Jake Vanderplas

> The JAX behavior isn't an "in-place" change The spec doesn't seem to make this distinction currently; it talks about requirements for `x1 += x2`, and later says that `__iadd__`...

Pathological cases aside, I think the root of the issue is that the current guidance is self-contradictory. i.e. these three statements in the [spec](https://data-apis.org/array-api/latest/API_specification/array_object.html#in-place-operators) cannot all be true: > An...

dynamic shapes are still very experimental and don't have much support in JAX APIs. Assigning to @mattjj because he may know whether or not it's expected to work here.

Hi Patrick – these are public symbols in JAX, so it should be safe to add them here. Thanks!

Thanks for the question! We've talked about this, but have not implemented any support for `__array_ufunc__` or other custom dispatch mechanisms. The thinking is that it adds too much complexity...

Here's an example of how this might look using the array API: ```python import jax.experimental.array_api # side-effecting import required now, but won't be needed in the future import jax.numpy as...

(2) will never be feasible in the long term. The numpy team tried many approaches over the years to make this work (starting with python operator dispatch & `__array__`, then...

I'd add that (1) is probably the best way forward for what you have in mind too: if the downstream tools that you use adapt their implementations to make use...

I'm going to close this, becuase `__array_ufunc__`-style dispatch is out-of-scope for JAX.

Thanks for the contribution! See #14688 for a past attempt at this, the things that went wrong, and some discussion of the considerations around a change like this. In particular,...