provide on_unimplemented hint for ops::Index

[durka]

As discussed on IRC. cc @steveklabnik

If you do slice[2i32] you get an error that tells you what is wrong (Index<i32> is not implemented for [T]) but not how to fix it (slices are indexed by usize). A quick fix would be to use #[rustc_on_unimplemented]; it would look like this.

If you are indexing something that isn't [T] then the message would be a false positive.

[GuillaumeGomez]

It looks like fun, I take it.

[durka]

Fixed by #33401 \o/

[pnkfelix]

PR #33491 is going to regress this issue; reopen when it lands

[arielb1]

@pnkfelix

No it's not. It is going to give the same error message on ranges (which is debatable), but it will give the correct error message with slices.

[GuillaumeGomez]

@arielb1: This feature isn't just for ranges and slices. So for now, this is a regression.

[pnkfelix]

@arielb1 I inferred that something regressed from the fact that you removed one of unit tests in #33491, but I guess I am not clear on how much of a regression occurred.