Fix the error handling code to work with how nice has been implemented for a while.  Change to clearing errno and checking it after the call as proscribed by the man page.

The daqd has carried code for at least 9 years to set the GPS leap seconds in the frameH structure that is not used, too small, and the configs where never updated.

Remove the code.  Putting the correct leap second value into the FrameH structure is done by FrameCPP in the call to SetGTime.  There is no need to keep this old code around.

* For now turning the set gps-leaps into a no-op in the config.
* Removing the leap seconds field from the circular buffer
* Removing some member variables and a helper function from daqd.

these are code copies of an install script from another program that
are not needed afaict

Update the daq parser to consume a port number as well as a broadcast interface name.

* This introduces a address:port parsing call into the net_writer code to allow it to handle both the port information as well.
* Removes a minor memory leak, by moving from strdup w/o a free to a std::string.
* As we no longer transmit data via daqd data concentrators this removes the DC code path from the net_writer broadcast section.
* Also removes to unneeded constants.

Changing the defaults for shmem_input/size to good defaults so they do not need to be mentioned in the daqdrc.

This will detect a missing libfl-dev package on Debian.  Find it at configure time, not build time.

This check can be turned off by setting the parameter "allow_any_dcu" = "1" in the daqdrc.

This allows dcuid's in the channel list that are not in the TP group or the MYRICOM group.  (We may be able to restrict the TP group as well).

Also removed some old code that is specific to the internal edcu, which along with not being used anymore gives a more obscure error message.

This fits better with recent packaging changes and auto configuration of the device files.

Move to using the systemd RuntimeDirectory for creating the temporary
nds/daqd run directory in /run.  Use the new nds command line argument
to specify the path to this directory.  The daqd and nds code (and
example daqdrc file) are also updated to use this by default if not
specified.

This also removes the LD_BIND_NOW variable from the nds systemd unit,
which is of unknown use by anyone anymore, and which was cargo-culted
from the old launcher script that was apparently created before the
advligorts used source control.

 * --socketdir changed to --rundir
 * renamed internal field job_dir to run_dir

* The socket has been renamed from pipe to daqd_socket
* Defaults to a working dir of /var/run/nds
 * socket path is then /var/run/nds/daqd_socket
 * jobs dir is then /var/run/nds/jobs
* Set the --socketdir option on nds to specify the working dir

The zmq streamers where kept around as a off the shelf method to move data for small (test) systems.  However the pub/sub code appears to work better, and has no distribution limitations either.  So remove the zmq code in favor of the pub/sub.

Mostly replacing magic numbers with symbolic constants and expanding comments.

This is to help properly track lifetimes of the work queues.

Put in a wrapper around pthread_create to allow specifying an arbitrary job to run in the guise of a std::function.  This keeps the flexibility (scheduler, stack size, ...) of pthreads but gives the us the ability to call more than just a void* (*)(void*).

With these changes the work queues have been moved out of object scope for both the producer and the daqd objects.  A few small static routines that were used to start worker threads in the producer/daqd are removed.

Added some debugging output to the cycle continuity checks and adjusted code so that the glitch warnings are terminations for daqd.

Broke the code out into its own header for easier testing.  Updated the code to allow catching up if it falls behind, not just jumping to the front if there is a data gap.

Updated the work queue to use C++11 constructs, and to use compile time determined sizes.

Moved the dcu_move_address structure to the global scope and into the producer.hh file.

It now catches the processing time of all the received dcus.  So it is a superset of the crc time.

Removed a producer cycle measurement that was recently added as the epics reporting variant was fixed.  Updated the crc timing measurements.

Add the crc checks to the ctest batch of tests.

The checksum_crc object needs to have the add method called to import data.
The recv time was being written to the recv time pv and the full cycle time pv.
Updated a checksum test to make sure the new and old code produce the same checksum.

 * Reflect that this is typically on a TW instead of a FW
 * Clarify what the _SEC applies to, the write time for the raw minute trends

RAW_MTREND_FW_STATE -> RAW_MTREND_TW_STATE
RAW_MTREND_FW_SEC -> RAW_MTREND_TW_WRITE_SEC

* Increases the DAQ_TRANSIT_MAX_DCU to 256
* Updates the tests of daqd/nds to run with more dcus and sample more channels
* Fixes a spurious failure in the edcu test to keep tests passing

The FRAMECPP_VERSION_NUMBER is not properly defined in v2.7.0.  If it is empty, assume version 2.7.0.

Allow spurious failures.  Just require that we exceed the number of successful connections that used to be in a 32 bit bit mask.

This addresses the issue by expanding the size of the bitmask.
* Replace the 'int' bitmask with a generic structure.  This allows the
type system to help us find all the places it is used.
* Replace direct bitmask manipulation with member functions to increase
local readability.
* Allow the new bitmask structure to have an increased capacity that can
be set (at compile time).
* Add unit tests to new bitmask code.
* Add test code to look at nds1 load.

Use transport-independent RPC package (tirpc) if available,
otherwise use RPC headers if still available.
Mostly for non-Debian compiles, as Debian asserts non-default
flag in glibc to retain the old RPC support