Matt Haberland

Thanks @steppi @xuefeng-xu!

Re: https://github.com/scipy/scipy/pull/20765#issuecomment-2132193699 The thought was that it might be worth creating a new function for all five changes to avoid disruption. That said, I, too, would really prefer to just...

I made some changes that I think makes this mergeable. Should have pushed with `[docs only]` but oh well. If the render looks good (and if the table looks correct...

Looks accurate. I'll let @dschmitz89 close gh-17404 if this addresses all points. I think there was some information that gh-17491 would have added that is not here, for instance. In...

I'll consider this issue to be about the infinite loop in `_drv2_ppfsingle`. It will be closed by gh-21354, but we'll leave gh-3890 open about improvements to `logser.cdf` and friends.

~~Use `xp_assert_close` with the `accept_0d=True` option.~~ Never mind. I didn't remember the context. Use whatever you need. I don't remember what happened with the NumPy strict checks. I tried to...

With `accept_0d=True`, `actual` and `desired` need to both be either 0d arrays or scalars, not mixed, to pass. `accept_0d=False` is a behavior that insists that the result be a scalar...

Ok, would you open a PR that changes the name of the argument and adds the logic for complex NaNs? I think the only place that uses the non-default is...

Oh, great. Yes, the code is in [`_lib/_array_api.py`](https://github.com/scipy/scipy/blob/628a6b3bf3e1a3149b8a6390322ae976bba930fa/scipy/_lib/_array_api.py#L312). BTW the keyword I was referring to is *`allow_0d`*, not `accept_0d`. That's much easier to interpret, I think. Hopefully that resolves that...

> if we want to avoid 0d-outputs passing silently, then just make strict=True the default for xp_assert_equal It is, and that is the default.