ValueError getting torch optimizers help message

[carmocca]

# repro.py
import jsonargparse
import torch

parser = jsonargparse.ArgumentParser()
parser.add_subclass_arguments(torch.optim.Optimizer, "optimizer", skip={"params"})
args = parser.parse_args()

$ python repro.py --optimizer.help torch.optim.Optimizer
Traceback (most recent call last):
  File "/Users/carlosmocholi/git/pytorch-lightning/kk2.py", line 6, in <module>
    args = parser.parse_args()
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/envs/lightning-3.9/lib/python3.9/site-packages/jsonargparse/deprecated.py", line 112, in patched_parse
    cfg = parse_method(*args, _skip_check=_skip_check, **kwargs)
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/envs/lightning-3.9/lib/python3.9/site-packages/jsonargparse/core.py", line 371, in parse_args
    cfg, unk = self.parse_known_args(args=args)
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/envs/lightning-3.9/lib/python3.9/site-packages/jsonargparse/core.py", line 260, in parse_known_args
    namespace, args = self._parse_known_args(args, namespace)
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/lib/python3.9/argparse.py", line 2060, in _parse_known_args
    start_index = consume_optional(start_index)
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/lib/python3.9/argparse.py", line 2000, in consume_optional
    take_action(action, args, option_string)
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/lib/python3.9/argparse.py", line 1928, in take_action
    action(self, namespace, argument_values, option_string)
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/envs/lightning-3.9/lib/python3.9/site-packages/jsonargparse/actions.py", line 344, in __call__
    self.print_help(args, self._baseclass, dest)
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/envs/lightning-3.9/lib/python3.9/site-packages/jsonargparse/actions.py", line 375, in print_help
    super().print_help(call_args, val_class, dest+'.init_args')
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/envs/lightning-3.9/lib/python3.9/site-packages/jsonargparse/actions.py", line 348, in print_help
    tmp.add_class_arguments(val_class, dest)
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/envs/lightning-3.9/lib/python3.9/site-packages/jsonargparse/signatures.py", line 74, in add_class_arguments
    return self._add_signature_arguments(inspect.getmro(theclass),
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/envs/lightning-3.9/lib/python3.9/site-packages/jsonargparse/signatures.py", line 248, in _add_signature_arguments
    self._add_signature_parameter(
  File "/Users/carlosmocholi/.pyenv/versions/3.9.1/envs/lightning-3.9/lib/python3.9/site-packages/jsonargparse/signatures.py", line 361, in _add_signature_parameter
    raise ValueError(f'Required parameter without a type for {obj.__name__} parameter {name}.')
ValueError: Required parameter without a type for Optimizer parameter params.

I understand what the error message says, but shouldn't the help message still work?

Noticed this while checking out Lightning-AI/pytorch-lightning#10725

[mauvilsa]

The code snippet is not equivalent to what is in lightning, in particular missing the skip. But from what I see the class help action is not considering the skip. This is a bug.

[mauvilsa]

Fixed in 0f02e94