1230428 - Crash [@ SpeechTaskCallback::OnWillSpeakWord]

Status: RESOLVED FIXED

(Core :: Web Speech, defect)

Description

[Jesse Ruderman]

Attachment: testcase (crashes Firefox)

Crash [@ SpeechTaskCallback::OnWillSpeakWord]

Comment 1

[Jesse Ruderman]

Attachment: stack

Comment 2

[Eitan Isaacson [:eeejay]]

Makoto,

This looks like it is crashing in the mac bits. Want to take a look?

Comment 3

[Makoto Kato [:m_kato]]

I take this.  Also, this reproduces on Windows too.

Comment 4

[Makoto Kato [:m_kato]]

This issue seems to be e10s only

Comment 5

[Makoto Kato [:m_kato]]

Oops, mac may not e10s only.   I should test it on mac.

Comment 6

[Bob Clary [:bc] (inactive)]

Bughunter reproduced this with Nightly using the attachment on OS X 10.8, 10.9. Haven't seen it yet on other OSs.

Comment 7

[Makoto Kato [:m_kato]]

Attachment: Part 1. Check mTask since end event is posted asynchronized

Comment 8

[Makoto Kato [:m_kato]]

Attachment: Part 2. Add crashtest

Comment 9

[Makoto Kato [:m_kato]]

Comment on attachment 8697063 [details] [diff] [review]
Part 1. Check mTask since end event is posted asynchronized

The callback into nsSpeechTask will be released by end event.  But since end event is posted asynchronized, when calling pause and resume, web content doesn't receive end event yet.   So we should check mTask on Windows and cocoa backend.

Comment 10

[Makoto Kato [:m_kato]]

Comment on attachment 8697065 [details] [diff] [review]
Part 2. Add crashtest

Part 2 test will causes leak on Linux build.  So mark as obsolete.

Comment 11

[Makoto Kato [:m_kato]]

Attachment: Part 2. Fix memory leak on speechd backend

When adding crashtest, I found memory leak on linux build.  mTask has to release when speaking is end.

Comment 12

[Makoto Kato [:m_kato]]

Attachment: Part 3. Add crashtest

add this fuzzing test

Comment 13

[Makoto Kato [:m_kato]]

Comment on attachment 8697085 [details] [diff] [review]
Part 3. Add crashtest

Add this fuzzing test

Comment 14

[Makoto Kato [:m_kato]]

Comment on attachment 8697083 [details] [diff] [review]
Part 2. Fix memory leak on speechd backend

I found another issue.  Canceling.  Sorry

Comment 15

[Makoto Kato [:m_kato]]

Comment on attachment 8697085 [details] [diff] [review]
Part 3. Add crashtest

Canceling too.  Sorry for spam

Comment 16

[Makoto Kato [:m_kato]]

Attachment: Part 2. Add crashtest

Comment 17

[Makoto Kato [:m_kato]]

Comment on attachment 8697238 [details] [diff] [review]
Part 2. Add crashtest

This test doesn't work on Android, gtkWidget and B2G, so I add skip list.  Android has no backend, and Linux and B2G doesn't have speech library on our test environment.

Comment 18

[Eitan Isaacson [:eeejay]]

Comment on attachment 8697063 [details] [diff] [review]
Part 1. Check mTask since end event is posted asynchronized

Review of attachment 8697063 [details] [diff] [review]:
-----------------------------------------------------------------

This looks good. Thank you!

Comment 19

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/f75a98aa4ef0
https://hg.mozilla.org/integration/mozilla-inbound/rev/5a889d2cc25b

Comment 20

[Phil Ringnalda (:philor)]

Backed out just the crashtest in https://hg.mozilla.org/integration/mozilla-inbound/rev/a5fa1e187b98

Apparently has a race that you lose when the test machine is too fast: only on Windows 7 (among the runs that happened while it was in the tree, anyway), more commonly on opt and PGO than on debug, which is why I think you're losing when it's running quicker, as often as not you got https://treeherder.mozilla.org/logviewer.html#?job_id=18472557&repo=mozilla-inbound

Comment 21

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/f75a98aa4ef0

Comment 22

[Makoto Kato [:m_kato]]

Humm, it isn't opt issue, and it is VM issue.  Cancel spends more time (1s?), but evne if opt build, test is successful.  Maybe, failed VM doesn't have any voice of speech engine or configuration.

So I will think another hack to run this test.

Comment 23

[Petruta Horea [:phorea], Desktop QA]

https://crash-stats.mozilla.com/report/index/bab65bf6-277c-4a2d-a687-be2db2160128

Encountered this crash on Firefox 45 beta 1 on Mac OS X 10.9.5.
I set up media.webspeech.synth.enabled to true in order to test Text to Speech feature.

STR:
On http://mdn.github.io/web-speech-api/speak-easy-synthesis/ or http://eeejay.github.io/webspeechdemos/ test pages, after entering some text and selecting a voice, start the speech and then select another voice. Refresh the page several times if needed.

Comment 24

[Makoto Kato [:m_kato]]

Attachment: crashtest v2

Comment 25

[Makoto Kato [:m_kato]]

Comment on attachment 8714633 [details] [diff] [review]
crashtest v2

test only on OSX.  Some Win7 VM is always failure due to unknown reason.　(ex. t-w732-ix-036 is always failure, but t-w732-ix-036 is successful).

