Date: Tue, 22 Jun 1999 10:54:24 -0500
From: Rajeev Thakur <thakur@XXXXXXXXXXX>
To: treumann@XXXXXXXXXX
CC: rusrabe@XXXXXXXXXXXXXXXXXXXXXXXXXX,raja@XXXXXXXXXX,linda@XXXXXXXXXXX,In-reply-to: <85256798.00543910.00@XXXXXXXXXXXXXXXXXXXX>
Subject: Re: Request for interpretation
Sender: owner-mpi-core@XXXXXXXXXXX
Precedence: bulk
X-UIDL: 1f0aca84bdc578852f8edb69b035726b

Since info was intended to be "optional", i.e. an implementation may
or may not pay attention to it, an alternative is to allow both
interpretations (as it currently is).  Then those implementations that
choose to allow layerability will not use MPI_Info_set as a filter,
and those that choose not to allow layerability will use MPI_Info as a
filter. 

Users of an implementation can, of course, indicate their
happiness or unhappiness with the choice of interpretation, and the
implementation can decide whether or not to pay attention to users :-).

Allowing both interpretations does not take away any portability since
info key-values have limited portability anyway --- an implementation
can ignore any predefined key-values and support additional key-values
of its own. It's only a question of layerability, and an
implementation may choose not to be layerable.

Rajeev