"""Private logic for creating pydantic dataclasses."""
from __future__ import annotations as _annotations
import copy
import dataclasses
import sys
import warnings
from collections.abc import Generator
from contextlib import contextmanager
from functools import partial
from typing import TYPE_CHECKING, Any, ClassVar, Protocol, cast
from pydantic_core import (
ArgsKwargs,
SchemaSerializer,
SchemaValidator,
core_schema,
)
from typing_extensions import TypeAlias, TypeIs
from ..errors import PydanticUndefinedAnnotation
from ..fields import FieldInfo
from ..plugin._schema_validator import PluggableSchemaValidator, create_schema_validator
from ..warnings import PydanticDeprecatedSince20
from . import _config, _decorators
from ._fields import collect_dataclass_fields
from ._generate_schema import GenerateSchema, InvalidSchemaError
from ._generics import get_standard_typevars_map
from ._mock_val_ser import set_dataclass_mocks
from ._namespace_utils import NsResolver
from ._signature import generate_pydantic_signature
from ._utils import LazyClassAttribute
if TYPE_CHECKING:
from _typeshed import DataclassInstance as StandardDataclass
from ..config import ConfigDict
class PydanticDataclass(StandardDataclass, Protocol):
"""A protocol containing attributes only available once a class has been decorated as a Pydantic dataclass.
Attributes:
__pydantic_config__: Pydantic-specific configuration settings for the dataclass.
__pydantic_complete__: Whether dataclass building is completed, or if there are still undefined fields.
__pydantic_core_schema__: The pydantic-core schema used to build the SchemaValidator and SchemaSerializer.
__pydantic_decorators__: Metadata containing the decorators defined on the dataclass.
__pydantic_fields__: Metadata about the fields defined on the dataclass.
__pydantic_serializer__: The pydantic-core SchemaSerializer used to dump instances of the dataclass.
__pydantic_validator__: The pydantic-core SchemaValidator used to validate instances of the dataclass.
"""
__pydantic_config__: ClassVar[ConfigDict]
__pydantic_complete__: ClassVar[bool]
__pydantic_core_schema__: ClassVar[core_schema.CoreSchema]
__pydantic_decorators__: ClassVar[_decorators.DecoratorInfos]
__pydantic_fields__: ClassVar[dict[str, FieldInfo]]
__pydantic_serializer__: ClassVar[SchemaSerializer]
__pydantic_validator__: ClassVar[SchemaValidator | PluggableSchemaValidator]
@classmethod
def __pydantic_fields_complete__(cls) -> bool: ...
def set_dataclass_fields(
cls: type[StandardDataclass],
config_wrapper: _config.ConfigWrapper,
ns_resolver: NsResolver | None = None,
) -> None:
"""Collect and set `cls.__pydantic_fields__`.
Args:
cls: The class.
config_wrapper: The config wrapper instance.
ns_resolver: Namespace resolver to use when getting dataclass annotations.
"""
typevars_map = get_standard_typevars_map(cls)
fields = collect_dataclass_fields(
cls, ns_resolver=ns_resolver, typevars_map=typevars_map, config_wrapper=config_wrapper
)
cls.__pydantic_fields__ = fields # type: ignore
def complete_dataclass(
cls: type[Any],
config_wrapper: _config.ConfigWrapper,
*,
raise_errors: bool = True,
ns_resolver: NsResolver | None = None,
_force_build: bool = False,
) -> bool:
"""Finish building a pydantic dataclass.
This logic is called on a class which has already been wrapped in `dataclasses.dataclass()`.
This is somewhat analogous to `pydantic._internal._model_construction.complete_model_class`.
Args:
cls: The class.
config_wrapper: The config wrapper instance.
raise_errors: Whether to raise errors, defaults to `True`.
ns_resolver: The namespace resolver instance to use when collecting dataclass fields
and during schema building.
_force_build: Whether to force building the dataclass, no matter if
[`defer_build`][pydantic.config.ConfigDict.defer_build] is set.
Returns:
`True` if building a pydantic dataclass is successfully completed, `False` otherwise.
