else is figured out.  Seemed cleaner.

instead of a fixed number when dealing with Usage.

src/common/slurm_jobacct_gather.h is needed for CPU_TIME_ADJ

they are used almost everywhere.

This reverts commit 78699d7b.

Bug 4702.

Remove partition time limits, run the test, then restore
partition time limits with reconfigure.

Bug 4276.

Remove partition MaxTime limit at the beginning of the test,
run the rest of the test, then restore the partition configuration
with scontrol reconfigure.

Bug 4915.

Bug 4915.

Bug 4915.

Bug 4926.

Bug 3567.

Missed from 500e23b5

Bug 4988

job_ptr was still being referenced after releasing the read locks. Also
prolog_running_decr() needs to have a job write lock and a fed read
lock.

Bug 4947

Avoids confusion with the %u header of USER.

Bug 4948.

Bug 4984.

Bug 4990.

When a capmc program is stuck a node can remain in IDLE*+POWER state
despite the admin manually rebooted and cleaned it. The status can be
changed then by issuing scontrol update ... state=power_down.
Added this example in docs.

Bug 4788

Bug 4411

While restarting slurmctld with -R flag, features were emptied.
This patch just does nothing when recover=2 (-R) which means active and
available features are preserved from the state files. Needed for
backup slurmctld controller.

Bug 4927

global variables for sizes instead of calculating the same size over
and over again.

Bug 4882

When setting environment variables in jobs we were tracking other job's
indexes and information, while the scope of the environment is totally
different for each job. The static variables have been moved to non static.

This fixes CUDA_VISIBLE_DEVICES indexing while running different jobs on
the same node.

Bug 4851

If block_map_size is zero (as seen on FreeBSD when not using HWLOC),
xmalloc(0) will return NULL. Writing to str[0] then is invalid.

Bug 4845.