btw, this was one of the major things i aimed at in my work since june... writing a nostr codec that stays binary in memory

and writing a max-fast binary encoder for the database

the binary codec is faster than anything and the memory utilization on events being natively binary saves typically over 128 bytes of space for each event, and even more if they have e or p tags, those also are stored binary in the encoder... i haven't written handling for that in the runtime form though, still hexadecimal string arrays

the binary codec: 

https://github.com/mleku/nostrbench

part of the reason my encodings are faster is also because the runtime binary form is also raw binary and not hex, so there is a missing step, and further, i use xhex, a SIMD hex codec

this is what the revised replicatr will be using, of course

i intend to make a branch specifically for the minimally configurable raw version once it's done, with no fancy, no limits, no nothing, something you can build a monster out of

fiatjaf was so far off the mark in his dismissing badger as the data store

it's only second to in-memory hashmaps