Hi!, When generating code with IR which seems ok in 1.3.61 it produces binaries that when loaded at...

# compiler

[raulraja]

Hi!,

When generating code with IR which seems ok in 1.3.61 it produces binaries that when loaded at runtime result in this kinds of exceptions related to kotlin.internal.ir.Intrinsic . I suppose this is an issue in code generation and not that I actually need an IR runtime to run that code and it’s missing as dependency. Is that the case? Or does anyone else know where this may be coming from? Any help is appreciated, thanks!

Exception in thread "main" java.lang.NoClassDefFoundError: kotlin/internal/ir/Intrinsic
	at consumer.MainKt.<clinit>(main.kt:5)
Caused by: java.lang.ClassNotFoundException: kotlin.internal.ir.Intrinsic
	at <http://java.net|java.net>.URLClassLoader.findClass(URLClassLoader.java:382)
	at java.lang.ClassLoader.loadClass(ClassLoader.java:424)
	at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:349)
	at java.lang.ClassLoader.loadClass(ClassLoader.java:357)
	... 1 more

[shikasd]

I remember seeing something like that when I tried to generate a function directly in an IrFile some time ago. Maybe it is related.

[raulraja]

I checked and we are not generating functions just inserting calls which are resolved in IR correctly

[shikasd]

@raulraja Do you know which calls got replaced with intrinsic?
From bytecode for example

[raulraja]

Source.kt

[raulraja]

IR Tree

[raulraja]

In the case where it works it generates :

2: invokestatic  #48                 // Method twitterHandle:(Ljava/lang/String;)Ltest/TwitterHandle;

[raulraja]

And in the case it doesn’t it generates:

[raulraja]

2: invokestatic  #50                 // Method kotlin/internal/ir/Intrinsic.twitterHandle:(Ljava/lang/String;)Ltest/TwitterHandle;

[raulraja]

so it thinks the twitterHandle functions is inside a class called kotlin/internal/ir/Intrinsic

[raulraja]

But the IR tree is the same in both cases

[raulraja]

PROPERTY name:x visibility:public modality:FINAL [val]
    FIELD PROPERTY_BACKING_FIELD name:x type:test.TwitterHandle? visibility:private [final,static]
      EXPRESSION_BODY
        CALL 'public final fun twitterHandle (): test.TwitterHandle? declared in test' type=test.TwitterHandle? origin=null
          $receiver: CONST String type=kotlin.String value="@admin"

[raulraja]

I think I’m missing some button I need to push but I’m weirded out that given the same IR tree it generates different bytecode

[shikasd]

I did a quick research on that yesterday, so what I found is:

companion object {
    val FAKE_OWNER_TYPE = Type.getObjectType("kotlin/internal/ir/Intrinsic")
}

No other mentions of that guy anywhere else

[shikasd]

Essentially it cannot find a class which is a method owner, maybe you are referencing the function before it gets wrapped into class?

[shikasd]

Also run a debugger on your example and it gets into this branch.
Apparently the method you are referencing still has PackageFragment as a parent 🤔

fun mapToCallableMethod(expression: IrFunctionAccessExpression): IrCallableMethod {
        val callee = expression.symbol.owner.getOrCreateSuspendFunctionViewIfNeeded(context)
        val calleeParent = callee.parent
        if (calleeParent !is IrClass) {
            // Non-class parent is only possible for intrinsics created in IrBuiltIns, such as dataClassArrayMemberHashCode. In that case,
            // we still need to return some IrCallableMethod with some owner instance, but that owner will be ignored at the call site.
            // Here we return a fake type, but this needs to be refactored so that we never call mapToCallableMethod on intrinsics.
            // TODO: get rid of fake owner here
            return IrCallableMethod(FAKE_OWNER_TYPE, Opcodes.INVOKESTATIC, mapSignatureSkipGeneric(callee), false)
        }

[raulraja]

the method is a global extension function

[raulraja]

so it’s parent is the PackageFragment

[raulraja]

Perhaps I’m missing something in the way I replace the call

[raulraja]

https://github.com/arrow-kt/arrow-meta/blob/56f28f0757ba4c4a1ca7eff94e2f43aa96841428/compiler-plugin/src/main/kotlin/arrow/meta/phases/codegen/ir/IrUtils.kt#L43-L54

[shikasd]

Try this, it fails with function is not declared. Apparently, it is not in the symbol table?

fun FunctionDescriptor.irCall(): IrCall {
        val irFunctionSymbol = backendContext.ir.symbols.externalSymbolTable.referenceDeclaredFunction(this)
        return IrCallImpl(
            startOffset = UNDEFINED_OFFSET,
            endOffset = UNDEFINED_OFFSET,
            type = irFunctionSymbol.owner.returnType,
            symbol = irFunctionSymbol,
            descriptor = irFunctionSymbol.descriptor,
            typeArgumentsCount = irFunctionSymbol.owner.descriptor.typeParameters.size,
            valueArgumentsCount = irFunctionSymbol.owner.descriptor.valueParameters.size
        )
    }

[shikasd]

So its parent should be the fragment on your stage, but then in jvmPhases, lowering replaces function parent for generation

[shikasd]

FileClassLowering or something like that`

[shikasd]

@raulraja Ok, apparently descriptor is there, but the instance is different from you have.

[shikasd]

I removed synthetic() in

fun List<Proof>.synthetic(): List<Proof> =
  mapNotNull { proof ->
    Proof(proof.from, <http://proof.to|proof.to>, (proof.through as SimpleFunctionDescriptor).synthetic(), proof.proofType)
  }

and it works 🙂

[raulraja]

Awesome!

[raulraja]

Thanks so much!

[raulraja]

I owe you a 🍺 when travelling is back allowed !

[raulraja]

I wonder if those unbounded symbols I was getting in 1.3.7 are related, I'll take a look at it tomorrow. Thanks again!

[shikasd]

My pleasure 🍻