Closed Bug 1170675 Opened 11 years ago Closed 7 years ago

Freeze, then crash when using Google Keyboard on Google search page

Categories

(Firefox for Android Graveyard :: Keyboards and IME, defect)

Product:
Firefox for Android Graveyard
Component:
Keyboards and IME
Version:
38 Branch
Platform:
ARM
Android
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

(Not tracked)

Status:
RESOLVED WONTFIX

People

(Reporter: nicolae_albu2004, Assigned: jnicol)

Details

Crash Data

User Agent: Mozilla/5.0 (Android; Mobile; rv:38.0) Gecko/38.0 Firefox/38.0 Build ID: 20150528015112 Steps to reproduce: Use Google Keyboard. Open Google search page. Tap on the search text box. The keyboard comes up. Actual results: Firefox freezes. After a couple of seconds, the crash reporter comes up. Expected results: The keyboard should have worked properly. It looks like it works with keyboards other than Google. But Google Keyboard works fine in any other app. I can consistently reproduce on Samsung S4 mini 4.4.2. On a tablet (where the keyboard layout is different), this doesn't reproduce.
OS: Unspecified → Android
Hardware: Unspecified → ARM
Crash Signature: https://crash-stats.mozilla.com/report/index/1bdcb3f7-c570-4ea0-a9de-452372150602
Crash Signature: https://crash-stats.mozilla.com/report/index/1bdcb3f7-c570-4ea0-a9de-452372150602 → [@ OOM | large | mozalloc_abort(char const*) | mozalloc_handle_oom(unsigned int) | moz_xmalloc | nsTArray_base<nsTArrayInfallibleAllocator, nsTArray_CopyWithConstructors<mozilla::layers::TileClient> >::EnsureCapacity(unsigned int, unsigned int) | nsTArray…
No one can reproduce to change the bug status? I can reproduce it every time on S4 mini and on another 4 inch "chinese" phone on landscape.
affected devices - 47 crashes in 28 days suggest that this bug is not very reproducible or only happens on uncommon devices samsung SCH-I535 LGE LG-E980 LGE LG-D802 samsung GT-I9505 samsung SGH-T889 Blackview Crown samsung SM-N9005 HTC One samsung GT-I9195 samsung GT-N8000 samsung GT-I9305 LGE LG-D838 zte Z987 samsung SM-G900V LGE LG-D851 HTC HTC One_M8 asus Nexus 7 Sony C6903 samsung GT-I9506 samsung SM-G7102 samsung SM-G360V samsung GT-I9300I samsung SCH-I605 Xiaomi HM NOTE 1LTEW samsung SPH-L710T bq Aquaris E6 Sony SO-05F samsung SCH-R970Xsamsung SCH-I535 LGE LG-E980 LGE LG-D802 samsung GT-I9505 samsung SGH-T889 Blackview Crown samsung SM-N9005 HTC One samsung GT-I9195 samsung GT-N8000 samsung GT-I9305 LGE LG-D838 zte Z987 samsung SM-G900V LGE LG-D851 HTC HTC One_M8 asus Nexus 7 Sony C6903 samsung GT-I9506 samsung SM-G7102 samsung SM-G360V samsung GT-I9300I samsung SCH-I605 Xiaomi HM NOTE 1LTEW samsung SPH-L710T bq Aquaris E6 Sony SO-05F samsung SCH-R970X Sony C6603 Sony D6708 samsung SM-N7502 Acer A1-810 samsung GT-N7100 samsung GT-N5110 Micromax Micromax A310 HUAWEI G7-L01 zte Z970 kyocera C6530N Sony C6603 Sony D6708 samsung SM-N7502 Acer A1-810 samsung GT-N7100 samsung GT-N5110 Micromax Micromax A310 HUAWEI G7-L01 zte Z970 kyocera C6530N
(In reply to Kevin Brosnan [:kbrosnan] from comment #4) > affected devices - 47 crashes in 28 days suggest that this bug is not very > reproducible or only happens on uncommon devices Since the bug completely makes Firefox unusable with the keyboard, users either stop using Firefox or Google Keyboard... Those devices look like fairily common - variants of Samsung S3, S4, Note phones & tablet. I reproduced it on every 4 inch phone that I tried...
