[Bitcoin-development] Relative CHECKLOCKTIMEVERIFY (was CLTV proposal)
pete at petertodd.org
Tue Apr 21 07:59:12 UTC 2015
On Mon, Mar 16, 2015 at 10:22:13PM +0000, Matt Corallo wrote:
> In building some CLTV-based contracts, it is often also useful to have a
> method of requiring, instead of locktime-is-at-least-N,
> locktime-is-at-least-N-plus-the-height-of-my-input. ie you could imagine
> an OP_RELATIVECHECKLOCKTIMEVERIFY that reads (does not pop) the top
> stack element, adds the height of the output being spent and then has
> identical semantics to CLTV.
Depending on what you mean by "identical" this isn't actually reorg
safe. For instance consider this implementation:
nLockTime = stack[-1] + prevout.nHeight
if (nLockTime > txTo.nLockTime):
Used with this scriptPubKey:
10 RCLTV DROP <pubkey> CHECKSIG
If I create that output in tx1 which is mined at height 42 I can spend
it in a tx2 at height > 42+10 by setting tx2's nLockTime to >42+10, for
instance 53. However if a reorg happens and tx1 ends up at height 43
after the reorg I'm stuck - tx2's nLockTime is set at 42.
Thus RCTLV is only reorg safe if the height is compared against the
actual block height of the block containing the spending transaction,
not the spending transaction's nLockTime.
> A slightly different API (and different name) was described by maaku at
> which does a better job of saving softfork-available opcode space.
> There are two major drawbacks to adding such an operation, however.
> 1) More transaction information is exposed inside the script (prior to
> CLTV we only had the sigchecking operation exposed, with a CLTV and
> RCLTV/OP_CHECK_MATURITY_VERIFY we expose two more functions).
> 2) Bitcoin Core's mempool invariant of "all transactions in the mempool
> could be thrown into one overside block and aside from block size, it
> would be valid" becomes harder to enforce. Currently, during reorgs,
> coinbase spends need checked (specifically, anything spending THE
> coinbase 100 blocks ago needs checked) and locktime transactions need
> checked. With such a new operation, any script which used this new
> opcode during its execution would need to be re-evaluated during reorgs.
Yup, definitely kinda ugly.
If the above style of RCTLV was used, one possibility might be to make
the relative locktime difference be required to be at least 100 blocks,
same as the coinbase maturity, and just accept that it's probably not
going to cause any problems, but could in an extremely big reorg. But
re-orgs that big might be big enough that we're screwed anyway...
With the 100 block rule, during a sufficiently large reorg that
coinbases become unavailble, simply disconnect entire blocks - all
txouts created by them.
> I think both of these requirements are reasonable and not particularly
> cumbersome, and the value of such an operation is quite nice for some
> protocols (including settings setting up a contest interval in a
> sidechain data validation operation).
So to be clear, right now the minimal interface to script execution is
int bitcoinconsensus_verify_script(const unsigned char *scriptPubKey, unsigned int scriptPubKeyLen,
const unsigned char *txTo , unsigned int txToLen,
unsigned int nIn, unsigned int flags, bitcoinconsensus_error* err);
Where scriptPubKey is derived from the unspent coin in the UTXO set and
txTo is the transaction containing the script that is being executed.
The UTXO set itself currently contains CCoins entries, one for each
transaction with unspent outputs, which basically contain:
nVersion - tx nVersion
nHeight - Height of the block the transaction is contained in.
vout - Unspent CTxOut's of the transaction.
The block nTime isn't directly available through the UTXO set, although
it can be found in the block headers. This does require nodes to have
the block headers, but at 4MB/year growth it's reasonable to assume the
UTXO set will grow faster.
Script execution does not have direct access to the current block
height/block time, however it does have indirect access via nLockTime.
Thus we have a few possibilities:
1) RCLTV against nLockTime
Needs a minimum age > COINBASE_MATURITY to be safe.
2) RCLTV against current block height/time
Completely reorg safe.
3) GET_TXOUT_HEIGHT/TIME <diff> ADD CLTV
To be reorg safe GET_TXOUT_HEIGHT/TIME must fail if minimum age <
COINBASE_MATURITY. This can be implemented by comparing against
All three possibilities require us to make information about the
prevout's height/time available to VerifyScript(). The only question is
if we want VerifyScript() to also take the current block height/time - I
see no reason why it can't. As for the mempool, keeping track of what
transactions made use of these opcodes so they can be reevaluated if
their prevouts are re-organised seems fine to me.
If we are going to make the block height/time available to
VerifyScript() to implement RCLTV, should absolute CLTV should continue
to have the proposed behavior of checking against nLockTime? If we go
with RCLTV against current block height/time, I'm going to vote no,
because doing so needlessly limits it to only being able to compare
against a block height or a block time in a single transaction.
Similarly it can complicate multi-party signatures in some
circumstances, as all parties must agree on a common nLockTime.
Do we want to support them at all? May cause incentive issues with
mining, see #bitcoin-wizards discussion, Jul 17th 2013:
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 650 bytes
Desc: Digital signature
More information about the bitcoin-dev