261dc6b933 was apparently already fixed in artiq-zynq.
740543d4e2 was also fixed a different way; unified with the upstream ARTIQ solution in https://git.m-labs.hk/M-Labs/artiq-zynq/commit/5c054…
(I can/will look into porting the alignment fixes across.)
Likely not related to 22.11 at all; this is the same bug that's fixed in the upstream commit that added the tests (which still needs to be ported to the artiq-zynq copy of the firmware).
A subscriptable np.ndarray has been merged into NumPy https://github.com/numpy/numpy/pull/19879, although the extra np.dtype wrapper seems undesirable, and I'm not sure what form the shape type…
With the general move towards "proper" Python type annotations, it would be nice if we could.
Unfortunately, this is very much still brewing in upstream NumPy, and what seems like a tentative…
Not sure how should we handle num_dims otherwise.
It's easily possible to implement template value parameters if you don't need to do global type inference for them (as long as you have function…
Do we want to try to implement this as a library, or hard-code an array[element_type, num_dims] type into the compiler like before? The first option would involve adding enough in terms of template…