fix: fix output to JSON schema [WIP]

[tlambert03]

This is a WIP, trying to fix OME.model_json_schema() which is broken:

In [1]: OME.model_json_schema()

File ~/miniforge3/envs/ome-types/lib/python3.11/site-packages/pydantic/json_schema.py:2110, in GenerateJsonSchema._build_definitions_remapping(self)
   2107         json_ref = JsonRef(self.ref_template.format(model=defs_ref))
   2108         defs_to_json[defs_ref] = json_ref
-> 2110 return _DefinitionsRemapping.from_prioritized_choices(
   2111     self._prioritized_defsref_choices, defs_to_json, self.definitions
   2112 )

File ~/miniforge3/envs/ome-types/lib/python3.11/site-packages/pydantic/json_schema.py:175, in _DefinitionsRemapping.from_prioritized_choices(prioritized_choices, defs_to_json, definitions)
    172 # Deduplicate the schemas for each alternative; the idea is that we only want to remap to a new DefsRef
    173 # if it introduces no ambiguity, i.e., there is only one distinct schema for that DefsRef.
    174 for defs_ref, schemas in schemas_for_alternatives.items():
--> 175     schemas_for_alternatives[defs_ref] = _deduplicate_schemas(schemas_for_alternatives[defs_ref])
    177 # Build the remapping
    178 defs_remapping: dict[DefsRef, DefsRef] = {}

File ~/miniforge3/envs/ome-types/lib/python3.11/site-packages/pydantic/json_schema.py:2216, in _deduplicate_schemas(schemas)
   2215 def _deduplicate_schemas(schemas: Iterable[JsonDict]) -> list[JsonDict]:
-> 2216     return list({_make_json_hashable(schema): schema for schema in schemas}.values())

File ~/miniforge3/envs/ome-types/lib/python3.11/site-packages/pydantic/json_schema.py:2216, in <dictcomp>(.0)
   2215 def _deduplicate_schemas(schemas: Iterable[JsonDict]) -> list[JsonDict]:
-> 2216     return list({_make_json_hashable(schema): schema for schema in schemas}.values())

TypeError: unhashable type: 'dict'

the reason is our lists of objects StructuredAnnotationsList and ShapesUnion, which use fields that look like this:

Field(
            default_factory=list,
            metadata={  # type: ignore[call-arg]
                "type": "Elements",
                "choices": tuple(
                    {"name": kind.title(), "type": cls} for kind, cls in _KINDS.items()
                ),
            },
        )

the problem is choices...which contains an unhashable dict. That field is really there for xsdata though, which uses it to bind data at runtime. But it clashes with the field of the same name in the json schema.

Currently looking into how best to resolve this

[codspeed-hq]

CodSpeed Performance Report

Merging #235 will not alter performance

_{⚠️ No base runs were found}

_{Falling back to comparing fix-bugs (f4e5d7b) with main (33bcc63)}

Summary

✅ 9 untouched benchmarks

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (33bcc63) 86.38% compared to head (f4e5d7b) 86.38%.
Report is 3 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #235   +/-   ##
=======================================
  Coverage   86.38%   86.38%           
=======================================
  Files          26       26           
  Lines        1256     1256           
=======================================
  Hits         1085     1085           
  Misses        171      171           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[tlambert03]

for now, I'm fixing this problem by expressing choices as a (hashable) tuple of 2-tuples. that does not work for xsdata, but it is easy enough to cast it to a dict in our internal xsdata-pydantic-basemodel handling