[omniORB] Segfault in omniOrbPOA::dispatch

[Jean Aunis]

Le 18/06/2020 à 16:44, Duncan Grisby a écrit :
> On Wed, 2020-06-17 at 17:06 +0200, Jean Aunis via omniORB-list wrote:
>
>> I'm experiencing a crash with omniORB 4.2.0. After re-compiling with
>> debug flags I could trace it back to omni::omniOrbPOA::dispatch :
>>
>> #0  0x00007f776e127f5c in omni::omniOrbPOA::dispatch (this=0x3aff560,
>> handle=..., id=0x7f77b800ab30) at poa.cc:1712
> This is the point at which the POA invokes a _dispatch() method on the
> servant object that your application provided. The most likely
> explanation is that you have your servant reference counting wrong, and
> the servant has therefore been deleted.
>
> I suggest that you instrument your code to check the lifetime of your
> servant objects, or that you run in a memory debugger like valgrind.
>
> [...]
>> Have you already encountered this issue ? May it have been fixed in a
>> later release ?
> I don't think it is pertinent here, but you really ought to upgrade to
> 4.2.4, which is the latest on the 4.2 branch. 4.2.0 was released 6
> years ago.
>
> Duncan.
>
Thank you for your response. I ran my code with valgrind and as you 
wrote, the issue looks to be linked to a reference-counting issue.

4.2.0 is the version packaged with the latest CentOS release. 
Unfortunately CentOS is usually slow at integrating package updates.

Jean