made some more progress on new hashmaps ^

assoc, dissoc, get/find, seq, reduce done. Next on the roster is transient impl (straightforward) & iterator (challenging)

also need a better way to reimplement fold() with perhaps better task forking

@ghadi: I've made a port of Lean Hash Maps for ClojureScript https://github.com/bendyworks/lean-map

Oh that was you. Awesome.

Yep the iteration is pretty straight forward (compared to the current impl) what looks challenging about it to you?

The tricky part for me right now is getting assoc and dissoc to current perf specs. They are around 2 - 4x slower in some cases right now . The transient assoc! and dissoc! are also pretty slow.

Feel free to PM if you want help / someone to rubber duck your ideas.

I wonder why assoc / dissoc are slower. Any theories besides the extra branch?

AFAICT assoc that requires making a node seems to be the slow down (at least in JS land). If you look at the reference impl you also see that a nodes array can contain 0 - 64 items (32 Key Value pairs) as opposed to 0 - 32 items for the current HAMT implementation.

so it might be too big for cpu caches

Another possibility is that it's due to running on JavaScript not the JVM. According to the paper their implementation of assoc / dissoc was comparable to Clojure's. So you might not see what I'm currently seeing

interesting. size 64 makes sense on javascript w 64 bits