When loading URL in a mozilla-central debug build, I get many copies of: ###!!! ASSERTION: Resample dirty flag set during sample!: '!mResampleNeeded', file ../../../mozilla/content/smil/nsSMILAnimationController.cpp, line 387 assertion-spam from full-animate-elem-30-t.html. It looks like we get one assertion-failure per SMIL sample. I think the strategy described at the end of bug 483584 comment 12 would fix this. To summarize, I'm pretty sure the problem here is that in nsSVGTransformSMILAttr::SetAnimValue, we're calling DidModify to issue notifications for the transform attribute changing, and that makes us request a new SMIL sample (while we're already sampling). Instead of DidModify, we should be calling a new method ("DidAnimateTransform"?) that would trigger the necessary invalidation & repainting, but that would NOT make us request a new sample (nor would it make us re-clone the content for the <use> element).

Here's a reduced testcase that reproduces the assertion-spam about 50% of the times that I reload it. (The nondeterminism is probably due to the random ordering in which we traverse our animation targets during a sample.)

This just adds a "DidAnimateTransform" method, just like all the other DidAnimateXXX methods, and it appears to work fine.

(In reply to comment #2) > and it appears to work fine. That is to say: it fixes the assertions (since we no longer notify), and it doesn't break any tests -- passed a TryServer run, earlier today.

pushed: http://hg.mozilla.org/mozilla-central/rev/6776f09c1bc6