Seems like assert_configuration() (and presumably also assert_reading()) are broken for numpy array types. I guess they probably were already broken before these changes, but one of the tests in dodal uses assert_configuration and I encountered this as I'm currently in the process of going through dodal and mx-bluesky to test out this PR.

[coretl]

          Seems like `assert_configuration()` (and presumably also `assert_reading()`) are broken for numpy array types. I guess they probably were already broken before these changes, but one of the tests in dodal uses assert_configuration and I encountered this as I'm currently in the process of going through dodal and mx-bluesky to test out this PR.

I will keep you posted if I find any other issues.

Originally posted by @rtuck99 in #594 (comment)

I suggest that we move the assert functions into a new ophyd_async.testing package, make pytest a dependency of it, then use pytest.approx on the value

[coretl]

The other alternative is to document that you should use pytest.approx in your test code when testing for arrays:

async def test_assert_reading_array(mock_signal_array: SignalRW):
    set_mock_value(mock_signal_array, [1, 2, 4.5])
    dummy_reading = {
        "mock_signal": Reading(
            {"alarm_severity": 0, "timestamp": ANY, "value": pytest.approx([1, 2, 4.5])}
        )
    }
    await assert_reading(mock_signal_array, dummy_reading)

@rtuck99 do you prefer moving ophyd_async.core.asset_* to ophyd_async.testing.assert_* and adding the pytest approx there, or writing docs that tell people to put pytest.approx in their code?