> So the sha256 of the span of the group doesn't commit to start and end
> -- it just serializes a vector, so commits to the number of elements,
> the order, and the elements themselves.

Gotcha wasn't clear to me that the new state pair isn't committed as part of the annex.

Have been confused by "Introduce a new SIGHASH_GROUP flag, as an alternative to ALL/SINGLE/NONE, that commits to each output i, start <= i < end."

> Does the above resolve that?

I think so. It shouldn't be susceptible to any spend replay attack, as the state pair prevents output group overlapping though you might still have to be careful about siphoning ? Something you should already care about if you use SIGHASH_SINGLE and your x's amount > y's value.

Le ven. 9 juil. 2021 à 21:47, Anthony Towns <aj@erisian.com.au> a écrit :
On Fri, Jul 09, 2021 at 09:19:45AM -0400, Antoine Riard via bitcoin-dev wrote:
> > The easy way to avoid O(n^2) behaviour in (3) is to disallow partial
> > overlaps. So let's treat the tx as being distinct bundles of x-inputs
> > and y-outputs, and we'll use the annex for grouping, since that is
> > committed to by singatures. Call the annex field "sig_group_count".
> > When processing inputs, setup a new state pair, (start, end), initially
> > (0,0).
> > When evaluating an input, lookup sig_group_count. If it's not present,
> > then set start := end. If it's present and 0, leave start and end
> > unchanged. Otherwise, if it's present and greather than 0, set
> > start := end, and then set end := start + sig_group_count.
> IIUC the design rationale, the "sig_group_count" lockdowns the hashing of
> outputs for a given input, thus allowing midstate reuse across signatures
> input.

No midstates, the message being signed would just replace
SIGHASH_SINGLE's:

  sha_single_output: the SHA256 of the corresponding output in CTxOut
  format

with

  sha_group_outputs: the SHA256 of the serialization of the group
  outputs in CTxOut format.

ie, you'd take span<CTxOut>{start,end}, serialize it (same as if it were
a vector of just those CTxOuts), and sha256 it.

> Let's say you want to combine {x_1, y_1} and {x_2, y_2} where {x, y} denotes
> bundles of Lightning commitment transactions.
> x_1 is dual-signed by Alice and Bob under the SIGHASH_GROUP flag with
> `sig_group_count`=3.
> x_2 is dual-signed by Alice and Caroll under the SIGHASH_GROUP flag, with
> `sig_group_count`=2.
> y_1 and y_2 are disjunctive.
> At broadcast, Alice is not able to combine {x_1,y_1} and {x_2, y_2} for the
> reason that x_1, x_2 are colliding on the absolute output position.

So the sha256 of the span of the group doesn't commit to start and end
-- it just serializes a vector, so commits to the number of elements,
the order, and the elements themselves. So you're taking serialize(y_1)
and serialize(y_2), and each of x_1 signs against the former, and each
of x_2 signs against the latter.

(Note that the annex for x_1_0 specifies sig_group_count=len(y_1)
and the annex for x_1_{1..} specifies sig_group_count=0, for "reuse
previous input's group", and the signatures for each input commit to
the annex anyway)

> One fix could be to skim the "end > num_ouputs" semantic,

That's only there to ensure the span doesn't go out of range, so I don't
think it makes any sense to skip it?

> I think this SIGHASH_GROUP proposal might solve other use-cases, but if I
> understand the semantics correctly, it doesn't seem to achieve the batch
> fee-bumping of multiple Lightning commitment with O(1) onchain footprint I was
> thinking of for IOMAP...

Does the above resolve that?

Cheers,
aj