Closed Bug 1128517 Opened 9 years ago Closed 9 years ago

layout/reftests/css-gradients/aja-linear-1b.html has unexpected pass on OS X 10.10 debug

Categories

(Core :: Graphics, defect)

Product:
Core
Component:
Graphics
Platform:
x86
macOS
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

()

Status:
RESOLVED WORKSFORME
Tracking Flags:
Tracking Status
firefox38 --- wontfix
firefox38.0.5 --- wontfix
firefox39 --- fixed
firefox40 --- unaffected
firefox-esr38 --- fixed

People

(Reporter: kmoir, Unassigned)

Details

(Whiteboard: [gfx-noted])

Attachments

(1 file)

test.log
18.43 KB, text/plain
Details
Summary: layout/reftests/css-gradients/aja-linear-1b.html has unexpected pass on Macosx 10.10 → layout/reftests/css-gradients/aja-linear-1b.html has unexpected pass on OS X 10.10 debug
Does this pass in non debug tests?
No it has an expected fail in non debug tests
I must be missing something because the reftest.list is showing that there's only special rules for D2D for that test:

Your push:
https://hg.mozilla.org/try/file/4c71b1e7fc52/layout/reftests/css-gradients/reftest.list

Central:
http://mxr.mozilla.org/mozilla-central/source/layout/reftests/css-gradients/reftest.list?raw=1&ctype=

It should be an expected pass on all of OS X unless I'm looking at the wrong thing.
Just looking at the double-negative backward: it fails-if(!d2d) and thus passes-only-if(d2d), so for not-d2d, it's expected to fail and does fail, REFTEST TEST-KNOWN-FAIL, and thus the run passes, except that for debug 10.10, it's expected to fail but the test and reference are identical and thus it passes and the run fails. Hmm, but are they identical gradients, which would be rather weird debug-only, or are they identical white canvases that both fail to render anything at all?
Ahh right, thanks for that!

We should have another look. It's possible the CoreGraphics changed in 10.10. We shouldn't have rendering differences between debug and not however. We should get someone to look at why, but that wont be me this quarter :(.
:BenWa: Do you have any suggestions on who else would be able to look at this issue this quarter?  We have a quarterly goal to get 10.10 tests into production for Q1.
Flags: needinfo?(bgirard)
https://treeherder.mozilla.org/logviewer.html#?job_id=4657715&repo=try says they aren't identically blank, we really do somehow draw this gradient correctly only on debug 10.10.
Flags: needinfo?(bgirard) → needinfo?(milan)
During 38 is not the best timing, but there may be a chance for 39, which is still Q2.  Unless Markus has a chance to look at it before?
Flags: needinfo?(milan) → needinfo?(mstange)
No worries, I annotated the incomprehensible pass in https://hg.mozilla.org/integration/mozilla-inbound/rev/5351f97609aa, and one of you can look at it after these are the only >10.6 slaves running, and you try to land something and bounce due to completely incomprehensible test results.
Keywords: leave-open
No longer blocks: 1121199
Whiteboard: [gfx-noted]
The fact that this is debug-only is surprising, but I don't think we'll gain useful information by trying to debug this, so I think keeping the reftest annotation is the right thing to do.

The switch to skia canvas required the fuzz for skia, so I added a "&&azureQuartz" to the condition in bug 932958.
Flags: needinfo?(mstange)
Oh, I know how we could stop starring bug 1145475 instances on aurora as this!
Keywords: intermittent-failure
Status: NEW → RESOLVED
Closed: 9 years ago
status-firefox38: --- → wontfix
status-firefox38.0.5: --- → wontfix
status-firefox40: --- → unaffected
status-firefox-esr38: --- → affected
Keywords: leave-open
Resolution: --- → WORKSFORME
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: