Animation of "width" on <svg> does not cause a repaint.

Oops, that should be setTimeAndSnapshot(1, true), not setTimeAndSnapshot(5, true).

Should/does attributeType="XML" make any difference?

Not on inner-<svg>, since the CSS property 'width' isn't relevant there. The underlying change does happen, but this is just an invalidation issue. If you resize the window then it updates correctly. In fact even on outer-<svg> attributeType wouldn't be necessary, since the CSS property 'width' defaults to 'auto' effectively allowing the XML attribute to be "used".

Does setting the width attribute from script invalidate correctly? If so, where does that happen?

As it happens, no, setting the 'width' attribute on inner-<svg> doesn't work either. That's really a separate bug and a regression from bug 545550 though, so I've filed bug 571969 for that. Regarding this bug, probably the issue is that this comment is now wrong: http://hg.mozilla.org/mozilla-central/annotate/7e685b61a291/layout/svg/base/src/nsSVGInnerSVGFrame.h#l59 since nsSVGElement::DidAnimateLength triggers invalidation by calling AttributeChanged on the element's frame. The strange thing is that when that comment was added (see bug 301628 comment 16) nsSVGSVGElement::DidModifySVGObservable didn't do anything explicit with width/height. I guess it did fall through blindly and call NotifyViewportChange though, but that didn't take account of the viewport established by the inner-<svg>. We should do that when we add nsSVGInnerSVGFrame::AttributeChanged to fix this bug. If someone doesn't get to writing a patch before me, I'll make one after I've fixed the length list bug.

http://bonsai.mozilla.org/cvsview2.cgi?command=DIFF&subdir=mozilla%2Flayout%2Fsvg%2Fbase%2Fsrc&file=nsSVGInnerSVGFrame.cpp&rev1=1.23&rev2=1.24&whitespace_mode=show&diff_mode=full Is the state of nsSVGSVGElement.cpp back when the nsSVGInnerSVGFrame.h comment was added.

You've been a long time with this Jonathan.

Comment on attachment 481158 [details] [diff] [review] updated patch with x, y fix Looks good!

(Needs reftest(s) before landing, though.)

low risk regression from firefox 3.6 with reftest

Probably worthy of blocker status (rather than needing approval), since this is a rendering regression w.r.t. Gecko 1.9.2 / Firefox 3.6.

Landed: http://hg.mozilla.org/mozilla-central/rev/1529931ada9e