Jeff Muizelaar [:jrmuizel] Reporter
	Description • 7 years ago

It would be nice if we could avoid this:
https://perfht.ml/2ApmBjH

	Ethan Lin[:ethlin]
	Comment 1 • 7 years ago

I think we don't need to clear invalidation state bits for the whole frame tree. We just need to clear the bits for fallback items.

	Ethan Lin[:ethlin]
	Comment 3 • 7 years ago

I can't reproduce comment 1, but I think this way should be correct. Jeff, can you check if the patch helps the performance?

	Ethan Lin[:ethlin]
	Comment 4 • 7 years ago

Comment on attachment 8933601 [details]
Bug 1422032 - Clear the invalidation state bits for fallback items only but not for all frames.

Ah..we also check the flag in nsIFrame::InvalidateFrameWithRect. I should find another way to erase the flag efficiently.

	Ethan Lin[:ethlin]
	Comment 5 • 7 years ago

For WR, we only need to clear the bits for fallback items and all parent frames of the fallback items. So we need a special clear function for WR.

	Ethan Lin[:ethlin]
	Comment 6 • 7 years ago

I did some test and studies. It looks like we should clear the flag for either non-fallback and fallback frames since frame tree also checks this flag. So we shouldn't partially clear the flag. Matt, can you give some suggestions if we want to optimize this part?

	Jeff Muizelaar [:jrmuizel] Reporter
	Comment 7 • 7 years ago

We talked about this in Austin. In the facebook-refresh case we're setting a flag on the body element and we need to then propagate this down to all descendants. We could have, but don't currently have the infrastructure to avoid doing this.

The style change could propagate down into a blob image so we basically need to have this for blob image invalidation to work properly.

Because we set the flag on all of the descendants we end up having to traverse the whole frame tree to clear the flag.

	Jeff Muizelaar [:jrmuizel] Reporter
	Comment 8 • 7 years ago

This isn't needed for the initial release.