PR: Use static calls of exec_ elsewhere where needed, and test them

[StSav012]

The help for QMenu.exec_ reads something like the following (the order might be different, and fully qualified Qt class names might appear):

exec_(...)
    exec_(actions: typing.Sequence[QAction], pos: QPoint, at: typing.Optional[QAction] = None, parent: typing.Optional[QWidget] = None) -> QAction
    exec_(self) -> QAction
    exec_(self, pos: QPoint, at: typing.Optional[QAction] = None) -> QAction

The first option here is a static call to QMenu.exec_. In the QtPy tests, such a call is never tested. So, here it is. As one might expect, it fails. Use the _possibly_static_exec function to fix the errors.

[StSav012]

The mistake of mine was in the following lines:

        QtCore.QTimer.singleShot(100, lambda: qtbot.keyClick(
            QtWidgets.QApplication.widgetAt(1, 1),
            QtCore.Qt.Key.Key_Escape)
            # namely, ↑↑↑ up here
        )

In PyQt5=5.9, there was no QtCore.Qt.Key.Key_Escape yet. Surprisingly, the attribute error was suppressed, and no timer shot was fired. That caused the stall. When I changed the parameter to QtCore.Qt.Key_Escape, it worked.

As I don't use Windows, and the problem appears only on Windows, I couldn't catch the problem locally. The third time's a charm.

[StSav012]

Oh, it's ERROR module 'importlib_metadata' has no attribute 'distributions' again. I see the error is so persistent that @dalthviz has merged df95964 into master despite the failed check.

[CAM-Gerlach]

Oh, it's ERROR module 'importlib_metadata' has no attribute 'distributions' again.

Thanks, I opened #423 to document that.

[CAM-Gerlach]

(Fixed the importlib error in #425 , BTW, while also speeding up the Conda=yes CI jobs by roughly 2x and simplifying the CI script a bit.)

[CAM-Gerlach]

Sorry, didn't realize I hadn't actually reviewed this.

If it might be used in multiple modules, maybe better to move _possibly_static_exec to utils? Thoughts?

Beyond that, I didn't spot any obvious blocking issues with this, so otherwise leaving this to @dalthviz 's expertise to review further.

[StSav012]

If it might be used in multiple modules, maybe better to move _possibly_static_exec to utils?

I've looked through

• QApplication (Qt5, Qt6),
• QCoreApplication (Qt5, Qt6),
• QDialog (Qt5, Qt6),
• QDrag (Qt5, Qt6),
• QEventLoop (Qt5, Qt6),
• QGuiApplication (Qt5, Qt6),
• QMenu (Qt5, Qt6),
• QTextStreamManipulator (PySide2, PySide6),
• and QThread (Qt5, Qt6),

and found that QCoreApplication, QGuiApplication, QApplication, and QMenu indeed have such static calls. Thank you for the remark!

As for QTextStreamManipulator, I can't compose a test call. I just can't make an instance of QTextStreamManipulator last. According to help(QTextStreamManipulator.exec_), it isn't static, but still. That's an odd class.

[StSav012]

(Fixed the importlib error in #425 , BTW, while also speeding up the Conda=yes CI jobs by roughly 2x and simplifying the CI script a bit.)

Thanks a lot! That's a marvelous solution!

[StSav012]

Surprisingly, the tests for QtCore.QCoreApplication.exec_ don't result in a segfault on Ubuntu with conda=No.

[CAM-Gerlach]

Thanks! All looks pretty reasonable to me now, but I'll leave it to @dalthviz 's greater expertise to review further, thanks.

[dalthviz]

Thanks @StSav012 ! LGTM 👍 But, just in case, what do you think @ccordoba12 ?

[ccordoba12]

Thanks @StSav012!