I have a Pydantic model representing a bank account holding transactions, which themselves are models that are nested on the account model. The transactions are held in a list on a hidden field and accessed through a computed field that applies some checks in the setter method.

Simplified Models:

class Account(BaseModel):
    id: UUID = Field(default_factory=uuid4)
    owner: Customer
    _transactions: List[Transaction] = list()

    @fields.computed_field()
    @property
    def transactions(self) -> List[Transaction]:
        return self._transactions

    @transactions.setter
    def transactions(self, list_: List[Transaction]) -> None:
        # DO SOME CHECKS
        self._transactions = sorted_list

class Transaction(BaseModel):
    id: UUID = Field(default_factory=uuid4)
    date: datetime.date = Field(frozen=False)
    amount: PositiveFloat = Field(frozen=False)

class Customer(BaseModel):
    id: UUID = Field(default_factory=uuid4)
    name: datetime.date = Field(frozen=False)

Serialization

I can successfully serialize an account to JSON, including the transactions:

x = my_account.model_dump_json()
x

Formatted output:

'{
    "id":"1eeed00b-3bd7-48fa-ab6a-6979618ef723",
    "owner":{      
        "id":"0cad242dad03-492b-9c6b-86f0b75f9c00",
        "name":"Bob"
    },
    "transactions":[
    {"id":"42738f1d-e998-4add-94b8-713afe25b525","date":"2012-01-01","amount":1.0}
  ]
}'

De-serialization (unexpected behavior)

However, when I try to reconstruct the model from the JSON string, the transactions field is left empty. No issue with the owner field, which holds the nested Customer model:

Account.model_validate_json(x)

Formatted output:

Account(
    id=UUID('1eeed00b-3bd7-48fa-ab6a-6979618ef723'), 
    owner=Customer(
        id=UUID('0cad242d-ad03-492b-9c6b-86f0b75f9c00'), 
        name='Bob'
    ), 
    transactions=[]
)

Is there a way to make the serialization process for a computed field reversible?