[stevevanhooser]

Item 3 of #11: brings EMPTYProvider to parity with src/ndi/+ndi/+ontology/EMPTY.m (ndi-ontology-matlab @ 8e109b0). Only EMPTY changes, in code, tests and bridge entry. No other provider or bridge entry is touched.

What changed

• Source file: the provider now reads main/empty-base.owl instead of empty-base.json.
• Format detection: it parses either format, as EMPTY.m does. If the content contains <?xml or <rdf:RDF it goes to parseXMLFormat; anything else goes to parseFunctionalFormat.
• Failover: any failure on main, whether the term is missing or the download fails, tries development/src/ontology/empty-edit.owl.
  • If the development file 404s, main's error is re-raised.
  • Any other failure there raises LookupFailed, which carries both errors (EMPTY.m:46-51).
• Matching now follows EMPTY.m:89/:160: the bare numeric id or the whole label, case-insensitive and exact, with the first match winning.
  • The old zero-padding, EMPTY_0000074 form and label-substring fallback are removed, because MATLAB has none of them.
  • Behaviour change: EMPTY:2 and partial labels no longer resolve.
• Returned fields: EMPTY:<digits>, the label, the first IAO_0000115 definition, and every exact or related synonym.
  • XML values go through unescapeXML in MATLAB's order: & before </>, leftover tags stripped, result trimmed. It is kept separate from the module's _unescape_xml, which other providers use and which gives different results.
  • Functional-syntax values are returned raw, as in MATLAB.
• Method names follow MATLAB: lookupTermOrID, performRemoteLookup, parseXMLFormat, parseFunctionalFormat, unescapeXML, plus the constants OWL_URL_MAIN and OWL_URL_DEV.

Kept from #10

• A failed download raises and is not cached.
• clearCache() resets EMPTYProvider._cache. The cache now holds each file's body, keyed by URL.

Deliberate differences (recorded in the bridge entry)

• Per-process cache: kept. EMPTY.m caches nothing, but a failure can no longer end up in the cache, so the cache is safe to keep.
• 404 detection: EMPTY.m:39 checks whether the error message contains 404 or Not Found. Python checks the HTTP status code instead, so a term containing 404 is not mistaken for a missing branch. Only which error gets raised can differ; both versions raise.

Testing

• 17 new offline tests, all with mocked downloads. They cover:
  • small OWL excerpts in RDF/XML and in functional syntax
  • lookup by id and by name
  • the matches that are no longer accepted
  • unescaping
  • every failover path: development has the term, main times out, development 404s, both fail
  • caching
• All 17 failed before the change and pass after it.
• black, ruff and pytest -m "not live" (166 passed) are clean.
• The six EMPTY cases in the shared case table pass live against GitHub raw. The suite's network probe targets EBI, which this sandbox can't reach, so I forced the probe to true and ran only -k EMPTY.
• The development branch currently 404s, so live lookups exercise only the main path. The fallback is covered by the mocked tests only.

---

Generated by Claude Code