From: Douglas Gregor <dgregor@XXXXXXXXXX>
Date: January 14, 2008 8:35:12 AM CST
To: mpi-21@XXXXXXXXXXXXX
Subject: Re: [mpi-21] Proposal EH2: add const keyword to the C bindings
Reply-To: mpi-21@XXXXXXXXXXXXX


On Fri, 2008-01-11 at 16:58 -0500, David Gingold wrote:
> [I'm still trying to catch up on all of the e-mail messages before
> next week, so please excuse me if this has come up before.]
>
> Here is a way in which adding "const" to the C bindings could break
> an existing MPI program.  Suppose I write this:
>
> 	#include <mpi.h>
>
> 	void foo(void)
> 	{
> 	    typedef int (* sendfunc_t)(void *, int, MPI_Datatype, int, int,
> MPI_Comm);
> 	    sendfunc_t s;
> 	    s = MPI_Send;
> 	}
>
> If we re-define MPI_Send() so that its first argument has type "const
> void *", then this code becomes incorrect.

Just for reference, the C++ standard itself does not consider the
description of function signatures in the standard library to be exact.
C++ standard library implementations are permitted to add extra,
defaulted parameters to library functions, for example. So, if a program
takes the address of a C++ standard library function, it isn't
guaranteed what the type of that function pointer will be.

I mention this because your example is holding MPI implementations to a
higher level of consistency than is expected of the standard libraries
shipped with one of MPI's underlying languages. Perhaps we don't want to
guarantee so much.

   - Doug