Description
It took me a while to find a way to reproduce this bug, and it doesn't quite happen here in the way it's happening in our larger codebase, however I'd like to give it a go:
🐛 Bug report
Specifying types which have forward references included can lead to errors when the forward-referenced type is not in scope.
In this case ForwardReferenced is in scope when defining Direct, so adding class arguments will work. However, in the case of Indirect ForwardReferenced is not in scope when using NamedType leading to an error when adding class variables.
To reproduce
types_file
NamedType = list["ForwardReferenced"]
class ForwardReferenced:
pass
class Direct:
def __init__(self, data_type: NamedType):
passminimal_reproduction.py
from jsonargparse import ArgumentParser
from types_file import NamedType, Direct
class Indirect:
def __init__(self, data_type: NamedType):
pass
parser = ArgumentParser()
parser.add_class_arguments(Direct, "direct")
parser.add_class_arguments(Indirect, "indirect") # <---- will raise errorExpected behavior
I expect both Direct and Indirect class arguments to be addable, instead adding Indirect will throw a ValueError:
ValueError: With fail_untyped=True, all mandatory parameters must have a supported type. Parameter 'data_type' from '__main__.Indirect.__init__' does not specify a type.
resolving it is trivial, you can either import the forward-referenced type within the current scope or remove the forward reference from the type.
I've found a similar case when using parser.link_arguments but have not quite been able to reproduce it. In our case it was quite hard to figure out the cause of the bug. At the very least there should be some impovement to error reporting.
Environment
- jsonargparse version: 4.37.0
- Python version: 3.11
- How jsonargparse was installed: UV
- OS: Pop!_OS 22.04 LTS