Fix mapping of redeclared type parameters

[sbomer]

Ported over from dotnet/linker#3167, and fixed in ILCompiler as well.

We were skipping the type parameter mapping of compiler-generated types whose only generic parameters were those implicitly created for the declaring type's type parameters.

For the testcase in question, the nested state machine inherited a generic parameter from the display class. This was causing unnecessary warnings in a field assignment that assigned this (an instance of the display class) to a field on the state machine. In IL, that assignment references a field type like DisplayClass<T> where T is the generic parameter on the state machine. Here we were properly mapping type parameters of the display class back to the annotated enclosing method's type parameters, so we could tell that the "target" required PublicMethods. But the substituted T from the state machine was not mapped, causing the mismatch.

Fixes the issue mentioned in dotnet/roslyn#46646 (comment).

[MichalStrehovsky]

I'll leave reviewing this to Vitek, but I have a general question:

When ILLink was in a separate repo, the approach we had was that things got fixed in ILLink repo and at some point we'd go, copy the files from linker repo over to ReferenceSource directories in this repo and apply the diffs manually.

ILLink is now here so it's feasible to just update these in one go, but we didn't fully catch up with other changes. Changing things now will make ReferenceSource diffs harder to apply.

The question I have is whether we'd want to prioritize catching up and deleting ReferenceSource.

[ghost]

Tagging subscribers to this area: @agocke, @sbomer, @vitek-karas
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Ported over from dotnet/linker#3167, and fixed in ILCompiler as well.

We were skipping the type parameter mapping of compiler-generated types whose only generic parameters were those implicitly created for the declaring type's type parameters.

For the testcase in question, the nested state machine inherited a generic parameter from the display class. This was causing unnecessary warnings in a field assignment that assigned this (an instance of the display class) to a field on the state machine. In IL, that assignment references a field type like DisplayClass<T> where T is the generic parameter on the state machine. Here we were properly mapping type parameters of the display class back to the annotated enclosing method's type parameters, so we could tell that the "target" required PublicMethods. But the substituted T from the state machine was not mapped, causing the mismatch.

Fixes the issue mentioned in dotnet/roslyn#46646 (comment).

Author:	sbomer
Assignees:	sbomer
Labels:	area-Tools-ILLink, needs-area-label
Milestone:	-

[sbomer]

Seems like a good idea to me. I can work on this if @vitek-karas hasn't already started.

[vitek-karas]

The source code should be very close for the most part. I did a sync before I started working on test bringup.
There might be some small refactorings which are present only in one and not the other.

As we do this I would like to think about sharing the code for real. This will obviously require some additional effort, but currently the amount of code which needs to be synced is pretty big. It's definitely good that it looks similar, but true sharing would be even better.

[vitek-karas]

[sbomer]

Failures are #80619 and #84995