Re: Some (negative) thoughts about did:key and multicodec.

On 3/10/21 8:44 PM, Nikos Fotiou wrote:
>> 1. Fork a single community working in peace on codec tables into two 
>> communities that are effectively doing the same thing, except for the
>> way that they encode bytes.
> Don't we do this right now?

No, I don't think we've split the two communities. We are re-using the
multicodec table for did:key. The thing that did:key does is choose to only
use base58btc and only a handful of the multicodec entries... which allows us
to hardcode values.

In the future, let's imagine that there is a post-quantum key format that
did:key wants to support. We put that public key format into the multicodec
table and then use the hard coded value in the code.

Think of it as doing things at two layers. Specifications should generalize,
so that they future-proof themselves and ideally, build on top of other
specifications that allow people to reason about them.

The generalization that did:key builds on top of is multibase and multicodec
to identify the public key type.

However, we subset both of those specifications... and only use base58 and
only a handful of public key formats encoded in multicodec to make
implementations easier.

> Transmute does that. Digital Bazaar does that They don't parse a multicodec
> header, they don't decode varints,  they just read the value of two bytes.
> Even did:key specification says more or less "use 6Mk" to denote a Ed25519
> public key :)

Yes, correct... and I'd argue that this is the best of both worlds.

You had the curiosity to go in and figure out how all of this works, which is
great. A non-trivial number of developers don't have the time or energy to do
that, and instead take short cuts and cargo-cult their way through
implementations. While that's less ideal, when people do that, you still want
the system to work... and in this case it does. People don't have to
understand varints to create a working implementation of did:key.

The ones that try to understand it at depth, like you and me, dip into
insanity for a brief spell before coming out of it and going "meh, I guess the
trade-off is acceptable".

So that's why I think we've made the right call here... we're trying to solve
a multivariate equation that results in a proper implementation (and achieves
the other goals listed in the previous email).

> And of course I don't blame them, it is much faster to transform 
> "Mutlicodec table" entries in an encoding that humans understand and all 
> programming languages can trivially implement, than doing varint maths :)

I skipped going into this detail before, as I didn't think it was worth
highlighting, but it might be worth highlighting now. You had mentioned
something to the effect of "this is what CoAP does" in a previous email.

I wanted to point out that CBOR (and CoAP) do their own variation of varints.
In a single byte, you can have major type bits followed by the value in a
single byte:

In a perfect world, we would've just used this (and not varints)... but that's
not the path that IPFS took (for the reasons previously mentioned).

All that to say, if you rewind time and go with the way CBOR encodes variable
length integers... you will still confuse people... because the second your
integer value gets to be larger than 23, you need two bytes. So, taking the
exact example that tripped you up -- 0xed becomes 0x18ed instead of 0xed01
(which is arguably better/worse).

So, all that to say, I'm not sure we'd end up in a different place if we went
the CoAP/CBOR unsigned integer encoding route. It's just a different way of doing
byte encoding that ends up not really resulting in anything different. Am I
missing something?

-- manu

Manu Sporny -
Founder/CEO - Digital Bazaar, Inc.
blog: Veres One Decentralized Identifier Blockchain Launches

Received on Thursday, 11 March 2021 15:40:52 UTC