820578 - B2G_DEBUG=1 for beta/nightly needs help

Status: RESOLVED INCOMPLETE

(Firefox OS Graveyard :: General, defect)

Description

[Michael Vines [:m1] [:evilmachines]]

Beta + nightly fall down pretty badly at the moment when B2G_DEBUG=1 is enabled.

Some immediate issues:
* We fail a couple assertions right at boot
** widget/gonk/nsWindow
** widget/xpwidgets/nsBaseScreen.cpp
* FTU app crashes

I've seen other transitory assertion failures in as well but no solid STR yet.

Comment 1

[Michael Vines [:m1] [:evilmachines]]

I've seen these two as well but no solid STR to share (other that boot a couple times and play with the device):
* [Parent 602] ###!!! ASSERTION: Existing entry in disk StartupCache.: 'NS_SUCCEEDED(rv) && hasEntry == false', file .../gecko/startupcache/StartupCache.cpp, line 398
* http://mxr.mozilla.org/mozilla-beta/source/toolkit/crashreporter/nsExceptionHandler.cpp#1897

Comment 2

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Just for some background (this was IRC ping), we have two forms of "assertion".
 - MOZ_ASSERT(): same as C assert().  Fatal error.  Any of these failing is serious.
 - NS_ASSERTION(): not the same semantics as C assert().  We use this to mark unexpected program states that should be investigated when hit, but aren't fatal errors.  Unfortunately it's used somewhat inconsistently.  We have to address these on a case by case basis.

Comment 3

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

(In reply to Michael Vines [:m1] from comment #0)
> Beta + nightly fall down pretty badly at the moment when B2G_DEBUG=1 is
> enabled.
> 
> Some immediate issues:
> * We fail a couple assertions right at boot
> ** widget/gonk/nsWindow
> ** widget/xpwidgets/nsBaseScreen.cpp

Hm, I don't see these.  Are you by chance running with HwcComposer2D enabled?

> * FTU app crashes
> 

Ooh, I do reproduce this, but only on device bootup.  Looks like a bad one.

(In reply to Michael Vines [:m1] from comment #1)
> I've seen these two as well but no solid STR to share (other that boot a
> couple times and play with the device):
> * [Parent 602] ###!!! ASSERTION: Existing entry in disk StartupCache.:
> 'NS_SUCCEEDED(rv) && hasEntry == false', file
> .../gecko/startupcache/StartupCache.cpp, line 398

mwu, is this an assert() or a warn()?

> *
> http://mxr.mozilla.org/mozilla-beta/source/toolkit/crashreporter/
> nsExceptionHandler.cpp#1897

Couldn't reproduce but this would be bad.

Overall this isn't an unusual number of NS_ASSERTION warnings, but the two MOZ_ASSERT() failures above look bad.

Comment 4

[Michael Vines [:m1] [:evilmachines]]

(In reply to Chris Jones [:cjones] [:warhammer] from comment #3)
> (In reply to Michael Vines [:m1] from comment #0)
> > Beta + nightly fall down pretty badly at the moment when B2G_DEBUG=1 is
> > enabled.
> > 
> > Some immediate issues:
> > * We fail a couple assertions right at boot
> > ** widget/gonk/nsWindow
> > ** widget/xpwidgets/nsBaseScreen.cpp
> 
> Hm, I don't see these.  Are you by chance running with HwcComposer2D enabled?

I disabled that just to be sure.  Removed some other noncommon bits as well.  I'm running with bootanimation.zip so maybe that's the diff.

> 
> > * FTU app crashes
> > 
> 
> Ooh, I do reproduce this, but only on device bootup.  Looks like a bad one.
> 
> (In reply to Michael Vines [:m1] from comment #1)
> > I've seen these two as well but no solid STR to share (other that boot a
> > couple times and play with the device):
> > * [Parent 602] ###!!! ASSERTION: Existing entry in disk StartupCache.:
> > 'NS_SUCCEEDED(rv) && hasEntry == false', file
> > .../gecko/startupcache/StartupCache.cpp, line 398
> 
> mwu, is this an assert() or a warn()?

