parse_input fails with KeyError instead of validation error

[virtuald]

Your latest release has a regression.

#!/usr/bin/env python3

import validobj.validation
import dataclasses
import typing as T


@dataclasses.dataclass(frozen=True)
class FooBase:
    ignore: bool = True


@dataclasses.dataclass(frozen=True)
class Foo(FooBase):
    pass


@dataclasses.dataclass(frozen=True)
class Spec:
    foo: T.Dict[str, Foo] = dataclasses.field(default_factory=dict)


value = {"foo": {"bar": {}}}

v = validobj.validation.parse_input(value, Spec)
print(v)

On 1.3 I get:

$ python test.py
Spec(foo={'bar': Foo(ignore=True)})

On 1.4 I get an exception:

$ python test.py
Traceback (most recent call last):
  File "test.py", line 22, in <module>
    validobj.validation.parse_input(value, Spec)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
  File "site-packages/validobj/validation.py", line 500, in parse_input
    return _parse_dataclass(value, spec)
  File "site-packages/validobj/validation.py", line 223, in _parse_dataclass
    res[k] = parse_input(v, field_dict[k])
             ~~~~~~~~~~~^^^^^^^^^^^^^^^^^^
  File "site-packages/validobj/validation.py", line 512, in parse_input
    return _handle_typing_spec(value, spec)
  File "site-packages/validobj/validation.py", line 372, in _handle_typing_spec
    return _parse_typing_mapping(value, spec)
  File "site-packages/validobj/validation.py", line 154, in _parse_typing_mapping
    parsed_v = parse_input(v, value_tp)
  File "site-packages/validobj/validation.py", line 500, in parse_input
    return _parse_dataclass(value, spec)
  File "site-packages/validobj/validation.py", line 218, in _parse_dataclass
    f.name: types[f.name] if not isinstance(f.type, dataclasses.InitVar) else f.type.type
            ~~~~~^^^^^^^^
KeyError: 'ignore'

Actual dataclasses at https://github.com/robotpy/semiwrap/blob/main/src/semiwrap/config/autowrap_yml.py if you're curious.

[Zaharid]

Thanks a lot for the report. I hope this is fixed now.

The fix is a bit awkward, and it is something that IMO the stdlib dataclasses module should handle. I believe it matches how fields are derived ( https://github.com/python/cpython/blob/7b7b90d1ce5116f29ad6c8120c0490824baa54e0/Lib/dataclasses.py#L962 ) closely enough to be correct.

[virtuald]

It works for me now, thanks for the quick fix!