[xgbtxy]

Fixes #2064.

What was wrong

transform_schema (src/anthropic/lib/_parse/_transform.py) had three issues on type: "array" schemas using JSON Schema Draft 2020-12 prefixItems, boolean items, or Draft-07 list items:

1. Crash on boolean items (items: False / items: True) and Draft-07 list items:

   # Before: raised TypeError: 'bool' object is not a mapping
   transform_schema({
       "type": "array",
       "prefixItems": [{"type": "integer"}, {"type": "string"}],
       "items": False,
   })

2. Unsatisfiable strict schema on open typed-tail tuples (prefixItems + dict items):
   For {"type": "array", "prefixItems": [{"type": "integer"}], "items": {"type": "string"}, "minItems": 2}, Draft 2020-12 applies items only to indices >= len(prefixItems). Previously, transform_schema emitted {"type": "array", "items": {"type": "string"}, "description": "{prefixItems: [{'type': 'integer'}], minItems: 2}"}, forcing index 0 to be a string and preventing valid [42, "ok"] outputs.
3. Unconstrained array and unreferenced $defs on Pydantic tuple[Step, int]:
   Pydantic v2 emits {"type": "array", "prefixItems": [{"$ref": "#/$defs/Step"}, {"type": "integer"}], "minItems": 2, "maxItems": 2}. Previously, strict_schema omitted "items" altogether, leaving #/$defs/Step unreferenced in $defs:

   # After:
   # properties.pair -> {
   #   "type": "array",
   #   "title": "Pair",
   #   "items": {"anyOf": [{"$ref": "#/$defs/Step"}, {"type": "integer"}]},
   #   "description": "{maxItems: 2, prefixItems: [{'$ref': '#/$defs/Step'}, {'type': 'integer'}], minItems: 2}",
   # }

What changed

• Added transform_tuple_items in src/anthropic/lib/_parse/_transform.py to lower closed tuples (items: False or maxItems <= len(prefixItems)) and open typed-tail tuples (prefixItems + dict items, as well as Draft-07 list items + additionalItems) into strict_schema["items"] (deduplicated single schema when homogeneous, {"anyOf": [...]} when heterogeneous), while keeping open untyped-tail tuples (prefixItems without items and without maxItems <= len(prefixItems), or items: True) unconstrained on items.
• Preserved non-dict items (False/True or list) and prefixItems in leftover description hints instead of passing non-dict items to transform_schema.
• Guarded minItems with not isinstance(min_items, bool) so True/False are not treated as 1/0.

Wire-level / compatibility notes

• Standard homogeneous array schemas ({"type": "array", "items": {...}}) are unchanged on the wire.
• Closed Pydantic / JSON Schema tuple arrays now emit a sound and complete "items" constraint ({"anyOf": [...]} or single schema) on the wire instead of omitting "items" or raising TypeError.

Verification

• Added unit tests in tests/lib/_parse/test_transform.py covering closed Draft 2020-12 tuples (items: False), open typed-tail tuples (prefixItems + dict items), open untyped-tail tuples, Draft-07 tuple lists, standalone items: False, and Pydantic v2 tuple[Step, int] / tuple[int, int].
• Verified with pytest -n0 tests/lib/_parse/ (40 passed), ruff check, ruff format --check, and pyright strict mode (0 errors, 0 warnings).

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.