[daip] CALIB: problem with TABIO?

[Eric Greisen]

Dmitrijs Docenko writes:

 > Day numbers are indeed 308-311 (and not 1308 or similar), as
 > the project included measurements in all four VLA configurations
 > and was carried out during one year.
 > 
 > It seems strange that the date might be the cause of the problem,
 > as other data (e.g., with day number 197) are calibrated without
 > such error message.

Generally, people reduce the configurations individually - echa with
its own reference date.  It is not necessary to set REDATE so long as
you load only one configuration at a time in FILLM.  The problem is
related to an old, but still meaningful choice of format.  If UV data
were stored in tables, then the TIME column could be double precision.
But tables were invented long after UV data were in use - so the only
practical way to have double precision times is to have double
precision everything.  With data sets now approaching 20 Gbytes that
would be a disaster.  In testing I notice that the real time for
processing an uncompressed data set is frequently longer, although the
cpu time is less, tan for an umcompressed data set.  Getting the
better weighting between IFs still jsutifies uncompressed in most
cases. 

I suspect we have hit some sort of computational problem with times.
One user with data every 0.4 sec, trying to process one record at a
time, put CALIB in an infinite loop since time + 0.4 = time inside the
computer in single precision.  This problem is probably somewhat
similar...

Eric Greisen