Allow for BPM to read only a single position

[GamelinAl]

Solve #457

The idea is to add a new CSSparseScalarAggregator class (subclass of CSScalarAggregator) to allow for aggregator with missing values. The missing values are replaced by nan values.

ControlSystem.create_bpm_aggregators now uses this new CSSparseScalarAggregator to avoid the bug described in #457.

AI Summary

Description

A pyaml.bpm.bpm.BPM may define only one position plane, e.g. a vertical-only X-BPM (y_pos only). Such a BPM works in a simulator mode, but building the control-system mode failed as soon as it was part of a pyaml.arrays.bpm array (PyAMLException: All devices must be instances of Attribute), because a None position device was passed to the backend aggregator. Reading the BPM alone also failed (None.get()).

With this PR, a missing plane reads NaN, both for a single BPM and in BPM arrays. Grouped (parallel) reads are kept for all accessors, so mixing single-plane BPMs into an array costs nothing in performance.

Related Issue

• Fixes ControlSystem fails on BPMs with a single position plane #457

Features/issues described there are:

• bugfix: ControlSystem.create_bpm_aggregators now handles BPMs with a missing position device. It uses a new CSSparseScalarAggregator that only adds the existing devices to the backend aggregator and scatters the values back into a NaN-filled array. This keeps one grouped read per accessor (positions, h, v). Falling back to per-BPM reads would cost 2N sequential round trips: with a simulated 1 ms per Tango read, ~260 ms instead of ~1 ms for 120 BPMs + 4 XBPMs.
• bugfix: RBpmArray (control system) returns NaN for a plane without device and takes unit() from whichever device is defined.

Changes to existing functionality

• create_bpm_aggregators now returns CSSparseScalarAggregator instances instead of CSScalarAggregator. Behaviour is unchanged for BPMs with both planes; the only overhead is one array copy per read.
• CSSparseScalarAggregator is a subclass rather than a change to CSScalarAggregator, on purpose:
  • CSStrengthScalarAggregator inherits from CSScalarAggregator, adds devices through self._devs directly and uses nb_device() as the number of power supplies, so slot tracking in the base class would be bypassed or ambiguous.
  • For magnets, a missing device is a configuration error and must keep raising, not turn into silent NaN reads or skipped writes.

Testing

The following tests (compatible with pytest) were added in tests/bpm/test_bpm_controlsystem.py, using the dummy tango-pyaml backend:

• test_controlsystem_bpm_single_plane: a vertical-only BPM reads [nan, y] and reports the unit of its y device.
• test_controlsystem_bpm_array_with_single_plane_bpm: an array of two full BPMs + one XBPM builds; positions, h and v return the right values with NaN at the missing plane; all three aggregators exist and read through the grouped path.
• test_controlsystem_bpm_array_without_horizontal_plane: an array of XBPMs only returns all-NaN h and correct v.

Verify that your checklist complies with the project

• New and existing unit tests pass locally (324 passed, 6 skipped)
• Tests were added to prove that all features/changes are effective
• The code is commented where appropriate
• Any existing features are not broken (unless there is an explicit change to an existing functionality)

🤖 Generated with Claude Code