how to make sure Quantity cannot wrap a user defined object

[seb5g]

Hi there,
I'm developing a data acquisition framework (PyMoDAQ) and data handling. The main objects (let's say DataBase holds data as a list of numpy arrays and an attribute being units. So I'm using quite a bit of pint functionalities under the hood, simply adding compatibility for list of ndarrays. By implementing array_ufunc and array_function and using their implementation in pint, I could quickly get it working for many functions. However when I execute:

Quantity(1, 's') * DataBase(...)

I'm getting a quantity with a magnitude being my DataBase object.

When I'm doing the other way around:

 DataBase(...) * Quantity(1, 's')

I properly (what I want at least) get the product of the underlying quantities (with units handling).

Here comes the question:

How can I be sure that this is always my DataBase that will have the priority?

I tried the __array_priority__ attribute (I saw you set it to 17 for your numpy wrapping quantities) but it doesn't work.

I also almost understand the idea of upcast_type but i'm actually not sure if this is or not related to my issue...

Thxs

[andrewgsavage]

yea have a try at adding it to upcast types

see #1888

[seb5g]

That's working! however my DataBase object is actually a base class inherited by many subclass. If I add the DataBase in upcast then only object of this type will work but not the inheriting ones. I could add all those in the upcast_type_names but do you know of a way to do it for all subclass at once?

[andrewgsavage]

something like

class DataBase():
   def __new__(cls):
      if not pint.compat.is_upcast_type(cls):
         # add class to upcast types here

[seb5g]

nice, I wrote:

    def __new__(cls, *args, **kwargs):
        if not pint.compat.is_upcast_type(cls):
            upcast_type_map.update({f'{cls.__module__}.{cls.__name__}': None})
            # this will make sure, these objects are not wrapped by pint
        return super().__new__(cls)

and all subclass are upcasted! Maybe it could be nice and clearer to add this in the doc rather than digging into past issues?