Enable runtime-async feature switch for NativeAOT System.Private.CoreLib

[Copilot]

Description

Enables the runtime-async compiler feature for the core libraries when building for Native AOT.

Changes

• Added <Features>$(Features);runtime-async=on</Features> to NativeAOT System.Private.CoreLib csproj
• Suppressed CA2252 (preview features usage) since runtime-async=on causes the compiler to inject AsyncHelpers.Await calls which are marked [RequiresPreviewFeatures]

<!-- Enable runtime async for Native AOT -->
<PropertyGroup>
<Features>$(Features);runtime-async=on</Features>
<NoWarn>$(NoWarn);CA2252</NoWarn>
</PropertyGroup>

Notes

The CA2252 suppression is project-scoped because runtime-async=on transforms all async methods via compiler-generated calls—individual method suppression isn't practical or meaningful here. This is consistent with how other runtime-async-enabled projects handle it (e.g., tests disable analyzers entirely).

Original prompt

I'd like to enable the runtimeasync feature switch for the core libraries if and only if we're building for native aot. I think we might need to plumb down another msbuild switch.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[agocke]

@MichalStrehovsky Some interesting failures in loongarch and riscv here. Is there work we have to do for those platforms?

[MichalStrehovsky]

@MichalStrehovsky Some interesting failures in loongarch and riscv here. Is there work we have to do for those platforms?

I think this is a JIT assert, we likely haven't done any work on Risc/Loong (we certainly haven't done things like hijacking either and there's a tracking issue somewhere).

I'd just not enable runtime-async corelib on those two architectures for now.

Btw, we also have a private copy of the framework assemblies, so we could theoretically also enable runtime-async when building native AOT runtime packs.

[MichalStrehovsky]

Btw, we also have a private copy of the framework assemblies, so we could theoretically also enable runtime-async when building native AOT runtime packs.

Ah, maybe we stopped building those in a separate leg with source build. So maybe can't do that one anymore.

[agocke]

I excluded loongarch and riscv and that seems to have worked

[Copilot]

Pull request overview

This PR enables the runtime-async compiler feature for NativeAOT's System.Private.CoreLib by adding the Features=runtime-async=on property in the csproj file. The feature switch transforms async methods to use the runtime's native async support instead of compiler-generated state machines, improving performance. The PR also suppresses CA2252 warnings that arise from compiler-generated calls to AsyncHelpers.Await methods marked with preview feature attributes. The feature is currently disabled for riscv64 and loongarch64 architectures.

Changes:

• Enabled runtime-async=on feature flag for NativeAOT System.Private.CoreLib (excluding riscv64/loongarch64)
• Suppressed CA2252 analyzer warning for preview feature usage at the project level

[jkotas]

we certainly haven't done things like hijacking either and there's a tracking issue somewhere

I am not able to find it. If there is not one, we should create it (one per arch) and link it from this change

[MichalStrehovsky]

we certainly haven't done things like hijacking either and there's a tracking issue somewhere

I am not able to find it. If there is not one, we should create it (one per arch) and link it from this change

It's #122492. But we're disabling runtimeasync on LoongArch/RiscV because of RyuJIT asserts when compiling. We don't even get to hijacking.