Not sure how to handle path on Windows. Should get_search_path returns CString directly? Or just throw an invalid UTF-8 error like util::byte2path?

Also, ccb48dd makes it thread-safe to access global state of search path, but have no idea is that necessary.

Not sure how to handle path on Windows. Should get_search_path returns CString directly? Or just throw an invalid UTF-8 error like util::byte2path?

e5d8643 fix: get_search_path will returns CString

I think returning a CString is more straight foward since set_search_search also accept an IntoCString type.

This seems generally fine but the global-state-ness is worrisome with respect to how this is being used in rust-lang/cargo#9035. If this doesn't work for that PR would you still like to land this?

This seems generally fine but the global-state-ness is worrisome with respect to how this is being used in rust-lang/cargo#9035. If this doesn't work for that PR would you still like to land this?

If i didn't get it wrong, from the view of git2-rs itself, the global state in git2-rs just reflects the global state in libgit2. To make git2-rs as safe as possible (that's one of it's goal 😂), I would choose an approach similar to this one no matter how callers use it.

Ok, in that case though I think it may be better to flag these functions as unsafe. They're not safe to mix with other configuration functions in multiple threads and this probably has to happen near the front of libgit2's usage rather than later. I think the synchronization can only really be left up to the caller since we don't know about all the calls to libgit2's configuration functions.

Got it. I'll remove the synchronization part. Thanks!

Looks good! As one final thing though, I think the test may wantt to move to an integration test in something like tests/foo.rs? Otherwise I don't think we can guarantee that it's synchronized with all other tests.

Thanks for your advice. The test has been moved to a integration test at adc0ee8.

I have a dumb question to task. If I understand it correctly, the reason moving it to a integration test is that if people what to mutate the same global state, it can edit the "integration test" directly without touching "unit test" part, right?

👍