Python classes: keep instance state separate from class state

Operation contract

ImportSession allocates a rejected-id list in its constructor. Two instances therefore have different lists. Adding a rejection to one session leaves the other empty. Putting that list on the class would instead provide one shared object unless the code deliberately replaced it for each instance.

Failure and ownership boundary

The class does not validate who may record a rejection or persist session state. A constructor can establish a local invariant, but a shared object used by multiple threads still needs a concurrency policy. Python thread pools: collect results and observe worker failures and Python dataclasses: frozen fields require an immutable value model use that same distinction.

Working program

class ImportSession:
    def __init__(self):
        self.rejected_ids = []

    def reject(self, receipt_id):
        self.rejected_ids.append(receipt_id)

first_session = ImportSession()
second_session = ImportSession()
first_session.reject("R41")
print(first_session.rejected_ids)
print(second_session.rejected_ids)

Output

['R41']
[]

Costs and limits

Each instance retains its own list, which grows with rejected IDs. The fixture imposes no retention limit; a long-running importer needs one.

Common Mistakes

• Do not store per-request mutable lists on the class.
• Object construction does not establish thread safety.

Connected lessons

Python dataclasses: frozen fields require an immutable value model, Python properties: validate mutation at the public attribute boundary, Python thread pools: collect results and observe worker failures.

Apply this boundary

Python descriptors: field validation and attribute precedence, Python super and method resolution: cooperate across mixins, Python Protocol: structural intent without runtime validation.

Follow the related contract

Python staticmethod and classmethod: choose the bound receiver, Python ABC: prevent construction of an incomplete implementation.