False positive IL2026 on MakeGenericType/Method when there are no annotations/constraints

[roji]

When using MakeGenericType/MakeGenericMethod, IL2026 is understandably generated since some generic type parameters may be annotated with DynamicallyAccessedMemberAttribute, but that can't be taken into account by the linker when the type/method isn't known at compile-time.

However, when the type/method is known at compile-time, it seems like the linker could verify whether annotations exist on generic type parameters, and not emit the warning if not:

_ = typeof(Foo<>).MakeGenericType(typeof(string));

class Foo<T>
{
}

From other issues in this repo, it seems that improved support for MakeGenericType/MakeGenericMethod is maybe already in the works - feel free to close this issue if this is already planned etc.

[marek-safar]

/cc @vitek-karas

[vitek-karas]

This is analyzer-only problem. Meaning the warning is only generated by the roslyn analyzer either in IDE or build. It should NOT be generated by the linker when trimming.

#2209 fixes the analyzer to not issue the warning, but that fix didn't make it into the SDK yet. It's in the 6.0.2xx branch in dotnet/linker, so hopefully it should make it to 6.0.200 SDK eventually.

[roji]

@vitek-karas I can confirm that setting <EnableTrimAnalyzer>false</EnableTrimAnalyzer> makes this go away, so that corresponds to what you wrote, thanks!