Eric Wieser

Well in this case, it doesn't fit in a date time, so we don't have that option. Is it better to return: 1. Only datetime64 2. Only rounded / truncated...

.astype(np.generic) might be an interesting way to allow spelling that, @seberg On Sun, Dec 30, 2018, 3:13 PM Sebastian Berg I noticed that years ago I once glanced at changing...

This is a duplicate of #8546 (closed) and #7619

Caused by `np.ndarray.round` using floats internally

I wonder if there'd be any value in creating an `DeprecationWarning` subclass with a `version` property?

Is this table up to date? Is there a better way of tracking this?

> It would be really nice to have some way of optionally outputing a simple nested-list format For simple dtypes, you can get that today with `json.dumps(arr.tolist())`

I don't think that `copy` is needed there. These seems like an improvement that could be built directly in to the `sum` function if #8773 were implemented

> I see the same deficiencies. Yes, that's because you're doing exactly the same operation over memory, and just calling the axes different things `np.copy(c, order='f').sum(axis=0)` is the relevent test.

Let's solicit a few more votes on https://leanprover.zulipchat.com/#narrow/stream/287929-mathlib4/topic/naming.20conventions.3A.20natCast.20vs.20nat_cast/near/346323810 first.