How to use the larray.AxisCollection.union function in larray

# transform slice keys to Array too IF they refer to axes present in advanced key (so that those axes
    # broadcast together instead of being duplicated, which is not what we want)
    def get_axes(value):
        return value.axes if isinstance(value, la.Array) else la.AxisCollection([])

    def slice_to_sequence(axis, axis_key):
        if isinstance(axis_key, slice) and axis in la_key_axes:
            # TODO: sequence assumes the axis in the la_key is in the same order. It will be easier to solve when
            # make_numpy_broadcastable automatically aligns all arrays
            start, stop, step = axis_key.indices(len(axis))
            return la.sequence(axis.subaxis(axis_key), initial=start, inc=step)
        else:
            return axis_key

    # XXX: can we avoid computing this twice? (here and in make_numpy_broadcastable)
    la_key_axes = la.AxisCollection.union(*[get_axes(k) for k in key])
    key = tuple(slice_to_sequence(axis, axis_key) for axis, axis_key in zip(self, key))

    # start with the simple (slice) keys
    # scalar keys are ignored since they do not produce any resulting axis
    res_axes = la.AxisCollection([axis.subaxis(axis_key)
                                  for axis, axis_key in zip(self, key)
                                  if isinstance(axis_key, slice)])
    transpose_indices = None

    # if there are only simple keys, do not bother going via the "advanced indexing" code path
    # MONKEY PATCH CHANGED LINE
    if all(isinstance(axis_key, (int, long, np.integer, slice)) for axis_key in key):
        bcasted_adv_keys = key
    else:
        # Now that we know advanced indexing comes into play, we need to compute were the subspace created by the
        # advanced indexes will be inserted. Note that there is only ever a SINGLE combined subspace (even if it