netbox-branching 1.2.0: branch revert fails for deleted COTs and dropped fields

[bctiemann]

Summary

netbox-branching 1.2.0 (released 2026-09-02, adds NetBox v4.7 support) fixes the CI version-gate crash the "main, branching" leg was hitting under NetBox 4.7.0, but introduces two new, genuine failures in netbox_custom_objects/tests/test_branching.py: 61 tests run, 2 failures + 2 errors. Confirmed on main with no other code changes (dispatched lint-tests.yaml directly against main: https://github.com/netboxlabs/netbox-custom-objects/actions/runs/33695787237), so this is a genuine compatibility gap with 1.2.0, not specific to any open PR.

Failure 1: revert() on a deleted COT fails with a NOT NULL violation

ERROR: test_cot_deleted_in_branch_merge_and_revert (IterativeBranchingTestCase and SquashBranchingTestCase)

psycopg.errors.NotNullViolation: null value in column "cache_timestamp" of
relation "netbox_custom_objects_customobjecttype" violates not-null constraint
DETAIL:  Failing row contains (19, null, null, {}, , , doomed_cot, ...).

  File ".../netbox_branching/models/changes.py", line 189, in undo
    deserialized.save(using=using)
  File ".../django/core/serializers/base.py", line 294, in save
    models.Model.save_base(self.object, using=using, raw=True, **kwargs)

branch.revert() replays a serialized ObjectChange to recreate the deleted CustomObjectType row, via Django's deserializer -> save_base(..., raw=True). cache_timestamp is a NOT NULL column on CustomObjectType that's presumably populated by a save() override or a signal under normal circumstances, but raw=True skips those, and the serialized snapshot captured at delete time apparently didn't include a value for it -- or something in 1.2.0's undo/redeserialization path handles that column differently than 1.1.x did.

Failure 2: dropped field's column is not restored on revert

FAIL: test_single_field_delete_merge_and_revert (IterativeBranchingTestCase and SquashBranchingTestCase)

AssertionError: 'drop_me' not found in {'keep_me', 'owner_id', 'created', 'id', 'last_updated'}
: drop_me column must be restored after revert

After deleting one field from a COT inside a branch and reverting, the dropped column doesn't come back. netbox-branching 1.2.0's release notes state it "replaces django-mptt with a PostgreSQL ltree implementation and moves the maintenance of ltree and denormalized fields into database triggers," and that "branch provisioning now replicates those triggers into each branch schema" -- this rework of trigger/schema provisioning is a plausible culprit for column-level DDL not being correctly captured/replayed during revert, though this hasn't been root-caused yet.

Scope

Neither failure touches code from #687 (the get_models() reentrancy guard) or any other currently-open PR -- confirmed by reproducing both on main directly. This needs investigation against netbox-branching 1.2.0's actual behavior change (ltree/trigger migration) before a fix can be scoped.

[bctiemann]

Root-caused and fixed upstream: netboxlabs/netbox-branching#648. Both failures trace to the same line -- #647 (v1.2.0) switched ObjectChange.undo()'s non-MPTT restore path from instance.save(using=using) to deserialized.save(using=using), and the latter calls Model.save_base(..., raw=True) directly, bypassing both auto_now population (breaking CustomObjectType.cache_timestamp, which is deliberately excluded from serialize_object()) and any model-defined save() override (breaking CustomObjectTypeField.save()'s schema DDL). Full root-cause writeup and fix in the linked PR.

[bctiemann]

Resolved by upstream release: netboxlabs-netbox-branching v1.2.1 (published 2026-09-09) includes the fix from netboxlabs/netbox-branching#648, which addresses issue #650 — the ObjectChange.undo() raw-save bypass root-caused in this issue's earlier comment.

Confirmed: reran tests (main, branching) on #687 against the new release (this repo's CI pin, netboxlabs-netbox-branching>=1.0.4,<2.0.0, already covers v1.2.1 without any change needed) — both previously-failing scenarios (test_cot_deleted_in_branch_merge_and_revert, test_single_field_delete_merge_and_revert) now pass, and the full tests (main, branching) matrix leg is green.

Closing.

[bctiemann]

Closing as Not Planned because the fix was in netbox-branching.