Raises:
PydanticUndefinedAnnotation: If `raise_error` is `True` and there is an undefined annotations.
"""
original_init = cls.__init__
# dataclass.__init__ must be defined here so its `__qualname__` can be changed since functions can't be copied,
# and so that the mock validator is used if building was deferred:
def __init__(__dataclass_self__: PydanticDataclass, *args: Any, **kwargs: Any) -> None:
__tracebackhide__ = True
s = __dataclass_self__
s.__pydantic_validator__.validate_python(ArgsKwargs(args, kwargs), self_instance=s)
__init__.__qualname__ = f'{cls.__qualname__}.__init__'
cls.__init__ = __init__ # type: ignore
cls.__pydantic_config__ = config_wrapper.config_dict # type: ignore
set_dataclass_fields(cls, config_wrapper=config_wrapper, ns_resolver=ns_resolver)
if not _force_build and config_wrapper.defer_build:
set_dataclass_mocks(cls)
return False
if hasattr(cls, '__post_init_post_parse__'):
warnings.warn(
'Support for `__post_init_post_parse__` has been dropped, the method will not be called',
PydanticDeprecatedSince20,
)
typevars_map = get_standard_typevars_map(cls)
gen_schema = GenerateSchema(
config_wrapper,
ns_resolver=ns_resolver,
typevars_map=typevars_map,
)
# set __signature__ attr only for the class, but not for its instances
# (because instances can define `__call__`, and `inspect.signature` shouldn't
# use the `__signature__` attribute and instead generate from `__call__`).
cls.__signature__ = LazyClassAttribute(
'__signature__',
partial(
generate_pydantic_signature,
# It's important that we reference the `original_init` here,
# as it is the one synthesized by the stdlib `dataclass` module:
init=original_init,
fields=cls.__pydantic_fields__, # type: ignore
validate_by_name=config_wrapper.validate_by_name,
extra=config_wrapper.extra,
is_dataclass=True,
),
)
try:
schema = gen_schema.generate_schema(cls)
except PydanticUndefinedAnnotation as e:
if raise_errors:
raise
set_dataclass_mocks(cls, f'`{e.name}`')
return False
core_config = config_wrapper.core_config(title=cls.__name__)
try:
schema = gen_schema.clean_schema(schema)
except InvalidSchemaError:
set_dataclass_mocks(cls)
return False
# We are about to set all the remaining required properties expected for this cast;
# __pydantic_decorators__ and __pydantic_fields__ should already be set
cls = cast('type[PydanticDataclass]', cls)
cls.__pydantic_core_schema__ = schema
cls.__pydantic_validator__ = create_schema_validator(
schema, cls, cls.__module__, cls.__qualname__, 'dataclass', core_config, config_wrapper.plugin_settings
)
cls.__pydantic_serializer__ = SchemaSerializer(schema, core_config)
cls.__pydantic_complete__ = True
return True
def is_stdlib_dataclass(cls: type[Any], /) -> TypeIs[type[StandardDataclass]]:
"""Returns `True` if the class is a stdlib dataclass and *not* a Pydantic dataclass.
Unlike the stdlib `dataclasses.is_dataclass()` function, this does *not* include subclasses
of a dataclass that are themselves not dataclasses.
Args:
cls: The class.
Returns:
`True` if the class is a stdlib dataclass, `False` otherwise.
"""
return '__dataclass_fields__' in cls.__dict__ and not hasattr(cls, '__pydantic_validator__')
def as_dataclass_field(pydantic_field: FieldInfo) -> dataclasses.Field[Any]:
field_args: dict[str, Any] = {'default': pydantic_field}
# Needed because if `doc` is set, the dataclass slots will be a dict (field name -> doc) instead of a tuple:
if sys.version_info >= (3, 14) and pydantic_field.description is not None:
field_args['doc'] = pydantic_field.description
# Needed as the stdlib dataclass module processes kw_only in a specific way during class construction:
if sys.version_info >= (3, 10) and pydantic_field.kw_only:
field_args['kw_only'] = True
# Needed as the stdlib dataclass modules generates `__repr__()` during class construction:
if pydantic_field.repr is not True:
field_args['repr'] = pydantic_field.repr
return dataclasses.field(**field_args)
DcFields: TypeAlias = dict[str, dataclasses.Field[Any]]
@contextmanager
def patch_base_fields(cls: type[Any]) -> Generator[None]:
"""Temporarily patch the stdlib dataclasses bases of `cls` if the Pydantic `Field()` function is used.
