load_entity: an empty result is not a list, and resolve() pairs IRIs with the wrong entities

[LukasGold]

Two defects in how the result of OSW.load_entity is shaped and consumed. Both produce wrong results.

1. An empty result is an OswBaseModel, not a list

LoadEntityResult.entities is typed Union[model.OswBaseModel, List[model.OswBaseModel]]:

osw-python/src/osw/core.py

Lines 1280 to 1284 in d7f0116

	class LoadEntityResult(BaseModel):
	"""Result of load_entity()"""
	entities: Union[model.OswBaseModel, List[model.OswBaseModel]]
	"""The dataclass instance(s)"""

pydantic v1 tries the first member of the Union first. BaseModel.validate([]) succeeds, because it falls back to dict([]), which is {}. So when no page yields an entity, the result is an empty model:

>>> r = OSW.LoadEntityResult(entities=[])
>>> r.entities
OswBaseModel()
>>> bool(r.entities)
True
>>> r.entities[0]
TypeError: 'OswBaseModel' object is not subscriptable
>>> len(r.entities)
TypeError: object of type 'OswBaseModel' has no len()

A non-empty list is not affected. dict() needs each element to yield exactly two items, and an entity yields one item per field (19 for a plain Item), so validation falls through to List[...]. Checked with lists of one and two entities.

Affected caller: export_entity_jsonld wraps a non-list result in a list, so its if not entities check never fires:

osw-python/src/osw/service/ops/entities.py

Lines 88 to 95 in d7f0116

	result = ctx.osw.load_entity(
	OSW.LoadEntityParam(titles=[title], autofetch_schema=True)
	)
	entities = result.entities
	if not isinstance(entities, list):
	entities = [entities]
	if not entities:
	raise errors.NotFound(f"Entity '{title}' not found.")

For a title that yields no entity, it therefore passes [OswBaseModel()] to export_jsonld instead of raising NotFound. This part follows from the code; I did not run it end to end.

Suggested fix: type the field as List[model.OswBaseModel], or at least put List[...] first in the Union. load_entity always passes a list.

2. resolve() pairs IRIs with the wrong entities

The oold backend's resolve() pairs request.iris with load_entity(...).entities by position:

osw-python/src/osw/core.py

Lines 265 to 276 in d7f0116

	def resolve(self, request: ResolveParam):
	# print("RESOLVE", request)
	osw_obj: OSW = self.osw_obj
	entities = osw_obj.load_entity(
	OSW.LoadEntityParam(titles=request.iris)
	).entities
	# create a dict with request.iris as keys and the loaded entities as values
	# by iterating over both lists
	nodes = {}
	for iri, entity in zip(request.iris, entities):
	nodes[iri] = entity
	return ResolveResult(nodes=nodes)

load_entity skips a page it cannot build, both when a schema is missing and when construction fails:

osw-python/src/osw/core.py

Lines 1387 to 1388 in d7f0116

	if not schemas_fetched:
	continue

osw-python/src/osw/core.py

Lines 1412 to 1426 in d7f0116

	if entity is not None:
	# make sure we do not override existing metadata
	if not hasattr(entity, "meta") or entity.meta is None:
	entity.meta = model.Meta()
	if (
	not hasattr(entity.meta, "wiki_page")
	or entity.meta.wiki_page is None
	):
	entity.meta.wiki_page = model.WikiPage()
	entity.meta.wiki_page.namespace = namespace_from_full_title(
	page.title
	)
	entity.meta.wiki_page.title = title_from_full_title(page.title)
	entities.append(entity)

The list is then shorter than the titles, and every later IRI is paired with the entity of the following page. Reproduced offline, with a page that has no jsondata followed by a valid page:

titles:               ['Item:OSWAlignBad', 'Item:OSWAlignGood']
entities:             ['good']
zip as in resolve():  {'Item:OSWAlignBad': 'good'}

A reference to the first IRI resolves to the second entity, and the second IRI resolves to nothing.

Suggested fix: key the result by page title instead of by position. load_entity already sets entity.meta.wiki_page.namespace and .title on every entity it returns.

Related

#166 changes how load_entity chooses a class, and does not change either behaviour above.