[MPICH] empty group returned by MPI_Group_excl

[William Gropp]

I've added it to the official list of items, so no more should be  
necessary.

Bill

On Jul 6, 2007, at 2:39 AM, Toon Knapen wrote:

> so do you advise that I'll mention this on the mpi-forum ml?
>
> toon
>
>
> William Gropp wrote:
>> This is a good point, and one that deserves an errata item in the  
>> standard.  I would argue that for internal consistency,  
>> MPI_GROUP_EMPTY was intended by the Forum, and this case was  
>> overlooked.
>> Bill
>> On Jul 5, 2007, at 3:25 PM, Toon Knapen wrote:
>>> Indeed it makes sense that if Group_excl returns an empty group  
>>> that it returns MPI_GROUP_EMPTY and of course nothing in the  
>>> standard forbids this.
>>>
>>> But because you test explicitly on this, I just wanted to know if  
>>> this is standard-required or an mpich2-convention.
>>>
>>> thanks,
>>>
>>> toon
>>>
>>> Rajeev Thakur wrote:
>>>> It is not spelled out in the definition of MPI_Group_excl, but  
>>>> it is in
>>>> others. For example,
>>>> MPI_Group_difference: "The new group can be empty, that is,  
>>>> equal to
>>>> MPI_GROUP_EMPTY."
>>>> MPI_Group_incl: "If n = 0, then newgroup is MPI_GROUP_EMPTY." So  
>>>> I would guess that if MPI_Group_excl results in an empty group,  
>>>> one would
>>>> expect to get back MPI_GROUP_EMPTY.
>>>> Rajeev
>>>>> -----Original Message-----
>>>>> From: owner-mpich-discuss at mcs.anl.gov [mailto:owner-mpich- 
>>>>> discuss at mcs.anl.gov] On Behalf Of Toon Knapen
>>>>> Sent: Thursday, July 05, 2007 2:50 AM
>>>>> To: mpich-discuss at mcs.anl.gov
>>>>> Subject: [MPICH] empty group returned by MPI_Group_excl
>>>>>
>>>>> Hi all,
>>>>>
>>>>> I noticed that there is a test in test/mpi/group/grouptest.c of  
>>>>> mpich2-1.0.5p4 that verifies that a group of which all ranks  
>>>>> are excluded (using MPI_Group_excl) is equal to MPI_GROUP_EMPTY.
>>>>>
>>>>> Is this really required by the standard? AFAICT the standard  
>>>>> does not require that an empty group is equal to  
>>>>> MPI_GROUP_EMPTY (both groups of course will be MPI_IDENT).
>>>>>
>>>>> Thanks in advance,
>>>>>
>>>>> Toon Knapen
>>>>>
>>>>>
>>>
>