[markthomaschurch]

Fixes #377

Summary

TM.report() raised AttributeError: 'Data' object has no attribute 'name' for any model where one Data object is carried by more than one Dataflow.

The cause is a reference cycle, Data -> carriedBy -> Dataflow -> data (DataSet) -> Data. Pydantic v2's BaseModel.__deepcopy__ registers the new instance in the memo before copying its fields, so the DataSet copy hashed a Data with no attributes yet.

Changes

• pytm/data.py: Data.__hash__ now reads its fields with getattr(..., None), so hashing a partially built instance cannot raise.
• tests/test_pydantic_models.py: two regression tests:
  • test_deepcopy_with_dataflow_cycle: calls copy.deepcopy on a Data that is part of a carriedBy ↔ Dataflow.data cycle.
  • test_report_with_data_carried_by_multiple_flows: end-to-end test of check() → resolve() → report() with one Data on two flows.

Testing

• Both new tests fail on master with the AttributeError and pass with this change.
• Full suite: pytest tests → 263 passed.

Known limitation

In the copied graph, the copied Data is stored in the copied DataSet under the hash it had while its fields were empty. Iterating the set works, but in checks on the copy return False. Report rendering doesn't depend on this; details are in the linked issue.

🤖 Generated with Claude Code

[izar]

Failing under Python 3.12:

__________________ TestData.test_deepcopy_with_dataflow_cycle __________________
self = <tests.test_pydantic_models.TestData object at 0x7232cbd26480>
def test_deepcopy_with_dataflow_cycle(self):
# Data -> carriedBy -> Dataflow -> data (DataSet) -> same Data.
# Pydantic's deepcopy memoizes the new Data before its fields are
# copied, so the DataSet copy hashes a Data with no attributes yet.
token = Data("token", classification=Classification.SECRET)
flow = Dataflow(Actor("user"), Server("web"), "login", data=token)
token.carriedBy = [flow]

    copied = copy.deepcopy(token)

    assert copied is not token
    assert copied.name == "token"
    assert copied.classification == Classification.SECRET
    copied_flow = copied.carriedBy[0]
    assert copied_flow is not flow

  assert any(d is copied for d in copied_flow.data)

E assert False
E + where False = any(<generator object TestData.test_deepcopy_with_dataflow_cycle.. at 0x7232c9849ff0>)
tests/test_pydantic_models.py:357: AssertionError
=========================== short test summary info ============================
FAILED tests/test_pydantic_models.py::TestData::test_deepcopy_with_dataflow_cycle - assert False

• where False = any(<generator object TestData.test_deepcopy_with_dataflow_cycle.. at 0x7232c9849ff0>)
  ======================== 1 failed, 262 passed in 3.71s =========================

[izar]

Needs fix.

[markthomaschurch]

Thanks for catching this. The failure comes from the pydantic version rather than Python 3.12: CI fails on every Python version because poetry.lock pins pydantic 2.12.5, and I had tested with 2.14.3.

The bug itself only happens from pydantic 2.14, whose BaseModel.__deepcopy__ puts the new instance in the memo before copying its fields. pyproject.toml allows pydantic>=2.10.0,<3.0.0, so a pip install of pytm today gets 2.14 and TM.report() raises the AttributeError. With 2.12 and 2.13 the cycle makes a second copy of the Data, and the test's d is copied check failed on that difference.

f9f6b91 changes the test to check that the flow carries a copy of the token rather than the original, which holds on every version. The test suite passes with pydantic 2.12.5, 2.13.5 and 2.14.3, and the regression test still fails without the fix on 2.14.3.

Fixed.