Beginning on October 25th, 2016, Persona will no longer be an option for authentication on BMO. For more details see Persona Deprecated.
Last Comment Bug 722014 - large number of reftest failures, possibly related to incorrect font-size inflation?
: large number of reftest failures, possibly related to incorrect font-size inf...
Product: Core
Classification: Components
Component: Layout: Text (show other bugs)
: Trunk
: x86 Mac OS X
: -- normal (vote)
: mozilla12
Assigned To: Jonathan Kew (:jfkthame)
Depends on:
  Show dependency treegraph
Reported: 2012-01-28 01:45 PST by Jonathan Kew (:jfkthame)
Modified: 2012-01-30 16:48 PST (History)
5 users (show)
See Also:
Crash Signature:
QA Whiteboard:
Iteration: ---
Points: ---
Has Regression Range: ---
Has STR: ---


Description Jonathan Kew (:jfkthame) 2012-01-28 01:45:41 PST
Seen on mozilla-inbound:

Total of 123 reftest failures, beginning with

REFTEST TEST-UNEXPECTED-FAIL | file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/css-calc/max-width-block-intrinsic-1.html | image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL | file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/css-calc/min-width-block-intrinsic-1.html | image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL | file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/css-calc/text-indent-intrinsic-1.html | image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL | file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/css-calc/vertical-align-1.html | image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL | file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/css-calc/width-block-intrinsic-1.html | image comparison (==)
REFTEST TEST-UNEXPECTED-PASS | file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/css-gradients/height-dependence-2.html | image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL | file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/text-overflow/marker-basic.html | image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL | file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/text-overflow/quirks-decorations.html | image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL | file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/text-overflow/quirks-line-height.html | image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL | file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/text-overflow/standards-decorations.html | image comparison (==)
REFTEST TEST-UNEXPECTED-FAIL | file:///Users/cltbld/talos-slave/test/build/reftest/tests/layout/reftests/text-overflow/standards-line-height.html | image comparison (==)

Looking at a sampling of the failures, many of them appear to show incorrect font-size inflation; so I wonder whether this could be a new intermittent failure, perhaps triggered by bug 698652, which was the push immediately preceding this one (but didn't exhibit the failures itself).
Comment 1 Chris Jones [:cjones] inactive; ni?/f?/r? if you need me 2012-01-28 02:00:54 PST
The first linked failure layout/reftests/css-calc/max-width-block-intrinsic-1.html doesn't touch <canvas> at all, which is where the substantive changes in bug 698652 were made.  That would suggest the reftest-manifest pref() annotation is buggy.  Sigh.

If this is more than "once in a blue moon" intermittent, then I suggest we back out the test and fix the reftest harness.
Comment 2 Jonathan Kew (:jfkthame) 2012-01-28 02:34:51 PST
Doesn't the reftest harness use canvas all the time?

Yes, it's entirely possible that the real problem lies with the pref() annotation - that's fairly new (bug 696585) and not yet heavily used, afaik.

I re-triggered the (green) OS X Opt reftest run on changeset c8ea6b7164bc (the bug 698652 push), and got the failures there as well. OTOH, a couple of re-runs on the following changeset, where this first occurred, came up green.
Comment 3 Chris Jones [:cjones] inactive; ni?/f?/r? if you need me 2012-01-28 02:51:10 PST
(In reply to Jonathan Kew (:jfkthame) from comment #2)
> Doesn't the reftest harness use canvas all the time?

Yes, but it uses canvas as a pixel-buffer target for drawWindow().  The end result is approximately saving a "screenshot" of the top-level XUL <window> to canvas, no repainting (except in extreme edge cases that we shouldn't hit in reftests).

What's your guess at the percent of intermittent failing runs right now?
Comment 4 Jonathan Kew (:jfkthame) 2012-01-28 03:40:26 PST
Looks like about 30% at the moment - 3 out of 10 runs on the relevant changesets have come up orange.

The failures are only on OS X opt, however; it hasn't happened on any other platform, for some reason.

Just checking - is the font-size inflation pref supposed to be fully "live apply"? If not, then it shouldn't be expected to work with the pref(...) annotation.
Comment 5 Jonathan Kew (:jfkthame) 2012-01-28 03:42:23 PST
(In reply to Jonathan Kew (:jfkthame) from comment #4)

> The failures are only on OS X opt, however; it hasn't happened on any other
> platform, for some reason.

I take that back - we just got an instance on WinXP opt:
Comment 6 Chris Jones [:cjones] inactive; ni?/f?/r? if you need me 2012-01-28 04:28:31 PST
OK.  I'm barely awake, but

is the patch to back out.  If that eliminates the orange, then the bug is in the new reftest harness support for pref().
Comment 7 Jonathan Kew (:jfkthame) 2012-01-28 08:13:21 PST
The failures happened several more times in the last few hours, so I backed out this test in

Let's confirm whether this resolves the orange, then file a bug on the reftest harness if necessary.
Comment 8 Matt Brubeck (:mbrubeck) 2012-01-28 09:54:10 PST (from right before the backout)
Comment 9 Joe Drew (not getting mail) 2012-01-28 18:50:52 PST

I'll leave this open because of comment 7
Comment 10 Chris Jones [:cjones] inactive; ni?/f?/r? if you need me 2012-01-30 16:48:10 PST
Filed bug 722569.  This bug was fixed by backing out the test.

Note You need to log in before you can comment on or make changes to this bug.