Need regression test for wasm linker override=clang (108910)

[pnkfelix]

As noted in #108996 (comment), I don't want to delay the fix further on the CI issues I was having with my regression test.

This issue is just tracking a (hopefully simple?) work item to make such a regression test, e.g. using the one from 568b722 after figuring out how to make the CI happy (see error CI posted here: #108996 (comment) )

[agaraman0]

@pnkfelix i would like to pick this

[jyn514]

@agaraman0 go for it :) you can use @rustbot claim to assign yourself. See https://rustc-dev-guide.rust-lang.org/getting-started.html for more instructions.

[ndrewxie]

Full disclaimer, I'm not an experienced contributor to open source, nor will I pretend to be, so take the below with a grain of salt:

If we're just checking to see if the link-arg=-nostartfiles is being emitted, would it suffice to just capture the link flags (e.g. using RUSTC_LOG=rustc_codegen_ssa::back::link=info) and check if this option is present through a simple string find? That way, even if clang isn't present in the CI environment, the functionality can still be verified. This is certainly brittle to some degree, so not ideal...

Although it does seem strange that clang isn't available... Your wording of "make the CI happy" seems to imply that there's a workaround of sorts - if so, could you provide a few pointers? I'd be happy to investigate

[jyn514]

@ndrewxie feel free to install clang on any of the CI builders, look for apt install in src/ci for examples.

[ndrewxie]

@jyn514 Oh dang I didn't expect to have the perms to do that in a CI environment. Thank you! I'll test that approach and add a test :)

[ndrewxie]

@ndrewxie feel free to install clang on any of the CI builders, look for apt install in src/ci for examples.

Hmm, there appears to already be a script that installs clang, located in /src/ci/scripts/install-clang.sh. However, it chooses to not install it for linux under the assumption that every system would already have clang installed and configured properly.

Note that we don't install clang on Linux since its compiler story is just so different. Each container has its own toolchain configured appropriately already.

Would it be safe for me to apt install it anyways, or is there a risk that clang acquired from the package manager would be configured improperly and mess something up?

[jyn514]

As long as the new test passes and no other test regresses, it's probably ok. We don't have a hard dependency on the version of clang like we do for LLVM and the bootstrap compiler.

[b-ncMN]

Is this being worked on ? I'd like to give it a try

[ndrewxie]

Is this being worked on ? I'd like to give it a try

Afaik it's not being worked on. Good luck!

[b-ncMN]

@rustbot claim

[b-ncMN]

I'm not sure about the goal, after checking some stuff out
The goal would be to take the test (this test 568b722), and make it work by installing clang?

[jyn514]

@InfRandomness yup, that sounds right :)

[reez12g]

@rustbot claim

[Dylan-DPC]

Closing as #116264 resolves it