Do not pass -lc to the Emscripten linker

[hoodmane]

Hack: drop -lc to work around buggy Emscripten behaviors when -lc is passed. This can be removed if/when the problems are resolved in Emscripten and we drop support for the currently existing versions of Emscripten. See
emscripten-core/emscripten#22758

Unblocks #131736

This is a repeat of #98303, but the bug we were trying to resolve then went away and this is supposed to work around a different bug.

[rustbot]

r? @TaKO8Ki

rustbot has assigned @TaKO8Ki.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[hoodmane]

According to Emscripten maintainer @sbc100, when linking with emcc we shouldn't pass -lc:

your best bet is to remove any/all explicit additions of -lc, since
the compiler driver (emcc) will add it if/when its needed.

emscripten-core/emscripten#17191 (comment)

[sbc100]

I think a better solution would be to remove the explict -lc from all platforms. Unless somebody knows of some reason why its needed/useful? Are there platforms where libc is not linked by default?

[hoodmane]

I am not sure where the -lc comes from so I don't actually know how to cleanly implement that. It seems to be listed in a lib.rmeta file somewhere? What I could do is apply this hack to all 9 linkers and see if anything breaks in CI. But even if CI doesn't break, there could potentially be someone somewhere relying on this? Would be great to find out.

[workingjubilee]

I think a better solution would be to remove the explict -lc from all platforms. Unless somebody knows of some reason why its needed/useful? Are there platforms where libc is not linked by default?

There are platforms where libc is not even a thing that makes sense to talk about, yes.

[workingjubilee]

There are platforms where libc is not even a thing that makes sense to talk about, yes.

Hm, I think the answer from the Jubilee of 30 minutes ago might have been a bit confused there.

[sbc100]

I think a better solution would be to remove the explict -lc from all platforms. Unless somebody knows of some reason why its needed/useful? Are there platforms where libc is not linked by default?

There are platforms where libc is not even a thing that makes sense to talk about, yes.

In that case doesn't it make sense to remove the generic addition of the explicit -lc? I don't know of any platform that requires this, but I could be wrong.

[hoodmane]

Personally, in interest of getting something merged, I'd prefer we'd start with this change which definitely doesn't affect anyone else. This libc problem has been around for years. It would be nice to fix it so we can enable WASM_BIGINT and other features that require Safari 15 when linking Rust.

[workingjubilee]

@hoodmane Can you open an issue about "it's kind of weird that we pass -lc in the first place, isn't it...?" explaining the problem here.

[hoodmane]

After a bit more investigation, it turns out that rust/libc specifically asks to pass -lc to the linker. I'm having trouble using a local override version of libc, but I think deleting these two lines may fix the problem:
https://github.com/rust-lang/libc/blob/main/src/unix/mod.rs#L392-L393

[sbc100]

After a bit more investigation, it turns out that rust/libc specifically asks to pass -lc to the linker. I'm having trouble using a local override version of libc, but I think deleting these two lines may fix the problem: https://github.com/rust-lang/libc/blob/main/src/unix/mod.rs#L392-L393

Awesome! That looks like the correct fix! Nice find.

[workingjubilee]

neat, thanks.

[hoodmane]

New PR to rust-lang/libc:
rust-lang/libc#4002

[hoodmane]

Closed as superseded by rust-lang/libc#4002. Thanks @sbc100 and @workingjubilee!