Closed Bug 629196 Opened 10 years ago Closed 10 years ago

[OS X 10.6] Sporadic failure in layout/reftests/bugs/579323-1.html

Categories

(Core :: Graphics, defect)

x86_64
macOS
defect
Not set
normal

Tracking

()

RESOLVED FIXED
mozilla8

People

(Reporter: dholbert, Assigned: roc)

References

()

Details

(Keywords: intermittent-failure)

http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1296080853.1296082257.8143.gz
Rev3 MacOSX Snow Leopard 10.6.2 mozilla-central debug test reftest on 2011/01/26 14:27:33
s: talos-r3-snow-014

I think the reftest failure log looks exactly like bug 627560 -- it's a 4px difference, with a mismatch at each corner of the box.

But that bug is a 10.5-specific permaorange, whereas this appears to be a sporadic orange on 10.6.  So I'm filing this new bug on it.
This happens on 10.6.2 when layersGPUAccelerated is false.
So, these rendering problems look to be a rendering error.  Bug 670338 and bug 670111 have landed around the time this got very frequent.  I'm going to back them out to see if they change the volume of these oranges.  If they do, I guess we can blame this bug on one of those patches...
Blocks: 670338, 670111
(In reply to comment #23)
> So, these rendering problems look to be a rendering error.  Bug 670338 and
> bug 670111 have landed around the time this got very frequent.  I'm going to
> back them out to see if they change the volume of these oranges.  If they
> do, I guess we can blame this bug on one of those patches...

I backed both of these bugs out, and will keep an eye on the frequency of this bug to see if it goes back down.  If not, I'll reland those two bugs again next week.
(In reply to comment #23)
> So, these rendering problems look to be a rendering error.  Bug 670338 and
> bug 670111 have landed around the time this got very frequent.  I'm going to
> back them out to see if they change the volume of these oranges.  If they
> do, I guess we can blame this bug on one of those patches...

I'm not overly convinced there is any correlation. Though if there is any error in my logic I'm quite happy to be corrected! :-)

I've pushed to try the following reftest runs:

m-c tip (includes bug 670338 and bug 670111):
http://dev.philringnalda.com/tbpl/?tree=Try&rev=9e056fe3c757
- OS X 10.6 opt: 0 orange out of 20 runs
- OS X 10.6 debug: 1 orange out of 20 runs
- OS X 10.5 opt: 0 orange out of 10 runs

m-c tip with bug 670338 backed out (ie just includes bug 670111):
- OS X 10.6 opt: 4 orange out of 20 runs
- OS X 10.6 debug: 0 orange out of 20 runs
- OS X 10.5 opt: 0 orange out of 10 runs

m-c tip with bug 670111 backed out (ie just includes bug 670338):
- OS X 10.6 opt: 4 orange out of 20 runs
- OS X 10.6 debug: 1 orange out of 20 runs
- OS X 10.5 opt: 0 orange out of 10 runs

m-c tip with both bug 670111 and bug 670338 backed out: (note try being try doesn't list both changesets, but they are there: http://hg.mozilla.org/try/graph/87912)
- OS X 10.6 opt: 4 orange out of 20 runs
- OS X 10.6 debug: 2 orange out of 20 runs
- OS X 10.5 opt: 0 orange out of 10 runs
I've re-triggered a load more reftest runs to try and make the results more statistically significant: (whether the below is sufficient, I'll leave for someone else to decide)

m-c tip (includes bug 670338 and bug 670111):
- OS X 10.6 opt: 2 orange out of 50
- OS X 10.6 debug: 2 orange out of 50
= 4/100

m-c tip with bug 670338 backed out (ie just includes bug 670111):
- OS X 10.6 opt: 9 orange out of 50
- OS X 10.6 debug: 0 orange out of 50
= 9/100

m-c tip with bug 670111 backed out (ie just includes bug 670338):
- OS X 10.6 opt: 7 orange out of 50
- OS X 10.6 debug: 2 orange out of 50
= 9/100

m-c tip with both bug 670111 and bug 670338 backed out:
- OS X 10.6 opt: 8 orange out of 50
- OS X 10.6 debug: 4 orange out of 50
= 12/100
This is just 4 corner pixels being off by 1/255. It's not relevant to the test itself. I have no insight into what might be causing the problem, just some wild guesses. I think we should just censor the corners.
Whiteboard: [orange] → [orange][inbound]
Given comment 27 and comment 30, can bug 670338 and bug 670111 be relanded?
http://hg.mozilla.org/mozilla-central/rev/37cb0a1bc2ce
Assignee: nobody → roc
Status: NEW → RESOLVED
Closed: 10 years ago
Resolution: --- → FIXED
Whiteboard: [orange][inbound] → [orange]
Target Milestone: --- → mozilla8
Whiteboard: [orange]
You need to log in before you can comment on or make changes to this bug.