Hello list, TL;DR: the core opcodes of MATT can build vaults with a very similar design to OP_VAULT. Code example here: https://github.com/bitcoin-inquisition/bitcoin/compare/24.0...bigspider:bitcoin-inquisition:matt-vault In my previous emails about the MATT proposal for smart contracts in bitcoin [1], I mostly focused on proving its generality; that is, it allows arbitrary smart contracts thanks to fraud proofs. While I still find this "completeness" result compelling, I spent more time thinking about the framework itself; the construction is not very interesting if it turns simple things into complicated ones. Luckily, this is not the case. In particular, in this email we will not merkleize anything (other than taptrees). This post describes some progress into formalizing the semantics of the core opcodes, and demonstrates how they could be used to create vaults that seem comparable to the ones built with OP_VAULT [2], despite using general purpose opcodes. An implementation and some minimal tests matching the content of this e-mail can be found in the link above, using the bitcoin-inquisition as the base branch. Note that the linked code is not well tested and is only intended for exploratory and demonstrative purposes; therefore, bugs are likely at this stage. ########################## # PART 1: MATT's core ########################## In this section, I will discuss plausible semantics for the core opcodes for MATT. The two core opcodes are defined below as OP_CHECKINPUTCONTRACTVERIFY and OP_CHECKOUTPUTCONTRACTVERIFY. (the initial posts named them OP_CHECK{INPUT,OUTPUT}COVENANTVERIFY) They enhance Script with the following capabilities: - decide the taptree of the output - embed some (dynamically computed) data in the output - access the embedded data in the current UTXO (if any) The opcodes below are incomplete, as they only control the output's Script and not the amounts; more on that below. Other than that, the semantics should be quite close to the "right" one for the MATT framework. ### The opcodes case OP_CHECKINPUTCONTRACTVERIFY: { // OP_CHECKINPUTCONTRACTVERIFY is only available in Tapscript if (sigversion == SigVersion::BASE || sigversion == SigVersion::WITNESS_V0) return set_error(serror, SCRIPT_ERR_BAD_OPCODE); // (x d -- ) if (stack.size() < 2) return set_error(serror, SCRIPT_ERR_INVALID_STACK_OPERATION); valtype& x = stacktop(-2); valtype& d = stacktop(-1); if (x.size() != 32 || d.size() != 32) return set_error(serror, SCRIPT_ERR_INVALID_STACK_OPERATION); const XOnlyPubKey nakedXOnlyKey{Span{x.data(), x.data() + 32}}; const uint256 data(d); if (!execdata.m_internal_key.has_value()) return set_error(serror, SCRIPT_ERR_UNKNOWN_ERROR); // TODO // Verify that tweak(lift_x(x), d) equals the internal pubkey if (!execdata.m_internal_key.value().CheckDoubleTweak(nakedXOnlyKey, &data, nullptr)) return set_error(serror, SCRIPT_ERR_WRONGCONTRACTDATA); popstack(stack); popstack(stack); } break; case OP_CHECKOUTPUTCONTRACTVERIFY: { // OP_CHECKOUTPUTCONTRACTVERIFY is only available in Tapscript if (sigversion == SigVersion::BASE || sigversion == SigVersion::WITNESS_V0) return set_error(serror, SCRIPT_ERR_BAD_OPCODE); // (out_i x taptree d -- ) if (stack.size() < 4) return set_error(serror, SCRIPT_ERR_INVALID_STACK_OPERATION); int out_i = CScriptNum(stacktop(-4), fRequireMinimal).getint(); valtype& x = stacktop(-3); valtype& taptree = stacktop(-2); valtype& d = stacktop(-1); auto outps = checker.GetTxvOut(); // Return error if the evaluation context is unavailable if (!outps) return set_error(serror, SCRIPT_ERR_UNKNOWN_ERROR); // TODO if (x.size() != 32 || taptree.size() != 32 || (d.size() != 0 && d.size() != 32)) return set_error(serror, SCRIPT_ERR_INVALID_STACK_OPERATION); if (out_i < 0 || out_i >= (int)outps->size()) return set_error(serror, SCRIPT_ERR_INVALID_STACK_OPERATION); const XOnlyPubKey nakedXOnlyKey{Span{x.data(), x.data() + 32}}; const uint256 data(d); const uint256 *data_ptr = (d.size() == 0 ? nullptr : &data); const uint256 merkle_tree(taptree); CScript scriptPubKey = outps->at(out_i).scriptPubKey; if (scriptPubKey.size() != 1 + 1 + 32 || scriptPubKey[0] != OP_1 || scriptPubKey[1] != 32) return set_error(serror, SCRIPT_ERR_WRONGCONTRACTDATA); const XOnlyPubKey outputXOnlyKey{Span{scriptPubKey.data() + 2, scriptPubKey.data() + 34}}; // Verify that taptweak(tweak(lift_x(x), d), taptree) equals the