVar)
                 addToResult(result, iterableTr, iterableTr.resultReg, -1)
                 val lengthReg = codeGen.registers.nextFree()
-                val iterableLength = codeGen.symbolTable.getLength(check.iterable.name)
+                val iterableLength = codeGen.symbolTable.getLength(haystackVar.name)
                 addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1=lengthReg, immediate = iterableLength!!), null)
                 result += codeGen.makeSyscall(IMSyscall.BYTEARRAY_CONTAINS, listOf(IRDataType.BYTE to elementTr.resultReg, IRDataType.WORD to iterableTr.resultReg, IRDataType.BYTE to lengthReg), IRDataType.BYTE to elementTr.resultReg)
+                addInstr(result, IRInstruction(Opcode.CMPI, IRDataType.BYTE, reg1=elementTr.resultReg, immediate = 0), null)
                 return ExpressionCodeResult(result, IRDataType.BYTE, elementTr.resultReg, -1)
             }
             DataType.ARRAY_UW, DataType.ARRAY_W -> {
                 addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 3), null)
-                val elementTr = translateExpression(check.element)
+                val elementTr = translateExpression(check.needle)
                 addToResult(result, elementTr, elementTr.resultReg, -1)
-                val iterableTr = translateExpression(check.iterable)
+                val iterableTr = translateExpression(haystackVar)
                 addToResult(result, iterableTr, iterableTr.resultReg, -1)
                 val lengthReg = codeGen.registers.nextFree()
-                val iterableLength = codeGen.symbolTable.getLength(check.iterable.name)
+                val iterableLength = codeGen.symbolTable.getLength(haystackVar.name)
                 addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1=lengthReg, immediate = iterableLength!!), null)
                 result += codeGen.makeSyscall(IMSyscall.WORDARRAY_CONTAINS, listOf(IRDataType.WORD to elementTr.resultReg, IRDataType.WORD to iterableTr.resultReg, IRDataType.BYTE to lengthReg), IRDataType.BYTE to elementTr.resultReg)
+                addInstr(result, IRInstruction(Opcode.CMPI, IRDataType.BYTE, reg1=elementTr.resultReg, immediate = 0), null)
                 return ExpressionCodeResult(result, IRDataType.BYTE, elementTr.resultReg, -1)
             }
             DataType.ARRAY_F -> {
                 addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 3), null)
-                val elementTr = translateExpression(check.element)
+                val elementTr = translateExpression(check.needle)
                 addToResult(result, elementTr, -1, elementTr.resultFpReg)
-                val iterableTr = translateExpression(check.iterable)
+                val iterableTr = translateExpression(haystackVar)
                 addToResult(result, iterableTr, iterableTr.resultReg, -1)
                 val lengthReg = codeGen.registers.nextFree()
                 val resultReg = codeGen.registers.nextFree()
-                val iterableLength = codeGen.symbolTable.getLength(check.iterable.name)
+                val iterableLength = codeGen.symbolTable.getLength(haystackVar.name)
                 addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1=lengthReg, immediate = iterableLength!!), null)
                 result += codeGen.makeSyscall(IMSyscall.FLOATARRAY_CONTAINS, listOf(IRDataType.FLOAT to elementTr.resultFpReg, IRDataType.WORD to iterableTr.resultReg, IRDataType.BYTE to lengthReg), IRDataType.BYTE to resultReg)
+                addInstr(result, IRInstruction(Opcode.CMPI, IRDataType.BYTE, reg1=resultReg, immediate = 0), null)
                 return ExpressionCodeResult(result, IRDataType.BYTE, resultReg, -1)
             }
-            else -> throw AssemblyError("weird iterable dt ${check.iterable.type} for ${check.iterable.name}")
+            else -> throw AssemblyError("weird iterable dt ${haystackVar.type} for ${haystackVar.name}")
         }
     }
 
@@ -451,7 +490,36 @@ internal class ExpressionGen(private val codeGen: IRCodeGen) {
     }
 
     fun translate(fcall: PtFunctionCall): ExpressionCodeResult {
-        when (val callTarget = codeGen.symbolTable.flat.getValue(fcall.name)) {
+        val callTarget = codeGen.symbolTable.flat.getValue(fcall.name)
+
+        if(callTarget.scopedName in listOf("sys.push", "sys.pushw", "sys.pop", "sys.popw")) {
+            // special case, these should be inlined, or even use specialized instructions. Instead of doing a normal subroutine call.
+            return translateStackFunctions(fcall, callTarget)
+        }
+        when(callTarget.scopedName) {
+            "sys.clear_carry" -> {
+                val chunk = mutableListOf<IRCodeChunkBase>()
+                addInstr(chunk, IRInstruction(Opcode.CLC), null)
+                return ExpressionCodeResult(chunk, IRDataType.BYTE, -1, -1)
+            }
+            "sys.set_carry" -> {
+                val chunk = mutableListOf<IRCodeChunkBase>()
+                addInstr(chunk, IRInstruction(Opcode.SEC), null)
+                return ExpressionCodeResult(chunk, IRDataType.BYTE, -1, -1)
+            }
+            "sys.clear_irqd" -> {
+                val chunk = mutableListOf<IRCodeChunkBase>()
+                addInstr(chunk, IRInstruction(Opcode.CLI), null)
+                return ExpressionCodeResult(chunk, IRDataType.BYTE, -1, -1)
+            }
+            "sys.set_irqd" -> {
+                val chunk = mutableListOf<IRCodeChunkBase>()
+                addInstr(chunk, IRInstruction(Opcode.SEI), null)
+                return ExpressionCodeResult(chunk, IRDataType.BYTE, -1, -1)
+            }
+        }
+
+        when (callTarget) {
             is StSub -> {
                 val result = mutableListOf<IRCodeChunkBase>()
                 addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = callTarget.parameters.size), null)
@@ -590,6 +658,39 @@ internal class ExpressionGen(private val codeGen: IRCodeGen) {
         }
     }
 
+    private fun translateStackFunctions(fcall: PtFunctionCall, callTarget: StNode): ExpressionCodeResult {
+        val chunk = mutableListOf<IRCodeChunkBase>()
+        when(callTarget.scopedName) {
+            "sys.push" -> {
+                // push byte
+                val tr = translateExpression(fcall.args.single())
+                chunk += tr.chunks
+                addInstr(chunk, IRInstruction(Opcode.PUSH, IRDataType.BYTE, reg1=tr.resultReg), null)
+                return ExpressionCodeResult(chunk, IRDataType.BYTE, -1, -1)
+            }
+            "sys.pushw" -> {
+                // push word
+                val tr = translateExpression(fcall.args.single())
+                chunk += tr.chunks
+                addInstr(chunk, IRInstruction(Opcode.PUSH, IRDataType.WORD, reg1=tr.resultReg), null)
+                return ExpressionCodeResult(chunk, IRDataType.WORD, -1, -1)
+            }
+            "sys.pop" -> {
+                // pop byte
+                val popReg = codeGen.registers.nextFree()
+                addInstr(chunk, IRInstruction(Opcode.POP, IRDataType.BYTE, reg1=popReg), null)
+                return ExpressionCodeResult(chunk, IRDataType.BYTE, popReg, -1)
+            }
+            "sys.popw" -> {
+                // pop word
+                val popReg = codeGen.registers.nextFree()
+                addInstr(chunk, IRInstruction(Opcode.POP, IRDataType.WORD, reg1=popReg), null)
+                return ExpressionCodeResult(chunk, IRDataType.WORD, popReg, -1)
+            }
+            else -> throw AssemblyError("unknown stack subroutine called")
+        }
+    }
+
     private fun callRomSubWithMultipleReturnValues(
         callTarget: StRomSub,
         fcall: PtFunctionCall,

codeGenIntermediate/src/prog8/codegen/intermediate/IRCodeGen.kt

@@ -5,7 +5,6 @@ import prog8.code.ast.*
 import prog8.code.core.*
 import prog8.intermediate.*
 import kotlin.io.path.readBytes
-import kotlin.math.pow
 
 
 class IRCodeGen(
@@ -391,10 +390,10 @@ class IRCodeGen(
         val result = mutableListOf<IRCodeChunkBase>()
         when(iterable) {
             is PtRange -> {
-                if(iterable.from is PtNumber && iterable.to is PtNumber)
-                    result += translateForInConstantRange(forLoop, loopvar)
+                result += if(iterable.from is PtNumber && iterable.to is PtNumber)
+                    translateForInConstantRange(forLoop, loopvar)
                 else
-                    result += translateForInNonConstantRange(forLoop, loopvar)
+                    translateForInNonConstantRange(forLoop, loopvar)
             }
             is PtIdentifier -> {
                 require(forLoop.variable.name == loopvar.scopedName)
@@ -488,49 +487,76 @@ class IRCodeGen(
         val loopLabel = createLabelName()
         val result = mutableListOf<IRCodeChunkBase>()
 
-        val toTr = expressionEval.translateExpression(iterable.to)
-        addToResult(result, toTr, toTr.resultReg, -1)
-        val fromTr = expressionEval.translateExpression(iterable.from)
-        addToResult(result, fromTr, fromTr.resultReg, -1)
-
-        val labelAfterFor = createLabelName()
-        val precheckInstruction = if(loopvarDt in SignedDatatypes) {
-            if(step>0)
-                IRInstruction(Opcode.BGTSR, loopvarDtIr, fromTr.resultReg, toTr.resultReg, labelSymbol=labelAfterFor)
-            else
-                IRInstruction(Opcode.BGTSR, loopvarDtIr, toTr.resultReg, fromTr.resultReg, labelSymbol=labelAfterFor)
-        } else {
-            if(step>0)
-                IRInstruction(Opcode.BGTR, loopvarDtIr, fromTr.resultReg, toTr.resultReg, labelSymbol=labelAfterFor)
-            else
-                IRInstruction(Opcode.BGTR, loopvarDtIr, toTr.resultReg, fromTr.resultReg, labelSymbol=labelAfterFor)
-        }
-        addInstr(result, precheckInstruction, null)
-
-        addInstr(result, IRInstruction(Opcode.STOREM, loopvarDtIr, reg1=fromTr.resultReg, labelSymbol=loopvarSymbol), null)
-        result += labelFirstChunk(translateNode(forLoop.statements), loopLabel)
-        if(step==1 || step==-1) {
-            // if endvalue == loopvar, stop loop, else iterate
+        if(loopvarDtIr==IRDataType.BYTE && step==-1 && iterable.to.asConstInteger()==0) {
+            // downto 0 optimization (byte)
+            val fromTr = expressionEval.translateExpression(iterable.from)
+            addToResult(result, fromTr, fromTr.resultReg, -1)
+            addInstr(result, IRInstruction(Opcode.STOREM, loopvarDtIr, reg1=fromTr.resultReg, labelSymbol=loopvarSymbol), null)
+            result += labelFirstChunk(translateNode(forLoop.statements), loopLabel)
+            result += addConstMem(loopvarDtIr, null, loopvarSymbol, -1)
             result += IRCodeChunk(null, null).also {
                 it += IRInstruction(Opcode.LOADM, loopvarDtIr, reg1 = fromTr.resultReg, labelSymbol = loopvarSymbol)
-                it += IRInstruction(Opcode.CMP, loopvarDtIr, reg1=toTr.resultReg, reg2=fromTr.resultReg)
-                it += IRInstruction(Opcode.BSTEQ, labelSymbol = labelAfterFor)
+                it += IRInstruction(Opcode.CMPI, loopvarDtIr, reg1 = fromTr.resultReg, immediate = 255)
+                it += IRInstruction(Opcode.BSTNE, labelSymbol = loopLabel)
             }
-            result += addConstMem(loopvarDtIr, null, loopvarSymbol, step)
-            addInstr(result, IRInstruction(Opcode.JUMP, labelSymbol = loopLabel), null)
-        } else {
-            // ind/dec index, then:
-            // ascending: if endvalue >= loopvar, iterate
-            // descending: if loopvar >= endvalue, iterate
-            result += addConstMem(loopvarDtIr, null, loopvarSymbol, step)
-            addInstr(result, IRInstruction(Opcode.LOADM, loopvarDtIr, reg1=fromTr.resultReg, labelSymbol = loopvarSymbol), null)
-            if(step > 0)
-                addInstr(result, IRInstruction(Opcode.CMP, loopvarDtIr, reg1 = toTr.resultReg, fromTr.resultReg), null)
-            else
-                addInstr(result, IRInstruction(Opcode.CMP, loopvarDtIr, reg1 = fromTr.resultReg, toTr.resultReg), null)
-            addInstr(result, IRInstruction(Opcode.BSTPOS, labelSymbol = loopLabel), null)
         }
-        result += IRCodeChunk(labelAfterFor, null)
+        else if(step==-1 && iterable.to.asConstInteger()==1) {
+            // downto 1 optimization (byte and word)
+            val fromTr = expressionEval.translateExpression(iterable.from)
+            addToResult(result, fromTr, fromTr.resultReg, -1)
+            addInstr(result, IRInstruction(Opcode.STOREM, loopvarDtIr, reg1=fromTr.resultReg, labelSymbol=loopvarSymbol), null)
+            result += labelFirstChunk(translateNode(forLoop.statements), loopLabel)
+            result += addConstMem(loopvarDtIr, null, loopvarSymbol, -1)
+            result += IRCodeChunk(null, null).also {
+                it += IRInstruction(Opcode.LOADM, loopvarDtIr, reg1 = fromTr.resultReg, labelSymbol = loopvarSymbol)
+                it += IRInstruction(Opcode.BSTNE, labelSymbol = loopLabel)
+            }
+        }
+        else {
+            val toTr = expressionEval.translateExpression(iterable.to)
+            addToResult(result, toTr, toTr.resultReg, -1)
+            val fromTr = expressionEval.translateExpression(iterable.from)
+            addToResult(result, fromTr, fromTr.resultReg, -1)
+
+            val labelAfterFor = createLabelName()
+            val precheckInstruction = if(loopvarDt in SignedDatatypes) {
+                if(step>0)
+                    IRInstruction(Opcode.BGTSR, loopvarDtIr, fromTr.resultReg, toTr.resultReg, labelSymbol=labelAfterFor)
+                else
+                    IRInstruction(Opcode.BGTSR, loopvarDtIr, toTr.resultReg, fromTr.resultReg, labelSymbol=labelAfterFor)
+            } else {
+                if(step>0)
+                    IRInstruction(Opcode.BGTR, loopvarDtIr, fromTr.resultReg, toTr.resultReg, labelSymbol=labelAfterFor)
+                else
+                    IRInstruction(Opcode.BGTR, loopvarDtIr, toTr.resultReg, fromTr.resultReg, labelSymbol=labelAfterFor)
+            }
+            addInstr(result, precheckInstruction, null)
+
+            addInstr(result, IRInstruction(Opcode.STOREM, loopvarDtIr, reg1=fromTr.resultReg, labelSymbol=loopvarSymbol), null)
+            result += labelFirstChunk(translateNode(forLoop.statements), loopLabel)
+            if(step==1 || step==-1) {
+                // if endvalue == loopvar, stop loop, else iterate
+                result += IRCodeChunk(null, null).also {
+                    it += IRInstruction(Opcode.LOADM, loopvarDtIr, reg1 = fromTr.resultReg, labelSymbol = loopvarSymbol)
+                    it += IRInstruction(Opcode.CMP, loopvarDtIr, reg1=toTr.resultReg, reg2=fromTr.resultReg)
+                    it += IRInstruction(Opcode.BSTEQ, labelSymbol = labelAfterFor)
+                }
+                result += addConstMem(loopvarDtIr, null, loopvarSymbol, step)
+                addInstr(result, IRInstruction(Opcode.JUMP, labelSymbol = loopLabel), null)
+            } else {
+                // ind/dec index, then:
+                // ascending: if endvalue >= loopvar, iterate
+                // descending: if loopvar >= endvalue, iterate
+                result += addConstMem(loopvarDtIr, null, loopvarSymbol, step)
+                addInstr(result, IRInstruction(Opcode.LOADM, loopvarDtIr, reg1=fromTr.resultReg, labelSymbol = loopvarSymbol), null)
+                if(step > 0)
+                    addInstr(result, IRInstruction(Opcode.CMP, loopvarDtIr, reg1 = toTr.resultReg, fromTr.resultReg), null)
+                else
+                    addInstr(result, IRInstruction(Opcode.CMP, loopvarDtIr, reg1 = fromTr.resultReg, toTr.resultReg), null)
+                addInstr(result, IRInstruction(Opcode.BSTPOS, labelSymbol = loopLabel), null)
+            }
+            result += IRCodeChunk(labelAfterFor, null)
+        }
         return result
     }
 
@@ -559,9 +585,25 @@ class IRCodeGen(
         result += chunk
         result += labelFirstChunk(translateNode(forLoop.statements), loopLabel)
         val chunk2 = addConstMem(loopvarDtIr, null, loopvarSymbol, iterable.step)
-        chunk2 += IRInstruction(Opcode.LOADM, loopvarDtIr, reg1 = indexReg, labelSymbol = loopvarSymbol)
-        chunk2 += IRInstruction(Opcode.CMPI, loopvarDtIr, reg1 = indexReg, immediate = rangeEndExclusiveWrapped)
-        chunk2 += IRInstruction(Opcode.BSTNE, labelSymbol = loopLabel)
+        if(loopvarDtIr==IRDataType.BYTE && iterable.step==-1 && iterable.last==0) {
+            // downto 0 optimization (byte)
+            if(loopvarDt==DataType.BYTE || iterable.first<=127) {
+                chunk2 += IRInstruction(Opcode.BSTPOS, labelSymbol = loopLabel)
+            } else {
+                chunk2 += IRInstruction(Opcode.LOADM, loopvarDtIr, reg1 = indexReg, labelSymbol = loopvarSymbol)
+                chunk2 += IRInstruction(Opcode.CMPI, loopvarDtIr, reg1 = indexReg, immediate = rangeEndExclusiveWrapped)
+                chunk2 += IRInstruction(Opcode.BSTNE, labelSymbol = loopLabel)
+            }
+        }
+        else if(iterable.step==-1 && iterable.last==1) {
+            // downto 1 optimization (byte and word)
+            chunk2 += IRInstruction(Opcode.BSTNE, labelSymbol = loopLabel)
+        } else {
+            // downto some other value
+            chunk2 += IRInstruction(Opcode.LOADM, loopvarDtIr, reg1 = indexReg, labelSymbol = loopvarSymbol)
+            chunk2 += IRInstruction(Opcode.CMPI, loopvarDtIr, reg1 = indexReg, immediate = rangeEndExclusiveWrapped)
+            chunk2 += IRInstruction(Opcode.BSTNE, labelSymbol = loopLabel)
+        }
         result += chunk2
         return result
     }
@@ -682,13 +724,11 @@ class IRCodeGen(
         return code
     }
 
-    internal val powersOfTwo = (0..16).map { 2.0.pow(it.toDouble()).toInt() }
-
     internal fun multiplyByConst(dt: IRDataType, reg: Int, factor: Int): IRCodeChunk {
         val code = IRCodeChunk(null, null)
         if(factor==1)
             return code
-        val pow2 = powersOfTwo.indexOf(factor)
+        val pow2 = powersOfTwoInt.indexOf(factor)
         if(pow2==1) {
             // just shift 1 bit
             code += IRInstruction(Opcode.LSL, dt, reg1=reg)
@@ -712,7 +752,7 @@ class IRCodeGen(
         val code = IRCodeChunk(null, null)
         if(factor==1)
             return code
-        val pow2 = powersOfTwo.indexOf(factor)
+        val pow2 = powersOfTwoInt.indexOf(factor)
         if(pow2==1) {
             // just shift 1 bit
             code += if(knownAddress!=null)
@@ -785,19 +825,32 @@ class IRCodeGen(
         val code = IRCodeChunk(null, null)
         if(factor==1)
             return code
-        val pow2 = powersOfTwo.indexOf(factor)
-        if(pow2==1 && !signed) {
-            code += IRInstruction(Opcode.LSR, dt, reg1=reg)     // simple single bit shift
-        }
-        else if(pow2>=1 &&!signed) {
-            // just shift multiple bits
-            val pow2reg = registers.nextFree()
-            code += IRInstruction(Opcode.LOAD, dt, reg1=pow2reg, immediate = pow2)
-            code += if(signed)
-                IRInstruction(Opcode.ASRN, dt, reg1=reg, reg2=pow2reg)
-            else
-                IRInstruction(Opcode.LSRN, dt, reg1=reg, reg2=pow2reg)
+        val pow2 = powersOfTwoInt.indexOf(factor)
+        if(pow2>=0) {
+            if(signed) {
+                if(pow2==1) {
+                    // simple single bit shift (signed)
+                    code += IRInstruction(Opcode.ASR, dt, reg1=reg)
+                } else {
+                    // just shift multiple bits (signed)
+                    val pow2reg = registers.nextFree()
+                    code += IRInstruction(Opcode.LOAD, dt, reg1=pow2reg, immediate = pow2)
+                    code += IRInstruction(Opcode.ASRN, dt, reg1=reg, reg2=pow2reg)
+                }
+            } else {
+                if(pow2==1) {
+                    // simple single bit shift (unsigned)
+                    code += IRInstruction(Opcode.LSR, dt, reg1=reg)
+                } else {
+                    // just shift multiple bits (unsigned)
+                    val pow2reg = registers.nextFree()
+                    code += IRInstruction(Opcode.LOAD, dt, reg1 = pow2reg, immediate = pow2)
+                    code += IRInstruction(Opcode.LSRN, dt, reg1 = reg, reg2 = pow2reg)
+                }
+            }
+            return code
         } else {
+            // regular div
             code += if (factor == 0) {
                 IRInstruction(Opcode.LOAD, dt, reg1=reg, immediate = 0xffff)
             } else {
@@ -806,39 +859,56 @@ class IRCodeGen(
                 else
                     IRInstruction(Opcode.DIV, dt, reg1=reg, immediate = factor)
             }
+            return code
         }
-        return code
     }
 
     internal fun divideByConstInplace(dt: IRDataType, knownAddress: Int?, symbol: String?, factor: Int, signed: Boolean): IRCodeChunk {
         val code = IRCodeChunk(null, null)
         if(factor==1)
             return code
-        val pow2 = powersOfTwo.indexOf(factor)
-        if(pow2==1 && !signed) {
-            // just simple bit shift
-            code += if(knownAddress!=null)
-                IRInstruction(Opcode.LSRM, dt, address = knownAddress)
-            else
-                IRInstruction(Opcode.LSRM, dt, labelSymbol = symbol)
+        val pow2 = powersOfTwoInt.indexOf(factor)
+        if(pow2>=0) {
+            // can do bit shift instead of division
+            if(signed) {
+                if(pow2==1) {
+                    // just simple bit shift (signed)
+                    code += if (knownAddress != null)
+                        IRInstruction(Opcode.ASRM, dt, address = knownAddress)
+                    else
+                        IRInstruction(Opcode.ASRM, dt, labelSymbol = symbol)
+                } else {
+                    // just shift multiple bits (signed)
+                    val pow2reg = registers.nextFree()
+                    code += IRInstruction(Opcode.LOAD, dt, reg1 = pow2reg, immediate = pow2)
+                    code += if (knownAddress != null)
+                                IRInstruction(Opcode.ASRNM, dt, reg1 = pow2reg, address = knownAddress)
+                            else
+                                IRInstruction(Opcode.ASRNM, dt, reg1 = pow2reg, labelSymbol = symbol)
+                }
+            } else {
+                if(pow2==1) {
+                    // just simple bit shift (unsigned)
+                    code += if(knownAddress!=null)
+                        IRInstruction(Opcode.LSRM, dt, address = knownAddress)
+                    else
+                        IRInstruction(Opcode.LSRM, dt, labelSymbol = symbol)
+                }
+                else {
+                    // just shift multiple bits (unsigned)
+                    val pow2reg = registers.nextFree()
+                    code += IRInstruction(Opcode.LOAD, dt, reg1=pow2reg, immediate = pow2)
+                    code += if(knownAddress!=null)
+                                IRInstruction(Opcode.LSRNM, dt, reg1 = pow2reg, address = knownAddress)
+                            else
+                                IRInstruction(Opcode.LSRNM, dt, reg1 = pow2reg, labelSymbol = symbol)
+                }
+            }
+            return code
         }
-        else if(pow2>=1 && !signed) {
-            // just shift multiple bits
-            val pow2reg = registers.nextFree()
-            code += IRInstruction(Opcode.LOAD, dt, reg1=pow2reg, immediate = pow2)
-            code += if(signed) {
-                if(knownAddress!=null)
-                    IRInstruction(Opcode.ASRNM, dt, reg1 = pow2reg, address = knownAddress)
-                else
-                    IRInstruction(Opcode.ASRNM, dt, reg1 = pow2reg, labelSymbol = symbol)
-            }
-            else {
-                if(knownAddress!=null)
-                    IRInstruction(Opcode.LSRNM, dt, reg1 = pow2reg, address = knownAddress)
-                else
-                    IRInstruction(Opcode.LSRNM, dt, reg1 = pow2reg, labelSymbol = symbol)
-            }
-        } else {
+        else
+        {
+            // regular div
             if (factor == 0) {
                 val reg = registers.nextFree()
                 code += IRInstruction(Opcode.LOAD, dt, reg1=reg, immediate = 0xffff)
@@ -863,8 +933,8 @@ class IRCodeGen(
                         IRInstruction(Opcode.DIVM, dt, reg1 = factorReg, labelSymbol = symbol)
                 }
             }
+            return code
         }
-        return code
     }
 
