The way we have nsSVGContainerFrame set NS_STATE_SVG_NONDISPLAY_CHILD, and then have nsSVGDisplayContainerFrame inherit nsSVGContainerFrame is not great design. Worse, nsSVGDisplayContainerFrame and its sub-classes do not clear that bit which makes the code confusing when you're trying to understand how NS_STATE_SVG_NONDISPLAY_CHILD works. The (very easy to miss) trick is that nsSVGDisplayContainerFrame::Init calls nsSVGContainerFrameBase::Init, and _not_ nsSVGDisplayContainerFrameBase::Init, thereby skipping over the nsSVGContainerFrame::Init method which sets the NS_STATE_SVG_NONDISPLAY_CHILD bit. I assume the reason for this ugly way of doing things was to avoid adding extra virtual Init methods to the frames that need NS_STATE_SVG_NONDISPLAY_CHILD set. However, we can set the bit in their ctors instead, which is what we should do I think. Setting the bit on the frames than need it helps clarify which frames have this bit set rather than obfuscating it, and in fact it allows us to _remove_ the virtual Init method on nsSVGContainerFrame.

This adds handling for the <defs> and unknown SVG element case, which I forgot.

https://hg.mozilla.org/mozilla-central/rev/2a4ff84e15cd https://hg.mozilla.org/mozilla-central/rev/eff92312950a