diff --git a/codeGenCpu6502/src/prog8/codegen/cpu6502/assignment/AssignmentAsmGen.kt b/codeGenCpu6502/src/prog8/codegen/cpu6502/assignment/AssignmentAsmGen.kt
index 6f67bd486..2ed2cdff2 100644
--- a/codeGenCpu6502/src/prog8/codegen/cpu6502/assignment/AssignmentAsmGen.kt
+++ b/codeGenCpu6502/src/prog8/codegen/cpu6502/assignment/AssignmentAsmGen.kt
@@ -344,8 +344,10 @@ internal class AssignmentAsmGen(private val program: Program,
                     require(elementDt.isBytes)
                     val stringVal = variable.value as StringLiteral
                     val varname = asmgen.asmVariableName(containment.iterable as IdentifierReference)
-                    assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, DataType.UWORD, containment.definingSubroutine, "P8ZP_SCRATCH_W1"), varname)
                     assignExpressionToRegister(containment.element, RegisterOrPair.A, elementDt istype DataType.BYTE)
+                    asmgen.saveRegisterLocal(CpuRegister.A, containment.definingSubroutine!!)
+                    assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, DataType.UWORD, containment.definingSubroutine, "P8ZP_SCRATCH_W1"), varname)
+                    asmgen.restoreRegisterLocal(CpuRegister.A)
                     asmgen.out("  ldy  #${stringVal.value.length}")
                     asmgen.out("  jsr  prog8_lib.containment_bytearray")
                     return
@@ -361,8 +363,10 @@ internal class AssignmentAsmGen(private val program: Program,
                     val varname = asmgen.asmVariableName(containment.iterable as IdentifierReference)
                     when(dt) {
                         in ByteDatatypes -> {
-                            assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, DataType.UWORD, containment.definingSubroutine, "P8ZP_SCRATCH_W1"), varname)
                             assignExpressionToRegister(containment.element, RegisterOrPair.A, elementDt istype DataType.BYTE)
+                            asmgen.saveRegisterLocal(CpuRegister.A, containment.definingSubroutine!!)
+                            assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, DataType.UWORD, containment.definingSubroutine, "P8ZP_SCRATCH_W1"), varname)
+                            asmgen.restoreRegisterLocal(CpuRegister.A)
                             asmgen.out("  ldy  #${arrayVal.value.size}")
                             asmgen.out("  jsr  prog8_lib.containment_bytearray")
                         }
@@ -383,8 +387,10 @@ internal class AssignmentAsmGen(private val program: Program,
         when(variable.datatype) {
             DataType.STR -> {
                 // use subroutine
-                assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, DataType.UWORD, containment.definingSubroutine, "P8ZP_SCRATCH_W1"), varname)
                 assignExpressionToRegister(containment.element, RegisterOrPair.A, elementDt istype DataType.BYTE)
+                asmgen.saveRegisterLocal(CpuRegister.A, containment.definingSubroutine!!)
+                assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, DataType.UWORD, containment.definingSubroutine, "P8ZP_SCRATCH_W1"), varname)
+                asmgen.restoreRegisterLocal(CpuRegister.A)
                 val stringVal = variable.value as StringLiteral
                 asmgen.out("  ldy  #${stringVal.value.length}")
                 asmgen.out("  jsr  prog8_lib.containment_bytearray")
@@ -393,8 +399,10 @@ internal class AssignmentAsmGen(private val program: Program,
             DataType.ARRAY_F -> throw AssemblyError("containment check of floats not supported")
             DataType.ARRAY_B, DataType.ARRAY_UB -> {
                 val arrayVal = variable.value as ArrayLiteral
-                assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, DataType.UWORD, containment.definingSubroutine, "P8ZP_SCRATCH_W1"), varname)
                 assignExpressionToRegister(containment.element, RegisterOrPair.A, elementDt istype DataType.BYTE)
+                asmgen.saveRegisterLocal(CpuRegister.A, containment.definingSubroutine!!)
+                assignAddressOf(AsmAssignTarget(TargetStorageKind.VARIABLE, program, asmgen, DataType.UWORD, containment.definingSubroutine, "P8ZP_SCRATCH_W1"), varname)
+                asmgen.restoreRegisterLocal(CpuRegister.A)
                 asmgen.out("  ldy  #${arrayVal.value.size}")
                 asmgen.out("  jsr  prog8_lib.containment_bytearray")
                 return
diff --git a/docs/source/todo.rst b/docs/source/todo.rst
index 3f9f6ab9a..1e61518bf 100644
--- a/docs/source/todo.rst
+++ b/docs/source/todo.rst
@@ -3,7 +3,6 @@ TODO
 
 For next release
 ^^^^^^^^^^^^^^^^
-- fix assembler application included file loading problem (hello4.asm)
 - x16: add new keyboard APIs https://github.com/commanderx16/x16-docs/blob/master/Commander%20X16%20Programmer%27s%20Reference%20Guide.md#keyboard
 - x16: optimize diskio load_raw because headerless files are now supported https://github.com/commanderx16/x16-rom/pull/216
   note: must still work on c64/c128 that don't have this!
diff --git a/examples/test.p8 b/examples/test.p8
index d0b311b36..cba0c4597 100644
--- a/examples/test.p8
+++ b/examples/test.p8
@@ -4,69 +4,30 @@
 ; NOTE: meant to test to virtual machine output target (use -target vitual)
 
 main {
+
     sub start() {
-        txt.clear_screen()
-
-        txt.setcc2(10,3,'*',$12)
-        txt.setcc2(11,3,'*',$02)
-        txt.setcc2(12,3,'*',$10)
-        txt.setcc2(10,4,'*',$d5)
-        txt.setcc2(11,4,'*',$05)
-        txt.setcc2(12,4,'*',$d0)
-
-        sys.wait(100)
-        cx16.screen_mode(0, false)
-        sys.wait(100)
-
-
-;        byte[] barr = [-1,-2,-3]
-;        uword[] uwarr = [1111,2222,3333]
-;
-;        txt.print_b(barr[2])
-;        txt.spc()
-;        txt.print_uw(uwarr[2])
-;        txt.nl()
-;
-;        barr[2] --
-;        uwarr[2] --
-;
-;        txt.print_b(barr[2])
-;        txt.spc()
-;        txt.print_uw(uwarr[2])
-;        txt.nl()
-;
-;        barr[2] ++
-;        uwarr[2] ++
-;
-;        txt.print_b(barr[2])
-;        txt.spc()
-;        txt.print_uw(uwarr[2])
-;        txt.nl()
-;
-;        sys.exit(99)
-
 
         ; the "pixelshader":
-;        syscall1(8, 0)      ; enable lo res creen
-;        ubyte shifter
-;
-;        shifter >>= 1
-;
-;        repeat {
-;            uword xx
-;            uword yy = 0
-;            repeat 240 {
-;                xx = 0
-;                repeat 320 {
-;                    syscall3(10, xx, yy, xx*yy + shifter)   ; plot pixel
-;                    xx++
-;                }
-;                yy++
-;            }
-;            shifter+=4
-;
-;            txt.print_ub(shifter)
-;            txt.nl()
-;        }
+        syscall1(8, 0)      ; enable lo res creen
+        ubyte shifter
+
+        shifter >>= 1
+
+        repeat {
+            uword xx
+            uword yy = 0
+            repeat 240 {
+                xx = 0
+                repeat 320 {
+                    syscall3(10, xx, yy, xx*yy + shifter)   ; plot pixel
+                    xx++
+                }
+                yy++
+            }
+            shifter+=4
+
+            txt.print_ub(shifter)
+            txt.nl()
+        }
     }
 }