Regression in 2024-11-28: llvm-cov export errors with "truncated coverage data"

[alex]

Code

Beginning with today's nightly, processing coverage produced by rustc binaries fails. Here is an example CI run: https://github.com/pyca/cryptography/actions/runs/12077661037/job/33681030705?pr=12066

The relevant snippet:

nox > /home/runner/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llvm-cov export /home/runner/work/cryptography/cryptography/.nox/tests/lib/python3.12/site-packages/cryptography/hazmat/bindings/_rust.abi3.so -instr-profile=rust-cov.profdata '--ignore-filename-regex=[/\].cargo[/\]' '--ignore-filename-regex=[/\]rustc[/\]' '--ignore-filename-regex=[/\].rustup[/\]toolchains[/\]' '--ignore-filename-regex=[/\]target[/\]' --format=lcov
nox > Command /home/runner/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llvm-cov export /home/runner/work/cryptography/cryptography/.nox/tests/lib/python3.12/site-packages/cryptography/hazmat/bindings/_rust.abi3.so -instr-profile=rust-cov.profdata '--ignore-filename-regex=[/\].cargo[/\]' '--ignore-filename-regex=[/\]rustc[/\]' '--ignore-filename-regex=[/\].rustup[/\]toolchains[/\]' '--ignore-filename-regex=[/\]target[/\]' --format=lcov failed with exit code 1:
error: failed to load coverage: '/home/runner/work/cryptography/cryptography/.nox/tests/lib/python3.12/site-packages/cryptography/hazmat/bindings/_rust.abi3.so': truncated coverage data
error: could not load coverage information

#133418 is the only change in the regression range with a commit message mentioning coverage. cc: @Zalathar

Happy Thanksgiving!

Version it worked on

rustc 1.85.0-nightly (6b6a867ae 2024-11-27)

Version with regression

rustc 1.85.0-nightly (a2545fd6f 2024-11-28)

[Zalathar]

Hmm, my guess is that 619a272 is making it possible for all of a CGU's functions to be discarded at a late stage, so parts of the metadata end up being empty in a way that the llvm-cov isn't happy with.

Finding and testing a fix is probably going to be non-trivial (especially without a small repro), so for now it's probably wisest to revert.

[alex]

Regrettably, I don't have a reproducer smaller than "our entire large project with a bunch of deps" :-(

[Zalathar]

Yeah, minimizing this sort of coverage regression is annoying, so I generally don't expect issue reporters to do it themselves.

[jieyouxu]

Revert PR #133608 landed.

[arnodb]

Hi!

It's not minimized but https://github.com/arnodb/quirky_binder/actions/runs/12090701392/job/33718146447 faces exactly this problem.

It occurs with a build script. Unfortunately it's one of the last modules of the workspace to compile.

Commit 7c0ae33c25363b8b4637bb6501e10f2d2117f24c crashes but it is also possible that other commits crash in the same way.

Hope this helps.

[apiraino]

WG-prioritization assigning priority (Zulip discussion).

@rustbot label -I-prioritize +P-medium

[Zalathar]

The revert has landed and seems to have fixed the regression, so closing as fixed.

(I'll need to be mindful of this failure when working on a new version of #133418.)

[Zalathar]

A new version of the original PR has been merged in #134497, and will be included in the next nightly.

Based on my testing, I believe that this issue should not reoccur, but let me know if something does go wrong.