1196115 - Intermittent Crash Autophone S1S2Test beginning 2015-08-18

Status: RESOLVED FIXED

(Firefox for Android Graveyard :: General, defect)

Description

[Bob Clary [:bc] (inactive)]

Up until

https://treeherder.mozilla.org/#/jobs?repo=mozilla-inbound&revision=18f274867bcc&exclusion_profile=false&filter-searchStr=autophone

Autophone had fairly infrequent failures due to either failure to get measurements for an unknown reason or due to crashing.

Beginning with

https://treeherder.mozilla.org/#/jobs?repo=mozilla-inbound&revision=20c6f74296e2&exclusion_profile=false&filter-searchStr=autophone

Autophone began experiencing a much more frequent failure mode. Unfortunately, the crashes have dumps which are not parseable and show as unknown frame.

The pattern is clear when looking at a larger range:

https://treeherder.mozilla.org/#/jobs?repo=mozilla-inbound&filter-searchStr=autophone&exclusion_profile=false&fromchange=2b213c98f8a8

This appears to implicate bug 1195154.

Comment 1

[Bob Clary [:bc] (inactive)]

backing out bug 1195154 did not help. I'll try to do some local runs and see if I can pin point the real regressing changeset.

Comment 2

[Bob Clary [:bc] (inactive)]

I ran tests locally for the relevant time period:

https://treeherder.allizom.org/#/jobs?repo=mozilla-inbound&exclusion_profile=false&filter-searchStr=autophone&fromchange=18f274867bcc&tochange=08dd09256e7e

this flags bug 1098064 and 874167. bug 1098064 comment 5 says the check in should have been flagged bug 1194780

bsmedberg: Do either of these seem like they could be responsible for the intermittent crashes in Android?

Comment 3

[Benjamin Smedberg]

Bug 1194780 removed support for in-process plugins. Is Android still allowing/supporting Flash? That might be what's involved here if those autophone tests include Flash.

Comment 4

[Bob Clary [:bc] (inactive)]

I did a series of tests where I started with everything backed out, then applied each patch.

https://treeherder.allizom.org/#/jobs?repo=try&author=[e-mail]&startdate=2015-08-19&exclusion_profile=false&filter-searchStr=autophone&fromchange=9bb55dbad5b9&tochange=26a61c43cfb4

part A - remove nsNPAPIPlugin::RunPluginOOP and unused code paths, r=jimm does appear to be the culprit.

We do have flash enabled and use it on nytimes pages which are the failing pages.

snorp?

Comment 5

[James Willcox (:snorp) ([e-mail]) (he/him)]

(In reply to Benjamin Smedberg  [:bsmedberg] from comment #3)
> Bug 1194780 removed support for in-process plugins. Is Android still
> allowing/supporting Flash? That might be what's involved here if those
> autophone tests include Flash.

We do allow and support Flash for the time being. Can you please back this out? Or add an exception for Android?

Comment 6

[Bob Clary [:bc] (inactive)]

bsmedberg, do you think we can resolve this soon ? I can make Flash click to play to work around this if it won't be resolved relatively quickly.

Comment 7

[Bob Clary [:bc] (inactive)]

I'm making flash click to play until this is resolved.

https://github.com/mozilla/autophone/commit/c9eb3e206583d9598dbd92a98135c99293ce63ac

Comment 8

[Benjamin Smedberg]

I sent around an email thread about this. Since the goal is to remove a bunch of support for in-process NPAPI, I'd like to remove Flash support from FxAndroid, but need to understand how that affects our market situation.

Comment 9

[Kevin Brosnan [Ex-Mozilla]]

FYI Bug 1194780 has removed the ability of Firefox for Android to play Flash in version 43.

Comment 10

[Barbara Bermes [:barbara] - NI please!]

Copied from Karen's email:

The short answer is that Flash much remain for Firefox for Android.


It is a very real point of differentiation between Firefox and other browsers. We know that people use Firefox because we support Flash (where Chrome doesn’t) and we have feedback saying that Flash support is a reason for being ‘happy’ with our application, and we have an increased spike in low ratings when it breaks (see below).

 

Specifically:

·       When Bu 1112086 happened (when we accidentally blocked flash on Android Dec 16 – 29 2014)

o   Flash complaints in English increased from 0.5% to 1.4% (of total Play feedback)

o   Japan’s Play ratings dropped from 4.1 to 3.2 because their mobile gaming sites are Flash heavy and they rely on the Flash user experience

·       The percentage of positive Play comments that include Flash is approximate to a percentage calling out another feature. This is pretty neat considering Flash isn’t a marketed feature, yet users still see it as such and are proactively telling us that Flash supported has provided a positive experience.



This is never the whole story – many users still talk about ‘video’ which we can’t tell is Flash specific or not, but we have enough evidence that blocking Flash

·       Gives us a really negative experience for countries that still rely on mobile Flash

·       Mobile sites still support Flash

·       Users are happy that Flash is supported on Firefox

·       Chrome doesn’t support Flash, giving us an edge to use us vs Chrome


So no, we’re not ready to stop supporting Flash yet on Firefox for Android.

Comment 11

[Benjamin Smedberg]

I filed bug 1200361 to re-enable OOPP on Android, and bclary can use this one to track the test failure in particular. I believe from email that the Android team is going to implement more comprehensive Flash functional tests. bc, do you know if that has a bug number to track that work or if not can you file one?

Comment 12

[Bob Clary [:bc] (inactive)]

Filed bug 1201061. I will re-enable Flash on the production devices when the devices catch up to the check in for bug 1200361

Comment 13

[Bob Clary [:bc] (inactive)]

Attachment: Bug-1196115-v1.patch

re-enable flash, r=self.

Comment 14

[Bob Clary [:bc] (inactive)]

https://github.com/mozilla/autophone/commit/4e9a4ba10620c7e2a7512fea6d73f697d3262712

Comment 15

[Bob Clary [:bc] (inactive)]

deployed 2015-09-09.