We clobber the channel notification callbacks with the cache validator, which means the CORS code will never see the redirect. I don't know how to write a test that exercises this, sadly....

Comment on attachment 602096 [details] [diff] [review] Set up the image cache validator before the CORS listener. Review of attachment 602096 [details] [diff] [review]: ----------------------------------------------------------------- ::: image/src/imgLoader.cpp @@ -1294,5 @@ > > listener = corsproxy; > } > > - newChannel->SetNotificationCallbacks(hvc); You could also just change this to SetNotificationCallbacks(listener). Might need a QueryInterface, though, and thus might not be worthwhile. I don't know whether that'll cause trouble with an nsCORSListenerProxy that fails to initialize in its constructor, though; maybe Jonas can weigh in on that.

Comment on attachment 602096 [details] [diff] [review] Set up the image cache validator before the CORS listener. Review of attachment 602096 [details] [diff] [review]: ----------------------------------------------------------------- They way this patch does it is the right way. You have to set up the notifications callback before constructing the CORS proxy since the proxy ctor grabs the existing notification listener.

Oh, and I'd still love to know how to test this. I pushed to try, and at first glance it looks like this patch is leaking on reftests (or at least all debug reftests are orange with leaks of 3 nsCORSListenerProxy and 3 imgCacheEntry and 3 imgRequest). Oddly enough, none of the leaked nsStandardURLs are obvious CORS-enabled images: file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/ogg-video/poster-2.html file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/ogg-video/not-a-valid-file file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/webm-video/poster-2.html file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/webm-video/not-a-valid-file file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/svg/image/image-svg-inline-01.html There is also an nsSimpleURI leaked....

Actually, nevermind. That may be fallout from bug 696301. Though the fact that we get leaks there is .... disturbing.

Yes, indeed. The issue was bug 696301 plus a prexisting leak in imagelib. Filed bug 732319 on that.

From our read of this bug, we don't believe ESR 10 is affected. Please email if that's not the case.