[RuntimeAsync] ilasm/ildasm support for the MethodImpl.Async

[hez2010]

Implementing ilasm and ildasm support for RuntimeAsync following the spec.

Closes#115093

cc: @VSadov

[Copilot]

Pull Request Overview

This pull request adds support for the "async" method implementation attribute to both ilasm and ildasm, in accordance with the RuntimeAsync specification. The changes include:

• In il_kywd.h, a new keyword "async" is declared.
• In dasm.cpp, the disassembler now checks for the async attribute and prints " async" in the output.

Reviewed Changes

Copilot reviewed 3 out of 5 changed files in this pull request and generated no comments.

File	Description
src/coreclr/inc/il_kywd.h	Added the "async" keyword definition to support async methods.
src/coreclr/ildasm/dasm.cpp	Updated method dump logic to include "async" when the attribute is active.

Files not reviewed (2)

• src/coreclr/ilasm/asmparse.y: Language not supported
• src/coreclr/ilasm/prebuilt/asmparse.grammar: Language not supported

[huoyaoyuan]

Combining MethodImplOptions.Async with MethodImplOptions.Synchronized is invalid.
Applying MethodImplOptions.Async to methods with a byref or ref-like return value is invalid.
Applying MethodImplOptions.Async to vararg methods is invalid.

Does ilasm do this sort of check?

[hez2010]

Does ilasm do this sort of check?

I don't think so. We don't have any of such check in ilasm, and it just emits whatever you write.
For example, today ilasm will happily take a byreflike type for a generic parameter that doesn't accept byreflike types. Such checks are done at the runtime.

[jkotas]

I don't think so. We don't have any of such check in ilasm, and it just emits whatever you write.

Right, it is a by-design feature that ilasm can produce invalid IL.

[VSadov]

This looks like a fairly mechanical change. I am not very familiar with ilasm/ildasm, but supporting async looks very similar to supporting synchronized.

I am wondering - are there any tests for ilasm/ildasm?

[hez2010]

I am wondering - are there any tests for ilasm/ildasm?

almost no. the only ilasm test in https://github.com/dotnet/runtime/tree/main/src/tests/ilasm is to test MethodImpl.AggressiveOptimization. and there's nothing for ildasm.

[jkotas]

We test ilasm/ildasm by roundtripping all runtime test binaries through ildasm/ilasm, and verifying that the result still works. So the test coverage for this will come online as side-effect of having any tests for async.

[VSadov]

LGTM. Thanks!

[jkotas]

This introduced regression dotnet/dotnet#544 (comment) that is breaking the dotnet/dotnet build.

Reverting in #115621 until we figure out what to do about this.

[hez2010]

Seems that we cannot make async a keyword in IL as it may be used as an identifier.
I'll work on a fix by making async a non-keyword in IL.

[VSadov]

I wonder if we could just treat it as numerical value if ilasm can take methodimpl as a number.
It would not be pretty, but might be ok for IL