fix(validation): reject None in validate() and validate_json()

The non-discriminated union was built with

    reduce(or_, non_discriminated_models, None)

whose None initializer made the resulting union None | Segment | ..., so
validate(None) and validate_json("null") returned successfully instead
of raising ValidationError.

Guard the empty case instead, matching what the CLI's equivalent code in
overture-schema-cli already does. Three annotation details follow from
that: the generator is materialized into a tuple because a generator's
truthiness cannot be tested for emptiness; non_discriminated_union
widens to type[BaseModel] | UnionType | None because reduce over a
single-element tuple returns the bare model type; and model_union gets
an explicit type[BaseModel] | UnionType declaration, because mypy infers
a branch variable from its first assignment and so read the narrower
UnionType off the both-present branch.

The non-discriminated bucket is non-empty in practice: Segment is an
Annotated discriminated union rather than a class, so it fails
_can_discriminate's isinstance(model_class, type) gate.

Tests assert both directions -- the two null forms raise, and a genuine
feature still round-trips -- so a regression that broke the adapter into
rejecting everything could not pass the negative tests by accident.

Signed-off-by: Seth Fitzsimmons <seth@mojodna.net>

Author: sethfitz

packages/overture-schema-validation/changelog.d/720.bugfix.md

@@ -0,0 +1,4 @@
+`validate()` and `validate_json()` no longer accept `None`. The non-discriminated
+union was built with `reduce(or_, models, None)`, whose `None` initializer made the
+resulting union admit `None`, so `validate(None)` and `validate_json("null")`
+returned successfully instead of raising `ValidationError`.

packages/overture-schema-validation/src/overture/schema/validation/__init__.py

@@ -1,4 +1,3 @@
-from collections.abc import Generator
 from functools import reduce
 from operator import or_
 from types import UnionType
@@ -80,13 +79,14 @@ def _union_type_adapter() -> TypeAdapter:
     )
     discriminated_union: UnionType | None = _discriminated_union(discriminated_models)
 
-    non_discriminated_models: Generator[type[BaseModel], None, None] = (
+    non_discriminated_models: tuple[type[BaseModel], ...] = tuple(
         m for m in models.values() if not _can_discriminate(m)
     )
-    non_discriminated_union: UnionType | None = reduce(
-        or_, non_discriminated_models, None
+    non_discriminated_union: type[BaseModel] | UnionType | None = (
+        reduce(or_, non_discriminated_models) if non_discriminated_models else None
     )
 
+    model_union: type[BaseModel] | UnionType
     if discriminated_union and non_discriminated_union:
         model_union = discriminated_union | non_discriminated_union
     elif discriminated_union:

packages/overture-schema-validation/tests/test_schema_validation.py

@@ -247,3 +247,34 @@ def test_counterexample_validation_flat(counterexample_file: str) -> None:
     assert not is_valid, (
         f"Counterexample should have failed validation (Python): {counterexample_file}"
     )
+
+
+def test_validate_none_raises() -> None:
+    """
+    `validate(None)` must raise `ValidationError` rather than succeed. Regression test for the
+    non-discriminated union's `reduce(or_, non_discriminated_models, None)` call, which seeded the
+    reduction with `None` and so built a union that admitted `None` as valid.
+    """
+    with pytest.raises(ValidationError):
+        validate(None)
+
+
+def test_validate_json_null_raises() -> None:
+    """`validate_json("null")` must raise `ValidationError` rather than succeed."""
+    with pytest.raises(ValidationError):
+        validate_json("null")
+
+
+def test_validate_still_accepts_a_genuine_feature() -> None:
+    """
+    Paired with `test_validate_none_raises` and `test_validate_json_null_raises`: confirms the
+    union still validates a genuinely valid feature, so a regression that broke the adapter into
+    rejecting everything could not pass those tests by accident.
+    """
+    example_file = EXAMPLES_DIR / "buildings" / "empire-state-building.json"
+    json_input = load_example_file(str(example_file))
+
+    model = validate_json(json.dumps(json_input))
+    assert (
+        model.model_dump(exclude_unset=True, by_alias=True, mode="json") == json_input
+    )