ColumnDataSource.patch() logic

[mattpap]

@bryevdv, the first isinstance() check makes the last one impossible:

bokeh/bokeh/models/sources.py

Lines 663 to 678 in cfd5a79

	if not isinstance(ind[0], int):
	raise ValueError("Initial patch sub-index may only be integer, got: %s" % ind[0])
	if ind[0] > col_len or ind[0] < 0:
	raise ValueError("Out-of bounds initial sub-index (%d) in patch for column: %s" % (ind, name))
	if not isinstance(self.data[name][ind[0]], np.ndarray):
	raise ValueError("Can only sub-patch into columns with NumPy array items")
	if len(self.data[name][ind[0]].shape) != (len(ind)-1):
	raise ValueError("Shape mismatch between patch slice and sliced data")
	elif isinstance(ind[0], slice):
	_check_slice(ind[0])
	if ind[0].stop is not None and ind[0].stop > col_len:
	raise ValueError("Out-of bounds initial slice sub-index stop (%d) in patch for column: %s" % (ind.stop, name))

Should that be removed or else what should be the correct logic here?

[bryevdv]

I think maybe that elif can just be deleted. For mulit-index into sub-arrays I don't think we ever had any intention of allowing one patch to straddle multiple sub-arrays, i.e. the first index should be an int index into the column (to access and patch one sub-array, possibly via slices in the sub-indexes)

For 1-d columns, the first index can be a slice, and that is handled above, so this is probably just copy-pasta.