Using config_path with subparsers does not seem to work

[aliounis]

Describe the bug
I am attempting to use subparsers and the config_path option together, however, I cannot seem to get the config path to actually set the defaults. I have tried multiple "paths" in the json file for the config file, posted below, but non seem to work. I've tried to investigate myself and think that the issue might be because the subparser does not copy the defaults for the command attribute (which chooses the subparser).

To Reproduce

import logging

from simple_parsing import ArgumentParser, Serializable


@dataclass
class DoA(Serializable):
    a_config: str = "taco"
    """
    what does a want?
    """

    def execute(self):
        print(f'A says {self.a_config}')


@dataclass
class DoB(Serializable):
    b_config: str = "pie"
    """
    what does b want?
    """

    def execute(self):
        print(f'B says {self.b_config}')


@dataclass
class Subprogram(Serializable):

    command: DoA | DoB
    """
    Do A or B?
    """

    def execute(self):
        self.command.execute()

if __name__ == "__main__":

    parser = ArgumentParser(add_config_path_arg=True)

    parser.add_arguments(Subprogram, dest="sub")

    # logging.basicConfig(level=logging.DEBUG, 
    #                     format="%(filename)s:%(lineno)s:%(name)s:%(message)s")
    args = parser.parse_args()

    sub: Subprogram = args.sub

    with open('./mwe_config_out.json', 'w') as ofile:
        sub.dump_json(ofile)

    sub.execute()

Expected behavior
Running

python sp_mwe.py doa --config_path mwe_config.json

with sp_mew.py being the above source code and mwe_config.json containing either:

    "sub": {
        "command": {
            "doa": {
                "a_config": "taco cheese"
            },
            "dob": {
                "b_config": "apple pie"
            }
        }
    }
}

or

    "sub": {
        "command": {
                "a_config": "taco cheese",
                "b_config": "apple pie"
            }
        }
    }

(I don't care which format, either would be fine) should produce the output

A says taco cheese

Actual behavior
Instead, using either of the json formats we get the default output of

A says taco

Desktop (please complete the following information):

• Version 0.1.0 (3122f6e)
• Python version: 3.11.0

[aliounis]

After some investigating, I can confirm that adding

            if isinstance(self.default, dict) and self.default.get(subcommand, None) is not None:
                subparser.add_arguments(dataclass_type, dest=self.dest, default=dataclass_type(**self.default.get(subcommand)))
            else:
                subparser.add_arguments(dataclass_type, dest=self.dest)

right after https://github.com/lebrice/SimpleParsing/blob/master/simple_parsing/wrappers/field_wrapper.py#L1033
produces the anticipated behavior when using the first config file in the original post (with doa and dob as entries). I am not sure if this would break other things, but at the very least it seems to work for my use case. If people think this is a good fix I can submit a PR

[lebrice]

Hello there @aliounis , thanks so much for posting this!

I'll try to look into this soon, but if you feel up to it and have time, sure, go ahead and create a PR!

This config_path argument seems to be causing issues (see #239 ), so hopefully we can kill two birds with one stone here.

Thanks again for posting!

[aliounis]

I submitted the above PR. This also grabs a second bug fix where subdataclasses had to be fully specified int he config file or not at all (partial specification was not allowed) instead of falling back to the standard default as it does for non-dataclass types. Let me know if you'd like me to open a second issue to track this other bug or if the PR/this comment is fine.

[tcapelle]

Is this going to be fixed? is there are workaround?

[lebrice]

Hello there @tcapelle, sorry no work has been done on this in a while. I will take a look when I find the time, but you're more than welcome to take a crack at it in the meantime! Let me know if you encounter any issues. #243 is an attempt at a fix, but broke a few tests. I'd start from there if I were you.