Multiple matching crates for parent crate [E0464]

[CAD97]

See https://travis-ci.org/Robbepop/string-interner/jobs/339735884 for the build log (config.link_deps()).
And https://travis-ci.org/Robbepop/string-interner/jobs/339733480 for an earlier one (config.target_ruscflags = Some("-L target/debug/deps").

Locally on my Windows machine, config.link_deps() fails due to #81 and config.target_rustcflags = Some("-L target/debug/deps") works fine. config.clean_rmeta is called after setting the link arguments in both cases.

Compile test harness here: https://github.com/CAD97/string-interner/blob/d95952459bb0b0a99448c9f5c84ff182b2ef7e38/tests/compiletests.rs
Singular compile-fail test here: https://github.com/CAD97/string-interner/blob/d95952459bb0b0a99448c9f5c84ff182b2ef7e38/tests/compile-fail/pooled_str_outlives.rs

[laumann]

Is this a bug in clean_rmeta()?

Currently, link_deps() shouldn't be used if your PATH variable contains spaces - I have kind of decided not to fix that issue here (#81), but rather try to get it fixed in rustc/compiletest. So for the moment, I would just say, don't use paths with spaces in conjunction with link_deps().

But the problem you're pointing to was observed when mixing cargo check and cargo test. Both commands produce .rmeta files, but they are not compatible and rustc then gets confused when two versions of the same crate was found on PATH (IIRC). So is this a case where some .rmeta files are left behind when they shouldn't be?

For reference, here's the implementation of config.clean_rmeta().