issue_comments

Do we still want to deprecate this behavior in v0.10?

Sure, that seems reasonable to me.

o I think we should keep the deprecation warning around for a while before making this behavior an error (probably in v0.10)

@shoyer - Do we still want to deprecate this behavior in v0.10?

@SpghttCd Yes, I was confused. I missed that you were writing np.zeros, so reply doesn't make much sense!

On master (and in the next release of xarray), we do indeed issue a warning here as @fmaussion points out. Unfortunately, we did support this behavior in prior releases of xarray (especially with an OrderedDict), so I think we should keep the deprecation warning around for a while before making this behavior an error (probably in v0.10).

@shoyer maybe changing the warning into an error would be the way to go now?

Note that the recommended to build data arrays is to specify the dimensions names explicitly:

In [11]: xr.DataArray(np.zeros([1, 2, 3]), {'x': [1], 'y': [1, 2], 'z': [1, 2, 3]}, dims=['x', 'y', 'z']) Out[11]: <xarray.DataArray (x: 1, y: 2, z: 3)> array([[[ 0., 0., 0.], [ 0., 0., 0.]]]) Coordinates: * y (y) int64 1 2 * z (z) int64 1 2 3 * x (x) int64 1

on master, failing to do so results in a FutureWarning (which is the important message, but gets lost in the traceback) and then in a ValueError (which might be confusing):

` In [12]: xr.DataArray(np.zeros([1, 2, 3]), {'x': [1], 'y': [1, 2], 'z': [1, 2, 3]}) /home/mowglie/.pyvirtualenvs/py3/bin/ipython:1: FutureWarning: inferring DataArray dimensions from dictionary likecoordshas been deprecated. Use an explicit list ofdims`` instead. #!/home/mowglie/.pyvirtualenvs/py3/bin/python3

ValueError Traceback (most recent call last) <ipython-input-12-501c20b8b785> in <module>() ----> 1 xr.DataArray(np.zeros([1, 2, 3]), {'x': [1], 'y': [1, 2], 'z': [1, 2, 3]})

/home/mowglie/Documents/git/xarray/xarray/core/dataarray.py in init(self, data, coords, dims, name, attrs, encoding, fastpath) 219 220 data = as_compatible_data(data) --> 221 coords, dims = _infer_coords_and_dims(data.shape, coords, dims) 222 variable = Variable(dims, data, attrs, encoding, fastpath=True) 223

/home/mowglie/Documents/git/xarray/xarray/core/dataarray.py in _infer_coords_and_dims(shape, coords, dims) 83 raise ValueError('conflicting sizes for dimension %r: ' 84 'length %s on the data but length %s on ' ---> 85 'coordinate %r' % (d, sizes[d], s, k)) 86 87 assert_unique_multiindex_level_names(new_coords)

ValueError: conflicting sizes for dimension 'y': length 1 on the data but length 2 on coordinate 'y' ```

You shouldn't be able to make a DataArray like this -- you have three dimensions on coordinates, but only one dimension on the array itself.

With xarray 0.8.2, you get an error message with your example: import xarray as xr arr = xr.DataArray(np.zeros([1, 2, 3]), {'x': [1], 'y': [1, 2], 'z': [1, 2, 3]}) ValueError: conflicting sizes for dimension 'z': length 1 on the data but length 3 on coordinate 'z' However, this error message is not quite accurate. You need to explicitly provide dims to get a sensible error: arr = xr.DataArray(np.zeros([1, 2, 3]), {'x': [1], 'y': [1, 2], 'z': [1, 2, 3]}, dims=['x']) ValueError: coordinate z has dimensions ('z',), but these are not a subset of the DataArray dimensions ['x']

I still need to test this on the dev version, but for now I will make this as an error reporting bug.