Feature request: into_group_btreemap

[HollayHorvath]

A BTreeMap version of the into_group* function group would be a nice addition.

The idea is that sometimes working with Hash is not feasible or even impossible, e.g. when you have to work with float-based datatypes as key or you would like to group your data by some custom criteria without wrapping the type and implementing custom Hash.

The into_group_btreemap() and into_group_btreemap_by() functions should work similarly how into_group_map() and into_group_btreemap_by() work, but would require K: Ord + Eq instead of K: Hash + Eq and would return BTreeMap<K, Vec<V>> instead of HashMap<K, Vec<V>>.

I have the code patched into my codebase, so I can also start a PR is this feature is required.

[phimuemue]

Hi, nice idea. It appears there is a certain need for this, as the request for other containers comes up (in different flavors) from time to time:

• Use FNV hasher #276
• Support different hash algorithms in into_group_map #322
• Specify hasher for unique? #462
• Add Itertools.counts_by. #515 (comment)
• Is there any way to make the into_grouping_map family of functions support custom hashers #588

I'd like to put on the record (see #462 (comment)) that imho in an ideal world we would solve this problem by abstracting over the kind of the container -- leading to uniform support for HashMaps, BTreeMaps, sorted Vecs, EnumMap<Enum, Option<_>>, and possibly more, instead of having sporadic support for some of them.

It might not be easily doable, but I am unsure if duplicating functions here and there is better than exploring the possibilities of the aforementioned abstraction.

[loarca]

Hey 👋🏻

Any updates on this?

[Philippe-Cholet]

I did not see any update related to it. The only recent thing is me adding the label "generic-container".
Personally, I tried some approaches to have more generic containers but nothing satisfying yet (by far).