Re: MPI code error raises a question

Richard Maine wrote:
> Gib Bogle <bogle@[EMAIL PROTECTED]
> wrote:
> 
>> Since the MPI library is accessed via a USE statement, doesn't this 
>> imply that the same information is available to the compiler as an 
>> explicit interface would provide?
> 
> No, that doesn't follow. It might happen to be true for MPI in specific;
> I don't know. But it does not follow from being accessed via a USE
> statement.
> 
> It is perfectly possible to access implicit-interface procedures via
> USE. Admitedly, the most common type of procedure to access via use
> would be a module procedure, in which case the USE does provide an
> explicit interface. But "Access via USE" does not directly imply "module
> procedure".
> 
> For a trivial example, consider
> 
>   module some_module
>     external some_procedure
>   end module
> 
> This is a perfectly fine module that provides access to the external
> procedure some_procedure. USEing this module will give you an implicit
> interface to the procedure, as there sure isn't any information here to
> provide anything else.
> 
> Yes, by the way, there are occasionally reasons to do things like that.
> They aren't common, but they do come up. And it is even vaguely possible
> that MPI might sit in one of the corners where such a thing is useful
> (in particular, if there is type cheating going on that would cause
> compilers to ***** if they noticed).
> 

Thanks Richard, my understanding was deficient.  If I understand you
correctly, this seems to imply 
that I need to create explicit interfaces for all the MPI procedures I'm
using, or, better, persuade 
the MPICH2 developers that it would be useful to provide, in the Fortran
distribution, a file 
containing explicit interfaces to all the MPI procedures.  I wasted an
enormous amount of time 
trying to track down my error, and I hate to think of others having to go
through the same process.