Serialization support

[kyrias]

This crate is pretty much exactly what I could really use for a project of mine, but I'd need to easily be able to serialize it to disk in a portable way. Is there any chance of a serde serialization implementation happening?

[michaelsproul]

Yeah, it's very much on the cards! My other trie (radix_trie) has a Serde serialisation implementation that someone contributed, and we could probably copy much of the code from there.

I'm on vacation at the moment, and probably wouldn't get around to it for 2-3 weeks. Do you feel like giving it a shot in the meantime?

[kyrias]

I'm not sure if I'm knowledgeable enough to do it, but I'll see if I can give it a shot this weekend. Have a nice vacation!

[kyrias]

Actually seem to have got it working by copying the radix_trie serde implementation, and doing some small changes, but one problem is that the HashMap iteration order is arbitrary, so the tests are currently failing most of the time.

Would need to somehow sort the key iterator, hm.

[kyrias]

Ah, I can just pass the same instance of RandomState to ::with_hasher.

Edit: Erhm, nevermind, I got confused by other issues. I blame my insomnia! Still unsure how to get a consistent test suite run for iterating over the hashmap, hm.

[kyrias]

So, I worked around the HashMap iteration order problem by switching to BTreeMap. WIP serde branch is available here.

[kyrias]

... And after fighting with Serde to get the deserialization implementaiton working for a few hours, I now realize that the auto-derived ones work just fine. sigh.

---

Edit: Is there any particular reason that HashMap is used rather than a BTreeMap btw? serde_derive can derive the implementations for the BTreeMap version of the code, but for the HashMap version of the code it chokes on the S trait bound which it doesn't know what to do with.

[michaelsproul]

The HashMap option is mostly for speed. Thanks for looking into this! I'm back from vacation now so I should have some time to look into it soon