     private fun translate(ifElse: PtIfElse): IRCodeChunks {
@@ -1291,6 +1361,10 @@ class IRCodeGen(
         val result = mutableListOf<IRCodeChunkBase>()
 
         fun translateSimple(condition: PtExpression, jumpFalseOpcode: Opcode) {
+
+            if(condition is PtBuiltinFunctionCall && condition.name.startsWith("prog8_ifelse_bittest_"))
+                throw AssemblyError("IR codegen doesn't have special instructions for dedicated BIT tests and should just still use normal AND")
+
             val tr = expressionEval.translateExpression(condition)
             result += tr.chunks
             if(ifElse.hasElse()) {
@@ -1503,17 +1577,16 @@ class IRCodeGen(
     private fun translate(jump: PtJump): IRCodeChunks {
         val result = mutableListOf<IRCodeChunkBase>()
         val chunk = IRCodeChunk(null, null)
-        if(jump.address!=null) {
-            chunk += IRInstruction(Opcode.JUMP, address = jump.address!!.toInt())
+        chunk += if(jump.address!=null) {
+            IRInstruction(Opcode.JUMP, address = jump.address!!.toInt())
         } else {
             if (jump.identifier != null) {
                 if(isIndirectJump(jump)) {
-                    chunk += IRInstruction(Opcode.JUMPI, labelSymbol = jump.identifier!!.name)
+                    IRInstruction(Opcode.JUMPI, labelSymbol = jump.identifier!!.name)
                 } else {
-                    chunk += IRInstruction(Opcode.JUMP, labelSymbol = jump.identifier!!.name)
+                    IRInstruction(Opcode.JUMP, labelSymbol = jump.identifier!!.name)
                 }
-            }
-            else
+            } else
                 throw AssemblyError("weird jump")
         }
         result += chunk
@@ -1618,7 +1691,7 @@ class IRCodeGen(
 
     private fun translate(parameters: List<PtSubroutineParameter>) =
         parameters.map {
-            val flattenedName = it.definingSub()!!.name + "." + it.name
+            val flattenedName = it.definingISub()!!.name + "." + it.name
             if(symbolTable.lookup(flattenedName)==null)
                 TODO("fix missing lookup for: $flattenedName   parameter")
             val orig = symbolTable.lookup(flattenedName) as StStaticVariable

codeGenIntermediate/src/prog8/codegen/intermediate/IRPeepholeOptimizer.kt

@@ -263,6 +263,24 @@ class IRPeepholeOptimizer(private val irprog: IRProgram) {
                     }
                 }
             }
+
+            if(ins.opcode== Opcode.SEI || ins.opcode== Opcode.CLI) {
+                if(idx < chunk.instructions.size-1) {
+                    val insAfter = chunk.instructions[idx+1]
+                    if(insAfter.opcode == ins.opcode) {
+                        chunk.instructions.removeAt(idx)
+                        changed = true
+                    }
+                    else if(ins.opcode== Opcode.SEI && insAfter.opcode== Opcode.CLI) {
+                        chunk.instructions.removeAt(idx)
+                        changed = true
+                    }
+                    else if(ins.opcode== Opcode.CLI && insAfter.opcode== Opcode.SEI) {
+                        chunk.instructions.removeAt(idx)
+                        changed = true
+                    }
+                }
+            }
         }
         return changed
     }

codeGenIntermediate/test/TestIRPeepholeOpt.kt

@@ -80,21 +80,52 @@ class TestIRPeepholeOpt: FunSpec({
         instr[1].opcode shouldBe Opcode.INC
     }
 
-    test("remove double sec/clc") {
+    test("remove double sec/clc/sei/cli") {
         val irProg = makeIRProgram(listOf(
             IRInstruction(Opcode.SEC),
             IRInstruction(Opcode.SEC),
             IRInstruction(Opcode.SEC),
             IRInstruction(Opcode.CLC),
             IRInstruction(Opcode.CLC),
-            IRInstruction(Opcode.CLC)
+            IRInstruction(Opcode.CLC),
+            IRInstruction(Opcode.SEI),
+            IRInstruction(Opcode.SEI),
+            IRInstruction(Opcode.SEI),
+            IRInstruction(Opcode.CLI),
+            IRInstruction(Opcode.CLI),
+            IRInstruction(Opcode.CLI),
         ))
-        irProg.chunks().single().instructions.size shouldBe 6
+        irProg.chunks().single().instructions.size shouldBe 12
         val opt = IRPeepholeOptimizer(irProg)
         opt.optimize(true, ErrorReporterForTests())
         val instr = irProg.chunks().single().instructions
-        instr.size shouldBe 1
+        instr.size shouldBe 2
         instr[0].opcode shouldBe Opcode.CLC
+        instr[1].opcode shouldBe Opcode.CLI
+    }
+
+    test("remove double sec/clc/sei/cli reversed") {
+        val irProg = makeIRProgram(listOf(
+            IRInstruction(Opcode.CLC),
+            IRInstruction(Opcode.CLC),
+            IRInstruction(Opcode.CLC),
+            IRInstruction(Opcode.SEC),
+            IRInstruction(Opcode.SEC),
+            IRInstruction(Opcode.SEC),
+            IRInstruction(Opcode.CLI),
+            IRInstruction(Opcode.CLI),
+            IRInstruction(Opcode.CLI),
+            IRInstruction(Opcode.SEI),
+            IRInstruction(Opcode.SEI),
+            IRInstruction(Opcode.SEI),
+        ))
+        irProg.chunks().single().instructions.size shouldBe 12
+        val opt = IRPeepholeOptimizer(irProg)
+        opt.optimize(true, ErrorReporterForTests())
+        val instr = irProg.chunks().single().instructions
+        instr.size shouldBe 2
+        instr[0].opcode shouldBe Opcode.SEC
+        instr[1].opcode shouldBe Opcode.SEI
     }
 
     test("push followed by pop") {

codeOptimizers/src/prog8/optimizer/ConstExprEvaluator.kt

@@ -66,7 +66,7 @@ class ConstExprEvaluator {
 //            DataType.WORD -> result = result.toShort().toInt()
 //            else -> { /* keep as it is */ }
 //        }
-        return NumericLiteral.optimalNumeric(result.toDouble(), left.position)
+        return NumericLiteral.optimalNumeric(left.type, null, result.toDouble(), left.position)
     }
 
     private fun bitwiseXor(left: NumericLiteral, right: NumericLiteral): NumericLiteral {
@@ -133,7 +133,7 @@ class ConstExprEvaluator {
         val error = "cannot add $left and $right"
         return when (left.type) {
             in IntegerDatatypes -> when (right.type) {
-                in IntegerDatatypes -> NumericLiteral.optimalInteger(left.number.toInt() + right.number.toInt(), left.position)
+                in IntegerDatatypes -> NumericLiteral.optimalInteger(left.type, right.type, left.number.toInt() + right.number.toInt(), left.position)
                 DataType.FLOAT -> NumericLiteral(DataType.FLOAT, left.number.toInt() + right.number, left.position)
                 else -> throw ExpressionError(error, left.position)
             }
@@ -150,7 +150,7 @@ class ConstExprEvaluator {
         val error = "cannot subtract $left and $right"
         return when (left.type) {
             in IntegerDatatypes -> when (right.type) {
-                in IntegerDatatypes -> NumericLiteral.optimalInteger(left.number.toInt() - right.number.toInt(), left.position)
+                in IntegerDatatypes -> NumericLiteral.optimalInteger(left.type, right.type, left.number.toInt() - right.number.toInt(), left.position)
                 DataType.FLOAT -> NumericLiteral(DataType.FLOAT, left.number.toInt() - right.number, left.position)
                 else -> throw ExpressionError(error, left.position)
             }
@@ -167,7 +167,7 @@ class ConstExprEvaluator {
         val error = "cannot multiply ${left.type} and ${right.type}"
         return when (left.type) {
             in IntegerDatatypes -> when (right.type) {
-                in IntegerDatatypes -> NumericLiteral.optimalInteger(left.number.toInt() * right.number.toInt(), left.position)
+                in IntegerDatatypes -> NumericLiteral.optimalInteger(left.type, right.type, left.number.toInt() * right.number.toInt(), left.position)
                 DataType.FLOAT -> NumericLiteral(DataType.FLOAT, left.number.toInt() * right.number, left.position)
                 else -> throw ExpressionError(error, left.position)
             }
@@ -190,7 +190,7 @@ class ConstExprEvaluator {
                 in IntegerDatatypes -> {
                     if(right.number.toInt()==0) divideByZeroError(right.position)
                     val result: Int = left.number.toInt() / right.number.toInt()
-                    NumericLiteral.optimalInteger(result, left.position)
+                    NumericLiteral.optimalInteger(left.type, right.type, result, left.position)
                 }
                 DataType.FLOAT -> {
                     if(right.number==0.0) divideByZeroError(right.position)
@@ -219,7 +219,7 @@ class ConstExprEvaluator {
             in IntegerDatatypes -> when (right.type) {
                 in IntegerDatatypes -> {
                     if(right.number.toInt()==0) divideByZeroError(right.position)
-                    NumericLiteral.optimalNumeric(left.number.toInt().toDouble() % right.number.toInt().toDouble(), left.position)
+                    NumericLiteral.optimalNumeric(left.type, right.type, left.number.toInt().toDouble() % right.number.toInt().toDouble(), left.position)
                 }
                 DataType.FLOAT -> {
                     if(right.number ==0.0) divideByZeroError(right.position)

codeOptimizers/src/prog8/optimizer/ConstantFoldingOptimizer.kt

@@ -33,6 +33,14 @@ class ConstantFoldingOptimizer(private val program: Program, private val errors:
     }
 
     override fun after(numLiteral: NumericLiteral, parent: Node): Iterable<IAstModification> {
+
+        if(numLiteral.type==DataType.LONG) {
+            // see if LONG values may be reduced to something smaller
+            val smaller = NumericLiteral.optimalInteger(numLiteral.number.toInt(), numLiteral.position)
+            if(smaller.type!=DataType.LONG)
+                return listOf(IAstModification.ReplaceNode(numLiteral, smaller, parent))
+        }
+
         if(parent is Assignment) {
             val iDt = parent.target.inferType(program)
             if(iDt.isKnown && !iDt.isBool && !iDt.istype(numLiteral.type)) {
@@ -91,7 +99,7 @@ class ConstantFoldingOptimizer(private val program: Program, private val errors:
                         program.encoding.encodeString(leftString.value, leftString.encoding) + program.encoding.encodeString(rightString.value, rightString.encoding),
                         leftString.encoding)
                 }
-                val concatStr = StringLiteral(concatenated, leftString.encoding, expr.position)
+                val concatStr = StringLiteral.create(concatenated, leftString.encoding, expr.position)
                 return listOf(IAstModification.ReplaceNode(expr, concatStr, parent))
             }
             else if(expr.operator=="*" && rightconst!=null && expr.left is StringLiteral) {
@@ -99,7 +107,7 @@ class ConstantFoldingOptimizer(private val program: Program, private val errors:
                 val part = expr.left as StringLiteral
                 if(part.value.isEmpty())
                     errors.warn("resulting string has length zero", part.position)
-                val newStr = StringLiteral(part.value.repeat(rightconst.number.toInt()), part.encoding, expr.position)
+                val newStr = StringLiteral.create(part.value.repeat(rightconst.number.toInt()), part.encoding, expr.position)
                 return listOf(IAstModification.ReplaceNode(expr, newStr, parent))
             }
         }
@@ -164,13 +172,13 @@ class ConstantFoldingOptimizer(private val program: Program, private val errors:
         if(leftconst==null && rightconst!=null && rightconst.number<0.0) {
             if (expr.operator == "-") {
                 // X - -1 ---> X + 1
-                val posNumber = NumericLiteral.optimalNumeric(-rightconst.number, rightconst.position)
+                val posNumber = NumericLiteral.optimalNumeric(rightconst.type, null, -rightconst.number, rightconst.position)
                 val plusExpr = BinaryExpression(expr.left, "+", posNumber, expr.position)
                 return listOf(IAstModification.ReplaceNode(expr, plusExpr, parent))
             }
             else if (expr.operator == "+") {
                 // X + -1 ---> X - 1
-                val posNumber = NumericLiteral.optimalNumeric(-rightconst.number, rightconst.position)
+                val posNumber = NumericLiteral.optimalNumeric(rightconst.type, null, -rightconst.number, rightconst.position)
                 val plusExpr = BinaryExpression(expr.left, "-", posNumber, expr.position)
                 return listOf(IAstModification.ReplaceNode(expr, plusExpr, parent))
             }
@@ -384,12 +392,13 @@ class ConstantFoldingOptimizer(private val program: Program, private val errors:
         val numval = decl.value as? NumericLiteral
         if(decl.type== VarDeclType.CONST && numval!=null) {
             val valueDt = numval.inferType(program)
+            if(valueDt istype DataType.LONG) {
+                return noModifications  // this is handled in the numericalvalue case
+            }
             if(valueDt isnot decl.datatype) {
-                if(decl.datatype!=DataType.BOOL || valueDt.isnot(DataType.UBYTE)) {
-                    val cast = numval.cast(decl.datatype, true)
-                    if (cast.isValid)
-                        return listOf(IAstModification.ReplaceNode(numval, cast.valueOrZero(), decl))
-                }
+                val cast = numval.cast(decl.datatype, true)
+                if (cast.isValid)
+                    return listOf(IAstModification.ReplaceNode(numval, cast.valueOrZero(), decl))
             }
         }
         return noModifications

codeOptimizers/src/prog8/optimizer/ConstantIdentifierReplacer.kt

@@ -154,11 +154,10 @@ class VarConstantValueTypeAdjuster(
         if(func==listOf("clamp")) {
             val t1 = functionCallExpr.args[0].inferType(program)
             if(t1.isKnown) {
-                val replaceFunc: String
-                if(t1.isBytes) {
-                    replaceFunc = if(t1.istype(DataType.BYTE)) "clamp__byte" else "clamp__ubyte"
+                val replaceFunc = if(t1.isBytes) {
+                    if(t1.istype(DataType.BYTE)) "clamp__byte" else "clamp__ubyte"
                 } else if(t1.isInteger) {
-                    replaceFunc = if(t1.istype(DataType.WORD)) "clamp__word" else "clamp__uword"
+                    if(t1.istype(DataType.WORD)) "clamp__word" else "clamp__uword"
                 } else {
                     errors.err("clamp builtin not supported for floats, use floats.clamp", functionCallExpr.position)
                     return noModifications
@@ -491,8 +490,7 @@ internal class ConstantIdentifierReplacer(
                 if(numericLv!=null) {
                     // arraysize initializer is a single value, and we know the array size.
                     if(numericLv.type!=DataType.BOOL) {
-                        if(options.strictBool || numericLv.type !in ByteDatatypes)
-                            errors.err("initializer value is not a boolean", numericLv.position)
+                        errors.err("initializer value is not a boolean", numericLv.position)
                         return null
                     }
                     val array = Array(size) {numericLv.number}.map { NumericLiteral(DataType.BOOL, it, numericLv.position) }.toTypedArray<Expression>()

codeOptimizers/src/prog8/optimizer/ExpressionSimplifier.kt

@@ -11,12 +11,9 @@ import prog8.ast.walk.IAstModification
 import prog8.code.core.*
 import kotlin.math.abs
 import kotlin.math.log2
-import kotlin.math.pow
 
-class ExpressionSimplifier(private val program: Program, private val options: CompilationOptions, private val errors: IErrorReporter) : AstWalker() {
-    private val powersOfTwo = (1..16).map { (2.0).pow(it) }.toSet()
-    private val negativePowersOfTwo = powersOfTwo.map { -it }.toSet()
 
+class ExpressionSimplifier(private val program: Program, private val errors: IErrorReporter) : AstWalker() {
     override fun after(typecast: TypecastExpression, parent: Node): Iterable<IAstModification> {
         val mods = mutableListOf<IAstModification>()
 
@@ -104,7 +101,7 @@ class ExpressionSimplifier(private val program: Program, private val options: Co
         val leftIDt = expr.left.inferType(program)
         val rightIDt = expr.right.inferType(program)
         if (!leftIDt.isKnown || !rightIDt.isKnown)
-            throw FatalAstException("can't determine datatype of both expression operands $expr")
+            return noModifications
 
         // X + (-A)  -->  X - A
         if (expr.operator == "+" && (expr.right as? PrefixExpression)?.operator == "-") {
@@ -249,20 +246,15 @@ class ExpressionSimplifier(private val program: Program, private val options: Co
                 }
             }
             if (rightVal?.number == 1.0) {
-                if (options.strictBool) {
-                    if (rightDt != leftDt) {
-                        val right = NumericLiteral(leftDt, rightVal.number, rightVal.position)
-                        return listOf(IAstModification.ReplaceNode(expr.right, right, expr))
-                    }
-                } else
-                    return listOf(IAstModification.ReplaceNode(expr, expr.left, parent))
+                if (rightDt != leftDt) {
+                    val right = NumericLiteral(leftDt, rightVal.number, rightVal.position)
+                    return listOf(IAstModification.ReplaceNode(expr.right, right, expr))
+                }
             }
             else if (rightVal?.number == 0.0) {
-                if (options.strictBool) {
-                    if (rightDt != leftDt) {
-                        val right = NumericLiteral(leftDt, rightVal.number, rightVal.position)
-                        return listOf(IAstModification.ReplaceNode(expr.right, right, expr))
-                    }
+                if (rightDt != leftDt) {
+                    val right = NumericLiteral(leftDt, rightVal.number, rightVal.position)
+                    return listOf(IAstModification.ReplaceNode(expr.right, right, expr))
                 }
             }
         }
@@ -277,21 +269,16 @@ class ExpressionSimplifier(private val program: Program, private val options: Co
                 }
             }
             if (rightVal?.number == 1.0) {
-                if(options.strictBool) {
-                    if(rightDt!=leftDt) {
-                        val right = NumericLiteral(leftDt, rightVal.number, rightVal.position)
-                        return listOf(IAstModification.ReplaceNode(expr.right, right, expr))
-                    }
+                if(rightDt!=leftDt) {
+                    val right = NumericLiteral(leftDt, rightVal.number, rightVal.position)
+                    return listOf(IAstModification.ReplaceNode(expr.right, right, expr))
                 }
             }
             else if (rightVal?.number == 0.0) {
-                if(options.strictBool) {
-                    if(rightDt!=leftDt) {
-                        val right = NumericLiteral(leftDt, rightVal.number, rightVal.position)
-                        return listOf(IAstModification.ReplaceNode(expr.right, right, expr))
-                    }
-                } else
-                    return listOf(IAstModification.ReplaceNode(expr, expr.left, parent))
+                if(rightDt!=leftDt) {
+                    val right = NumericLiteral(leftDt, rightVal.number, rightVal.position)
+                    return listOf(IAstModification.ReplaceNode(expr.right, right, expr))
+                }
             }
         }
 
@@ -686,7 +673,7 @@ class ExpressionSimplifier(private val program: Program, private val options: Co
                     if(!idt.isKnown)
                         throw FatalAstException("unknown dt")
                     return NumericLiteral(idt.getOr(DataType.UNDEFINED), 0.0, expr.position)
-                } else if (cv in powersOfTwo) {
+                } else if (cv in powersOfTwoFloat) {
                     expr.operator = "&"
                     expr.right = NumericLiteral.optimalInteger(cv!!.toInt()-1, expr.position)
                     return null
@@ -711,6 +698,7 @@ class ExpressionSimplifier(private val program: Program, private val options: Co
                 return null
             val leftDt = leftIDt.getOr(DataType.UNDEFINED)
             when (cv) {
+                0.0 -> return null // fall through to regular float division to properly deal with division by zero
                 -1.0 -> {
                     //  '/' -> -left
                     if (expr.operator == "/") {
@@ -738,12 +726,10 @@ class ExpressionSimplifier(private val program: Program, private val options: Co
                         else -> return null
                     }
                 }
-                in powersOfTwo -> {
-                    if (leftDt==DataType.UBYTE || leftDt==DataType.UWORD) {
-                        // Unsigned number divided by a power of two => shift right
-                        // Signed number can't simply be bitshifted in this case (due to rounding issues for negative values),
-                        // so we leave that as is and let the code generator deal with it.
-                        val numshifts = log2(cv).toInt()
+                in powersOfTwoFloat -> {
+                    val numshifts = powersOfTwoFloat.indexOf(cv)
+                    if (leftDt in IntegerDatatypes) {
+                        // division by a power of two => shift right (signed and unsigned)
                         return BinaryExpression(expr.left, ">>", NumericLiteral.optimalInteger(numshifts, expr.position), expr.position)
                     }
                 }
@@ -795,14 +781,14 @@ class ExpressionSimplifier(private val program: Program, private val options: Co
                     // left
                     return expr2.left
                 }
-                in powersOfTwo -> {
+                in powersOfTwoFloat -> {
                     if (leftValue.inferType(program).isInteger) {
                         // times a power of two => shift left
                         val numshifts = log2(cv).toInt()
                         return BinaryExpression(expr2.left, "<<", NumericLiteral.optimalInteger(numshifts, expr.position), expr.position)
                     }
                 }
-                in negativePowersOfTwo -> {
+                in negativePowersOfTwoFloat -> {
                     if (leftValue.inferType(program).isInteger) {
                         // times a negative power of two => negate, then shift
                         val numshifts = log2(-cv).toInt()

codeOptimizers/src/prog8/optimizer/Extensions.kt

@@ -59,8 +59,8 @@ fun Program.inlineSubroutines(options: CompilationOptions): Int {
     return inliner.applyModifications()
 }
 
-fun Program.simplifyExpressions(errors: IErrorReporter, options: CompilationOptions) : Int {
-    val opti = ExpressionSimplifier(this, options, errors)
+fun Program.simplifyExpressions(errors: IErrorReporter) : Int {
+    val opti = ExpressionSimplifier(this, errors)
     opti.visit(this)
     return opti.applyModifications()
 }

codeOptimizers/src/prog8/optimizer/Inliner.kt

@@ -30,11 +30,11 @@ class Inliner(private val program: Program, private val options: CompilationOpti
         }
 
         override fun visit(subroutine: Subroutine) {
-            if(!subroutine.isAsmSubroutine && !subroutine.inline && subroutine.parameters.isEmpty()) {
-                val containsSubsOrVariables = subroutine.statements.any { it is VarDecl || it is Subroutine}
-                if(!containsSubsOrVariables) {
-                    if(subroutine.statements.size==1 || (subroutine.statements.size==2 && isEmptyReturn(subroutine.statements[1]))) {
-                        if(subroutine !== program.entrypoint) {
+            if (!subroutine.isAsmSubroutine && !subroutine.inline && subroutine.parameters.isEmpty()) {
+                val containsSubsOrVariables = subroutine.statements.any { it is VarDecl || it is Subroutine }
+                if (!containsSubsOrVariables) {
+                    if (subroutine.statements.size == 1 || (subroutine.statements.size == 2 && isEmptyReturn(subroutine.statements[1]))) {
+                        if (subroutine !== program.entrypoint) {
                             // subroutine is possible candidate to be inlined
                             subroutine.inline =
                                 when (val stmt = subroutine.statements[0]) {
@@ -87,9 +87,9 @@ class Inliner(private val program: Program, private val options: CompilationOpti
                                                 } else
                                                     false
                                             targetInline || valueInline
-                                        } else if(stmt.target.identifier!=null && stmt.isAugmentable) {
+                                        } else if (stmt.target.identifier != null && stmt.isAugmentable) {
                                             val binExpr = stmt.value as BinaryExpression
-                                            if(binExpr.operator in "+-" && binExpr.right.constValue(program)?.number==1.0) {
+                                            if (binExpr.operator in "+-" && binExpr.right.constValue(program)?.number == 1.0) {
                                                 makeFullyScoped(stmt.target.identifier!!)
                                                 makeFullyScoped(binExpr.left as IdentifierReference)
                                                 true
@@ -121,8 +121,8 @@ class Inliner(private val program: Program, private val options: CompilationOpti
                         }
                     }
 
