--quality-delta gates a clean tree on Python: 58 dead-code false positives for Django test hooks dispatched via cls./self.

[hnipps]

What is wrong

On a Django monolith (~7,100 tracked files, Python), ripwire . --quality-delta on a clean working tree (identical to git HEAD, only untracked directories present) reports 58 gating regressions, all kind="dead-code", all preexisting-worse, exit 2:

<quality-delta baseline="git-HEAD" regressions="58" minor="0" acked="0" stale="0"
  preexisting-worse="58" new-symbol="0" gating="58" register-macro-excluded="0"
  api-new-surface="0" at="a2c192068+dirty+shallow">
<r kind="dead-code" sym="api_v2/tests/test_boundary_meter_status_logs.py::BoundaryMeterStatusLogTests::create_test_objects" p="api_v2/tests/test_boundary_meter_status_logs.py:82" gating="1" .../>
<r kind="dead-code" sym="api_v2/tests/test_boundary_meter_status_logs.py::BoundaryMeterStatusLogTests::detail_url_kwargs" p="api_v2/tests/test_boundary_meter_status_logs.py:99" gating="1" .../>
...

Two problems stacked:

1. A no-op diff should never gate. Working tree == HEAD, yet the baseline snapshot and the working-tree snapshot disagree on whether these symbols are dead. Whatever differs between the two parses (ordering, resolver tie-breaks, shallow-clone history) makes the exit code unreliable as a pre-commit hook on this repo. --quality-delta=HEAD~1..HEAD on a one-file test commit also reports 28 preexisting-worse dead-code rows in unrelated files.

2. The symbols are not dead. They are Django test-class hooks dispatched dynamically. A base class in api_v2/tests/utils.py does, inside setUpTestData:

   cls.create_test_objects()

   and ~30 test classes each define their own create_test_objects. Likewise detail_url_kwargs is called as self.detail_url_kwargs(...) from base helpers and overridden per subclass. The name-based resolver sees one call site with ~30 same-language definitions in other files, declines the bind, and every override reads as unreachable. This is the Python equivalent of the C++ self-registering test macro case that register_macros already exempts, but there is no Python-side hook for it.

The --callers=create_test_objects verb reflects the same blind spot (counts_floor="1", callers found only via super().create_test_objects() chains).

Environment

• ripwire 0.6.0 (Release, AppleClang 16.0.0.16000026, built_from=2d2f10e62), macOS arm64
• Repo is a shallow clone (--doctor reports shallow="1"); private, so I cannot share it, but the shape is any Django project using TestCase subclasses with @classmethod hooks overridden per class.
• --doctor: 8 checks, 7 passed (the one failure is an unrelated stale tracked PNG).

Reproduction shape

# tests/base.py
class BaseAPITest(TestCase):
    @classmethod
    def setUpTestData(cls):
        cls.create_test_objects()          # dynamic dispatch on cls
    @classmethod
    def create_test_objects(cls): ...

# tests/test_a.py .. tests/test_z.py  (many files)
class ATests(BaseAPITest):
    @classmethod
    def create_test_objects(cls): ...      # reported dead

Run ripwire . --quality-delta on a clean tree. Expected: gating="0". Observed: one gating dead-code row per override.

Workaround that works

ripwire . --quality-delta --quality-ack="Django test hooks dispatched via cls/super" --ack-only=dead-code

After that, gating="0" acked="58". But acking 58 findings to make a clean tree pass is a ratchet built on a false floor.

Suggestions

• Treat a cls.NAME() / self.NAME() call site whose name has N same-language method definitions across subclasses as reaching all of them for dead-code purposes (receiver is provably an instance/class of the hierarchy). Or, cheaper: extend .ripwire_config with a Python analogue of register_macros naming method names to exempt from dead-code.
• Whatever the resolver decides, --quality-delta against an identical tree should be deterministic and report zero. That part looks like a bug independent of the dispatch question.