Closed Bug 769975 Opened 12 years ago Closed 11 years ago

Frequent REFTEST TEST-UNEXPECTED-FAIL | gfx/tests/reftest/611498-1.html | image comparison (==), max difference: 62 (or 63), number of differing pixels: 312 (or 812)

Categories

(Core :: Graphics, defect)

Product:
Core
Component:
Graphics
Platform:
All
Windows 7
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

()

Status:
RESOLVED WORKSFORME

People

(Reporter: RyanVM, Assigned: mattwoodrow)

References

Details

(Keywords: intermittent-failure) 

I think this is from DLBI, but it appears that some of the earliest failures were mixed up with another unrelated set of failures. Happens on Windows 7 debug & opt with and without acceleration.

https://tbpl.mozilla.org/php/getParsedLog.php?id=13140793&tree=Mozilla-Inbound&full=1

Rev3 WINNT 6.1 mozilla-inbound debug test reftest on 2012-06-30 15:47:15 PDT for push 3b62d36c68c5

builder: mozilla-inbound_win7-debug_test-reftest
slave: talos-r3-w7-075
starttime: 1341096435.88
results: warnings (1)
buildid: 20120630141850
builduid: 8b683c8d04c24ea2813ac7a19bf32b2d
revision: 3b62d36c68c5

REFTEST TEST-UNEXPECTED-FAIL | file:///C:/talos-slave/test/build/reftest/tests/gfx/tests/reftest/611498-1.html | image comparison (==)
Assignee: nobody → matt.woodrow
This is a completely different failure to what the bug was originally filed for.

Different platform, actual difference between the images is entirely different too.
(In reply to Matt Woodrow (:mattwoodrow) from comment #71)
> This is a completely different failure to what the bug was originally filed
> for.
> 
> Different platform, actual difference between the images is entirely
> different too.

Would imagine the new failure is fallout from bug 685516.

khuey please can you disable :-)
Blocks: 685516
(In reply to Ed Morley [:edmorley] from comment #82)
> (In reply to Matt Woodrow (:mattwoodrow) from comment #71)
> > This is a completely different failure to what the bug was originally filed
> > for.
> > 
> > Different platform, actual difference between the images is entirely
> > different too.
> 
> Would imagine the new failure is fallout from bug 685516.
> 
> khuey please can you disable :-)

I've landed:
https://hg.mozilla.org/mozilla-central/rev/f5f29adc6d30
Not happening anymore.
Status: NEW → RESOLVED
Closed: 12 years ago
Resolution: --- → WORKSFORME
Beg to differ - same row of chins as comment 58, which I presume was still the "is this bug" thing.
Status: RESOLVED → REOPENED
Resolution: WORKSFORME → ---
Summary: Frequent REFTEST TEST-UNEXPECTED-FAIL | gfx/tests/reftest/611498-1.html | image comparison (==) → Frequent REFTEST TEST-UNEXPECTED-FAIL | gfx/tests/reftest/611498-1.html | image comparison (==), max difference: 62, number of differing pixels: 208
Whiteboard: [orange]
Summary: Frequent REFTEST TEST-UNEXPECTED-FAIL | gfx/tests/reftest/611498-1.html | image comparison (==), max difference: 62, number of differing pixels: 208 → Frequent REFTEST TEST-UNEXPECTED-FAIL | gfx/tests/reftest/611498-1.html | image comparison (==), max difference: 62, number of differing pixels: 312
Summary: Frequent REFTEST TEST-UNEXPECTED-FAIL | gfx/tests/reftest/611498-1.html | image comparison (==), max difference: 62, number of differing pixels: 312 → Frequent REFTEST TEST-UNEXPECTED-FAIL | gfx/tests/reftest/611498-1.html | image comparison (==), max difference: 62 (or 63), number of differing pixels: 312 (or 812)
Matt, can you take a look at this please or else find a new owner? :-)
Flags: needinfo?(matt.woodrow)
I don't have any ideas here. I've pushed a debugging patch to try:

https://tbpl.mozilla.org/?tree=Try&rev=a0616add2f41

Hopefully I can get it to fail there and see what it shows.
Flags: needinfo?(matt.woodrow)
This is us getting bitten by bug 702504, yet again.

Relevant section of the log: http://pastebin.mozilla.org/2207512

There are 3 paints shown here.

The first is when we draw the page for the first time, using the retained d3d10 layer manager.

We draw the background of frame 0x070FF178 at (8, 50, 104, 104) - lines 11, 17, 18.

The second paint is us taking a snapshot of the page (drawn using a temporary BasicLayerManager).

This time around the background of frame 0x070FF178 is at (8, 58, 104, 104) - lines 55, 63, 64. Note that this is 8 pixels further down that it actually is on the screen LayerManager.

The third paint is when we move the background and do a new paint for it's new position and generate invalidations for this. This is again with the d3d10 layer manager.

Here we can see us removing the background, and marking the pixels it covered as invalid (8,50,104,104 - line 98/98).

We then send this invalid region to the reftest harness (via MozAfterPaint) which requests draws a new snapshot (using the BasicLayerManager) of those pixels. Unfortunately the background was drawn onto pixels outside that region in the BasicLayerManager version of events, and these don't get redrawn.

Fixing bug 702504 would be the obvious solution here (and what I intend to work on), but figuring out why the two LayerManagers are positioning things differently would work too. roc, do you have any ideas for that part?
Depends on: 702504
(OrangeWFM for bugs not modified in > 2 months)
Status: REOPENED → RESOLVED
Closed: 12 years ago → 11 years ago
Resolution: --- → WORKSFORME
You need to log in before you can comment on or make changes to this bug.