Unmapped windows with size constraints need to adjust them for SSDs

[AlanGriffiths]

Fixes: #3778

[tarek-y-ismail]

So if I understand correctly, for umapped windows (basically, hidden or minimized windows), this assumes they're 640x480? What happens when they're mapped? I assume they can override this size, right?

[AlanGriffiths]

So if I understand correctly, for umapped windows (basically, hidden or minimized windows), this assumes they're 640x480? What happens when they're mapped? I assume they can override this size, right?

The case encountered with Qt was "no buffer submitted yet" + "min & max size set"

[tarek-y-ismail]

The case encountered with Qt was "no buffer submitted yet" + "min & max size set"

Huh, very weird. So I guess the application is telling us to choose a size between the min and max? If so, shouldn't we take that into account when adjusting?

[AlanGriffiths]

Huh, very weird. So I guess the application is telling us to choose a size between the min and max? If so, shouldn't we take that into account when adjusting?

Not weird. Unlike X11 toolkits Wayland ones often commit() before attaching buffers. What was new to us was that the min/max size was set and we were using SSDs.

The existing code only adjusted the min/max size when we also had a size (from the submitted buffer)

[tarek-y-ismail]

The existing code only adjusted the min/max size when we also had a size (from the submitted buffer)

Oh, ho! So this is so we execute these statements?

mir/src/server/shell/abstract_shell.cpp

Lines 95 to 102 in 02c7506

	if (wm_relevant_mods.max_width.is_set())
	wm_relevant_mods.max_width.value() += horiz_frame_padding;
	if (wm_relevant_mods.max_height.is_set())
	wm_relevant_mods.max_height.value() += vert_frame_padding;
	if (wm_relevant_mods.min_width.is_set())
	wm_relevant_mods.min_width.value() += horiz_frame_padding;
	if (wm_relevant_mods.min_height.is_set())
	wm_relevant_mods.min_height.value() += vert_frame_padding;

[AlanGriffiths]

Oh, ho! So this is so we execute these statements?

Yes, we needed to relax the condition on L232. After which we also fake any values that are not currently set. We already handle the client modifying things after that

[tarek-y-ismail]

Alright. Now makes sense. Would prefer if we handle cases like these explicitly but I'm sure the code would be more spaghettified with all the jank clients throw at us :)