BUG: np.equal does not work where == does

[abalkin]

>>> import numpy as np
>>> x = np.array('abc', 'c')
>>> x == x
array([ True,  True,  True], dtype=bool)
>>> np.equal(x, x)
NotImplemented

[ddasilva]

I took a look at this. Both use the equal ufunc. However, x==x goes through array_richcompare() first, and array_richcompare() has extra code that is recovering from the failure.

The end of array_richcompare() (arrayobject.c, line 1433):

    if (result == Py_NotImplemented) {
        /* Try to handle string comparisons */
        ...
    }
    return result;

[abalkin]

So it appears that the underlying problem is that ufuncs do not work with flex arrays. For example,

>>> x = numpy.zeros(1, dtype='d,d')
>>> numpy.add(x, x)
NotImplemented

[charris]

See also

#4470
#3322
#3015

[seberg]

Going to close this, while still true in some other cases, it isn't for strings since 1.25 or so.