chore: add a unittest template for testing bindings

[ryanking13]

This adds a test template for testing bindings such as r2, kv, d1.

I'll start adding tests for those bindings as a follow-up. This PR is mostly a preparation with a small smoke test using kv.

[ask-bonk]

I'm Bonk, and I've done a quick review of your PR.

This PR adds a test harness for exercising Cloudflare bindings (KV, R2, D1, etc.) by running an in-worker test suite via a live pywrangler dev server and mapping results back to pytest.

Issues found (highest severity first):

1. Test names are double-prefixed with test_ — KV_TESTS contains "test_put_and_get", but both _make_test and the setattr call add another test_ prefix, producing test_test_put_and_get as the pytest method name. See inline suggestion.

2. Each pytest test case re-runs the entire in-worker suite — _run_suite() is called independently by every generated test method. With N tests this means N full HTTP round-trips executing the same suite. Consider caching the results per suite (e.g. with a session/module-scoped fixture or functools.lru_cache).

3. Zombie process after process.kill() — In the dev_server fixture teardown, if process.wait(timeout=10) times out, process.kill() is called but process.wait() is not called afterward, which can leave a zombie process.

[ask-bonk]

I'm Bonk, and I've done a quick review of your PR.

I posted a review on PR #105 with 3 issues found:

1. Double test_ prefix on generated test names — KV_TESTS contains "test_put_and_get" but both _make_test and the setattr call add another test_ prefix, producing test_test_put_and_get. Posted a suggestion to drop the prefix from the list entries.

2. Redundant suite execution — Each generated pytest test method calls _run_suite() independently, so the full in-worker suite runs N times for N tests. Suggested caching via lru_cache or a module-scoped fixture.

3. Zombie process after process.kill() — In the fixture teardown, if process.wait(timeout=10) times out, process.kill() is called without a subsequent process.wait(), which can leave a zombie. Posted a suggestion fix.

github run

[dom96]

Interesting approach. I suppose we need to run wrangler to be able to use the bindings. I wonder if we should freeze the workerd/wrangler version we run these tests against, to ensure our CI is less susceptible to the world breaking it.