NativeAOT-LLVM: Review decision to include pthreads in WASI build and Wait

[yowl]

I've been regretting my decision to use pthreads in WASI, so was looking again at going to the non-pthreads configuration of the Wasi-SDK.

What would be a good approach to deal with things like

https://github.com/dotnet/runtimelab/blob/588b1e37a5caeefd187bc8ddc2e4735c3da2c183/src/coreclr/nativeaot/Runtime/unix/PalRedhawkUnix.cpp#L219

It uses pthread_mutex_lock so is a problem.

Any thoughts on these options:

• Attempt to remove the whole function.
• Reimplement with whatever is available (spinning as required maybe), mutexes are required without threads.
• Something else

Thanks,

cc @dotnet/nativeaot-llvm

[yowl]

Another option

• stub pthread_mutex_lock with a noop

[SingleAccretion]

What does the Browser build do, and how much shimming is needed to make this work?

It seems like there are two options, broadly: 1) Implement PAL-level functions and constructs for a single-threaded environment. 2) Shim pthread APIs for a single-threaded environment - logically copy what Emscripten does.

pthread is lower-level than PAL so it is in some sense easier to shim it, but if all of it needs shimming, that becomes less attractive.

[yowl]

Emscripten stubs it. It starts here with the PTHREADS option setting is_mt

https://github.com/emscripten-core/emscripten/blob/96a832cee0cb3c028687dd03fa761264c67bf7b9/tools/system_libs.py#L703

In the same file, this brings in the stubs

https://github.com/emscripten-core/emscripten/blob/96a832cee0cb3c028687dd03fa761264c67bf7b9/tools/system_libs.py#L1120

Where __pthread_mutex_timedlock is stubbed

https://github.com/emscripten-core/emscripten/blob/96a832cee0cb3c028687dd03fa761264c67bf7b9/system/lib/pthread/library_pthread_stub.c#LL78C1-L82C1

I will see how many of these we would need.

[SingleAccretion]

Fixed in #2317.