Variable length encoding

[klaidliadon]

This is a proposal for variable length encoding for access keys.

• First byte is version (3)
• Second byte is 4 bit for length of each IDs
  • 1-4 Bytes for the first ID
  • 1-4 Bytes for the second ID
• The rest is random bytes

The current version is 32 bits (at least 22 random bytes) but we can change that

Sample keys

• v0: 7vShFoC0qe2eMQKeSde1hqBAAAAAAAAAA
• v1: AQAAAAAAAAAB40UGtqT8QqnbqhyGohFvk7Y
• v2 (old): AgAAAAAAAAABAAAAAAAAAAJ9voA9iD-LUCZ0UF1rYD4Also
• v2 (new): AxEBAlv8_WVyNz2lgc1Eltn3jzVIgkVw0_HxE_RtGXI

[patrislav]

I like this, this increases the entropy by a lot and makes the keys more aesthetically pleasing by removing all the screams ("AAAAA"). The v2 was very short-lived - is it even used in production? Maybe this should be v2?

[klaidliadon]

The v2 was very short-lived - is it even used in production? Maybe this should be v2?

No it's in yet, so we can use this as v2 if there's quorum.

[klaidliadon]

Possible improvements

We could compress this even more by using the first byte for version and length:

• 4 bits (0-16) for version
• 2 bits per id length - we save the log2 of the bitsize:
  • 0 = 1 byte
  • 1 = 2 bytes
  • 2 = 4 bytes
  • 3 = 8 bytes

For instance:

• version 2
• 2 byte ID1
• 2 byte ID2
  Byte 1 will be 0x25 (0010 for the version, 01 and 01 for the lengths)

[patrislav]

Not sure it's worth it for just one byte. I like the version as a separate byte, it's easy to decode, understand, and even see at a glance. We don't need to squeeze every bit of performance out of it IMO