Compiling with `RUSTFLAGS="-C instrument-coverage"` changes dependency tracking for doctest builds

[scouten-adobe]

I have a (currently) closed-source crate. I'm trying to use the new coverage tools built into Rust 1.60 stable and I'm seeing some unexpected behavior when trying to run unit tests with coverage enabled.

When I run:

RUSTFLAGS="-C instrument-coverage" cargo test --all-features

I get hundreds of errors of the form …

error[E0463]: can't find crate for `serde_derive` which `ciborium` depends on
 --> (path redacted)
  |
4 | use ciborium::value::Value;
  |     ^^^^^^^^ can't find crate

… when it gets to the doctest portion of the test suite. (Ironically, this crate currently has no doc tests.)

If I add …

[lib]
doctest = false

… to Cargo.toml or compile without the instrument-coverage flag, I have no such errors.

Rust 1.60 stable, macOS 12.1 (Intel), closed-source (currently) crate.

Before I start digging in deeper, I would like to know if:

(a) Anyone else is seeing anything like this?
(b) Would it help the Rust core team if I could find a publicly-visible repro case?

Meta

rustc --version --verbose:

rustc 1.60.0 (7737e0b5c 2022-04-04)
binary: rustc
commit-hash: 7737e0b5c4103216d6fd8cf941b7ab9bdbaace7c
commit-date: 2022-04-04
host: x86_64-apple-darwin
release: 1.60.0
LLVM version: 14.0.0

[ehuss]

(b) Would it help the Rust core team if I could find a publicly-visible repro case?

This might be a good idea. I tried in a few projects and couldn't find a way to reproduce.

[wesleywiser]

Yeah, if you can point to a public crate that shows the issue that would be very helpful!

[scouten-adobe]

I am working on a repro case; hope to have something over the weekend.

[scouten-adobe]

Creating a repro case is proving "interesting." I've lost reproducibility several times by cutting away different pieces of the original case.

[scouten-adobe]

Please see https://github.com/scouten-adobe/rust-bug-95825. 100% reproducible for me with this project.

[apiraino]

@scouten-adobe Thank you so much for providing a reproducible example. Assigning priority as discussed in the Zulip thread of the Prioritization Working Group.

@rustbot label -I-prioritize +P-medium

Can someone please bisect the regression happening on macOS?
@rustbot ping icebreakers-cleanup-crew