[mythtv] DVB SI rework - Help needed

Roger James roger at beardandsandals.co.uk
Mon Jan 23 17:24:33 UTC 2017


On 22 January 2017 9:49:40 pm Peter Bennett <pgbennett at comcast.net> wrote:

> On 01/22/2017 11:26 AM, Roger James wrote:
>> On 22 January 2017 2:58:23 pm Peter Bennett <pgbennett at comcast.net>
>> wrote:
>>
>>> On 01/22/2017 09:40 AM, roger wrote:
>>>> As a complete aside. If anyone reading this is interested in looking
>>>> at another fascinating bug in myth. Try running mythfrontend on a fast
>>>> 64bit linux frontend with a ssd drive, and starting the frontend from
>>>> an autostart file. Lots of interesting thread safety issues here! Just
>>>> remember that exit code 139 is probably a SIGSEGV and not a
>>>> GENERIC_EXIT_INVALID_TIME (Darn those overlaps between uncaught
>>>> signals and myth error codes when you are in a shell script ;-) ).
>>>
>>> I am working on the frontend startup, fixing the cases where the
>>> frontend starts before the database and similar issues. There is not
>>> normally a seg fault, just confusing prompts about entering your
>>> language and such. I don't know how to recreate your error. I don't have
>>> an ssd drive. Maybe if you have some more info I can look at it at the
>>> same time.
>>>
>>> Peter
>>>
>>
>> Hi Peter
>>
>> I did raise an issue on this a while back here
>>
>> https://code.mythtv.org/trac/ticket/12909
>>
>> And here
>>
>> https://bugs.launchpad.net/mythbuntu/+bug/1635685
>>
>> I coded up a fix for the crash in QImage but that just moved the crash
>> a bit further down the mythfrontend initialisation sequence into a
>> number of random other crashes all of which seem to be Qt signalling
>> related. I can send you stack traces for these if you wish. I think I
>> still have the cores.
>>
>> This all seems to relate to Qt not being thread safe or reentrant in a
>> lot of areas. This exacerbated by running on a fast multi core 64bit
>> processor with a fast SSD drive.
>>
>> Cheers
>>
>> Roger
>>
>
> I saw that ticket and I have also had some issues with the threaded
> QImage. Did you try setting the environment variable
> DISABLETHREADEDMYTHUIIMAGE ? There is some code that disables the
> threaded load if you set that variable to any value (see
> libs/libmythui/mythuiimage.cpp).
>
> Peter
>

Peter,

Thanks, I tried that setting with my current build which has my own 
ImageLoad patch in it. The behaviour was unchanged, random crash in the Qt 
event signalling code. Had a quick look at the core, 30+ threads running a 
large number in QtWebKit. I think that probably shows that these random 
crashes are not an after effect of the a corruption in the Image loading 
process, but a separate threading problem. The second run of mythfrontend 
which the startup script loops round to always works fine. I suspect this 
means that there is some other external (to mythfrontend) boot time 
activity that is playing a part.

Roger




More information about the mythtv-dev mailing list