RFC on SciPy array types support

Hey all,

I just posted a long RFC at https://github.com/scipy/scipy/issues/18286. It
brings together a number of related efforts and discussions that we've had
going on for a while. Quite a bit of that has been focused at "other array
library" support - primarily CuPy, PyTorch, and Dask. However, I think
there's also an opportunity to deal with lots of NumPy inputs that we've
been struggling with - np.matrix, masked arrays, and object arrays in
particular.

I'd like to post a couple of questions here. The main ones are

   1. Do y'all like the main design principle, "array type in == array type
   out" (and raise TypeError otherwise)?
   2. Does the introduction strategy - develop behind a global flag and env
   var first, flip the switch when complete - seem reasonable?
   3. For the process of discussing and deciding on this, what do you think
   we need (if anything) beyond discussion here, on the issue, and in the next
   community meeting(s)?

Cheers,
Ralf

_______________________________________________
SciPy-Dev mailing list -- [email protected]
To unsubscribe send an email to [email protected]
https://mail.python.org/mailman3/lists/scipy-dev.python.org/
Member address: [email protected]