Speaking of np.remainder, I also recently noticed that it treats integer zero strangely:

>>> np.reciprocal(np.array(0))
-9223372036854775808

Probably should return 0 with a warning, like np.int32(1)/np.int32(0). Thus for ints, reciprocal would return 1 if x==1, otherwise 0.

Oh oops misread, you are talking about np.remainder

fwiw that's a float nan interpreted as an integer (sorry for continuing off-topic..)

>>> np.float64(np.nan).astype(int)
-9223372036854775808

Yes.

Sorry for noise, please continue any side discussion at #6318.

The problem with how the remainder is calculated also affects divmod -- see #6127

The problem with how the remainder is calculated also affects divmod -- see #6127

I'm not sure this is the same issue. Here the result's value is correct, just the sign is incorrect (note that 0.0 == -0.0).

@pitrou - you may be right that the problems are separate. My link was triggered by the problem in divmod being in part because the sign of the remainder is checked and then the remainder is changed without changing the quotient. It does seem that part of the code needs some looking after...

I think the root of the problem is how the sign is calculated here:

From the most recent commit to that file it looks like the author's intention was to copy the behavior of the python function float_divmod, however the sign check there looks like this:

/* div is zero - get the same sign as the true quotient */
floordiv = copysign(0.0, vx / wx); /* zero w/ sign of vx/wx */

and copysign is defined as:

double
copysign(double x, double y)
{
    /* use atan2 to distinguish -0. from 0. */
    if (y > 0. || (y == 0. && atan2(y, -1.) > 0.)) {
        return fabs(x);
    } else {
        return -fabs(x);
    }
}

however the numpy implementation does not use the atan trick.

Closing as I think this has been fixed, on current master:

In [75]: >>> np.remainder(-1.,1.)                                                                                       
Out[75]: 0.0

In [76]: >>> (-1.) % 1.                                                                                                 
Out[76]: 0.0