add SLURM_GRES environment variable to srun

The information describing the select plugin APIs did not reflect
modifications made in SLURM version 2.3. Documentation for many
functions was added and other function descriptions were modified.

On Bluegene systems, smap's command-line mode would generate an invalid
memory reference due to an uninitialized variable.

In test8.8, set a time limit and explicitly spawn bash in order to
insure we don't just use SallocDefault

Modify test1.45 to work on BlueGene systems. Minimum job allocation size
is set equal to the smallest block size.

Remove the following select plugin functions: select_p_ba_reset(),
select_p_ba_request_apply() and select_p_ba_remove_block(). They are
no longer used.

will try to register when we are running from the slurmctld.

a POLLERR the dbd_fail callback is called.

the DBD where both remained up but were disconnected the slurmctld would
get registered again with the DBD.

Modify the error message if srun's task count exceeds the node count.
Log using info() rather than error() and explain how it it handled.

This adds another argument to the test using SLURM's internal API
assoc_mgr_init(), modified in commit 1b838865
on July 29, 2011

clarify job step use of gres in web pages.

With sched/wiki or sched/wiki2 (Maui or Moab scheduler), insure that a
requeued job's priority is reset to zero.

Set the absolute pathnames for the salloc and srun commands in the
srun wrapper for the aprun command (used only on Cray systems).
Fix for bug 774375.

all that aren't SLURM_SUCCESS (meaning no database connection)

comments that has quotes.

When pty terminal emulation is used (srun --pty), the user's pseudo
terminal has the potential to cause IO to block. Therefore, the code
must avoid connecting slurmstepd's STDERR_FILENO to user pty to avoid
a potential slurmstepd hang.
Patch from Chris Morrone, LLNL.

Creating unnecssary output is poor RPM behavior. Quiet the instructions
added in the post-install scriptlets of the slurm specfile.
Patch from Mark Grondona, LLNL.

Similar to the fix for the reason timestamp, the reason user
should also be compared when creating sinfo -R and sinfo -Rl output,
or else incorrect usernames could end up in output if everything
else matches.
Patch from Mark Grondona, LLNL.

sinfo does not check for differing reason timestamps when coalescing output
for sinfo -R and -Rl. This leads to incorrect timestamps in output if two
nodes are down with the same reason string but at a different time.

This fix adds a timestamp_reason_flag and a test to determine if
timestamps are equivalent before coalescing lines of output.

(Note: This patch is not an ideal fix for this issue. It would be much
better if the sinfo output handlers (_print_* in sinfo/print.c) could
also be called by sinfo to see if two lines of output were going
to be equivalent and could be coalesced. This would do away with the
separately kept "match_flags" which (as is seen here) are cumbersome
to maintain.)
Patch from Mark Grondona, LLNL.