1436642 - With runByManifest, dom/animation/test/chrome/test_animation_performance_warning.html fails

Status: RESOLVED FIXED

(Testing :: Mochitest, enhancement)

Description

[Geoff Brown [:gbrown]]

In bug 1393326 we are trying to run Android mochitests, including mochitest-chrome, by-manifest, meaning that each manifest of tests is run in a separate browser instance.

When we run mochitest-chrome by manifest, there is a single new test failure, in dom/animation/test/chrome/test_animation_performance_warning.html, on both Android 4.3 API16+ opt and Android 4.3 API16+ debug; it runs and passes on Android x86.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=3fded1dca347246a9ca659b38c2e0e32108d9155

Comment 1

[Hiroyuki Ikezoe (:hiro)]

We are setting a pref value in the test [1].  If the pref was set in chrome.ini, what happens?

[1] https://hg.mozilla.org/mozilla-central/file/8cc2427a322c/dom/animation/test/chrome/test_animation_performance_warning.html#l41

Comment 2

[Geoff Brown [:gbrown]]

Thanks :hiro, but I don't think it is possible to set a pref in chrome.ini (prefs can only be set in reftest manifests, I think).

Comment 3

[Andrew Halberstadt [:ahal]]

Actually, it is possible. But it can only be set in the DEFAULT section, and if the manifest gets used when --run-by-manifest is off, then the harness will error out. (So if it's a shared manifest between android and desktop, the android run will fail as run-by-manifest isn't turned on there yet).

E.g:
https://dxr.mozilla.org/mozilla-central/source/browser/base/content/test/performance/browser.ini#7

Comment 4

[Geoff Brown [:gbrown]]

Oh, right, that's why I'm doing this run-by-manifest-on-android work anyway! How silly!

Comment 5

[Geoff Brown [:gbrown]]

(In reply to Hiroyuki Ikezoe (:hiro) from comment #1)
> We are setting a pref value in the test [1].  If the pref was set in
> chrome.ini, what happens?
> 
> [1]
> https://hg.mozilla.org/mozilla-central/file/8cc2427a322c/dom/animation/test/
> chrome/test_animation_performance_warning.html#l41

The test still fails if the pref is set in chrome.ini:

https://treeherder.mozilla.org/#/jobs?repo=try&revision=56169000bb949d45519893c73cf6653e25e6489b

Comment 6

[Hiroyuki Ikezoe (:hiro)]

Thanks Geoff, I will take a look.

Comment 7

[Hiroyuki Ikezoe (:hiro)]

I could confirm the failure on local emulator when I run "./mach mochitest dom/animation/test/chrome", so I'd say the test has been broken in the first place.  I don't quite understand why the width value which is used in the failure case has been changed and why 5200px was chosen in bug 1364221 .

CCing Jared and keep NI to me.

Comment 8

[Hiroyuki Ikezoe (:hiro)]

I think the value, 5200px, was too small for Android arm emulator.  (The scale value on the emulator is 0.5x?)

Using 10000px for the area test and using 20000px for the other case work fine;

https://treeherder.mozilla.org/#/jobs?repo=try&revision=162b3df18b892de9bdc29d1f42ea50a257b70600

Comment 9

[Hiroyuki Ikezoe (:hiro)]

Attachment: Bug 1436642 - Make target element size larger to be unable to send transform animations to the compositor.

The original value were too small for Android.

Review commit: https://reviewboard.mozilla.org/r/219194/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/219194/

Comment 10

[Hiroyuki Ikezoe (:hiro)]

Here is a try for all platforms (other than Android x86);
https://treeherder.mozilla.org/#/jobs?repo=try&revision=148700594786dcdfbdbebe9e94b8549c5737218e

And for Android x86;
https://treeherder.mozilla.org/#/jobs?repo=try&revision=8728018c4e34310c2a48cc9af9bca8d4e34237d8

Comment 11

[Jared Wein [:jaws] (please needinfo? me)]

Comment on attachment 8949871 [details]
Bug 1436642 - Make target element size larger to be unable to send transform animations to the compositor.

https://reviewboard.mozilla.org/r/219194/#review225026

If I remember right I had to use the lower number because I was getting crashes on Android with larger numbers. Though much may have changed since then so let's try it again :)

Comment 12

[Hiroyuki Ikezoe (:hiro)]

(In reply to Jared Wein [:jaws] (please needinfo? me) from comment #11)
> Comment on attachment 8949871 [details]
> Bug 1436642 - Make target element size larger to be unable to send transform
> animations to the compositor.
> 
> https://reviewboard.mozilla.org/r/219194/#review225026
> 
> If I remember right I had to use the lower number because I was getting
> crashes on Android with larger numbers. Though much may have changed since
> then so let's try it again :)

Thanks!  Let's see what happens for a while. :)

Comment 13

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/f5282133fa20
Make target element size larger to be unable to send transform animations to the compositor. r=jaws

Comment 14

[Narcis Beleuzu [:NarcisB]]

https://hg.mozilla.org/mozilla-central/rev/f5282133fa20