[Mpi-forum] Reserved MPI_ prefix & namespace in C and Fortran
nmm1 at cam.ac.uk
Mon Aug 27 02:29:18 CDT 2012
On Aug 26 2012, Jeremiah Willcock wrote:
>Does "linker names" come under any of the categories so far? I know that
>some of the C compiler extensions to set linker names explicitly do not
>treat them as identifiers in the source code. Are those worth adding
>explicitly (to all languages, probably)?
The technical term in both C and Fortran is "external names", and Fortran
2003 BIND(C,NAME=...) sets ones that are not identifiers. So the current
wording is fine.
Just to be really nasty, many systems did/do have the concept of "linker
names" that are beyond any name concept in any of the languages they
support and, despite that, they can clash. But there is no reasonable
way to say "An MPI implementation must avoid provoking its linker into
creating artificial names beginning MPI_". This has arisen in various
language standards, and the universal decision was to keep the standard's
wording simple - even C never got into this one, though it was raised!
More information about the mpi-forum