[release/8.0.1xx] Don't require AppHost when publishing AOT

[github-actions]

Backport of #38644 to release/8.0.1xx

/cc @ilonatommy @MichalStrehovsky

Customer Impact

• Customer reported
• Found internally

Backport of #38644. In order to fixe issue dotnet/runtime#99423 that has OOM failures in NativeAOT lanes on net8 branch, we have to backport this one and dotnet/runtime#101270. Runtime's PR backport fails without the change in SDK from this PR.

Regression

• Yes
• No

Testing

CI.

Risk

Low, it's only adding an opt-in option to build without host.

[jkotas]

Could you please fill in servicing template? Why does this need backporting?

[ilonatommy]

@jkotas, updated

[MichalStrehovsky]

This was a set of changes and I'm not sure what the dependencies between them were. I cannot comment on the risk, just feel this is risky.

Can you help me understand how this is related to dotnet/runtime#99423? The issue is about browser-wasm linux Release AllSubsets_Mono_RuntimeTests monointerpreter. We are not supposed to build any of native AOT in that queue. Unfortunately the build logs in the top post that started the issue already expired so I cannot see whether native AOT was getting built there.

[ilonatommy]

I don't follow why it's marked with that name, all the logs pinned to the issue (all from the table that are non-expired) use:
__w/1/s/build.sh -ci -arch x64 -os linux -cross -s clr.aot+host.native+libs+libs.tests -c Release -rc Checked /p:TestNativeAot=true /p:ArchiveTests=true /p:OptimizationPreference=Size . Is this not a nativeaot build command?

[MichalStrehovsky]

all the logs pinned to the issue (all from the table are non-expired) use:
__w/1/s/build.sh -ci -arch x64 -os linux -cross -s clr.aot+host.native+libs+libs.tests -c Release -rc Checked /p:TestNativeAot=true /p:ArchiveTests=true /p:OptimizationPreference=Size

The issue is conflating all docker OOM failures into one. We only have 1 month of history.

The native AOT legs issues would likely be fixed by backporting dotnet/runtime#95896.

Not sure about the Mono AOT issues.

It is not true that all of the logs in the table are for native AOT. I can still see some that are likely Mono - all of those that have "Step Name" = "Build Tests" in the table are likely the Mono OOM, they just have expired logs. Here is the only one that has Step name == build tests and is not yet expired. It is for the wasm interpreter:

Report

Build	Definition	Step Name	Console log	Pull Request
739573	dotnet/runtime	Build Tests	Log

[ilonatommy]

Thanks, I missed that one. Let's close this backports then. If you don't mind, I would like to backport the GC one. The remaining lanes will need more investigation, we have a similar issue opened long time ago that was trying to increase the memory but in the end, did not: dotnet/runtime#60234.

[MichalStrehovsky]

If you don't mind, I would like to backport the GC one.

That sounds good to me, thanks! That's a test-build-only change so it's probably just a tell mode only change and will help with quite a few hits in the OOM issue.