[daip] imaging problem

[Eric Greisen]

I have, due to my current testing, released a rattlesnake and I was 
hoping you might have a suggestion.   I am attempting to make gridded 
model subtraction faster by reducing the number of passes through the uv 
data set.  I have code that appears very proficient at this and it 
produces binary equal results with older versions.  But, in one test, it 
did not.  The upshot was that the model facets were subtracted in a 
different order.  The actual test was IMAGR which even went into a 
slightly different path in the Clean after a while.  Isolating this to 
UVSUB where I can control the order of subtraction I find that the 
residual data sets agree in almost all points to better than 10 micro Jy 
(source is 7 Jy, residuals get as high as 1 Jy) but that there is a 
small group of samples - 24 or 36 or so that disagree by > 10 microJy 
with 7 - 9 disagreeing by > 1 mJy.  Looking at them, there does not 
appear to be some pattern (e.g. last sample in a buffer) within the data 
set or some pattern in the u-v values.  Rerunning UVSUB using an XY 
sorted version of the data produces a different collection of equally 
bad samples - different both in position within the data set and in 
which samples they are - even the number differs by 50%.

Note that I find this in 31DEC06 UVSUB (identically to my newer 
version).  Do you have any suggestions about where I might look for a 
possible actual error?

I hope you had fun in Oz.  I wish I could go bird in western Australia - 
I never got there.


Thanks,

Eric