I have a Samsung GS3 and can't repro this [0]. There was a vaguely related patch [1] [2] just put in to fix an issue with keyboard (Google and Swift), but a) it wasn't crashing, and b) it didn't start until v41 ... Your original report had you running v38, given the Play Store has release v39, and Beta v40, have you tried updating / testing those? Also, have you reviewed your add-ons for possible conflicts? Maybe disable all as part of your test? [0] https://www.dropbox.com/s/yf5gcic96ls3uic/bug1170675.png?dl=0 [1] Bug 1175838 - Swift Keyboard broken on www.google.com [2] Bug 1177011 - In IMDB search field, text is duplicated each time you type a character
I am now on 39 and it still crashes. I installed 40 beta, without any addons, and still crashes.
(In reply to Mark Capella [:capella] from comment #6) > I have a Samsung GS3 and can't repro this [0]. Here is a way to reproduce it: 0.1. Make sure that Google Keyboard is set as default. 0.2. Make sure that you are on LANDSCAPE all the time (NOT rotating it). 1. Close all tabs. 2. Close Firefox. Swipe it from recent apps. Force stop it from the Android menu. 3. Open Firefox. 4. Go to www.google.com (mobile site). 5. Tap on the text box for the keyboard to come up. 6. The text cursor does not blink. Inputed text does not enter the field. After a couple of seconds, Firefox freezes. I can't reproduce it 100% of tries, but it does like 2 of 3 tries.
Flags: needinfo?(markcapella)
Oh yah ... I see it ... :/
Flags: needinfo?(markcapella)
Build w/enable-debug and traced with gdb shows [0] Looks gfx related (?) I think I'll semi-random ping kats first. kats ... any ideas on this? Or who might look if not you? https://www.dropbox.com/s/9q30l06by01we9v/bug1170675_crash.log?dl=0
Flags: needinfo?(bugmail.mozilla)
fyi, I ran several more crashes and find this is a more representative trace: https://www.dropbox.com/s/t0hszwfwvb7on2e/bug1170675_crash3.log?dl=0
It would help if those logs had actual gdb backtraces. Regardless this looks like a question for nical looking at the crash stack from the crash report.
Status: UNCONFIRMED → NEW
Ever confirmed: true
Flags: needinfo?(bugmail.mozilla) → needinfo?(nical.bugzilla)
The OOM allocation size is 263272 on the crash report from comment 1. That's a crazy size for a tile buffer.
It could be that we take a stupidly big invalid region ClientTiledLayerBuffer::update (for some reason that needs to be figured out) and allocate tiles for it without restricting it to the display port? Still digging.
first day with the new debugger :) maybe this helps https://www.dropbox.com/s/sfw1xgcgoy9i695/bug1170675_crash_bt.log?dl=0
So that I don't forget some important info: Mark is helping me debug this over irc (he's got a gdb session open with the crash) while I am getting a phone that can reproduces the issue. The critical displayport is *huge* x:0 y:0 width 1280: height:16777842 (that's what's in mPaintData.mCriticalDisplayPort in RenderLayer).
Assignee: nobody → nical.bugzilla
Flags: needinfo?(nical.bugzilla)
That's assuming gdb isn't lying - it does that sometimes so be careful! It might be worth reproducing it with some logging enabled - I would want to see layers.dump=true, and ENABLE_APZCTM_LOGGING (at the top of APZCTreeManager.cpp) and ENABLE_APZC_LOGGING (at the top of AsyncPanZoomController.cpp) enabled. That should at least confirm the displayport size issue and might also tell us why it's happening.
(In reply to Kartikaya Gupta (email:kats@mozilla.com) from comment #17) > That's assuming gdb isn't lying - it does that sometimes so be careful! true that, but a crazy invalid region or displayport is the only situation I can think of that would cause us to allocate a huge nsTArray for the tile buffer (which appears in several of the crash reports).
Assignee: nical.bugzilla → jnicol
Closing because no crash reported since 12 weeks.
Status: NEW → RESOLVED
Closed: 7 years ago
Resolution: --- → WONTFIX
Product: Firefox for Android → Firefox for Android Graveyard
You need to log in before you can comment on or make changes to this bug.