for those researching this topic, I keep a list of these UUID/GUID implementatio...

crdrost · on June 28, 2023

Thanks for this!

I have one idea which is perhaps nerdy enough to make the list but I've never fully fleshed it out, it's that one can encode the nonnegative integers {0, 1, 2, ...} into the finite bitstrings {0, 1}* in a way which preserves ordering.

So if we use hexits for the encoding the idea would be that 0=0, 1=1, ... E=14, then

    F00 = 15
    F01 = 16
    ...
    F0F = 30
    F100 = 31
    F101 = 32
    ...
    F1FF = 286
    F2000 =

so the format is F, which is the overflow sigil, followed by a recursive representation of the length of the coming string, followed by a string of hexits that long.

What if you need 16 hexits? That's where the recursion comes in,

    F F00 0123456789ABCDEF
     \   \    \
      \   \    \----- 16 hexits
       \   \
        \   \-- the number 15, "there are 15+1 digits to follow"
         \       (consisting of overflow, 0+1 digits to follow, and hex 0)
          \ 
           \--- overflow sigil

Kind of goofy but would allow a bunch of things like "timestamp * 1024 + 10-bit machine ID" etc without worrying about the size of the numbers involved

Ironchefpython · on June 29, 2023

https://datatracker.ietf.org/doc/html/rfc2550

Note that this RFC also supports lexicographic sorting for negative numbers using the 10k complement notation.

CMCDragonkai · on June 29, 2023

We have a uuidv7 implementation that we've been using with rocksdb for over a year https://github.com/matrixai/js-id