There is the following comment at
// FIXME: If this SVG ends up loading itself, we might leak the world.
// The comment said that the Cache code does not know about CachedImages
// holding Frames and won't know to break the cycle. But
I tested SVG images with various kinds of content, running testshell under valgrind, but I could not see any leak.
I believe that comment is no longer applicable. (It was written in 2009.)
Created attachment 84040 [details]
bug 15443 suggests that images loading images has not been implemented yet. If so the comment could be true when it is implemented couldn't it?
(In reply to comment #2)
> bug 15443 suggests that images loading images has not been implemented yet. If so the comment could be true when it is implemented couldn't it?
SVG images loading other SVG images, via the <image> tag, is supported, albeit in a limited manner.
You can see in the layout test that I submitted that it works. The layout test is, in fact, meant to be run under valgrind, to show that it does not leak memory.
I wonder what happened, the bubbles have been purple for the whole day.
Created attachment 84626 [details]
Forgot to use option --binary in git-diff last time. That explains why the previous patch failed to apply.
Instead of removing that comment altogher, I will submit a patch explaining why it cannot currently happen, and when (i.e. at what event in the future) it will happen.
Created attachment 84636 [details]
Comment on attachment 84636 [details]
Clearing flags on attachment: 84636
Committed r80371: <http://trac.webkit.org/changeset/80371>
All reviewed patches have been landed. Closing bug.