internal pubkey if (!outputXOnlyKey.CheckDoubleTweak(nakedXOnlyKey, data_ptr, &merkle_tree)) return set_error(serror, SCRIPT_ERR_WRONGCONTRACTDATA); popstack(stack); popstack(stack); popstack(stack); popstack(stack); } break; ### Commentary CheckDoubleTweak function (implemented in the branch) gets an x-only pubkey, optionally some data, and optionally taptree's merkle root. It verifies that the x-only pubkey being tested equals the given naked pubkey, optionally tweaked with the embedded data, optionally tweaked with the tagged hash of the merkle tree per BIP-0341 [3]. Making both the tweaks optional allows to simplify the code, and also to obtain more compact scripts in some spending paths. In words: - OP_CHECKINPUTCONTRACTVERIFY: verify that the current input's internal key contains some embedded data (which would typically be passed through the witness stack) - OP_CHECKOUTPUTCONTRACTVERIFY: verify that a given output is a certain P2TR output script containing the desired embedded data. TBD if the tweaking used for the embedded data tweak should use a tagged hash; omitted for simplicity in this demo implementation. ### Amount preservation In the code above and in the linked demo implementation, the opcodes only operate on the scriptPubkey; a complete implementation would want to make sure that amounts are correctly preserved. The most direct and general way to address this would be to allow direct introspection on the output amounts. This has the complication that output amounts require 64-bits arithmetics, as discussed in the context of other proposals, for example: [4]. One more limited approach that works well for many interesting contracts is that of the deferred checks, implemented in OP_VAULT [2]. The idea is that all the amounts of the inputs that commit to the same output script with OP_CHECKOUTPUTCONTRACTVERIFY are added together, and the script interpreter requires that the amount of that output is not smaller than the total amount of those inputs. This check is therefore transaction-wide rather than being tested during the input's script evaluation. This behaviour is adequate for vaults and likely suitable for many other applications; however, it's not the most general approach. I didn't try to implement it yet, and defer the decision on the best approach to a later time. ### Extensions The opcodes above are not enough for the full generality of MATT: one would need to add an opcode like OP_SHA256CAT to allow the data embedding to commit to multiple pieces of data. This is not used in today's post, therefore I left it out of these code examples. It would be easy to extend OP_CHECKOUTPUTCONTRACTVERIFY to also apply for an arbitrary input (typically, different from the currently executed one); there are likely use cases for that, allowing to define contracts with more complex cross-input semantics, but I preferred to keep things simple. Of course, one could also entirely replace CICV/COCV with generic full introspection on inputs/output's program, plus opcodes for elliptic curve math and tagged hashes. ########################## # PART 2: Vaults with MATT ########################## In the rest of this post, I will document the first attempt at creating a vault using the opcodes described. While not an attempt at cloning exactly the functionality of OP_VAULT [2], it borrows heavily from the excellent work that was done there. In particular, it also inherits the choice of using OP_CTV as a primitive, building on top of the bitcoin-inquisition's current branch that has already merged OP_CTV. Reasonable vaults would be possible without CTV, but they would be less efficient, particularly in the case of sending to many addresses in a single unvaulting flow. ### Distilling OP_VAULT Abstracting from the implementation details, I mentally model a vault as a simple state machine with 2 states: [V] and [U]: [V]: the initial vault UTXO(s); [U]: the utxo produced by the "trigger transaction" during unvaulting. On the typical path: one or more [V] UTXOs are sent to the [U] state, and after a timelock set on [U] expires, [U] is spent to one or several destinations. Crucially, the destination outputs and amounts are already decided when [V] is spent into [U]. At any time before the funds are spent from [U], they can always be spent by sending them to some specified recovery path. There are two key elements that are part