When creating a Pydantic dataclass, it is possible to inherit from stdlib dataclasses, where
the Pydantic `Field()` function is used. To create this Pydantic dataclass, we first apply
the stdlib `@dataclass` decorator on it. During the construction of the stdlib dataclass,
the `kw_only` and `repr` field arguments need to be understood by the stdlib *during* the
dataclass construction. To do so, we temporarily patch the fields dictionary of the affected
bases.
For instance, with the following example:
```python {test="skip" lint="skip"}
import dataclasses as stdlib_dc
import pydantic
import pydantic.dataclasses as pydantic_dc
@stdlib_dc.dataclass
class A:
a: int = pydantic.Field(repr=False)
# Notice that the `repr` attribute of the dataclass field is `True`:
A.__dataclass_fields__['a']
#> dataclass.Field(default=FieldInfo(repr=False), repr=True, ...)
@pydantic_dc.dataclass
class B(A):
b: int = pydantic.Field(repr=False)
```
When passing `B` to the stdlib `@dataclass` decorator, it will look for fields in the parent classes
and reuse them directly. When this context manager is active, `A` will be temporarily patched to be
equivalent to:
```python {test="skip" lint="skip"}
@stdlib_dc.dataclass
class A:
a: int = stdlib_dc.field(default=Field(repr=False), repr=False)
```
!!! note
This is only applied to the bases of `cls`, and not `cls` itself. The reason is that the Pydantic
dataclass decorator "owns" `cls` (in the previous example, `B`). As such, we instead modify the fields
directly (in the previous example, we simply do `setattr(B, 'b', as_dataclass_field(pydantic_field))`).
!!! note
This approach is far from ideal, and can probably be the source of unwanted side effects/race conditions.
The previous implemented approach was mutating the `__annotations__` dict of `cls`, which is no longer a
safe operation in Python 3.14+, and resulted in unexpected behavior with field ordering anyway.
"""
# A list of two-tuples, the first element being a reference to the
# dataclass fields dictionary, the second element being a mapping between
# the field names that were modified, and their original `Field`:
original_fields_list: list[tuple[DcFields, DcFields]] = []
for base in cls.__mro__[1:]:
dc_fields: dict[str, dataclasses.Field[Any]] = base.__dict__.get('__dataclass_fields__', {})
dc_fields_with_pydantic_field_defaults = {
field_name: field
for field_name, field in dc_fields.items()
if isinstance(field.default, FieldInfo)
# Only do the patching if one of the affected attributes is set:
and (field.default.description is not None or field.default.kw_only or field.default.repr is not True)
}
if dc_fields_with_pydantic_field_defaults:
original_fields_list.append((dc_fields, dc_fields_with_pydantic_field_defaults))
for field_name, field in dc_fields_with_pydantic_field_defaults.items():
default = cast(FieldInfo, field.default)
# `dataclasses.Field` isn't documented as working with `copy.copy()`.
# It is a class with `__slots__`, so should work (and we hope for the best):
new_dc_field = copy.copy(field)
# For base fields, no need to set `doc` from `FieldInfo.description`, this is only relevant
# for the class under construction and handled in `as_dataclass_field()`.
if sys.version_info >= (3, 10) and default.kw_only:
new_dc_field.kw_only = True
if default.repr is not True:
new_dc_field.repr = default.repr
dc_fields[field_name] = new_dc_field
try:
yield
finally:
for fields, original_fields in original_fields_list:
for field_name, original_field in original_fields.items():
fields[field_name] = original_field
