Tango improved descriptor

[burkeds]

This improves the descriptor returned by Tango devices. If specified in the device server, the DataKey will now include limits including the Tango RDS range, choices, dims, precision, and units.

An important note here is that we are still treating Tango commands with input/output types as SignalRW instead of SignalX. Tango commands have limited configuration information so their DataKeys can only carry the keys object_name, source, dtype, and shape. Applications using the document stream must be aware of this and not try to, for example, read the units of a SignalRW backed by a Tango command.

[coretl]

Just wondering if you could make any use of

ophyd-async/src/ophyd_async/core/_signal_backend.py

Lines 180 to 195 in 90209e1

	def make_datakey(
	datatype: type[SignalDatatypeT],
	value: SignalDatatypeT,
	source: str,
	metadata: SignalMetadata,
	) -> DataKey:
	"""Make a DataKey for a given datatype."""
	dtn = _datakey_dtype_numpy(datatype, value)
	return DataKey(
	dtype=_datakey_dtype(datatype),
	shape=_datakey_shape(value),
	# Ignore until https://github.com/bluesky/event-model/issues/308
	dtype_numpy=dtn.descr if len(dtn.descr) > 1 else dtn.str, # type: ignore
	source=source,
	**metadata,
	)

Like we do for EPICS:

ophyd-async/src/ophyd_async/epics/core/_p4p.py

Lines 391 to 396 in 90209e1

	async def get_datakey(self, source: str) -> DataKey:
	value = await context().get(self.read_pv)
	metadata = _metadata_from_value(self.converter.datatype, value)
	return make_datakey(
	self.converter.datatype, self.converter.value(value), source, metadata
	)

[burkeds]

Does this make the datakey from the signal itself? Is this because an EPICS PV does not easily provide this information? It doesn't look like this enforces that the signal datatype match the PV. With Tango, the descriptor can simply return information from the device server and check that the type matches.

[coretl]

It uses the signal value for 2 reasons:

1. To get the dtype, both the loose json form ('integer`) and the numpy datatype ('<i8')
2. To get the shape, needed especially for waveforms

The first could be got from the EPICS metadata, but the second requires the current value

[burkeds]

@coretl I am trying to use this now. Is make_datakey only intended to be used for attributes? Commands of course have the None datatype and value which make_datakey can't handle.

[tacaswell]

It is better to provide no dims rather than dummy place holder dims.

The current primary consumer of this is generating xarrays from the collected data which can leverage dims if they exist (which in turn helps make sure that array operations are correctly aligned rather than blindly zipping by dimension index). If dims aren't provided xarray will auto-generate them at the end. Providing dummy dims reduces the value of intentionally provided ones.

[burkeds]

It is better to provide no dims rather than dummy place holder dims.

The current primary consumer of this is generating xarrays from the collected data which can leverage dims if they exist (which in turn helps make sure that array operations are correctly aligned rather than blindly zipping by dimension index). If dims aren't provided xarray will auto-generate them at the end. Providing dummy dims reduces the value of intentionally provided ones.

Thanks @tacaswell, I removed the dummy dims. Tango devices won't have dims in the descriptor by default.