rustc 1.32 using system LLVM-7 crashes with: error: failed to load bc of "core.9adngr8s-cgu.0": Invalid record

[Whissi]

On Gentoo Linux we have the same problem like reported in #57762, downstream bug: https://bugs.gentoo.org/675752

I now applied the two patches (llvm-mirror/llvm@da1fb72 and llvm-mirror/llvm@cc1f2a5) to our llvm-7 version but fd (https://github.com/sharkdp/fd) is still failing to build. New error is:

INFO 2019-01-25T18:55:20Z: rustc_codegen_llvm::back::lto: linking "core.9adngr8s-cgu.0"
error: failed to load bc of "core.9adngr8s-cgu.0": Invalid record

error: aborting due to previous error
[...]
Could not compile `fd-find`.), unknown: false, exit_code: 101 }
error: Could not compile `fd-find`.

Please see the attached verbose build.log.
build.log.zip

[Whissi]

Patching rust iwth 1c95f5a doesn't help.

[nagisa]

There is a somewhat conflicting signal – the comment 16 seems to say that the patch does fix the compilation issue.

Would it be possible to get the sample file that is referenced in the error message (use -Csave-temps and -Ccodegen-units=1)?

[Whissi]

Not sure if I did it the right way, I set RUSTFLAGS="-Csave-temps -Ccodegen-units=1" environment variable. Here's the complete target directory, I hope it is included: https://dev.gentoo.org/~whissi/stuff/fd-7.2.0_target_201901252245.tar.xz

[nagisa]

Ah, my bad. It seems that the failing bytecode comes from libcore, so it won’t be included in the target/.

[gyakovlev]

looks like we figured it out, 1c95f5a actually fixes rust with system installed llvm-7

[Whissi]

Right. The problem error: failed to load bc of "core.9adngr8s-cgu.0": Invalid record seems to be caused by sccache usage I tried while I started to investigate the original issue #57762 before the bug + solution was filed here to speed up bisect a little bit :/