Wow, lots of effort here! I don’t think I see how something like this could be merged into DataScript database. My concerns are existing clients, breaking changes, compatibility and JS portability. But I’m happy to see it exists as a separate project on its own.

Couple of questions though, purely out of curiosity:

1. Why double? Wouldn’t simply using long give you more bits to work with?

This patch should reduce memory usage by at least 50%.

2. Have you actually measured it delivers on that goal?

I could not figure out how to benchmark Datascript.

Add to profiles.clj:

:bench  { :source-paths ["bench/src"] }

Then run (in datascript dir):

lein with-profiles +bench trampoline run -m datascript.bench/bench-all

Thanks for the feedback. I will close this PR.
If anybody is interested, my fork can be found at https://github.com/ronbu/datascript

Why double? Wouldn’t simply using long give you more bits to work with?

It's the only number type available on JavaScript Engines.
Using two 32bit number values is another option i need to investigate.

Have you actually measured it delivers on that goal?

I measured memory usage in Chrome with the following code:

Regular Datascript (167 MB)

  (def db
    (d/init-db (for [i (range 1e6)]
                (d/datom i :name i))
              {:name {:db/order 0}}))

With optimized Datom type (76 MB)

(def db
    (d/init-db (for [i (range 1e6)]
                (d/datom i 0 i))
              {:name {:db/order 0}}))