[canvaskit] read pixels back in Picture.toImage

[yjbanov]

This undoes the optimization made in #38573, and adds a test-case that would have failed when that optimization was made. The issue is that CanvasKit loses track of which resources belong to which context that leads to WebGL errors followed by UI corruption.

Fixes flutter/flutter#121758

[jonahwilliams]

Nit: this says its about toByteData, but the changes are in toImage.

Is there a way to defer the usage of MakeRasterDirectSurface until we're calling toByteData? ITs more or less expected that toByteData can be slow, but toImageSync is useful for composite effects using fragment shaders - and running those in software is not going to be feasible.

[yjbanov]

Nit: this says its about toByteData, but the changes are in toImage.

Is there a way to defer the usage of MakeRasterDirectSurface until we're calling toByteData? ITs more or less expected that toByteData can be slow, but toImageSync is useful for composite effects using fragment shaders - and running those in software is not going to be feasible.

Couldn't find a way to defer it. I'm going to start a thread with the Skia team to figure this out. In the meantime, I'm reverting #38573 and adding a new unit-test that catches the issue. Hopefully, next time we attempt it, it won't regress.

[yjbanov]

For posterity, I tried the following solutions that didn't work:

• Use CanvasKit.MakeRasterDirectSurface - this fixes [web:canvaskit] Image.toByteData loses track of WebGL contexts corrupting rendering pipeline state flutter#121758 but it regresses texture-backed images.
• I tried getting the VideoFrame pixels so SkImages can be instantiated from in-memory pixels - that fixed both problems, but it required that I manually convert YUV encoded images to RGBA. Due to Orientation metadata for VideoFrame w3c/webcodecs#351 there's no way to tell if images need to be reoriented during the RGBA conversion, so some of them came out wrong. This method is also likely to slow down image decoding (a more common use-case than Picture.toImage), so it was only good for experimentation.

[harryterkelsen]

LGTM

[flutter-dashboard]

Golden file changes have been found for this pull request. Click here to view and triage (e.g. because this is an intentional change).

If you are still iterating on this change and are not ready to resolve the images on the Flutter Gold dashboard, consider marking this PR as a draft pull request above. You will still be able to view image results on the dashboard, commenting will be silenced, and the check will not try to resolve itself until marked ready for review.

Changes reported for pull request #40004 at sha d679b87