It looks like the test crashes the process on Ubuntu with conda=No. The test stalls on Python3.7 on macOS. There are several tests which express this very behavior. Any thoughts on that?

For the issue with Ubuntu, the most related thread I found on SO was https://stackoverflow.com/questions/56281631/qapplication-instance-qtbot-fixture-causes-travis-ci-to-abort-and-core-dump. Maybe, it'll help.

From the logs, I see that the crash occurs when pytest creates an instance of QApplication with an empty list being the only argument. At home, I get warnings (I don't remember what they read) when I create an app like this: QApplication([]), but not a crash. For now, I can't reproduce the crash locally to prevent it before pushing here.

Oh, wow, I get the same crash in GitHub tests when I create an instance of QApplication manually. That is, there is nothing wrong with pytest. It can't influence QApplication that much, can it?

I had to skip the test for the config for now.

And I can't actually get the coordinates I move the mouse to from the event methods. I can't predict the window border dimensions. Possibly, I missed something. Anyway, the mapping I use should prevent any default values (like QPoint(-1, -1)) from passing the test.

(Sidenote: @dalthviz , if/when this is merged, it should be presumably done via the Squash strategy (or with rebase cleanup first) to avoid all the repeated fixup commits for a relatively small change. I enabled a setting that will still preserve a list of the commit messages in the commit description, and the full commit history of this branch can still be referred to on the PR in perpetuity, while the history on the main branch is kept clean.)