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.

Update test suite for BGQ emulation. Some command output has changed
slightly. srun command only launches one task (runjob) now.

We only launch one user task on an emulated bluegene system, so modify test
to only check for the currently expected file(s) for one task.

Due to a race condition, a job may be cancelled before the launch completes.
In that case, an error message may be logged by slurmctld. This change makes
that condition be logged using info() rather than error().

One new file was not packaged in the RPM and compiler reported that one
variable could be used without being initialized and another variable
was never used.

third patch adds the man page and an example.
0003_bull_cgroup_devices_doc_add_allowed_devices_support-2.3.0-0.pre7.patch
Patch from Yiannis Georgiou, Bull.

adds the support of a file to declare the default allowed devices for
all the jobs.
0002_bull_cgroup_devices_add_allowed_devices_support-2.3.0-0.pre7.patch
Patch from Yiannis.Georgiou, Bull.

bug correction that I found when using sbatch,
0001_bull_cgroup_devices_correct_memory_leak_with_sbatch-2.3.0-0.pre7.patch
Patch from Yiannis.Georgiou, Bull.

code base by setting the SLURM_LOCAL_GLOBALS_FILE env var.  This way if you
are running off a git repo and want to test if your changes work on
multiple architectures you only have to set the env var to point at a
different install/build instead of having to check things out again.

code base by setting the SLURM_LOCAL_GLOBALS_FILE env var.  This way if you
are running off a git repo and want to test if your changes work on
multiple architectures you only have to set the env var to point at a
different install/build instead of having to check things out again.

BlueGene: Permit users to specify a separate connection type for each
dimension (e.g. "--conn-type=torus,mesh,torus").

If the user does not set the job name for srun, then set its name to the
executable files name rather than having it always be "runjob".

When reading an old state file, if the old coordinates would be out of bounds in
the new array, return a NULL pointer rather than going off the end of the array
and getting a segv.

If unpacking BGQ state information and a bad coordinate is read, then
return a NULL pointer rather than going off the end of an array and
causing an invalid memory error.

The latest version of man2html used to make web pages from man pages
produces slightly different output. This patch from Rod Schultz (Bull)
will work with the latest version of man2html plus older versions.

This reverts commit 48f468ea7136d27bee8c598eb58c0390be7f9ca8.

Different versions of man2html generated slightly different output. This change
supports both known varieties of output.