Explicitly exclude the few libraries that are packed elsewhere.
Otherwise grab all libraries at once rather than carry around
this huge list of libraries that may potentially be of interest.

This also makes long-term maintenance much simpler.

MUNGE is the only support auth plugin, so always require it.

Also, always require python at build (used for docs currently).

It is unlikely this version of the spec file would
work on such a machine due to rhel7/suse12 requirements.
If needed, these should just be rolled into the normal
packages rather than being standalone.

Running with auth/none is a terrible idea, and I
refuse to make a package to let you do it.

accounting_storage_mysql is only used by slurmdbd,
jobcomp_mysql is only used by slurmd

sgijob is only kept around in slurm for cray/alps support,
so drop both at the same time.

Remove db2 configure option.
Add select_bluegene.so to file deletion list.

Set the package name, release, fix the license field, and
correct the source package name.

Switch to using %{buildroot} as well
Do not install the init scripts, always use systemd service files.

Preserve an older copy of slurm.spec. The shipped version
will be overhauled, and targeted at RHEL 7+ / SuSE 12+ only
and assume systemd is being used.

This copy is deprecated, not maintained, and will be removed
in a future release.

While relocating, update the header block with the package name,
correct the license description, and set the release.

Important note: do not be tempted to rename this to slurm-legacy.spec.
Doing so will confuse rpmbuild, as it attempts to build everything in
the archive file with a .spec extension on it and this will immediately
conflict with the shipped version of slurm.spec.

Start of work for bug 3396.

Continuation of 3c97b7d7

Bug 4252

Bug 4252

CID 177939.

when dealing with this scenario.

If any component of a pack job launch fails, then fail all components

If any node in any component of a job fails, make all components of the
  job fail

only the fields we care about from the database instead of always getting
everything.

The function should probably be a extern so we can use it in other packing.
But that will wait for some future day.

Requires a bit of shuttling around to pass the sbcast_cred_arg_t
to where it's needed. Adds additional checks that someone hasn't
managed to steal an sbcast credential somehow by comparing uid/gid
between auth header and the cred.

Remove some stray xfree(gids) calls that cause a double-free; before
the gids were fetched only for this call path, but now they're
handled as part of the cred.

It's not clear how this would ever happen - there is no
'sbcast --no-alloc', and 'srun --no-alloc --bcast' doesn't work.

Don't keep looking them up on each step / sbcast cred creation.
If they're not there, then slurmd will fill them in for us.

(Same technique used on user_name field.)

Much tidier. Should have done that many commits ago.

I'm tired of doing this manually.