Parsing dataclass construction

[liamhuber]

Dataclasses are just callables, and for a given dataclass one can straightforwardly construct an unambiguous atomic recipe to transform individual inputs -> dataclass instance outputs. Thus, it would be nice to be able to parse workflows like:

import dataclasses
import flowrep as fr

@fr.dataclass
class Payload:
    item: str
    multiple: int

def order(payload: Payload):
    to_buy = f"Please buy {payload.multiple} {payload.item}s"

@fr.workflow
def shopping_for_two(item, n):
    p = Payload(item, n)
    to_buy = order(p)
    return to_buy

I think if @fr.dataclass attaches a .flowrep_recipe: AtomicRecipe, and a partial function for that recipe to reference, this should work out-of-the-box. Compiling it might be slightly more of a headache, but I think it's just bookkeeping and not any fundamental obstacle.

I've already done something similar in pyiron_workflow to overload dataclasses.dataclass with workflow power-ups, so I know it's very doable and should be able to mostly just port parts of that solution here.

[liamhuber]

It might even be possible to avoid attaching some partialized creator function, and just reference the dataclass itself, as long as some types for the AtomicRecipe branch get relaxed from types.FunctionType to collections.abc.Callable.