Speed probably isn’t going to differ that much. Uuids have a space-efficient encoding in fressian and transit and are compactly represented in java. An alternative id scheme will have to be represented as a string

Thanks for pointing to representation and transferring data, just didn’t think about it. However, the main question about lookup speed is still opened.

again, it’s unlikely to make a difference, except insofar as things with smaller representations in memory tend to be faster

comparing two longs vs comparing a non-interned string for instance

or just being able to fit more of them into a cpu cache

if speed is absolutely critical you should benchmark, but my hunch is it doesn’t matter

people routinely use string identifiers in datomic

Thanks a lot.

The question comes from https://github.com/zelark/nano-id/issues/12

I’m not sure what you mean? The datom is the data. Do you just want something with same fields but type “vector”?

Datoms support nth and get access (using keys :e :a :v :tx :added) so it should be rare you need to convert them