592941 - Fix performance regression from bug 579323

Status: RESOLVED FIXED

(Core :: Layout, defect)

Description

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Attachment: fix

Turns out that bug 579323 tripped over an existing bug: we failed to create a ThebesLayerInvalidRegionProperty() when setting up a new container layer in StoreNewDisplayItemData. This means that the next time we paint, in the new code in bug 579323 we'll find no ThebesLayerInvalidRegionProperty and repaint all the ThebesLayer children. (But after that, UpdateDisplayItemDataForFrame will set up the ThebesLayerInvalidRegionProperty so this won't happen again.) This is obviously very bad. Also, there's an existing problem which is that invalidates passed to FrameLayerBuilder::InvalidateThebesLayerContents between the first and second paint are currently being dropped on the floor!

Comment 1

[Robert O'Callahan (:roc) (email my personal email if necessary)]

http://hg.mozilla.org/mozilla-central/rev/15ae10f019c2

One Linux number has come in so far, the regression is gone:
http://graphs.mozilla.org/graph.html#tests=[[71,1,433],[71,1,434],[71,1,435],[71,1,436],[71,1,437],[71,1,438],[71,1,439],[71,1,440],[71,1,441],[71,1,442],[71,1,443],[71,1,444],[71,1,445],[71,1,446],[71,1,447],[71,1,448],[71,1,449],[71,1,450],[71,1,688],[71,1,689],[71,1,690],[71,1,691],[71,1,692],[71,1,693],[71,1,694],[71,1,695],[71,1,696],[71,1,697],[71,1,698],[71,1,699],[71,1,700],[71,1,701],[71,1,702],[71,1,703],[71,1,704],[71,1,705],[71,1,706],[71,1,707],[71,1,708],[71,1,709],[71,1,710],[71,1,711],[71,1,712],[71,1,713],[71,1,714],[71,1,715],[71,1,716],[71,1,717],[71,1,895],[71,1,896],[71,1,897],[71,1,898],[71,1,899]]&sel=1283076625,1283405619