Inconsistent Channel Operator Behavior

[cjw85]

The channel operator API in nextflow could do with rationalisation and cleanup

Usage scenario

Here's one example of an inconsistency in the operators. From the docs:

Join: docs

left = Channel.from(['X', 1], ['Y', 2], ['Z', 3], ['P', 7])
right= Channel.from(['Z', 6], ['Y', 5], ['X', 4])
left.join(right).view()
[Z, 3, 6]
[Y, 2, 5]
[X, 1, 4]

Combine: docs

left = Channel.from(['A',1], ['B',2], ['A',3])
right = Channel.from(['B','x'], ['B','y'], ['A','z'], ['A', 'w'])

left
    .combine(right, by: 0)
    .view()
[A, 1, z]
[A, 3, z]
[A, 1, w]
[A, 3, w]
[B, 2, x]
[B, 2, y]

Cross: docs

source = Channel.from( [1, 'alpha'], [2, 'beta'] )
target = Channel.from( [1, 'x'], [1, 'y'], [1, 'z'], [2,'p'], [2,'q'], [2,'t'] )
source.cross(target).view()
[ [1, alpha], [1, x] ]
[ [1, alpha], [1, y] ]
[ [1, alpha], [1, z] ]
[ [2, beta],  [2, p] ]
[ [2, beta],  [2, q] ]
[ [2, beta],  [2, t] ]

All three of the above operators I would consider as types of product on their input channels, however the output of cross is at odds with the output of join and combine: cross forms a simple tuple from inputs verbatim whereas join and combine flatten the inputs and bring the joining key to the front of the resulting tuple.

I believe I've come across other implicit flattenings of structures but I don't recall the example.

I and others spend a lot of time viewing and dumping channnels in order to check whats going on: often because of confusions like the above where operators have different flattening behaviours. This creates unnecessary cognitive overhead, constant reference to the docs, and adds to the learning curve for nextflow.