ubuntu-phone team mailing list archive

[Ricardo Salveti de Araujo <ricardo.salveti@xxxxxxxxxxxxx>]

On Wed, Feb 26, 2014 at 2:27 PM, Didier Roche <didrocks@xxxxxxxxxx> wrote:
> Hey,
>
> Some desktop only update are still flowing in through CI Train, however, we
> are still blocked by not being able to get one image promotable.
>
> Otherwise, two images were produced:
> #208 contained the second part of the sdk fix for the alarm
> #209 contains a libhybris change to protect bionic properties call.
>
> Here are the updates on current known regressions:
>
> 2. Unity8 (Michał/Ricardo)
> It seems there is a crash, debug symbols on the android side are building,
> Michał and Ricardo are looking into those. Seems it's maybe not Android-side
> related but a gcc 4.7 vs 4.8 ABI mismatch. Ricardo is trying to disable tls
> for more information.
>
> The investigation is still ongoing with Ricardo trying to disable multiple
> flags to see if that helps.

We spent some more time on this issue, and were able to check that
it's not related with Unity8 itself, but with Qt (v8 engine). It
crashes in the heap when it's trying to find the inline function
generator, which is the same crash we had with qmlscene and
gallery-app.

That's probably why removing the android TLS blocks, and also dropping
any pthread allocation inside libhybris didn't make any difference.
We're not yet sure what might be causing this behavior now, but hard
to say it's really because of 4.4.2 at this point.

ricmm was also trying to understand the crash a bit better, but not
sure if he got any further.

At the same time I also tried to reproduce this issue with Qt5.2, and
noticed that this crash is gone, but there is still another crash when
running the unity8 autopilot test cases. Opened bug
https://bugs.launchpad.net/ubuntu/+source/qtdeclarative-opensource-src/+bug/1285385
for this one (on the good side we finally have a useful stack trace
now).

> 3. QMLScene crash (Timo + SDK team)
> -> Timo told that this is in the V8 code and the experts don't have time and
> want to get Qt 5.2 asap rather.
>
> Another crash today, same than the one yesterday.

This seems to be the only crash actually :-) As we have a completely
different V8 implementation with Qt5.2 (old one is now dead), it's
going to be hard to find a fix for this guy with the 5.0 stack.

Cheers,
-- 
Ricardo Salveti de Araujo