When joining two recarrays and the keys are not in the same order in the two recarrays the join_by function fails, because it assumes the key fields to be in the same order.

Example:

from numpy.lib.recfunctions import rec_join
from numpy import recarray

a = recarray((1,), [('key1', int), ('key2', int), ('x', int)])
b = recarray((1,), [('key2', int), ('key1', int), ('y', int)])
a[0] = (0,1,50)
b[0] = (1,0,100)

ab = rec_join(('key1', 'key2'), a, b, jointype='inner', r1postfix='', r2postfix='')

This outputs:

TypeError: invalid type promotion

because the wrong keys are concatenated.

The problem is lines 910-911 in recfunctions.py:
r1k = drop_fields(r1, [n for n in r1names if n not in key])
r2k = drop_fields(r2, [n for n in r2names if n not in key])

After this the fields should also be sorted for both arrays.

Version: Numpy 1.8, python 2.7.1