[ILLink analyzer] Move generic parameter analysis to dataflow

[sbomer]

And align behavior more closely with NativeAot. I tried to roughly follow the implementation approach used by NativeAot in GenericArgumentDataFlow.cs to make it easier to share the logic (without actually sharing the code for now).

Fixes #95121.

Some things to note:

• Gets rid of some analyzer warnings in cases where NativeAot doesn't warn either
• Doesn't warn for generics in typeof
• Doesn't warn for generics in signatures of reflectable members (ILC does this to work around an incorrect suppression in DI, as discussed in NativeAOT Data flow annotations are not applied to DI created generic types #81358).
• Generics in base types or interface types are still analyzed outside of dataflow

[ghost]

Tagging subscribers to 'linkable-framework': @eerhardt, @vitek-karas, @LakshanF, @sbomer, @joperezr, @marek-safar
See info in area-owners.md if you want to be subscribed.

Issue Details

---

And align behavior more closely with NativeAot.

Fixes #95121.

Some things to note:

• Gets rid of some analyzer warnings in cases where NativeAot doesn't warn either
• Doesn't warn for generics in typeof
• Doesn't warn for generics in signatures of reflectable members (ILC does this to work around an incorrect suppression in DI, as discussed in NativeAOT Data flow annotations are not applied to DI created generic types #81358).
• Generics in base types or interface types are still analyzed outside of dataflow

Author:	sbomer
Assignees:	-
Labels:	linkable-framework
Milestone:	-

[vitek-karas]

Nice!

[ShreyasJejurkar]

This is great. I just had one question can't we share the trimming logic between the trimmer, NativeAOT, and the rest of the app models where we do trimming? This is hard to do or is it technically impossible to do so?

[sbomer]

We've been making progress on sharing some of the code - see https://github.com/dotnet/runtime/tree/main/src/tools/illink/src/ILLink.Shared. The three tools use three different type systems, so a big part of the sharing effort is using wrapper types that hide some of the type system differences.

[vitek-karas]

The additional complication is that trimmer (illink) and NativeAOT see IL, while the analyzer sees the Roslyn's operations/CFG. We could probably share more if(when) we implement CFG reconstruction from IL in trimmer and NativeAOT. But currently the two approaches have to do some things differently.

That said, we already share quite a lot of the core data flow logic, and almost all of the tests.