[jimsynz]

What was timing-based

Three tests in test/bb/pid/controller_test.exs used refute_receive {:bb, [:actuator, :base_link, :joint1, :servo], _}, 50 to stand in for "the controller didn't publish":

• "controller filters messages by type"
• "controller does not publish when only setpoint is received"
• "controller does not publish when only measurement is received"

Fifty milliseconds is about five ticks at 100 Hz, but nothing proved the inputs had been consumed or that a tick had stepped in that window. Under load the tests would pass without testing anything.

What they wait on now

A shared refute_output_after_a_step/0 helper:

1. :sys.get_state/1 on the controller, so the published inputs have been handled (BB.PubSub.publish/3 sends from the test process).
2. Attaches a [:bb, :loop, :tick] telemetry handler filtered on robot: TestRobot, path: [:test_pid]. It is attached after the sync, so any event it reports comes from a tick that saw the inputs. BB.Loop only emits on ticks with a dt, which are the ones that reach step/2.
3. assert_receives that event (1 s upper bound), then :sys.get_state/1 again so that tick's step/2 has finished. Anything it published reached the test before the reply.
4. refute_received the output, which checks the mailbox without waiting.

As a check that the helper isn't vacuous, I temporarily added a measurement to the setpoint-only test. It then failed on refute_received 3/3 times. No lib changes.

The assert_receive …, 100 calls elsewhere in the file are upper bounds on an output that must arrive, so I left them alone.

Left

Nothing.

Results

• The three converted tests, mix test test/bb/pid/controller_test.exs:174 :201 :218 --seed 0 ×10: 10/10 runs passed (3 tests each).
• mix check --no-retry: all green, before and after the change. No pre-existing failures.