507550 - Invalidation issues resizing windows with scaled tiled backgrounds

Status: RESOLVED DUPLICATE of bug 509329

(Core :: CSS Parsing and Computation, defect, P2)

Description

[Jeff Walden [:Waldo]]

<Waldo> bz: so, I'm seeing hideous artifacts on Windows if I load <http://whereswalden.com/>, then <javascript:document.body.id="experimental";[].v>, then resize the window -- could it be a ireflow problem? I don't remember seeing this on OS X when testing background-size patches
<Waldo> screenshot: http://whereswalden.com/files/temp/artifacts.png (note a few rects with proper rendering showing through where another window in the OS obscured the page at one point)

bz sez doesn't look like ireflow, maybe invalidation, so looks like I get to investigate and determine what's wrong.

Comment 1

[Jeff Walden [:Waldo]]

This doesn't reproduce in OS X, hence why I never saw it during my own testing.

Comment 2

[Jeff Walden [:Waldo]]

Part of the story appears to be nsFrame::CheckInvalidateSizeChange.  If I breakpoint in that when resizing the window, I get the interesting result that aOldRect has zero height and apparently-reasonable width.  Of course, invalidating that (with a tweak to the background loop in that method, to accommodate contain/cover) doesn't have much effect, and if I hop up a couple levels in the stack I don't see anything in nsBlockFrame::Reflow that invalidates anything else in view (assuming that you start with about:blank following the instructions in comment 0).  The method has comments that claim aNewDesiredSize - aOldRect should have been invalidated elsewhere, but I don't see where that would happen.

Comment 3

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Maybe more info on the problem in bug 509329? (which might be the same thing?)

Comment 4

[Jeff Walden [:Waldo]]

(As I noted in bug 509329 I plan to get to this next week.)

Comment 5

[Jeff Walden [:Waldo]]

Attachment: Reduced testcase

Here's a testcase reduced from the original errors.  With respect to the proto-patch in bug 509329, two bits are important: first, it's a fixed background, and second, the body has a width.  Remove those, or the latter, and with that first patch there this testcase will work correctly.

Comment 6

[Jeff Walden [:Waldo]]

This bug is cross-platform; its visibility seems to depend on the frequency of OS-level whole-window invalidates.  OS X invalidates often enough that this never shows up there, but Linux and Windows are more judicious and thus demonstrate problems during (and for Windows, after) window resizing.

Comment 7

[Jeff Walden [:Waldo]]

Fixed in bug 509329.