While re-doing the image bases for https://bugs.webkit.org/show_bug.cgi?id=21821, we noticed that these pixel tests are currently failing (SVG does not seem to be rendering): media/video-aspect-ratio.html media/video-controls-rendering.html media/video-display-toggle.html media/video-layer-crash.html media/video-transformed.html
Ignore the SVG comment, that was copy-paste error...
Still happens on r38790 and the tests work fine when run directly in Safari. There's however a delay of about half-a-second before the image appears in the video element.
Created attachment 25545 [details] Patch v1
Created attachment 25639 [details] Patch v2
Sending LayoutTests/ChangeLog Sending LayoutTests/media/video-aspect-ratio.html Sending LayoutTests/media/video-controls-rendering.html Sending LayoutTests/media/video-display-toggle.html Sending LayoutTests/media/video-layer-crash.html Sending LayoutTests/media/video-paint-test.js Sending LayoutTests/media/video-transformed.html Sending LayoutTests/platform/mac/media/video-aspect-ratio-expected.checksum Sending LayoutTests/platform/mac/media/video-aspect-ratio-expected.png Sending LayoutTests/platform/mac/media/video-controls-rendering-expected.checksum Sending LayoutTests/platform/mac/media/video-controls-rendering-expected.png Sending LayoutTests/platform/mac/media/video-display-toggle-expected.checksum Sending LayoutTests/platform/mac/media/video-display-toggle-expected.png Sending LayoutTests/platform/mac/media/video-layer-crash-expected.checksum Sending LayoutTests/platform/mac/media/video-layer-crash-expected.png Sending LayoutTests/platform/mac/media/video-transformed-expected.checksum Sending LayoutTests/platform/mac/media/video-transformed-expected.png Transmitting file data ................. Committed revision 38879.
The new bases do not work on the pixel bots (images are too dark for some unknown reason).
Created attachment 25647 [details] Patch v3 Here are new bases generated on the bot itself. I'm not sure why the bases I generated on my machine are not matching what renders on the bot: the version of WK is really close, it's the same QT 7.5.5 and same HW. The bot is 10.5.5 and my machine 10.5.4 however. One thing to notice is that it looks as if QT was rendering the frames in the wrong colorspace. Considering DRT changes temporarily the display colorspace, and the script launches and stop DRT a few times, maybe we have some kind of race condition and QT is not picking the right colorspace. It's either that, or a difference in display colorspace & QT behavior in 10.5.5 versus 10.5.4.
Sending LayoutTests/ChangeLog Sending LayoutTests/platform/mac/media/video-aspect-ratio-expected.checksum Sending LayoutTests/platform/mac/media/video-aspect-ratio-expected.png Sending LayoutTests/platform/mac/media/video-controls-rendering-expected.checksum Sending LayoutTests/platform/mac/media/video-controls-rendering-expected.png Sending LayoutTests/platform/mac/media/video-display-toggle-expected.checksum Sending LayoutTests/platform/mac/media/video-display-toggle-expected.png Sending LayoutTests/platform/mac/media/video-layer-crash-expected.checksum Sending LayoutTests/platform/mac/media/video-layer-crash-expected.png Sending LayoutTests/platform/mac/media/video-transformed-expected.checksum Sending LayoutTests/platform/mac/media/video-transformed-expected.png Transmitting file data ........... Committed revision 38886.
To conclude on the above issue, my machine actually had a custom version of QT installed instead of the regular 7.5.5. This very likely explains why the bases I checked in were incorrect. However, the new bases are randomly failing on the bot: https://bugs.webkit.org/show_bug.cgi?id=22593
Finally! I found the exact reason for the difference between my machine and the bot: I have Perian installed and the bot doesn't. If I remove Perian from my machine, then the video test.mp4 renders with the same brightness as on the bot. Here's a new bug that tracks this issue: https://bugs.webkit.org/show_bug.cgi?id=22615