Pydantic V2 Pre Release

#Ready for experimentation

• BaseModel - the core of validation in Pydantic V1 remains, albeit with new method names.
• Dataclasses - Pydantic dataclasses are improved and ready to test.
• Serialization - dumping/serialization/marshalling is significantly more flexible, and ready to test.
• Strict mode - one of the biggest additions in Pydantic V2 is strict mode, which is ready to test.
• JSON Schema - generation of JSON Schema is much improved and ready to test.
• Generic Models - are much improved and ready to test.
• Recursive Models - and validation of recursive data structures is much improved and ready to test.
• Custom Types - custom types have a new interface and are ready to test.
• Custom Field Modifiers - used via Annotated[] are working and in use in Pydantic itself.
• Validation without a BaseModel - the new TypeAdapter class allows validation without the need for a BaseModel class, and it's ready to test.
• TypedDict - we now have full support for TypedDict via TypeAdapter, it's ready to test.

#Still under construction

• Documentation - we're working hard on full documentation for V2, but it's not ready yet.
• Conversion Table - a big addition to the documentation will be a conversion table showing how types are coerced, this is a WIP.
• BaseSettings - BaseSettings will move to a separate pydantic-settings package, it's not yet ready to test. Notice: since pydantic-settings is not yet ready to release, there's no support for BaseSettings in the first alpha release.
• validate_arguments - the validate_arguments decorator remains and is working, but hasn't been updated yet.
• Hypothesis Plugin - the Hypothesis plugin is yet to be updated.
• computed fields - we know a lot of people are waiting for this, we will include it in Pydantic V2.
• Error messages - could use some love, and links to docs in error messages are still to be added.
• Migration Guide - we have some pointers below, but this needs completing.

#Changes to BaseModel

• Various method names have been changed; BaseModel methods all start with model_ now. Where possible, we have retained the old method names to help ease migration, but calling them will result in DeprecationWarnings.
  • Some of the built-in data loading functionality has been slated for removal. In particular, parse_raw and parse_file are now deprecated. You should load the data and then pass it to model_validate.
• The from_orm method has been removed; you can now just use model_validate (equivalent to parse_obj from Pydantic V1) to achieve something similar, as long as you've set from_attributes=True in the model config.
• The __eq__ method has changed for models; models are no longer considered equal to the dicts.
• Custom __init__ overrides won't be called. This should be replaced with a @root_validator.
• Due to inconsistency with the rest of the library, we have removed the special behavior of models using the __root__ field, and have disallowed the use of an attribute with this name to prevent confusion. However, you can achieve equivalent behavior with a "standard" field name through the use of @root_validator, @model_serializer, and __pydantic_modify_json_schema__. You can see an example of this here.

#Changes to Pydantic Dataclasses

• The __post_init__ in Pydantic dataclasses will now be called after validation, rather than before.
• We no longer support extra='allow' for Pydantic dataclasses, where extra attributes passed to the initializer would be stored as extra fields on the dataclass. extra='ignore' is still supported for the purposes of allowing extra fields while parsing data; they just aren't stored.
• __post_init_post_parse__ has been removed.
• Nested dataclasses no longer accept tuples as input, only dict.

#Changes to Config

• To specify config on a model, it is now deprecated to create a class called Config in the namespace of the parent BaseModel subclass. Instead, you just need to set a class attribute called model_config to be a dict with the key/value pairs you want to be used as the config.

The following config settings have been removed:

• allow_mutation — this has been removed. You should be able to use frozen equivalently (inverse of current use).
• error_msg_templates.
• fields — this was the source of various bugs, so has been removed. You should be able to use Annotated on fields to modify them as desired.
• getter_dict — orm_mode has been removed, and this implementation detail is no longer necessary.
• schema_extra — you should now use the json_schema_extra keyword argument to pydantic.Field.
• smart_union.
• underscore_attrs_are_private — the Pydantic V2 behavior is now the same as if this was always set to True in Pydantic V1.

The following config settings have been renamed:

• allow_population_by_field_name → populate_by_name
• anystr_lower → str_to_lower
• anystr_strip_whitespace → str_strip_whitespace
• anystr_upper → str_to_upper
• keep_untouched → ignored_types
• max_anystr_length → str_max_length
• min_anystr_length → str_min_length
• orm_mode → from_attributes
• validate_all → validate_default

#Changes to Validators

• Raising a TypeError inside a validator no longer produces a ValidationError, but just raises the TypeError directly. This was necessary to prevent certain common bugs (such as calling functions with invalid signatures) from being unintentionally converted into ValidationError and displayed to users. If you really want TypeError to be converted to a ValidationError you should use a try: except: block that will catch it and do the conversion.
• each_item validators are deprecated and should be replaced with a type annotation using Annotated to apply a validator or with a validator that operates on all items at the top level.
• Changes to @validator-decorated function signatures.
• The stricturl type has been removed.
• Root validators can no longer be run with skip_on_failure=False.

#Changes to Validation of specific types

• Integers outside the valid range of 64 bit integers will cause ValidationErrors during parsing. To work around this, use an IsInstance validator (more details to come).
• Subclasses of built-ins won't validate into their subclass types; you'll need to use an IsInstance validator to validate these types.

#Changes to Generic models

• While it does not raise an error at runtime yet, subclass checks for parametrized generics should no longer be used. These will result in TypeErrors and we can't promise they will work forever. However, it will be okay to do subclass checks against non-parametrized generic models

#Other changes

• GetterDict has been removed, as it was just an implementation detail for orm_mode, which has been removed.

#TypeAdapter

Pydantic V1 didn't have good support for validation or serializing non-BaseModel. To work with them you had to create a "root" model or use the utility functions in pydantic.tools (parse_obj_as and schema_of). In Pydantic V2 this is a lot easier: the TypeAdapter class lets you build an object that behaves almost like a BaseModel class which you can use for a lot of the use cases of root models and as a complete replacement for parse_obj_as and schema_of.

from typing import List

from pydantic import TypeAdapter

validator = TypeAdapter(List[int])
assert validator.validate_python(['1', '2', '3']) == [1, 2, 3]
print(validator.json_schema())
#> {'items': {'type': 'integer'}, 'type': 'array'}

Note that this API is provisional and may change before the final release of Pydantic V2.