xr.merge fails when passing dict

[mathause]

Code Sample

import xarray as xr

a = xr.DataArray([1, 2], dims=('foo', ), name='a')
b = xr.DataArray([3], dims=('bar', ), name='b')

objects = dict(a=a, b=b)
xr.merge(objects)

Problem description

This returns an AttributeError, while the documentation states that a dict is a valid input.

def merge(objects, compat='no_conflicts', join='outer'):
    """Merge any number of xarray objects into a single Dataset as variables.

    Parameters
    ----------
    objects : Iterable[Union[xarray.Dataset, xarray.DataArray, dict]]
        Merge together all variables from these objects. If any of them are
        DataArray objects, they must have a name.

...

Expected Output

Either the docs or the dict_like_objects loop should be changed.

Output of xr.show_versions()

INSTALLED VERSIONS ------------------ commit: None python: 3.7.1 | packaged by conda-forge | (default, Feb 18 2019, 01:42:00) [GCC 7.3.0] python-bits: 64 OS: Linux OS-release: 4.4.138-59-default machine: x86_64 processor: x86_64 byteorder: little LC_ALL: None LANG: en_GB.UTF-8 LOCALE: en_GB.UTF-8 libhdf5: 1.10.4 libnetcdf: 4.6.2

xarray: 0.12.1
pandas: 0.24.2
numpy: 1.16.2
scipy: 1.2.1
netCDF4: 1.5.0.1
pydap: None
h5netcdf: 0.7.1
h5py: 2.9.0
Nio: None
zarr: None
cftime: 1.0.3.4
nc_time_axis: 1.2.0
PseudonetCDF: None
rasterio: 1.0.22
cfgrib: 0.9.6.2
iris: 2.2.0
bottleneck: 1.2.1
dask: 1.1.5
distributed: 1.26.1
matplotlib: 3.0.3
cartopy: 0.17.0
seaborn: 0.9.0
setuptools: 41.0.0
pip: 19.0.3
conda: None
pytest: 4.4.0
IPython: 7.4.0
sphinx: 2.0.1

[shoyer]

You can pass an iterable of dicts into merge, not a single dict, e.g.,

>>> xr.merge([objects])
<xarray.Dataset>
Dimensions:  (bar: 1, foo: 2)
Dimensions without coordinates: bar, foo
Data variables:
    a        (foo) int64 1 2
    b        (bar) int64 3

So I think this is working as expected.

[max-sixty]

+1

Though I think for these we should have better error messages.

Ideally I think we'd a) check it were an iterable, and b) check each item as processed for one of the valid types, raising with a clear message on failures. Thoughts?

[shoyer]

Though I think for these we should have better error messages.

Ideally I think we'd a) check it were an iterable, and b) check each item as processed for one of the valid types, raising with a clear message on failures. Thoughts?

Yes, absolutely!

[mathause]

Yes, you are right - I did not take the Iterable[...] part in the docs seriously enough.

Would it make sense to pack them?

if isinstance(objects, (xr.DataArray, xr.Dataset, dict)):
    objects = [objects]

Because the alternative is to raise and then the user has to do [objects] himself, which seems a bit nonsensical to me. Although it is also nonsensical to pass a single DataArray or a Dataset to xr.merge, in contrast to dict.

Then we would need to check isinstance(obj, (xr.DataArray, xr.Dataset, dict)) in the in the dict_like_objects-loop and if not raise a ValueError?

[max-sixty]

I'm trying to think through if there would ever be ambiguity between packed and as specified; I think it's fine.

One note is that I'm not sure it's that helpful though. In the case listed above, it's just as good to put them in a Dataset:

In [2]: xr.merge([objects])
Out[2]:
<xarray.Dataset>
Dimensions:  (bar: 1, foo: 2)
Dimensions without coordinates: bar, foo
Data variables:
    a        (foo) int64 1 2
    b        (bar) int64 3

In [3]: xr.Dataset(objects)
Out[3]:
<xarray.Dataset>
Dimensions:  (bar: 1, foo: 2)
Dimensions without coordinates: bar, foo
Data variables:
    a        (foo) int64 1 2
    b        (bar) int64 3