fix: use shape of grouping and not shape of grouping + shape of signal - #262

Here's the instrument/detector_panel_0:

DataGroup(sizes={'x_pixel_offset': 1280, 'y_pixel_offset': 1280, 'face': 6, 'vertex': 8, 'winding_order': 24, 'time': 0, 'event_time_zero': 1}, keys=[
    data: DataArray({'x_pixel_offset': 1280, 'y_pixel_offset': 1280}),
    depends_on: TransformationChain(parent='/entry/instrument/detector_panel_1', value='/entry/instrument/detector_panel_1/transformations/orientation', transformations=DataGroup(sizes={'time': 0}, keys=[
    /entry/instrument/detector_panel_1/transformations/orientation: Transform({}),
    /entry/instrument/detector_panel_1/transformations/axis6: Transform({'time': 0}),
    /entry/instrument/detector_panel_1/transformations/axis5: Transform({'time': 0}),
    /entry/instrument/detector_panel_1/transformations/axis4: Transform({'time': 0}),
    /entry/instrument/detector_panel_1/transformations/axis3: Transform({'time': 0}),
    /entry/instrument/detector_panel_1/transformations/axis2: Transform({'time': 0}),
    /entry/instrument/detector_panel_1/transformations/axis1: Transform({'time': 0}),
    /entry/instrument/detector_panel_1/transformations/stageZ: Transform({'time': 0}),
])),
    detector_number: Variable({'x_pixel_offset': 1280, 'y_pixel_offset': 1280}),
    efu_publication: DataGroup(4, {}),
    geometry: DataGroup(3, {'face': 6, 'vertex': 8, 'winding_order': 24}),
    transformations: DataGroup(8, {'time': 0}),
    x_pixel_offset: Variable({'x_pixel_offset': 1280}),
    y_pixel_offset: Variable({'y_pixel_offset': 1280}),
    z_pixel_offset: Variable({'event_time_zero': 1}),

The only issue I see there is: z_pixel_offset: Variable({'event_time_zero': 1}),

Okay I think I got the gist of the problem now.

• The axis attribute on z_pixel_offset tells us that this is a dimension of the NXdetector.
• The signal in the NXdetector does not have this dimension, it has the dimensions of detector_numbers and its own intrinsic dimensions (for event data that is event_time_zero).
• For the result to be a DataArray the data field needs to have all dimensions of all coordinates.
• Then to fix this either we have to exclude the z_pixel_offset coordinate from the resulting DataArray or we need to return a DataGroup instead of a DataArray if the dimensions of the signal and the dimensions of the detector don't match.

@SimonHeybrock do you have an opinion about the best approach here?
I think dropping the z_pixel_offset coordinate makes most sense.

@SimonHeybrock do you have an opinion about the best approach here? I think dropping the z_pixel_offset coordinate makes most sense.

The coord can be there, if it is a scalar, i.e., the length-1 dimension is squeezed out. But isn't the issue arising even before the data array assembly is attempted?

But isn't the issue arising even before the data array assembly is attempted?

It does, because the NXdetector code assumes that the dimensions of the signal field contains all dimensions found in the NXdetector group. But to fix it we need to know how extra dimensions should be handled, that is, if they should be dropped or included.

How about this?

• If the coordinate is length 1 it's treated as a scalar and the dimension is removed.
• If the coordinate is non-scalar we raise.

Just to clarify, the result of loading a NXdetector should be a DataArray right?

Just to clarify, the result of loading a NXdetector should be a DataArray right?

No. It is a DataGroup, containing a DataArray. The latter contains the signal dataset as well as all its coords and masks. NXdetector may have additional bits that are no coords, those are just places in the data group.