Mark “Murch” Erhardt and Mike Schmidt are joined by Dave Harding, Clara Shikhelman, Peter Todd, Josie Baker, and Eduardo Quintana to discuss Newsletter #263.
The Bitcoin Optech Podcast and transcription content is licensed Creative Commons CC BY-SA 2.0
Bitcoin Core PR Review Club
Releases and release candidates
Notable code and documentation changes
Mike Schmidt: Welcome everyone to Bitcoin Optech Newsletter #263 Recap on Twitter Spaces. It’s Thursday, August 10, and we’ll be talking about a vulnerability in Libbitcoin that has resulted in some loss of funds; a discussion of DoS and Lightning; Clara is going to be speaking about Hash Time Locked Contract (HTLC) endorsements and plans for implementations on Lightning to collect that data; Peter Todd is going to talk about some of his proposed changes to Bitcoin Core’s default relay policy; Josie’s going to join us to walk through Bitcoin Core PR Review Club related to silent payments; and then we have a Bitcoin Development Kit (BDK) release and some notable code changes, including the renepay plugin for Core Lightning (CLN).
We’ll go through some introductions. Mike Schmidt, Contributor at Optech and Executive Director at Brink, funding Bitcoin open-source developers. Murch?
Mark Erhardt: Hey, I’m Murch, I work at Chaincode Labs.
Mike Schmidt: Dave?
Dave Harding: I’m Dave, I help with Optech and I’m also the co-author of a third edition of Mastering Bitcoin.
Mike Schmidt: Peter?
Peter Todd: Hey, so I’ve been working on and off Bitcoin Core for quite a few years, more recently off. But even more recently, I did do two PRs as people have noted, and I’m also involved with timestamping with OpenTimestamps and a bunch of consulting stuff.
Severe Libbitcoin Bitcoin Explorer vulnerability
Libbitcoin Bitcoin Explorer security disclosure
Mike Schmidt: The first item of the newsletter, which this week happens to be an action item, which we have not seen in a while, we have it as an action item and we also cover it in the news section in more detail; I think we could probably cover all of the discussion here initially and then skip it when we go to the news. So Severe Libbitcoin Bitcoin Explorer vulnerability involving the bx seed command, which creates BIP32 seeds. Well, apparently this seed command only used 32 bit of system time to generate the seed, which is an incredibly small space, and it seems that someone has discovered this and has potentially or definitely stolen approximately 30 Bitcoin worth of funds from users using this exploit.
Dave, I saw that you had posted on Twitter about raising awareness for this. I believe maybe some of the researchers who have discovered this reached out to you since you were doing some work on Mastering Bitcoin. Do you want to walk through a bit of the detail there?
Dave Harding: Sure, actually I first heard about it from Murch, who apparently heard about it from one of the researchers. So, first of all, we saw the action item which is if you used Libbitcoin’s bx seed command, or you think you may have used a process that involved it, you definitely want to go check them on your wallet right now. If you don’t know what you’re doing, if you think you might have used software that used it and you’re not aware of the software, go talk to the people who made the software. Just go out there, investigate if you think this may have been involved in your process at all.
So, there were a number of popular instructions, including in the second edition of the book Mastering Bitcoin and also on the Libbitcoin website, that suggested that you use the bx seed command as the first step in a process to creating either private keys directly or creating seeds for hierarchical derivation wallets, or for creating mnemonics that will also be used to seed an HD wallet. And like Mike said in the introduction, it’s only using 32 bits of entry, that’s about 4 billion combinations; it’s possible to check the generated addresses for an HD seed for 4 billion of those in a day or so on a computer. It could be a bit slower, depending on what optimizations you used or not, but it’s possible to do using a single computer in a reasonable amount of time. So it’s very, very insecure.
Most of the instructions, pretty much all the instructions, did not have any disclaimer about the lack of security on this command. There was one set of instructions that did, and the author of the command has claimed that because that one set of instructions had a disclaimer that was pretty hard to understand unless you were already well-versed in cryptography and cryptographic security, and even then, if you were well-versed, it was sort of ambiguous, but yeah, so it was one set of instructions that had a disclaimer, the author has claimed that this is intentional behavior that this is not secure. However, none of the sets of instructions that the authors of Libbitcoin wrote provided a safe way to use the command. So it’s a little disingenuous, I think, to claim that this is intentional behavior. It may be intentioned, but it was not very helpful to the users, and as I said in the introduction, about $1 million have been lost in Bitcoin, quite likely possibly more, because of this attack.
