Trying to debug a bug that only happens in CI... fun

[marc0246]

No description provided.

[marc0246]

The environment variables all seem fine. I really can't fathom what could be different about the environment in the CI compared to my local environment.

[marc0246]

At least we can update proc-macro-crate now as of #2391, but that still doesn't answer why proc-macro-crate would fail with having an extra workspace, and only since 2.0, and only with certain examples, and different ones between the OSes...

[marc0246]

Turns out, since proc-macro-crate 2.0, it spawns is own cargo process. Because our examples/Cargo.toml was misconfigured, and didn't exclude examples/target from the examples workspace, this would cause that spawned cargo process to error out.

That said, the way this manifested itself is extremely non-obvious. proc-macro-crate should print cargo's stderr, instead of returning its stdout which is the empty string you can see in the first commit's CI runs. I guess the platform differences are due cargo behaving slightly differently between them and compiling them in a different order/parellelizing differently.