-                    if(subroutine.inline && subroutine.statements.size>1) {
-                        require(subroutine.statements.size==2 && isEmptyReturn(subroutine.statements[1]))
+                    if (subroutine.inline && subroutine.statements.size > 1) {
+                        require(subroutine.statements.size == 2 && isEmptyReturn(subroutine.statements[1]))
                         subroutine.statements.removeLast()      // get rid of the Return, to be able to inline the (single) statement preceding it.
                     }
                 }
@@ -147,6 +147,7 @@ class Inliner(private val program: Program, private val options: CompilationOpti
         }
 
         private fun makeFullyScoped(call: FunctionCallStatement) {
+            makeFullyScoped(call.target)
             call.target.targetSubroutine(program)?.let { sub ->
                 val scopedName = IdentifierReference(sub.scopedName, call.target.position)
                 val scopedArgs = makeScopedArgs(call.args)
@@ -169,6 +170,7 @@ class Inliner(private val program: Program, private val options: CompilationOpti
         }
 
         private fun makeFullyScoped(call: FunctionCallExpression) {
+            makeFullyScoped(call.target)
             call.target.targetSubroutine(program)?.let { sub ->
                 val scopedName = IdentifierReference(sub.scopedName, call.target.position)
                 val scopedArgs = makeScopedArgs(call.args)

codeOptimizers/src/prog8/optimizer/StatementOptimizer.kt

@@ -21,7 +21,8 @@ class StatementOptimizer(private val program: Program,
         if(functionCallStatement.target.nameInSource.size==1) {
             val functionName = functionCallStatement.target.nameInSource[0]
             if (functionName in functions.purefunctionNames) {
-                errors.warn("statement has no effect (function return value is discarded)", functionCallStatement.position)
+                if("ignore_unused" !in parent.definingBlock.options())
+                    errors.warn("statement has no effect (function return value is discarded)", functionCallStatement.position)
                 return listOf(IAstModification.Remove(functionCallStatement, parent as IStatementContainer))
             }
         }
@@ -199,35 +200,6 @@ class StatementOptimizer(private val program: Program,
             }
         }
 
-        val loopvarDt = forLoop.loopVarDt(program)
-        if(loopvarDt.istype(DataType.UWORD) || loopvarDt.istype(DataType.UBYTE)) {
-
-            fun incOrDec(inc: Boolean): Assignment {
-                val pos = forLoop.position
-                val loopVar = forLoop.loopVar
-                val addSubOne = BinaryExpression(loopVar.copy(), if(inc) "+" else "-", NumericLiteral.optimalInteger(1, pos), pos, false)
-                return Assignment(AssignTarget(loopVar.copy(), null, null, null, false, pos), addSubOne, AssignmentOrigin.USERCODE, pos)
-            }
-
-            if (range != null && range.to.constValue(program)?.number == 0.0 && range.step.constValue(program)?.number==-1.0) {
-                val fromExpr = range.from
-                if(fromExpr.constValue(program)==null) {
-                    // FOR X = something DOWNTO 0 {...} -->  X=something,  DO { ... , X-- } UNTIL X=255 (or 65535 if uword)
-                    val pos = forLoop.position
-                    val checkValue = NumericLiteral(loopvarDt.getOr(DataType.UNDEFINED), if(loopvarDt.istype(DataType.UBYTE)) 255.0 else 65535.0, pos)
-                    val condition = BinaryExpression(forLoop.loopVar.copy(), "==", checkValue, pos)
-                    val decOne = incOrDec(false)
-                    forLoop.body.statements.add(decOne)
-                    val replacement = AnonymousScope(mutableListOf(
-                        Assignment(AssignTarget(forLoop.loopVar.copy(), null, null, null, false, pos),
-                            fromExpr, AssignmentOrigin.OPTIMIZER, pos),
-                        UntilLoop(forLoop.body, condition, pos)
-                    ), pos)
-                    return listOf(IAstModification.ReplaceNode(forLoop, replacement, parent))
-                }
-            }
-        }
-
         return noModifications
     }
 

codeOptimizers/src/prog8/optimizer/UnusedCodeRemover.kt

@@ -149,7 +149,8 @@ class UnusedCodeRemover(private val program: Program,
                                         val declIndex = (parent as IStatementContainer).statements.indexOf(decl)
                                         val singleUseIndex = (parent as IStatementContainer).statements.indexOf(singleUse.parent)
                                         if(declIndex==singleUseIndex-1) {
-                                            errors.info("replaced unused variable '${decl.name}' with void call, maybe this can be removed altogether", decl.position)
+                                            if("ignore_unused" !in decl.definingBlock.options())
+                                                errors.info("replaced unused variable '${decl.name}' with void call, maybe this can be removed altogether", decl.position)
                                             val fcall = assignment.value as IFunctionCall
                                             val voidCall = FunctionCallStatement(fcall.target, fcall.args, true, fcall.position)
                                             return listOf(

compiler/build.gradle

@@ -2,7 +2,8 @@ plugins {
     id 'java'
     id 'application'
     id 'org.jetbrains.kotlin.jvm'
-    id 'com.github.johnrengelman.shadow' version '8.1.1'
+    // id 'com.github.johnrengelman.shadow' version '8.1.1'
+    id 'io.github.goooler.shadow' version '8.1.7'
     id 'com.peterabeles.gversion' version '1.10.2'
 }
 
@@ -31,7 +32,7 @@ dependencies {
     implementation project(':codeGenIntermediate')
     implementation project(':codeGenExperimental')
     implementation project(':virtualmachine')
-    implementation "org.antlr:antlr4-runtime:4.13.1"
+    implementation "org.antlr:antlr4-runtime:4.13.2"
     implementation "org.jetbrains.kotlin:kotlin-stdlib-jdk8"
     // implementation "org.jetbrains.kotlin:kotlin-reflect"
     implementation 'org.jetbrains.kotlinx:kotlinx-cli:0.3.6'
@@ -39,7 +40,8 @@ dependencies {
 
     testImplementation project(':codeCore')
     testImplementation project(':intermediate')
-    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.9.0'
+    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.9.1'
+    testImplementation 'io.kotest:kotest-framework-datatest:5.9.1'
     testImplementation 'org.junit.jupiter:junit-jupiter:5.9.2'
     testRuntimeOnly 'org.junit.platform:junit-platform-launcher'
 

compiler/compiler.iml

@@ -15,7 +15,7 @@
     <orderEntry type="library" name="michael.bull.kotlin.result.jvm" level="project" />
     <orderEntry type="library" name="io.kotest.assertions.core.jvm" level="project" />
     <orderEntry type="library" name="io.kotest.runner.junit5.jvm" level="project" />
-    <orderEntry type="library" name="antlr.antlr4" level="project" />
+    <orderEntry type="library" name="io.kotest.framework.datatest" level="project" />
     <orderEntry type="module" module-name="codeCore" />
     <orderEntry type="module" module-name="compilerAst" />
     <orderEntry type="module" module-name="codeOptimizers" />
@@ -24,5 +24,6 @@
     <orderEntry type="module" module-name="codeGenIntermediate" />
     <orderEntry type="module" module-name="virtualmachine" />
     <orderEntry type="module" module-name="intermediate" scope="TEST" />
+    <orderEntry type="library" name="antlr.antlr4" level="project" />
   </component>
 </module>

compiler/res/prog8lib/anyall.p8

@@ -0,0 +1,82 @@
+; any() and all()  checks on arrays/memory buffers.
+; These were builtin functions in older versions of the language.
+
+%option no_symbol_prefixing, ignore_unused
+
+anyall {
+    sub any(uword arrayptr, uword num_elements) -> bool {
+        ; -- returns true if any byte in the array is not zero.
+        cx16.r1 = arrayptr
+        if msb(num_elements)==0 {
+            for cx16.r0L in 0 to lsb(num_elements)-1 {
+                if cx16.r1[cx16.r0L]!=0
+                    return true
+            }
+            return false
+        }
+        repeat num_elements {
+            if @(cx16.r1)!=0
+                return true
+            cx16.r1++
+        }
+        return false
+    }
+
+    sub all(uword arrayptr, uword num_elements) -> bool {
+        ; -- returns true if all bytes in the array are not zero.
+        cx16.r1 = arrayptr
+        if msb(num_elements)==0 {
+            for cx16.r0L in 0 to lsb(num_elements)-1 {
+                if cx16.r1[cx16.r0L]==0
+                    return false
+            }
+            return true
+        }
+        repeat num_elements {
+            if @(cx16.r1)==0
+                return false
+            cx16.r1++
+        }
+        return true
+    }
+
+    sub anyw(uword arrayptr, uword num_elements) -> bool {
+        ; -- returns true if any word in the array is not zero.
+        ;    doesn't work on @split arrays.
+        cx16.r1 = arrayptr
+        if msb(num_elements)==0 {
+            repeat lsb(num_elements) {
+                if peekw(cx16.r1)!=0
+                    return true
+                cx16.r1+=2
+            }
+            return false
+        }
+        repeat num_elements {
+            if peekw(cx16.r1)!=0
+                return true
+            cx16.r1+=2
+        }
+        return false
+    }
+
+    sub allw(uword arrayptr, uword num_elements) -> bool {
+        ; -- returns true if all words in the array are not zero.
+        ;    doesn't work on @split arrays.
+        cx16.r1 = arrayptr
+        if msb(num_elements)==0 {
+            repeat lsb(num_elements) {
+                if peekw(cx16.r1)==0
+                    return false
+                cx16.r1+=2
+            }
+            return true
+        }
+        repeat num_elements {
+            if peekw(cx16.r1)==0
+                return false
+            cx16.r1+=2
+        }
+        return true
+    }
+}

compiler/res/prog8lib/buffers.p8

@@ -0,0 +1,152 @@
+; experimental buffer data structures
+
+%option no_symbol_prefixing, ignore_unused
+
+smallringbuffer {
+    ; -- A ringbuffer (FIFO queue) that occupies a single page in memory, containing 255 bytes maximum.
+    ;    You can store and retrieve words too.
+    ;    It's optimized for speed and depends on the byte-wrap-around feature when doing incs and decs.
+
+    ubyte fill
+    ubyte head
+    ubyte tail
+    ubyte[256] buffer
+
+    sub init() {
+        ; -- (re)initialize the ringbuffer, you must call this before using the other routines
+        head = fill = 0
+        tail = 255
+    }
+
+    sub put(ubyte value) -> bool {
+        ; -- store a byte in the buffer, returns success
+        if fill==255
+            return false
+        buffer[head] = value
+        head++
+        fill++
+        return true
+    }
+
+    sub putw(uword value) -> bool {
+        ; -- store a word in the buffer, returns success
+        if fill>=254
+            return false
+        fill += 2
+        buffer[head] = lsb(value)
+        head++
+        buffer[head] = msb(value)
+        head++
+        return true
+    }
+
+    sub get() -> ubyte {
+        ; -- retrieves a byte from the buffer. Also sets Carry flag: set=success, clear=buffer was empty
+        if fill==0 {
+            sys.clear_carry()
+            return 0
+        }
+        fill--
+        tail++
+        sys.set_carry()
+        return buffer[tail]
+    }
+
+    sub getw() -> uword {
+        ; -- retrieves a word from the buffer. Also sets Carry flag: set=success, clear=buffer was empty
+        if fill<2 {
+            sys.clear_carry()
+            return 0
+        }
+        fill -= 2
+        tail++
+        cx16.r0L = buffer[tail]
+        tail++
+        cx16.r0H = buffer[tail]
+        sys.set_carry()
+        return cx16.r0
+    }
+}
+
+
+ringbuffer {
+    ; -- A ringbuffer (FIFO queue) that occupies a single page in memory, containing 8 KB maximum.
+    ;    You can store and retrieve words too.
+
+    uword fill
+    uword head
+    uword tail
+    uword buffer_ptr = memory("ringbuffer", 8192, 0)
+
+    sub init() {
+        ; -- (re)initialize the ringbuffer, you must call this before using the other routines
+        head = fill = 0
+        tail = 8191
+    }
+
+    sub put(ubyte value) -> bool {
+        ; -- store a byte in the buffer, returns success
+        if fill==8192
+            return false
+        buffer_ptr[head] = value
+        inc_head()
+        fill++
+        return true
+    }
+
+    sub putw(uword value) -> bool {
+        ; -- store a word in the buffer, returns success
+        if fill>=8191
+            return false
+        fill += 2
+        buffer_ptr[head] = lsb(value)
+        inc_head()
+        buffer_ptr[head] = msb(value)
+        inc_head()
+        return true
+    }
+
+    sub get() -> ubyte {
+        ; -- retrieves a byte from the buffer. Also sets Carry flag: set=success, clear=buffer was empty
+        if fill==0 {
+            sys.clear_carry()
+            return 0
+        }
+        fill--
+        inc_tail()
+        cx16.r0L = buffer_ptr[tail]
+        sys.set_carry()
+        return cx16.r0L
+    }
+
+    sub getw() -> uword {
+        ; -- retrieves a word from the buffer. Also sets Carry flag: set=success, clear=buffer was empty
+        if fill<2 {
+            sys.clear_carry()
+            return 0
+        }
+        fill -= 2
+        inc_tail()
+        cx16.r0L = buffer_ptr[tail]
+        inc_tail()
+        cx16.r0H = buffer_ptr[tail]
+        sys.set_carry()
+        return cx16.r0
+    }
+
+    sub inc_head() {
+        head++
+        if msb(head)==$20
+            head=0
+    }
+
+    sub inc_tail() {
+        tail++
+        if msb(tail)==$20
+            tail=0
+    }
+}
+
+
+; TODO ringbuffer (FIFO queue) should use banked ram on the X16, but still work on virtual
+; TODO stack (LIFO queue) using more than 1 page of ram (maybe even banked ram on the x16)

compiler/res/prog8lib/c64/floats.p8

@@ -36,7 +36,7 @@ romsub $b1bf = AYINT() clobbers(A,X,Y)          ; fac1-> signed word in 100-101
 ; there is also floats.GIVUAYFAY - unsigned word in A/Y (lo/hi) to fac1
 ; there is also floats.FREADS32  that reads from 98-101 ($62-$65) MSB FIRST
 ; there is also floats.FREADUS32  that reads from 98-101 ($62-$65) MSB FIRST
-; there is also floats.FREADS24AXY  that reads signed int24 into fac1 from A/X/Y (lo/mid/hi bytes)
+; there is also floats.FREAD{S,U}24AXY  that read (un)signed int24 into fac1 from A/X/Y (lo/mid/hi bytes)
 romsub $b391 = GIVAYF(ubyte lo @ Y, ubyte hi @ A) clobbers(A,X,Y)
 
 romsub $b3a2 = FREADUY(ubyte value @ Y) clobbers(A,X,Y)     ; 8 bit unsigned Y -> float in fac1
@@ -104,7 +104,6 @@ asmsub  FREADUS32  () clobbers(A,X,Y)  {
 
 asmsub  FREADS24AXY  (ubyte lo @ A, ubyte mid @ X, ubyte hi @ Y) clobbers(A,X,Y)  {
 	; ---- fac1 = signed int24 (A/X/Y contain lo/mid/hi bytes)
-	;      note: there is no FREADU24AXY (unsigned), use FREADUS32 instead.
 	%asm {{
 		sty  $62
 		stx  $63
@@ -119,6 +118,39 @@ asmsub  FREADS24AXY  (ubyte lo @ A, ubyte mid @ X, ubyte hi @ Y) clobbers(A,X,Y)
 	}}
 }
 
+asmsub FREADU24AXY(ubyte lo @ A, ubyte mid @ X, ubyte hi @ Y) clobbers(A, X, Y) -> float @FAC1 {
+    ; ---- fac1 = unsigned int24 (A/X/Y contain lo/mid/hi bytes)
+        %asm{{
+                 FAC = $61
+                 sty FAC+1
+                 stx FAC+2
+                 sta FAC+3
+
+                 cpy #$00
+                 bne +
+                 cpx #$00
+                 bne +
+                 cmp #$00
+                 beq ++
+
+              +  ldx #$98
+                 bit FAC+1
+                 bmi +
+
+              -  dex
+                 asl FAC+3
+                 rol FAC+2
+                 rol FAC+1
+                 bpl -
+
+              +  stx FAC
+                 lda #$00
+                 sta FAC+4
+                 sta FAC+5
+                 rts
+       }}
+}
+
 asmsub  GIVUAYFAY  (uword value @ AY) clobbers(A,X,Y)  {
 	; ---- unsigned 16 bit word in A/Y (lo/hi) to fac1
 	%asm {{
@@ -177,6 +209,13 @@ asmsub normalize(float value @FAC1) -> float @ FAC1 {
     }}
 }
 
+; get the jiffy clock as a float
+asmsub time() -> float @ FAC1 {
+    %asm {{
+        jsr cbm.RDTIM
+        jmp floats.FREADU24AXY
+    }}
+}
 
 %asminclude "library:c64/floats.asm"
 %asminclude "library:c64/floats_funcs.asm"

compiler/res/prog8lib/c64/floats_funcs.asm

@@ -6,108 +6,6 @@ func_sign_f_into_A	.proc
 		jmp  SIGN
 		.pend
 
-func_swap_f	.proc
-		; -- swap floats pointed to by SCRATCH_ZPWORD1, SCRATCH_ZPWORD2
-		ldy  #4
--               lda  (P8ZP_SCRATCH_W1),y
-		pha
-		lda  (P8ZP_SCRATCH_W2),y
-		sta  (P8ZP_SCRATCH_W1),y
-		pla
-		sta  (P8ZP_SCRATCH_W2),y
-		dey
-		bpl  -
-		rts
-		.pend
-
-func_reverse_f	.proc
-		; --- reverse an array of floats (array in P8ZP_SCRATCH_W1, num elements in A)
-_left_index = P8ZP_SCRATCH_W2
-_right_index = P8ZP_SCRATCH_W2+1
-_loop_count = P8ZP_SCRATCH_REG
-		pha
-		jsr  a_times_5
-		sec
-		sbc  #5
-		sta  _right_index
-		lda  #0
-		sta  _left_index
-		pla
-		lsr  a
-		sta  _loop_count
-_loop		; push the left indexed float on the stack
-		ldy  _left_index
-		lda  (P8ZP_SCRATCH_W1),y
-		pha
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		pha
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		pha
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		pha
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		pha
-		; copy right index float to left index float
-		ldy  _right_index
-		lda  (P8ZP_SCRATCH_W1),y
-		ldy  _left_index
-		sta  (P8ZP_SCRATCH_W1),y
-		inc  _left_index
-		inc  _right_index
-		ldy  _right_index
-		lda  (P8ZP_SCRATCH_W1),y
-		ldy  _left_index
-		sta  (P8ZP_SCRATCH_W1),y
-		inc  _left_index
-		inc  _right_index
-		ldy  _right_index
-		lda  (P8ZP_SCRATCH_W1),y
-		ldy  _left_index
-		sta  (P8ZP_SCRATCH_W1),y
-		inc  _left_index
-		inc  _right_index
-		ldy  _right_index
-		lda  (P8ZP_SCRATCH_W1),y
-		ldy  _left_index
-		sta  (P8ZP_SCRATCH_W1),y
-		inc  _left_index
-		inc  _right_index
-		ldy  _right_index
-		lda  (P8ZP_SCRATCH_W1),y
-		ldy  _left_index
-		sta  (P8ZP_SCRATCH_W1),y
-		; pop the float off the stack into the right index float
-		ldy  _right_index
-		pla
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		pla
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		pla
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		pla
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		pla
-		sta  (P8ZP_SCRATCH_W1),y
-		inc  _left_index
-		lda  _right_index
-		sec
-		sbc  #9
-		sta  _right_index
-		dec  _loop_count
-		bne  _loop
-		rts
-
-		.pend
-
-
 
 a_times_5	.proc
 		sta  P8ZP_SCRATCH_B1
@@ -118,26 +16,6 @@ a_times_5	.proc
 		rts
 		.pend
 
-func_any_f_into_A	.proc
-		jsr  a_times_5
-		jmp  prog8_lib.func_any_b_into_A
-		.pend
-
-func_all_f_into_A	.proc
-		jsr  a_times_5
-		jmp  prog8_lib.func_all_b_into_A
-		.pend
-
-func_any_f_stack	.proc
-		jsr  a_times_5
-		jmp  prog8_lib.func_any_b_stack
-		.pend
-
-func_all_f_stack	.proc
-		jsr  a_times_5
-		jmp  prog8_lib.func_all_b_stack
-		.pend
-
 func_abs_f_into_FAC1    .proc
         jsr  MOVFM
         jmp  ABS

compiler/res/prog8lib/c64/graphics.p8

@@ -288,17 +288,30 @@ hline_zero2
 
     sub disc(uword xcenter, ubyte ycenter, ubyte radius) {
         ; Warning: NO BOUNDS CHECKS. Make sure circle fits in the screen.
-        ; Midpoint algorithm, filled
+        ; Midpoint algorithm, filled.
         if radius==0
             return
         ubyte @zp yy = 0
         word decisionOver2 = (1 as word)-radius
+        ubyte last_y3 = ycenter+radius
+        ubyte last_y4 = ycenter-radius
+        ubyte new_y3, new_y4
 
         while radius>=yy {
             horizontal_line(xcenter-radius, ycenter+yy, radius*2+1)
             horizontal_line(xcenter-radius, ycenter-yy, radius*2+1)
-            horizontal_line(xcenter-yy, ycenter+radius, yy*2+1)
-            horizontal_line(xcenter-yy, ycenter-radius, yy*2+1)
+
+            new_y3 = ycenter+radius
+            if new_y3 != last_y3 {
+                horizontal_line(xcenter-yy, last_y3, yy*2+1)
+                last_y3 = new_y3
+            }
+            new_y4 = ycenter-radius
+            if new_y4 != last_y4 {
+                horizontal_line(xcenter-yy, last_y4, yy*2+1)
+                last_y4 = new_y4
+            }
+
             yy++
             if decisionOver2>=0 {
                 radius--
@@ -307,6 +320,10 @@ hline_zero2
             decisionOver2 += yy*$0002
             decisionOver2++
         }
+        ; draw the final two spans
+        yy--
+        horizontal_line(xcenter-yy, last_y3, yy*2+1)
+        horizontal_line(xcenter-yy, last_y4, yy*2+1)
     }
 
 

compiler/res/prog8lib/conv.p8

@@ -517,16 +517,16 @@ _stop
 
 asmsub internal_ubyte2decimal(ubyte value @A) -> ubyte @Y, ubyte @X, ubyte @A {
     %asm {{
-        ldy #'0'-1
-        ldx #'9'+1
+        ldy  #'0'-1
+        ldx  #'9'+1
         sec
 -       iny
-        sbc #100
-        bcs -
+        sbc  #100
+        bcs  -
 -       dex
-        adc #10
-        bmi -
-        adc #'0'-1
+        adc  #10
+        bmi  -
+        adc  #'0'-1
         rts
     }}
 }

compiler/res/prog8lib/cx16/diskio.p8

@@ -379,6 +379,7 @@ m_in_buffer     sta  $ffff
     ; optimized for Commander X16 to use MACPTR block read kernal call
     sub f_read_all(uword bufferpointer) -> uword {
         ; -- read the full contents of the file, returns number of bytes read.
+        ;    It is assumed the file size is less than 64 K.
         ;    NOTE: cannot be used to load into VRAM.  Use vload() or call cx16.MACPTR() yourself with the vera data register as address.
         if not iteration_in_progress
             return 0
@@ -393,12 +394,12 @@ m_in_buffer     sta  $ffff
         return total_read
     }
 
-    asmsub f_readline(uword bufptr @AY) clobbers(X) -> ubyte @Y {
+    asmsub f_readline(uword bufptr @AY) clobbers(X) -> ubyte @Y, ubyte @A {
         ; Routine to read text lines from a text file. Lines must be less than 255 characters.
         ; Reads characters from the input file UNTIL a newline or return character (or EOF).
         ; The line read will be 0-terminated in the buffer (and not contain the end of line character).
         ; The length of the line is returned in Y. Note that an empty line is okay and is length 0!
-        ; I/O error status should be checked by the caller itself via READST() routine.
+        ; The I/O error status byte is returned in A.
         %asm {{
             sta  P8ZP_SCRATCH_W1
             sty  P8ZP_SCRATCH_W1+1
@@ -415,7 +416,8 @@ _loop       jsr  cbm.CHRIN
 _line_end   dey     ; get rid of the trailing end-of-line char
             lda  #0
             sta  (P8ZP_SCRATCH_W1),y
-_end        rts
+_end        jsr  cbm.READST
+            rts
         }}
     }
 
