Previously, an image that failed to load (m_failed == true) would trigger an assertion failure when WebKit tried to get its size.
Created attachment 44338 [details]
(contains a binary file, so please don't cq+!)
Comment on attachment 44338 [details]
> // Zero-height images can cause problems for some ports. If we have an
> // empty image dimension, just bail.
> - if (size().isEmpty())
> + if (isSizeAvailable() && size().isEmpty())
> return 0;
This changes the logic. If the size is not available, this now continues and creates a native image. But the old code would return 0 if the size was not available. You should reverse the logic.
if (!isSizeAvailable() || ...
Created attachment 44503 [details]
I originally thought that size() would return random memory in the case that it hadn't been set, but now I notice that it's constructed to 0, 0. So I've changed the logic as suggested.
Created attachment 44504 [details]
Crap. Wrong patch. Corrected.
Attachment 44504 [details] was posted by a committer and has review+, assigning to Adam Langley for commit.
Comment on attachment 44504 [details]
This can't be cq'd because it's missing the binary data.
This test seems to be (at least recently) crashing on Gtk.