where

[jpivarski]

numpy.where takes three arguments (ignore the single-argument version: it's equivalent to nonzero and we can do that other ways):

• condition, a boolean array
• x, any array
• y, any array

and returns a single array with x values if condition is true and y values if condition is false.

We can do this in Awkward even if x and y are not the same type; ak.where would return a UnionArray (my issue #54).

If x and y do have the same type, we can return something simpler. Actually, we can do well by implementing the general case with UnionArray first, then add special cases like "both are NumpyArrays of the same format" or "both have ListType with compatible content," one by one as needed.

[jpivarski]

This was a frequently requested feature (see old issue, also frequently requested in person).

[jpivarski]

Fixed in PR #89.