Ambient context and testing

The unscoped namespace and the isolation helper.

context

nodrill.context

An ambient attribute namespace, isolated per thread and per asyncio task.

Set an attribute anywhere, read it anywhere below. Backed by its own ContextVar, so a new thread starts empty and a new task inherits a snapshot then diverges.

context.request_id = "r-42"
...
log(f"[{context.request_id}]")

Writes are unscoped, so nothing restores them at the end of a block, and a value set in a pooled worker thread survives into that thread’s next job. Use provider() for anything scoped.

The object deliberately exposes no methods, only dunders, so no attribute name your application chooses can collide with the API. These are the supported operations.

context.name

Read. Raises AttributeError listing the names currently set if it was never assigned.

context.name = value

Write.

del context.name

Delete. Raises AttributeError if the name is not set.

"name" in context

Membership test, without raising.

iter(context), len(context)

The names currently set, in sorted order, and how many there are. Dunders cannot collide with an application’s attribute names, which is why these are safe to have and keys() is not.

isolate

nodrill.isolate()

Run a block against fresh context state, restoring the outer state on exit. Returns a context manager.

Inside the block no providers are active and the ambient namespace is empty. Registrations made with set_default() inside the block are rolled back on exit. Registrations made outside stay visible inside, since they are configuration rather than state. Refs built with ref() inside the block are rolled back the same way, so a path one test made up on purpose is not a path the next test’s resolve_refs() has to import.

Intended for test fixtures.

conftest.py

import nodrill
import pytest


@pytest.fixture(autouse=True)
def _clean_context():
    with nodrill.isolate():
        yield