[VicSheCodes]

Why

A router test needs a node that answers an error status with a body that the simulator cannot build today, for example a 503 whose body holds both result and error. mode: "error" builds its own error envelope, and the canned body of a responses entry took a 2xx status only. A test in smart-router-automation checks that such a reply reaches the caller unchanged, and it waits on this change.

What changed

• provider_simulator/control_api.py: the canned body of a responses entry takes a status from 200 to 299, as before, and an error status from 400 to 599. A 1xx (not a final reply), a 3xx (neither a success nor an error) and a status that is not an int are still refused with HTTP 400.
• provider_simulator/listeners/base.py: the history row of a canned reply follows its status: success inside 2xx, error outside it. The body is never read for the label. JSON-RPC and GraphQL serve the canned reply, so both follow the rule.
• The docstrings of jsonrpc.py, graphql.py and Listener.build_body_override, the comment above the status check, and docs/using_the_simulator.md state the rule.
• tests/test_simulator_per_method_overrides.py: a canned body with status 400, 503 or 599 is accepted, sent with that status and the exact body, and recorded as error. Status 199, 300, 399, 600 and the text "503" are refused with HTTP 400. These two tests replace the test that refused a 500.
• tests/test_listener_graphql.py: a canned GraphQL body is labelled by its status and not by its body: a 200 that holds errors reads success, and a 503 reads error.

How to verify

pytest tests/test_simulator_per_method_overrides.py tests/test_listener_graphql.py -v

Results

• The new cases failed before the change: the control API refused 400, 503 and 599, and a canned GraphQL 503 was recorded as success.
• Each new check fails on a broken copy of the code: a label that always reads success fails 4 cases, and a status range of 200 to 599 fails the 300 and 399 cases.
• Whole suite, locally: 2546 passed. The last CI run on main had 2538; the difference is the 8 new cases. One warning came from a WebSocket test that closes a connection: a server thread hit "bad file descriptor" while it printed an error. That test does not reach the changed code, and the warning did not come back in three reruns of its class.
• black, ruff and mypy: clean.