napari: [Qt6 pyside6 6.5] TypeError on start

Hi! So not totally sure what changed in PySide6 >=6.5.0 for the error to happen but basically changing the order of inheritance seems like changes also the order used when you call super to call the respective __init__ methods of the inherited classes. I think the MRO (Method Resolution Order) depends on the order you put the classes your class inherits.

My guess is that leaving last the plain Python class while also inherithing a PySide6>=6.5.0 causes super to call the class __init__ passing all the args/kwargs which causes the TypeError. Changing the other seems like workarounds this since possibly the args/kwargs don’t get passed to the plain Python class __init__ (maybe the MRO mechanism detects is not necessary to pass the args/kwargs in that call?)

As mentioned, this is more like a guess because I’m not completly sure to fully get the way MRO works to be honest, but tests passed over the superqt PR when I encountered the problem there so I took that as a win 😃

Edit: Edited this because I got lost in my own explanation 😅

Hi all 👋 - thanks for your efforts here. It’s currently a big blocker for us at deeplabcut, so if our dev team can allocate time/efforts please let us know and cc @jeylau ❤️

Just as an update, pyside6 6.5.2 was released and this issue remains. The fix in https://github.com/napari/napari/pull/6062 still works, but alas so do the test failures from https://github.com/napari/napari/issues/5657

Ok building 6.5.1.1 https://github.com/conda-forge/pyside2-feedstock/pull/195 to help with testing.

I have a branch with the swap of position and it seems to work fine. However, we hit a different issue: https://github.com/napari/napari/issues/5657 I will make a PR, but I expect that tests will fail. Edit: note pyside6 6.5.1 is bad regardless of this fix, with TypeError: actionTriggered(int) only accepts 1 argument(s), 2 given! when for example opening samples, so it requires 6.5.1.1

class _BaseEventedItemModel(Generic[ItemType], QAbstractItemModel):

Would swapping the order of the subclasses be acceptable for napari? The presumption is that it would be accompanied with a comment as to why this is important for PySide6.

Thanks @dalthviz I think maybe we should take the same approach in napari, so we can test on pyside >6.5

Maybe, Issue 2354 has been getting good attention. The developers translated my example and adapted it for their internal validation and reuploaded it. so i’m pretty confident that there will be some improvement.

To recreate the PySide6 issue

from PySide6.QtWidgets import QApplication, QMainWindow, QLabel
from typing import Generic, TypeVar

ItemType = TypeVar("ItemType")


class MainWindow(QMainWindow, Generic[ItemType]):
    def __init__(self):
        QMainWindow.__init__(self, parent=None)
        self.resize(400, 300)
        self.lbl = QLabel(self)
        self.lbl.setText("XXX")


app = QApplication([])
window = MainWindow()
window.show()
app.exec()

now i need to figure out their issue reporting system:

https://bugreports.qt.io/browse/PYSIDE-2354

(not officially involved with deeplabcut but thanks for the ping)

Lets see if we can backport the patch: https://github.com/conda-forge/pyside2-feedstock/pull/191

It looks like it is working with PyQt6, so the problem is only touching PySide6.