[bitcoin-dev] [PROPOSAL] OP_TX: generalized covenants reduced to OP_CHECKTEMPLATEVERIFY

From: Rusty Russell <rusty@rustcorp•com.au>
To: "Bitcoin Protocol Discussion" <bitcoin-dev@lists•linuxfoundation.org>
Subject: [bitcoin-dev] [PROPOSAL] OP_TX: generalized covenants reduced to OP_CHECKTEMPLATEVERIFY
Date: Tue, 10 May 2022 20:05:54 +0930	[thread overview]
Message-ID: <87h75xoet1.fsf@rustcorp.com.au> (raw)

Hi all,

	TL;DR: a v1 tapscript opcode for generic covenants, but
OP_SUCCESS unless it's used a-la OP_CHECKTEMPLATEVERIFY.  This gives an
obvious use case, with clean future expansion.  OP_NOP4 can be
repurposed in future as a shortcut, if experience shows that to be a
useful optimization.

(This proposal builds on Russell O'Connor's TXHASH[1], with Anthony
Towns' modification via extending the opcode[2]; I also notice on
re-reading that James Lu had a similar restriction idea[3]).

Details
-------

OP_TX, when inside v1 tapscript, is followed by 4 bytes of flags.
Unknown flag patterns are OP_SUCCESS, though for thoroughness some future
potential uses are documented here.  Note that pushing more than 1000
elements on the stack or an element more than 512 bytes will hit the
BIP-342 resource limits and fail.

Defined bits
------------

(Only those marked with * have to be defined for this soft fork; the
 others can have semantics later).

OPTX_SEPARATELY: treat fields separately (vs concatenating)
OPTX_UNHASHED: push on the stack without hashing (vs SHA256 before push)

- The first nicely sidesteps the lack of OP_CAT, and the latter allows
  OP_TXHASH semantics (and avoid stack element limits).

OPTX_SELECT_VERSION*: version
OPTX_SELECT_LOCKTIME*: nLocktime
OPTX_SELECT_INPUTNUM*: current input number
OPTX_SELECT_INPUTCOUNT*: number of inputs
OPTX_SELECT_OUTPUTCOUNT*: number of outputs

OPTX_INPUT_SINGLE: if set, pop input number off stack to apply to
		OPTX_SELECT_INPUT_*, otherwise iterate through all.
OPTX_SELECT_INPUT_TXID: txid
OPTX_SELECT_INPUT_OUTNUM: txout index
OPTX_SELECT_INPUT_NSEQUENCE*: sequence number
OPTX_SELECT_INPUT_AMOUNT32x2: sats in, as a high-low u31 pair
OPTX_SELECT_INPUT_SCRIPT*: input scriptsig
OPTX_SELECT_INPUT_TAPBRANCH: ?
OPTX_SELECT_INPUT_TAPLEAF: ?

OPTX_OUTPUT_SINGLE: if set, pop input number off stack to apply to
		OPTX_SELECT_OUTPUT_*, otherwise iterate through all.
OPTX_SELECT_OUTPUT_AMOUNT32x2*: sats out, as a high-low u31 pair
OPTX_SELECT_OUTPUT_SCRIPTPUBKEY*: output scriptpubkey

OPTX_SELECT_19...OPTX_SELECT_31: future expansion.

OP_CHECKTEMPLATEVERIFY is approximated by the following flags:
	OPTX_SELECT_VERSION
	OPTX_SELECT_LOCKTIME
	OPTX_SELECT_INPUTCOUNT
	OPTX_SELECT_INPUT_SCRIPT
	OPTX_SELECT_INPUT_NSEQUENCE
	OPTX_SELECT_OUTPUTCOUNT
	OPTX_SELECT_OUTPUT_AMOUNT32x2
	OPTX_SELECT_OUTPUT_SCRIPTPUBKEY
	OPTX_SELECT_INPUTNUM

All other flag combinations result in OP_SUCCESS.

Discussion
----------

By enumerating exactly what can be committed to, it's absolutely clear
what is and isn't committed (and what you need to think about!).

The bits which separate concatenation and hashing provide a simple
mechanism for template-style (i.e. CTV-style) commitments, or for
programatic treatment of individual elements (e.g. amounts, though the
two s31 style is awkward: a 64-bit push flag could be added in future).

The lack of double-hashing of scriptsigs and other fields means we
cannot simply re-use hashing done for SIGHASH_ALL.

The OP_SUCCESS semantic is only valid in tapscript v1, so this does not
allow covenants for v0 segwit or pre-segwit inputs.  If covenants prove
useful, dedicated opcodes can be provided for those cases (a-la
OP_CHECKTEMPLATEVERIFY).

Cheers,
Rusty.

[1] https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2022-January/019813.html
[2] https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2022-January/019819.html
[3] https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2022-January/019816.html