The re-use of error_code results in the function returning an error
  when it really functions properly. Added new variable to use.

Use part_is_visible instead, which does not modify the partitions
(and thus require the part write lock).

No longer require part write lock.

They now only need the part read lock, not the write lock.

Only needs part read lock, whereas part_filter_set/clear require
the part write lock which limits concurrency.

Convert one location in pack_all_part() while here.

Also remove long-hidden test program from parse_time.c.

Fix one location in sacctmgr that was altering the value to use a
heap-allocated value instead.

Previous implementation is not thread-safe.

Replace with direct references to the struct.

Improve backfill scheduling algorithm with respect to starting jobs as soon
    as possible while avoiding advanced reservations.
bug 3757

bug 3886

Not necessarily fatal(), but of potential interest when debugging
odd slurmctld crashes. Cannot go where the limit is originally set,
as the logging infrastructure is not avaiable at that point.

Bug 3886.

test1.91 fails with non-default binding

Bug 3581.

Fix for commit 250378c2
  test7.3 was failing without this patch
  bug 3502

test 17.12 was leaving slurm-#.out files around. Explicitly set
  output file to /dev/null and set time limit to 1 minute to avoid
  vestigial jobs.

Handle syncing between siblings (not the origin). Job could have been
canclled or started while the origin was down. The sibling should see
this and remove it's copy if the job is running or cancelled on the
remote cluster.

Handle case where job was cancelled or finished while the origin was
down and the siblings were up.

The controller will keeps job in the job_list until the origin comes
back up and will find out about it then.

This allows the origin to be able to sync up jobs after it has been
down.