@@ -467,6 +469,7 @@ _end        rts
 
     sub f_write(uword bufferpointer, uword num_bytes) -> bool {
         ; -- write the given number of bytes to the currently open file
+        ;    you can call this multiple times to append more data
         if num_bytes!=0 {
             reset_write_channel()
             do {
@@ -863,6 +866,7 @@ io_error:
 
     sub f_seek(uword pos_hiword, uword pos_loword) {
         ; -- seek in the reading file opened with f_open, to the given 32-bits position
+        ;    Note: this will not work if you have already read the last byte of the file! Then you must close and reopen the file first.
         ubyte[6] command = ['p',0,0,0,0,0]
         command[1] = READ_IO_CHANNEL       ; f_open uses this secondary address
         command[2] = lsb(pos_loword)
@@ -931,15 +935,14 @@ io_error:
             return
 
         cx16.r0 = cx16.r1 = cx16.r2 = cx16.r3 = 0
-
-        sub read4hex() -> uword {
-            str hex = "0000"
-            hex[0] = cbm.CHRIN()
-            hex[1] = cbm.CHRIN()
-            hex[2] = cbm.CHRIN()
-            hex[3] = cbm.CHRIN()
-            return conv.hex2uword(hex)
-        }
     }
 
+    sub read4hex() -> uword {
+        str hex = "0000"
+        hex[0] = cbm.CHRIN()
+        hex[1] = cbm.CHRIN()
+        hex[2] = cbm.CHRIN()
+        hex[3] = cbm.CHRIN()
+        return conv.hex2uword(hex)
+    }
 }

compiler/res/prog8lib/cx16/emudbg.p8

@@ -62,4 +62,25 @@ emudbg {
         if is_emulator()
             EMU_DBGOUT2 = value
     }
+
+    sub reset_cpu_cycles() {
+        if is_emulator()
+            EMU_CPUCLK_L = 0
+    }
+
+    asmsub cpu_cycles() -> uword @R0, uword @R1 {
+        ; -- returns the 32 bits cpu clock counter in R1:R0,
+        ;    returns lower 16 bits in R0, upper 16 bits in R1.
+        %asm {{
+            lda  p8v_EMU_CPUCLK_L
+            sta  cx16.r0L
+            lda  p8v_EMU_CPUCLK_M
+            sta  cx16.r0H
+            lda  p8v_EMU_CPUCLK_H
+            sta  cx16.r1L
+            lda  p8v_EMU_CPUCLK_U
+            sta  cx16.r1H
+            rts
+        }}
+    }
 }

compiler/res/prog8lib/cx16/floats.p8

@@ -90,6 +90,38 @@ asmsub  FREADSA  (byte value @A) clobbers(A,X,Y) {
     }}
 }
 
+asmsub FREADU24AXY(ubyte lo @ A, ubyte mid @ X, ubyte hi @ Y) clobbers(A, X, Y) -> float @FAC1 {
+    ; ---- fac1 = unsigned int24 (A/X/Y contain lo/mid/hi bytes)
+        %asm{{
+                 FAC = $C3
+                 sty FAC+1
+                 stx FAC+2
+                 sta FAC+3
+
+                 cpy #$00
+                 bne +
+                 cpx #$00
+                 bne +
+                 cmp #$00
+                 beq ++
+
+              +  ldx #$98
+                 bit FAC+1
+                 bmi +
+
+              -  dex
+                 asl FAC+3
+                 rol FAC+2
+                 rol FAC+1
+                 bpl -
+
+               + stx FAC
+                 stz FAC+4
+                 stz FAC+5
+                 rts
+       }}
+}
+
 asmsub  GIVUAYFAY  (uword value @ AY) clobbers(A,X,Y)  {
 	; ---- unsigned 16 bit word in A/Y (lo/hi) to fac1
 	%asm {{
@@ -157,7 +189,6 @@ _msg    .text 13,"?rom 47+ required for val1",13,0
     }}
 }
 
-
 &uword AYINT_facmo = $c6      ; $c6/$c7 contain result of AYINT   (See "basic.sym" kernal symbol file)
 
 sub rnd() -> float {
@@ -173,6 +204,14 @@ asmsub normalize(float value @FAC1) -> float @ FAC1 {
     }}
 }
 
+; get the jiffy clock as a float
+asmsub time() -> float @ FAC1 {
+    %asm {{
+        jsr cbm.RDTIM
+        jmp floats.FREADU24AXY
+    }}
+}
+
 %asminclude "library:c64/floats.asm"
 %asminclude "library:c64/floats_funcs.asm"
 

compiler/res/prog8lib/cx16/gfx2.p8

@@ -13,6 +13,10 @@
 ; NOTE: For sake of speed, NO BOUNDS CHECKING is performed in most routines!
 ;       You'll have to make sure yourself that you're not writing outside of bitmap boundaries!
 ;
+; NOTE: the bitmap screen data is positioned in vram at $0:0000
+;
+; NOTE: In the future, these routines might be split out to separate modules, 1 for each screen mode,
+;       so they can be optimized a lot better.  There's already a "gfx_lores" module with a few routines for lores 256C mode.
 ;
 ; SCREEN MODE LIST:
 ;   mode 0 = reset back to default text mode
@@ -254,6 +258,7 @@ gfx2 {
                 ;; color <<= gfx2.plot.shift4c[lsb(xx) & 3]
                 cx16.r2L = lsb(xx) & 3
                 when color & 3 {
+                    0 -> color = 0
                     1 -> color = gfx2.plot.shiftedleft_4c_1[cx16.r2L]
                     2 -> color = gfx2.plot.shiftedleft_4c_2[cx16.r2L]
                     3 -> color = gfx2.plot.shiftedleft_4c_3[cx16.r2L]
@@ -272,10 +277,10 @@ gfx2 {
 
         sub set_both_strides(ubyte stride) {
             stride <<= 4
-            cx16.VERA_CTRL = 0
-            cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & %00000111 | stride
             cx16.VERA_CTRL = 1
             cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & %00000111 | stride
+            cx16.VERA_CTRL = 0
+            cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & %00000111 | stride
         }
 
     }
@@ -283,6 +288,9 @@ gfx2 {
     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.
+
+        ; NOTE: if you are using lores 256c (mode 1), the line routine in the fastline module can be used and is MUCH faster than this one.
+
         if y1>y2 {
             ; make sure dy is always positive to have only 4 instead of 8 special cases
             cx16.r0 = x1
@@ -477,12 +485,25 @@ gfx2 {
             return
         ubyte @zp yy = 0
         word @zp decisionOver2 = (1 as word)-radius
+        uword last_y3 = ycenter+radius
+        uword last_y4 = ycenter-radius
+        uword new_y3, new_y4
 
         while radius>=yy {
             horizontal_line(xcenter-radius, ycenter+yy, radius*$0002+1, color)
             horizontal_line(xcenter-radius, ycenter-yy, radius*$0002+1, color)
-            horizontal_line(xcenter-yy, ycenter+radius, yy*$0002+1, color)
-            horizontal_line(xcenter-yy, ycenter-radius, yy*$0002+1, color)
+
+            new_y3 = ycenter+radius
+            if new_y3 != last_y3 {
+                horizontal_line(xcenter-yy, last_y3, yy*$0002+1, color)
+                last_y3 = new_y3
+            }
+            new_y4 = ycenter-radius
+            if new_y4 != last_y4 {
+                horizontal_line(xcenter-yy, last_y4, yy*$0002+1, color)
+                last_y4 = new_y4
+            }
+
             yy++
             if decisionOver2>=0 {
                 radius--
@@ -491,6 +512,11 @@ gfx2 {
             decisionOver2 += yy*$0002
             decisionOver2++
         }
+
+        ; draw the final two spans
+        yy--
+        horizontal_line(xcenter-yy, last_y3, yy*$0002+1, color)
+        horizontal_line(xcenter-yy, last_y4, yy*$0002+1, color)
     }
 
     sub safe_disc(uword @zp xcenter, uword @zp ycenter, ubyte @zp radius, ubyte color) {
@@ -500,12 +526,23 @@ gfx2 {
             return
         ubyte @zp yy = 0
         word @zp decisionOver2 = (1 as word)-radius
+        uword last_y3 = ycenter+radius
+        uword last_y4 = ycenter-radius
+        uword new_y3, new_y4
 
         while radius>=yy {
             safe_horizontal_line(xcenter-radius, ycenter+yy, radius*$0002+1, color)
             safe_horizontal_line(xcenter-radius, ycenter-yy, radius*$0002+1, color)
-            safe_horizontal_line(xcenter-yy, ycenter+radius, yy*$0002+1, color)
-            safe_horizontal_line(xcenter-yy, ycenter-radius, yy*$0002+1, color)
+            new_y3 = ycenter+radius
+            if new_y3 != last_y3 {
+                safe_horizontal_line(xcenter-yy, last_y3, yy*$0002+1, color)
+                last_y3 = new_y3
+            }
+            new_y4 = ycenter-radius
+            if new_y4 != last_y4 {
+                safe_horizontal_line(xcenter-yy, last_y4, yy*$0002+1, color)
+                last_y4 = new_y4
+            }
             yy++
             if decisionOver2>=0 {
                 radius--
@@ -514,6 +551,10 @@ gfx2 {
             decisionOver2 += yy*$0002
             decisionOver2++
         }
+        ; draw the final two spans
+        yy--
+        safe_horizontal_line(xcenter-yy, last_y3, yy*$0002+1, color)
+        safe_horizontal_line(xcenter-yy, last_y4, yy*$0002+1, color)
     }
 
     sub plot(uword @zp xx, uword @zp yy, ubyte @zp color) {
@@ -547,6 +588,7 @@ gfx2 {
                 ; color &= 3
                 ; color <<= shift4c[cx16.r2L]
                 when color & 3 {
+                    0 -> color = 0
                     1 -> color = shiftedleft_4c_1[cx16.r2L]
                     2 -> color = shiftedleft_4c_2[cx16.r2L]
                     3 -> color = shiftedleft_4c_3[cx16.r2L]
@@ -626,10 +668,18 @@ gfx2 {
     }
 
     sub fill(uword x, uword y, ubyte new_color) {
+        ; reuse a few virtual registers in ZP for variables
+        &ubyte fillm = &cx16.r7L
+        &ubyte seedm = &cx16.r8L
+        &ubyte cmask = &cx16.r8H
+        &ubyte vub   = &cx16.r13L
+        &ubyte nvub  = &cx16.r13H
+        ubyte[4] amask = [$c0,$30,$0c,$03] ; array of cmask bytes
+
         ; Non-recursive scanline flood fill.
         ; based loosely on code found here https://www.codeproject.com/Articles/6017/QuickFill-An-efficient-flood-fill-algorithm
         ; with the fixes applied to the seedfill_4 routine as mentioned in the comments.
-        const ubyte MAXDEPTH = 64
+        const ubyte MAXDEPTH = 100
         word @zp xx = x as word
         word @zp yy = y as word
         word[MAXDEPTH] @split @shared stack_xl
@@ -641,6 +691,7 @@ gfx2 {
         word x2
         byte dy
         cx16.r10L = new_color
+
         sub push_stack(word sxl, word sxr, word sy, byte sdy) {
             if cx16.r12L==MAXDEPTH
                 return
@@ -704,39 +755,27 @@ gfx2 {
             return
         if xx<0 or xx>width-1 or yy<0 or yy>height-1
             return
+        if gfx2.active_mode == 2 set_color_masks()
         push_stack(xx, xx, yy, 1)
         push_stack(xx, xx, yy + 1, -1)
         word left = 0
         while cx16.r12L!=0 {
             pop_stack()
             xx = x1
-            ; possible speed optimization: if mode==1 (256c) use vera autodecrement instead of pget(), but code bloat not worth it?
-            while xx >= 0 {
-                if pget(xx as uword, yy as uword) != cx16.r11L
-                    break
-                xx--
+            when active_mode {
+                1 -> if fill_scanline_left_8bpp() goto skip
+                2 -> if fill_scanline_left_2bpp() goto skip
             }
-            if x1!=xx
-                horizontal_line(xx as uword+1, yy as uword, x1-xx as uword, cx16.r10L)
-            else
-                goto skip
-
             left = xx + 1
             if left < x1
                 push_stack(left, x1 - 1, yy, -dy)
             xx = x1 + 1
 
             do {
-                cx16.r9s = xx
-                ; possible speed optimization: if mode==1 (256c) use vera autoincrement instead of pget(), but code bloat not worth it?
-                while xx <= width-1 {
-                    if pget(xx as uword, yy as uword) != cx16.r11L
-                        break
-                    xx++
+                when active_mode {
+                    1 -> fill_scanline_right_8bpp()
+                    2 -> fill_scanline_right_2bpp()
                 }
-                if cx16.r9s!=xx
-                    horizontal_line(cx16.r9, yy as uword, xx-cx16.r9s as uword, cx16.r10L)
-
                 push_stack(left, xx - 1, yy, dy)
                 if xx > x2 + 1
                     push_stack(x2 + 1, xx - 1, yy, -dy)
@@ -750,6 +789,133 @@ skip:
                 left = xx
             } until xx>x2
         }
+
+        sub set_vera_address() {
+            ; set both data0 and data1 addresses (expects H in R1L, M/L in R0)
+            cx16.VERA_CTRL = 0
+            cx16.VERA_ADDR_H = cx16.r1L
+            cx16.VERA_ADDR = cx16.r0
+            cx16.VERA_CTRL = 1
+            cx16.VERA_ADDR_H = cx16.r1L
+            cx16.VERA_ADDR = cx16.r0
+            cx16.VERA_CTRL = 0
+        }
+
+        sub fill_scanline_left_8bpp() -> bool {
+            void addr_mul_24_for_lores_256c(yy as uword, xx as uword)      ; 24 bits result is in r0 and r1L (highest byte)
+            cx16.r1L |= %00011000     ; auto decrement enabled
+            set_vera_address()
+            cx16.r9s = xx
+            while xx >= 0 {
+                if cx16.VERA_DATA0 != cx16.r11L
+                    break
+                cx16.VERA_DATA1 = cx16.r10L
+                xx--
+            }
+            return xx==cx16.r9s
+        }
+
+        sub fill_scanline_right_8bpp() {
+            void addr_mul_24_for_lores_256c(yy as uword, xx as uword)      ; 24 bits result is in r0 and r1L (highest byte)
+            cx16.r1L |= %00010000    ; auto increment enabled
+            set_vera_address()
+            while xx <= width-1 {
+                if cx16.VERA_DATA0 != cx16.r11L
+                    break
+                cx16.VERA_DATA1 = cx16.r10L
+                xx++
+            }
+        }
+
+        sub fill_scanline_left_2bpp() -> bool {
+            uword vx = xx as uword
+            void gfx2.addr_mul_24_for_highres_4c(yy as uword,vx)
+            cx16.r1L |= %0001_1000  ; auto decrement
+            set_vera_address()
+            cmask = amask[lsb(vx) & 3]  ; set the color mask for the first color pel
+
+            repeat {
+                vub = cx16.VERA_DATA0  ; read the VERA color data for 4 pixels
+                if cmask == $03 {     ; only speed fill from far right
+                    ; speed fill
+                    if vub == seedm { ; all four colors match the seed
+                        nvub = fillm  ; replace all four colors at once
+                        xx -= 4
+                        goto set_byte  ; go on
+                    }
+                }
+
+                ; replace one color at a time
+                nvub = vub
+                while cmask != 0 {
+                    if vub & cmask == seedm & cmask {
+                        nvub &= ~cmask
+                        nvub |= cmask & fillm
+;                        %asm{{
+;                            lda  p8v_cmask
+;                            trb  p8v_nvub
+;                            and  p8v_fillm
+;                            tsb  p8v_nvub
+;                        }}
+                        xx--
+                        cmask <<= 2
+                    } else { ; not the seed color, finish here
+                        cx16.VERA_DATA1 = nvub
+                        return vx == xx
+                    }
+                }
+set_byte:
+                cx16.VERA_DATA1 = nvub
+                if xx <= 0 break
+                cmask = $03
+            }
+            return vx == xx
+        }
+
+        sub fill_scanline_right_2bpp() {
+            void gfx2.addr_mul_24_for_highres_4c(yy as uword,xx as uword)
+            cx16.r1L |= %00010000    ; auto increment
+            set_vera_address()
+            cmask = amask[lsb(xx) & 3]  ; set the color mask for the first color pel
+
+            repeat {
+                vub = cx16.VERA_DATA0  ; read the VERA color data for 4 pixels
+                ; speed fill
+                if vub == seedm { ; all four colors match the seed
+                    nvub = fillm  ; replace all four colors at once
+                    xx += 4
+                    goto set_byte   ; go on
+                }
+                ; replace one color at a time
+                nvub = vub
+                while cmask != 0 {
+                    if vub & cmask == seedm & cmask {
+                        nvub &= ~cmask
+                        nvub |= cmask & fillm
+;                        %asm{{
+;                            lda  p8v_cmask
+;                            trb  p8v_nvub
+;                            and  p8v_fillm
+;                            tsb  p8v_nvub
+;                        }}
+                        xx++
+                        cmask >>= 2
+                    } else { ; not the seed color finish here
+                        cx16.VERA_DATA1 = nvub
+                        return
+                    }
+                }
+set_byte:
+                cx16.VERA_DATA1 = nvub
+                if xx >= gfx2.width-1 break
+                cmask = $C0
+            }
+        }
+
+        sub set_color_masks() {
+            seedm = cx16.r11L | (cx16.r11L<<2) | (cx16.r11L<<4) | (cx16.r11L<<6) ; seed mask
+            fillm = cx16.r10L | (cx16.r10L<<2) | (cx16.r10L<<4) | (cx16.r10L<<6) ; fill mask
+        }
     }
 
     sub position(uword @zp xx, uword yy) {

compiler/res/prog8lib/cx16/gfx_lores.p8

@@ -0,0 +1,235 @@
+; optimized graphics routines for just the single screen mode: lores 320*240, 256c  (8bpp)
+; bitmap image needs to start at VRAM addres $00000.
+; This is compatible with the CX16's screen mode 128.  (void cx16.set_screen_mode(128))
+
+
+gfx_lores {
+
+    %option ignore_unused
+
+    sub set_screen_mode() {
+        cx16.VERA_CTRL=0
+        cx16.VERA_DC_VIDEO = (cx16.VERA_DC_VIDEO & %11001111) | %00100000      ; enable only layer 1
+        cx16.VERA_DC_HSCALE = 64
+        cx16.VERA_DC_VSCALE = 64
+        cx16.VERA_L1_CONFIG = %00000111
+        cx16.VERA_L1_MAPBASE = 0
+        cx16.VERA_L1_TILEBASE = 0
+        clear_screen(0)
+    }
+
+    sub clear_screen(ubyte color) {
+        cx16.VERA_CTRL=0
+        cx16.VERA_ADDR=0
+        cx16.VERA_ADDR_H = 1<<4    ; 1 pixel auto increment
+        repeat 240
+            cs_innerloop320(color)
+        cx16.VERA_ADDR=0
+        cx16.VERA_ADDR_H = 0
+    }
+
+    sub line(uword x1, ubyte y1, uword x2, ubyte y2, ubyte color) {
+        ; Bresenham algorithm.
+        ; This code special-cases various quadrant loops to allow simple ++ and -- operations.
+        ; NOTE:  this is about twice as fast as the kernal routine GRAPH_draw_line, and ~3-4 times as fast as gfx2.line()
+        ;        it trades memory for speed (uses inline plot routine and multiplication lookup tables)
+        ;
+        ; NOTE:  is currently still a regular 6502 routine, could likely be made much faster with the VeraFX line helper.
+
+        cx16.r3L = y2    ; ensure zeropage
+        cx16.r1L = y1    ; ensure zeropage
+
+        if cx16.r1L > cx16.r3L {
+            ; make sure dy is always positive to have only 4 instead of 8 special cases
+            cx16.r0 = x1
+            x1 = x2
+            x2 = cx16.r0
+            cx16.r0L = cx16.r1L
+            cx16.r1L = cx16.r3L
+            cx16.r3L = cx16.r0L
+        }
+        word @zp dx = x2 as word
+        word @zp dy = cx16.r3L
+        dx -= x1
+        dy -= cx16.r1L
+
+        if dx==0 {
+            vertical_line(x1, cx16.r1L, lsb(dy)+1, color)
+            return
+        }
+        if dy==0 {
+            if x1>x2
+                x1=x2
+            horizontal_line(x1, cx16.r1L, abs(dx) as uword +1, color)
+            return
+        }
+
+        word @zp d = 0
+        bool positive_ix = true
+        if dx < 0 {
+            dx = -dx
+            positive_ix = false
+        }
+        word @zp dx2 = dx*2
+        word @zp dy2 = dy*2
+
+        cx16.r0  = x1    ; ensure zeropage
+        cx16.r2  = x2    ; ensure zeropage
+
+        cx16.VERA_CTRL = 0
+        if dx >= dy {
+            if positive_ix {
+                repeat {
+                    plot()
+                    if cx16.r0==cx16.r2
+                        return
+                    cx16.r0++
+                    d += dy2
+                    if d > dx {
+                        cx16.r1L++
+                        d -= dx2
+                    }
+                }
+            } else {
+                repeat {
+                    plot()
+                    if cx16.r0==cx16.r2
+                        return
+                    cx16.r0--
+                    d += dy2
+                    if d > dx {
+                        cx16.r1L++
+                        d -= dx2
+                    }
+                }
+            }
+        }
+        else {
+            if positive_ix {
+                repeat {
+                    plot()
+                    if cx16.r1L == cx16.r3L
+                        return
+                    cx16.r1L++
+                    d += dx2
+                    if d > dy {
+                        cx16.r0++
+                        d -= dy2
+                    }
+                }
+            } else {
+                repeat {
+                    plot()
+                    if cx16.r1L == cx16.r3L
+                        return
+                    cx16.r1L++
+                    d += dx2
+                    if d > dy {
+                        cx16.r0--
+                        d -= dy2
+                    }
+                }
+            }
+        }
+
+        asmsub plot() {
+            ; x in r0,  y in r1,   color.
+            %asm {{
+                ldy  cx16.r1L
+                clc
+                lda  times320_lo,y
+                adc  cx16.r0L
+                sta  cx16.VERA_ADDR_L
+                lda  times320_mid,y
+                adc  cx16.r0H
+                sta  cx16.VERA_ADDR_M
+                lda  #0
+                adc  times320_hi,y
+                sta  cx16.VERA_ADDR_H
+                lda  p8v_color
+                sta  cx16.VERA_DATA0
+                rts
+            }}
+        }
+
+        %asm {{
+
+; multiplication by 320 lookup table
+times320 := 320*range(240)
+
+times320_lo     .byte <times320
+times320_mid    .byte >times320
+times320_hi     .byte `times320
+
+            }}
+    }
+
+    sub horizontal_line(uword xx, ubyte yy, uword length, ubyte color) {
+        if length==0
+            return
+        vera_setaddr(xx, yy)
+        ; set vera auto-increment to 1 pixel
+        cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & %00000111 | (1<<4)
+
+        %asm {{
+            lda  p8v_color
+            ldx  p8v_length+1
+            beq  +
+            ldy  #0
+-           sta  cx16.VERA_DATA0
+            iny
+            bne  -
+            dex
+            bne  -
++           ldy  p8v_length     ; remaining
+            beq  +
+-           sta  cx16.VERA_DATA0
+            dey
+            bne  -
++
+        }}
+    }
+
+    sub vertical_line(uword xx, ubyte yy, ubyte lheight, ubyte color) {
+        vera_setaddr(xx,yy)
+        ; set vera auto-increment to 320 pixel increment (=next line)
+        cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & %00000111 | (14<<4)
+        %asm {{
+            ldy  p8v_lheight
+            beq  +
+            lda  p8v_color
+-           sta  cx16.VERA_DATA0
+            dey
+            bne  -
++
+        }}
+    }
+
+
+    asmsub cs_innerloop320(ubyte color @A) clobbers(Y) {
+        ; using verafx 32 bits writes here would make this faster but it's safer to
+        ; use verafx only explicitly when you know what you're doing.
+        %asm {{
+            ldy  #40
+-           sta  cx16.VERA_DATA0
+            sta  cx16.VERA_DATA0
+            sta  cx16.VERA_DATA0
+            sta  cx16.VERA_DATA0
+            sta  cx16.VERA_DATA0
+            sta  cx16.VERA_DATA0
+            sta  cx16.VERA_DATA0
+            sta  cx16.VERA_DATA0
+            dey
+            bne  -
+            rts
+        }}
+    }
+
+    inline asmsub vera_setaddr(uword xx @R0, ubyte yy @R1) {
+        ; set the correct vera start address (no auto increment yet!)
+        %asm {{
+            jsr  p8s_line.p8s_plot
+        }}
+    }
+}
+

compiler/res/prog8lib/cx16/graphics.p8

@@ -76,83 +76,30 @@ graphics {
 
     sub circle(uword xcenter, ubyte ycenter, ubyte radius) {
         ; Warning: NO BOUNDS CHECKS. Make sure circle fits in the screen.
-
-        ;cx16.r0 = xcenter - radius/2
-        ;cx16.r1 = ycenter - radius/2
-        ;cx16.r2 = radius*2
-        ;cx16.r3 = radius*2
-        ;cx16.GRAPH_draw_oval(false)          ; currently this call is not implemented on cx16, does a BRK
-
-        ; Midpoint algorithm
         if radius==0
             return
-        ubyte @zp xx = radius
-        ubyte @zp yy = 0
-        word @zp decisionOver2 = (1 as word)-xx
-
-        while xx>=yy {
-            cx16.r0 = xcenter + xx
-            cx16.r1 = ycenter + yy
-            FB_cursor_position2()
-            cx16.FB_set_pixel(stroke_color)
-            cx16.r0 = xcenter - xx
-            FB_cursor_position2()
-            cx16.FB_set_pixel(stroke_color)
-            cx16.r0 = xcenter + xx
-            cx16.r1 = ycenter - yy
-            FB_cursor_position2()
-            cx16.FB_set_pixel(stroke_color)
-            cx16.r0 = xcenter - xx
-            FB_cursor_position2()
-            cx16.FB_set_pixel(stroke_color)
-            cx16.r0 = xcenter + yy
-            cx16.r1 = ycenter + xx
-            FB_cursor_position2()
-            cx16.FB_set_pixel(stroke_color)
-            cx16.r0 = xcenter - yy
-            FB_cursor_position2()
-            cx16.FB_set_pixel(stroke_color)
-            cx16.r0 = xcenter + yy
-            cx16.r1 = ycenter - xx
-            FB_cursor_position2()
-            cx16.FB_set_pixel(stroke_color)
-            cx16.r0 = xcenter - yy
-            FB_cursor_position2()
-            cx16.FB_set_pixel(stroke_color)
-
-            yy++
-            if decisionOver2>=0 {
-                xx--
-                decisionOver2 -= xx*$0002
-            }
-            decisionOver2 += yy*$0002
-            decisionOver2++
-        }
+        cx16.GRAPH_draw_oval(xcenter - radius, ycenter - radius, radius*2, radius*2, false)
     }
 
     sub disc(uword xcenter, ubyte ycenter, ubyte radius) {
         ; Warning: NO BOUNDS CHECKS. Make sure circle fits in the screen.
-        ; Midpoint algorithm, filled
-
         if radius==0
             return
-        ubyte @zp yy = 0
-        word decisionOver2 = (1 as word)-radius
+        cx16.GRAPH_draw_oval(xcenter - radius, ycenter - radius, radius*2, radius*2, true)
+    }
 
