Add handling of generic attributes to CustomAttributeDecoder

[MichalStrehovsky]

If the attribute constructor refers to the generic T in its signature, we would throw BadImageFormatException. This adds handling by capturing the generic context and interpreting the signature variables when needed.

I've hit this in the src\tests\reflection\GenericAttribute tests with NativeAOT since the NativeAOT compiler uses S.R.Metadata to read custom attributes.

No tests yet because I would like to discuss testing strategy.

• There's notes about pre-existing test debt in CustomAttributeDecoderTests.
• I don't want to write an IL test for this because they're a maintenance nightmare.
• We crosstarget (and test?) on .NET Framework. I don't know the support plans for generic attributes when targeting net461 with Roslyn. I don't know the support status of generic attributes in this repo.

I tested it with the src\tests\reflection\GenericAttribute test and NativeAOT, but that test is not much more than a smoke test, really.

Cc @tmat @RikkiGibson

[tmat]

    private ArgumentTypeInfo DecodeNamedArgumentType(ref BlobReader valueReader, bool isElementType = false)

Named argument can't by of generic type argument type?

class A<S, T>
{
    S[] P { get; }
    A(T arg) {} 
}

[A<int, bool>(1, P = new[] { true })]
class C {}

---

Refers to: src/libraries/System.Reflection.Metadata/src/System/Reflection/Metadata/Ecma335/CustomAttributeDecoder.cs:230 in 5204378. [](commit_id = 5204378, deletion_comment = False)

[tmat]

We should definitely add unit tests for this.

You can use MetadataBuilder to construct the metadata image in the test that has generic attributes and then read them back.

[MichalStrehovsky]

Named argument can't by of generic type argument type?

Named arguments encode their type inline - we don't need to resolve the members to find their signature.

[tmat]

Named arguments encode their type inline - we don't need to resolve the members to find their signature.

So DecodeNamedArgumentType does not need to handle SignatureTypeCode.GenericTypeParameter? Seems odd

Never mind, the types used there will be specific instantiations, not generic type parameters.

[krwq]

consider also checking for negative parameterIndex and numGenericParameters

[krwq]

isn't there any existing subroutine for checking simple types?

[krwq]

considering it's not valid in C# to create generic attributes I'm ok with no extra tests and just unblocking scenarios for problematic assemblies. If they're ever supported then we should add more tests and fix any rough edges if there are any

[tmat]

@krwq Generic attributes are going to be supported in C# 10 or 11. We need tests.

[RikkiGibson]

Generic attributes are shipping in preview in .NET 6/C# 10, and will hopefully ship stable in C# 11.

[MichalStrehovsky]

I thought I would be able to get to the tests, but it probably won't happen anytime soon. I filed #58073. It's better if it's handled by someone else. The fix I have works and I can just fork the file where I need this. I have test coverage through other CoreCLR tests from that.

[krwq]

@tmat @RikkiGibson if that's the case then agreed we need tests