Seems like a bug to me.  Certain branches in _array_richcompare return False to fail rather than Py_NotImplemented, which means the string-understanding comparison fallbacks don't run.  Attached is a (simple) patch that resolves this bug, and doesn't seem to cause any of the unit tests to fail.  Does this make sense to someone with a better understanding of the rich comparison code than I?

Mike

On 05/25/2010 12:54 PM, Keith Goodman wrote:

a1 = np.array(['a', 'b'], dtype=object)
>>  a2 = np.array(['a', 'b'])
>>
>>  a1 == a2
     

   array([ True,  True], dtype=bool)  # Looks good
 

>>  a2 == a1
     

   False  # Should I have expected this?