I’ve been thinking more about the interfaces to containers.
Now, for all of the linear data types (lists, sets, multisets, queues, priority queues, arrays, maps, other kinds of associative arrays, etc.) it is easily possible to implement Collectable and Enumerable.
However, a closer look at Enumerable shows that it is not the inverse of Collectable at all: The caller is able to temporarily suspend the enumeration process, but it is impossible to only extract a single item, because the ‘enumerated-collection-in-progress’ is thrown away and only the accumulator is returned.
That is, Enumerable should maybe more accurately be named Foldable or Reducable. (It has two extra features on top of a pure reducable, which are the ability for the reducing function to stop before the end of the reducable is reached, and the ability to temporarily suspend an in-process enumeration if the caller asks for it. Only for things wrapping e.g. file handles or sockets the suspend-case might involve some extra clean-up code.)
However, there are many cases in which we are only interested in extracting one element from the current collection we have at hand. For lists, this should be the head, for a priority queue, the minimum(/maximum, depending on what kind of queue it is), for queues the front, for stacks the top, for unordered collections like sets and maps any element is equally fine, etc.
The signature of, let’s say: Extractable.extract is as follows:
extract(Extractable.t) :: {:ok, {item :: any, Extractable.t}} | :error
The implementations of this for the built-in types List and Tuple will be fast. Map and MapSet unfortunately not, as Erlang does not expose a way to get an arbitrary element from a Map, so we have to do a round-robin list conversion.
In other words: For Map and MapSet, the Enumerable implementation is basically an optimized version of wrapping Extractable!
