Looks like same issue as here: https://bugs.kde.org/show_bug.cgi?id=343285 And here: https://bugs.kde.org/show_bug.cgi?id=343365 Though for a 8 bit/int picture this time. Basic selection (done with Select opaque command): http://i57.tinypic.com/6q9lxe.jpg Inverted selection: http://i59.tinypic.com/7asl.jpg I know there may be some not-perfectly-opaque parts there, but definitely not THAT many... Reproducible: Always Steps to Reproduce: 1. Create partially transparent 8 bit/int picture. 2. Select the opaque part with Select opaque command under R-click menu on the layer list. 3. Invert the selection (Ctrl+Shift+I). Actual Results: Selection changes shape. A lot. Expected Results: Selection gets simply inverted. A file I was working with when met the issue: https://www.dropbox.com/s/hq2dq5m4u5rd0lx/couronne.kra?dl=0
To make things less clear - after several repeats of the above now* it works correctly... (*now = after several saves and loads + a lot of changes in the image)
Just the displayed outline changes a lot, the selection itself is inverted correctly. The difference is just the way in which Krita defines the outline. Krita allows to have various degrees of selectedness. If a pixel is selected it will be inside of the outline no matter if it's 1% or 100% selected. If a pixel is 99% selected it will be inside the selection, after inverting it's 1% selected and still in the selection. Only inverting a 100% selected pixel will result in a totally unselected pixel. That exactly what happens in the image above.
So in a word there's no way to get inverted selection only inverted "selectedness"? Quite sad, since it makes a lot of things impossible to achieve... Guess I'll have to add it to wishlist then.
We implemented an anti-alias checkbox so this can be worked around(so you can get sharper selections)