Content-Type: text/plain; charset=utf-8
From: vjudeu@gazeta.pl
To: bitcoin-dev@lists.linuxfoundation.org
Message-ID: <75be7180-709a-b735-a27b-50d0c6a2af5b@gazeta.pl>
Content-Transfer-Encoding: quoted-printable
Date: Sat, 07 May 2022 04:50:12 +0000
MIME-Version: 1.0
Subject: [bitcoin-dev] Adding SIGHASH to TXID
Precedence: list

For now, we have txid:vout as a previous transaction output. This means =
that to have a stable TXID, we are forced to use SIGHASH_ALL somewhere, =
just to prevent any transaction modifications that can happen during adding=
 some inputs and outputs. But it seems that new sighashes could be far more=
 powerful than we expected: it is technically possible to not only remove =
previous transaction output by using SIGHASH_ANYPREVOUT. We can do more and=
 do it better, we could decide, how to calculate this txid at all!

So, something like SIGHASH_PREVOUT_NONE would be similar to SIGHASH_NONE =
(applied to the previous transaction, taken from txid). To have =
SIGHASH_ANYPREVOUT, we need to remove absolutely everything, I don't know =
any such sighashes, because even SIGHASH_NONE | SIGHASH_ANYONECANPAY will =
commit at least to some fields, for example to the locktime. But, if we =
introduce SIGHASH_PREVOUT_XYZ flags for all existing sighashes, we would =
have this:

SIGHASH_PREVOUT_NONE
SIGHASH_PREVOUT_SINGLE
SIGHASH_PREVOUT_ALL
SIGHASH_PREVOUT_ANYONECANPAY

Then, the procedure is as follows: we use =
txid:vout to find our previous transaction. Then, we apply those sighashes =
to this previous transaction, to form a new txid, that will be checked =
during every OP_CHECKSIG-based opcode. In this way, our txid:vout is used =
just to do transaction lookup, after that, sighashes can be applied to the =
previous transaction, so our txid could remain stable, even if someone will=
 add some inputs and outputs.

By default, we could use SIGHASH_PREVOUT_ALL=
, that would mean our txid:vout remains unchanged. Then, =
SIGHASH_PREVOUT_SINGLE would obviously mean, that we want to commit only to=
 this particular previous transaction output. That would allow adding any =
new outputs to the previous transaction, without affecting our replaced =
txid, but also without blindly accepting any txid, because some data of the=
 previous transaction would be still hashed.

Then, SIGHASH_PREVOUT_NONE is=
 an interesting case, because it would mean that no outputs of the previous=
 transaction are checked. But still, the inputs will be! That would mean: =
"I don't care about in-between addresses, but I care that it was initiated =
from these inputs". In this case, it is possible to choose some input =
without those flags, and then apply SIGHASH_PREVOUT_NONE many times, to =
make sure that everything started from that input, but everything =
in-between can be anything.

All of those three SIGHASH_PREVOUT_XYZ flags =
could be combined with SIGHASH_PREVOUT_ANYONECANPAY. That would mean all =
inputs of the previous transaction are discarded, except from the input =
number matching "vout". Or we could just use SIGHASH_PREVOUT_ANY instead =
and discard all inputs from that previous transaction, that could also be =
combined with other sighashes.

So, to sum up, by applying sighashes to the=
 previous transaction, instead of allowing for any transaction, we could =
still have some control of our txid, and I think it could be better than =
just saying "give me any txid, I will accept that". I think in most cases =
we don't want to allow any txid: we want to only "control the flow", just =
to make sure that our signatures will sign what we want and will not be =
invalidated by changing some transaction inputs and outputs, unrelated to =
the currently-checked signature.