Date: Fri, 9 Feb 2001 11:18:40 -0500
From: Nicholas Nevin - Sun HPC High Performance Computing <Nicholas.Nevin@XXXXXXXXXXXX>
To: William Gropp <gropp@XXXXXXXXXXX>
Cc: Nicholas Nevin - Sun HPC High Performance Computing <Nicholas.Nevin@XXXXXXXXXXXX>,Subject: Re: MPI-2 standard issues
References: <20010209095646.C20056@caja> <4.2.2.20010209090721.034ea2c0@XXXXXXXXX>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.3.9i
In-Reply-To: <4.2.2.20010209090721.034ea2c0@XXXXXXXXX>

On Fri, Feb 09, 2001 at 09:19:17AM -0600, William Gropp wrote:
> At 09:56 AM 2/9/2001 -0500, Nicholas Nevin - Sun HPC High Performance 
> Computing wrote:
> >A have a couple of questions regarding the MPI-2 standard.
> >
> >1.  In section 8.5 is the following
> >
> >         "Instead, a predefined attribute key MPI_LASTUSEDCODE is
> >associated with MPI_COMM_WORLD. The attribute value corresponding to
> >this key is the current maximum error class including the user-defined
> >ones."
> >
> >Shouldn't "maximum error class" read "maximum error code"?
> 
> This reflects the statement from the MPI-1 standard that MPI_ERR_LASTCODE 
> is greater or equal to the value of every error *class*.  The text of the 
> description says "last standard error code", not "last error code".  One 
> reading of this is that MPI_ERR_LASTCODE (and the MPI-2 attribute 
> MPI_LASTUSEDCODE) defines the maximum defined error class.  Error codes can 
> be much larger; we encode information about the specific error case into 
> the code.  Our maximum error code is effectively MAX_INT.  But our maximum 
> error class is much smaller.
> 
> 

Yes but I think it is also quite reasonable to interpret the following

    "The error classes are a subset of the error codes: an MPI function
    may return an error class number; and the function MPI_ERROR_STRING
    can be used to compute the error string associated with an error
    class.

    The error codes satisfy, 0 = MPI_SUCCESS < MPI_ERR... <= MPI_ERR_LASTCODE".

to mean that MPI_ERR_LASTCODE is greater than or equal to all error
codes. Indeed the name suggests this. Otherwise why not
MPI_ERR_LASTCLASS? I can't find the text "last standard error code" in
the 1.1 standard.  Did I miss it?

The advice to users in MPI-2 8.5, "The value returned by the key
MPI_LASTUSEDCODE will not change unless the user calls a function to
explicitly add an error class/code." implies that adding an error code
might change the value associated with MPI_LASTUSEDCODE which appears
to contradict its definition as the "maximum error class".

> 
> >2. Why the restriction that types obtained from calls to
> >MPI_Type_create_f90_{real,complex,integer} cannot be freed? This
> >restriction doesn't appear to serve any useful purpose and the
> >requirement that MPI_Type_get_envelope return the precision/range used
> >to create such a type seems to either require an implementation to
> >maintain a potentially very large table or allocate dynamic storage
> >(which cannot be freed) to store the precision/range.
> 
> This looks like a mistake.  A correction could say that the type returned 
> is effectively a dup of the predefined type, it can (and should be) freed 
> by the user, and implementer could use internal attributes to store the 
> info needed for MPI_Type_get_envelope.  Note that for layered libraries on 
> heterogeneous system, it is necessary to provide the features specified for 
> MPI_Type_get_envelope.
> 

I'm happy with such a correction.

Thanks, Nick.