+++ This bug was initially created as a clone of Bug #989462 +++ Android 2.3 reftests are now running reliably on Ash and meet the requirements on https://wiki.mozilla.org/Sheriffing/Job_Visibility_Policy: https://tbpl.mozilla.org/?tree=Ash&jobname=Android.*2.3.* All of {C1 C2 C3 J1 J2 J3 J4 J5 J6 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12} can now be run on trunk. Mochitest-gl is the only remaining job on Ash that is not ready for trunk. Android 2.3 crashtests, js-reftests, and plain-reftests must be run on ix slaves to avoid excessive run times and timeouts. From my perspective, Android 2.3 mochitests, robocop, and xpcshell tests may continue running on aws slaves, or may also be run on ix slaves.

mochitest-gl is running reliably on ash now and can also be scheduled on trunk.

patch to enable additional tests gbrown: As a side note, in this patch I enabled these tests where gecko is at least 32. Previously there were limited tests on gecko31 i.e. limited tests on mozilla-aurora. However, to clean up our code so there weren't stanzas for disabled and enabled 2.3 tests, I just enabled all tests on 2.3 going forward where gecko is at least 32. This means that these tests will be disabled on m-a until the next merge. Let me know if this is okay, but I felt that it was important to clean up mobile_config.py. I'll attach a builder diff so you can see the difference.

builder diff

(In reply to Kim Moir [:kmoir] from comment #2) > gbrown: As a side note, in this patch I enabled these tests where gecko is > at least 32. Previously there were limited tests on gecko31 i.e. limited > tests on mozilla-aurora. However, to clean up our code so there weren't > stanzas for disabled and enabled 2.3 tests, I just enabled all tests on 2.3 > going forward where gecko is at least 32. This means that these tests will > be disabled on m-a until the next merge. Let me know if this is okay, but I > felt that it was important to clean up mobile_config.py. I'll attach a > builder diff so you can see the difference. That seems reasonable. Of course there is interest in retiring the tegras as soon as possible, and I suppose we cannot retire them fully until these changes ride the trains and 32 is released. But I don't see a way around that: We only know that m-gl passes on 32, for example. It would probably take a lot of uplifts and re-testing to accelerate the tegra retirement.

Comment on attachment 8421182 [details] [diff] [review] bug1004791.patch Review of attachment 8421182 [details] [diff] [review]: ----------------------------------------------------------------- This is great! On another note, if armv6, armv7 and androidx86 are going to be tested on talos-linux64-ix we *might* need another round of iX purchases. We might also have some suites that we can move to AWS without too much of a wall time hit compared to iX machines.

Live in production.

oops, they were not running on ix slaves, but rather ec2 instances which will lead to long test times. I've tested this on my master, will attach builder diff

builder diff

Comment on attachment 8421819 [details] [diff] [review] bug1004791fix.patch and reconfiged

verified on tbpl