Previous logic was not properly setting env vars for pack jobs. In
  some cases the env vars were being set in slurmctld and NOT only
  if the parameter was set on sbatch command line.

When the "scontrol update jobid=#" command specifies a pack job
leader, then modify all components of the pack job. To modiify only
the pack job leader, specify "scontrol update jobid=#+0".

Modify scontrol job hold/release and update to operate with heterogeneous
      job id specification (e.g. "scontrol hold 123+4").

If scancel is called only with job filter options (e.g. --user=..)
  and that results in an attempt to cancel a pending pack job component
  then do not log the ESLURM_NOT_PACK_WHOLE error code.

Prior logic would report "No error" rather than "already complete"

Doing so would break the current scheduling logic.

Coverity CID 44817

Coverity CID 44731

Coverity CID 45279, 44884

Coverity CID 44976

Coverity CID 45198

Coverity CID 44820

Coverity CID 45161

Coverity CID 45166

Coverity CID 44847

Properly set up debugger symbols if an srun command tries to
  launch multiple, different applications in different components
  of a heterogeneous job (i.e. a different application in each
  pack job component).

Modify srun logic so each pack job component can have a different
  executable program and arguments.

This is needed to run a single MPI_COMM_WORLD spread across multiple
  job allocations. The work is incomplete, but well underway in this
  commit (much of the code in place, test suite runs sucessfully).

If the system configuration prevents the tests job from starting,
  terminate the test with a warning.

If submit failed, tried to cancel job ID 0

Systems with less than 3 CPUs per node would fail the test

Logic recently introduced would cancel an entire job allocation if
  a step launch failed, even if the srun command did not create the
  allocation (running under salloc or sbatch).

The srun message changed for pack jobs and the test failed.
  Tweak both message and test for better clarity.

Modify the signal handling and termination functions to operate on
  a list of job steps (pack job) rather than a single record
Add job ID to some srun log messages for pack jobs

Add the pack count argument to slurm_step_launch() function in
  test suite and perl api. The argument was added in commit
  71a34f56

Modify launch/slurm plugin to signal all components of a pack job rather
   than just the one (modify to use a list of step context records).

If prolog is running when attempting to signal a step, then return EAGAIN
   and retry rather than simply returning SLURM_ERROR and aborting.