RFC: Task initialization syntax

[effigies]

Let's break things!

Right now, when you instantiate a Task, you can give it a name:

@pydra.mark.task
def greet(name : str) -> str:
    return f"Hi, {name}"

greeter = greet(name="greeter")
print(greeter(name="@effigies").output.out)

Hi, @effigies

So this works, but it might be worth considering whether this is going to cause confusion in a workflow context:

wf = pydra.Workflow(name="myworkflow", input_spec=["name"])
wf.add(greet(name="greeter"))
wf.greeter.inputs.name = wf.lzin.name
wf.set_output([("out", wf.greeger.lzout.out")])
wf.inputs.name = "@effigies"

with pydra.Submitter() as sub: 
    print(sub(wf).output.out)

Hi, myworkflow

So this is both confusing and actually doesn't work.

In any event, I wanted to open discussion about possibly rethinking the syntax, or possibly just some names.

[satra]

a couple of options are:

1. greet(name="greeter", [other_init_keywords,] inputs={"name": wf.lzin.name})
2. greet(name="greeter", [other_init_keywords,] pydin_name=wf.lzin.name)

[satra]

@effigies - we already had this in the base class:
https://github.com/nipype/pydra/blob/master/pydra/engine/core.py#L84
https://github.com/nipype/pydra/blob/master/pydra/engine/core.py#L157

@djarecka - somehow this was not adhered to in derived classes. any memories why? but in general we should follow this pattern. note that this issue is only for init

[djarecka]

@satra - i don't remember why, but it was a time that input was the only option to provide input in __init__.

I'm a bit confuse your second option... is it possible that you wanted to write:
2. greet(pydin_name="greeter", [other_init_keywords,], name=wf.lzin.name)?

[effigies]

I think I would be inclined to keep inputs explicit at __init__, and continue passing them as keywords at __call__(). I think for a subclass to change the syntax just invites confusion.

[satra]

2. greet(pydin_name="greeter", [other_init_keywords,], name=wf.lzin.name)?

i did not mean that. i did mean the prefix to indicate input keywords.

i think at this point i would take the decision of restricting input values during init, not during run or function call, only through the inputs keyword.

[djarecka]

I got used to the shorter syntax, but this can cause problems, so probably we should change it.
I guess we already proofed that the option 1 is better than the option 2 ;-)

[djarecka]

should I work on this next week, or we still wanna take more time to think?

[effigies]

Is there a hive mind to ping that isn't already getting notifications from here? If not, I'd say go ahead.

[satra]

i say go ahead too.

[tclose]

no longer relevant