o Split OPEN into simpler OPEN and VIEW.
o Added nonblocking IOPEN and ICLOSE.
o Replaced ``hints'' interface with ``info''.
o Added ``_EXPLICIT'' to explicit access routines, and eliminated
``_NEXT'' from READ/WRITE to be more UNIX-like.
o Added C++ bindings (not marked with change bars).
o Added proposed alternative to MPI_FILE_CONTROL.
In answer to specific comments on the draft:
John Wilson <jwilson@hal300.cs.colorado.edu> wrote:
> 1. I was looking over the type constructor functions associated
> with MPI-IO and have the following comment. It appears that
> MPI_Type_subarray assumes that the start of the array is at
> zero in C and one in Fortran, but no statement of this fact
> appears in the description. This should be made clear in the
> specification.
This statement has been added to MPI_TYPE_SUBARRAY.
John Wilson <jwilson@hal300.cs.colorado.edu> also wrote:
> 2. I also looked at File_control and it would appear from the
> specification that it is possible to change the file name and
> communicator associated with a file handle. I doubt this would be
> desirable or possible. I know that some other facility will be replacing
> File_control, but I thought I would still make the comment.
Hopefully, we won't have MPI_FILE_CONTROL around much longer...
John M May <johnmay@coral.llnl.gov> wrote:
> The MPI_View function's description of the parameters has a confusing
> sentence at lines 42-43 on page 237: "Offsets into the file are
> measured in units of etype." I believe this should say "offsets into
> the view," since the displacement parameter specifies absolute
> locations in the file, and it's measured in bytes.
Changed.
John M May <johnmay@coral.llnl.gov> also wrote:
> Also, this section should describe what happens to file pointers when a
> view is changed (they are reset to zero, right?).
Yes; a note stating that the file pointers are reset to zero
has been added.
John M May <johnmay@coral.llnl.gov> also wrote:
> The introduction to shared file pointers needs to be updated to
> reflect the use of views -- it's erroneous to use a shared file
> pointer if the views are not identical.
Text has been added to both MPI_VIEW and the shared file pointer
introduction to clarify this.
- bill
Bill Nitzberg nitzberg@nas.nasa.gov
NAS Parallel Systems, MRJ, Inc.
NASA Ames Research Center, M/S 258-6 Tel: (415) 604-4513
Moffett Field, CA 94035-1000 FAX: (415) 966-8669