I really don’t know what to say besides, this is a really important thing to do to have your process checked by an expert. Even in this case, I don’t know, I suspect a lot of people would have missed this problem. I certainly didn’t notice it was a problem. Not great, definitely a security vulnerability. Like I said in the beginning, if you think Libbitcoin’s bx seed command was used in your process at all, you absolutely need to go out, check your wallet, and try to see what you can do, if there’s anything you can do.
Mark Erhardt: I think one thing that at least my cursory look at this revealed was that it had an option where you could set the length of the key material generated, and it was like 128, 192 or 256, however there was no way to use it with more than 32-bit input material. It would just, whatever you give it, cut it down to 32 bit. So, even if you gave it a different seed from some other better entropy source, it would still curtail the entropy to 32 bits, which as we said, is only 4 billion possible tries, regardless of where the entropy originally came from. And so, at least in the Mastering Bitcoin example, it was presented in the context that the operating system provides the entropy, but later when that was changed, that comment in the book was not updated. So, I don’t know, I don’t think that saying this was intentional is a good way of wrapping it.
Mike Schmidt: A couple of notable things from the discovery of this. This class of error or bug is use of cryptographically weak, pseudo random, number generating. It’s also similar to an issue that happened with another wallet, called Trust Wallet, last year, where I believe there was a similar issue with a similar type of attack, and I think maybe even using one of the same underlying libraries which had the same issue. There is a website with details on this research that was done by some folks, called milksad.info. And if you’re wondering why it’s called Milk Sad, if you run this command on the versions that are affected with a system time of 0.0, since the seeds were essentially derived from the system time, you could just put in a fake system time of zero, and it would always generate the same exact secret. And it was a mnemonic with “milk sad wage cup” as the first words. So, they decided to name this website and this uncovered research, Milk Sad.
The website also has a lookup service where you can look up a hashed version of your seed to see if it’s in the list of affected seeds. Obviously, all sorts of caveats if you’re going to be doing that and handling key material as well. Murch or Dave, anything else to add?
Peter Todd: I mean, I think this is one of those things where the problem is, this kind of apparently bug is indistinguishable from an intentional backdoor. And my advice to anyone using Libbitcoin is just stop at this point. The developers are trustworthy and that’s that.
Mike Schmidt: Maybe a similar comment, it looks like also that there’s been a significant drop in the GitHub activity since around the time that this bug was being exploited. So, not sure what that has to do with any of this attack, but maybe be cautious of using this tool at this point, or these tools. Murch?
Mark Erhardt: No, I think that Libbitcoin has been less actively maintained in the last four years already, so I think it was always a fairly small team and I think that maybe featuring it this strongly in a book like Mastering Bitcoin is a little much of an endorsement that misled some people into thinking this is a widespread tool, but I don’t think that Libbitcoin was ever that well reviewed. So, just seconding Peter’s comment here, I don’t think you should be using this tool in your process.
Dave Harding: I can say that as of yesterday, all mentions of Libbitcoin will not occur in the third edition of Mastering Bitcoin; they’ve all been removed as of the commit yesterday. So, that is definitely changed. I had already removed the appendix that was the problem in an earlier change, but like Mark said Peter said, this is not a well-reviewed library. It’s got other problems, and it’s got this really big problem. So, yeah.
Peter Todd: And I want to go say in support of Mastering Bitcoin here, I mean this kind of mistake in a book, it happens. I mean, books are an enormous amount of effort to write. Expecting book authors to in-depth review every single thing that they go reference, well in theory, yes, they should do it. If they don’t, consider yourself warned. Writing a book is a lot of effort and people don’t always catch every single issue there. The responsibility here is on Libbitcoin and if you copy and paste something out of a book, you probably should get it reviewed by someone else.
Dave Harding: I’ve been writing about Bitcoin for about ten years now. I’ve made lots of mistakes, including there’s a CVE out on some documentation I wrote about Pete’s opt-in RBF implementation, and so there’s a CVE out on that, so I’m very well aware that making mistakes to documentation is absolutely something that’s going to happen. that. So I’m very well aware that making mistakes in documentation is absolutely something that’s going to happen, it’s just really sad. The whole thing I think is just very sad. And I don’t know, I just wish it had broken differently.
