On Sun, Jul 28, 2013 at 07:39:08PM +0000, John Dillon wrote:
> Peter Todd recently came up with two related, and IMO very good, uses for
> non-standard transactions to implement both oracles and one-time-password
> protection of wallet funds. While the wallet fund case could be implemented as
> only a single standard type, at the cost of generality, the oracle case would
> be most useful with more arbitrary rules. More generally it is also useful to
> be able to have scriptPubKeys like the following:
> 
>     n <pubkey>...<pubkey> m CHECKMULTISIG <master pubkey> CHECKSIG BOOLOR
> 
> and many other similar constructions.
> 
> What are your thoughts on creating a whitelist for specific opcodes that would
> apply to scripts serialized using P2SH, retaining the existing standard
> whitelist for scriptPubKeys? (I would still recommend dropping pay-to-pubkey
> and pay-to-multisig due to their potential for dumping data in the UTXO set)

One subtlety of what you are proposing is that we should still retain
the IsStandard() check, or to be exact the AreInputsStandard() check, if
a P2SH serialized script follows a standard form.

The reason is transaction mutability. Right now other than BIP11
CHECKMULTISIG only miners can mutate transactions because any change to
the scriptSig will render the transaction non-standard. As you know this
is a good thing because it means unconfirmed transaction chains don't
get broken in flight.

BIP11 is an interesting case because CHECKMULTISIG consumes one extra
stack item, so when you spend a BIP11 n <pubkey>...<pubkey> m
CHECKMULTISIG scriptPubKey you have to provide an additional item prior
to the signatures; usually OP_0 is used.

But we don't actually check that! You can put anything there provided it
doesn't make the scriptSig go over the standard allowed scriptSig size
of 500 bytes; for instance I (ab)used that feature just now to timestamp
my Litecoin v0.8.3.6 audit report SHA256 hash:

d0dfe270e8e8e4c0196f780d42e34d8a1121f2f8a249586aa1a2c5ebcada10b1

in transaction:

15bb08318335f94a8de154dc39b03db2cdebcc7a96ab6cec0379978676d00301

It's been suggested that we consider transactions non-standard, or just
now allowed at all in a future soft-fork, if at the end of execution
there is more than one stack item left; a opcode whitelist should
probably do this. On the other hand I've come up with some soft-fork
upgrade mechanisms that would leave extra items on the stack for
non-upgraded nodes, suggesting a soft-fork imposing this is a bad idea.
(though note how it suggests considering such tx's non-standard is
reasonable in a few ways)

CHECKMULTISIG isn't helped here because the value really is ignored - a
soft-fork to force it to always be zero might not be a bad idea, though
it's far from the only example of mutability.

I'd be interested if you can come up with an example where imposing a
one stack item at the end of execution rule causes problems.


More generally, and getting a bit off topic, I think Bitcoin should have
been designed so that CHECKSIG signed hashes of scriptPubKeys, rather
than txid:vout outputs, so that malleability wouldn't affect the
validity of a signature. Of course, this would mean that signatures
could be reused if scriptPubKeys were reused, but address re-use is a
bad thing anyway! Not that I'll fault Satoshi here, type 2 deterministic
wallets were unknown back then. (though we should be careful that a
future CHECKSIG design can go back to txid:vout references - ECC is
unique in allowing for type 2 wallets)

-- 
'peter'[:-1]@petertodd.org
0000000000000053ef658095fb45c7a86955d70c76b44264c7abce79683a8a90