760439 - OMTC: Nightly and Thunderbird Daily builds crash with abort message: "creating ThebesLayer 'back buffer' failed!"

Status: RESOLVED FIXED

(Core :: Graphics: Layers, defect)

Description

[IU]

With OMTC enabled, neither Firefox nor Thunderbird will start.  They keep crashing.  I had to manually edit pref.js and change the value of layers.offmainthreadcomposition.enabled to false

This happens on both Windows XP and Windows 7.

https://crash-stats.mozilla.com/report/index/bp-9dff1ee9-1eda-46c3-89f0-5b8112120601

Crashing Thread
Frame 	Module 	Signature 	Source
0 	mozalloc.dll 	mozalloc_abort 	memory/mozalloc/mozalloc_abort.cpp:46
1 	xul.dll 	NS_DebugBreak_P 	xpcom/base/nsDebugImpl.cpp:369
2 	nspr4.dll 	nspr4.dll@0x8fcf 	
3 		@0x1b

Comment 1

[Ali Juma [:ajuma]]

Removing the regression keyword since OMTC has never worked on Windows yet.

Comment 2

[IU]

There's the following in that crash report:

AdapterVendorID: 0x0000, AdapterDeviceID: 0x0000, AdapterSubsysID: 00000000, AdapterDriverVersion: 

xpcom_runtime_abort(###!!! ABORT: creating ThebesLayer 'back buffer' failed! width=1003, height=76, type=1000: file e:/builds/moz2_slave/m-cen-w32-ntly/build/gfx/layers/basic/BasicLayers.cpp, line 2525)

Comment 3

[Benoit Girard (:BenWa)]

Just to be clear you enabled the pref manually right? Please re-open if you did not.

In that case I'm not concerned because the feature is still in development and some critical patches have not landed.

Thanks for trying this out anyways, please try it again once we clear some of the dependent issues and part 2 lands.

Comment 5

[Benoit Girard (:BenWa)]

I've decide to remove the pref from the listing. This way the users will have to jump towards more hoops to get a broken state.

This will help since we will be asking people to turn on OMTC in central in 16-17 and when 15 rides the train people who will try it there will get a broken state and have to modify pref.js to fix it.

Comment 6

[Benoit Girard (:BenWa)]

Attachment: Remove pref from default listing

Comment 7

[Benoit Girard (:BenWa)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/c4d9a10d000b

Comment 8

[Brian Carpenter [:geeknik]]

Not sure if it matters, but setting layers.offmainthreadcomposition.enabled to true in the 5/30 and 5/31 Nightly builds didn't result in a crash, just started in the 6/1 build. x64 here (again, if that matters).

Comment 9

[Ali Juma [:ajuma]]

(In reply to Brian Carpenter [:geeknik] from comment #8)
> Not sure if it matters, but setting layers.offmainthreadcomposition.enabled
> to true in the 5/30 and 5/31 Nightly builds didn't result in a crash, just
> started in the 6/1 build. x64 here (again, if that matters).

That's because up until that point, the pref was entirely ignored on Windows. Starting with the 6/1 build, the pref is used, and since OMTC on Windows doesn't actually work yet, we crash.

Comment 10

[Benoit Girard (:BenWa)]

(In reply to Brian Carpenter [:geeknik] from comment #8)
> Not sure if it matters, but setting layers.offmainthreadcomposition.enabled
> to true in the 5/30 and 5/31 Nightly builds didn't result in a crash, just
> started in the 6/1 build. x64 here (again, if that matters).

It's expected, we went from not supporting anything to an in progress implementation. OMTC is a bit to big to implement as a patch queue so we hide the progress behind a pref rather then use a branch.

Comment 11

[Brian Carpenter [:geeknik]]

Good to know, thanks for clarifying. =)

Comment 12

[(no longer active)]

https://hg.mozilla.org/mozilla-central/rev/c4d9a10d000b

Comment 13

[Scoobidiver (away)]

The fix doesn't prevent crashes.

Comment 14

[Benoit Girard (:BenWa)]

It's not going to. These users are running an unsupported feature that we never asked anyone to test on desktop.

Comment 15

[Andreas Gal :gal]

BenWa, we are working on off-main-thread animations as a critical feature for B2G. Development is happening on mac, and will then be ported to Gonk. While maybe not a default configuration yet, this sudden breakage is massively breaking our work. Can we please back it out until we have a fix?

Comment 16

[Benoit Girard (:BenWa)]

It shouldn't be hard to fix but I'm fine if we backout if we're in a rush.

Comment 17

[Scoobidiver (away)]

It's similar to bug 717658.

Comment 18

[Marco Castelluccio [:marco]]

I think this is bug 717658. Testing on Windows, the problem is in ShareToProcess (just as in bug 717658).

Comment 19

[Benoit Girard (:BenWa)]

This should of been closed when the patch merged.