Minor compile bugs, and LAM update

[Pete Wyckoff]

ben at bellatrix.pcl.ox.ac.uk said:
> 1. If your PBS headers are installed in /usr/include, the mpiexec
>    Makefile adds -I/usr/include to a gcc invocation. This is a bit dodgy
>    as it alters the search path (if gcc has modified any system headers,
>    and put them in gcc-specific directories, this use of -I will cause
>    the old, buggy, headers to be used instead). gcc 3.1 complains about
>    it, anyway. I use the attached patch to work around this.
> 
> 2. The LAM patch is out of date. An updated version is available at
>    http://bellatrix.pcl.ox.ac.uk/~ben/pbs/lam-mpiexec-tm.patch. (The
>    version included with mpiexec won't, I think, work at all, due to a
>    bug in my use of waitpid() for hboot's child processes.) I would have
>    noticed this earlier except that there appears to be a problem with the
>    mpiexec CVS server. Whenever I do a "cvs -z3 update -dP" I get:-
> 
> in.cvsd server: Updating .
> in.cvsd server: Updating patch
> in.cvsd server: failed to create lock directory in repository
> `/cvs/mpiexec/patch': Permission denied
> in.cvsd server: failed to obtain dir lock in repository
> `/cvs/mpiexec/patch'
> in.cvsd [server aborted]: read lock failed - giving up
> 
> This doesn't look like a problem at my end...

Somehow all my checkins were getting owned by the wrong group.  I fixed
it, and tried putting g+s on the directories in the CVS repo on the
sunos box hoping that might keep the group perms correct.

Thanks for the other changes.  I'll bump the version number again
soon now since LAM was broken.

		-- Pete

P.S.  Version 0.68 is available now.  No new functionality over 0.67,
which in turn has nothing new over 0.66 unless you want to use a
compiler different than gcc.