Satrajit Ghosh

i think the challenge is with combining, and something needs to retain the shape of the original splitter operation. in the above example, technically one should be able to combine...

if i remember the implementation, one could combine over `a`, `b`, `[a, b]` or `c` as valid options after such a split. combining with any of the first two would...

@dafrose - have you tried inserting a passthrough task whose only function is to combine? i vaguely remember discussing this with @djarecka in the past, but left it as an...

essentially something like this. ```python @pydra.mark.task def passthrough(arg1): return arg1 passthroughtask = passthrough(arg1=prev_task.lzout.out).combine(splitter_info) ``` now the output of this task should have the combined state. effectively you are asking for...

it's a bug if it doesn't work without a `Workflow`. i'll let @djarecka comment on both, since she is much more familiar with the internals.

for a combiner or splitter exception, it would be useful to list the names of available states, which should be accessible given the inputs to the task.

i think this is a good starting point. some comments below. > so I'm not quite sure why the user should have to provide an OutputSpec to get more than...

i think there are two questions implicit in this discussion: 1. what is the list of output keys? this would be used to allow users to connect outputs of the...

> But the zenodo spec does not give you answer to this questions i think the zenodo spec gives you the answer to (1) just like nipype outputspec. but it...

and would you automatically add that to cache locations for the entire workflow? i think this may be getting into a can of worms. also people can set cache dir...