Unable to pickle pydantic model #565
Description
After pydantic 2.11.0 release.
It looks like the issue is with references now used for optimizations in Rust code after this change in pydantic-core: pydantic/pydantic-core#1616.
How to reproduce
model-write.py:
from cloudpickle import dumps
from pydantic import BaseModel
class MyModel(BaseModel):
name: str = ""
serialized = dumps(MyModel)
with open("model.pkl", "wb") as f:
f.write(serialized)model-read.py
from cloudpickle import loads
with open("model.pkl", "rb") as f:
serialized = f.read()
MyModel = loads(serialized)
model = MyModel(name="bar")
assert model.name == "bar"$ uv pip install cloudpickle pydantic
Using Python 3.13.2 environment at: /Users/vlad/.virtualenvs/test-pydantic
Resolved 6 packages in 131ms
Installed 6 packages in 5ms
+ annotated-types==0.7.0
+ cloudpickle==3.1.1
+ pydantic==2.11.1
+ pydantic-core==2.33.0
+ typing-extensions==4.13.0
+ typing-inspection==0.4.0
$ python model-write.py
$ python model-read.py
Traceback (most recent call last):
File "/Users/vlad/playground/model-read.py", line 9, in <module>
assert model.name == "bar"
^^^^^^^^^^
File "/Users/vlad/.virtualenvs/test-pydantic/lib/python3.13/site-packages/pydantic/main.py", line 984, in __getattr__
raise AttributeError(f'{type(self).__name__!r} object has no attribute {item!r}')
AttributeError: 'MyModel' object has no attribute 'name'
$
More details
MyModel.__pydantic_validator__ before serialization:
SchemaValidator(title="MyModel", validator=Model(
ModelValidator {
revalidate: Never,
validator: ModelFields(
ModelFieldsValidator {
fields: [
Field {
name: "name",
lookup_key_collection: LookupKeyCollection {
by_name: Simple(
LookupPath {
first_item: PathItemString {
key: "name",
py_key: Py(
0x0000000101d1bd50,
),
},
rest: [],
},
),
by_alias: None,
by_alias_then_name: None,
},
name_py: Py(
0x00000001013da278,
),
validator: WithDefault(
WithDefaultValidator {
default: Default(
Py(
0x00000001013d33d8,
),
),
on_error: Raise,
validator: Str(
StrValidator {
strict: false,
coerce_numbers_to_str: false,
},
),
validate_default: false,
copy_default: false,
name: "default[str]",
undefined: Py(
0x00000001009f4730,
),
},
),
frozen: false,
},
],
model_name: "MyModel",
extra_behavior: Ignore,
extras_validator: None,
extras_keys_validator: None,
strict: false,
from_attributes: false,
loc_by_alias: true,
validate_by_alias: None,
validate_by_name: None,
},
),
class: Py(
0x000000011f61bd10,
),
generic_origin: None,
post_init: None,
frozen: false,
custom_init: false,
root_model: false,
undefined: Py(
0x00000001009f4730,
),
name: "MyModel",
},
), definitions=[], cache_strings=True)
MyModel.__pydantic_validator__ after deserialization:
SchemaValidator(title="MyModel", validator=Prebuilt(
PrebuiltValidator {
schema_validator: Py(
0x00000001520eddc0,
),
},
), definitions=[], cache_strings=True)
Even more details
Naive approaches with using register_pickle_by_value does not works.
pydantic_coreonly:
import cloudpickle
import pydantic_core
cloudpickle.register_pickle_by_value(pydantic_core)Gives this error on deserialization:
Traceback (most recent call last):
File "/Users/vlad/playground/model-read.py", line 6, in <module>
MyModel = cloudpickle.loads(serialized)
AttributeError: type object 'SchemaSerializer' has no attribute 'to_json'
- Both
pydanticandpydantic_core:
import cloudpickle
import pydantic
import pydantic_core
cloudpickle.register_pickle_by_value(pydantic)
cloudpickle.register_pickle_by_value(pydantic_core)Gives this error on serialization:
Traceback (most recent call last):
File "/Users/vlad/playground/model-write.py", line 15, in <module>
serialized = cloudpickle.dumps(MyModel)
File "/Users/vlad/.virtualenvs/test-pydantic/lib/python3.13/site-packages/cloudpickle/cloudpickle.py", line 1537, in dumps
cp.dump(obj)
~~~~~~~^^^^^
File "/Users/vlad/.virtualenvs/test-pydantic/lib/python3.13/site-packages/cloudpickle/cloudpickle.py", line 1303, in dump
return super().dump(obj)
~~~~~~~~~~~~^^^^^
TypeError: cannot pickle 'classmethod_descriptor' object
(related cloudpickle issue: #486)