BIPs for MuSig2 derivation, descriptors, and PSBT fields

[achow101]

This PR contains 3 proposed BIPs, the main contents of which were sent to the bitcoin-dev mailing list in October. There have been a few changes, notably the addition of a new BIP for the derivation methodology which was split from the descriptors BIP.

I've also changed the PSBT BIP to use 33 byte plain aggregate pubkeys rather than xonly.

[benma]

I've also changed the PSBT BIP to use 33 byte plain aggregate pubkeys rather than xonly.

It would be helpful to document the reasoning in the commit msg or here in the PR to have a record of design decisions.

[achow101]

I've also changed the PSBT BIP to use 33 byte plain aggregate pubkeys rather than xonly.

It would be helpful to document the reasoning in the commit msg or here in the PR to have a record of design decisions.

I've added a rationale section which addresses this.

[Sjors]

An additional benefit is that such wallets can be watched by a musig2-unaware wallet, or a tapleaf co-signer who does not need to know about the individual keys, by casting the tr(musig()) descriptor to tr(). Similarly it's compatible with the privacy-preserving rawtr descriptor (needs a BIP).

[achow101]

I've made a minor change with the descriptors regarding sorting. Keys in a musig() can be specified in any order, but will be sorted prior to aggregation, similar to sortedmulti(). However for PSBTs, keys must be supplied in the expected order for aggregation (this is unchanged), which means sorting them if they weren't already.

[bigspider]

I've made a minor change with the descriptors regarding sorting. Keys in a musig() can be specified in any order, but will be sorted prior to aggregation, similar to sortedmulti(). However for PSBTs, keys must be supplied in the expected order for aggregation (this is unchanged), which means sorting them if they weren't already.

What is the rationale for this change? It seems to me that this complicates parsing/processing the descriptor unnecessarily. The only practical advantage I know of sortedmulti versus multi is that with multi an external observer can possibly fingerprint who signed different UTXOs even without knowing the descriptor; but that's not the case anyway for musig.

[achow101]

What is the rationale for this change?

I had a discussion involving @jonasnick and @sipa about this. Since BIP 327 suggests to do sorting, we decided it made sense to do in descriptors as well.

[bigspider]

I had a discussion involving @jonasnick and @sipa about this. Since BIP 327 suggests to do sorting, we decided it made sense to do in descriptors as well.

BIP 327 discusses the possibility of sorting and describes a canonical sorting, but I didn't read it as 'suggesting' one way or another.

In general, I'd prefer maximum simplicity and less chances of malleability in descriptors, and this imho goes against both (albeit slightly, not a huge deal).

Note that this removes functionality, as one of $n! - 1$ of the $n!$ possible combinations become inexpressible with descriptors.

[bigspider]

The functionality that I would suggest to consider for removal is derivation before aggregation: https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2023-November/022132.html

Unless there are use cases for it that I'm not aware of?

[rot13maxi]

have you thought about putting [] or something around the final hash here to make it clear that it will be omitted if the aggregate key is either the taproot output key or the taproot internal key

[rot13maxi]

the specification for the descriptor says that keys must be sorted with KeySort. If unsorted keys are allowed here, then its possible to have a psbt for a musig aggregate key that's not expressible in a descriptor. Maybe that's ok, just flagging it.

[achow101]

Yes, that is allowed. PSBTs is intentionally less restrictive than descriptors.

[Ademan]

What's wrong with taking the ordering from the descriptor directly? That seems simpler and more flexible (Not that I can imagine a specific reason why a user would want a particular order).

[rot13maxi]

Yes, that is allowed. PSBTs is intentionally less restrictive than descriptors.

makes sense, you want PSBTs to be a super-set of what you can express in a descriptor

[achow101]

What's wrong with taking the ordering from the descriptor directly? That seems simpler and more flexible (Not that I can imagine a specific reason why a user would want a particular order).

There's no descriptor in psbts.

[Ademan]

Right I was commenting on KeySort-ing the participant keys in the descriptor. Seemed on-topic for this thread but I'll put the comment where it belongs.

[Ademan]

Why KeySort when we already have a explicit ordering from the descriptor? It seems like an unnecessary complication.

[achow101]

As I mentioned in #1540 (comment), there was a discussion about this and that was the conclusion. But I'm also having a hard time remembering the exact reason, maybe @jonasnick and/or @sipa can help me out here.

[achow101]

After talking about this again with @sipa, the conclusion is that philosophically, MuSig2 operates over a set of keys, so the order should not matter. For serialization as an implementation detail, it does, so having it sorted is a logical order for an ostensibly unordered set.

Another argument for sorting is to make the recovery process easier,. As we know from multisig usage that exists currently, users are often surprised to learn that they had to know more than just their key and their cosigners. By sorting, we can eliminate this as a concern. Ostensibly, users should be backing up their descriptors, but in practice, especially with the prevalance of hardware signers which espouse that only the seed needs to be backed up, I don't think that's something that happens that often.

[murchandamus]

The formatting seems to be in order on these documents. What is the status of the discussion on these proposals?

[murchandamus]

In #1434 it was mentioned that this proposal may conflict with the way PSBT is used by an existing implementation. Has there been any communication to reconcile that?

[achow101]

In #1434 it was mentioned that this proposal may conflict with the way PSBT is used by an existing implementation. Has there been any communication to reconcile that?

I've changed to skip the conflicting 0x19 fiend number, and have also emailed them about using proprietary fields in the future.

[murchandamus]

I have taken a look at the first commit, MuSig2 BIP 32 derivation BIP.

I noticed a formatting issue and suggested a couple alternative phrasings. The language in this document feels a bit roundabout. Some parts of the Motivation that describe alternative designs would better fit the Rationale section. I would generally recommend the use of active voice in a specification document. It would be great if at least a first test vectors could be added.

I would recommend to split this PR to make each BIP a separate pull request.

[achow101]

It would be great if at least a first test vectors could be added.

Given that there are still a few active questions and some things may still end up being changed, I prefer to wait a bit longer before committing to the current specs with test vectors.

There are a few for the descriptors BIP.

I would recommend to split this PR to make each BIP a separate pull request.

This was considered originally, however there is a dependency on the derivation BIP from both the descriptor and psbt BIPs, so I'm not sure it makes sense to have them be separate.

[bigspider]

BIP-0327 allows participant pubkeys to be duplicates.

Since the psbt fields are using the pubkeys (and not the signer's position in the MuSig) in order to identify the nonce/partial_sig contributions, and given that duplicate pubkeys are AFAIK of no practical value, musig() key expressions should perhaps explicitly forbid repeated pubkeys.