A sensor must not contradict the page it sits on

[zios0007]

Reported by @billjes in #1043, on an Amcrest NV5232E-16P running 1.3.0-rc.1:

on both channels the Diagnostic firmware/serial sensors show the recorder's values (4.000.00AC000.0.R, AMR044032F4F06B33E), while Device info shows each camera's own firmware (2.460.AC01.0.R) and serial.

Two answers to one question, a few centimetres apart on the same page.

The page was right. #1044 made a recorder channel's device page describe the camera in that slot, and #1054 gave the box a page of its own to be described on. These two sensors were never moved with it.

One rule, not two copies

The condition that decides whose identity a page shows was written out twice, which is how the page and the sensors drifted apart. It is now stands_in_for_the_host() on the coordinator, defined once and read by device_info and by both sensors:

return (
    self.get_serial_number() == self.get_device_serial_number()
    and not self.uses_separate_host_device()
)

True only for the lowest channel of a device with no separate host device, which is the one page that has to describe the box, because the box has nowhere else to be described. A standalone camera is unaffected: it is that page.

The build date

get_build_date is the box's, and #1044 deliberately strips the date from the camera version it reads out of the slot table. Showing both would have paired the recorder's build date with a camera's version, which is the same disagreement one level quieter and much less likely to get reported. The attribute is now only shown alongside the box's own firmware.

What does not move

unique_id on both sensors is built from get_serial_number, the channel-suffixed bookkeeping key, and nothing here touches it. A changed unique_id orphans the entity and takes its history and its automations with it, so there is a test pinning both keys.

Tests

A new file with nine tests, and four mistakes reintroduced to check they are caught:

reintroduced	caught by
the firmware sensor reports the box on every page	test_the_firmware_sensor_reports_the_camera_on_a_channel_page
the serial sensor reports the box on every page	test_the_serial_sensor_reports_the_camera_on_a_channel_page
the box's build date beside a camera's firmware	test_the_build_date_is_dropped_when_the_firmware_is_the_cameras
the camera's value wins on the box's own page	test_the_sensors_report_the_box_on_the_page_that_is_the_box

Three existing doubles learned the new question. One of them is in test_sensor_entities.py, which the device_info double guard does not police because it never mentions device_info; it was found by grepping for the sensor classes instead of waiting for CI to say so.

test_the_serial_sensor_reports_the_device_serial_not_the_entity_key carried the rationale "every channel of one NVR is the same box and must say the same serial", which this makes false. Its surviving claim, that the value is a serial and not the _15-suffixed entity key, is still asserted under a name that matches it.

🤖 Generated with Claude Code