Can now only be used to launch steps in an existing job allocation;
it can no longer be used by SlurmUser to create a new job allocation
with a given jobid value.

Bug 6691.

Bug 6691.

If running within salloc or a job step, both SLURM_JOB_ID and SLURM_JOBID
would be set to the same value, and SLURM_JOB_ID would overwrite SLURM_JOBID
as it is parsed second.

The only time this would cause a difference in behavior is if SLURM_JOBID
were set, and SLURM_JOB_ID were not. Which is not something that can be
caused by Slurm itself.

This configuration previously caused test15.10 to fail
(but only when run as part of full regression test due
to NHC running on previous jobs).

Non-operators are not permitted to submit jobs with a negative
nice value. Rather than silently reset the value, reject it outright.

Without moving the function, it is undefined when referenced

Bug 6286

This reverts commit 082c59dd.

Bug 6286

Continuation of 6fc7298f

Bug 6286

The key and keylen are already cached inside uthash's
UT_hash_handle. So the memory is wasted.
Bug 6620

Hash functions no longer take the bucket as parameter.
Use of the HASH_FCN can be changed, since the
1,000,000 parameters is not needed. The hash is used for
comparison.
Bug 6620

Under some conditions the requested 4 tasks could not land 2 per
node. This change explicitly sets --ntasks-per-node=2 so it should
always work.

This fixes 24c04bce. Related commit c8c9694f.

Bug 3808
Bug 6681

without this change the task distribution logic errors out and
quite a few regression tests fail

Bug 6409.

Needs to use the acctg_freq argument, instead of snagging optarg
from the global environment. Since this function is exclusively used
inside a getopt_long() loop, this still managed to function correctly.

Correction to commit 72ff192a
test17.19 was failing consistently

I neglected to add a pre4 section after tagging pre3.

Because we skip select_p_select_nodeinfo_set(), we never needed to use
this data, keeping it can be confusing (e.g. backtraces).

Bug 6437

And SALLOC_WAIT as well.

Not serialized into RPC, so do not bother collecting it.

Change it so that we can resize arrays conveniently. Renaming
similarly to slurm_xcalloc().

This name is something of my own invention - there is no direct libc
equivalent that will resize allocations and ensure the contents are
zero'd out, let alone for arrays, and reallocarray() itself is
already non-standard.

Add an extra bool argument so we can introduce try_xcalloc() while here.

EMUNGE_CRED_REPLAYED is expected, and we should still mark the credential
as verified.

The bug here is that g_slurm_auth_verify was still returning SLURM_SUCCESS,
while the credential internally had verify = false, resulting in
g_slurm_auth_get_*() hitting an assertion failure.

Bug 6676.