DRF fields, compilers and tuned profiles¶

The same DRF model fields under normal, strict-msgspec, strict-pydantic and tuned profiles; per-serializer override, field-template caching, relation batching and explicit read/write serialization.

Run with Docker¶

From the repository root:

docker compose -f examples/compose.yaml up --build --wait serializer-backends
docker compose -f examples/compose.yaml exec serializer-backends python manage.py check

The API is available on http://127.0.0.1:8106. See container setup for port overrides, required services, optional profiles, tests and data retention. The commands below run the same application directly with uv.

Run¶

From this directory (Python 3.12+ and uv):

uv venv
uv pip install --python .venv/bin/python -r pyproject.toml --group test -e ../..
uv run --no-sync python manage.py check
uv run --no-sync python manage.py migrate
uv run --no-sync uvicorn project.asgi:application --host 127.0.0.1 --port 8106

Change --host and --port as needed. For another hostname also set EXAMPLE_ALLOWED_HOSTS=localhost,127.0.0.1,your-host. The default is loopback. This project depends on this checkout through [tool.uv.sources]; it does not import another example. Its virtual environment and SQLite database are local.

Requests¶

curl -X POST http://127.0.0.1:8106/articles/ -H 'Content-Type: application/json' -d '{"title":"First","tags":[]}'
curl http://127.0.0.1:8106/articles/
curl http://127.0.0.1:8106/profile/
# Restart with a profile; profiles are never switched on a live app.
AIODRF_EXAMPLE_PROFILE=tuned uv run --no-sync uvicorn project.asgi:application --host 127.0.0.1 --port 8106
uv run --no-sync python manage.py aiodrf_inspect_serializers --serializer demo.views.Read --format json

Tests¶

uv run --no-sync pytest -q

The tests send requests through the project's ASGI application, with lifespan startup/shutdown. Database tests use an isolated test database.

Compatibility and limits¶

Profile	Behavior
normal	DRF validation and output, default settings
strict-msgspec	Eligible output uses msgspec with strict DRF parity
strict-pydantic	Eligible output uses Pydantic with strict DRF parity
tuned	Fast compiler parity, field-template caching and relation batching
tuned-clone	Tuned plus opt-in FIELD_COPY_MODE="clone"; eligible exact scalar fields avoid constructor replay
tuned-compiled	Tuned plus recursive copy plans; nested/container constructor behavior and unsupported custom copies stay on DRF

/selected-articles/ enables cached compiled field copying only for that view. PerSerializer.Meta demonstrates serializer-level selection on the explicit_backend action. These options do not change process-wide settings; the nearest serializer Meta takes precedence over view attributes.

Select a profile with AIODRF_EXAMPLE_PROFILE before starting the process. The listed profiles have request/response equivalence tests for this model. Cloning retains DRF deepcopy for custom, nested and relational fields. It requires the static field-template cache and does not modify DRF classes.

Tuned is opt-in: fast parity is not universally DRF-equivalent (notably Decimal handling). This model deliberately contains no Decimal field. Both the per-class backend and the global backend have explicit endpoints. POST re-reads saved relations, preserving database ordering and signal effects. Serialization is CPU work, not asynchronous I/O.

The endpoints are public and the settings use a development secret key: do not deploy this project as it is. See the feature guide and example catalogue for related examples.