unintuitive behaviour interacting with numpy scalars

[rlhycd]

import pint, numpy
si = pint.UnitRegistry()

a = numpy.ones(1)
b = 10. * si.gram / si.kilogram
print(a[0]+b, "vs.", float(a[0])+b)

11.0 gram / kilogram vs. 1.01 dimensionless

[jthielen]

I think this is a bigger issue in general with NumPy's deferring to pint for operator precedence not working properly. The following code (with arrays) demonstrates the same issue:

import pint, numpy
si = pint.UnitRegistry()

a = numpy.array([0, 1, 2])
b = 10. * si.gram / si.kilogram
print(a + b, "vs.", b + a)

[10.0 11.0 12.0] gram / kilogram vs. [0.01 1.01 2.01] dimensionless

It looks like some more subtle debugging of why the bare NumPy array/scalar is not treated as dimensionless in the first example is need.

[Tigraan]

I guess "unintuitive behaviour" is the British way to say "buggy"...

Even more generally, the problem seems to be that (numpy array) + (pint array) is a valid operation regardless of the dimensions which is surely a bug:

import pint, numpy
si = pint.UnitRegistry()

numpy.asarray([1.0, 2.0]) + 10.* si.gram  # [11.0, 12.0] gram
1.0 + 10.0 * si.gram  # pint.errors.DimensionalityError: Cannot convert from 'gram' to 'dimensionless'

One could argue that the first behaviour (add to the magnitude) is proper (I do not think so), but surely one cannot argue that floats and ndarrays should behave differently.

Looking at the definition in pint (

pint/pint/quantity.py

Line 719 in e6ae929

def _add_sub(self, other, op):

) the addition operation seems kosher. My best guess is that numpy accepts operations of the form (array) + (some other object) with restrictions lax enough that pint arrays qualify, but I have not checked.

[jthielen]

This should be resolved with #905 (see test added below):

pint/pint/testsuite/test_numpy.py

Lines 327 to 331 in 43fbae2

	def test_addition_with_scalar(self):
	a = np.array([0, 1, 2])
	b = 10. * self.ureg('gram/kilogram')
	self.assertQuantityAlmostEqual(a + b, self.Q_([0.01, 1.01, 2.01], self.ureg.dimensionless))
	self.assertQuantityAlmostEqual(b + a, self.Q_([0.01, 1.01, 2.01], self.ureg.dimensionless))

@Michaeldz36 and @Tigraan would you be able to confirm if this is resolved for you on the current master branch?

[jthielen]

Unless there are complaints, I think this can be closed now @hgrecco? There was a test added in #905 for this issue (see above comment).

[hgrecco]

Closing. @Michaeldz36 and @Tigraan , feel free to write back if needed.