Remove or shrink unnecessary page rerender calls

[hrdl-github]

This requires #4158 that requires fewer explicit redraws when layers are deleted or their visibility is modified. Addresses #4580 in combination with #4158. This reduces the number of page repaints required. Full repaints are still used upon

• finalisation of text edits
• layer changes
• initialisation

TODO:

• More extensive testing
• Remove debug messages

[hrdl-github]

This requires #4158 that requires fewer explicit redraws when layers are deleted or their visibility is modified.

I think I was wrong about this one, as they behave the same now. This means that this PR is independent of #4158. This PR requires #4158 after all to rerender layers in the correct order.

I've restored repainting at a few points where it is needed.

[hrdl-github]

Do you have any idea why a repaint would be performed here?

xournalpp/src/core/control/layer/LayerController.cpp

Lines 333 to 334 in 1038ec5

	// Repaint page
	control->getWindow()->getXournal()->layerChanged(selectedPage);

If it's not needed I would just remove it to save a repaint.

[hrdl-github]

There are some additional buttons triggering ACTION_GOTO_{NEXT,PREVIOUS,TOP}_LAYER, which change the selected page AND some layers' visibility. I don't think a repaint is needed if the selection but not the visibility changes, hence 3228571.

[hrdl-github]

After a bit more testing this weekend I don't see any more issues.

[bhennion]

To be honest, I'm a bit conflicted about this PR.
One the one hand, it seems to work fine without the repainting you removed. On the other hand, what's the point of refreshing the page buffer if it's not to display it?

For instance, when an EditSelection is created (meaning some Elements just got selected), the Elements are:

• removed from the page, added to a selection, painted to a selection buffer, and all of that is then painted on screen in the main loop
• erased from the page buffer (via rerenderPage() which is an overkill) in a separated thread, followed by a full repaint.

With this PR, you removed the asynchronous full repaint, while I think the synchronous repaint is the superfluous one: we should wait for the page buffer to be updated before repainting.

That's just an example, and I think the other instances are similar. At this point, I can't imagine a situation where the page buffer needs updating without a follow up repaint being needed.

[hrdl-github]

Thanks, this makes perfect sense. Instead of complicating the repainting logic I could just remove the call to rerenderPage() where it is not needed and do a more limited rerendering where necessary, e.g. in the EditSelection logic, which I indeed managed to break.

[hrdl-github]

I've removed the rerenderPage() calls that I don't think are needed and replaced the ones in the EditSelection constructor with range-based calls.

[bhennion]

LGTM!
Squashing and merging in 48 hours unless an objection is raised.