Test with overrides¶

override_dependency replaces a factory everywhere it is used, including nested dependencies, and restores the previous state on exit, even after exceptions. Nested contexts restore the outer override correctly.

A pytest fixture per swap¶

Wrap common swaps in fixtures so tests declare intent, not mechanics:

tests/conftest.py

from collections.abc import Iterator

import pytest

from wireme import override_dependency

from myapp.dependencies import get_connection
from tests.fakes import FakeConnection


def get_fake_connection() -> Iterator[FakeConnection]:
    connection = FakeConnection()
    try:
        yield connection
    finally:
        connection.assert_all_released()


@pytest.fixture
def fake_connection() -> Iterator[None]:
    with override_dependency(get_connection, get_fake_connection):
        yield

tests/test_reports.py

def test_report_uses_the_fake(fake_connection: None) -> None:
    assert make_report() == "report against fake"

Replacement factories may have different parameter lists and any form: sync, async, generator, or async-generator. A generator replacement keeps its cleanup semantics, so the fake above can verify its own teardown.

Layered overrides¶

Overrides nest, so a test can specialize what a fixture set up:

def test_slow_database(fake_connection: None) -> None:
    with override_dependency(get_connection, get_slow_connection):
        assert make_report() == "report against slow"
    # the fixture's fake is active again here

One-off values beat context managers sometimes¶

For a single call, pass the dependency explicitly by keyword instead of overriding:

make_report(config=Settings(database_url="sqlite://:memory:"))

Process-level provider

Overrides mutate a provider shared across the process. Use them for isolated tests and application setup, not concurrent request-level mutation; parallel test processes (pytest-xdist) each get their own.