Fix Control rect coordinate system inconsistency

[Sauermann]

Fix get_rect, get_global_rect and get_screen_rect to take Controls scale into account.

supersedes & extracted from #37765 authored by @ThakeeNathees (no response in a year)
Also implemented comments and additionally simplified get_screen_rectand get_screen_position.

related to #66602

Current behavior is, that the returned values are in different coordinate systems.

• position in canvas/screen
• size in local space

I would expect, that a function returns values in the same coordinate system.
MRP: BugGlobalRectScale.zip

Updated 2022-10-22: Fix merge conflict.

[kleonc]

Note that with this implementation of get_rect the returned size is still not necessarily in the parent's (or containing node's) coordinate space. Control's local to parent transform (result of get_transform()) is composed of (right to left):
translation(position) * translation(pivot_offset) * rotation * scale * translation(-pivot_offset).

Meaning return Rect2(get_position(), get_scale() * get_size()) is correctly in the parent space only if: (rotation is zero) and ((pivot_offset is zero) or (scale is default)).

It can be changed to:

Transform2D t = get_transform();
return Rect2(t.get_origin, t.get_scale() * get_size())

which should be in the parent space when rotation is zero.

For non-zero rotation the position obtained from get_transform() will be correctly in the parent space but the size calculated like that won't be meaningful (since get_size() is in the Control's local space and t.get_scale() is in the parent's space and these are oriented differently). Similar is true for other similar methods like get_global_rect or get_screen_rect. So it's probably worth adding notes to these that the resulting size is not meaningful in case of non-default rotation.

Another thing which might be worth noting is that what Control::get_rect returns is unaffected by the gui/common/snap_controls_to_pixels project setting and hence it can differ a little from the actual rendering result in case this setting is turned on (same for global/screen).

Also another thing is there's Sprite.get_rect method which returns local rect, so there's naming discrepancy. I wrote plenty why the local rect is important and that's what it should return in #65112. So I think we should rename Sprite.get_rect -> Sprite.get_local_rect and probably add Control.get_local_rect returning Rect2(Vector2(0, 0), get_size()).

[Sauermann]

@kleonc thanks for your detailed analysis. I will go through your notes and adjust my PR accordingly.

Regarding your suggestion Control::get_local_rect, it should be noted, that there is already Control::get_anchorable_rect, which returns exactly this.

[Sauermann]

I have updated the PR with the comments.
I found of note, that if the node itself or any parent CanvasItem between the node and the canvas have a non default rotation or skew, the resulting size for get_global_rect and get_screen_rect is likely not meaningful.

[groud]

Approved in PR review meeting.

[akien-mga]

Thanks!