Open Bug 1072866 Opened 10 years ago Updated 2 years ago

Intermittent cross-iframe-1.html | image comparison (==), max difference: 255, number of differing pixels: 222

Categories

(Core :: Layout, defect, P3)

Product:
Core
Component:
Layout
Platform:
x86
macOS
Type:
defect

Tracking

()

People

(Reporter: cbook, Unassigned)

References

(
URL
)

Details

(Keywords: intermittent-failure, regression)

Attachments

(1 file)

Relevant part of the log
22.67 KB, text/plain
Details 
Rev4 MacOSX Snow Leopard 10.6 mozilla-inbound opt test reftest on 2014-09-25 01:12:55 PDT for push ab566bc21805

slave: t-snow-r4-0079

https://tbpl.mozilla.org/php/getParsedLog.php?id=48847330&tree=Mozilla-Inbound

REFTEST TEST-UNEXPECTED-FAIL | file:///builds/slave/talos-slave/test/build/tests/reftest/tests/layout/reftests/font-face/cross-iframe-1.html | image comparison (==), max difference: 255, number of differing pixels: 222
This is pretty frequent. Dan, do you have cycles to look into this?
Flags: needinfo?(dholbert)
(In reply to Ryan VanderMeulen [:RyanVM UTC-4] from comment #113)
> This is pretty frequent. Dan, do you have cycles to look into this?

From the reftest snapshot, it looks like we're failing to load a font ("markC.ttf" specifically).  This also seems to be a regression (first report 9/25, and we've been having reports every few days to several times a day ever since).  So, might as well track down what broke it.

Liberal regression pushlog (5 days before the report in comment 0):
http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?startdate=2014-09-20&enddate=2014-09-25+07%3A10

There are several font-related changes in that range. The most suspicious-sounding is bug 998869 ("lazily load userfonts"), from jdaggett -- but there are also bug 1072072, bug 1070316, bug 1070260, & bug 1069761 from heycam.

Transfering needinfo to jdaggett.
Flags: needinfo?(dholbert) → needinfo?(jdaggett)
Keywords: regression
Observation: The reference case ALSO uses an external font, but in the (admittedly few) test-failure snapshots I've looked at, it's always the *testcase* that missed loading the font.  So, there's something special about the testcase that makes it fail. Also, it always seems to be just the first line of the testcase -- the line that's *not* in an iframe.

So, it seems that the iframes' external-font-loads are somehow responsible for the missing font in the first line. (outside of the iframe).  I wonder if there's some sort of race condition here? (Or if the iframes' loads make us purge some cache, which clears the outside-the-iframe font?)

Just spitballing; hopefully jdaggett can take it from here. :)
What's odd here is that this is *always* a failure under 10.6, not other OSX versions.
Flags: needinfo?(jdaggett)
Assignee: nobody → jdaggett
Blocks: 998869
This seems to have gotten worse on or around 3/24 (comment 268) -- this has been failing 1-4 times per day since then.

(Technically 3/29 had no reports, but it was a Sunday & probably just had a lighter load of pushes than other days.)
Assignee: jd.bugzilla → nobody
Bulk assigning P3 to all open intermittent bugs without a priority set in Firefox components per bug 1298978.
Priority: -- → P3
Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: