Fixed issue in TextureAtlas import of images with wrong size.

[dankan1890]

Fix #42057

[akien-mga]

For the reference, last commit on this code was #39271 (@RandomShaper).

[akien-mga]

The fix seems to work fine, and I could reproduce the issue before applying it.

For the reference, last commit on this code was #39271 (@RandomShaper).

I checked and #39271 did not cause this issue though, it's already present in 3.2.1-stable.

[RandomShaper]

I'm not sure this fix is totally correct. I think that if the <s are changed to <=s, the the - 1s applied to width and height - p_offset.y should be removed, since otherwise the cases where the right and/or bottom edges of the atlas are reached the last 1-pixel row or column would be left transparent.

However, I haven't tested it. and I'm not sure I'll have time reasonably soon. If one of you was able to test it, that'd help.

[akien-mga]

However, I haven't tested it. and I'm not sure I'll have time reasonably soon. If one of you was able to test it, that'd help.

It does solve the problem in #42057, but it would likely be worth testing on more configurations to see how it behaves.

This code is not the easiest to parse and indeed, it seems a good spot for having off-by-one errors here and there which might counterbalance each other.

[boruok]

@akien-mga can this PR be merged in 3.x branch? compiled it with 3.3.3-stable and it does solve #42057 issue.

[akien-mga]

I've tested that the changes fix the bug last time - couldn't get anyone to review the code change in depth as it probably gives everyone a headache, me included. Let's merge and see if we spot any regression in testing builds.

[akien-mga]

Thanks!

[akien-mga]

Cherry-picked for 3.4.