It was aborting the system process.  I had to whack /data/b2g to recover.  Will try to get STR on this.

> 
> > *
> > http://mxr.mozilla.org/mozilla-beta/source/toolkit/crashreporter/
> > nsExceptionHandler.cpp#1897
> 
> Couldn't reproduce but this would be bad.

I'll try harder to get STR on this one too.

Comment 5

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

(In reply to Michael Vines [:m1] from comment #4)
> (In reply to Chris Jones [:cjones] [:warhammer] from comment #3)
> > (In reply to Michael Vines [:m1] from comment #0)
> > > Beta + nightly fall down pretty badly at the moment when B2G_DEBUG=1 is
> > > enabled.
> > > 
> > > Some immediate issues:
> > > * We fail a couple assertions right at boot
> > > ** widget/gonk/nsWindow
> > > ** widget/xpwidgets/nsBaseScreen.cpp
> > 
> > Hm, I don't see these.  Are you by chance running with HwcComposer2D enabled?
> 
> I disabled that just to be sure.  Removed some other noncommon bits as well.
> I'm running with bootanimation.zip so maybe that's the diff.

Aha, seems likely.  I don't have that.

> > (In reply to Michael Vines [:m1] from comment #1)
> > > I've seen these two as well but no solid STR to share (other that boot a
> > > couple times and play with the device):
> > > * [Parent 602] ###!!! ASSERTION: Existing entry in disk StartupCache.:
> > > 'NS_SUCCEEDED(rv) && hasEntry == false', file
> > > .../gecko/startupcache/StartupCache.cpp, line 398
> > 
> > mwu, is this an assert() or a warn()?
> 
> It was aborting the system process.  I had to whack /data/b2g to recover. 
> Will try to get STR on this.

NS_ASSERTION is nonfatal unless you enable some power-user options.  Did you do that, or maybe was there something else in the logcat just afterwards?

> > 
> > > *
> > > http://mxr.mozilla.org/mozilla-beta/source/toolkit/crashreporter/
> > > nsExceptionHandler.cpp#1897
> > 
> > Couldn't reproduce but this would be bad.
> 
> I'll try harder to get STR on this one too.

I have a guess for what this might be so STR might not be needed.

Comment 6

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

(In reply to Chris Jones [:cjones] [:warhammer] from comment #5)
> (In reply to Michael Vines [:m1] from comment #4)
> > (In reply to Chris Jones [:cjones] [:warhammer] from comment #3)
> > > (In reply to Michael Vines [:m1] from comment #0)
> > > > *
> > > > http://mxr.mozilla.org/mozilla-beta/source/toolkit/crashreporter/
> > > > nsExceptionHandler.cpp#1897
> > > 
> > > Couldn't reproduce but this would be bad.
> > 
> > I'll try harder to get STR on this one too.
> 
> I have a guess for what this might be so STR might not be needed.

If my guess is right, the code running after this would probably just crash hard so the logcat would look like a failed fatal assertion.

Comment 7

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

s/would/might/

Comment 8

[Michael Wu [:mwu]]

The startup cache warning/assertion itself shouldn't be fatal but could mean very bad things. The particular variant you got is especially weird.

Comment 9

[Andrew Overholt [:overholt]]

Andrea, please investigate.

Comment 10

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

This is a metabug.

Comment 11

[Mike Habicher [:mikeh] (high bugzilla latency)]

Filed bug 821444 about debug builds recently becoming unusably slow.

Comment 12

[Andrew Overholt [:overholt]]

Having this block bug 817946 is causing a weird dependency graph in another bug so I'm dropping it from the blocks list.

Comment 13

[Andrea Marchesini [:baku]]

I'm not actually working on it. Is this bug still valid?

Comment 14

[Mike Habicher [:mikeh] (high bugzilla latency)]

I recently filed bug 873573 and updated bug 841206 for hamachi debug builds.

Comment 15

[Michael Vines [:m1] [:evilmachines]]

This might still be an issue but is not something I care a lot about anymore.