Incorrect answers with numpy ufuncs and the out keyword

[ngoldbaum]

Consider the following example:

import numpy as np
from pint import UnitRegistry

u = UnitRegistry()

a = [1, 2, 3]*u.g
b = [1, 2, 3]*u.kg
out = np.zeros(3)

print(np.add(a, b, out=out))

With Pint 0.8.1, I get [2. 4. 6.] as the output of this script, when I would expect either [1001. 1002. 1003.], [1.001 1.002 1.003] or similar with units attached. Pint returns the correct answer when out is not specified.

[hgrecco]

Can you check without the out keyword?

[ngoldbaum]

This script:

import numpy as np
from pint import UnitRegistry

u = UnitRegistry()

a = [1, 2, 3]*u.g
b = [1, 2, 3]*u.kg

print(np.add(a, b))

prints [1001. 2002. 3003.] gram, so indeed it has to do with using out.

[hgrecco]

I wonder if when #630 is merged, this issue will be fixed.

[ngoldbaum]

I get the same wrong answer with numpy 1.12.1 installed.

[jthielen]

Just to note it down so it doesn't get forgotten, this issue (along with most other issues with optional arguments in NumPy ufuncs) remains after the refactor in #905.

[hgrecco]

This works in
Numpy: 1.20.1
Pint: 0.17