I cannot reproduce the problem. It also seems strange that you get a QResizeEvent in enterEvent?

Yes, it seems strange to me too.
I have installed it in several different conda environments (from PyPi) to test if it was a version problem or if the piece of code I put was functional. In all cases, I get the error.
I tried with PyQt5 versions: 5.15.2, 5.15.4, and 5.15.6 and in all of them, I get the same error. I tried to debug but since enterEvent is a virtual function it's hard to keep track of it (at least for me). I printed the event type to confirm and interestingly, only in this combination do I get QResizeEvent, in the rest, I always get QEnterEvent.
With the try-except clause work as expected

I attach a small video with the procedure to reproduce the error.

Mario S.

That's strange :/ (but I'd think blindly ignoring the error with a try... except is wrong too, as something weird is going on).

I've got the same issue. Do you have any idea about how to avoid this before issue fixed?

Please provide your repro. Also include the environment, all relevant versions of modules, installation methods (conda/pip/linux package), etc.

I ran into the same traceback using Matplotlib 3.5.3 and PyQt 5.15.6.

Here is the minimal (admittedly, quite non-minimal) example, run it, click on the second tab, then move the mouse over FigureCanvas:

import sys

from PyQt5.QtWidgets import QApplication, QHBoxLayout, QScrollArea, QTabWidget, QVBoxLayout, QWidget

from matplotlib.backends.backend_qtagg import FigureCanvas


if __name__ == '__main__':
    app = QApplication(sys.argv)
    layout = QVBoxLayout()

    tabs = QTabWidget()
    tabs.addTab(FigureCanvas(), 'a')
    tabs.addTab(QWidget(), 'b')

    def tabChanged(index):  # create second tab lazily
        if index != 1:
            return
        widget = tabs.widget(index)
        if widget.layout() is not None:
            return
        vboxlayout = QVBoxLayout()
        vboxlayout.addWidget(FigureCanvas())
        wrapper = QWidget()
        wrapper.setLayout(vboxlayout)
        scrollarea = QScrollArea()
        scrollarea.setWidget(wrapper)
        hboxlayout = QHBoxLayout()
        hboxlayout.addWidget(scrollarea)
        widget.setLayout(hboxlayout)
    tabs.currentChanged.connect(tabChanged)
    tabs.show()
    sys.exit(app.exec())

I did some digging and the culprit seems to be frame = sys._getframe() in FigureCanvas' resizeEvent method, which causes a memory leak, which in turn leads to the wrong event (QResizeEvent) being sent to enterEvent. Sadly, I didn't dig further than that - I have no idea why this only rarely leads to an exception - the leak is pretty consistent. Also, the exception doesn't happen using PyQt6. If I had to guess I'd say there's a race condition somewhere.

I wasn't able to reproduce the problem originally reported here, so I cannot be certain the root cause is the same, but I think it's safe enough to assume it is.

For anyone else facing this issue, the workaround I've used is:

class FigureCanvas(FigureCanvas):
    def resizeEvent(self, event):
      super().resizeEvent(event)
      import gc
      gc.collect()

Not the prettiest solution, and it does come with a slight performance penalty, but at least it manages to avoid the exception. By explicitly forcing garbage collection, we ensure frame variable is cleaned up before any other Qt code is executed. A faster option for someone who doesn't care about enter/leave events would be to override enterEvent and leaveEvent and simply ignore them. Filtering out events in enterEvent by type is not an option, since when this occurs, all subsequent enter events also receive the same QResizeEvent.

The solution I would suggest is quite simple, I would explicitly delete the variable frame once it's no longer needed in FigureCanvas' resizeEvent method:

        recursion = frame.f_code is getattr(frame.f_back, 'f_code', None)
        del frame
        if recursion:
            return

Perhaps the cleaner way to check for recursion would be something like

def resizeEvent(self, event):
    if self._in_resize_event: return  # where the attribute is initialized to False in init.
    self._in_resize_event = True
    try: ...  # the actual resize logic
    finally: self._in_resize_event = False

?

Sounds good. Should I create a PR or will you handle it?

@anthepro If you are willing to open a PR that would be great!

Closed by #23729, I guess.