Is there a way to fill empty strings / np.nan

[ahuang11]

This one errors

import xarray as xr
import numpy as np
xr.DataArray(["", np.nan, "a", "b"]).bfill("dim_0")

~/anaconda3/lib/python3.8/site-packages/bottleneck/slow/nonreduce_axis.py in push(a, n, axis)
     49     elif ndim == 0:
     50         return y
---> 51     fidx = ~np.isnan(y)
     52     recent = np.empty(y.shape[:-1])
     53     count = np.empty(y.shape[:-1])

TypeError: ufunc 'isnan' not supported for the input types, and the inputs could not be safely coerced to any supported types according to the casting rule ''safe''

This one just does astype(str)

import xarray as xr
import numpy as np
xr.DataArray(["", np.nan, "a", "b"]).fillna("a")

[max-sixty]

I think this may be a bottleneck issue — specifically we use it for bfill but not fillna and it doesn't support strings. Potentially we could disable bottleneck for string types.

We recently added an option for disabling bottleneck, if you want a short-term workaround. (Though can't quite remember whether it was released yet).

[ahuang11]

I tried bfill and ffill with bottleneck off:

~/anaconda3/envs/py3/lib/python3.7/site-packages/xarray/core/missing.py in bfill(arr, dim, limit)
    430     if not OPTIONS["use_bottleneck"]:
    431         raise RuntimeError(
--> 432             "bfill requires bottleneck to be enabled."
    433             " Call `xr.set_options(use_bottleneck=True)` to enable it."
    434         )

RuntimeError: bfill requires bottleneck to be enabled. Call `xr.set_options(use_bottleneck=True)` to enable it.

[keewis]

right, we considered what to do with that and decided that silently using bottleneck functions with use_bottleneck=False is inconsistent. Once we have a different implementation (numbagg? or a numpy function?) this might change.