Brandon Black <freedom@reardencode.com> writes:
> Hey list,
>
> On 2024-12-09 (Mon) at 19:08:51 +0000, Andrew Poelstra wrote:
>> On Mon, Dec 09, 2024 at 05:27:54AM -0800, Weikeng Chen wrote:
>> > When I am implementing fraud proofs in Bitcoin script, I find it useful to
>> > have an opcode "OP_SUCCESS" that will mark the execution to be successful
>> > without running the rest of the script, if this opcode is being executed.
>> > This is useful for writing code for fraud proofs such as BitVM, where the
>> > verifier wins if it finds one mismatch, and the verifier does not need to
>> > show the other mismatches.
>> >
>> > This OP_SUCCESS is weaker version of the OP_SUCCESSx in the Taproot
>> > upgrade, which marks the execution as successful for the mere presence of
>> > OP_SUCCESSx anywhere in the script. Rusty Russell in a 2023 article,
>> > "Covenants: Examining ScriptPubkeys in Bitcoin Script", also mentioned
>> > about the usefulness of such an opcode.
>> >
>> > <snip>
>>
>> In short, for purpose of softforking upgrade mechanism, the existing
>> SUCCESS codes give us way more freedom of action.
>>
>> But it sounds like you want a "weak SUCCESS" opcode in order to use the
>> success semantics, not as an upgrade mechanism. Maybe it makes sense to
>> propose that one of the existing OP_SUCCESSx opcodes should be
>> softforked to become OP_WEAK_SUCCESS?
>
> An alternative that Rusty Russel has discussed wanting as part of his
> script restoration work is "OP_SEGMENT" which would split the script
> execution for purposes of SUCCESS checking, allowing (for example) a
> prefix to be required to execute before an arbitrary user provided
> script that might contain an OP_SUCCESS.
>
> It occurred to me today when thinking about Weikeng's post that we can
> slightly weaken the existing OP_SUCCESS behavior while retaining
> essentially all of its benefits in practice without introducing
> OP_SEGMENT by leveraging OP_CODESEPARATOR. Redefine OP_SUCCESS with a
> soft fork from "make the script unconditionally valid" to "make the
> script segment unconditionally valid", and define a script segment as
> "each lexicographic section of the script containing no
> OP_CODESEPARATOR".
>
> The script interpreter can perform SUCCESS checking as it currently does
> until it encounters an OP_CODESEPARATOR. Each OP_CODESEPARATOR gets a
> "SUCCESS" flag defaulted to false and SUCCESS checking now sets that
> flag to true on the most recently encountered OP_CODESEPARATOR.
>
> During script execution, whenever an OP_CODESEPARATOR is popped (not
> executed) its "SUCCESS" flag value is copied to the interpreter state.
> After this state setting conditional, if the interpreter "SUCCESS" flag
> is true, and fExec is true, the script immediately succeeds.

Beware success inside branches?  This is why I preferred to segment the
script and scan for OP_SUCCESS and evaluate each part in order (if you
have part of an if statement inside one segment, you fail as expected).
This is actually not that different inside Bitcoin's script.cpp.

But that's kind of a detail.  IMHO there's nothing fundamentally wrong
with runtime success opcodes, in fact several proposals work better if
you allow them (e.g. "undefined bit patterns in operand to OP_TXHASH
cause immediate success" lets you reserve some bits for future
extension).

(Also: OP_CODESEPARATOR is cursed, so I chose a different name :)

Cheers,
Rusty.