GraphicsContext::drawImageBuffer has logic inside it to set up a FloatSize properly based on (tw, th) and (tsw, tsh) if the dimensions are -1, -1 for the source rect. However, these variables are never used and the (-1, -1) rect is passed to ImageBuffer anyway. This is clearly in contrast with, for example, drawImage.
Created attachment 81536 [details] Send correctly sized rects in drawImageBuffer Send the correct sized rects in drawImageBuffer to the ImageBuffer.
Comment on attachment 81536 [details] Send correctly sized rects in drawImageBuffer Does this change have any effect? How can we test that the fix is effective? Normally we do not accept code changes that fix bugs without regression tests.
(In reply to comment #2) > (From update of attachment 81536 [details]) > Does this change have any effect? How can we test that the fix is effective? Normally we do not accept code changes that fix bugs without regression tests. It's exactly the same logic as is applied in drawImage. This is clearly what the original author of drawImageBuffer intended, but forgot to set up the FloatRects accordingly (evident by the fact that drawImageBuffer already calculates th, tw, tsh and tsw but never actually uses them). This fixes rendering issues with certain versions of Skia.
(In reply to comment #3) > This fixes rendering issues with certain versions of Skia. Can you make a regression test which shows those issues, or are they with some version of Skia not available to us on buildbots?
(In reply to comment #4) > (In reply to comment #3) > > This fixes rendering issues with certain versions of Skia. > > Can you make a regression test which shows those issues, or are they with some version of Skia not available to us on buildbots? yes, it's a version of skia that's not available on the build bots
(In reply to comment #3) > (In reply to comment #2) > > Does this change have any effect? How can we test that the fix is effective? Normally we do not accept code changes that fix bugs without regression tests. > > It's exactly the same logic as is applied in drawImage. This is clearly what the original author of drawImageBuffer intended [...] > > This fixes rendering issues with certain versions of Skia. For the record, the requirement for a regression test is not any reflection on how good or obvious a fix is, or how silly the original error was. I think it’s OK for us to take this change that has no effect with any current configurations and is useful to Google in your work with newer fancier versions of Skia. But in the future, change log should explain why there is no test when you propose a change without a test.
(In reply to comment #6) > I think it’s OK for us to take this change that has no effect with any current configurations and is useful to Google in your work with newer fancier versions of Skia. But in the future, change log should explain why there is no test when you propose a change without a test. Heh, RIM, not Google.
Comment on attachment 81536 [details] Send correctly sized rects in drawImageBuffer review+ given that George assures me this causes no problems with any other ports, and is useful to RIM in some improved version of Skia they are working with
(In reply to comment #6) > For the record, the requirement for a regression test is not any reflection on how good or obvious a fix is, or how silly the original error was. > > I think it’s OK for us to take this change that has no effect with any current configurations and is useful to Google in your work with newer fancier versions of Skia. But in the future, change log should explain why there is no test when you propose a change without a test. Noted. Thanks!
Comment on attachment 81536 [details] Send correctly sized rects in drawImageBuffer cq+ as requested by gw280 on IRC
Comment on attachment 81536 [details] Send correctly sized rects in drawImageBuffer Clearing flags on attachment: 81536 Committed r77945: <http://trac.webkit.org/changeset/77945>
All reviewed patches have been landed. Closing bug.