1121835 - OS X 10.10 is very angry about gfx 893572 crashtests "Error: this application, or a library it uses, has passed an invalid numeric value (NaN, or not-a-number) to CoreGraphics API"

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Phil Ringnalda (:philor)]

See https://treeherder.mozilla.org/logviewer.html#?job_id=90780&repo=cedar or once that expires any debug 10.10 crashtest run on Cedar (which is the only place they are running so far).

16:03:56 INFO - REFTEST TEST-LOAD | file:///builds/slave/talos-slave/test/build/tests/reftest/tests/gfx/tests/crashtests/893572-2.html | 872 / 2776 (31%)
16:03:56 INFO - ++DOMWINDOW == 23 (0x11f062800) [pid = 1631] [serial = 2078] [outer = 0x11c28fc00]
16:03:56 INFO - Jan 14 16:03:56 t-yosemite-r5-0001.test.releng.scl3.mozilla.com firefox[1631] <Error>: Error: this application, or a library it uses, has passed an invalid numeric value (NaN, or not-a-number) to CoreGraphics API. This is a serious error and contributes to an overall degradation of system stability and reliability. This notice is a courtesy: please fix this problem. It will become a fatal error in an upcoming update.
(repeated many times)
...
16:03:57 INFO - REFTEST TEST-LOAD | file:///builds/slave/talos-slave/test/build/tests/reftest/tests/gfx/tests/crashtests/893572-3.html | 873 / 2776 (31%)
...
16:03:57 INFO - REFTEST TEST-LOAD | file:///builds/slave/talos-slave/test/build/tests/reftest/tests/gfx/tests/crashtests/893572-4.html | 874 / 2776 (31%)

Comment 1

[Kartikaya Gupta (email:[e-mail])]

We'd need somebody with OS X 10.10 to repro this and get a stack to where the bad value is getting passed in. [1] has a tip that may be useful. I'm not sure which gfx devs are running 10.10; Milan, do you know?

[1] http://stackoverflow.com/questions/19187344/how-to-detect-where-nan-is-passing-to-coregraphics-api-on-mac-os-x-10-9

Comment 2

[Milan Sreckovic [:milan] (needinfo for best results)]

I'm guessing this is us not protecting when using path, the way we protect when using other canvas operations.

Comment 3

[Milan Sreckovic [:milan] (needinfo for best results)]

Attachment: Deal with non-finite values in Path calls. Add IsFinite() to BasePoint. r=bschouten

Comment 4

[Milan Sreckovic [:milan] (needinfo for best results)]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=a6704634abaa

Comment 5

[Phil Ringnalda (:philor)]

You only get 10.8 on try with your 10.6 if you choose -u all[10.6,10.8].

You only get 10.10 on try with your 10.6 and 10.8 if you choose -u all[10.6,10.8,10.10] and push sometime after the next buildbot reconfig.

Pushed to Cedar, where 10.10 already is, in https://treeherder.mozilla.org/#/jobs?repo=cedar&revision=852f84996843

Comment 6

[Milan Sreckovic [:milan] (needinfo for best results)]

Thanks for the Cedar run Phil; I was mostly making sure I didn't break anything in the "standard" builds with my try run.
The run from comment 4 is green enough for this to land. I imagine the Cedar run isn't, but I imagine it isn't green without this change either.  However, the reported problem did seem to go away with this change, if I'm looking in the right logs (http://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/cedar-macosx64-debug/1421705578/cedar_yosemite-debug_test-crashtest-bm108-tests1-macosx-build1.txt.gz), so we should be safe to land.

Comment 7

[Phil Ringnalda (:philor)]

Right, I've been looking at them long enough now I no longer remember how horrifying they are at first glance :) Make that

"Pushed to Cedar, where 10.10 already is, in https://treeherder.mozilla.org/#/jobs?repo=cedar&revision=852f84996843 with encouraging results, please do land this"

Comment 8

[Phil Ringnalda (:philor)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/7bb2058c0cda

Comment 9

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/7bb2058c0cda