[bananas] 31DEC07 Patch

[Eric Greisen]

An error in VISDFT has been discovered.  This subroutine does DFT model 
subtraction and division for tasks like IMAGR, CALIB, UVSUB, et al. 
When it was asked to do multiple facets, it looked at the first facet's 
Clean Component file to see what type of model to use (point, Gaussian, 
or Sphere).  Thereafter, it silently ignored any CCs that were not of 
that type.  Thus, in multi-scale Clean, a restart in which point and 
Gaussian Clean Components should have been subtracted, had the Gaussian 
components quietly ignored.  This should have been noticeable in the 
initial images or in the fit gains from CALIB etc. and no one has 
reported the problem.  Nonetheless, it is a pretty serious oversight.
The fix involves making points into 0-size Gaussians when needed to make 
all models Gaussians (or spheres) and also reporting any CCs that are to 
be ignored.  That will happen only if Gaussians and Spheres are mixed 
into one call to VISDFT.

The patch is available in the old way from the web - start with
        http://www.aips.nrao.edu/
Alternatively, you may now run a MNJ in 31DEC07 to update your 31DEC07 
version.  The fix will be in 31DEC08 after 10pm Mountain Daylight time 
tonight (5 hours or so from now).

Eric Greisen