If this indeed portable this way, can we just decorate the methods directly? Even if we have to add a bunch of methods like

@QtCore.Slot
def foo(self):
    return super().foo()

that seems better than doing the multi-layered meta-programming here!

Yes, can do. A considerable amount of meta-programming is still done in the loop just around the snippet, so maybe a test is needed to make sure that all functions in self.toolitems have been decorated...

Wow, that's interesting. I tried adding

    @QtCore.Slot()
    def home(self, *args):
        return super().home(*args)

    @QtCore.Slot()
    def back(self, *args):
        return super().back(*args)

    @QtCore.Slot()
    def forward(self, *args):
        return super().forward(*args)

    @QtCore.Slot()
    def pan(self, *args):
        return super().pan(*args)

    @QtCore.Slot()
    def zoom(self, *args):
        return super().zoom(*args)

    @QtCore.Slot()
    def configure_subplots(self, *args):
        return super().configure_subplots(*args)

    @QtCore.Slot()
    def edit_parameters(self, *args):
        return super().edit_parameters(*args)

    @QtCore.Slot()
    def save_figure(self, *args):
        return super().save_figure(*args)

and this works for home, back and forward - but not for the others. Irrespective of the position inside NavigationToolbar2QT, and irrespective of the order of the functions. This is with PySide6-Essentials v6.6.0.

I put this code into https://github.com/bersbersbers/matplotlib-27214/commit/a3d97f8eb2b40a46fe2d4fc4cf60f9d631fd3cf2 if anyone wants to try it themselves.

My guess is multiple inheritance is related, but do not have a good theory as to why (other than general suspicion that MI always causes problems....).

If the metaprogramming approach works then let's just stick to it.