Fix nightly WASM target

[robertbastian]

No description provided.

[sffc]

Context, please?

Older nightlies, including the one we were using in CI when I last checked a few months ago, do not support the wasip1 target.

[robertbastian]

https://unicode-org.slack.com/archives/C06K24H8XHQ/p1731075306932339

CI passes on the PR

[robertbastian]

https://unicode-org.slack.com/archives/C06K24H8XHQ/p1730988954956999

[Manishearth]

@sffc Rustc has been working on renaming wasm-wasi to wasm-wasip1 for a while (there's a wasip2 target being introduced as well, hence the rename). Our build has warned on this for a while but we haven't noticed since it's in CI.

[sffc]

Okay, I understand "wasip1" and why we want to use it; I just don't know "why now". The pinned nightly that we had been using for a long time, about a year old, did not support "wasip1" last I checked. I don't know how big the window was of nightlies that supported both "wasi" and "wasip1". I guess the latest nightly out last night dropped "wasi" backwards compatibility support?

[Manishearth]

Yes.

[Manishearth]

See rust-lang/rust#132562: wasip1 introduced in January, and wasi removed a few days ago.

[sffc]

That seems like a fairly aggressive schedule, especially considering wasm32-wasi is a target on stable...

Currently our LLVM_COMPATIBLE_NIGHTLY ?= "nightly-2024-01-01" which is still too old. I think we picked that version because it was the latest version that was compatible with LLVM in Xcode; I assume it is a coincidence that it is January 1.

So this is a bit unfortunate of timing. But I guess if CI works, we can ignore it and hope things are okay

[Manishearth]

@sffc I think it would be fine to mention the Xcode thing on that PR, people who care about LLVM versions have heavy overlap with people doing wasm stuff. I agree that the schedule feels aggressive, it felt aggressive to me as well.

[Manishearth]

Surprised it didn't show up with deny(warnings) though

[sffc]

I think this PR might not break things in the short term because I believe our WASM LLVM size tests use wasm32-unknown-unknown, not wasm32-wasi. However, I did notice one bug in one of the changes, which should have been caught earlier.

[sffc]

Issue: This is impacting the include path of clang++, not Rust

[sffc]

Well we do link C++ together with Rust on a WASI target in the codepointtrie builder code. So it may be difficult to rebuild the codepointtrie builder WASM with this change. Hopefully we won't need to do that again soon.

[bjorn3]

Surprised it didn't show up with deny(warnings) though

#[deny] only denies lints and this warning is not a lint. If it was a lint, you did see a note with the name of the lint and the current lint level attached to the warning.

[riking]

@sffc Target availability is explicitly excluded from the stability promises, due to it needing to match timing decisions of other organizations outside of Rust's control https://doc.rust-lang.org/rustc/target-tier-policy.html?highlight=hard%20stability%20guarantee#general