Hi everyone, I am working on the bug of ambiguous dataclass for Union type (Here is the draft PR). Some questions to discuss: 1. I’m using two Go packages, one licensed under LGPL-2.1 and the other under Apache-2.0. Does anyone know if these licenses permit commercial use? 2. We are planning to support superset dataclasses in Union type, meaning that the expected type of the workflow’s input could be a superset of the user’s input. I am wondering if there are any detailed examples or expected behaviors for nested cases. Please feel free to comment on the PR as well. cc @damp-lion-88352 @cool-lifeguard-49380 @thankful-minister-83577 @damp-rain-31363 Thanks!

Hi can you explain question 2 more?

This is the example I’ve come up with so far. For nested cases, will both of them be considered a superset?

from dataclasses import dataclass
from typing import Optional, Union
from flytekit import task, workflow

@dataclass
class A:
    i: int

@dataclass
class B:
    nested_class: A

@dataclass
class C:
    i: int
    j: Optional[int]
    k: int = 5

@task 
def my_task() -> A:
    return A(i=1)

@task 
def foo(input: Union[B, C]):
    print(input)

@workflow
def my_wf():
    a = my_task()
    foo(input=a)

No, this should not work, right? You pass an A where you expect a B or C and the only commonality would be that B has an attr of type A. The question is whether inheritance should be supported, meaning B is

class B(A)

.

LGPL might be a problem

Good call, @freezing-airport-6809. Let's not use any copyleft dependency, please. That includes LGPL.

key question: is LGPL compatible with Apache License?

the specific case you’re referring to should not work right? B is not a subclass of A, it just contains an A. i think just a schema check is the most appropriate and is completely sufficient. what are the two libraries you’re looking at?

cc @nutritious-telephone-58961

the specific case you’re referring to should not work right? B is not a subclass of A, it just contains an A. i think just a schema check is the most appropriate and is completely sufficient. what are the two libraries you’re looking at?

So will class C match with an input of class A? Something like the example below:

@dataclass
class C(A):
    j: Optional[int]
    k: int = 5

The main problem is that if we plan to support comparing jsonschemas generated from

marshmallow

and

mashumaro

, they have huge differences in their structure. To compare them, we need additional processes. The two libraries I originally used are json_schema_compare and jsonschema, but actually they can not handle all the differences, which means we still need to do some processing before or after by ourselves. We are still finding other useful packages as well. Btw thanks for the clarification of LGPL license!