flaky repo.reset() behavior

This issue is about repo.reset() not seeming to work reliably 100% of the time.

I have some CI tests that modify a file in a non-bare repo, commit the change to a branch, and then call repo.reset() with GIT_RESET_HARD to restore to a different commit. After calling reset, to confirm that the reset was successful, the test creates a new Repository object and checks the repository's status using repo.status(). Occasionally (but very rarely), the status check comes back showing the original file still as WT_MODIFIED. A subsequent call to repo.diff() confirms the difference in the file. Is there something pygit2 is or isn't doing that could cause reset() to behave flakily like this? Or is libgit2 the more likely culprit?

This is using pygit2 1.2.0. I can try upgrading, but is there a reason to think this would be fixed in a newer version (e.g. a known issue that was fixed)? I haven't found anything like this in pygit2 or libgit2's issue tracker or release notes.

Commencez par reproduire la séquence de CI signalée avec pygit2 1.2.0 : modifiez et committez un fichier dans un dépôt non bare, appelez repo.reset() avec GIT_RESET_HARD, puis créez un nouveau Repository et comparez repo.status() avec repo.diff(). Le travail est terminé lorsqu’il a été déterminé si le résultat intermittent WT_MODIFIED provient de pygit2 ou de libgit2 et que la cause ou une voie de correction confirmée a été documentée.