-        while radius>=yy {
-            horizontal_line(xcenter-radius, ycenter+yy, radius*2+1)
-            horizontal_line(xcenter-radius, ycenter-yy, radius*2+1)
-            horizontal_line(xcenter-yy, ycenter+radius, yy*2+1)
-            horizontal_line(xcenter-yy, ycenter-radius, yy*2+1)
+    sub oval(uword xcenter, ubyte ycenter, uword h_radius, ubyte v_radius) {
+        ; specific for the X16 which has a kernal routine that can draw ovals
+        if h_radius==0 or v_radius==0
+            return
+        cx16.GRAPH_draw_oval(xcenter - h_radius, ycenter - v_radius, h_radius*2, v_radius*2, false)
+    }
 
-            yy++
-            if decisionOver2>=0 {
-                radius--
-                decisionOver2 -= radius*$0002
-            }
-            decisionOver2 += yy*$0002
-            decisionOver2++
-        }
+    sub filled_oval(uword xcenter, ubyte ycenter, uword h_radius, ubyte v_radius) {
+        ; specific for the X16 which has a kernal routine that can draw ovals
+        if h_radius==0 or v_radius==0
+            return
+        cx16.GRAPH_draw_oval(xcenter - h_radius, ycenter - v_radius, h_radius*2, v_radius*2, true)
     }
 
     inline asmsub  plot(uword plotx @R0, uword ploty @R1) clobbers(A, X, Y) {

compiler/res/prog8lib/cx16/monogfx.p8

@@ -32,7 +32,7 @@ monogfx {
         width = 320
         height = 240
         mode = MODE_NORMAL
-        clear_screen(0)
+        clear_screen(false)
     }
 
     sub hires() {
@@ -47,7 +47,7 @@ monogfx {
         width = 640
         height = 480
         mode = MODE_NORMAL
-        clear_screen(0)
+        clear_screen(false)
     }
 
     sub textmode() {
@@ -61,16 +61,16 @@ monogfx {
         mode = dm
     }
 
-    sub clear_screen(ubyte color) {
+    sub clear_screen(bool draw) {
         position(0, 0)
         when width {
             320 -> {
                 repeat 240/2/8
-                    cs_innerloop640(color)
+                    cs_innerloop640(draw)
             }
             640 -> {
                 repeat 480/8
-                    cs_innerloop640(color)
+                    cs_innerloop640(draw)
             }
         }
         position(0, 0)
@@ -90,8 +90,8 @@ monogfx {
     }
 
     sub fillrect(uword xx, uword yy, uword rwidth, uword rheight, bool draw) {
-        ; Draw a filled rectangle of the given size and color.
-        ; To fill the whole screen, use clear_screen(color) instead - it is much faster.
+        ; Draw a filled rectangle of the given size.
+        ; To fill the whole screen, use clear_screen(draw) instead - it is much faster.
         if rwidth==0
             return
         repeat rheight {
@@ -358,10 +358,10 @@ drawmode:               ora  cx16.r15L
 
         sub set_both_strides(ubyte stride) {
             stride <<= 4
-            cx16.VERA_CTRL = 0
-            cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & %00000111 | stride
             cx16.VERA_CTRL = 1
             cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & %00000111 | stride
+            cx16.VERA_CTRL = 0
+            cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & %00000111 | stride
         }
 
     }
@@ -563,12 +563,23 @@ drawmode:               ora  cx16.r15L
             return
         ubyte @zp yy = 0
         word @zp decisionOver2 = (1 as word)-radius
+        uword last_y3 = ycenter+radius
+        uword last_y4 = ycenter-radius
+        uword new_y3, new_y4
 
         while radius>=yy {
             horizontal_line(xcenter-radius, ycenter+yy, radius*$0002+1, draw)
             horizontal_line(xcenter-radius, ycenter-yy, radius*$0002+1, draw)
-            horizontal_line(xcenter-yy, ycenter+radius, yy*$0002+1, draw)
-            horizontal_line(xcenter-yy, ycenter-radius, yy*$0002+1, draw)
+            new_y3 = ycenter+radius
+            if new_y3 != last_y3 {
+                horizontal_line(xcenter-yy, last_y3, yy*$0002+1, draw)
+                last_y3 = new_y3
+            }
+            new_y4 = ycenter-radius
+            if new_y4 != last_y4 {
+                horizontal_line(xcenter-yy, last_y4, yy*$0002+1, draw)
+                last_y4 = new_y4
+            }
             yy++
             if decisionOver2>=0 {
                 radius--
@@ -577,6 +588,10 @@ drawmode:               ora  cx16.r15L
             decisionOver2 += yy*$0002
             decisionOver2++
         }
+        ; draw the final two spans
+        yy--
+        horizontal_line(xcenter-yy, last_y3, yy*$0002+1, draw)
+        horizontal_line(xcenter-yy, last_y4, yy*$0002+1, draw)
     }
 
     sub safe_disc(uword @zp xcenter, uword @zp ycenter, ubyte @zp radius, bool draw) {
@@ -586,12 +601,23 @@ drawmode:               ora  cx16.r15L
             return
         ubyte @zp yy = 0
         word @zp decisionOver2 = (1 as word)-radius
+        uword last_y3 = ycenter+radius
+        uword last_y4 = ycenter-radius
+        uword new_y3, new_y4
 
         while radius>=yy {
             safe_horizontal_line(xcenter-radius, ycenter+yy, radius*$0002+1, draw)
             safe_horizontal_line(xcenter-radius, ycenter-yy, radius*$0002+1, draw)
-            safe_horizontal_line(xcenter-yy, ycenter+radius, yy*$0002+1, draw)
-            safe_horizontal_line(xcenter-yy, ycenter-radius, yy*$0002+1, draw)
+            new_y3 = ycenter+radius
+            if new_y3 != last_y3 {
+                safe_horizontal_line(xcenter-yy, last_y3, yy*$0002+1, draw)
+                last_y3 = new_y3
+            }
+            new_y4 = ycenter-radius
+            if new_y4 != last_y4 {
+                safe_horizontal_line(xcenter-yy, last_y4, yy*$0002+1, draw)
+                last_y4 = new_y4
+            }
             yy++
             if decisionOver2>=0 {
                 radius--
@@ -600,6 +626,10 @@ drawmode:               ora  cx16.r15L
             decisionOver2 += yy*$0002
             decisionOver2++
         }
+        ; draw the final two spans
+        yy--
+        safe_horizontal_line(xcenter-yy, last_y3, yy*$0002+1, draw)
+        safe_horizontal_line(xcenter-yy, last_y4, yy*$0002+1, draw)
     }
 
     sub plot(uword @zp xx, uword @zp yy, bool @zp draw) {
@@ -708,7 +738,7 @@ invert:
         ; Non-recursive scanline flood fill.
         ; based loosely on code found here https://www.codeproject.com/Articles/6017/QuickFill-An-efficient-flood-fill-algorithm
         ; with the fixes applied to the seedfill_4 routine as mentioned in the comments.
-        const ubyte MAXDEPTH = 64
+        const ubyte MAXDEPTH = 100
         word @zp xx = x as word
         word @zp yy = y as word
         word[MAXDEPTH] @split @shared stack_xl
@@ -957,20 +987,20 @@ cdraw_mod2              ora  cx16.VERA_DATA1
         sub set_autoincrs() {
             ; set autoincrements to go to next pixel row (40 or 80 increment)
             if width==320 {
-                cx16.VERA_CTRL = 0
-                cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & $0f | (11<<4)
                 cx16.VERA_CTRL = 1
                 cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & $0f | (11<<4)
+                cx16.VERA_CTRL = 0
+                cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & $0f | (11<<4)
             } else {
-                cx16.VERA_CTRL = 0
-                cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & $0f | (12<<4)
                 cx16.VERA_CTRL = 1
                 cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & $0f | (12<<4)
+                cx16.VERA_CTRL = 0
+                cx16.VERA_ADDR_H = cx16.VERA_ADDR_H & $0f | (12<<4)
             }
         }
     }
 
-    asmsub cs_innerloop640(ubyte color @A) clobbers(Y) {
+    asmsub cs_innerloop640(bool draw @A) clobbers(Y) {
         ; using verafx 32 bits writes here would make this faster but it's safer to
         ; use verafx only explicitly when you know what you're doing.
         %asm {{

compiler/res/prog8lib/cx16/sprites.p8

@@ -23,7 +23,7 @@ sprites {
              ubyte databank, uword dataaddr,
              ubyte width_flag, ubyte height_flag,
              ubyte colors_flag, ubyte palette_offset) {
-        hide(spritenum)
+        pos(spritenum, -64, -64)                    ; move sprite off-screen initially
         cx16.VERA_DC_VIDEO |= %01000000             ; enable sprites globally
         dataaddr >>= 5
         dataaddr |= (databank as uword)<<11
@@ -42,9 +42,15 @@ sprites {
         cx16.vpoke_mask(1, sprite_reg+1, %11110000, msb(addr))    ; address 16:13
     }
 
-    sub get_data_ptr(ubyte spritenum) {
+    inline asmsub get_data_ptr(ubyte spritenum @A) -> ubyte @R1, uword @R0 {
         ; -- returns the VRAM address where the sprite's bitmap data is stored
         ;    R1 (byte) = the vera bank (0 or 1), R0 (word) = the address.
+        %asm {{
+            jsr  p8b_sprites.p8s_get_data_ptr_internal
+        }}
+    }
+
+    sub get_data_ptr_internal(ubyte spritenum) {
         sprite_reg = VERA_SPRITEREGS + spritenum*$0008
         cx16.r0L = cx16.vpeek(1, sprite_reg)
         cx16.r0H = cx16.vpeek(1, sprite_reg+1)
@@ -132,7 +138,15 @@ sprites {
     }
 
     sub hide(ubyte spritenum) {
-        pos(spritenum, -64, -64)
+        cx16.vpoke_and(1, VERA_SPRITEREGS + 6 + spritenum*$0008, %11110011)
+    }
+
+    sub show(ubyte spritenum) {
+        cx16.vpoke_or(1, VERA_SPRITEREGS + 6 + spritenum*$0008, %00001100)
+    }
+
+    sub zdepth(ubyte spritenum, ubyte depth) {
+        cx16.vpoke_mask(1, VERA_SPRITEREGS + 6 + spritenum*$0008, %11110011, depth<<2)
     }
 
     sub flipx(ubyte spritenum, bool flipped) {
@@ -176,7 +190,7 @@ sprites {
     }
 
     sub set_mousepointer_image(uword data, bool compressed) {
-        get_data_ptr(0)  ; the mouse cursor is sprite 0
+        get_data_ptr_internal(0)  ; the mouse cursor is sprite 0
         if cx16.r1L==0 and cx16.r0==0
             return    ; mouse cursor not enabled
         ubyte vbank = cx16.r1L

compiler/res/prog8lib/cx16/syslib.p8

@@ -6,6 +6,39 @@
 cbm {
     ; Commodore (CBM) common variables, vectors and kernal routines
 
+; irq, system and hardware vectors  (common across cbm machines):
+    &uword  IERROR      = $0300
+    &uword  IMAIN       = $0302
+    &uword  ICRNCH      = $0304
+    &uword  IQPLOP      = $0306
+    &uword  IGONE       = $0308
+    &uword  IEVAL       = $030a
+    &ubyte  SAREG       = $030c     ; register storage for A for SYS calls
+    &ubyte  SXREG       = $030d     ; register storage for X for SYS calls
+    &ubyte  SYREG       = $030e     ; register storage for Y for SYS calls
+    &ubyte  SPREG       = $030f     ; register storage for P (status register) for SYS calls
+    &uword  USRADD      = $0311     ; vector for the USR() basic command
+    ; $0313 is unused.
+    &uword  CINV        = $0314     ; IRQ vector (in ram)
+    &uword  CBINV       = $0316     ; BRK vector (in ram)
+    &uword  NMINV       = $0318     ; NMI vector (in ram)
+    &uword  IOPEN       = $031a
+    &uword  ICLOSE      = $031c
+    &uword  ICHKIN      = $031e
+    &uword  ICKOUT      = $0320
+    &uword  ICLRCH      = $0322
+    &uword  IBASIN      = $0324
+    &uword  IBSOUT      = $0326
+    &uword  ISTOP       = $0328
+    &uword  IGETIN      = $032a
+    &uword  ICLALL      = $032c
+    ; $032e has a X16 specific function (KEYHDL) so you'll find this as cx16.KEYHDL
+    &uword  ILOAD       = $0330
+    &uword  ISAVE       = $0332
+    &uword  NMI_VEC     = $FFFA     ; 65c02 nmi vector, determined by the kernal if banked in
+    &uword  RESET_VEC   = $FFFC     ; 65c02 reset vector, determined by the kernal if banked in
+    &uword  IRQ_VEC     = $FFFE     ; 65c02 interrupt vector, determined by the kernal if banked in
+
 
 ; STROUT --> use txt.print
 ; CLEARSCR -> use txt.clear_screen
@@ -117,42 +150,14 @@ asmsub kbdbuf_clear() {
 
 cx16 {
 
-; irq, system and hardware vectors:
-    &uword  IERROR      = $0300
-    &uword  IMAIN       = $0302
-    &uword  ICRNCH      = $0304
-    &uword  IQPLOP      = $0306
-    &uword  IGONE       = $0308
-    &uword  IEVAL       = $030a
-    &ubyte  SAREG       = $030c     ; register storage for A for SYS calls
-    &ubyte  SXREG       = $030d     ; register storage for X for SYS calls
-    &ubyte  SYREG       = $030e     ; register storage for Y for SYS calls
-    &ubyte  SPREG       = $030f     ; register storage for P (status register) for SYS calls
-    &uword  USRADD      = $0311     ; vector for the USR() basic command
-    ; $0313 is unused.
-    &uword  CINV        = $0314     ; IRQ vector (in ram)
-    &uword  CBINV       = $0316     ; BRK vector (in ram)
-    &uword  NMINV       = $0318     ; NMI vector (in ram)
-    &uword  IOPEN       = $031a
-    &uword  ICLOSE      = $031c
-    &uword  ICHKIN      = $031e
-    &uword  ICKOUT      = $0320
-    &uword  ICLRCH      = $0322
-    &uword  IBASIN      = $0324
-    &uword  IBSOUT      = $0326
-    &uword  ISTOP       = $0328
-    &uword  IGETIN      = $032a
-    &uword  ICLALL      = $032c
+; cx16 specific vectors and variables
     &uword  KEYHDL      = $032e     ; keyboard scan code handler see examples/cx16/keyboardhandler.p8
-    &uword  ILOAD       = $0330
-    &uword  ISAVE       = $0332
-    &uword  NMI_VEC     = $FFFA     ; 65c02 nmi vector, determined by the kernal if banked in
-    &uword  RESET_VEC   = $FFFC     ; 65c02 reset vector, determined by the kernal if banked in
-    &uword  IRQ_VEC     = $FFFE     ; 65c02 interrupt vector, determined by the kernal if banked in
 
     &uword  edkeyvec    = $ac03     ; (ram bank 0): for intercepting BASIN/CHRIN key strokes. See set_chrin_keyhandler()
     &ubyte  edkeybk     = $ac05     ; ...the RAM bank of the handler routine, if not in low ram
 
+    &ubyte  stavec      = $03b2     ; argument for stash()
+
 
 ; the sixteen virtual 16-bit registers in both normal unsigned mode and signed mode (s)
     &uword r0  = $0002
@@ -395,8 +400,8 @@ romsub $ff5c = LKUPSA(ubyte sa @Y)  clobbers(A,X,Y)
 romsub $ff5f = screen_mode(ubyte mode @A, bool getCurrent @Pc) -> ubyte @A, ubyte @X, ubyte @Y, bool @Pc        ; also see SCREEN or get_screen_mode()
 romsub $ff62 = screen_set_charset(ubyte charset @A, uword charsetptr @XY)  clobbers(A,X,Y)
 romsub $ff6e = JSRFAR()  ; following word = address to call, byte after that=rom/ram bank it is in
-romsub $ff74 = fetch(ubyte bank @X, ubyte index @Y)  clobbers(X) -> ubyte @A
-romsub $ff77 = stash(ubyte data @A, ubyte bank @X, ubyte index @Y)  clobbers(X)
+romsub $ff74 = fetch(ubyte zp_startaddr @A, ubyte bank @X, ubyte index @Y)  clobbers(X) -> ubyte @A
+romsub $ff77 = stash(ubyte data @A, ubyte bank @X, ubyte index @Y)  clobbers(X)     ;  note: The the zero page address containing the base address is passed in stavec ($03B2)
 romsub $ff7d = PRIMM()
 
 ; high level graphics & fonts
@@ -549,7 +554,8 @@ const ubyte  EXTAPI_ps2data_raw = $09
 const ubyte  EXTAPI_cursor_blink = $0A
 const ubyte  EXTAPI_led_update = $0B
 const ubyte  EXTAPI_mouse_set_position = $0C
-const ubyte  EXTAPI_scnsiz = $0D        ; rom R48+
+const ubyte  EXTAPI_scnsiz = $0D
+const ubyte  EXTAPI_kbd_leds = $0E
 
 ; extapi16 call numbers
 const ubyte  EXTAPI16_test = $00
@@ -598,6 +604,14 @@ asmsub mouse_pos() -> ubyte @A, uword @R0, uword @R1, byte @X {
     }}
 }
 
+sub mouse_present() -> bool {
+    ; -- check if a mouse is connected to the machine
+    cx16.r0L, void = cx16.i2c_read_byte($42, $22)         ; $22 = I2C_GET_MOUSE_DEVICE_ID
+    if_cs
+        return false
+    return cx16.r0L != $fc       ; $fc = BAT_FAIL
+}
+
 ; shims for the kernal routines called via the extapi call:
 
 asmsub mouse_set_pos(uword xpos @R0, uword ypos @R1) clobbers(X) {
@@ -1070,8 +1084,8 @@ asmsub  enable_irq_handlers(bool disable_all_irq_sources @Pc) clobbers(A,Y)  {
         trb  cx16.VERA_IEN      ; disable all IRQ sources
 +       lda  #<_irq_dispatcher
         ldy  #>_irq_dispatcher
-        sta  cx16.CINV
-        sty  cx16.CINV+1
+        sta  cbm.CINV
+        sty  cbm.CINV+1
         plp
         rts
 
@@ -1195,6 +1209,7 @@ asmsub set_sprcol_irq_handler(uword address @AY) clobbers(A) {
 asmsub set_aflow_irq_handler(uword address @AY) clobbers(A) {
     ; Sets the AFLOW irq handler to use with enable_irq_handlers().  Also enables AFLOW irqs.
     ; NOTE: unless a proper irq handler is already running, you should enclose this call in set_irqd() / clear_irqd() to avoid system crashes.
+    ; NOTE: the handler itself must fill the audio fifo buffer to at least 25% full again (1 KB) or the aflow irq will keep triggering!
     %asm {{
         php
         sei
@@ -1322,6 +1337,23 @@ _continue   iny
         void cx16.i2c_write_byte($42, $05, cx16.r0L)
     }
 
+    asmsub rom_version() clobbers(Y) -> ubyte @A, bool @Pc {
+        ; Returns the KERNEL ROM version. Carry set if pre-release, clear if offical release.
+        %asm{{
+            ; the ROM BANK is unknown on entry
+            ldy  $01
+            stz  $01        ; KERNEL ROM
+            clc             ; prepare for released ROM
+            lda  $FF80
+            bpl  _final     ; pre-release versions are negative
+            eor  #$FF       ; twos complement
+            ina
+            sec
+    _final:
+            sty  $01
+            rts
+        }}
+    }
 }
 
 sys {
@@ -1378,9 +1410,9 @@ asmsub  init_system()  {
 asmsub  init_system_phase2()  {
     %asm {{
         sei
-        lda  cx16.CINV
+        lda  cbm.CINV
         sta  restore_irq._orig_irqvec
-        lda  cx16.CINV+1
+        lda  cbm.CINV+1
         sta  restore_irq._orig_irqvec+1
         lda  #PROG8_VARSHIGH_RAMBANK
         sta  $00    ; select ram bank
@@ -1418,9 +1450,9 @@ asmsub  set_irq(uword handler @AY) clobbers(A)  {
         sta  _modified+1
         sty  _modified+2
         lda  #<_irq_handler
-        sta  cx16.CINV
+        sta  cbm.CINV
         lda  #>_irq_handler
-        sta  cx16.CINV+1
+        sta  cbm.CINV+1
         lda  #1
         tsb  cx16.VERA_IEN      ; enable the vsync irq
         cli
@@ -1449,9 +1481,9 @@ asmsub  restore_irq() clobbers(A) {
 	%asm {{
 	    sei
 	    lda  _orig_irqvec
-	    sta  cx16.CINV
+	    sta  cbm.CINV
 	    lda  _orig_irqvec+1
-	    sta  cx16.CINV+1
+	    sta  cbm.CINV+1
 	    lda  cx16.VERA_IEN
 	    and  #%11110000     ; disable all Vera IRQs but the vsync
 	    ora  #%00000001
@@ -1478,9 +1510,9 @@ asmsub  set_rasterirq(uword handler @AY, uword rasterpos @R0) clobbers(A) {
             ldy  cx16.r0+1
             jsr  set_rasterline
             lda  #<_raster_irq_handler
-            sta  cx16.CINV
+            sta  cbm.CINV
             lda  #>_raster_irq_handler
-            sta  cx16.CINV+1
+            sta  cbm.CINV+1
             cli
             rts
 

compiler/res/prog8lib/cx16/textio.p8

@@ -262,6 +262,12 @@ sub iso16() {
     cx16.screen_set_charset(10, 0)  ; charset
 }
 
