Catch exceptions in other Qt threads

[jdreaver]

I recently switched to pytest-qt, and I love it so far. Before, I used my own pytest fixture to create a QApplication. I found a way to catch exceptions on all threads by overriding sys.excepthook in the fixture and checking for errors:

caught_exceptions = []


@pytest.yield_fixture
def qtbot_(request, qtbot):
    # Set excepthook to catch pyside errors in other threads.
    global caught_exceptions
    caught_exceptions = []

    def pytest_excepthook(type_, value, tback):
        global caught_exceptions
        caught_exceptions.append(''.join(traceback.format_tb(tback)) + "\n" +
                                 str(type_.__name__) + ": " + str(value))
        sys.__excepthook__(type_, value, tback)
    sys.excepthook = pytest_excepthook

    yield qtbot

    sys.excepthook = sys.__excepthook__
    if caught_exceptions:
        pytest.fail("Caught qt exceptions:\n{}".format(
            "\n".join(caught_exceptions)))

I created this so I can see when my table model methods fail. Without this, if an error happens in the data() or header() methods of a table model, like when a table view is being populated using the model, the test will not fail.

I think something similar to this would be a nice feature of pytest-qt, but a better, less hacky implementation would probably be better.

[nicoddemus]

Done in 1.1.1, if you have a change to try it out let me know what you think. Cheers!

[jdreaver]

Wow, great implementation!

One thing I would consider doing: move the new the context manager outside of the QtBot class, and define it as its own top-level definition. It doesn't seem to require a QtBot class instance for anything. That's just a style opinion I suppose.

Thanks!

EDIT: I didn't even notice you already made the new 1.1.1 release too. I'll start using this immediately to help test it out.

[jdreaver]

It works great!

As a sidenote, I wish using pytest.fail() in a fixture caused the test to be marked as a failure, instead of an error. Oh well!

[nicoddemus]

Excellent!

Yes, I noticed that about pytest.fail() too, but that's how pytest works it seems.

Thanks again!

[nicoddemus]

Thanks for the suggestion about changing the context manager into a function. That allows me to test the code formatting code (see this commit if you want to check it out).

Cheers!

[jdreaver]

Nice! I like the formatting refactoring as well. It keeps the main fixture definition very clear.

[patrickkidd]

FWIW, it is probably more accurate to report these as "Exceptions caught in Qt event loop" rather than "Qt exceptions in virtual methods." This is because these exceptions can also be raised in a pyqt slot.

[nicoddemus]

@patrickkidd good point, thanks. Implemented it in #255.