[bitcoin-dev] Annex Purpose Discussion: OP_ANNEX, Turing Completeness, and other considerations

[Jeremy Rubin <jeremy.l.rubin@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 7776 bytes --]

I've seen some discussion of what the Annex can be used for in Bitcoin. For
example, some people have discussed using the annex as a data field for
something like CHECKSIGFROMSTACK type stuff (additional authenticated data)
or for something like delegation (the delegation is to the annex). I think
before devs get too excited, we should have an open discussion about what
this is actually for, and figure out if there are any constraints to using
it however we may please.

The BIP is tight lipped about it's purpose, saying mostly only:

*What is the purpose of the annex? The annex is a reserved space for future
extensions, such as indicating the validation costs of computationally
expensive new opcodes in a way that is recognizable without knowing the
scriptPubKey of the output being spent. Until the meaning of this field is
defined by another softfork, users SHOULD NOT include annex in
transactions, or it may lead to PERMANENT FUND LOSS.*

*The annex (or the lack of thereof) is always covered by the signature and
contributes to transaction weight, but is otherwise ignored during taproot
validation.*

*Execute the script, according to the applicable script rules[11], using
the witness stack elements excluding the script s, the control block c, and
the annex a if present, as initial stack.*

Essentially, I read this as saying: The annex is the ability to pad a
transaction with an additional string of 0's that contribute to the virtual
weight of a transaction, but has no validation cost itself. Therefore,
somehow, if you needed to validate more signatures than 1 per 50 virtual
weight units, you could add padding to buy extra gas. Or, we might somehow
make the witness a small language (e.g., run length encoded zeros) such
that we can very quickly compute an equivalent number of zeros to 'charge'
without actually consuming the space but still consuming a linearizable
resource... or something like that. We might also e.g. want to use the
annex to reserve something else, like the amount of memory. In general, we
are using the annex to express a resource constraint efficiently. This
might be useful for e.g. simplicity one day.

Generating an Annex: One should write a tracing executor for a script, run
it, measure the resource costs, and then generate an annex that captures
any externalized costs.

-------------------

Introducing OP_ANNEX: Suppose there were some sort of annex pushing opcode,
OP_ANNEX which puts the annex on the stack as well as a 0 or 1 (to
differentiate annex is 0 from no annex, e.g. 0 1 means annex was 0 and 0 0
means no annex). This would be equivalent to something based on <annex
flag> OP_TXHASH <has annex flag> OP_TXHASH.

Now suppose that I have a computation that I am running in a script as
follows:

OP_ANNEX
OP_IF
    `some operation that requires annex to be <1>`
OP_ELSE
    OP_SIZE
    `some operation that requires annex to be len(annex) + 1 or does a
checksig`
OP_ENDIF

Now every time you run this, it requires one more resource unit than the
last time you ran it, which makes your satisfier use the annex as some sort
of "scratch space" for a looping construct, where you compute a new annex,
loop with that value, and see if that annex is now accepted by the program.

In short, it kinda seems like being able to read the annex off of the stack
makes witness construction somehow turing complete, because we can use it
as a register/tape for some sort of computational model.

-------------------

This seems at odds with using the annex as something that just helps you
heuristically guess  computation costs, now it's somehow something that
acts to make script satisfiers recursive.

Because the Annex is signed, and must be the same, this can also be
inconvenient:

Suppose that you have a Miniscript that is something like: and(or(PK(A),
PK(A')), X, or(PK(B), PK(B'))).

A or A' should sign with B or B'. X is some sort of fragment that might
require a value that is unknown (and maybe recursively defined?) so
therefore if we send the PSBT to A first, which commits to the annex, and
then X reads the annex and say it must be something else, A must sign
again. So you might say, run X first, and then sign with A and C or B.
However, what if the script somehow detects the bitstring WHICH_A WHICH_B
and has a different Annex per selection (e.g., interpret the bitstring as a
int and annex must == that int). Now, given and(or(K1, K1'),... or(Kn,
Kn')) we end up with needing to pre-sign 2**n annex values somehow... this
seems problematic theoretically.

Of course this wouldn't be miniscript then. Because miniscript is just for
the well behaved subset of script, and this seems ill behaved. So maybe
we're OK?

But I think the issue still arises where suppose I have a simple thing
like: and(COLD_LOGIC, HOT_LOGIC) where both contains a signature, if
COLD_LOGIC and HOT_LOGIC can both have different costs, I need to decide
what logic each satisfier for the branch is going to use in advance, or
sign all possible sums of both our annex costs? This could come up if
cold/hot e.g. use different numbers of signatures / use checksigCISAadd
which maybe requires an annex argument.



------------

It seems like one good option is if we just go on and banish the OP_ANNEX.
Maybe that solves some of this? I sort of think so. It definitely seems
like we're not supposed to access it via script, given the quote from above:

*Execute the script, according to the applicable script rules[11], using
the witness stack elements excluding the script s, the control block c, and
the annex a if present, as initial stack.*

If we were meant to have it, we would have not nixed it from the stack, no?
Or would have made the opcode for it as a part of taproot...

But recall that the annex is committed to by the signature.

So it's only a matter of time till we see some sort of Cat and Schnorr
Tricks III the Annex Edition that lets you use G cleverly to get the annex
onto the stack again, and then it's like we had OP_ANNEX all along, or
without CAT, at least something that we can detect that the value has
changed and cause this satisfier looping issue somehow.

Not to mention if we just got OP_TXHASH



-----------

Is the annex bad? After writing this I sort of think so?

One solution would be to... just soft-fork it out. Always must be 0. When
we come up with a use case for something like an annex, we can find a way
to add it back.  Maybe this means somehow pushing multiple annexes and
having an annex stack, where only sub-segments are signed for the last
executed signature? That would solve looping... but would it break some
aggregation thing? Maybe.


Another solution would be to make it so the annex is never committed to and
unobservable from the script, but that the annex is always something that
you can run get_annex(stack) to generate the annex. Thus it is a hint for
validation rules, but not directly readable, and if it is modified you
figure out the txn was cheaper sometime after you execute the scripts and
can decrease the value when you relay. But this sounds like something that
needs to be a p2p only annex, because consensus we may not care (unless
it's something like preallocating memory for validation?).

-----------------------

Overall my preference is -- perhaps sadly -- looking like we should
soft-fork it out of our current Checksig (making the policy that it must 0
a consensus rule) and redesign the annex technique later when we actually
know what it is for with a new checksig or other mechanism. But It's not a
hard opinion! It just seems like you can't practically use the annex for
this worklimit type thing *and* observe it from the stack meaningfully.



Thanks for coming to my ted-talk,

Jeremy


--
@JeremyRubin <https://twitter.com/JeremyRubin>

[-- Attachment #2: Type: text/html, Size: 14856 bytes --]