The library changed after these were written and more changes are already
decided upstream. Records what is wrong today, what to write toward, the
one question still open, and the fast-JSON-field trap, rather than
silently leaving code that would fail on contact.
Agent review (source-verified against both repos) confirmed the spec's
claims accurate throughout, with one real gap: the JSON-subpath
tantivy-py carve-out (index.parse_query fallback for notes.*/
custom_fields.* until tantivy-py#716 ships) interacts with paperless's
pinned tantivy~=0.26.0 and wasn't mentioned. Also added two footnotes:
FieldRegistry forces date_only=True on any DATE spec regardless of the
PublicField default, and the date-grammar parity audit implicitly
grants new keyword vocabulary as a side effect.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Design for replacing _translate.py/_dates.py with whoosh-compat: shared
field-definition table driving both the Tantivy schema and the query
FieldRegistry, diagnostics->exception mapping (aggregating all errors,
not just the first), a 4-PR stack with no rollout flag, and a
result-level acceptance corpus + date-grammar parity audit as the
safety net instead.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
stumpylog
