regression: multiple definitions of function

[Mark-Simulacrum]

[INFO] [stdout]   = note: /usr/bin/ld: /opt/rustwide/target/debug/deps/libavif_parse-3944f88254156a01.rlib(avif_parse-3944f88254156a01.avif_parse.f31058791515eadf-cgu.3.rcgu.o): in function `avif_parse':
[INFO] [stdout]           /opt/rustwide/cargo-home/registry/src/index.crates.io-6f17d22bba15001f/avif-parse-0.13.2/src/c_api.rs:27: multiple definition of `avif_parse'; /opt/rustwide/target/debug/deps/libavif_parse-8988254f09aeef86.rlib(avif_parse-8988254f09aeef86.avif_parse.d6c78051b3aeaa35-cgu.3.rcgu.o):/opt/rustwide/cargo-home/registry/src/index.crates.io-6f17d22bba15001f/avif-parse-1.0.0/src/c_api.rs:27: first defined here
[INFO] [stdout]           /usr/bin/ld: /opt/rustwide/target/debug/deps/libavif_parse-3944f88254156a01.rlib(avif_parse-3944f88254156a01.avif_parse.f31058791515eadf-cgu.3.rcgu.o): in function `core::ptr::const_ptr::<impl *const T>::is_null::runtime_impl':
[INFO] [stdout]           /opt/rustwide/cargo-home/registry/src/index.crates.io-6f17d22bba15001f/avif-parse-0.13.2/src/c_api.rs:50: multiple definition of `avif_data_free'; /opt/rustwide/target/debug/deps/libavif_parse-8988254f09aeef86.rlib(avif_parse-8988254f09aeef86.avif_parse.d6c78051b3aeaa35-cgu.3.rcgu.o):/opt/rustwide/cargo-home/registry/src/index.crates.io-6f17d22bba15001f/avif-parse-1.0.0/src/c_api.rs:50: first defined here
[INFO] [stdout]           collect2: error: ld returned 1 exit status

• https://crater-reports.s3.amazonaws.com/beta-1.78-3/beta-2024-03-30/reg/avif-decode-0.2.2/log.txt

[saethlin]

avif-decode has two different versions of avif-parse in its dependency tree, and avif-parse exports the symbols avif_parse and avif_data_free. I think this crate has only ever managed to link due to optimistic Linux linker optimization shenaniganry.

The change bisects to #122233. I think there's a CGU partitioning that causes the build to successfully link, and that PR perturbed enough code to alter how CGUs get merged. The crate still builds on nightly and on stable if you set -Ccodegen-units=9 or any greater value, but fails to build on nightly and on stable with -Ccodegen-units=8 or any lesser value.

[saethlin]

@estebank fresh example of where it would be great to have better linker diagnostics

[Mark-Simulacrum]

Sounds good. I think there's no true regression here in that case, dropping various labels and I suspect we can just close (will do that unless we want to track it as diagnostics, but doesn't seem worth it to me)

[saethlin]

Esteban was musing about writing better linker diagnostics, and now that I've assigned myself to this I'll be better able to find it if I do cook up something later.

I agree this is better closed.