of OP_VAULT's semantics, and could be generalized: − Forcing the script/amount of the next stepon − Storing some data for future Script's access (in the vault's case, a hash that commits to the final withdrawal transaction). CICV/COCV generalize both to arbitrary scripts (taptrees) and state machines, and to dynamical and witness-dependent data embedded in the pubkey of a P2TR output. ### Vault parameters A contract that represents a vault has the following parameters (hardcoded in the script when the vault is created): - alternate_pk: a key that can be used any time. - spend_delay: the relative timelock before the withdrawal can be finalized; - recover_pk: a pubkey for a P2TR output where funds can be sent at any time. The alternate_pk is a pubkey that can optionally be used as the key-path spending condition for both states [V] and [U]. If such a spending condition is not desired, it can be replaced with a NUMS point, making the key-path unspendable. The spend_delay is the number of blocks that must be mined before the final withdrawal transaction In this example we also use an unvault_pk needed to authorize the unvaulting process (that is, spend [V] into [U]); this could be replaced with any miniscript or other conditions expressible in Script. ### P2TR structure for [V] (vault) internal key: alternate_pk Script 1: "trigger" # witness: { , , 2, OP_ROLL, OP_CHECKOUTPUTCONTRACTVERIFY, OP_CHECKSIG } Script 2: "recover" # witness: { recover_pk, OP_0, # no data tweak OP_0, # no taptweak OP_CHECKOUTPUTCONTRACTVERIFY, OP_TRUE } The "trigger" script requires in the witness an output index and the ctv-hash that describes the withdrawal transaction. COCV forces the output to contain the ctv-hash as embedded data. That's followed by the unvaulting condition − in this example, a simple signature check. The "recover" script doesn't require any signature, and it simply forces the output specified in the witness to be a P2TR output with recover_pk as its pubkey. (Omitting the "recover" script in [V] would reduce the size of the witness by 32 bytes in the expected case, and might be preferred for some users) ### P2TR structure for [U] (unvaulting state) internal key: alternate_pk (tweaked with ctv_hash) Script 1: "withdrawal" # witness: { OP_DUP, # check that the top of the stack is the # embedded data in the current input , OP_SWAP, OP_CHECKINPUTCONTRACTVERIFY, # Check timelock , OP_CHECKSEQUENCEVERIFY, OP_DROP, # Check that the transaction output is as expected OP_CHECKTEMPLATEVERIFY } Script 2: "recover" # witness: { , OP_0, OP_0, OP_CHECKOUTPUTCONTRACTVERIFY, OP_TRUE } The "withdrawal" finalizes the transaction, by checking that the timelock expired and the outputs satisfy the CTV hash that was committed to in the previous transaction. The "recover" script is identical as before. ### Differences with OP_VAULT vaults Here I refer to the latest version of OP_VAULT at the time of writing. [5] It is not a thorough analysis. Unlike the implementation based on OP_VAULT, the [V] utxos don't have an option to add an additional output that is sent back to the same exact vault. Supporting this use case seems to require a more general way of handling the distribution of amounts than what I discussed in the section above: that would in fact need to be generalized to the case of multiple OP_CHECKOUTPUTCONTRACTVERIFY opcodes executed for the same input. By separating the ctv-hash (which is considered "data") from the scripts in the taptree, one entirely avoids the need to dynamically create taptrees and replace leaves in the covenant-encumbered UTXOs; in fact, the taptrees of [V] and [U] are already set in stone when [V] utxos are created, and only the "data" portion of [U]'s scriptPubKey is dynamically computed. In my opinion, this makes it substantially easier to program "state machines" that control the behavior of coins, of which vaults are a special case. I hope you'll find this interesting, and look forward to your comments. Salvatore Ingala [1] - https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2022-November/021223.html [2] - https://github.com/bitcoin/bips/pull/1421 [3] - https://github.com/bitcoin/bips/blob/master/bip-0341.mediawiki [4] - https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2021-September/019420.html [5] - https://github.com/bitcoin/bips/blob/7112f308b356cdf0c51d917dbdc1b98e30621f80/bip-0345.mediawiki