#166 wsclean multiscale stuck in subminor loop on MeerKAT MS

[Oleg Smirnov]

Symptoms are a stuck iteration counter and peak value, e.g.:

Iteration 70346, scale 10 px : -150.03 µJy at 4994,5066
Iteration 70346, scale 10 px : -150.03 µJy at 4994,5066
Iteration 70346, scale 10 px : -150.03 µJy at 4994,5066
Iteration 70346, scale 10 px : -150.03 µJy at 4994,5066
Iteration 70346, scale 10 px : -150.03 µJy at 4994,5066

As commented by @anoko in an e-mail: "It means that the multiscale algorithm thinks it should continue iterating, but when calling the 'sub loop optimization' procedure as described in Offringa & Smirnov, that procedure doesn't find a single suitable component. Not sure why this would happen, and not sure how cleaning should proceed in this case -- I think the best I could do is detect it and stop the current major iteration."

I agree with the latter proposal. Print an appropriate message and stop the major loop. This way at least we're guaranteed to exit when the -nmiter limit is reached. If the user has set that to 0, too bad for them...

[Oleg Smirnov]

P.S. The problem does not arise with 2.6, interestingly.

[André Offringa]

This turned out to be an issue caused when not providing a zero scale (which should be allowed). This is fixed in 2.10.