522687 - Large scaled images will show white horizontal lines during rendering

Status: RESOLVED DUPLICATE of bug 703231

(Core :: Graphics, defect)

Description

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Attachment: screenshot

Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-US; rv:1.9.3a1pre) Gecko/20091015 Minefield/3.7a1pre ID:20091015030945

While running some layout tests from Litmus I noticed that one of the tests failed. Part of this test is to check if large images are scaled correctly for different actions. When I load such a file (see URL) the image shows up white horizontal lines during the rendering process. Those lines persist until you place another window in-front of the current one and switch back, or just switch the tabs. See the attachment how it looks like for me.

It's really hard to reproduce. I have figured out that copying two files into a new profile shows up the same issue again. Sadly this trick does not work on another macbook. I can only see it on my MacBook Pro with the accelerated graphics enabled. I will attach both files with the hope it can be reproduced by someone.

Further I did a regression test and the causing fix is on bug 489389 where we reduce the temporary surface for tiling. The checkin was between Firefox 3.5 beta 4 and beta99.

Comment 1

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Attachment: example files

For me the following steps reproduce the problem:

1. Create a new profile
2. Copy the files from this attachment into the profile folder
3. Start Firefox

After step 3 the loaded image shows the white stripes.

Comment 2

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Running in the debugger shows me the following assertion when having the tab with the image open:

Breakpoint 1, nsThebesImage::Draw (this=0x21bb7ae0, aContext=0x19e024d0, aUserSpaceToImageSpace=@0xbfffb350, aFill=@0xbfffb380, aPadding=@0xbfffb420, aSubimage=@0xbfffb450) at /data/build/shiretoko/gfx/src/thebes/nsThebesImage.cpp:463
463	    NS_ASSERTION(!aFill.IsEmpty(), "zero dest size --- fix caller");

Comment 3

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Attachment: GDB stack

Comment 4

[Joe Drew (not getting mail)]

Is this actually a regression? I don't think it is - I've seen this bug reported elsewhere. If it is, a regression window would help, but I'm pretty certain that this isn't new to 3.6.

Comment 5

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Joe, see comment 0. The patch on bug 489389 has been caused this regression.

Comment 6

[Joe Drew (not getting mail)]

Ah, okay. Well, still not a regression, then. :) Sorry for missing it.

Comment 7

[Mike Beltzner [:beltzner, not reading bugmail]]

This is a regression from 1.9.0 -> 1.9.1, not from 1.9.1 -> 1.9.2.

Comment 8

[Joe Drew (not getting mail)]

We should fix this. But we've released a couple of releases with it already, so we can't really justify blocking on it.

Comment 9

[Joe Drew (not getting mail)]

Sudden thought: we're probably mixing the top line of the blank area with the bottom line of the image. Then, when we get more data off the network, we send a dirty rect of only the added lines, leaving white-ish lines.

Another possible option: we're rounding the scaled dirty rect incorrectly.

Both of these options would account for the fact that causing a redraw, like by switching tabs, makes this problem go away.

This would also be pretty easy to test by inserting delays while sending the image with httpd.js.