fe_multi_stream_test can now write out individual models mbufs (the
rmIpcStr structure) as well as a combined daq_multi_cycle_t structure.

Remove fe_stream_test, rework the zmq/local_dc tests to use
fe_multi_stream.

The channel mix wasn't properly generating mid data rate range channels.

The test is to:

* start a simulated IFO (via fe_multi_stream_test)
* start daqd
* read live data from nds
* verify the data

This updates the NDS2 client detection, reimplements fe_stream_check_nds,
and adds some test harness logic.

fe_stream_test is a single front end outputing a rmIpcStr to mbuf.

fe_multi_stream_test can be used to simulate fe systems in bulk, it
outputs to a dac_multi_cycle_t in mbuf, along with populating a
ini directory and master file.  This is targeted for daqd development
and testing.

This adds a fe_generator_support library which holds some common code
between the two generators.

This also adds another check to fe_stream_check, verifying the data crc.

Fix usage message in local_dc.

Remove commented out code in the makefile and associated .c code

Looking for variables with ix_ prefix and removed an usused bit.

* Renamed the cmake target ix_xmit to dix_xmit.  Copied in code from mx2ix.c, removed the
mx specific bits.
* Reformatted local_dc.c, explicitly checking for 64bit alignment in tests
* fe_stream_check requires that the dataCycleSize is 8 byte aligned.
* Adding a simple test program to the ix_stream area.

ix_test -g <num> -s - sending/server side test
ix_test -g <num> -c - recieving/client side test
ix_test -g <num> [-c] -d - recieving side, dump first 64 bytes of data

* dix_xmit waits for the input buffer to be valid before checking buffers.

If you read the input mbuf before it is ready you will likely have invalid
input and will make bad assumptions about buffer lengths and data strides.

So wait until data is ready and then start up.

* Moved ix_fb_rcv.c to dix_recv.c.  Ran clang-format on the code.  Reviewed
the program arguments.

Primary change is to load TP/EXC via for loop instead of memcpy, as memcpy caused kernel panics with Debian 10 and gcc 8.3.
Also, I beutified the code using clang-format, which accounts for the multitude of diffs.

* Rewrote zmq_xmit.c to no longer do data concentration, it now depends
on local_dc.
* Removed some old test code that was not being used
* Added a new test (hooked into ctest) to run several simulated FE
computers, local_dc, zmq_xmit, and zmq_recv and then read the output
from the zmq_recv output buffer and verify each channel.

It was commented out so didn't impact anything.  However it should
not have been put in.

* Remove unneeded dolphin dependency
* Remove unneeded makefile target
* Put correct names in omx_recv.c
* Removed dead code ifdef #0'ed away.
* Update to help messages.
* Removed the dc restart code as it wasn't doing anything.

Rolf took the mx2ix code, pulled the dolphin code out to make mx2rcv.
This change moves the code to the mx_stream directory and renames
the file to omx_recv.c.

Rolf Bork authored 5 years ago and Jonathan Hanks committed 5 years ago

cherry-picked from rcgtest branch

When resyncing to the IOP, reset the counter for all models.
This helps mx_fe skip models that have not been started (ie
their mbuf is unitialized).

Corrected some index values.  The ixDataBlock dcuheaders where
indexed with the wrong value, and could cause a bad tpBlockSize
to be caculated if a required model was stopped (this happens when
the ii index != the db index) which can manifest itself with the
process aborting after seeing an overflow.

inconsistent tab/space indentation errors when run.

Missing return values and return types keep daq from building on debian 10.

This tool can scan a rmIpcStr based mbuf continually and print some
information out.  It was written to help debug fe or fe_stream_test
issues.

Fix a mixup of index variables.  TP size was be calculated with the
wrong models (possibily uninitialized memory) data.

This handles the case where a model is not running at startup and we
create an mbuf for it.  The contents are uninitialized and undefined.
Simply setting an impossible cycle counter for all models (not just the
IOP) protects us from processing uninitialized mbufs.  The buffers will
not be seen as ready until a model runs and sets a proper cycle counter.

Remove the requirement for -DDEV_BUILD, its tests can be used for any
cmake based build.

these are not executables

Local_dc reads RT model data from mbuf (rmIpcStr) concentrates the data
into one block and stores it in an output mbuf (daq_multi_cycle_data_t).

In a standalone system this would be the data concentrator.  On a larger
system, this prepares data for a streamer process.

This also adds a basic test for local_dc which starts several local
simulated FE streams, concentrates their data and checks that each
channel is available in the output with the expected values.

This removes fe_data_concentrator.
This updates code in fe_stream_test.
This adds a new fe_stream_check, renaming the old (and disabled) one
to fe_stream_check_nds.

This also enables cmake based testing.

!!Warning the tests fail unless mbuf and gpstime are enabled!!

Moved a bit of old zmq code to an 'attic' directory in the zmq area.

Ripped out the dolphin bits from zmq_recv_ix_xmit_delay.cc and renamed
it zmq_stream/zmq_recv.cc.

Moved ix_stream/recvr_utils.hh to zmq_stream as it was some debug code
for zmq_recv_ix_xmit.

ix_fb_rcv -> ix_recv
zmq_rcv_ix_xmit_delay -> zmq_recv
mx2ix -> omx_recv
ix_xmit - added as placeholder
local_dc - added as placeholder
mx_fe -> omx_xmit
zmq_fe -> zmq_xmit

No longer builder mx_stream.

"2to3 -w .", and changes to shebang lines

The following are now included in the install in a make install:

ix_fb_rcv
zmq_rcv_ix_xmit_delay
mx_fe
zmq_fe

Updated the FindEPICSBase.cmake to make full targets for ca, cas, Com, gdd.