CoordinateIndexer (.cx) is relying on pandas internals

[jorisvandenbossche]

See pandas-dev/pandas#27258, recent change in pandas broke some CoordinateIndexer tests.

The current implementation of CoordinateIndexer:

geopandas/geopandas/base.py

Lines 734 to 761 in 7c66c93

	class _CoordinateIndexer(_NDFrameIndexer):
	"""
	Coordinate based indexer to select by intersection with bounding box.
	Format of input should be ``.cx[xmin:xmax, ymin:ymax]``. Any of ``xmin``,
	``xmax``, ``ymin``, and ``ymax`` can be provided, but input must
	include a comma separating x and y slices. That is, ``.cx[:, :]`` will
	return the full series/frame, but ``.cx[:]`` is not implemented.
	"""
	def _getitem_tuple(self, tup):
	obj = self.obj
	xs, ys = tup
	# handle numeric values as x and/or y coordinate index
	if type(xs) is not slice:
	xs = slice(xs, xs)
	if type(ys) is not slice:
	ys = slice(ys, ys)
	# don't know how to handle step; should this raise?
	if xs.step is not None or ys.step is not None:
	warn("Ignoring step - full interval is used.")
	xmin, ymin, xmax, ymax = obj.total_bounds
	bbox = box(xs.start if xs.start is not None else xmin,
	ys.start if ys.start is not None else ymin,
	xs.stop if xs.stop is not None else xmax,
	ys.stop if ys.stop is not None else ymax)
	idx = obj.intersects(bbox)
	return obj[idx]

is relying on some pandas internals. We should see if we can make this cleaner.

I actually think this only needs the __getitem__ of its parent _NDFrameIndexer, so that is something we could implement rather easily ourselves.
In addition, we use DataFrame/Series._create_indexer, but this is also only 3 lines of code.