testing.md

Testing

Your application's tests

# conftest.py
import pytest, aggify
from aggify.testing import temporary_database

@pytest.fixture(scope="session", autouse=True)
def mongo():
    with temporary_database("mongodb://localhost:27017") as name:     # random db, dropped at the end
        yield name

@pytest.fixture(autouse=True)
def clean(mongo):
    for model in (User, Order):
        model.drop_collection()

Use pytest-asyncio (asyncio_mode = "auto") for async tests. A throw-away MongoDB for CI:

services:
  mongo:
    image: mongo:7
    ports: ["27017:27017"]
    options: >-
      --health-cmd "mongosh --quiet --eval 'db.runCommand({ping:1})'"

(add --replSet rs0 + rs.initiate() if you test transactions).

Count and inspect queries

from aggify.testing import assert_num_queries, capture_commands, assert_pipeline, assert_query

def test_no_n_plus_one():
    with assert_num_queries(1):
        list(Post.objects.select_related("owner"))

def test_what_was_sent():
    with capture_commands() as log:
        Post.objects.filter(published=True).count()
    assert log.names == ["aggregate"] or log.names == ["count"]
    print(log.records[0].command)

def test_compiled_form_without_a_database():
    assert_query(Post.objects.filter(published=True), filter={"published": True})
    assert_pipeline(Post.objects.filter(owner__country="DE"), [
        {"$lookup": {...}}, {"$unwind": {...}}, {"$match": {"_j_owner.country": "DE"}}, {"$unset": ["_j_owner"]},
    ])

qs.query / pipeline.to_list() need no connection at all, so query-building logic can be unit-tested offline.

How Aggify tests itself

• Unit tests (no database): fields, lookups, Q, expressions, pipeline stages, update building, security rules.
• Integration tests against a real replica set: CRUD, joins, grouping, transactions, signals, indexes (and that queries really use them), multiple event loops and threads, 40 concurrent requests...
• Differential tests (tests/integration/test_differential.py): Hypothesis generates random documents and random Q trees (AND/OR/NOT over 20+ lookups); the same predicate is evaluated by MongoDB through Aggify and by a small Python reference implementation - results must be identical, and exclude() must be the exact complement.
• Property tests for security: no string field can ever emit a $-keyed value; escaped regexes match literally.
• Framework tests: a real FastAPI app over ASGI (including injection attempts and transactions), an in-process grpc.aio server, Django views (sync and async), forms, pagination, management command, middleware.
• The runnable documentation examples are executed too.

make mongo          # throw-away replica set in Docker on :27018
make test           # pytest
make test-all       # + lint, mypy, 3.9 - 3.13 via tox

Set AGGIFY_TEST_MONGO_URI to use another server; tests marked mongo skip themselves when none is reachable and tests marked replica skip on a standalone server.