To address "The --help output needs to be made (much) better" I'd like to suggest that --help and -h each give the same BRIEF summary of under 24-lines of 80-column text. Additionally implement --help=[topic] to give more info on related groups of options. Of course, -h and --help would list the valid "topic" values in their brief output.

Just my USD 0.02
-Paul

P.S. arguments is misspelled in the subject line :-)

On 4/21/2011 3:52 PM, Ralph Castain wrote:
Might help if you were to list those CLI params you believe to be removable 
and/or those you propose to deprecate.

Just a quick glance: I don't see any options that are defunct, though we could 
argue about how many people use them. So perhaps what we discussed - having a 
short response for -h that lists the most commonly used ones, and a longer 
response to --help that contains them all - might be more appropriate?


On Apr 21, 2011, at 9:51 AM, Jeff Squyres wrote:

WHAT: Deprecate a bunch of old mpirun CLI parameters, remove *most*
      "single dash" long mpirun options, make "mpirun --help" much
      more user friendly

WHY: `mpirun --help` is currently 232 lines of output.  *Ouch*
     Additionally, the Josh/Terry/Jeff affinity re-work will end up
     creating new mpirun CLI options and deprecating the old ones.

WHEN: Maybe late in the 1.5 series (the new affinity stuff is
      tentatively scheduled for late in the 1.5 series).  1.7 for
      sure.

WHERE: Mostly in orte/tools/orterun/

TIMEOUT: ORNL face-to-face OMPI meeting (May 3)

-----

MORE DETAILS:

We simply have too many options to mpirun.

- Some could be removed
- Some should be removed
- Some should be deprecated (but still available)
- The --help output needs to be made (much) better

The new mapping/affinity options effectively replace a bunch of the
old options.  The old mapping/affinity options should be deprecated in
favor of the new stuff.

Additionally, there are at least a few old orterun options that are
either only of interest to developers and/or have an MCA parameter
backing them (and are fairly uncommonly used such that the MCA param
could be used instead).

Finally, we have many options that are available via both "-foo" and
"--foo" (mostly for hysterical raisins).  In most cases, the
single-dash version should be removed -- per GNU conventions, long
names should only be available via double-dash.  There are a small
number of single-dash options that must be retained, however, for
compatibility with other MPI implementations and for mpiexec options
specifically listed in MPI-2.2:

    -np, -host, -hostfile, -machinefile, -wd, -wdir, -path

I propose:

1. Remove all other single-dash long name options.

2. Make all deprecated options only available if the user
   specifies --deprecated-options on the command line (or invokes mpirun
   via mpirun.deprecated-options).
   -->  This allows users to keep their existing scripts that use
       deprecated options, but with a glaring signal that "hey, this
       option you're using may disappear in a future version!"

3. Revamp the --help output to show a short listing of the most common
   options, and a note that a) the mpirun(1) man page offers much more
   detail, and b) --help-all shows the original, exhaustive CLI option
   listing.

Extra bonus points would be given for anyone who'd like to implement
an svn/hg-like "help" command, perhaps something like:

  mpirun help machinefile
  ...help output specifically about the machinefile option...

I don't have time to do this last part, but it would be a great
usability feature.

--
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/

_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel

_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel

--
Paul H. Hargrove                          phhargr...@lbl.gov
Future Technologies Group
HPC Research Department                   Tel: +1-510-495-2352
Lawrence Berkeley National Laboratory     Fax: +1-510-486-6900

Reply via email to