[daip] Re: NOT a TVFLG bug -- cal. pos'n changes between fast-switching scans

[Ken Sowinski]

This certainly looks like it is my fault and I
think that I can offer an explanation.  

The explanation hinges on the precision of floats
and requires that scans with different program
sources use the same calibrator and that the 
calibrator be the secondary (given on the //OF card)
source.  When the scan is initialized, the two positions, 
represented as doubles, are differenced and the result 
saved as a float.  Later while nodding, the secondary
position(double) is computed as 
primary-position(double) - difference(float).
The precision of a float is such that I expect errors
of a few milli-arc-sec if the difference is a few degrees.

When I analyzed this ages ago I reasoned that errors of that
magnitude are irrelevant and are repeatable.  However I
neglected to consider the case of different program source
positions.

Is it a reasonable solution to ask AIPS to put a little fuzz
in the source position comparison?  The real solution is to
represent the difference internally as a double rahter than 
float.  That however means recompiling most programs in the
Modcomps.  That could be done, but is tedious and I worry
about any omissions.  Shall we talk about the possibilities
this week?

Ken