730504 - Runfield has bad invalidation behaviour

Status: NEW

(Core :: Graphics, defect)

Description

[Jeff Muizelaar [:jrmuizel]]

Currently on Maple we often invalidate runfield twice per draw call. This has a number of causes which I'll file a dependent bugs.

Comment 1

[Jeff Muizelaar [:jrmuizel]]

This was revealed by 729537 which caused the nsHTMLScrollFrame to have a display port. 

Here's how it all happens:
 - Canvas does not properly clip invalidations to the size of the canvas (bug 730508)
 - Previously, these invalidations were clipped by the scroll port of the nsHTMLScrollFrame, but with a display port they are not.
 - These invalidations outside cause us to get an invalid region that's not a rect.
 - This region is split into rects and passed to nsWindow::Invalidate (bug 730509)
 - Each Invalidate() call queues up a Draw event.
 - Between Draw events, we can get NATIVE_POKE events in the queue (e.g. from the TimerThread)
 - Draws are only coalesced if they are contiguous (bug 730515)
 - Each draw will schedule a composite.
 - The following draws will upload the entire unchanged canvas
 - Sadness