> That in argument against adopting the inferior version, as that will
> contribute more momentum to doing it in a way that doesn't make sense long
> term.

That was moreso an attempt at a disclosure, rather than may argument. But
also as noted further up in the thread, both approaches have a trade off:
one is better for light clients in a p2p "one honest peer mode", while the
other is more compact, but is less verifiable for the light clients. They're
"inferior" in different ways.

My argument goes more like: moving to prev scripts means clients cannot
verify in full unless a block message is added to include the prev outs.
This is a downgrade assuming a "one honest peer" model for the p2p
interactions. A commitment removes this drawback, but ofc requires a soft
fork. Soft forks take a "long" time to deploy. So what's the cost in using
the current filter (as it lets the client verify the filter if they want to,
or in an attempted "bamboozlement" scenario) in the short term (as we don't
yet have a proposal for committing the filters) which would allow us to
experiment more with the technique on mainnet before making the step up to
committing the filter. Also, depending on the way the commitment is done,
the filters themselves would need to be modified. 

> I don't agree at all, and I can't see why you say so.

Sure it doesn't _have_ to, but from my PoV as "adding more commitments" is
on the top of every developers wish list for additions to Bitcoin, it would
make sense to coordinate on an "ultimate" extensible commitment once, rather
than special case a bunch of distinct commitments. I can see arguments for
either really.

> This is inherent in how e.g. the segwit commitment is encoded, the initial
> bytes are an identifying cookies. Different commitments would have different
> cookies.

Indeed, if the filter were to be committed, using an output on the coinbase
would be a likely candidate. However, I see two issues with this: 
  
  1. The current filter format (even moving to prevouts) cannot be committed
     in this fashion as it indexes each of the coinbase output scripts. This
     creates a circular dependency: the commitment is modified by the
     filter, which is modified by the commitment (the filter atm indexes the
     commitment). So we'd need to add a special case to skip outputs with a
     particular witness magic. However, we don't know what that witness
     magic looks like (as there's no proposal). As a result, the type
     filters that can be served over the p2p network may be distinct from
     the type of filters that are to be committed, as the commitment may
     have an impact on the filter itself. 

  2. Since the coinbase transaction is the first in a block, it has the
     longest merkle proof path. As a result, it may be several hundred bytes
     (and grows with future capacity increases) to present a proof to the
     client. Depending on the composition of blocks, this may outweigh the
     gains had from taking advantage of the additional compression the prev
     outs allow.

In regards to the second item above, what do you think of the old Tier Nolan
proposal [1] to create a "constant" sized proof for future commitments by
constraining the size of the block and placing the commitments within the
last few transactions in the block?

> but with an added advantage of permitting expirementation ahead of the
> commitment.

Indeed! To my knowledge, lnd is the only software deployed that even has
code to experiment with the filtering proposal in general. Also, as I
pointed out above, we may require an additional modification in order to be
able to commit the filter. The nature of that modification may depend on how
the filter is to be committed. As a result, why hinder experimentation today
(since it might need to be changed anyway, and as you point out the filter
being committed can even be swapped) by delaying until we know what the
commitment will look like?

> You can still scan blocks directly when peers disagree on the filter
> content, regardless of how the filter is constructed

But the difference is that one options lets you fully construct the filter
from a block, while the other requires additional data.

> but it makes the attack ineffective and using outpoints considerably increases
> bandwidth for everyone without an attack

So should we optimize for the ability to validate in a particular model (better
security), or lower bandwidth in this case? It may also be the case that the
overhead of receiving proofs of the commitment outweigh the savings depending
on block composition (ofc entire block that re-uses the same address is super
small).

> It seems to me this point is being overplayed, especially considering the
> current state of non-existing validation in SPV software (if SPV software
> doesn't validate anything else they could be validating, why would they
> implement a considerable amount of logic for this?).

