Cleanup some code under FEATURE_STUBS_AS_IL

[huoyaoyuan]

There's no actually no IL stubs under this feature switch now, only asm stubs.

This PR addresses the simple cases. On all platforms except Windows x86, single cast delegate is written in invariant assembly code.

Remaining code under the feature switch is COM call stub, which is more complicated and is better separated out.

[jkotas]

Code quality

If this helps with code quality, could you please make the same change in

runtime/src/coreclr/tools/Common/TypeSystem/IL/Stubs/DelegateMethodILEmitter.cs

Lines 59 to 60 in 3f607bd

	codeStream.EmitLdArg(0);
	codeStream.Emit(ILOpcode.ldfld, emit.NewToken(functionPointerField.InstantiateAsOpen()));

as well? It is naot equivalent of this stub.

[huoyaoyuan]

The same tests in DI are still failing. Can it be related to the signature instantiation in ILC?

runtime/src/coreclr/tools/Common/TypeSystem/IL/Stubs/DelegateMethodILEmitter.cs

Lines 63 to 81 in 3f607bd

	if (method.OwningType.HasInstantiation)
	{
	// If the owning type is generic, the signature will contain T's and U's.
	// We need !0's and !1's.
	TypeDesc[] typesToReplace = new TypeDesc[method.OwningType.Instantiation.Length];
	TypeDesc[] replacementTypes = new TypeDesc[typesToReplace.Length];
	for (int i = 0; i < typesToReplace.Length; i++)
	{
	typesToReplace[i] = method.OwningType.Instantiation[i];
	replacementTypes[i] = context.GetSignatureVariable(i, method: false);
	}
	TypeDesc[] parameters = new TypeDesc[method.Signature.Length];
	for (int i = 0; i < parameters.Length; i++)
	{
	parameters[i] = method.Signature[i].ReplaceTypesInConstructionOfType(typesToReplace, replacementTypes);
	}
	TypeDesc returnType = method.Signature.ReturnType.ReplaceTypesInConstructionOfType(typesToReplace, replacementTypes);
	signature = new MethodSignature(signature.Flags, signature.GenericParameterCount, returnType, parameters);
	}

Is it also required for coreclr?

[jkotas]

The same tests in DI are still failing. Can it be related to the signature instantiation in ILC?

We should not need a special signature instantiation like that in CoreCLR. I suspect that it is related to a special delegate calling convention used for delegate Invocation in some cases. The delegate target cannot be called via pointer - it must be called as Delegate.Invoke method so that the special calling convention kicks in.

Do you know if the Invoke method is used in the DI on a performance critical path?

This is proving to be much harder than what I thought it is going to be.

[jkotas]

I have pushed a commit with a small change to handle the new IL stub in stub manager. The manual debugger test from #104731 (comment) works fine now.

[jkotas]

LGTM. Thank you!

[jkotas]

Remaining code under the feature switch is COM call stub, which is more complicated and is better separated out.

Do you plan to work on cleaning up the remaining code under the feature switch?

[huoyaoyuan]

Do you plan to work on cleaning up the remaining code under the feature switch?

Aaron suggested me to avoid converting COM interop related code because the effort needed for reviewing. I'm already busy now until October.

If you can confirm the review effort on behalf of interop team, I can start to investigate this when I finish urgent tasks.

[jkotas]

I gave it a shot in #108048.