fix: coerce by_alias to bool in model_dump to prevent TypeError

`model_dump()` in `_compat.py` declares `by_alias: bool | None = None`
but passes the value directly to Pydantic v2's `model_dump()`, which
expects a `bool`. When `by_alias` is `None` (the default), pydantic-core's
Rust serializer raises `TypeError: argument 'by_alias': 'NoneType' object
cannot be converted to 'PyBool'`.

This only manifests when DEBUG logging is enabled, because the affected
call path is inside `_build_request()`'s `log.isEnabledFor(logging.DEBUG)`
block.

Apply `bool(by_alias)` (matching the existing Pydantic v1 path on line 157)
to coerce `None` → `False` safely.

Fixes #2921

Signed-off-by: Giulio Leone <6887247+giulio-leone@users.noreply.github.com>

Author: giulio-leone

src/openai/_compat.py

@@ -149,7 +149,7 @@ def model_dump(
             exclude_defaults=exclude_defaults,
             # warnings are not supported in Pydantic v1
             warnings=True if PYDANTIC_V1 else warnings,
-            by_alias=by_alias,
+            by_alias=bool(by_alias),
         )
     return cast(
         "dict[str, Any]",

tests/test_models.py

@@ -157,7 +157,15 @@ def test_unknown_fields() -> None:
     assert model_dump(m2) == {"foo": "foo", "unknown": {"foo_bar": True}}
 
 
-def test_strict_validation_unknown_fields() -> None:
+def test_model_dump_by_alias_none() -> None:
+    """Ensure model_dump does not crash when by_alias defaults to None (GH-2921)."""
+    m = BasicModel.construct(foo="hello")
+    # by_alias=None is the default — must not raise TypeError
+    result = model_dump(m)
+    assert result == {"foo": "hello"}
+    # Explicit by_alias=None must also work
+    result = model_dump(m, by_alias=None)
+    assert result == {"foo": "hello"}
     class Model(BaseModel):
         foo: str