Create eyeball-im-util with FilteredVectorSubscriber API

[jplatte]

No description provided.

[bnjbvr]

Looks interesting! I've found a few places where there could be invalid subtracts (unless I'm missing something), can you add tests for those please?

Also, I think there might be a way to make all this code more robust, by testing invariants. Say something like fn assert_invariants, that would be either debug only (or guarded behind a compile-time cargo feature, enabled in tests only), that would check the following at the end of each handle_ function call:

• indices in filtered_indices are unique (don't appear twice or more)
• indices in filtered_indices refer to values that are in the original array, and those values pass the filter check

What do you think? This might mean storing more of the original array too, but at least for testing purposes, something akin to that would be useful and increase the confidence in this new code IMO.

[jplatte]

I think testing invariants would be nice, but I don't think doing it as part of the regular non-cfg(test) is a good idea, since app debug builds by default also compile all dependencies in debug mode.

A good way to do invariant testing would be adding another test suite IMO (as unit tests that can access internal stat), using the proptest crate to ensure the internal state is kept consistent in ~all cases. There's also the option of trying Kani, but it's probably a bit harder to do and possibly an overkill.

I'd prefer not blocking this PR on extra testing though.. We don't intend to use it in places where it would be a huge issue for some bugs to be present, right? And if there are bugs, they'll likely be pretty obvious (i.e. lead to a panic).

[Hywan]

I probably like it too much, thank you!