add support for _call_ = False in config_to_dict (so bare function or class)

[thomasloux]

No description provided.

[thomasloux]

@acecchini I need this code for my auto generation of config for Matune. But it also actually solves a TODO that was in comments in your code

[acecchini]

Ok I'll take care of this ASAP

[thomasloux]

I should add tests, but that's for later I think. Here is an example:

from math import sqrt
import fiddledyn
class Test:
  def __init__(self, fn):
    self.fn = fn

import fiddle
a = fiddle.Config(Test, fn=sqrt)
print(a)
b = fiddledyn.config_to_dict(a)
print(b)
c = fiddledyn.dict_to_config(b)
print(c)
d = fiddle.build(c)
assert a.fn == d.fn # assert passes

Which give

<Config[Test(fn=<built-in function sqrt>)]>
{'_target_': '__main__.Test', 'fn': {'_target_': 'math.sqrt', '_call_': False}}
<Config[Test(fn=<built-in function sqrt>)]>

[acecchini]

I have updated fiddledyn, the code has undergone substantial refactoring. A new test suite was developed and all tests pass now. Please check out the updated README.md and the new public API.

In particular parse_cli has now a as_dict option (which is by default set to False).

Furthermore, when setting include_defaults=True in config_to_dict method, fiddledyn will now check for manually unbound arguments which have a default value and will fetch the WHOLE subgraph found in the default value (so all of its nested callables will be interpreted as nested subgraphs in nested and will be wrapped in{fiddle, nemo_run}.{Config, Partial} objects).

Finally, nemo-run is now optional to reduce dependency graph of the repo. If you ever need nemo-run config objects please add it into your pyproject.toml with the square bracket syntax, i.e. fiddledyn[nemo]. Everything is expalined in the README.md