make it to the slurmctld when using message aggregation.

make sure we handle it correctly when the database comes back up.

ours in from the config.h which is now included.

Give test job an extra second to start. Test was failing by one
  second on kachina.

The Perl API didn't define HAVE_CONFIG, which prevented headers
from include config.h which has a different definition of the VERSION
macro than the Perl API build system. Add these two #undef's to
prevent build error and work around the namespace issue.

Get the maximum file pathname size from system include file rather
  than local #define. This was causing failures on kachina test.

strcasestr is a non-standard glibc extension.

Remove unused includes, and aliases from slurm_xlator.

Also remove local version of inet_pton/ntop functions hiding
behind HAVE_INET_NTOP/PTON. Those are both in POSIX1.2001.

Delete files, remove from Makefile.
Remove stray includes as well to keep build happy.

If HAVE_STRNDUP is undefined, please fix your compiler.
Replacing strndup with an unbounded strndup is a bad idea,
and you should be using xstrndup anyways.

Will need to audit for <stdbool.h> inclusion later, currently
this is indirect sourced through inclusion of "src/common/macros.h".

These functions are already out of sync with their data-structures;
removing lest anyone assume they'll be correct when uncommented.

Underlying xmalloc will log and abort for us on OOM, none of these
checks would ever fire. Remove custom log macros and tidy up.

Since the glibc-portion in the #ifdef above uses malloc, malloc here instead
to keep the same free() call below.

strsignal() is POSIX1.2008, and we already relied on <string.h> to provide
the correct definition in plugins/launch/slurm/launch_slurm.c .

O_CLOEXEC is POSIX1.2008, so the macro test and fallback code are no longer
necessary. Each functions was only used once, so replace with equivalent calls
to open().

This has never been used since addition a decade ago.

Automatically found by looking for files relying on macro definitions from
config.h.

Ugly one-liners used to automate this are:

tim@zoidberg:~/slurm$ for def in $(grep undef config.h.in |cut -f 2 -d ' ' );
do for f in $(grep -r $def src/ |cut -f 1 -d : | sort -u);
do if [ $(grep -q "#include \"config.h\"" $f; echo $? ) -ne 0 ];
then echo $def $f ; fi ; done ; done  > missing-includes
tim@zoidberg:~/slurm$ for aaa in $(grep -v \.in missing-includes |
grep -v \.am |grep SLURM_VERSION|grep -v malloc |cut -f 2 -d ' '|sort -u );
do vim $aaa ; done
tim@zoidberg:~/slurm$ for aaa in $(grep -v \.in missing-includes |
grep -v \.am |grep -v VERSION|grep -v malloc |cut -f 2 -d ' '|sort -u );
do vim $aaa ; done

abort() call removed back in commit 94384552.

It turns out that autogen.sh is just going to put these right back in.

This reverts commit cb37e64864b706b7c14851d03e509ef237db1813.

config.h.in: <inttypes.h> required and includes PRIu64.
smd_ns.h: remove local bool redefinition. Always pick stdbool.h. Alphabetize.
testsuite: cleanup includes as well while here.

Goodbye to:
HAVE_DIRENT_H HAVE_ERRNO_H HAVE_FLOAT_H HAVE_HSTRERROR HAVE_LIMITS_H
HAVE_MCHECK_H HAVE_MTRACE HAVE_NETDB_H HAVE_PTHREAD_H
HAVE_SECURITY_PAM_APPL_H HAVE_SOCKET_H HAVE_STRSIGNAL HAVE_SYSINT_H
HAVE_SYS_IPC_H HAVE_SYS_SEM_H HAVE_SYS_SHM_H HAVE_VALUES_H

<syslog.h> already in use throughout Slurm.

Also remove #include "config.h" where no longer required.

C99 says this keyword exists, stop hiding it behind a macro.

Continuation on previous patch, but for files that require review.

Most of these were just using a local #if _DEBUG and/or protecting against
multiple inclusion with #ifndef _PROTECT_H; #define _PROTECT_H.

Some also have __APPLE__ or __FreeBSD__ blocks, but those are not in config.h.

Found semi-automatically with:
for a in $( for i in $(grep -r \"config.h\"|cut -f 1 -d : ) ; \
do echo -n "$i "; grep '#if' $i | wc -l  ; done \
|grep \ 0 |cut -f 1 -d ' ' ) ; do vim $a ; done