Crash when undoing/redoing Image component scale changes

[tassaron]

The undo history window (QUndoStack) sometimes causes a segfault when rapidly clicking to undo a bunch of changes to the "scale" setting of the Image component. It's not actually a new bug (it exists in 2.0.0 as well), but I just discovered it.

It is probably a similar cause as one previous segfault I fixed related to Pillow images, so I'm optimistic I'll figure it out soon. Specifically the problem occurs within ComponentUpdate, a QUndoCommand used for (almost) every component setting change (in src/avp/component.py)

A more common crash related to the same core issue occurs on the Python side:

Traceback (most recent call last):
  File "/home/b/code/audio-visualizer-python/src/avp/component.py", line 951, in redo
    self.setWidgetValues(self.modifiedVals)
  File "/home/b/code/audio-visualizer-python/src/avp/component.py", line 941, in setWidgetValues
    else self.parent.pixelValForAttr(attr, val)
  File "/home/b/code/audio-visualizer-python/src/avp/component.py", line 735, in relativeWidgetAxis
    return func(self, attr, *args, **kwargs)
  File "/home/b/code/audio-visualizer-python/src/avp/component.py", line 753, in pixelValForAttr
    log.verbose(
  File "/home/b/code/audio-visualizer-python/src/avp/__init__.py", line 26, in verbose
    if self.isEnabledFor(5):
RecursionError: maximum recursion depth exceeded
Aborted                    (core dumped) avp

This happens often when scrolling up and down with my mousewheel on the "scale" setting of the Image component. The problem occurs because the Image component defines its scale widget to be a "relative widget" so it gets handled specially (especially brokenly, it turns out).

I'm hoping to get this fixed within a couple of weeks, because it's a good excuse to re-familiarize myself with sections of the codebase I want to refactor the most in future (after writing more tests).

[tassaron]

The source of this bug was the fact that the Image component had two spinboxes used for the scale setting, one for regular scale and one for scale when the image is stretched. Which had two problems:

1. The Image component didn't block Qt signals when updating one spinbox to match the other (could have been easily fixed by adding the blockSignals context manager), which triggered the relativeWidgets system and sometimes resulted in infinite recursion depending on how fast different threads did work

2. The undo system didn't merge scale-related actions correctly as two actions would occur in quick succession due to the first problem, and undoing would have side effects because of the relativeWidgets...

Basically, it was a bizarre design decision I made 10 years ago when creating the Image component that resulted in two complex systems combining in a confusing way. relativeWidgets is confusing enough as it is: any relative widget will have its value changed automatically when output resolution changes so that the appearance looks the same. This works okay, but really doesn't mesh easily with components trying to pull shenanigans by overriding what should be simple methods (update, savePreset)

I've fixed both the recursion and the segfault in a branch, and I'm going to change the Image component to use ImageOps which has better options for scaling the image while maintaining aspect ratio. Then probably merge and close this issue by tomorrow.