[Geany-Devel] GeanyLaTeX segfault

[Matthew Brush]

On 14-04-09 01:15 PM, Frank Lanitz wrote:
> On Wed, 9 Apr 2014 23:57:49 +0400
> Pavel Roschin <roshin at scriptumplus.ru> wrote:
>
>>> On Wed, 9 Apr 2014 23:17:02 +0400
>>> Pavel Roschin <roshin at scriptumplus.ru> wrote:
>>>
>>>>> On Wed, 9 Apr 2014 20:48:35 +0200
>>>>> Frank Lanitz <frank at frank.uvena.de> wrote:
>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> Thanks for the work, bad for putting all together in one big
>>>>>> blob.
>>>>>>
>>>>>> On Wed, 9 Apr 2014 21:58:59 +0400
>>>>>> Pavel Roschin <roshin at scriptumplus.ru> wrote:
>>>>>>
>>>>>>> - geanylatex at close segfault (see Gist for backtrace and
>>>>>>> IRC for investigation)
>>>>>>
>>>>>> Will not search backlog of IRC. To much noise. What did you
>>>>>> do to get this?
>>>>>
>>>>> Even I see what might could be the cause, I cannot reproduce
>>>>> it. I need more details please.
>>>>>
>>>>> Cheers,
>>>>> Frank
>>>>
>>>>
>>>> This not actually reproducible. I enabled all plugins, after time
>>>> Geany started to crash sometimes at exit.
>>>>
>>>> This is IRC history about this bug:
>>>>
>>>> https://gist.github.com/scriptum/10304696
>>>>
>>>> In short: devhelp + geanylatex = crash at exit
>>>
>>> I assume it only happens, when a TeX file having focus on closing? I
>>> think for some reasons, and this is what b4n also mention on IRC,
>>> devhelp is destroying/changing the doc-pointer at some point.
>>
>>
>> Not sure. It happens even with empty (few *.c files) Geany. But who
>> knows what happens internally. I do not use Geany for TeX, so
>> probably it crashes without TeX documents (probably because I don't
>> remember all details). Currently I have all plugins disabled and
>> crashes didn't appear.
>
> I had some crash with devhelp at booth at CLT when I wanted to show
> somebody some feature. But wasn't able to reproduce. Maybe some timing
> thing?
>

Devhelp plugin shouldn't change document pointer, it only reparents the 
document notebook into another notebook, but it doesn't change its name 
and ui_lookup_widget() can still find it and such. Actually the concept 
is mostly copied from SplitWindow plugin which does something quite 
similar, so if you are running SplitWindow + Devhelp, I can almost 
guarantee some problems (although it really shouldn't *crash*, but 
rather spew assertion failures to terminal and just not work correctly).

The geanylatex bug I looked at from pastebin on IRC the other day looked 
to me like Scintilla's notification was firing after the GeanyEditor, 
GeanyDocument or GeanyFiletype was destroyed/re-used/whatever and it was 
dereferencing a NULL or invalid pointer.

Cheers,
Matthew Brush