[evlatests] idcaf update

[Walter Brisken]

The problem Rick reported on earlier today has been found.  It is in 
IDCAF: New integration times were being used, sort of, too early.

A fix has been committed.  I ran a test file that changes integration time 
and all looks as expected, no complaints from IDCAF.

I'm afraid this would have affected other projects that change 
integration times.  The symptoms would be as follows:

1. Only scans immediately before a new integration time are affected; the 
problem does not cross project boundaries.

2. If the new integration time (call it tNew) is longer than the old (call 
it tOld), then only one integration would be recorded within each interval 
of length tNew, so for that scan the fraction of data written should be 
tOld/tNew -- the remainder will not appear in the archive; the cadence of 
integrations would be tNew, even though they only had tOld actual 
integration time.  No alerts are generated in this case.

3. If the new integration time is shorter than the old, then every 
integration is still written and the operator should have gotten alerts.

4. Any data written to the archive should have the correct integration 
time reported, but what AIPS does with observing duty cycle less than 1 I 
don't know.

Reference pointing might be the most common case for such a problem -- 
usually 10 second integrations on the pointing source and < 10 on the 
science targets and calibrators.  Only the very last pointing position in 
the reference pointing would have been affected -- it would have reduced 
integration time and might have confused telcal.  It is quite possible 
though that good pointing solutions are determined.  The scan prior to the 
reference pointing scan would also have been affected -- fewer 
integrations would be written to disk.

Please contact me if you suspect a project might have been affected. 
Looking through the logs, I suspect there might have been a couple, 
including: AK679, AK680, AR642 (2 segments).  These are the only 3 
non-test projects since last Monday that I can see clearly have problems. 
The extent of the loss to each project may vary from inconsequential to 
fairly bad depending on the circumstatnces of the integration time 
change.

Thanks Rick and Emmanuel for reporting this!

-Walter