I don't think its fair to compare those that wish to implement this proposal
(and actually do the validation) to the legacy SPV software that to my
knowledge is all but abandoned. The project I work on that seeks to deploy
this proposal (already has, but mainnet support is behind a flag as I
anticipated further modifications) indeed has implemented the "considerable"
amount of logic to check for discrepancies and ban peers trying to bamboozle
the light clients. I'm confident that the other projects seeking to implement
this (rust-bitcoin-spv, NBitcoin, bcoin, maybe missing a few too) won't find it
too difficult to implement "full" validation, as they're bitcoin developers
with quite a bit of experience. 

I think we've all learned from the past defects of past light clients, and
don't seek to repeat history by purposefully implementing as little validation
as possible. With these new projects by new authors, I think we have an
opprotunity to implement light clients "correctly" this time around.

[1]: https://github.com/TierNolan/bips/blob/00a8d3e1ac066ce3728658c6c40240e1c2ab859e/bip-aux-header.mediawiki

-- Laolu


On Fri, Jun 8, 2018 at 9:14 AM Gregory Maxwell <greg@xiph.org> wrote:
On Fri, Jun 8, 2018 at 5:03 AM, Olaoluwa Osuntokun via bitcoin-dev
<bitcoin-dev@lists.linuxfoundation.org> wrote:
> As someone who's written and reviews code integrating the proposal all the
> way up the stack (from node to wallet, to application), IMO, there's no
> immediate cost to deferring the inclusion/creation of a filter that includes
> prev scripts (b) instead of the outpoint as the "regular" filter does now.
> Switching to prev script in the _short term_ would be costly for the set of
> applications already deployed (or deployed in a minimal or flag flip gated
> fashion) as the move from prev script to outpoint is a cascading one that
> impacts wallet operation, rescans, HD seed imports, etc.

It seems to me that you're making the argument against your own case
here: I'm reading this as a "it's hard to switch so it should be done
the inferior way".  That in argument against adopting the inferior
version, as that will contribute more momentum to doing it in a way
that doesn't make sense long term.

> Such a proposal would need to be generalized enough to allow several components to be committed,

I don't agree at all, and I can't see why you say so.

> likely have versioning,

This is inherent in how e.g. the segwit commitment is encoded, the
initial bytes are an identifying cookies. Different commitments would
have different cookies.

> and also provide the necessary extensibility to allow additional items to be committed in the future

What was previously proposed is that the commitment be required to be
consistent if present but not be required to be present.  This would
allow changing whats used by simply abandoning the old one.  Sparsity
in an optional commitment can be addressed when there is less than
100% participation by having each block that includes a commitment
commit to the missing filters ones from their immediate ancestors.

Additional optionality can be provided by the other well known
mechanisms,  e.g. have the soft fork expire at a block 5 years out
past deployment, and continue to soft-fork it in for a longer term so
long as its in use (or eventually without expiration if its clear that
it's not going away).

> wallets which wish to primarily use the filters for rescan purposes can't
> just construct them locally for this particular use case independent of
> what's currently deployed on the p2p network.

Absolutely, but given the failure of BIP37 on the network-- and the
apparent strong preference of end users for alternatives that don't
scan (e.g. electrum and web wallets)-- supporting making this
available via P2P was already only interesting to many as a nearly
free side effect of having filters for local scanning.  If it's a
different filter, it's no longer attractive.

It seems to me that some people have forgotten that this whole idea
was originally proposed to be a committed data-- but with an added
advantage of permitting expirementation ahead of the commitment.

> Maintaining the outpoint also allows us to rely on a "single honest peer"security model in the short term.

You can still scan blocks directly when peers disagree on the filter
content, regardless of how the filter is constructed-- yes, it uses
more bandwidth if you're attacked, but it makes the attack ineffective
and using outpoints considerably increases bandwidth for everyone
without an attack.  These ineffective (except for increasing
bandwidth) attacks would have to be common to offset the savings. It
seems to me this point is being overplayed, especially considering the
current state of non-existing validation in SPV software (if SPV
software doesn't validate anything else they could be validating, why
would they implement a considerable amount of logic for this?).