FCI reader fails to load composites due to metadata issues

[ameraner]

Describe the bug
The FCI FDHSI reader fails to load a composite.

To Reproduce

sc = Scene(["/path_to_your_testdata/20130804_RC72/W_XX-EUMETSAT-Darmstadt,IMG+SAT,MTI1+FCI-1C-RRAD-FDHSI-FD--CHK-BODY--L2P-NC4E_C_EUMT_20130804115845_GTT_DEV_20130804115330_20130804115345_N__C_0072_0035.nc"], reader="fci_l1c_fdhsi")
sc.load(["ir_87"]) #success
sc.load(["fog"]) #fail

Expected behavior
Load both single datasets as well as composites successfully.

Actual results

  File "/tcenas/home/andream/PycharmProjects/dev_src/dev_src/fci_test_data/check_fci_test_data.py", line 43, in analyse_FCI
    scn.load(channel_list)
  File "/tcenas/home/andream/code/satpy_latest/satpy/satpy/scene.py", line 975, in load
    keepables = self.generate_composites()
  File "/tcenas/home/andream/code/satpy_latest/satpy/satpy/scene.py", line 888, in generate_composites
    return self._read_composites(nodes)
  File "/tcenas/home/andream/code/satpy_latest/satpy/satpy/scene.py", line 863, in _read_composites
    self._generate_composite(item, keepables)
  File "/tcenas/home/andream/code/satpy_latest/satpy/satpy/scene.py", line 838, in _generate_composite
    **self.attrs)
  File "/tcenas/home/andream/code/satpy_latest/satpy/satpy/composites/__init__.py", line 859, in __call__
    new_attrs = combine_metadata(*projectables)
  File "/tcenas/home/andream/code/satpy_latest/satpy/satpy/dataset.py", line 107, in combine_metadata
    elif all(val == values[0] for val in values[1:]):
  File "/tcenas/home/andream/code/satpy_latest/satpy/satpy/dataset.py", line 107, in <genexpr>
    elif all(val == values[0] for val in values[1:]):
  File "/tcenas/home/andream/anaconda3/envs/trolldev_latest/lib/python3.7/site-packages/xarray/core/common.py", line 120, in __bool__
    return bool(self.values)
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

Environment Info:

• Linux
• 0.20.1.dev233+g836af09c

[gerritholl]

The ancillary_variables attribute contains List[xarray.DataArray] (filled with the contents of data variable names listed in the original attribute). Based on initial investigations, I suspect that Satpy's combine_metadata cannot handle composites that combine different inputs that each have such an attribute. The FCI reader can handle the single-channel composite ir108_3d fine:

sc = Scene(["/media/nas/x21308/2020_04_MTG_unofficial_Testdata/20130804_RC72/W_XX-EUMETSAT-Darmstadt,IMG+SAT,MTI1+FCI-1C-RRAD-FDHSI-FD--CHK-BODY--L2P-NC4E_C_EUMT_20130804115845_GTT_DEV_20130804115330_20130804115345_N__C_0072_0035.nc"], reader="fci_l1c_fdhsi")
sc.load(["ir108_3d"])

As far as I know, the only other readers providing ancillary_variables are nwcsaf-nc and nucaps, which are both for level-2 data, and their only "composites" do not combine multiple variables that each have their own ancillary_variables attribute.

[gerritholl]

This regression was almost certainly caused by #1177.

[gerritholl]

It appears that combine_metadata does have a branch for arrays, but doesn't enter this because the ancillary_variables are not ndarrays but xarray.DataArrays:

satpy/satpy/dataset.py

Lines 97 to 108 in 836c657

	# combine all of the dictionaries
	shared_info = {}
	for k in shared_keys:
	values = [nfo[k] for nfo in info_dicts]
	any_arrays = any([isinstance(val, np.ndarray) for val in values])
	if any_arrays:
	if all(np.all(val == values[0]) for val in values[1:]):
	shared_info[k] = values[0]
	elif 'time' in k and isinstance(values[0], datetime) and average_times:
	shared_info[k] = average_datetimes(values)
	elif all(val == values[0] for val in values[1:]):
	shared_info[k] = values[0]

It would seem that this check needs to be replaced by some check testing positive for anything that supports the array interface, whether it's ndarray, xarray.DataArray, dask.array, pandas.DataFrame, etc.

[gerritholl]

I'll make an attempt at a fix.