Fix JSONSchema serialization inconsistency with alias field

[veeceey]

Summary

Fixes #310 - JSONSchema serialization returns null for schema field when model_dump() is called without by_alias=True.

Problem

The JSONSchema.model_serializer was attempting to retrieve the schema_definition field using its alias "schema", but when by_alias=False, the serialized dict contains the field name "schema_definition" instead. This caused the schema field to be set to None incorrectly.

Solution

Updated the serializer to check both the alias key and the field name when retrieving values, ensuring consistent behavior regardless of the by_alias parameter.

Changes

• Modified JSONSchema.serialize_model to try both alias and field name when getting values
• Added comprehensive test cases to verify serialization works with and without by_alias
• Ensured both serialization methods produce identical results

Test Results

$ python3 -m pytest tests/test_jsonschema_serialization.py -v
============================= test session starts ==============================
platform darwin -- Python 3.14.2, pytest-9.0.2, pluggy-1.6.0
cachedir: .pytest_cache
rootdir: /Users/vc/open source/client-python
configfile: pyproject.toml
plugins: anyio-4.12.1, mock-3.15.1, asyncio-1.3.0
collecting ... collected 3 items

tests/test_jsonschema_serialization.py::test_jsonschema_serialization_without_by_alias PASSED [ 33%]
tests/test_jsonschema_serialization.py::test_jsonschema_serialization_with_by_alias PASSED [ 66%]
tests/test_jsonschema_serialization.py::test_jsonschema_serialization_consistency PASSED [100%]

========================= 3 passed, 1 warning in 0.26s =========================

All existing tests continue to pass:

$ python3 -m pytest tests/ -v
============================= test session starts ==============================
platform darwin -- Python 3.14.2, pytest-9.0.2, pluggy-1.6.0
cachedir: .pytest_cache
rootdir: /Users/vc/open source/client-python
configfile: pyproject.toml
plugins: anyio-4.12.1, mock-3.15.1, asyncio-1.3.0
collecting ... collected 5 items

tests/test_jsonschema_serialization.py::test_jsonschema_serialization_without_by_alias PASSED [ 20%]
tests/test_jsonschema_serialization.py::test_jsonschema_serialization_with_by_alias PASSED [ 40%]
tests/test_jsonschema_serialization.py::test_jsonschema_serialization_consistency PASSED [ 60%]
tests/test_prepare_readme.py::test_rewrite_relative_links_keeps_absolute PASSED [ 80%]
tests/test_prepare_readme.py::test_main_prints_rewritten_readme_with_defaults PASSED [100%]

========================= 5 passed, 1 warning in 0.26s =========================

Impact

This fix ensures that:

• model_dump() and model_dump(by_alias=True) produce consistent results
• Debugging and logging code that uses model_dump() will now see the correct schema
• No breaking changes - existing code using by_alias=True continues to work

[veeceey]

Manual Test Results

Environment

• Python 3.14.2, macOS 15.4
• Pydantic 2.11.4

Test 1: model_dump() without by_alias (the bug)

>>> from mistralai.models.jsonschema import JSONSchema
>>> schema = JSONSchema(name="test", schema_definition={"type": "object", "properties": {"name": {"type": "string"}}})
>>> result = schema.model_dump()
>>> result["schema"]
{'type': 'object', 'properties': {'name': {'type': 'string'}}}

Before fix: result["schema"] was None because the serializer only checked for the alias key.
Result: PASS - schema field is correctly populated.

Test 2: model_dump(by_alias=True) still works

>>> result_alias = schema.model_dump(by_alias=True)
>>> result_alias["schema"]
{'type': 'object', 'properties': {'name': {'type': 'string'}}}

Result: PASS - by_alias=True continues to work as before.

Test 3: Both serialization methods produce identical results

>>> result = schema.model_dump()
>>> result_alias = schema.model_dump(by_alias=True)
>>> result["schema"] == result_alias["schema"]
True
>>> result["name"] == result_alias["name"]
True

Result: PASS - consistent results regardless of by_alias parameter.

Test 4: Full test suite

$ python3 -m pytest tests/test_jsonschema_serialization.py -v
collected 3 items

tests/test_jsonschema_serialization.py::test_jsonschema_serialization_without_by_alias PASSED
tests/test_jsonschema_serialization.py::test_jsonschema_serialization_with_by_alias PASSED
tests/test_jsonschema_serialization.py::test_jsonschema_serialization_consistency PASSED

3 passed in 0.26s

$ python3 -m pytest tests/ -v
collected 5 items

tests/test_jsonschema_serialization.py::test_jsonschema_serialization_without_by_alias PASSED
tests/test_jsonschema_serialization.py::test_jsonschema_serialization_with_by_alias PASSED
tests/test_jsonschema_serialization.py::test_jsonschema_serialization_consistency PASSED
tests/test_prepare_readme.py::test_rewrite_relative_links_keeps_absolute PASSED
tests/test_prepare_readme.py::test_main_prints_rewritten_readme_with_defaults PASSED

5 passed in 0.26s

Result: PASS - all existing and new tests pass.

[abdelhadi703]

Hi @veeceey, thanks for working on this fix! The root cause analysis is spot on — the serialize_model method only checks the alias key when by_alias=False, which causes the schema field to be None.

The fix itself (checking both alias and field name) is clean and the tests are solid.

One concern: the modified file src/mistralai/client/models/jsonschema.py is auto-generated by Speakeasy (Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.). This means the fix would be overwritten on the next code generation run. It might be worth flagging this to the maintainers so the fix can be applied upstream in the Speakeasy template or configuration as well, ensuring it persists across regenerations.

That said, +1 on merging this as-is since it fixes a real user-facing issue. The maintainers can decide how to handle the code-gen aspect.