Currently we only inject nsDisplayLayerEventRegions for stacking and pseudo-stacking contexts. With containerless scrolling, this means that all but the first processed layer as part of a scrollframe will be missing event regions. <roc> dvander: in BuildDisplayListForChild, we want to push a new nsDisplayEventRegion when nsLayoutUtils::GetAnimatedGeometryRootForFrame returns a different animated geometry root for the parent than for the child <roc> Before actually shipping that code I think we need to do some work to optimize GetAnimatedGeometryRootForFrmae

This adds an nsDisplayLayerEventRegions item for non-stacking contexts, if they have a different animated geometry root than their parent. The current animated geometry root for a frame is now cached as well. I had to move some stuff from nsLayoutUtils to nsDisplayList, there was one material change from that. Previously we could return an animated geometry root in between a sticky frame and its nearest scrollframe, and that's been fixed.

Comment on attachment 8525735 [details] [diff] [review] bug1101260-event-regions.patch Review of attachment 8525735 [details] [diff] [review]: ----------------------------------------------------------------- Can you split the patch into "cache animated geometry roots" and "create more nsDisplayEventRegions" please? Thanks!

Comment on attachment 8525759 [details] [diff] [review] part 1, cache the current animated geometry root Review of attachment 8525759 [details] [diff] [review]: ----------------------------------------------------------------- Excellent!!!

We should go further and actually store the animated geometry root in each display item. This would reduce overhead in FrameLayerBuilder. But that shouldn't be in this bug.

Okay, filed bug 1102676 as a follow-up https://hg.mozilla.org/integration/mozilla-inbound/rev/42c5ee6094ee https://hg.mozilla.org/integration/mozilla-inbound/rev/b4bff1b4f793

https://hg.mozilla.org/mozilla-central/rev/42c5ee6094ee https://hg.mozilla.org/mozilla-central/rev/b4bff1b4f793