subclassing quantity breaks __setitem__ with no useful error message

[Tigraan]

The problem

Code to reproduce: (pint 0.9)

import pint
UREG = pint.UnitRegistry()
Q_ = UREG.Quantity

class QClone(Q_):
    pass

arr = QClone([1.0, 2.0], 'm')
arr[0] = Q_(0.0, 'm')

causes the error

Traceback (most recent call last):
  File "<the test file>", line 9, in <module>
    arr[0] = Q_(0.0, 'm')
  File "...\anaconda2\lib\site-packages\pint\quantity.py", line 1425, in __setitem__
    '`obj.magnitude[%s] = %s`' % (key, value))
pint.errors.DimensionalityError: Cannot convert from '0.0 meter' to 'meter'. Assign a quantity with the same dimensionality or access the magnitude directly as `obj.magnitude[0] = 0.0 meter`

The error message is not very helpful. Inspecting the two objects reveals they share the unit registry, so it is not a case of unit systems.

What happens
The subclassing causes the __setitem__ method to take a certain conditional branch. Relevant excerpt for reference

# self is the array we set an item of, value is what we put in
if isinstance(value, self.__class__):  
    factor = self.__class__(value.magnitude, value._units / self._units).to_root_units()
else:
    factor = self.__class__(value, self._units ** (-1)).to_root_units()

if isinstance(factor, self.__class__):  # not sure that test is ever False BTW
    if not factor.dimensionless:
        raise DimensionalityError(value, self.units,
                extra_msg='. Assign a quantity with the same dimensionality or '
                        'access the magnitude directly as '
                         '`obj.magnitude[%s] = %s`' % (key, value))

Because the value we try to put in is of the parent class, it is not an instance of the child class the array has, and isinstance(value, self.__class__) == False. It causes factor to become a dimensional quantity and fail the subsequent test. The resulting error message is nonsensical ("cannot convert some_unit to some_unit" is hardly helpful for debugging).

What could be done

Option 1 : nothing. (Whoever subclasses the pint classes is responsible for making them work, redefining methods etc.)

Option 2 : if the test is actually here to detect whether value is a dimensional quantity, check the relevant attributes directly (EAFP): replace the first part of the snippet by

try:
    mag, un = value.magnitude, value._units / self._units
except AttributeError:
    mag, un = value, self._units ** (-1)
factor = self.__class__(mag, un).to_root_units()

I do not know enough about pint's design to confidently suggest a pull request for option 2 before asking first. There might also be another option I missed.

[hgrecco]

There is a plan to split Scalar from Array Quantities to simplify working with pandas and other tabular oriented frameworks. I think that might help.