FastAPI dependency injection fallback not restoring after Pydantic v2 upgrade

Answer

Yes – FastAPI’s dependency injection will still resolve overridden dependencies after you upgrade endpoint models to Pydantic v2, middleware sees the original request object unchanged, and OpenAPI tags remain static so you must regenerate the schema or use a separate router to reflect the fallback version.

Confirmed facts

• DI resolution is independent of the Pydantic version used by models; app.dependency_overrides works the same after a Pydantic v2 upgrade as long as the overridden dependency does not rely on v1‑specific methods.
• Middleware runs before the route handler; when a fallback endpoint is invoked via a dependency override, the original Request object is unchanged, so any request.state set by middleware persists.
• OpenAPI tags are static metadata; they do not auto‑version when you swap implementations. The generated schema must be regenerated (e.g., app.openapi()) or you must expose the fallback through a distinct APIRouter with its own prefix/tag to produce a schema that matches the Pydantic v2 models.

Steps to verify

1. Upgrade your endpoint’s Pydantic models to v2 (use model_dump() instead of .dict()).
2. Set a dependency override: app.dependency_overrides[YourDep] = lambda: fake_service.
3. Call the endpoint and assert that the fake service is used (e.g., via a mock or side‑effect).
4. Add middleware that sets request.state.test = "original" and, in the fallback endpoint, read request.state.test; the value should be unchanged.
5. Generate the OpenAPI schema before and after activating the fallback router; compare the JSON (e.g., with jsondiff) to confirm that tags and model definitions now reflect the v2 models.