MPIAbort should be called when DOLFIN throws an error

I guess this is a problem in the Python interface too, when Exceptions
are raised. If we implement a call to MPIAbort as part of dolfin_error,
it is another argument to not only py3fy all Python exceptions, but to
replace them with calls to dolfin_error.

Johan

On 01/06/2013 05:31 PM, Garth Wells wrote:
> Public bug reported:
>
> DOLFIN can deadlock when one process throws a error and others don't.
> The solution is probably to call MPIAbort when throwing an error.
>
> ** Affects: dolfin
> Importance: High
> Status: New
>
> ** Changed in: dolfin
> Importance: Undecided => High
>

Agree in principle. But keep in mind that some python exceptions are part
of special function protocols and are not errors.

Martin, on the train to Cambridge
Den 6. jan. 2013 19:15 skrev "Johan Hake" <email address hidden>
følgende:

> I guess this is a problem in the Python interface too, when Exceptions
> are raised. If we implement a call to MPIAbort as part of dolfin_error,
> it is another argument to not only py3fy all Python exceptions, but to
> replace them with calls to dolfin_error.
>
> Johan
>
> On 01/06/2013 05:31 PM, Garth Wells wrote:
> > Public bug reported:
> >
> > DOLFIN can deadlock when one process throws a error and others don't.
> > The solution is probably to call MPIAbort when throwing an error.
> >
> > ** Affects: dolfin
> > Importance: High
> > Status: New
> >
> > ** Changed in: dolfin
> > Importance: Undecided => High
> >
>
> --
> You received this bug notification because you are a member of DOLFIN
> Core Team, which is subscribed to DOLFIN.
> https://bugs.launchpad.net/bugs/1096627
>
> Title:
> MPIAbort should be called when DOLFIN throws an error
>
> Status in DOLFIN:
> New
>
> Bug description:
> DOLFIN can deadlock when one process throws a error and others don't.
> The solution is probably to call MPIAbort when throwing an error.
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/dolfin/+bug/1096627/+subscriptions
>

I see that, but I have silently ignored it...

We could extend the C++ interface to reflect this (templated function or
similar), but it will not cover the most important, TypeError.

To address this properly we need a separate bug report.

Johan

On 01/06/2013 10:57 PM, Martin Sandve Alnæs wrote:
> Agree in principle. But keep in mind that some python exceptions are part
> of special function protocols and are not errors.
>
> Martin, on the train to Cambridge
> Den 6. jan. 2013 19:15 skrev "Johan Hake" <email address hidden>
> følgende:
>
>> I guess this is a problem in the Python interface too, when Exceptions
>> are raised. If we implement a call to MPIAbort as part of dolfin_error,
>> it is another argument to not only py3fy all Python exceptions, but to
>> replace them with calls to dolfin_error.
>>
>> Johan
>>
>> On 01/06/2013 05:31 PM, Garth Wells wrote:
>>> Public bug reported:
>>>
>>> DOLFIN can deadlock when one process throws a error and others don't.
>>> The solution is probably to call MPIAbort when throwing an error.
>>>
>>> ** Affects: dolfin
>>> Importance: High
>>> Status: New
>>>
>>> ** Changed in: dolfin
>>> Importance: Undecided => High
>>>
>>
>> --
>> You received this bug notification because you are a member of DOLFIN
>> Core Team, which is subscribed to DOLFIN.
>> https://bugs.launchpad.net/bugs/1096627
>>
>> Title:
>> MPIAbort should be called when DOLFIN throws an error
>>
>> Status in DOLFIN:
>> New
>>
>> Bug description:
>> DOLFIN can deadlock when one process throws a error and others don't.
>> The solution is probably to call MPIAbort when throwing an error.
>>
>> To manage notifications about this bug go to:
>> https://bugs.launchpad.net/dolfin/+bug/1096627/+subscriptions
>>
>