• Tests added
• Passes isort . && black . && mypy . && flake8
• User visible changes (including notable bug fixes) are documented in whats-new.rst

Significantly speeds up _construct_dataarray by iterating over ._coord_names instead of .coords. This avoids unnecessarily constructing a DatasetCoordinates object and massively speeds up repr construction for datasets with large numbers of variables.

Construct a 2000 variable dataset

import numpy as np
import xarray as xr

a = np.arange(0, 2000)
b = np.core.defchararray.add("long_variable_name", a.astype(str))
coords = dict(time=np.array([0, 1]))
data_vars = dict()
for v in b:
    data_vars[v] = xr.DataArray(
        name=v,
        data=np.array([3, 4]),
        dims=["time"],
        coords=coords
    )
ds0 = xr.Dataset(data_vars)

Before:

%timeit ds0['long_variable_name1999']
%timeit ds0.__repr__()

1.33 ms ± 23 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
2.66 s ± 52.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

After:

%timeit ds0['long_variable_name1999']
%timeit ds0.__repr__()

10.5 µs ± 203 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)
84.2 ms ± 1.28 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)