Just in case something goes wrong with the test, this should clear
  the job relatively soon

The RESPONSE_JOB_PACK_ALLOCATION RPC returns a list of
  RESPONSE_RESOURCE_ALLOCATION data types. Error handling
  still needs to be added to slurmctld/agent.c, blocking
  for resources does not work yet, and the salloc code
  does not handle the List response type yet.

Expect seems to be sometimes truncating the output from "salloc --help"
  resulting in a test failure. This is a bug in Expect not Slurm. Modified
  the test to not parse all the way to the end of the expected output so
  the test always passes.

No change in logic

No change in logic yet, but added a function to determine if there
  are pending pack job components and moved the data structure
  build logic to a separate function so it can be called for
  each pack job component

Use syntax "scancel #+#". For example "scancel 1+2" cancels component
 number 2 from job 1

slurmctld modified to return all components of a pack job if the
  pack job leader job ID is specified
Modify squeue to properly handle job ID filtering for pack jobs

pack_job_id_set shows all job IDs associated with a pack job.
Logic added to include in show job RPC and display in scontrol.
Logic added to save/restore the field.
Logic still needed to validate on slurmctld restart and rebuild
  pack_job_list list of job pointers.

Also sort job ID's considering pack job ID and offset

Add pack_job_id and pack_job_offset fields to job record in slurmctld
  plus job_info structure
Add logic to save/restore pack_job_id and pack_job_offset fields on
  slurmctld restart
Add logging of pack_job_id and pack_job_offset fields in slurmctld's
  jobid2fmt() and jobid2str() functions
Add pack_job_id and pack_job_offset fields to "scontrol show job"
  output

cancel vestigial pack jobs when an error happens in some portion of
  the job request

Add a new RPC, REQUEST_JOB_PACK_ALLOCATION, that will send a List of
  job request descriptors in a single message. Each job descriptor
  will be logged and an job record created for it.

Refactor the salloc option parsing logic to build a separate set of
  job descriptor data structures for each portion of the heterogeneous
  job. Only the last job record is current submitted to slurmctld.

Bug 3695

bug 3779

If capmc reports a node name, but not mcdram_cfg for the node, then
  log the missing data rather than assume the value is zero and
  report a value mismatch with cnselect.

When requesting an operation on jobs, where the operation permits to specify
more than one job in the same request, and a job array appears before a
regular job (no-array job) in the list of jobs to operate with, the
job_array_resp_msg_t pointer was not properly NULL'ed and thus incorrectly
accessed when processing the no-array job. This fix prevents the crash from
happening in the following scontrol operations:

uhold, hold, suspend, requeue, requeuehold, update, release

when the same request has <array_jobid>,<non-array_jobid> in this order in
the job list to process.

Bug 3759

Job expansion example in FAQ enhanced to demonstrate operation in
    heterogeneous environments.
bug 2979

Do not attempt to schedule jobs after changing the power cap if there are
    already many active threads.

# Conflicts:
#	META
#	NEWS

Bug 3760

didn't work at all.

Bug 3712.

This reverts commit ecfd007f.

Continuation of 9a1370e3
CID 168995

It was noticed that while doing any update to a job the admin comment would
be blown away.  This patch fixes that.

Bug 3789

run multiple tasks on multiple nodes.  Changing the max nodes setting from
3 to 6 fixes the issue without apparent compromise to the test.

destroying a mutex.

When running sacct from a federated client, the db returns jobs for each
cluster with duplicate jobs removed on each cluster. A federated job could have
ran on a different cluster when the before the jobid's rolled. This patch
filters out past old federated jobs and leaves the newest ones.

Reverted d31965 which was too slow.