`cargo tests` fails on windows

[Urhengulas]

Currently cargo test fails on the windows CI (and are therefore disabled). See https://github.com/knurling-rs/probe-run/runs/2829124012?check_suite_focus=true#step:6:76.

Following tests are reported to fail:

• dep::cratesio::tests::end_to_end
• dep::rust_repo::tests::v1_52_path
• dep::rust_std::tests::end_to_end
• dep::rustc::tests::end_to_end
• dep::tests::from_std_path_returns_correct_variant

Git lens points me to @japaric, but I will look into this for now.

[japaric]

From a quick glance, these tests contain OS-specific details (the path delimiter). e.g. "/home/user/.cargo/registry/.." is not a valid path on Windows. It's also not the right folder on Windows either.

So either

1. the tests should become OS-agnostic: the expected strings should not use '/' but rather std::path::MAIN_SEPARATOR, or
2. the test should be duplicated: one Windows variant ("??\Users\.cargo") and one Unix variant ("/home/user/.cargo") marked with cfg(unix) / cfg(not(unix)) respectively.

Depending on each test we may want to do one (a) or the other (b).

[Urhengulas]

I've started to work on this, but it is a bit hard to test, if you are not working on a windows machine. See https://github.com/knurling-rs/probe-run/tree/fix-winsows-tests

[japaric]

@Urhengulas feel free to assign this issue to me if it's proving to be time consuming. I have a windows machine at hand.

[justahero]

Hello,

I tried to activate some of the tests for Windows to have them run on CI again. In general the use of PathBuf::join helps a lot to make some of the tests pass.

Unfortunately there are some checks inside the different Path::from_std_path implementations that make it pretty difficult to test correctly on a Windows system, for example the function Path::is_absolute is different on Unix & Windows. On Unix systems all path that start with / are considered absolute, while on Windows this requires a drive location, e.g. c:\. For some tests that is not a big issue, because their path use the home directory that can be obtained via the dirs crate to get the OS specific location.

On the other hand I'm not quite sure how to resolve the test dep::rustc::tests::end_to_end, because the path starts with "/rustc/...", where I don't know the equivalent on Windows. Maybe the input path is quite irrelevant and can be easily fixed, otherwise this specific test (& dep::tests::from_std_path_returns_correct_variant) could use a Windows specific setup?

[justahero]

Ok, added the other tests as well in #246, once I realised that the "rustc" part is parsed in a similar fashion as the other Path::from_std_path handle the path parameters. The path does not have to start with "/rustc/...".

[Urhengulas]

Fixed by #246