Deprecate the wrappers of the array iterator API.

[adamreichold]

Follow-up to #279, deprecating the npyiter module with a short explanation of why ndarray's facilities are preferable.

[adamreichold]

I also noticed that the implementations of Iterator::size_hint were incorrect (they did not consider the amount of iteration already performed) and added a commit fixing this. If we decide against the deprecation after all, this should probably be cherry-picked elsewhere.

[davidhewitt]

With apologies just saying that I've seen this and the other PRs but haven't had enough time to dedicate a read of these and form any opinion. Might be able to do so tomorrow evening, depending on how long you want to delay the 0.16 numpy release.

[adamreichold]

With apologies just saying that I've seen this and the other PRs but haven't had enough time to dedicate a read of these and form any opinion. Might be able to do so tomorrow evening, depending on how long you want to delay the 0.16 numpy release.

Please take your time. I think delaying the release until you or @kngwyu had time to look into the two PR related to the npyiter module (#279 and #280 here) is fine.

I think I should merge the already prepared #259 after you make the 0.16 release of PyO3, so people can upgrade using a Git dependency against our main branch here. After that we can take our time making a decision on the npyiter issue.

[davidhewitt]

👍 I think the rationale here makes a lot of sense, let's proceed with this.

One thing I'm unsure on, does using ndarray iteration require unsafe? (At least until #274 or similar.) It might be nice to have an example to point users at so that they know what to replace the deprecated code with.

[adamreichold]

One thing I'm unsure on, does using ndarray iteration require unsafe?

Only under the same circumstances that the npyiter module requires it, so it is does not when using PyReadonlyArray. Once an ArrayView is at hand, there is no additional need for unsafe. I will extend the module-level comment to point users at the PyReadonlyArray type and its as_array method.