+sub kata() {
+    ; -- switch to katakana character set (requires rom 48+)
+    cbm.CHROUT($0f)                 ; iso mode
+    cx16.screen_set_charset(12, 0)  ; charset
+}
+
 asmsub  scroll_left() clobbers(A, X, Y)  {
 	; ---- scroll the whole screen 1 character to the left
 	;      contents of the rightmost column are unchanged, you should clear/refill this yourself

compiler/res/prog8lib/cx16/verafx.p8

@@ -1,6 +1,6 @@
 ; Somewhat experimental Vera FX support.
 ; Docs:
-; https://github.com/X16Community/x16-docs/blob/101759f3bfa5e6cce4e8c5a0b67cb0f2f1c6341e/X16%20Reference%20-%2010%20-%20VERA%20FX%20Reference.md
+; https://github.com/X16Community/x16-docs/blob/fb63156cca2d6de98be0577aacbe4ddef458f896/X16%20Reference%20-%2010%20-%20VERA%20FX%20Reference.md
 ; https://docs.google.com/document/d/1q34uWOiM3Be2pnaHRVgSdHySI-qsiQWPTo_gfE54PTg
 
 verafx {
@@ -166,6 +166,8 @@ verafx {
     }
 
     sub transparency(bool enable) {
+        ; Set transparent write mode for VeraFX cached writes and also for normal writes to DATA0/DATA.
+        ; If enabled, pixels with value 0 do not modify VRAM when written (so they are "transparent")
         cx16.VERA_CTRL = 2<<1       ; dcsel = 2
         if enable
             cx16.VERA_FX_CTRL |= %10000000

compiler/res/prog8lib/diskio.p8

@@ -333,6 +333,7 @@ m_in_buffer     sta  $ffff
 
     sub f_read_all(uword bufferpointer) -> uword {
         ; -- read the full contents of the file, returns number of bytes read.
+        ;    It is assumed the file size is less than 64 K.
         if not iteration_in_progress
             return 0
 
@@ -346,12 +347,13 @@ m_in_buffer     sta  $ffff
         return total_read
     }
 
-    asmsub f_readline(uword bufptr @AY) clobbers(X) -> ubyte @Y {
+    asmsub f_readline(uword bufptr @AY) clobbers(X) -> ubyte @Y, ubyte @A {
         ; Routine to read text lines from a text file. Lines must be less than 255 characters.
         ; Reads characters from the input file UNTIL a newline or return character (or EOF).
         ; The line read will be 0-terminated in the buffer (and not contain the end of line character).
         ; The length of the line is returned in Y. Note that an empty line is okay and is length 0!
         ; I/O error status should be checked by the caller itself via READST() routine.
+        ; The I/O error status byte is returned in A.
         %asm {{
             sta  P8ZP_SCRATCH_W1
             sty  P8ZP_SCRATCH_W1+1
@@ -368,7 +370,8 @@ _loop       jsr  cbm.CHRIN
 _line_end   dey     ; get rid of the trailing end-of-line char
             lda  #0
             sta  (P8ZP_SCRATCH_W1),y
-_end        rts
+_end        jsr  cbm.READST
+            rts
         }}
     }
 
@@ -405,6 +408,7 @@ _end        rts
 
     sub f_write(uword bufferpointer, uword num_bytes) -> bool {
         ; -- write the given number of bytes to the currently open file
+        ;    you can call this multiple times to append more data
         if num_bytes!=0 {
             reset_write_channel()
             repeat num_bytes {

compiler/res/prog8lib/prog8_funcs.asm

@@ -1,60 +1,5 @@
 ; ---- builtin functions
 
-
-func_any_b_into_A	.proc
-		; -- any(array),  array in P8ZP_SCRATCH_W1, num bytes in A
-		sta  _cmp_mod+1		; self-modifying code
-		ldy  #0
--		lda  (P8ZP_SCRATCH_W1),y
-		bne  _got_any
-		iny
-_cmp_mod	cpy  #255		; modified
-		bne  -
-		lda  #0
-		rts
-_got_any	lda  #1
-		rts
-		.pend
-
-
-func_all_b_into_A	.proc
-		; -- all(array),  array in P8ZP_SCRATCH_W1, num bytes in A
-		sta  _cmp_mod+1		; self-modifying code
-		ldy  #0
--		lda  (P8ZP_SCRATCH_W1),y
-		beq  _got_not_all
-		iny
-_cmp_mod	cpy  #255		; modified
-		bne  -
-		lda  #1
-_got_not_all	rts
-		.pend
-
-func_any_w_into_A	.proc
-		asl  a
-		jmp  func_any_b_into_A
-		.pend
-
-func_all_w_into_A	.proc
-		; -- all(warray),  array in P8ZP_SCRATCH_W1, num bytes in A
-		asl  a			; times 2 because of word
-		sta  _cmp_mod+1		; self-modifying code
-		ldy  #0
--		lda  (P8ZP_SCRATCH_W1),y
-		bne  +
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		bne  ++
-		lda  #0
-		rts
-+		iny
-+		iny
-_cmp_mod	cpy  #255		; modified
-		bne  -
-		lda  #1
-		rts
-		.pend
-
 abs_b_into_A	.proc
 	; -- A = abs(A)
 		cmp  #0

compiler/res/prog8lib/prog8_lib.asm

@@ -248,31 +248,25 @@ _arg_s2		.word  0
 
 strcmp_mem	.proc
 		; --   compares strings in s1 (AY) and s2 (P8ZP_SCRATCH_W2).
-		;      Returns -1,0,1 in A, depeding on the ordering. Clobbers Y.
+		;      Returns -1,0,1 in A, depending on the ordering. Clobbers Y.
 		sta  P8ZP_SCRATCH_W1
 		sty  P8ZP_SCRATCH_W1+1
 		ldy  #0
-_loop		lda  (P8ZP_SCRATCH_W1),y
-		bne  +
-		lda  (P8ZP_SCRATCH_W2),y
-		bne  _return_minusone
-		beq  _return
-+		cmp  (P8ZP_SCRATCH_W2),y
-		bcc  _return_minusone
-		bne  _return_one
-		inc  P8ZP_SCRATCH_W1
-		bne  +
-		inc  P8ZP_SCRATCH_W1+1
-+		inc  P8ZP_SCRATCH_W2
-		bne  _loop
-		inc  P8ZP_SCRATCH_W2+1
-		bne  _loop
-_return_one
+_loop           lda  (P8ZP_SCRATCH_W1),y
+		beq  _c1_zero
+		cmp  (P8ZP_SCRATCH_W2),y
+		beq  _equal
+		bmi  _less
 		lda  #1
-_return		rts
-_return_minusone
-		lda  #-1
 		rts
+_less           lda  #-1
+		rts
+_equal          iny
+		bne  _loop
+_c1_zero        lda  (P8ZP_SCRATCH_W2),y
+		beq  +
+		lda  #-1
++		rts
 		.pend
 
 

compiler/res/prog8lib/shared_floats_functions.p8

@@ -96,7 +96,16 @@ sub atan(float value) -> float {
 ; two-argument arctangent that returns an angle in the correct quadrant
 ; for the signs of x and y, normalized to the range [0, 2π]
 sub atan2(float y, float x) -> float {
-    float atn = atan(y / x)
+    float atn
+    if x == 0 {
+        atn = π/2
+        if y == 0 return 0
+        if y < 0 {
+            atn += π
+        }
+    } else {
+        atn = atan(y / x)
+    }
     if x < 0 atn += π
     if atn < 0 atn += 2*π
     return atn

compiler/res/prog8lib/string.p8

@@ -151,6 +151,37 @@ _found	tya
         }}
     }
 
+    asmsub rfind(uword string @AY, ubyte character @X) -> ubyte @A, bool @Pc {
+        ; Locates the first position of the given character in the string, starting from the right.
+        ; returns Carry set if found + index in A, or Carry clear if not found (and A will be 255, an invalid index).
+        %asm {{
+            stx  P8ZP_SCRATCH_B1
+            sta  _str
+            sty  _str+1
+            jsr  string.length
+            dey
+            lda  _str
+    		sta  P8ZP_SCRATCH_W1
+    		lda  _str+1
+    		sta  P8ZP_SCRATCH_W1+1
+-           lda  (P8ZP_SCRATCH_W1),y
+            cmp  P8ZP_SCRATCH_B1
+            beq  _found
+            dey
+            cpy  #255
+            bne  -
+_notfound   lda  #255
+            clc
+            rts
+_found      tya
+            sec
+            rts
+
+_str    .word 0
+
+        }}
+    }
+
     asmsub contains(uword string @AY, ubyte character @X) -> bool @Pc {
         ; Just return true/false if the character is in the given string or not.
         %asm {{
@@ -199,9 +230,9 @@ _found	tya
 
     asmsub compare(uword string1 @R0, uword string2 @AY) clobbers(Y) -> byte @A {
         ; Compares two strings for sorting.
-        ; Returns -1 (255), 0 or 1 depending on wether string1 sorts before, equal or after string2.
+        ; Returns -1 (255), 0 or 1, meaning: string1 sorts before, equal or after string2.
         ; Note that you can also directly compare strings and string values with eachother using
-        ; comparison operators ==, < etcetera (it will use strcmp for you under water automatically).
+        ; comparison operators ==, < etcetera (this will use strcmp automatically).
         %asm {{
 		sta  P8ZP_SCRATCH_W2
 		sty  P8ZP_SCRATCH_W2+1

compiler/res/prog8lib/virtual/diskio.p8

@@ -11,7 +11,7 @@ diskio {
         %ir {{
             loadm.w r65534,diskio.load.filenameptr
             loadm.w r65535,diskio.load.address_override
-            syscall 61 (): r0.b
+            syscall 48 (): r0.b
             returnr.b r0
         }}
     }
@@ -72,6 +72,7 @@ diskio {
 
     sub f_read_all(uword bufferpointer) -> uword {
         ; -- read the full contents of the file, returns number of bytes read.
+        ;    It is assumed the file size is less than 64 K.
         txt.print("@TODO: f_read_all\n")
         return 0
     }
@@ -81,7 +82,7 @@ diskio {
         ; Reads characters from the input file UNTIL a newline or return character (or EOF).
         ; The line read will be 0-terminated in the buffer (and not contain the end of line character).
         ; The length of the line is returned in Y. Note that an empty line is okay and is length 0!
-        ; I/O error status should be checked by the caller itself via READST() routine.
+        ; This routine is not able here to return the status as well in a secondary return value, so you have to do that yourself.
         txt.print("@TODO: f_readline\n")
         return 0
     }
@@ -107,6 +108,7 @@ diskio {
 
     sub f_write(uword bufferpointer, uword num_bytes) -> bool {
         ; -- write the given number of bytes to the currently open file
+        ;    you can call this multiple times to append more data
         txt.print("@TODO: f_write\n")
         return false
     }
@@ -157,7 +159,7 @@ diskio {
             loadm.w r65533,diskio.save.filenameptr
             loadm.w r65534,diskio.save.start_address
             loadm.w r65535,diskio.save.savesize
-            syscall 58 (r65532.b, r65533.w, r65534.w, r65535.w): r0.b
+            syscall 45 (r65532.b, r65533.w, r65534.w, r65535.w): r0.b
             returnr.b r0
         }}
     }
@@ -169,7 +171,7 @@ diskio {
             loadm.w r65533,diskio.save.filenameptr
             loadm.w r65534,diskio.save.start_address
             loadm.w r65535,diskio.save.savesize
-            syscall 58 (r65532.b, r65533.w, r65534.w, r65535.w): r0.b
+            syscall 45 (r65532.b, r65533.w, r65534.w, r65535.w): r0.b
             returnr.b r0
         }}
     }
@@ -184,7 +186,7 @@ diskio {
         %ir {{
             loadm.w r65534,diskio.load.filenameptr
             loadm.w r65535,diskio.load.address_override
-            syscall 56 (r65534.w, r65535.w): r0.w
+            syscall 43 (r65534.w, r65535.w): r0.w
             returnr.w r0
         }}
     }
@@ -196,7 +198,7 @@ diskio {
         %ir {{
             loadm.w r65534,diskio.load_raw.filenameptr
             loadm.w r65535,diskio.load_raw.start_address
-            syscall 57 (r65534.w, r65535.w): r0.w
+            syscall 44 (r65534.w, r65535.w): r0.w
             returnr.w r0
         }}
     }
@@ -205,7 +207,7 @@ diskio {
         ; -- delete a file on the drive
         %ir {{
             loadm.w r65535,diskio.delete.filenameptr
-            syscall 59 (r65535.w)
+            syscall 46 (r65535.w)
         }}
     }
 
@@ -214,7 +216,7 @@ diskio {
         %ir {{
             loadm.w r65534,diskio.rename.oldfileptr
             loadm.w r65535,diskio.rename.newfileptr
-            syscall 60 (r65534.w, r65535.w)
+            syscall 47 (r65534.w, r65535.w)
         }}
     }
 }

compiler/res/prog8lib/virtual/floats.p8

@@ -13,7 +13,7 @@ sub print(float value) {
     ; ---- prints the floating point value (without a newline and no leading spaces).
     %ir {{
         loadm.f fr65535,floats.print.value
-        syscall 25 (fr65535.f)
+        syscall 15 (fr65535.f)
         return
     }}
 }
@@ -24,7 +24,7 @@ sub tostr(float value) -> str {
     %ir {{
         load.w r65535,floats.tostr.buffer
         loadm.f fr65535,floats.tostr.value
-        syscall 47 (r65535.w, fr65535.f)
+        syscall 34 (r65535.w, fr65535.f)
         load.w r0,floats.tostr.buffer
         returnr.w r0
     }}
@@ -34,7 +34,7 @@ sub parse(str value) -> float {
     ; -- parse a string value of a number to float
     %ir {{
         loadm.w  r65535,floats.parse.value
-        syscall 45 (r65535.w): fr0.f
+        syscall 32 (r65535.w): fr0.f
         returnr.f fr0
     }}
 }
@@ -147,7 +147,7 @@ sub ceil(float value) -> float {
 
 sub rnd() -> float {
     %ir {{
-        syscall 35 () : fr0.f
+        syscall 22 () : fr0.f
         returnr.f fr0
     }}
 }
@@ -155,7 +155,7 @@ sub rnd() -> float {
 sub rndseed(float seed) {
     %ir {{
         loadm.f  fr65535,floats.rndseed.seed
-        syscall 32 (fr65535.f)
+        syscall 19 (fr65535.f)
         return
     }}
 }

compiler/res/prog8lib/virtual/math.p8

@@ -164,14 +164,14 @@ math {
 
     sub rnd() -> ubyte {
         %ir {{
-            syscall 33 (): r0.b
+            syscall 20 (): r0.b
             returnr.b r0
         }}
     }
 
     sub rndw() -> uword {
         %ir {{
-            syscall 34 (): r0.w
+            syscall 21 (): r0.w
             returnr.w r0
         }}
     }
@@ -199,7 +199,7 @@ math {
         %ir {{
             loadm.w r65534,math.rndseed.seed1
             loadm.w r65535,math.rndseed.seed2
-            syscall 31 (r65534.w, r65535.w)
+            syscall 19 (r65534.w, r65535.w)
             return
         }}
     }
@@ -233,7 +233,7 @@ math {
     sub direction(ubyte x1, ubyte y1, ubyte x2, ubyte y2) -> ubyte {
         ; From a pair of positive coordinates, calculate discrete direction between 0 and 23 into A.
         ; This adjusts the atan() result  so that the direction N is centered on the angle=N instead of having it as a boundary
-        ubyte angle = atan2(x1, y1, x2, y2) - 256/48
+        ubyte angle = atan2(x1, y1, x2, y2) - (256/48 as ubyte)
         return 23-lsb(mkword(angle,0) / 2730)
     }
 
@@ -280,7 +280,7 @@ math {
             loadm.b r65533,math.atan2.y1
             loadm.b r65534,math.atan2.x2
             loadm.b r65535,math.atan2.y2
-            syscall 44 (r65532.b, r65533.b, r65534.b, r65535.b): r0.b
+            syscall 31 (r65532.b, r65533.b, r65534.b, r65535.b): r0.b
             returnr.b r0
         }}
     }
@@ -294,7 +294,7 @@ math {
         ;   - not all multiplications in the source code result in an actual multiplication call:
         ;     some simpler multiplications will be optimized away into faster routines. These will not set the upper 16 bits at all!
         %ir {{
-            syscall 46 (): r0.w
+            syscall 33 (): r0.w
             returnr.w r0
         }}
     }

compiler/res/prog8lib/virtual/monogfx.p8

@@ -23,7 +23,7 @@ monogfx {
         width = 320
         height = 240
         mode = MODE_NORMAL
-        clear_screen(0)
+        clear_screen(false)
     }
 
     sub hires() {
@@ -32,7 +32,7 @@ monogfx {
         width = 640
         height = 480
         mode = MODE_NORMAL
-        clear_screen(0)
+        clear_screen(false)
     }
 
     sub textmode() {
@@ -43,9 +43,10 @@ monogfx {
         mode = dm
     }
 
-    sub clear_screen(ubyte color) {
-        if color!=0
-            color=255
+    sub clear_screen(bool draw) {
+        ubyte color = 0
+        if draw
+            color = 255
         sys.gfx_clear(color)
     }
 
@@ -64,7 +65,7 @@ monogfx {
 
     sub fillrect(uword xx, uword yy, uword rwidth, uword rheight, bool draw) {
         ; Draw a filled rectangle of the given size and color.
-        ; To fill the whole screen, use clear_screen(color) instead - it is much faster.
+        ; To fill the whole screen, use clear_screen(draw) instead - it is much faster.
         if rwidth==0
             return
         repeat rheight {
@@ -300,12 +301,23 @@ monogfx {
             return
         ubyte @zp yy = 0
         word @zp decisionOver2 = (1 as word)-radius
+        uword last_y3 = ycenter+radius
+        uword last_y4 = ycenter-radius
+        uword new_y3, new_y4
 
         while radius>=yy {
             horizontal_line(xcenter-radius, ycenter+yy, radius*$0002+1, draw)
             horizontal_line(xcenter-radius, ycenter-yy, radius*$0002+1, draw)
-            horizontal_line(xcenter-yy, ycenter+radius, yy*$0002+1, draw)
-            horizontal_line(xcenter-yy, ycenter-radius, yy*$0002+1, draw)
+            new_y3 = ycenter+radius
+            if new_y3 != last_y3 {
+                horizontal_line(xcenter-yy, last_y3, yy*$0002+1, draw)
+                last_y3 = new_y3
+            }
+            new_y4 = ycenter-radius
+            if new_y4 != last_y4 {
+                horizontal_line(xcenter-yy, last_y4, yy*$0002+1, draw)
+                last_y4 = new_y4
+            }
             yy++
             if decisionOver2>=0 {
                 radius--
@@ -314,6 +326,10 @@ monogfx {
             decisionOver2 += yy*$0002
             decisionOver2++
         }
+        ; draw the final two spans
+        yy--
+        horizontal_line(xcenter-yy, last_y3, yy*$0002+1, draw)
+        horizontal_line(xcenter-yy, last_y4, yy*$0002+1, draw)
     }
 
     sub safe_disc(uword @zp xcenter, uword @zp ycenter, ubyte @zp radius, bool draw) {
@@ -323,12 +339,23 @@ monogfx {
             return
         ubyte @zp yy = 0
         word @zp decisionOver2 = (1 as word)-radius
+        uword last_y3 = ycenter+radius
+        uword last_y4 = ycenter-radius
+        uword new_y3, new_y4
 
         while radius>=yy {
             safe_horizontal_line(xcenter-radius, ycenter+yy, radius*$0002+1, draw)
             safe_horizontal_line(xcenter-radius, ycenter-yy, radius*$0002+1, draw)
-            safe_horizontal_line(xcenter-yy, ycenter+radius, yy*$0002+1, draw)
-            safe_horizontal_line(xcenter-yy, ycenter-radius, yy*$0002+1, draw)
+            new_y3 = ycenter+radius
+            if new_y3 != last_y3 {
+                safe_horizontal_line(xcenter-yy, last_y3, yy*$0002+1, draw)
+                last_y3 = new_y3
+            }
+            new_y4 = ycenter-radius
+            if new_y4 != last_y4 {
+                safe_horizontal_line(xcenter-yy, last_y4, yy*$0002+1, draw)
+                last_y4 = new_y4
+            }
             yy++
             if decisionOver2>=0 {
                 radius--
@@ -337,6 +364,10 @@ monogfx {
             decisionOver2 += yy*$0002
             decisionOver2++
         }
+        ; draw the final two spans
+        yy--
+        safe_horizontal_line(xcenter-yy, last_y3, yy*$0002+1, draw)
+        safe_horizontal_line(xcenter-yy, last_y4, yy*$0002+1, draw)
     }
 
     sub plot(uword @zp xx, uword @zp yy, bool @zp draw) {
@@ -377,7 +408,7 @@ monogfx {
         ; Non-recursive scanline flood fill.
         ; based loosely on code found here https://www.codeproject.com/Articles/6017/QuickFill-An-efficient-flood-fill-algorithm
         ; with the fixes applied to the seedfill_4 routine as mentioned in the comments.
-        const ubyte MAXDEPTH = 64
+        const ubyte MAXDEPTH = 100
         word @zp xx = x as word
         word @zp yy = y as word
         word[MAXDEPTH] @split @shared stack_xl
@@ -423,11 +454,10 @@ monogfx {
             while xx >= 0 {
                 if pget(xx as uword, yy as uword) as ubyte != cx16.r11L
                     break
+                plot(xx as uword, yy as uword, cx16.r10L as bool)
                 xx--
             }
-            if x1!=xx
-                horizontal_line(xx as uword+1, yy as uword, x1-xx as uword, cx16.r10L as bool)
-            else
+            if x1==xx
                 goto skip
 
             left = xx + 1
@@ -436,15 +466,12 @@ monogfx {
             xx = x1 + 1
 
             do {
-                cx16.r9 = xx as uword
                 while xx <= width-1 {
                     if pget(xx as uword, yy as uword) as ubyte != cx16.r11L
                         break
+                    plot(xx as uword, yy as uword, cx16.r10L as bool)
                     xx++
                 }
-                if cx16.r9!=xx
-                    horizontal_line(cx16.r9, yy as uword, (xx as uword)-cx16.r9, cx16.r10L as bool)
-
                 push_stack(left, xx - 1, yy, dy)
                 if xx > x2 + 1
                     push_stack(x2 + 1, xx - 1, yy, -dy)

compiler/res/prog8lib/virtual/string.p8

@@ -68,6 +68,21 @@ string {
         return 255
     }
 
+    sub rfind(uword stringptr, ubyte character) -> ubyte {
+        ; Locates the first position of the given character in the string, starting from the right.
+        ; returns Carry set if found + index in A, or Carry clear if not found (and A will be 255, an invalid index).
+        ; NOTE: because this isn't an asmsub, there's only a SINGLE return value here. On the c64/cx16 targets etc there are 2 return values.
+        ubyte ix
+        for ix in string.length(stringptr)-1 downto 0 {
+            if stringptr[ix]==character {
+                sys.set_carry()
+                return ix
+            }
+        }
+        sys.clear_carry()
+        return 255
+    }
+
     sub contains(str st, ubyte character) -> bool {
         void find(st, character)
         if_cs
@@ -83,7 +98,7 @@ string {
         %ir {{
             loadm.w r65534,string.copy.source
             loadm.w r65535,string.copy.target
-            syscall 52 (r65534.w, r65535.w): r0.b
+            syscall 39 (r65534.w, r65535.w): r0.b
             returnr.b r0
         }}
     }
@@ -97,13 +112,13 @@ string {
 
     sub compare(str st1, str st2) -> byte {
         ; Compares two strings for sorting.
-        ; Returns -1 (255), 0 or 1 depending on wether string1 sorts before, equal or after string2.
+        ; Returns -1 (255), 0 or 1, meaning: string1 sorts before, equal or after string2.
         ; Note that you can also directly compare strings and string values with eachother using
-        ; comparison operators ==, < etcetera (it will use strcmp for you under water automatically).
+        ; comparison operators ==, < etcetera (this will use strcmp automatically).
         %ir {{
             loadm.w r65534,string.compare.st1
             loadm.w r65535,string.compare.st2
-            syscall 29 (r65534.w, r65535.w) : r0.b
+            syscall 16 (r65534.w, r65535.w) : r0.b
             returnr.b r0
         }}
     }

compiler/res/prog8lib/virtual/syslib.p8

@@ -34,7 +34,7 @@ sys {
         %ir {{
             loadm.w r65534,sys.internal_stringcopy.source
             loadm.w r65535,sys.internal_stringcopy.tgt
-            syscall 52 (r65534.w, r65535.w): r0.b
+            syscall 39 (r65534.w, r65535.w): r0.b
         }}
     }
 
@@ -43,7 +43,7 @@ sys {
             loadm.w r65533,sys.memcopy.source
             loadm.w r65534,sys.memcopy.tgt
             loadm.w r65535,sys.memcopy.count
-            syscall 49 (r65533.w, r65534.w, r65535.w)
+            syscall 36 (r65533.w, r65534.w, r65535.w)
         }}
     }
 
