[Android] Spill out CHECK error message in official build

[Android] Spill out CHECK error message in official build

In M35 release, we had a lot of user feedbacks of Chrome native crash.
We had great difficulty to investigate these native crashes
because user feedbacks don't have minidumps.
    
By keeping the CHECK error message in logcat, it'd be a lot easier
to investigate these user reports of Chrome crashes.

The binary size increase: ~0.5% (188k out of 34.4M).

BUG=378974
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=281859

[Nico, 2014-06-17 18:28:30]

Why are clients unable to send minidumps?

[Feng Qian, 2014-06-17 18:47:19]

On 2014/06/17 18:28:30, Nico (away) wrote:
> Why are clients unable to send minidumps?

There are a lot of reasons, network, out of memory, we limit the number of
minidumps for saving bandwidth, etc.

[Nico, 2014-06-17 18:48:21]

Are they fixable?

None of them sound that android specific, and making CHECK behave different
on different platforms seems pretty weird.


On Tue, Jun 17, 2014 at 11:47 AM, <feng@chromium.org> wrote:

> On 2014/06/17 18:28:30, Nico (away) wrote:
>
>> Why are clients unable to send minidumps?
>>
>
> There are a lot of reasons, network, out of memory, we limit the number of
> minidumps for saving bandwidth, etc.
>
>
> https://codereview.chromium.org/336413005/
>

To unsubscribe from this group and stop receiving emails from it, send an email
to chromium-reviews+unsubscribe@chromium.org.

[Feng Qian, 2014-06-17 18:54:41]

On 2014/06/17 18:48:21, Nico (away) wrote:
> Are they fixable?
> 
> None of them sound that android specific, and making CHECK behave different
> on different platforms seems pretty weird.

A user may not be on WiFi for days, I don't know how to fix it.
In the past months, we received a lot of user reports from ConOps,
with truncated back traces. These reports don't have minidump
associated with them. Without CHECK message, it was almost
impossible to investigate.

Here is the latest incident: 
https://code.google.com/p/chromium/issues/detail?id=376072

We spent days on finding a reproducible case, and was luck to have on
in office. With CHECK message, it will take less than 5 seconds to
understand the cause.

There is another issue here:
https://code.google.com/p/chromium/issues/detail?id=381983
I don't have a clue yet.

Combination of two made us not able to investigate a lot of user
reported crashes at this point.



> 
> 
> On Tue, Jun 17, 2014 at 11:47 AM, <mailto:feng@chromium.org> wrote:
> 
> > On 2014/06/17 18:28:30, Nico (away) wrote:
> >
> >> Why are clients unable to send minidumps?
> >>
> >
> > There are a lot of reasons, network, out of memory, we limit the number of
> > minidumps for saving bandwidth, etc.
> >
> >
> > https://codereview.chromium.org/336413005/
> >
> 
> To unsubscribe from this group and stop receiving emails from it, send an
email
> to mailto:chromium-reviews+unsubscribe@chromium.org.

[Feng Qian, 2014-06-18 22:04:44]

On 2014/06/17 18:54:41, Feng Qian wrote:
> On 2014/06/17 18:48:21, Nico (away) wrote:
> > Are they fixable?
> > 
> > None of them sound that android specific, and making CHECK behave different
> > on different platforms seems pretty weird.
> 
> A user may not be on WiFi for days, I don't know how to fix it.
> In the past months, we received a lot of user reports from ConOps,
> with truncated back traces. These reports don't have minidump
> associated with them. Without CHECK message, it was almost
> impossible to investigate.
> 
> Here is the latest incident: 
> https://code.google.com/p/chromium/issues/detail?id=376072
> 
> We spent days on finding a reproducible case, and was luck to have on
> in office. With CHECK message, it will take less than 5 seconds to
> understand the cause.
> 
> There is another issue here:
> https://code.google.com/p/chromium/issues/detail?id=381983
> I don't have a clue yet.
> 
> Combination of two made us not able to investigate a lot of user
> reported crashes at this point.
> 
> 
> 
> > 
ping!


> > 
> > On Tue, Jun 17, 2014 at 11:47 AM, <mailto:feng@chromium.org> wrote:
> > 
> > > On 2014/06/17 18:28:30, Nico (away) wrote:
> > >
> > >> Why are clients unable to send minidumps?
> > >>
> > >
> > > There are a lot of reasons, network, out of memory, we limit the number of
> > > minidumps for saving bandwidth, etc.
> > >
> > >
> > > https://codereview.chromium.org/336413005/
> > >
> > 
> > To unsubscribe from this group and stop receiving emails from it, send an
> email
> > to mailto:chromium-reviews+unsubscribe@chromium.org.

[Feng Qian, 2014-06-24 17:54:12]

xianzhu & nico, are you OK with the change?

We need a way to get more information about CHECK failaures before M36 stable is
released.

We have another change that adds unwind table to binary, but that increases the
binary size too much.

Print out CHECK failure message seems a good trade-off.

[Nico, 2014-06-24 18:04:32]

I feel doing this differently on different platforms isn't a good idea. The wifi
issue is one we have on regular laptops too.

[Feng Qian, 2014-06-24 18:22:52]

On 2014/06/24 18:04:32, Nico (away) wrote:
> I feel doing this differently on different platforms isn't a good idea. The
wifi
> issue is one we have on regular laptops too.

Any suggestions on how to get CHECK failure message?

Android is already different from other platforms, in the sense we have mixed
Java
and C++. The minidump only captures one part of the crash, and Java exception is
left to logcat.

We have an initiative to capture Java exception stack traces and upload with
minidump,
but not sure when that effort will be done.

[Nico, 2014-07-07 20:14:52]

As discussed offline, the logcats come from places where it's difficult to get
access to minidumps. Since this only changes if the CHECK strings are included
and not the behavior of CHECKs, lgtm

One thing we discussed offline: I don't want to see is efforts to minimize the
CHECK strings after this goes in (say, I don't want people to say "please make
this CHECK message shorter, since it's included in the binary on Android). Feng
is happy with this (within reasons; folks shouldn't use needlessly long CHECK
strings. Looking through current CHECK usage, this shouldn't be a problem.)

[Feng Qian, 2014-07-08 17:18:22]

The CQ bit was checked by feng@chromium.org

[commit-bot: I haz the power, 2014-07-08 17:20:09]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/feng@chromium.org/336413005/20001

[commit-bot: I haz the power, 2014-07-08 22:52:40]

Change committed as 281859