fix(tests): wait for the attach instead of assuming it, and unify the inbound Indigo fakes (#304, #306)

[simons-plugins]

Two test-hygiene fixes. Both were found by the CI gate added in #305 — the first one within minutes of turning it on.

Closes #304. Closes #306.

#304 — four tests asserted on state they never waited for

task = asyncio.create_task(client.run())
await client.wait_connected(timeout=2)
assert client.status.endpoint_count == 0     # AttributeError: NoneType on 3.11

wait_connected is transport-level (ws_json_client, _connected_event). The attach that populates status and sets _attached lands later, on the run loop. On 3.13 the loop happened to have run it by the time the assertion executed; on 3.11 it had not.

The tests passed on 3.13 by luck. Pre-existing — confirmed failing at 913ad44 — and invisible until now precisely because no CI ran the suite and everyone develops on 3.13.

Fix

BridgeClient gains wait_attached(timeout=...), mirroring wait_connected: an asyncio.Event set at all three sites where _attached becomes True, cleared in _mark_disconnected beside the existing reset.

It follows _attached, not the attached property. That property is _attached and connected because an endpoint_map_invalid refusal leaves a live socket serving nothing — the distinction is deliberate and stays.

No assertion changed. Only what the four tests await:

-            await client.wait_connected(timeout=2)
+            await client.wait_attached(timeout=2)

×4, and nothing else in that file. A fix that needed an assertion edited would have been the wrong fix. The other 28 wait_connected sites are untouched — they test transport state and already passed on 3.11.

CI matrix restored

python-version: ["3.11", "3.13"]. 3.11 is the floor pyproject.toml declares and pylint targets; 3.13 is what jarvis runs. Both green — 3.11 for the first time.

#306 — a drifted copy of the inbound Indigo fakes

test_generic_switch.py carried its own FakeDev/FakeDevices/FakeDeviceFactory/FakeFolderFactory under a header reading "Helpers shared with DeviceSync tests". They were not shared, and had already lost the richer version's Supports* state seeding, replacePluginPropsOnServer simulation, fail_replace rollback, stateListOrDisplayStateIdChanged counting, the ADR-0009 device-group model, and real folder creation.

Fix

Authoritative versions move to a new tests/indigo_fakes.py — not fakes.py, which holds the export-side static doubles. The two families must not be conflated (one is stateful by necessity, the other deliberately isn't), and a separate module makes that structurally unmissable instead of relying on a comment banner.

Grep found two further importers reaching into test_device_sync directly — test_power_source.py and test_integration.py — both repointed.

One thing merged rather than moved: initial_states, which only the generic_switch copy had. It seeds matterButton's lastButtonEvent/pressCount, derived from neither Supports* nor the static state table; dropping it would have broken that suite.

Nothing failed once generic_switch used the richer fake. Every mechanism the copy lacked is inert for that suite — so the drift was latent, not already producing wrong answers. It would not have stayed that way.

Numbers

3873 passing on both 3.11 and 3.13. pylint 9.48, unchanged. tests/fakes.py untouched.

Net −35 lines: test_device_sync.py −304, test_generic_switch.py −80, new module +322.

Version 2026.28.8. No [release] marker.

🤖 Generated with Claude Code

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: dc4cee2d-8598-42fb-8f57-08d9db25d9ea

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}