Also, this is passed by https://treeherder.mozilla.org/#/jobs?repo=try&revision=06749a4a2afb

Comment 26

[Makoto Kato [:m_kato]]

(In reply to Petruta Rasa [QA] [:petruta] from comment #23)
> https://crash-stats.mozilla.com/report/index/bab65bf6-277c-4a2d-a687-
> be2db2160128
> 
> Encountered this crash on Firefox 45 beta 1 on Mac OS X 10.9.5.
> I set up media.webspeech.synth.enabled to true in order to test Text to
> Speech feature.
> 
> STR:
> On http://mdn.github.io/web-speech-api/speak-easy-synthesis/ or
> http://eeejay.github.io/webspeechdemos/ test pages, after entering some text
> and selecting a voice, start the speech and then select another voice.
> Refresh the page several times if needed.

Humm, this is another crash.  Could you file a new bug for Nightly?

Comment 27

[Eitan Isaacson [:eeejay]]

(In reply to Makoto Kato [:m_kato] from comment #25)
> Comment on attachment 8714633 [details] [diff] [review]
> crashtest v2
> 
> test only on OSX.  Some Win7 VM is always failure due to unknown reason.
> (ex. t-w732-ix-036 is always failure, but t-w732-ix-036 is successful).
> 
> Also, this is passed by
> https://treeherder.mozilla.org/#/jobs?repo=try&revision=06749a4a2afb

If it is a non intermittent failure, shouldn't it be possible to debug and figure out why it doesn't pass in windows?

Comment 28

[Petruta Horea [:phorea], Desktop QA]

(In reply to Makoto Kato [:m_kato] from comment #26)
> (In reply to Petruta Rasa [QA] [:petruta] from comment #23)
> > https://crash-stats.mozilla.com/report/index/bab65bf6-277c-4a2d-a687-
> > be2db2160128
> > 
> > Encountered this crash on Firefox 45 beta 1 on Mac OS X 10.9.5.
> > I set up media.webspeech.synth.enabled to true in order to test Text to
> > Speech feature.
> > 
> > STR:
> > On http://mdn.github.io/web-speech-api/speak-easy-synthesis/ or
> > http://eeejay.github.io/webspeechdemos/ test pages, after entering some text
> > and selecting a voice, start the speech and then select another voice.
> > Refresh the page several times if needed.
> 
> Humm, this is another crash.  Could you file a new bug for Nightly?

I couldn't reproduce it on latest Nightly 47.0a1 and latest Aurora 46.0a2 2016-02-02.
This only happens on Firefox 45 Beta 1 & 2 when enabling the webspeech feature.
Should I file it against Beta considering it won't be enabled by default in this version?

Comment 29

[Makoto Kato [:m_kato]]

(In reply to Eitan Isaacson [:eeejay] from comment #27)
> (In reply to Makoto Kato [:m_kato] from comment #25)
> > Comment on attachment 8714633 [details] [diff] [review]
> > crashtest v2
> > 
> > test only on OSX.  Some Win7 VM is always failure due to unknown reason.
> > (ex. t-w732-ix-036 is always failure, but t-w732-ix-036 is successful).
> > 
> > Also, this is passed by
> > https://treeherder.mozilla.org/#/jobs?repo=try&revision=06749a4a2afb
> 
> If it is a non intermittent failure, shouldn't it be possible to debug and
> figure out why it doesn't pass in windows?

I cannot reproduce on my environment.  But this occurs on our test infra for Win7 opt only.  I think that SAPI's Speak method spends more times to start it, then this occur.

Comment 30

[Makoto Kato [:m_kato]]

(In reply to Petruta Rasa [QA] [:petruta] from comment #28)
> (In reply to Makoto Kato [:m_kato] from comment #26)
> > (In reply to Petruta Rasa [QA] [:petruta] from comment #23)
> > > https://crash-stats.mozilla.com/report/index/bab65bf6-277c-4a2d-a687-
> > > be2db2160128
> > > 
> > > Encountered this crash on Firefox 45 beta 1 on Mac OS X 10.9.5.
> > > I set up media.webspeech.synth.enabled to true in order to test Text to
> > > Speech feature.
> > > 
> > > STR:
> > > On http://mdn.github.io/web-speech-api/speak-easy-synthesis/ or
> > > http://eeejay.github.io/webspeechdemos/ test pages, after entering some text
> > > and selecting a voice, start the speech and then select another voice.
> > > Refresh the page several times if needed.
> > 
> > Humm, this is another crash.  Could you file a new bug for Nightly?
> 
> I couldn't reproduce it on latest Nightly 47.0a1 and latest Aurora 46.0a2
> 2016-02-02.
> This only happens on Firefox 45 Beta 1 & 2 when enabling the webspeech
> feature.
> Should I file it against Beta considering it won't be enabled by default in
> this version?

This feature won't turn on 45.  So it is unnecessary.

Comment 31

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/6c2610f9334a

Comment 32

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/6c2610f9334a

Comment 33

[Makoto Kato [:m_kato]]

No crash, so I close this.  For OSX 10.6, we will continue by bug 1252673.