Run unit tests by passing a file path

[JohnTitor]

Currently, we have to invoke the following to run specific unit tests:

./x.py test compiler/rustc_data_structures/ --test-args thin_vec

This isn't straightforward, it'd be great if we could do with:

./x.py test compiler/rustc_data_structures/src/thin_vec/tests.rs

But unfortunately, it fails currently:

thread 'main' panicked at 'error: no rules matched compiler/rustc_data_structures/src/thin_vec/tests.rs', src/bootstrap/builder.rs:225:17
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

meta

This happens on b17d9c1.

(This was originally reported on rust-lang/rustc-dev-guide#1138)

[fee1-dead]

I would like to work on this.

@rustbot claim

[asquared31415]

Curiously, passing a file like ./x.py test ./src/test/blah/file.rs only recently started working for me, and never used to work. The only difference I see in what you have as an example is the leading ./ in the test path, however it also works to omit that for me like ./x.py test src/test/blah/file.rs

I understand "it works for me" is incredibly unhelpful, so if anyone can think of ways to try and narrow down why it only sometimes works for some people, I would be happy to assist in testing things. I am on a x86_64 linux system, and it is definitely working at commit 9fef8d9 (June 16, 2021) which is before the commit in the issue here. However the original issue was reported 14 days ago. I highly doubt that something fixed and then re-broke the test runner, so I imagine it's something setup-dependent.

[fee1-dead]

@asquared31415 src/test/* stuff are suites and unlike normal unit tests on crates. I believe that is what makes the difference.

[asquared31415]

Ohhhh my bad, I misinterpreted

[fee1-dead]

I have made a pull request that makes it accept a file path but runs all tests in that crate. I could not figure out how to test only the path/file specified.

@rustbot release-assignment

[Mark-Simulacrum]

As mentioned in #86459 (comment), I'm going to go ahead and close this issue -- I don't think this is something we want to track as rustbuild work.