Rust's libLLVM-8.so doesn't work with clang

[emilio]

Which means that programs that run clang from a build script, or from cargo in some other way, fail.

Steps to Reproduce:
0. Have a clang binary pointing to clang-8 installed and in your path.

1. rustup install nightly
2. git clone https://github.com/rust-lang/rust-bindgen && cd $_
3. cargo test

Actual results: /usr/bin/clang: relocation error: /lib64/libclangBasic.so.8: symbol _ZTIN4llvm14FoldingSetBaseE version LLVM_8 not defined in file libLLVM-8.so with link time reference

Expected results: not so

This only happens if ran from cargo, if I run the test executable manually then stuff works, so I suspect this is some LD_LIBRARY_PATH mungling that cargo or rustup do under the hood.

This is technically a regression from #58408, same problem does not exist with rust stable / beta and clang 7.

There's a slight chance that there's an issue on how clang-8 is built in Fedora, so I also filed https://bugzilla.redhat.com/show_bug.cgi?id=1687047, but I think it's unexpected that clang when called from cargo / build scripts uses rust's libLLVM.so, and probably should be avoided.

cc @alexcrichton

[nagisa]

Is clang invoked from a build script? It is possible that cargo neglects removing the unnecessary LD_LIBRARY_PATH or similar environment variables before invoking the script in question.

EDIT: oh you wrote that in the report already. I remember there being an issue like this in cargo repo from a long long time ago, but cannot find it now.

[nagisa]

cc rust-lang/cargo#2888

[cuviper]

See also #55737.

We should probably set a custom suffix per #53987. Maybe even include the Rust version to help avoid bootstrap issues like what commit 2d21df8 saw. (cc @Mark-Simulacrum)

[Mark-Simulacrum]

Yep, I think a suffix of rust version (or maybe LLVM submodule commit?) would be good.

I think we'd appreciate a PR for something like that; I would expect it to be a matter of making the version suffix default be -rust-1.33.0-<sha> or something along those lines.

[emilio]

Sure, that sounds good, I can give that a shot.

[emilio]

I sent #59173 with a change that I hope will address this, but see there why I cannot confirm it will :)

[pnkfelix]

triage: P-high, assigning to @emilio