X-Mailer: QUALCOMM Windows Eudora Version 5.1
Date: Thu, 20 Sep 2001 10:16:06 -0500
To: Peter Ganster <P.Ganster@XXXXXXXXXXXX>
From: William Gropp <gropp@XXXXXXXXXXX>
Subject: Re: blocklength zero in MPI_Type_struct 
Cc: mpi-comments@XXXXXXXXXXXXX
In-Reply-To: <3B5EAA9E.1743FEE3@XXXXXXXXXXXX>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-mpi-comments@XXXXXXXXXXXXX
Precedence: bulk

At 01:16 PM 7/25/2001 +0200, Peter Ganster wrote:
Hello,

I have a question:

What happens if one use blocklenghts of zero in the call of
MPI_Type_struct: Is the behavior implementation dependent or is it
definded by the MPI-1 or MPI-2 standard?

The standard seems to allow this.  Interestingly enough, other Datatype routines are careful to specify that the values in the array of blocklengths are non-negative rather than positive (see MPI_TYPE_HINDEXED).  This is strengthed by the comment at the end of 3.12.1 that states that a call to MPI_TYPE_HINDEXED is equivalent to a call to MPI_TYPE_STRUCT where all of the datatypes are of the same type, suggesting that since zero blocklenghts are allowed in MPI_TYPE_HINDEXED, they must be in MPI_TYPE_STRUCT.

Bill