LinearMapper.map_screen should return scalar output for scalar input.

[mdickinson]

The LinearMapper.map_screen method normally returns a scalar output given scalar input. However, if self._null_data_range is True, it'll return a 1d array instead of a scalar, given scalar input.

While we're here, there's also an isinstance check that makes little sense: it checks for either an ndarray, a tuple, or a list, but then looks up the .shape attribute in all three cases:

chaco/chaco/linear_mapper.py

Lines 47 to 48 in e0e4f7d

	if isinstance(data_array, (tuple, list, ndarray)):
	x = empty(data_array.shape)

[mdickinson]

@rkern points out on another repository that:

map_screen() is not a very coherent interface. In some of the AbstractMapper interfaces, like LinearMapper, map_screen() behaves inconsistently when given a scalar. If the data range is null, it returns a 1-element array. When it has a non-null data range, it computes the result via a very standard arithmetic expression involving only scalars, so the result is a scalar. In others, like LogMapper, it coerces the input to an array first so it always returns an array, regardless of branch. Returning an array is the intended interface of map_screen(), it seems.

[mdickinson]

Closing this; from discussion, map_screen should be returning arrays. (There's still the looking up .shape on a list/tuple part, but that's a separate issue.)