pandas deprecates Index.get_loc with method

[mathause]

pandas deprecates the method keyword in Index.get_loc, see pandas-dev/pandas#42269. Therefore we end up with about 5000 warnings in our upstream tests:

FutureWarning: Passing method to Index.get_loc is deprecated and will raise in a future version. Use index.get_indexer([item], method=...) instead

We should fix this before pandas releases because the warning will not be silent (FutureWarning) or ask pandas to give us more time and use a DeprecationWarning at the moment.

We use this here:

xarray/xarray/core/indexes.py

Lines 233 to 235 in 4bb9d9c

	indexer = self.index.get_loc(
	label_value, method=method, tolerance=tolerance
	)

Is this only ever called with one item? Then we might be able to use

 indexer = self.index.get_indexer( 
     [label_value], method=method, tolerance=tolerance 
 ).item()
if indexer == -1:
    raise KeyError(label_value)

---

xarray/xarray/core/missing.py

Lines 571 to 572 in 3956b73

	imin = index.get_loc(minval, method="nearest")
	imax = index.get_loc(maxval, method="nearest")

This one could be easy to fix (replace with imin = index.get_indexer([minval], method="nearest").item())

---

It is also defined in CFTimeIndex, which complicates things:

xarray/xarray/coding/cftimeindex.py

Lines 461 to 466 in eea7673

	def get_loc(self, key, method=None, tolerance=None):
	"""Adapted from pandas.tseries.index.DatetimeIndex.get_loc"""
	if isinstance(key, str):
	return self._get_string_slice(key)
	else:
	return pd.Index.get_loc(self, key, method=method, tolerance=tolerance)

because get_indexer expects an iterable and thus the if isinstance(key, str) test no longer works.

@benbovy @spencerkclark

[spencerkclark]

Thanks for the heads up @mathause. We'll need to think carefully about this with respect to partial datetime string indexing. DatetimeIndex.get_indexer and DatetimeIndex.get_loc behave differently with respect to datetime strings. get_indexer interprets strings as specific dates, while get_loc interprets them as ranges:

>>> import pandas as pd
>>> times = pd.date_range("2000", periods=5)
>>> times.get_indexer(["2000"])
array([0])
>>> times.get_loc("2000")
slice(0, 5, None)

In other words -- at least for partial datetime string indexing -- it may not be as simple as swapping in get_indexer for get_loc.

Perhaps @jbrockmendel has thoughts on how we should approach this?

[shoyer]

These lines seem to be the problematic place?

xarray/xarray/core/indexes.py

Lines 233 to 235 in e26aec9

	indexer = self.index.get_loc(
	label_value, method=method, tolerance=tolerance
	)

We might be able to do something like:

if method is not None:
    indexer = get_indexer_nd(self.index, label, method, tolerance)
else:
    indexer = self.index.get_loc(label_value)

[spencerkclark]

Exactly, yeah, those are the problematic lines. That's an elegant solution. I think it will work with the minor modification to raise a KeyError if any of the indices returned by get_indexer_nd are less than zero:

if method is not None:
    indexer = get_indexer_nd(self.index, label, method, tolerance)
    if np.any(indexer < 0):
        raise KeyError(f"not all values found in index {coord_name!r}")
else:
    indexer = self.index.get_loc(label_value)

[jbrockmendel]

The "if method is not None" approach seems reasonable.

[benbovy]

👍 I can take care of it in #5692 or in a separate PR.

[mathause]

👍 I can take care of it in #5692 or in a separate PR.

I don't have a strong opinion but would be good to do this before the next release.

[benbovy]

Not sure that #5692 will be ready for the next release so I'll do this in a separate PR.