Prefect and dataclasses: I want to refactor my tasks to pass around dataclasses instead of dicts and stuff. now, if I have a

dict

, prefect handles it by planting the

getItem

task in the DAG:

dict_output = generate_dict()
a = task_a(dict_output['a'])
b = task_b(dict_output['b'])

How would you recommend working with dataclasses? The following piece of code won’t work, naturally (since a task doesn’t have these attributes)

@dataclass
def MyClass:
    a: int
    b: int
...
...
dataclass_output = generate_dataclass()
a = task_a(dataclass_output.a)
b = task_b(dataclass_output.b)
c = another_task(dataclass_output)

Hey @Avi A - behind the scenes indexing is just creating one of these

GetItem

tasks (see here) - you could replicate this with a

GetAttribute

task that basically runs

lambda x: x.attr

. Adding sugar for that might be tricky, because we don’t want to allow magic attribute access to the task class. Maybe something like

dataclass_output.output.a

or something? Just idle thoughts.

Thanks @Jeremiah. So

dataclass_output.output

allows direct interaction with a `Task`’s result?

No, sorry for the confusion - I was trying to think of how we might implement something like what you’re looking for. Today that isn’t implemented.

yeah I got that part 🙂

👌 👌

follow-up: from what I could tell I see that

GetItem

doesn’t persist even though

Result

is defined on the flow, which makes a lot of sense b/c it just extracts data from existing data. How can I do the same for this task? would it help to just set

target=None

?

I think so - there’s nothing special about the

GetItem

task AFAIK