Merging empty channels with collect() and toList()

[glormph]

I am a happy user of the following pattern, which collects a number of process outputs and feeds them into a merge process:

process proc1 {
output:
tuple val(a), file(b), file(c) into chan1
...
}

chan1
  .toList() // create big list of all output ([a2, b2, c2], [a1, b1, c1], ...])
  .map { it.sort( { a, b -> a[0] <=> b[0] }) } // sort list on value a
  .transpose() // transpose list and output multiple items  [a1, a2, ...], [b2, b2, ], [c1, c2,  ]
  .toList() // one output again, for downstream merge process
  .set { merge_in }

This works great, until I ran into the problem where there is no output from proc1 (e.g. empty sample), which ideally would have to finish the pipeline as usual, just not execute anything downstream of proc1. This doesnt work, since the channel has been made non-empty by the two toList() operators, and output is now [[]].

If I instead use collect(), it runs fine, no empty lists are made and thus the merge process is not executed. Unfortunately collect creates a problem in that it seems to flatten the results to a single list, so the sort step doesnt work. So my question is, am I doing this wrong completely or is there some way to set a channel to empty (or close it) when detecting e.g. this:

...
.map { it == [[]] ? empty : it }
or 
.ifEmpty ( close_channel )

This strikes me as non-intuitive, but I don't know really. Until then I'm going with the alternative in which I do e.g. ifEmpty(['NA', 'NA', 'NA']), and detect the problem in downstream processes.

[bentsherman]

I didn't realize the collect operator automatically flattens its inputs, we'll have to update the docs with a note.

Anyway, the best solution I can think of at the moment is to add a filter step after the first toList:

  .filter { it.size() > 0 }