Mobile/Testing/02 19 14
From MozillaWiki
Contents
Previous Action Items
- Callek to followup on ant install - bug 971841
- Ongoing
-
gbrown will follow-up on startup crashes with :glandium- bug 959254 has crash logs with MOZ_DEBUG_LINKER
-
gbrown will invite jmaher for next week- sorry - he has a conflict
- snorp will talk to jmaher about talos improvements to find gradual, long-term regressions
- gbrown will find out about GeckoView on Talos
Status reports
Dev team
Rel Eng
- Deployed
ant
, andant-apache-regexp
to try machines- Unsuccessful, need ant 1.8+, todo: create new RPMs and deploy afresh.
IT
A Team
- tegra failure rate: [2.08%]
- last week 1.5
- panda failure rate: [15.03%]
- last week 14.19
- excluding retries 1.06%
Slow Reftests
- tried tweaking settings in test harness:
- disabled image comparison (always true)
- allowed arbitrarily many recycled canvases
- made canvases 1px x 1px
- none of this had any effect
- ran android profile (am profile) for mochitests and reftests
- reftest profile shows substantial time in:
- ViewRoot.handleMessage, ViewRoot.measureTraversals (27.3%)
- View.measure, View.layout
- ViewGroup.measureChildWithMargins
- FrameLayout.onMeasure
- RelativeLayout.onMeasure, RelativeLayout.measureChildHorizontal, RelativeLayout.measureChild
- FrameLayout.onLayout
- and a bunch of other layout stuff
- The first layout related function for mochitests is at 17th place
- ViewRoot.handleMessage, ViewRoot.measureTraversals (2.6%)
- raw traces: mochitest-trace.txt reftest-trace.txt
- reftest profile shows substantial time in:
- have a debug with profiling emulator build I will be trying out this afternoon
- hopefully correlate to android profiling
- slow floating point?
Android 2.3 emulator
- See "Android 2.3 Opt" at https://tbpl.mozilla.org/?tree=Ash&jobname=Android.*&rev=0ce5e6c6778f
- A few intermittent mochitest/crashtest failures still
- Many robocop tests skipped
Android x86 emulator
Autophone
- Landed bug 968905, phones keeping up with builds with new partitioning.
- Fixed bug 972346 mozbase - devicemanagerSUT - During _waitForRebootPing, do not accept multiple times which caused the outage on 2014-02-12.
- On 2014-02-14, directed all output from the local phonedash web server to /dev/null and changed the logging level to WARNING which will reduce extraneous output on the host server.
- On 2014-02-15, started using a phonedash server hosted on a separate machine for serving the remote files to remove any possible interference.
- Working on bug 968920 Autophone - use standard error percentage as a discriminator.
Eideticker
- New dashboard structure: faster to load (metadata for each testrun now stored seperately)
- Moving away from idea of using datazilla as a backend (lots of effort for little gain), sticking with static json
- Will still hopefully be able to piggyback on internals datazilla's notification system, somehow. Will talk to relevant people later this week.
- Considering hosting dashboard entirely on Amazon S3
- Will fix bug 971477 (startup-abouthome-dirty does not create a full profile), bug 967109 (Replace clock test on android with canvas box) hopefully this week.
Round Table
Action Items
- Callek to followup on ant install - bug 971841
- wlach will send email about Datazilla, notifications, etc (for improvements to find gradual, long-term regressions)
- Talos tests for GeckoView?