On Thu, Feb 23, 2017 at 03:13:43PM -0800, Bram Cohen wrote:
> On Thu, Feb 23, 2017 at 9:53 AM, Chris Priest via bitcoin-dev <
> bitcoin-dev@lists.linuxfoundation.org> wrote:
> 
> >
> > What problem does this try to solve, and what does it have to do with
> > bitcoin?
> >
> 
> I can't speak to MMRs (they look a bit redundant with the actual blockchain
> history to my eye) but circling back to utxo commitments, the benefits are

In what way do you see MMRs as redundant?

Remember that with UTXO commitments because access patterns are uniform, you'll
over time have a lot more "redundancy" in the form of lost-coins evenly spread
out across the whole keyspace.

> that it enables actual proofs of non-fraud: You can prove the validity of a
> block based on just the previous block (and maybe some previous headers
> because of mining rewards) and can prove to a light node that a utxo hasn't
> been spent yet.
>
> A major factor in the way of getting utxo commitments in blocks is
> performance. The txo set is of course vastly larger and more unwieldy. If

That statement is incorrect with pruning: you can maintain a commitment to the
TXO set, without actually storing the entire TXO set, because you don't need to
store anything for nodes that have already been spent.

Concretely, this can be done with nothing more than adding a FullySpent node
type to the MMR definition I published earlier, with the rule being that only a
left or right child of an inner node be a FullySpent node, not both; if both
sides are spent, the inner node itself becomes FullySpent. Equally, I think you
can re-use the Empty node for this, but I need to think a little about the
implications re: partial inner nodes.

Regardless, with a generalized commitment scheme, the serialization/commitment
to an Empty node is simply '0', the encoding of an unspent txout surrounded by
spent txouts will be similar in size to a position integer followed by the
txout...


A subtlety of this construction is that you can only prove that a specific
txout # is unspent, but that's actually sufficient, as you can also prove what
# a txout txid corresponds too with a previous version of the MMR.

> you make the utxo commitments trail by a small fixed number of blocks
> (between 2 and 5) their latency problems shouldn't be a big deal as long as
> the overall performance is good enough. My thesis is that with appropriate
> format and implementation tricks it's possible to get performance good
> enough to no longer be a gating factor to deployment.
> 
> Disappointingly there hasn't been any feedback about my implementation,
> just discussion about merkle sets generally.

Well, I think at this point there's still discussion over whether or not a UTXO
set commitment is the right approach to begin with; if it's not your
implementation isn't relevant.

-- 
https://petertodd.org 'peter'[:-1]@petertodd.org