What happened:

I started with a Dataframe that represented an identity matrix and used reindex with multiple dimensions and a fill_value. The goal was to produce a Dataset from a sparse dataframe and the identity matrix was the simplest example. The fill_value in reindex is defined as "Value to use for newly missing values." What I found was that fill_values were not applied at coordinates that were in the unique set of the any coord. For a pure identity matrix, that means fill_values were not applied anywhere (all rows are present, all cols are present). When I thin the identity matrix (skipping elements), the error is more obvious. On the rows and columns that have valid input data, the fill_value is not applied.

What you expected to happen:

I expected all new nan values to be filled with the fill value.

Minimal Complete Verifiable Example:

``` import numpy as np import pandas as pd

n = 10 thin = 2

df = pd.DataFrame.from_dict([ dict(ROW=v, COL=v, LAND=1) for v in np.arange(0, n, thin) ]).set_index(['ROW', 'COL'])

ds = df.to_xarray() rds = ds.reindex(ROW=np.arange(n), COL=np.arange(n), fill_value=0)

p = rds.LAND.plot()

p.axes.set_facecolor('red')

p.axes.figure.savefig('test.png')

print(rds.LAND[:]) print(rds.LAND[::thin, ::thin]) ```

Output: <xarray.DataArray 'LAND' (ROW: 10, COL: 10)> array([[ 1., 0., nan, 0., nan, 0., nan, 0., nan, 0.], [ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], [nan, 0., 1., 0., nan, 0., nan, 0., nan, 0.], [ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], [nan, 0., nan, 0., 1., 0., nan, 0., nan, 0.], [ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], [nan, 0., nan, 0., nan, 0., 1., 0., nan, 0.], [ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], [nan, 0., nan, 0., nan, 0., nan, 0., 1., 0.], [ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]]) Coordinates: * ROW (ROW) int64 0 1 2 3 4 5 6 7 8 9 * COL (COL) int64 0 1 2 3 4 5 6 7 8 9 <xarray.DataArray 'LAND' (ROW: 5, COL: 5)> array([[ 1., nan, nan, nan, nan], [nan, 1., nan, nan, nan], [nan, nan, 1., nan, nan], [nan, nan, nan, 1., nan], [nan, nan, nan, nan, 1.]]) Coordinates: * ROW (ROW) int64 0 2 4 6 8 * COL (COL) int64 0 2 4 6 8

Anything else we need to know?:

Environment:

Previous test relied on CF generated metadata, but this test is more robust.

• [x] Addresses #3409 #3420 #3434
• [x] Tests fixed

Development machine is missing python3.6 environment, using azure to check. This should help me get this fix in faster.

PNC is used for GEOS-Chem, CAMx, CMAQ and other atmospheric data formats including NOAA arl packed bit and HYSPLIT related files. The goal is to provide access to many file types that have their own file formats and meta-data conventions. It can provide a CF compliant netCDF-like interface.

• [x] Tests were run using example datasets, but datasets were not added to repository as they are large and diverse. Only files that provide CF compliant structures will successfully open. Confirmed by @bbakernoaa
• [x] Tests added (for all bug fixes or enhancements)
• [x] Tests passed (for all non-documentation changes)
• [x] Fully documented, including whats-new.rst for all changes and api.rst for new API (remove if this change should not be visible to users, e.g., if it is an internal clean-up, or if this is part of a larger project that will be documented later)

Code Sample, a copy-pastable example if possible

```bash

Your code here

pytest -v xarray/tests/test_backends.py::NetCDF4DataTest ```

Problem description

Instead of passing all tests, this fails on the test_88_character_filename_segmentation_fault function.

E AssertionError: exception ImportWarning("can't resolve package from __spec__ or __package__, falling back on __name__ and __path__",) did not match pattern 'segmentation fault'

It oddly doesn't seem to be related to the 88 character issue. I traced this down to xarray/backends/common.py::get_scheduler where an "ImportWarning" is not being caught on line 49.

Expected Output

Passes all tests. Adding ImportWarning to the tuple of exceptions in get_scheduler fixes the problem.

Output of xr.show_versions()