Linker error because `-wL as-needed` conflicts with `-C link-dead-code`

[xd009642]

When a project has the [lib] tag in the Cargo.toml the -wL as-needed flag is added to the projects linker flags. However, if a project uses -C link-dead-code the two flags conflict causing a linker error of undefined reference for any functions in the lib.

This was discovered because link-dead-code is used for code coverage in tarpaulin. I spent some time seeing if there was a way around it as a user but I didn't manage to solve it myself. Issue that lead me to this for reference: xd009642/tarpaulin#126

Any solutions I could currently implement would be appreciated if this requires a PR I'm also happy to contribute (though may need some mentorship)

[xd009642]

Is there any idea about what it would take to fix this? I'm interested in helping out if someone can help guide me 😄

[sharksforarms]

Getting a similar issue in suricata: https://github.com/OISF/suricata/tree/master/rust

I tried removing the [lib] but still getting undefined symbols... I wonder if it's the extern C stuff

[richkadel]

When compiling Rust binaries with -Z instrument-coverage and -C link-dead-code enabled, Rust binaries compiled for the Windows-MSVC target would typically crash (seg-fault) during the post-exit phase of writing the coverage-counter results to the *.profraw file (or at best, generate an empty file).

After learning of this known bug, I discovered a workaround: Disabling link-dead-code, via -C link-dead-code=no, resolves the problem with -Z instrument-coverage when targeting Windows-MSVC. I'm incorporating this workaround in PR #75828 , but note that this does not actually correct the bug.

It is an acceptable workaround for now, but I would prefer to enable link-dead-code, so dead code would still show up in the coverage reports (with zero execution counts).

Before learning of this workaround, I did some extensive investigation of the bug with -Z instrument-coverage on MSVC. I've documented my discoveries and a lot of the intermediate results, in case they are helpful to someone with more knowledge in this area. I think I understand why the program fails (and how it appears to be a link-time issue), but I was not able to determine why the linker is doing what it is doing, or how to fix it.

[richkadel]

After some sidebar discussions, I think the issue with -Z instrument-coverage and -C link-dead-code (which is only present when targeting MSVC and using the MSVC linker) is different. I've created a new bug report, Issue #76038 , and referenced this one, just in case someone sees a relationship.