Explicit support for `pickle`?

[Techcable]

I don't know very much about pickling, so I am not sure whether we need special support.

Is the automatic support sufficient? Are there any benefits to special support?

[Techcable]

Answer: There is a significant efficiency disadvantage to the automatic pickling support - It results in duplicate data.

The current implementation of OrderedSet is implemented in terms of two underlying builtin collections, a list and a set:

_unique: set[T]
_elements: list[T]

This has little overhead in-memory, since python objects have identity are stored by reference.

However, if we use the default pickle implementation, this means that both collections will be serialized separately.

Pickling preserves object identity, so automatic pickling results in correct behavior. However, a custom implementation would still be more space efficient.