The actual issue is most likely https://bugs.webkit.org/show_bug.cgi?id=47661. We have now a more strict decoding of base64 data with the recent land of the soupURILoader stuff. That's the reason why this test started to fail.
Created attachment 70730 [details] Fix for the failing test Reverted to old behaviour
Comment on attachment 70730 [details] Fix for the failing test r=me
Comment on attachment 70730 [details] Fix for the failing test Clearing flags on attachment: 70730 Committed r69775: <http://trac.webkit.org/changeset/69775>
All reviewed patches have been landed. Closing bug.
Note that Safari (or more precisely, CFNetwork) is strict in this respect - "data:image/png;base64,iVBORw0KGrkJggg==" fails to load, while "data:image/png;base64,iVBORw0KGrkJggg=" loads fine (but then fails to render, since the data itself is of course broken). This matches Firefox.
So perhaps instead of loading the URL, we just need to fail in a better way?
(In reply to comment #6) > So perhaps instead of loading the URL, we just need to fail in a better way? Yeah. We reverted the change in order to mimic the old behaviour but I think it'd be better to fail in a better way