Fix alpha bleeding when smooth-scaling images

[tronical]

After loading for example PNGs with an alpha, prefer passing it to renderers as pre-multiplied alpha, so that nearby transparent pixels (r=g=b=a=0) don't contribute anything as pre-multiplication will zero out their r/g/b.

Fixes #10469

[ogoffart]

Premultiplied is probably the right format to store the cached rendered image. But while this works for image loaded from file, it would still be a problem for image created by the user with the ARGB (not premultiplied) format.

[tronical]

Premultiplied is probably the right format to store the cached rendered image. But while this works for image loaded from file, it would still be a problem for image created by the user with the ARGB (not premultiplied) format.

Yeah, I was thinking the same. But at least there folks have a choice :)

[ogoffart]

Could we have a screenshot test with an image that highlights the wrong behavior?

[tronical]

Could we have a screenshot test with an image that highlights the wrong behavior?

That turns out to be rather difficult, given that already the smooth scaling with pre-multiplied alpha seems to result in differences with skia :-(

[tronical]

I'm trying a new PR that disables the smooth scaling for these tests - which also sucks :-(

[ogoffart]

smooth scaling with pre-multiplied alpha seems to result in differences with skia :-(

Ah oops. This might be why some test were skipped before:

slint/tests/screenshots/cases/image/images.slint

Lines 5 to 6 in 5c21a83

	// Skia doesn't have identical rendering across platforms
	//ignore:skia

(and more tests have that)

We could of course skip more tests but that'd be unfortunate. Or have different threshold for skia.

[ogoffart]

I'm trying a new PR that disables the smooth scaling for these tests - which also sucks :-(

Ah good idea.