Ensure backend creates sender callbacks with the correct controller

[GDYendell]

Fixes #130

This is not the ideal fix, but tried many different things and they were all quite horrible:

• Adding the controller back into ControllerAPI - the transport should not have access to it
• Adding the ScanCallbacks to ControllerAPI - this just doesn't really make sense, only the Backend needs them
• Make the Backend manually parse methods and attribute out of the Controller and only create the ControllerAPI at the end - this was OK, but meant repeating this code multiple times.

I think the interface of the handlers may not be correct, which is making this painful. It doesn't seem necessary for the whole controller to be passed into the handler put/update methods.

I don't know why that tango test is suddenly failing on 3.12 😢 It doesn't happen on main or locally.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 90.56%. Comparing base (759fa13) to head (7622301).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #134      +/-   ##
==========================================
+ Coverage   90.52%   90.56%   +0.04%     
==========================================
  Files          41       41              
  Lines        1920     1930      +10     
==========================================
+ Hits         1738     1748      +10     
  Misses        182      182              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[evvaaaa]

It doesn't seem necessary for the whole controller to be passed into the handler put/update methods.

I agree. I think that if a controller is necessary for the Handler to function then it should be an instance member on the Handler.

[evvaaaa]

I don't know why that tango test is suddenly failing on 3.12 😢 It doesn't happen on main or locally.

Feel free to cherry pick this.

I remember having this problem in ophyd-async too, I'm looking through pytango issues now.

Edit

I was thinking of a DeprecationWarning on pytango 3.12. This problem seems more fundamental to some slow cleanup when exiting a test... I tried playing around with the fixtures to no avai.

Main still passes, so this failure is probably a consequence of the new test added here.

[evvaaaa]

GTM once CI is passing...

[jsouter]

EigerHandler.update is still receiving the top controller in fastcs-eiger with this fix, I think you also need to update _get_scan_coros to use the correct controller.

[jsouter]

Something like

def _get_scan_coros(
    root_controller_api: ControllerAPI, root_controller: Controller
) -> list[Callable]:
    scan_dict: dict[float, list[Callable]] = defaultdict(list)

    for controller_api in root_controller_api.walk_api():
        controller = root_controller.get_controller_by_path(controller_api.path)
        _add_scan_method_tasks(scan_dict, controller_api)
        _add_attribute_updater_tasks(scan_dict, controller_api, controller)

    scan_coros = _get_periodic_scan_coros(scan_dict)
    return scan_coros

[GDYendell]

Closing in preference for #135