@@ -52,7 +52,7 @@ sys {
             loadm.w r65533,sys.memset.mem
             loadm.w r65534,sys.memset.numbytes
             loadm.b r65535,sys.memset.value
-            syscall 50 (r65533.w, r65534.w, r65535.b)
+            syscall 37 (r65533.w, r65534.w, r65535.b)
         }}
     }
 
@@ -61,7 +61,7 @@ sys {
             loadm.w r65533,sys.memsetw.mem
             loadm.w r65534,sys.memsetw.numwords
             loadm.w r65535,sys.memsetw.value
-            syscall 51 (r65533.w, r65534.w, r65535.w)
+            syscall 38 (r65533.w, r65534.w, r65535.w)
         }}
     }
 
@@ -85,6 +85,18 @@ sys {
         }}
     }
 
+    sub set_irqd() {
+        %ir {{
+            sei
+        }}
+    }
+
+    sub clear_irqd() {
+        %ir {{
+            cli
+        }}
+    }
+
     sub disable_caseswitch() {
         ; no-op
     }
@@ -128,7 +140,7 @@ sys {
         %ir {{
             loadm.w r65534,sys.gfx_getpixel.xx
             loadm.w r65535,sys.gfx_getpixel.yy
-            syscall 30 (r65534.w, r65535.w): r0.b
+            syscall 17 (r65534.w, r65535.w): r0.b
             returnr.b r0
         }}
     }

compiler/res/prog8lib/virtual/textio.p8

@@ -7,14 +7,14 @@ txt {
 
 sub width() -> ubyte {
     %ir {{
-        syscall 62 (): r0.w
+        syscall 49 (): r0.w
         returnr.b r0
     }}
 }
 
 sub height() -> ubyte {
     %ir {{
-        syscall 62 (): r0.w
+        syscall 49 (): r0.w
         msig.b r1,r0
         returnr.b r1
     }}

compiler/src/prog8/CompilerMain.kt

@@ -19,6 +19,11 @@ import kotlin.system.exitProcess
 
 
 fun main(args: Array<String>) {
+
+    // NOTE: if your compiler/IDE complains here about "Unresolved reference: buildversion",
+    //       it means that you have to run the gradle task once to generate this file.
+    //       Do that with this command:  gradle createVersionFile
+
     println("\nProg8 compiler v${prog8.buildversion.VERSION} by Irmen de Jong (irmen@razorvine.net)")
     if('-' in prog8.buildversion.VERSION) {
         println("Prerelease version from git commit ${prog8.buildversion.GIT_SHA.take(8)} in branch ${prog8.buildversion.GIT_BRANCH}")
@@ -45,7 +50,6 @@ private fun compileMain(args: Array<String>): Boolean {
     val dumpVariables by cli.option(ArgType.Boolean, fullName = "dumpvars", description = "print a dump of the variables in the program")
     val dumpSymbols by cli.option(ArgType.Boolean, fullName = "dumpsymbols", description = "print a dump of the variable declarations and subroutine signatures")
     val dontWriteAssembly by cli.option(ArgType.Boolean, fullName = "noasm", description="don't create assembly code")
-    val noStrictBool by cli.option(ArgType.Boolean, fullName = "nostrictbool", description = "allow implicit conversions between bool and bytes")
     val dontOptimize by cli.option(ArgType.Boolean, fullName = "noopt", description = "don't perform code optimizations")
     val outputDir by cli.option(ArgType.String, fullName = "out", description = "directory for output files instead of current directory").default(".")
     val quietAssembler by cli.option(ArgType.Boolean, fullName = "quietasm", description = "don't print assembler output results")
@@ -102,10 +106,14 @@ private fun compileMain(args: Array<String>): Boolean {
         }
     }
 
-    if(bytes2float!=null)
-        return convertBytesToFloat(bytes2float!!, compilationTarget!!)
-    if(float2bytes!=null)
-        return convertFloatToBytes(float2bytes!!, compilationTarget!!)
+    if(bytes2float!=null) {
+        convertBytesToFloat(bytes2float!!, compilationTarget!!)
+        return true
+    }
+    if(float2bytes!=null) {
+        convertFloatToBytes(float2bytes!!, compilationTarget!!)
+        return true
+    }
 
     if(varsHighBank==0 && compilationTarget==Cx16Target.NAME) {
         System.err.println("On the Commander X16, HiRAM bank 0 is used by the kernal and can't be used.")
@@ -139,7 +147,8 @@ private fun compileMain(args: Array<String>): Boolean {
     }
 
     if(startVm==true) {
-        return runVm(moduleFiles.first())
+        runVm(moduleFiles.first())
+        return true
     }
 
     val processedSymbols = processSymbolDefs(symbolDefs) ?: return false
@@ -173,7 +182,6 @@ private fun compileMain(args: Array<String>): Boolean {
                     breakpointCpuInstruction,
                     printAst1 == true,
                     printAst2 == true,
-                    noStrictBool != true,
                     processedSymbols,
                     srcdirs,
                     outputPath
@@ -254,7 +262,6 @@ private fun compileMain(args: Array<String>): Boolean {
                     breakpointCpuInstruction,
                     printAst1 == true,
                     printAst2 == true,
-                    noStrictBool != true,
                     processedSymbols,
                     srcdirs,
                     outputPath
@@ -297,21 +304,19 @@ private fun compileMain(args: Array<String>): Boolean {
     return true
 }
 
-fun convertFloatToBytes(number: String, target: String): Boolean {
+fun convertFloatToBytes(number: String, target: String) {
     val tgt = getCompilationTargetByName(target)
     val dbl = number.toDouble()
     val bytes = tgt.machine.convertFloatToBytes(dbl)
     print("$dbl in bytes on '$target': ")
     println(bytes.joinToString(","))
-    return true
 }
 
-fun convertBytesToFloat(bytelist: String, target: String): Boolean {
+fun convertBytesToFloat(bytelist: String, target: String) {
     val tgt = getCompilationTargetByName(target)
     val bytes = bytelist.split(',').map { it.trim().toUByte() }
     val number = tgt.machine.convertBytesToFloat(bytes)
     println("floating point value on '$target': $number")
-    return true
 }
 
 private fun processSymbolDefs(symbolDefs: List<String>): Map<String, String>? {
@@ -329,9 +334,8 @@ private fun processSymbolDefs(symbolDefs: List<String>): Map<String, String>? {
     return result
 }
 
-fun runVm(irFilename: String): Boolean {
+fun runVm(irFilename: String) {
     val irFile = Path(irFilename)
     val vmdef = VirtualMachineDefinition()
     vmdef.launchEmulator(0, irFile)
-    return true
 }

compiler/src/prog8/compiler/BuiltinFunctions.kt

@@ -19,8 +19,6 @@ internal val constEvaluatorsForBuiltinFuncs: Map<String, ConstExpressionCaller>
     "sqrt__ubyte" to { a, p, prg -> oneIntArgOutputInt(a, p, prg, false) { sqrt(it.toDouble()) } },
     "sqrt__uword" to { a, p, prg -> oneIntArgOutputInt(a, p, prg, false) { sqrt(it.toDouble()) } },
     "sqrt__float" to { a, p, prg -> oneFloatArgOutputFloat(a, p, prg) { sqrt(it) } },
-    "any" to { a, p, prg -> collectionArgBoolResult(a, p, prg) { array->array.any { it!=0.0 }  } },
-    "all" to { a, p, prg -> collectionArgBoolResult(a, p, prg) { array->array.all { it!=0.0 }  } },
     "lsb" to { a, p, prg -> oneIntArgOutputInt(a, p, prg, true) { x: Int -> (x and 255).toDouble() } },
     "msb" to { a, p, prg -> oneIntArgOutputInt(a, p, prg, true) { x: Int -> (x ushr 8 and 255).toDouble()} },
     "mkword" to ::builtinMkword,
@@ -77,17 +75,6 @@ private fun oneFloatArgOutputFloat(args: List<Expression>, position: Position, p
     return NumericLiteral(DataType.FLOAT, function(constval.number), args[0].position)
 }
 
-private fun collectionArgBoolResult(args: List<Expression>, position: Position, program: Program, function: (arg: List<Double>)->Boolean): NumericLiteral {
-    if(args.size!=1)
-        throw SyntaxError("builtin function requires one non-scalar argument", position)
-
-    val array= args[0] as? ArrayLiteral ?: throw NotConstArgumentException()
-    val constElements = array.value.map{it.constValue(program)?.number}
-    if(constElements.contains(null))
-        throw NotConstArgumentException()
-    return NumericLiteral.fromBoolean(function(constElements.mapNotNull { it }), args[0].position)
-}
-
 private fun builtinAbs(args: List<Expression>, position: Position, program: Program): NumericLiteral {
     // 1 arg, type = int, result type= uword
     if(args.size!=1)
@@ -140,6 +127,8 @@ private fun builtinLen(args: List<Expression>, position: Position, program: Prog
         return NumericLiteral.optimalInteger(arraySize, position)
     if(args[0] is ArrayLiteral)
         return NumericLiteral.optimalInteger((args[0] as ArrayLiteral).value.size, position)
+    if(args[0] is StringLiteral)
+        return NumericLiteral.optimalInteger((args[0] as StringLiteral).value.length, position)
     if(args[0] !is IdentifierReference)
         throw SyntaxError("len argument should be an identifier", position)
     val target = (args[0] as IdentifierReference).targetVarDecl(program)

compiler/src/prog8/compiler/Compiler.kt

@@ -54,7 +54,6 @@ class CompilerArguments(val filepath: Path,
                         val breakpointCpuInstruction: String?,
                         val printAst1: Boolean,
                         val printAst2: Boolean,
-                        val strictBool: Boolean,
                         val symbolDefs: Map<String, String>,
                         val sourceDirs: List<String> = emptyList(),
                         val outputDir: Path = Path(""),
@@ -91,7 +90,6 @@ fun compileProgram(args: CompilerArguments): CompilationResult? {
                 splitWordArrays = args.splitWordArrays
                 outputDir = args.outputDir.normalize()
                 symbolDefs = args.symbolDefs
-                strictBool = args.strictBool
             }
             resultingProgram = program
             importedFiles = imported
@@ -302,10 +300,9 @@ fun parseMainModule(filepath: Path,
         .filter { it.isFromFilesystem }
         .map { Path(it.origin) }
     val compilerOptions = determineCompilationOptions(program, compTarget)
-    // depending on the machine and compiler options we may have to include some libraries
-    for(lib in compTarget.machine.importLibs(compilerOptions, compTarget.name))
-        importer.importImplicitLibraryModule(lib)
 
+    // import the default modules
+    importer.importImplicitLibraryModule("syslib")
     if(compilerOptions.compTarget.name!=VMTarget.NAME && !compilerOptions.experimentalCodegen) {
         importer.importImplicitLibraryModule("math")
     }
@@ -413,7 +410,7 @@ private fun processAst(program: Program, errors: IErrorReporter, compilerOptions
     errors.report()
     program.reorderStatements(errors)
     errors.report()
-    program.desugaring(errors, compilerOptions)
+    program.desugaring(errors)
     errors.report()
     program.changeNotExpressionAndIfComparisonExpr(errors, compilerOptions.compTarget)
     errors.report()
@@ -439,7 +436,7 @@ private fun optimizeAst(program: Program, compilerOptions: CompilationOptions, e
     removeUnusedCode(program, errors,compilerOptions)
     while (true) {
         // keep optimizing expressions and statements until no more steps remain
-        val optsDone1 = program.simplifyExpressions(errors, compilerOptions)
+        val optsDone1 = program.simplifyExpressions(errors)
         val optsDone2 = program.optimizeStatements(errors, functions, compilerOptions)
         val optsDone3 = program.inlineSubroutines(compilerOptions)
         program.constantFold(errors, compilerOptions) // because simplified statements and expressions can result in more constants that can be folded away
@@ -453,7 +450,7 @@ private fun optimizeAst(program: Program, compilerOptions: CompilationOptions, e
     removeUnusedCode(program, errors, compilerOptions)
     if(errors.noErrors()) {
         // last round of optimizations because constFold may have enabled more...
-        program.simplifyExpressions(errors, compilerOptions)
+        program.simplifyExpressions(errors)
         program.optimizeStatements(errors, functions, compilerOptions)
         program.constantFold(errors, compilerOptions) // because simplified statements and expressions can result in more constants that can be folded away
     }
@@ -462,7 +459,7 @@ private fun optimizeAst(program: Program, compilerOptions: CompilationOptions, e
 }
 
 private fun postprocessAst(program: Program, errors: IErrorReporter, compilerOptions: CompilationOptions) {
-    program.desugaring(errors, compilerOptions)
+    program.desugaring(errors)
     program.addTypecasts(errors, compilerOptions)
     errors.report()
     program.variousCleanups(errors, compilerOptions)

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

@@ -163,9 +163,7 @@ internal class AstChecker(private val program: Program,
 
     override fun visit(ifElse: IfElse) {
         if(!ifElse.condition.inferType(program).isBool) {
-            val ctype = ifElse.condition.inferType(program).getOr(DataType.UNDEFINED)
-            if(compilerOptions.strictBool || ctype !in ByteDatatypes)
-                errors.err("condition should be a boolean", ifElse.condition.position)
+            errors.err("condition should be a boolean", ifElse.condition.position)
         }
 
         val constvalue = ifElse.condition.constValue(program)
@@ -506,9 +504,7 @@ internal class AstChecker(private val program: Program,
 
     override fun visit(untilLoop: UntilLoop) {
         if(!untilLoop.condition.inferType(program).isBool) {
-            val ctype = untilLoop.condition.inferType(program).getOr(DataType.UNDEFINED)
-            if(compilerOptions.strictBool || ctype !in ByteDatatypes)
-                errors.err("condition should be a boolean", untilLoop.condition.position)
+            errors.err("condition should be a boolean", untilLoop.condition.position)
         }
 
         super.visit(untilLoop)
@@ -516,9 +512,7 @@ internal class AstChecker(private val program: Program,
 
     override fun visit(whileLoop: WhileLoop) {
         if(!whileLoop.condition.inferType(program).isBool) {
-            val ctype = whileLoop.condition.inferType(program).getOr(DataType.UNDEFINED)
-            if(compilerOptions.strictBool || ctype !in ByteDatatypes)
-                errors.err("condition should be a boolean", whileLoop.condition.position)
+            errors.err("condition should be a boolean", whileLoop.condition.position)
         }
 
         super.visit(whileLoop)
@@ -780,12 +774,8 @@ internal class AstChecker(private val program: Program,
             if (iDt isnot decl.datatype) {
                 if(decl.isArray) {
                     val eltDt = ArrayToElementTypes.getValue(decl.datatype)
-                    if(iDt isnot eltDt) {
-                        if(compilerOptions.strictBool)
-                            err("initialisation value has incompatible type ($iDt) for the variable (${decl.datatype})")
-                        else if(!(iDt.isBool && eltDt==DataType.UBYTE || iDt.istype(DataType.UBYTE) && eltDt==DataType.BOOL))
-                            err("initialisation value has incompatible type ($iDt) for the variable (${decl.datatype})")
-                    }
+                    if(iDt isnot eltDt)
+                        err("initialisation value has incompatible type ($iDt) for the variable (${decl.datatype})")
                 } else {
                     if(!(iDt.isBool && decl.datatype==DataType.UBYTE || iDt.istype(DataType.UBYTE) && decl.datatype==DataType.BOOL))
                         err("initialisation value has incompatible type ($iDt) for the variable (${decl.datatype})")
@@ -1048,8 +1038,7 @@ internal class AstChecker(private val program: Program,
         }
         else if(expr.operator == "not") {
             if(dt!=DataType.BOOL) {
-                if(compilerOptions.strictBool || dt !in ByteDatatypes)
-                    errors.err("logical not is for booleans", expr.position)
+                errors.err("logical not is for booleans", expr.position)
             }
         }
         super.visit(expr)
@@ -1168,8 +1157,7 @@ internal class AstChecker(private val program: Program,
 
         if(expr.operator in LogicalOperators) {
             if (leftDt != DataType.BOOL || rightDt != DataType.BOOL) {
-                if(compilerOptions.strictBool || leftDt !in ByteDatatypes || rightDt !in ByteDatatypes)
-                    errors.err("logical operator requires boolean operands", expr.right.position)
+                errors.err("logical operator requires boolean operands", expr.right.position)
             }
         }
         else {
@@ -1248,7 +1236,7 @@ internal class AstChecker(private val program: Program,
                 errors.warn("sgn() of unsigned type is always 0 or 1, this is perhaps not what was intended", functionCallExpr.args.first().position)
         }
 
-        val error = VerifyFunctionArgTypes.checkTypes(functionCallExpr, program, compilerOptions)
+        val error = VerifyFunctionArgTypes.checkTypes(functionCallExpr, program)
         if(error!=null)
             errors.err(error.first, error.second)
 
@@ -1339,7 +1327,7 @@ internal class AstChecker(private val program: Program,
 
         }
 
-        val error = VerifyFunctionArgTypes.checkTypes(functionCallStatement, program, compilerOptions)
+        val error = VerifyFunctionArgTypes.checkTypes(functionCallStatement, program)
         if(error!=null)
             errors.err(error.first, error.second)
 
@@ -1418,6 +1406,13 @@ internal class AstChecker(private val program: Program,
         if(target is VarDecl) {
             if(target.datatype !in IterableDatatypes && target.datatype!=DataType.UWORD)
                 errors.err("indexing requires an iterable or address uword variable", arrayIndexedExpression.position)
+            val indexVariable = arrayIndexedExpression.indexer.indexExpr as? IdentifierReference
+            if(indexVariable!=null) {
+                if(indexVariable.targetVarDecl(program)?.datatype in SignedDatatypes) {
+                    errors.err("variable array indexing can't be performed with signed variables", indexVariable.position)
+                    return
+                }
+            }
             val arraysize = target.arraysize?.constIndex()
             val index = arrayIndexedExpression.indexer.constIndex()
             if(arraysize!=null) {
@@ -1720,8 +1715,7 @@ internal class AstChecker(private val program: Program,
             }
             DataType.BOOL -> {
                 if (value.type!=DataType.BOOL) {
-                    if (compilerOptions.strictBool || value.type !in ByteDatatypes)
-                        err("type of value ${value.type} doesn't match target $targetDt")
+                    err("type of value ${value.type} doesn't match target $targetDt")
                 }
             }
             in ArrayDatatypes -> {
@@ -1822,12 +1816,10 @@ internal class AstChecker(private val program: Program,
             // this is allowed: a pass-by-reference datatype into a uword (pointer value).
         }
         else if(sourceDatatype==DataType.BOOL && targetDatatype!=DataType.BOOL) {
-            if(compilerOptions.strictBool || targetDatatype !in ByteDatatypes)
-                errors.err("type of value $sourceDatatype doesn't match target $targetDatatype", position)
+            errors.err("type of value $sourceDatatype doesn't match target $targetDatatype", position)
         }
         else if(targetDatatype==DataType.BOOL && sourceDatatype!=DataType.BOOL) {
-            if(compilerOptions.strictBool || sourceDatatype !in ByteDatatypes)
-                errors.err("type of value $sourceDatatype doesn't match target $targetDatatype", position)
+            errors.err("type of value $sourceDatatype doesn't match target $targetDatatype", position)
         }
         else {
             errors.err("type of value $sourceDatatype doesn't match target $targetDatatype", position)

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

@@ -92,8 +92,8 @@ internal fun Program.addTypecasts(errors: IErrorReporter, options: CompilationOp
     caster.applyModifications()
 }
 
-fun Program.desugaring(errors: IErrorReporter, options: CompilationOptions) {
-    val desugar = CodeDesugarer(this, options, errors)
+fun Program.desugaring(errors: IErrorReporter) {
+    val desugar = CodeDesugarer(this, errors)
     desugar.visit(this)
     while(errors.noErrors() && desugar.applyModifications()>0)
         desugar.visit(this)

compiler/src/prog8/compiler/astprocessing/AstIdentifiersChecker.kt

@@ -48,22 +48,30 @@ internal class AstIdentifiersChecker(private val errors: IErrorReporter,
         if(decl.name in BuiltinFunctions)
             errors.err("builtin function cannot be redefined", decl.position)
 
-        if(decl.names.size<2) {
-            val existingInSameScope = decl.definingScope.lookup(listOf(decl.name))
+        fun checkNameForErrors(name: String) {
+            val existingInSameScope = decl.definingScope.lookup(listOf(name))
             if (existingInSameScope != null && existingInSameScope !== decl)
-                nameError(decl.name, decl.position, existingInSameScope)
+                nameError(name, decl.position, existingInSameScope)
 
-            val existingOuter = decl.parent.definingScope.lookup(listOf(decl.name))
+            val existingOuter = decl.parent.definingScope.lookup(listOf(name))
             if (existingOuter != null && existingOuter !== decl) {
                 if (existingOuter is VarDecl) {
                     if (existingOuter.parent !== decl.parent)
-                        nameShadowWarning(decl.name, decl.position, existingOuter)
+                        nameShadowWarning(name, decl.position, existingOuter)
                     else
-                        nameError(decl.name, decl.position, existingOuter)
+                        nameError(name, decl.position, existingOuter)
                 }
             }
         }
 
+        if(decl.names.size<2) {
+            checkNameForErrors(decl.name)
+        }
+        else {
+            for (name in decl.names)
+                checkNameForErrors(name)
+        }
+
         if(decl.definingBlock.name==decl.name)
             nameError(decl.name, decl.position, decl.definingBlock)
         if(decl.definingSubroutine?.name==decl.name)
@@ -72,6 +80,7 @@ internal class AstIdentifiersChecker(private val errors: IErrorReporter,
         super.visit(decl)
     }
 
+
     override fun visit(subroutine: Subroutine) {
         if(subroutine.name in BuiltinFunctions) {
             // the builtin functions can't be redefined
@@ -177,7 +186,7 @@ internal class AstIdentifiersChecker(private val errors: IErrorReporter,
                                 '_'
                         }.joinToString("")
                         val textEncoding = (call as Node).definingModule.textEncoding
-                        call.args[0] = StringLiteral(processed, textEncoding, name.position)
+                        call.args[0] = StringLiteral.create(processed, textEncoding, name.position)
                         call.args[0].linkParents(call as Node)
                     }
                 }

compiler/src/prog8/compiler/astprocessing/CodeDesugarer.kt

@@ -8,7 +8,7 @@ import prog8.ast.walk.IAstModification
 import prog8.code.core.*
 
 
-internal class CodeDesugarer(val program: Program, private val options: CompilationOptions, private val errors: IErrorReporter) : AstWalker() {
+internal class CodeDesugarer(val program: Program, private val errors: IErrorReporter) : AstWalker() {
 
     // Some more code shuffling to simplify the Ast that the codegenerator has to process.
     // Several changes have already been done by the StatementReorderer !
@@ -123,9 +123,7 @@ if not CONDITION
     override fun after(whileLoop: WhileLoop, parent: Node): Iterable<IAstModification> {
 
         if(!whileLoop.condition.inferType(program).isBool) {
-            val ctype = whileLoop.condition.inferType(program).getOr(DataType.UNDEFINED)
-            if(options.strictBool || ctype !in ByteDatatypes)
-                errors.err("condition should be a boolean", whileLoop.condition.position)
+            errors.err("condition should be a boolean", whileLoop.condition.position)
         }
 
         /*

compiler/src/prog8/compiler/astprocessing/IntermediateAstMaker.kt

@@ -11,9 +11,9 @@ import prog8.ast.statements.*
 import prog8.code.ast.*
 import prog8.code.core.*
 import prog8.compiler.builtinFunctionReturnType
-import java.io.File
 import kotlin.io.path.Path
 import kotlin.io.path.isRegularFile
+import java.io.File
 
 
 /**
@@ -623,11 +623,10 @@ class IntermediateAstMaker(private val program: Program, private val errors: IEr
         }
 
         when(srcCheck.iterable) {
-            is IdentifierReference -> {
+            is IdentifierReference, is ArrayLiteral -> {
                 val check = PtContainmentCheck(srcCheck.position)
                 check.add(transformExpression(srcCheck.element))
-                val iterable = transformExpression(srcCheck.iterable)
-                check.add(iterable)
+                check.add(transformExpression(srcCheck.iterable))
                 return check
             }
             is RangeExpression -> {

