Cache the delegate for static method group conversions.

[pawchen]

This is the new attempt of #6642:

1. Dropped module scoped containers to reduce complexities, by not dealing with module builders, and utilize the existing SynthesizedContainer.
2. The target method can be local functions marked as static.
3. Caching is enabled for C# preview version.
4. Probably need more tests.

Considering the complexities currently we have in this PR, we leave the following interesting ideas/improvements for future considerations:

• Improve handling for scenarios where the enclosing method is generic and the target method is a local function
• Adding type scoped generic containers grouped by the arity
• Take the branching op out of loops
• Use fields instead of containers
• Adding type scoped generic containers grouped by the arity
• Pulling the cache into the containing type
• Module scoped cache

Fixes #5835

[AlekseyTs]

@pawchen

Should we look for generic attributes? It's new to me, but probably no need?

What aspect do you have in mind?

[pawchen]

I see that now we have several tests targeting RegularNext version, but we still don't have any test in this file specifically verifying that caching is disabled for Regular10. I think we need at least one test like that.

In reply to: 1007055636

Refers to: src/Compilers/CSharp/Test/Emit2/CodeGen/CodeGenMethodGroupConversionCachingTests.cs:5293 in 9ca84f4. [](commit_id = 9ca84f4, deletion_comment = False)

Fair, added in 28ffe27. Please check.

[pawchen]

    G(Target<int>);

It would be good to test that we can share the cache field across multiple functions.

Do we have a test like that?

In reply to: 1007099643

Refers to: src/Compilers/CSharp/Test/Emit2/CodeGen/CodeGenMethodGroupConversionCachingTests.cs:4220 in 9ca84f4. [](commit_id = 9ca84f4, deletion_comment = False)

Also added in 28ffe27, please check.

[jcouv]

LGTM Thanks (iteration 72)

[AlekseyTs]

LGTM (commit 72)

[AlekseyTs]

@pawchen I think now it is a good time to rename files and do any other changes you would like to do before we merge this PR. Let us know when you are ready and we squash-merge this PR.

[pawchen]

@AlekseyTs Thanks, I apologize for renaming files in the middle. I think I changed my mind, the original DelegateCacheRewriter is actually good as it has Cache in its name, so I renamed the class instead of the file in 84551f1.

[AlekseyTs]

LGTM (commit 73)

[AlekseyTs]

@jcouv Any additional feedback?

[AlekseyTs]

@pawchen Are we good to merge from your point of view?

[pawchen]

Yes, we have possible future improvements documented and an issue of Expression Compiler filed. Test cases checked multiple times. Let's get this into preview early, in case we missed any edge cases we could fix them ASAP. Thank you for all the great feedback and guidance!

[AlekseyTs]

@pawchen Thank you for the contribution!