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.

color wasn't picked correctly, it was set correctly later, but this fixes
it so it is set correctly from the beginning.

Add the ability for a user to limit the number of leaf switches in a job's
allocation using the --switch option of salloc, sbatch and srun. There is
also a new SchedulerParameters value of max_switch_wait, which a SLURM
administrator can used to set a maximum job delay and prevent a user job
from blocking lower priority jobs for too long. Based on work by Rod
Schultz, Bull.

Modify test2.8 to prevent a problem with "expect" that fails to
process all of the information in a job output buffer if that
buffer is large (hundreds of lines).

If a user job includes a maximum node specification and a node list, report an
error if the number of nodes in that node list exceeds the job's maximum node
specification. This addresses Cray bug 774402.

This fixes the parsing in the srun wrapper for aprun (Cray systems)
when arguments on the command line are quoted. Fixes Cray bugs
774536 and 774544

Note that multiple partitions that overlap and gang schedule are
scheduled independently and that is not generally recommended.

Update job submit man pages to describe job's --shared/exclusive option
interaction partition Shared option

Major update to Cray web page based upon feedback from Cray.

If salloc is interupted by a SIGINT then set the job's start time
to now. Otherwise the original start time will be preserved. That
original start time could be in the future due to the backfill scheduler
setting its expected start time.

Do not attempt to change a job's state in the database until after we
are sure that it has been written to the database from slurmctld.

This patch can prevent an invalid memory references by slurmd daemon if a
node's configuration in slurm.conf differs from the hardware configuration.

If a node's configuration differs from the actual hardware configureation,
an internal bitmap may be referenced with an invalid index causing slurmd
to abort without this patch.

If system is configured with a default memory limit, that applies to both the job
and job step which can prevent the expected number of job steps from being started.
Two tests were modified to explicitly set a job step memory limit to start more steps.

This fixes a bug in parsing slurm.conf for node information if there
are more than one NodeName=DEFAULT value. This adds to existing default
values rather than clearing old default values that are not explicitly
set to new values on that configuration line.

If a node has fewer CPUs than configured and task/affinity is configured a
reference off of the end of a bitmap may result without this patch.

If hostname input had suffix with leading zeros then the suffix returned
had leading zeros, which performed octal arithmetic causing at least
test1.83 to fail in some cases.

Fix logic in how each node's  HostAddr and HostNodeName fields are
set on a BlueGene system. When values are set on the NodeName line
of slurm.conf, only the first node's  HostAddr and HostNodeName fields
were being set, others were NULL.