failing UVW decomposion in 3C286.BP.ms

[o-smirnov]

Ask @bennahugo for a copy. I think the subset of baselines is particularly perverted, and defeats your current two-pass UVW decomposition logic.

I think I also have the solition.

[bennahugo]

hugo@stevie ~/polcal_dec2018                                                                                                                                     [15:53:32] 
> $ du -sh msdir/3C286.BP.ms                                                                                                                                               
24G     msdir/3C286.BP.ms

A fix for this exact issue is already implemented in the source providers in DDFacet. It is as simple as padding up the chunk to contain all baselines and then resimulating UV corrdinates into the padded arrays - similar to CASA fixvis @o-smirnov @sjperkins

[bennahugo]

See
https://github.com/cyriltasse/DDFacet/blob/master/DDFacet/Imager/ClassMontblancMachine.py#L146
and
https://github.com/cyriltasse/DDFacet/blob/master/DDFacet/Imager/ClassMontblancMachine.py#L291
through
https://github.com/cyriltasse/DDFacet/blob/master/DDFacet/Imager/ClassMontblancMachine.py#L376

[o-smirnov]

Mine is simpler though. :)

[bennahugo]

lol ok :) anyway this is not the first case that this is a problem. All split data suffers this affliction