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

@@ -4,19 +4,14 @@ import prog8.ast.IFunctionCall
 import prog8.ast.IStatementContainer
 import prog8.ast.Node
 import prog8.ast.Program
-import prog8.ast.expressions.ArrayLiteral
-import prog8.ast.expressions.BinaryExpression
-import prog8.ast.expressions.IdentifierReference
-import prog8.ast.expressions.StringLiteral
+import prog8.ast.expressions.*
 import prog8.ast.statements.Assignment
 import prog8.ast.statements.VarDecl
 import prog8.ast.statements.WhenChoice
 import prog8.ast.walk.AstWalker
 import prog8.ast.walk.IAstModification
-import prog8.code.core.DataType
-import prog8.code.core.IErrorReporter
-import prog8.code.core.NumericDatatypesWithBoolean
-import prog8.code.core.SplitWordArrayTypes
+import prog8.code.ast.PtContainmentCheck
+import prog8.code.core.*
 
 
 internal class LiteralsToAutoVars(private val program: Program, private val errors: IErrorReporter) : AstWalker() {
@@ -54,6 +49,16 @@ internal class LiteralsToAutoVars(private val program: Program, private val erro
             }
         } else {
             val arrayDt = array.guessDatatype(program)
+            val elementDt = ArrayToElementTypes.getValue(arrayDt.getOr(DataType.UNDEFINED))
+            val maxSize = when(elementDt) {
+                in ByteDatatypesWithBoolean -> PtContainmentCheck.MAX_SIZE_FOR_INLINE_CHECKS_BYTE
+                in WordDatatypes -> PtContainmentCheck.MAX_SIZE_FOR_INLINE_CHECKS_WORD
+                else -> 0
+            }
+            if(parent is ContainmentCheck && array.value.size <= maxSize) {
+                // keep the array in the containmentcheck inline
+                return noModifications
+            }
             if(arrayDt.isKnown) {
                 val parentAssign = parent as? Assignment
                 val targetDt = parentAssign?.target?.inferType(program) ?: arrayDt

compiler/src/prog8/compiler/astprocessing/NotExpressionAndIfComparisonExprChanger.kt

@@ -59,16 +59,15 @@ internal class NotExpressionAndIfComparisonExprChanger(val program: Program, val
             // integer case (only if both are the same type)
             val leftC = expr.left as? BinaryExpression
             val rightC = expr.right as? BinaryExpression
-            if(leftC!=null && rightC!=null && leftC.operator=="==" && rightC.operator=="==") {
+            if(expr.operator=="and" && leftC!=null && rightC!=null && leftC.operator=="==" && rightC.operator=="==") {
                 if (leftC.right.constValue(program)?.number == 0.0 && rightC.right.constValue(program)?.number == 0.0) {
                     val leftDt = leftC.left.inferType(program).getOr(DataType.UNDEFINED)
                     val rightDt = rightC.left.inferType(program).getOr(DataType.UNDEFINED)
                     if(leftDt==rightDt && leftDt in IntegerDatatypes) {
                         if (rightC.left.isSimple) {
-                            // x==0 or y==0    ->  (x & y)==0
                             // x==0 and y==0   ->  (x | y)==0
-                            val newOperator = if(expr.operator=="or") "&" else "|"
-                            val inner = BinaryExpression(leftC.left, newOperator, rightC.left, expr.position)
+                            // the 'or' case cannot be easily optimized with a binary and like this!
+                            val inner = BinaryExpression(leftC.left, "|", rightC.left, expr.position)
                             val compare = BinaryExpression(inner, "==", NumericLiteral(leftDt, 0.0, expr.position), expr.position)
                             return listOf(IAstModification.ReplaceNode(expr, compare, parent))
                         }

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

@@ -140,7 +140,7 @@ internal class StatementReorderer(
         // change 'str' and 'ubyte[]' parameters into 'uword' (just treat it as an address)
         val stringParams = subroutine.parameters.filter { it.type==DataType.STR || it.type==DataType.ARRAY_UB }
         val parameterChanges = stringParams.map {
-            val uwordParam = SubroutineParameter(it.name, DataType.UWORD, it.position)
+            val uwordParam = SubroutineParameter(it.name, DataType.UWORD, it.zp, it.position)
             IAstModification.ReplaceNode(it, uwordParam, subroutine)
         }
         // change 'str' and 'ubyte[]' return types into 'uword' (just treat it as an address)

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

@@ -24,11 +24,6 @@ class TypecastsAdder(val program: Program, val options: CompilationOptions, val
             val valueDt = declValue.inferType(program)
             if(valueDt isnot decl.datatype) {
 
-                if(decl.isArray && !options.strictBool) {
-                    if(tryConvertBooleanArrays(decl, declValue, parent))
-                        return noModifications
-                }
-
                 if(valueDt.isInteger && decl.isArray) {
                     if(decl.datatype == DataType.ARRAY_BOOL) {
                         val integer = declValue.constValue(program)?.number
@@ -54,69 +49,6 @@ class TypecastsAdder(val program: Program, val options: CompilationOptions, val
         return noModifications
     }
 
-    private fun tryConvertBooleanArrays(decl: VarDecl, declValue: Expression, parent: Node): Boolean {
-        val valueNumber = declValue.constValue(program)
-        val valueArray = declValue as? ArrayLiteral
-        when (decl.datatype) {
-            DataType.ARRAY_BOOL -> {
-                if(valueNumber!=null) {
-                    decl.value = NumericLiteral.fromBoolean(valueNumber.number!=0.0, declValue.position)
-                    decl.linkParents(parent)
-                    return true
-                } else if(valueArray!=null) {
-                    val newArray = valueArray.value.map {
-                        if(it.inferType(program).isBytes) {
-                            TypecastExpression(it, DataType.BOOL, false, it.position)
-                        } else {
-                            it
-                        }
-                    }
-                    decl.value = ArrayLiteral(InferredTypes.InferredType.known(DataType.ARRAY_BOOL), newArray.toTypedArray(), valueArray.position)
-                    decl.linkParents(parent)
-                    return true
-                }
-            }
-            DataType.ARRAY_B -> {
-                if(valueNumber!=null) {
-                    decl.value = NumericLiteral(DataType.BYTE, if(valueNumber.asBooleanValue) 1.0 else 0.0, declValue.position)
-                    decl.linkParents(parent)
-                    return true
-                } else if(valueArray!=null) {
-                    val newArray = valueArray.value.map {
-                        if(it.inferType(program).isBool) {
-                            TypecastExpression(it, DataType.BYTE, false, it.position)
-                        } else {
-                            it
-                        }
-                    }
-                    decl.value = ArrayLiteral(InferredTypes.InferredType.known(DataType.ARRAY_B), newArray.toTypedArray(), valueArray.position)
-                    decl.linkParents(parent)
-                    return true
-                }
-            }
-            DataType.ARRAY_UB -> {
-                if(valueNumber!=null) {
-                    decl.value = NumericLiteral(DataType.UBYTE, if(valueNumber.asBooleanValue) 1.0 else 0.0, declValue.position)
-                    decl.linkParents(parent)
-                    return true
-                } else if(valueArray!=null) {
-                    val newArray = valueArray.value.map {
-                        if(it.inferType(program).isBool) {
-                            TypecastExpression(it, DataType.UBYTE, false, it.position)
-                        } else {
-                            it
-                        }
-                    }
-                    decl.value = ArrayLiteral(InferredTypes.InferredType.known(DataType.ARRAY_UB), newArray.toTypedArray(), valueArray.position)
-                    decl.linkParents(parent)
-                    return true
-                }
-            }
-            else -> { /* no casting possible */ }
-        }
-        return false
-    }
-
     override fun after(expr: BinaryExpression, parent: Node): Iterable<IAstModification> {
         val leftDt = expr.left.inferType(program)
         val rightDt = expr.right.inferType(program)
@@ -155,26 +87,6 @@ class TypecastsAdder(val program: Program, val options: CompilationOptions, val
             }
 
             if(leftDt!=rightDt) {
-                if(!options.strictBool) {
-                    if (expr.operator in LogicalOperators) {
-                        if (leftDt.isBool) {
-                            val cast = TypecastExpression(expr.right, DataType.BOOL, false, expr.right.position)
-                            return listOf(IAstModification.ReplaceNode(expr.right, cast, expr))
-                        } else {
-                            val cast = TypecastExpression(expr.left, DataType.BOOL, false, expr.left.position)
-                            return listOf(IAstModification.ReplaceNode(expr.left, cast, expr))
-                        }
-                    } else {
-                        if(leftDt.isBool && rightDt.isBytes) {
-                            val cast = TypecastExpression(expr.left, rightDt.getOr(DataType.UNDEFINED), false, expr.left.position)
-                            return listOf(IAstModification.ReplaceNode(expr.left, cast, expr))
-                        } else if(rightDt.isBool && leftDt.isBytes) {
-                            val cast = TypecastExpression(expr.right, leftDt.getOr(DataType.UNDEFINED), false, expr.right.position)
-                            return listOf(IAstModification.ReplaceNode(expr.right, cast, expr))
-                        }
-                    }
-                }
-
                 // convert a negative operand for bitwise operator to the 2's complement positive number instead
                 if(expr.operator in BitwiseOperators && leftDt.isInteger && rightDt.isInteger) {
                     if(leftCv!=null && leftCv.number<0) {
@@ -278,42 +190,7 @@ class TypecastsAdder(val program: Program, val options: CompilationOptions, val
                     }
                     val cvalue = assignment.value.constValue(program)
                     if(cvalue!=null) {
-                        val number = cvalue.number
-                        // more complex comparisons if the type is different, but the constant value is compatible
-                        if(!options.strictBool) {
-                            if (targettype == DataType.BOOL && valuetype in ByteDatatypes) {
-                                val cast = NumericLiteral.fromBoolean(number!=0.0, cvalue.position)
-                                return listOf(IAstModification.ReplaceNode(assignment.value, cast, assignment))
-                            }
-                            if (targettype in ByteDatatypes && valuetype == DataType.BOOL) {
-                                val cast = NumericLiteral(targettype, if(cvalue.asBooleanValue) 1.0 else 0.0, cvalue.position)
-                                return listOf(IAstModification.ReplaceNode(assignment.value, cast, assignment))
-                            }
-                        }
-                        if (valuetype == DataType.BYTE && targettype == DataType.UBYTE) {
-                            if(number>0)
-                                return castLiteral(cvalue)
-                        } else if (valuetype == DataType.WORD && targettype == DataType.UWORD) {
-                            if(number>0)
-                                return castLiteral(cvalue)
-                        } else if (valuetype == DataType.UBYTE && targettype == DataType.BYTE) {
-                            if(number<0x80)
-                                return castLiteral(cvalue)
-                        } else if (valuetype == DataType.UWORD && targettype == DataType.WORD) {
-                            if(number<0x8000)
-                                return castLiteral(cvalue)
-                        }
-                    } else {
-                        if(!options.strictBool) {
-                            if (targettype == DataType.BOOL && valuetype in ByteDatatypes) {
-                                val cast = TypecastExpression(assignment.value, targettype, false, assignment.value.position)
-                                return listOf(IAstModification.ReplaceNode(assignment.value, cast, assignment))
-                            }
-                            if (targettype in ByteDatatypes && valuetype == DataType.BOOL) {
-                                val cast = TypecastExpression(assignment.value, targettype, false, assignment.value.position)
-                                return listOf(IAstModification.ReplaceNode(assignment.value, cast, assignment))
-                            }
-                        }
+                        return castLiteral(cvalue)
                     }
                 }
             }
@@ -365,8 +242,7 @@ class TypecastsAdder(val program: Program, val options: CompilationOptions, val
                     } else if(argDt isAssignableTo targetDt) {
                         if(argDt!=DataType.STR || targetDt!=DataType.UWORD)
                             addTypecastOrCastedValueModification(modifications, it.second, targetDt, call as Node)
-                    } else if(!options.strictBool && targetDt in ByteDatatypes && argDt==DataType.BOOL)
-                        addTypecastOrCastedValueModification(modifications, it.second, targetDt, call as Node)
+                    }
                 }
             } else {
                 val identifier = it.second as? IdentifierReference
@@ -432,18 +308,6 @@ class TypecastsAdder(val program: Program, val options: CompilationOptions, val
             if(subroutine.returntypes.size==1) {
                 val subReturnType = subroutine.returntypes.first()
                 val returnDt = returnValue.inferType(program)
-
-                if(!options.strictBool) {
-                    if(subReturnType==DataType.BOOL && returnDt.isBytes) {
-                        val cast = TypecastExpression(returnValue, DataType.BOOL, false, returnValue.position)
-                        return listOf(IAstModification.ReplaceNode(returnValue, cast, returnStmt))
-                    }
-                    if(subReturnType in ByteDatatypes && returnDt.isBool) {
-                        val cast = TypecastExpression(returnValue, subReturnType, false, returnValue.position)
-                        return listOf(IAstModification.ReplaceNode(returnValue, cast, returnStmt))
-                    }
-                }
-
                 if (returnDt istype subReturnType or returnDt.isNotAssignableTo(subReturnType))
                     return noModifications
                 if (returnValue is NumericLiteral) {
@@ -466,7 +330,7 @@ class TypecastsAdder(val program: Program, val options: CompilationOptions, val
             val values = whenChoice.values
             values?.toTypedArray()?.withIndex()?.forEach { (index, value) ->
                 val valueDt = value.inferType(program)
-                if(valueDt!=conditionDt) {
+                if(valueDt.isKnown && valueDt!=conditionDt) {
                     val castedValue = value.typecastTo(conditionDt.getOr(DataType.UNDEFINED), valueDt.getOr(DataType.UNDEFINED), true)
                     if(castedValue.first) {
                         castedValue.second.linkParents(whenChoice)
@@ -479,19 +343,68 @@ class TypecastsAdder(val program: Program, val options: CompilationOptions, val
     }
 
     override fun after(range: RangeExpression, parent: Node): Iterable<IAstModification> {
+        val fromConst = range.from.constValue(program)
+        val toConst = range.to.constValue(program)
+
+        if(fromConst!=null) {
+            val smaller = NumericLiteral.optimalInteger(fromConst.number.toInt(), fromConst.position)
+            if(fromConst.type.largerThan(smaller.type)) {
+                val toType = range.to.inferType(program)
+                if(toType isnot smaller.type) {
+                    if(toConst!=null) {
+                        // can we make the to value into the same smaller type?
+                        val smallerTo = NumericLiteral.optimalInteger(toConst.number.toInt(), toConst.position)
+                        if(smaller.type==smallerTo.type) {
+                            val newRange = RangeExpression(smaller, smallerTo, range.step, range.position)
+                            return listOf(IAstModification.ReplaceNode(range, newRange, parent))
+                        }
+                    }
+                } else {
+                    val newRange = RangeExpression(smaller, range.to, range.step, range.position)
+                    return listOf(IAstModification.ReplaceNode(range, newRange, parent))
+                }
+            }
+        }
+        if(toConst!=null) {
+            val smaller = NumericLiteral.optimalInteger(toConst.number.toInt(), toConst.position)
+            if(toConst.type.largerThan(smaller.type)) {
+                val fromType = range.from.inferType(program)
+                if(fromType isnot smaller.type) {
+                    if(fromConst!=null) {
+                        // can we make the from value into the same smaller type?
+                        val smallerFrom = NumericLiteral.optimalInteger(fromConst.number.toInt(), fromConst.position)
+                        if(smaller.type==smallerFrom.type) {
+                            val newRange = RangeExpression(smallerFrom, smaller, range.step, range.position)
+                            return listOf(IAstModification.ReplaceNode(range, newRange, parent))
+                        }
+                    }
+                } else {
+                    val newRange = RangeExpression(range.from, smaller, range.step, range.position)
+                    return listOf(IAstModification.ReplaceNode(range, newRange, parent))
+                }
+            }
+        }
+
+        val modifications = mutableListOf<IAstModification>()
         val fromDt = range.from.inferType(program).getOr(DataType.UNDEFINED)
         val toDt = range.to.inferType(program).getOr(DataType.UNDEFINED)
-        val modifications = mutableListOf<IAstModification>()
         val (commonDt, toChange) = BinaryExpression.commonDatatype(fromDt, toDt, range.from, range.to)
         if(toChange!=null)
             addTypecastOrCastedValueModification(modifications, toChange, commonDt, range)
+
         return modifications
     }
 
-    override fun after(ifElse: IfElse, parent: Node): Iterable<IAstModification> {
-        if(!options.strictBool && ifElse.condition.inferType(program).isBytes) {
-            val cast = TypecastExpression(ifElse.condition, DataType.BOOL, false, ifElse.condition.position)
-            return listOf(IAstModification.ReplaceNode(ifElse.condition, cast, ifElse))
+    override fun after(arrayIndexedExpression: ArrayIndexedExpression, parent: Node): Iterable<IAstModification> {
+        val constIdx = arrayIndexedExpression.indexer.constIndex()
+        if(constIdx!=null) {
+            val smaller = NumericLiteral.optimalInteger(constIdx, arrayIndexedExpression.indexer.position)
+            val idxDt = arrayIndexedExpression.indexer.indexExpr.inferType(program).getOr(DataType.UNDEFINED)
+            if(idxDt.largerThan(smaller.type)) {
+                val newIdx = ArrayIndex(smaller, smaller.position)
+                val newIndexer = ArrayIndexedExpression(arrayIndexedExpression.arrayvar, newIdx, arrayIndexedExpression.position)
+                return listOf(IAstModification.ReplaceNode(arrayIndexedExpression, newIndexer, parent))
+            }
         }
         return noModifications
     }
@@ -505,18 +418,6 @@ class TypecastsAdder(val program: Program, val options: CompilationOptions, val
         val sourceDt = expressionToCast.inferType(program).getOr(DataType.UNDEFINED)
         if(sourceDt == requiredType)
             return
-        if(!options.strictBool) {
-            if(requiredType==DataType.BOOL && sourceDt in ByteDatatypes) {
-                val cast = TypecastExpression(expressionToCast, DataType.BOOL, false, expressionToCast.position)
-                modifications.add(IAstModification.ReplaceNode(expressionToCast, cast, parent))
-                return
-            }
-            if(requiredType in ByteDatatypes && sourceDt==DataType.BOOL) {
-                val cast = TypecastExpression(expressionToCast, requiredType, false, expressionToCast.position)
-                modifications.add(IAstModification.ReplaceNode(expressionToCast, cast, parent))
-                return
-            }
-        }
         if(requiredType==DataType.BOOL) {
             return
         }

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

@@ -97,16 +97,6 @@ internal class VariousCleanups(val program: Program, val errors: IErrorReporter,
             return listOf(IAstModification.ReplaceNode(expr, expr.expression, parent))
         }
 
-        if(!options.strictBool && expr.operator=="not") {
-            if(expr.expression.inferType(program).isBytes) {
-                // not bytevalue --> bytevalue==0
-                val cmp = BinaryExpression(expr.expression, "==",
-                    NumericLiteral(expr.expression.inferType(program).getOr(DataType.UNDEFINED), 0.0, expr.expression.position),
-                    expr.expression.position)
-                return listOf(IAstModification.ReplaceNode(expr, cmp, parent))
-            }
-        }
-
         return noModifications
     }
 
@@ -145,9 +135,25 @@ internal class VariousCleanups(val program: Program, val errors: IErrorReporter,
                 }
 
                 if(isMultiComparisonRecurse(leftBinExpr1)) {
-                    // replace it!
-                    val valueCopies = values.sortedBy { it.number }.map { it.copy() }
                     val elementType = needle.inferType(program).getOrElse { throw FatalAstException("invalid needle dt") }
+                    if(values.size==2 || values.size==3 && (elementType==DataType.UBYTE || elementType==DataType.UWORD)) {
+                        val numbers = values.map{it.number}.toSet()
+                        if(numbers == setOf(0.0, 1.0)) {
+                            // we can replace unsigned  x==0 or x==1 with x<2
+                            val compare = BinaryExpression(needle, "<", NumericLiteral(elementType, 2.0, expr.position), expr.position)
+                            return listOf(IAstModification.ReplaceNode(expr, compare, parent))
+                        }
+                        if(numbers == setOf(0.0, 1.0, 2.0)) {
+                            // we can replace unsigned  x==0 or x==1 or x==2 with x<3
+                            val compare = BinaryExpression(needle, "<", NumericLiteral(elementType, 3.0, expr.position), expr.position)
+                            return listOf(IAstModification.ReplaceNode(expr, compare, parent))
+                        }
+                    }
+                    if(values.size<2)
+                        return noModifications
+
+                    // replace x==1 or x==2 or x==3  with a containment check  x in [1,2,3]
+                    val valueCopies = values.sortedBy { it.number }.map { it.copy() }
                     val arrayType = ElementToArrayTypes.getValue(elementType)
                     val valuesArray = ArrayLiteral(InferredTypes.InferredType.known(arrayType), valueCopies.toTypedArray(), expr.position)
                     val containment = ContainmentCheck(needle, valuesArray, expr.position)
@@ -277,6 +283,10 @@ internal class VariousCleanups(val program: Program, val errors: IErrorReporter,
             val target = arrayIndexedExpression.arrayvar.targetVarDecl(program)
             val arraysize = target?.arraysize?.constIndex()
             if(arraysize!=null) {
+                if(arraysize+index < 0) {
+                    errors.err("index out of bounds", arrayIndexedExpression.position)
+                    return noModifications
+                }
                 // replace the negative index by the normal index
                 val newIndex = NumericLiteral.optimalNumeric(arraysize+index, arrayIndexedExpression.indexer.position)
                 arrayIndexedExpression.indexer.indexExpr = newIndex

compiler/src/prog8/compiler/astprocessing/VerifyFunctionArgTypes.kt

@@ -41,7 +41,7 @@ internal class VerifyFunctionArgTypes(val program: Program, val options: Compila
     private val memorySlabs = mutableListOf<Slab>()
 
     override fun visit(functionCallExpr: FunctionCallExpression) {
-        val error = checkTypes(functionCallExpr as IFunctionCall, program, options)
+        val error = checkTypes(functionCallExpr as IFunctionCall, program)
         if(error!=null)
             errors.err(error.first, error.second)
         else {
@@ -62,7 +62,7 @@ internal class VerifyFunctionArgTypes(val program: Program, val options: Compila
     }
 
     override fun visit(functionCallStatement: FunctionCallStatement) {
-        val error = checkTypes(functionCallStatement as IFunctionCall, program, options)
+        val error = checkTypes(functionCallStatement as IFunctionCall, program)
         if(error!=null)
             errors.err(error.first, error.second)
 
@@ -85,7 +85,7 @@ internal class VerifyFunctionArgTypes(val program: Program, val options: Compila
             return false
         }
 
-        fun checkTypes(call: IFunctionCall, program: Program, options: CompilationOptions): Pair<String, Position>? {
+        fun checkTypes(call: IFunctionCall, program: Program): Pair<String, Position>? {
             val argITypes = call.args.map { it.inferType(program) }
             val firstUnknownDt = argITypes.indexOfFirst { it.isUnknown }
             if(firstUnknownDt>=0) {
@@ -105,12 +105,7 @@ internal class VerifyFunctionArgTypes(val program: Program, val options: Compila
                 if(mismatch>=0) {
                     val actual = argtypes[mismatch]
                     val expected = consideredParamTypes[mismatch]
-                    return if(expected==DataType.BOOL && !options.strictBool && actual in ByteDatatypes)
-                        null
-                    else if(expected in ByteDatatypes && !options.strictBool && actual==DataType.BOOL)
-                        null
-                    else
-                        Pair("argument ${mismatch + 1} type mismatch, was: $actual expected: $expected", call.args[mismatch].position)
+                    return Pair("argument ${mismatch + 1} type mismatch, was: $actual expected: $expected", call.args[mismatch].position)
                 }
                 if(target.isAsmSubroutine) {
                     if(target.asmReturnvaluesRegisters.size>1) {

compiler/test/ModuleImporterTests.kt

@@ -1,4 +1,4 @@
-package prog8tests
+package prog8tests.compiler
 
 import com.github.michaelbull.result.getErrorOrElse
 import com.github.michaelbull.result.getOrElse

compiler/test/ProjectConfig.kt

@@ -1,4 +1,4 @@
-package prog8tests
+package prog8tests.compiler
 
 import io.kotest.core.config.AbstractProjectConfig
 import io.kotest.core.spec.SpecExecutionOrder

compiler/test/TestBuiltinFunctions.kt

@@ -1,4 +1,4 @@
-package prog8tests
+package prog8tests.compiler
 
 import io.kotest.core.spec.style.FunSpec
 import io.kotest.matchers.shouldBe

compiler/test/TestCallgraph.kt

@@ -1,4 +1,4 @@
-package prog8tests
+package prog8tests.compiler
 
 import io.kotest.assertions.withClue
 import io.kotest.core.spec.style.FunSpec

compiler/test/TestCompilerOnCharLit.kt

@@ -1,4 +1,4 @@
-package prog8tests
+package prog8tests.compiler
 
 import io.kotest.assertions.fail
 import io.kotest.assertions.withClue