Denial-of-Service (DoS) protection design philosophy
Mike Schmidt: Moving on to the News section of the newsletter, the first item is DoS protection design philosophy. We covered in the newsletter and in our recap discussion a few weeks ago, in Newsletter #261, a series of talking points that happened at the Lightning-Dev Summit in-person meeting. And one of those topics was heavily focused on channel jamming discussion. AJ responded to the email thread related to that discussion and he says, “The cost that will deter an attacker is unreasonable for an honest user, and the cost that is reasonable for an honest user is too low [to deter] an attacker”, and that was towards the beginning of his discussion about a potential alternate way to price out attackers with regards to channel jamming. Dave, did you want to say something; did you want to walk through this?
Dave Harding: I’m sorry the newsletter was unclear on that, that was actually in the notes. So that was in Carla Kirk Cohen’s notes about the –
Mike Schmidt: That’s right.
Dave Harding: AJ didn’t say that, sorry.
Mike Schmidt: AJ didn’t say that, you’re right. But yes, okay, so the notes did say that, and then AJ suggested an alternative to pricing out attackers. And he sort of resurrects this idea from a few years ago about forward commitment fees and reverse hold fees. Dave, you may be better at speaking to this than I would. Maybe you want to talk about this old idea and the revitalization of it?
Dave Harding: Sure. I mean, the main point I think being made in that thread by AJ, and I actually had a discussion with him about this on the Bitcoin-dev IRC channel, is that it’s just trying to set the costs in the protocol as similar as possible to the cost being borne by the users, the people running the nodes. And that way, as long as they’re earning a reasonable profit, attacks just aren’t economical. If somebody comes in and they attack, then nodes can increase their capacity and get paid for it. It’s a simple idea, hard to implement, of course. And AJ wanted to just kind of work through that idea, see how it would apply in practice.
In doing so, he found an old email that he had written, but not sent, to a thread from a couple of years ago about charging two sets of fees in the network. The first set of fees would be a forward commitment fee. So, this means that the sender of the payment pays the fee forward to the following hops, through each hop on the route that he chooses. That would be a very small fee to cover the cost of just processing an HTLC. So, there’s a little bit of cost that you bear. AJ was suggesting one millisatoshi (msat), so a thousandth of a sat. It’s just a very tiny value just to cover the small cost. And then the main cost in the channel jamming attack, and also in just holding invoices open, which is something that people want to do sometimes, is the cost of capital. It’s having your funds locked up and not doing anything else productive for that time. So, AJ wanted to have reverse hold fees, because the person who receives an HTLC has the ability to jam it. So, they would pay backwards along the route to the previous people.
The challenge with this is that that’s got to be based on time, and there is no way to have a completely universal clock. The blockchain does provide a clock that we all agree on, but it’s just blocks don’t propagate at the same speed to the network. There’s a lot of challenges there. But yeah, tiny forward commitment fees, and then a backwards fee going from receiver towards spender for the amount of time it took to settle a payment on Lightning. That would cover the cost of capital. Those wouldn’t be perfect, but they would recover some of the fundamental costs that Lightning nodes are faced with, from routing payments and having payments delayed. So, that was his basic idea.
Mike Schmidt: I believe we have Clara now. Clara, do you want to introduce yourself, and then we can work you into this discussion since you had some interaction on this particular mailing list post?
Clara Shikhelman: So hi, I’m Clara, I work at Chaincode. Yes, so we talked about this over the mailing list, and I think as with many ideas that have to do with jamming, there’s a lot of basic ideas floating around, but you really need to put in the work and figure out all of the details. And personally, I’m not sure the details in the solution can be worked out. Maybe they can. You know, it would be very, very relevant when we’ll get to talking about upfront fees and things like that, but that’s mostly what the discussion was about, it was about the details. So, I don’t know if anybody’s planning to actually jump into this and work them out, but it seems that currently that’s not on the table.
Mike Schmidt: Murch or Dave, any other notables from this news item before we move into other channel jamming discussion?
Dave Harding: I mean, I think Clara’s absolutely right that this is a really challenging thing to work out. This is a thread that AJ kind of brought back from two years ago. So two years have gone by, nobody has worked them out in that time. I think the summary in this newsletter is AJ is just trying to say he thinks this might be an interesting direction for research. He’s not saying that anybody should be doing it right now, but it is an interesting direction that people might want to look at in the future, even as we work right now to deploy easier solutions. It doesn’t mean they’re easy, they’re just easier, like reputation and stuff.
HTLC endorsement testing and data collection
Mike Schmidt: Well, speaking of reputation, the next item in the newsletter for this week is HTLC endorsement testing and data collection. And so, a Lightning Node has channel liquidity, like we just spoke about, and also HTLC slots, and both of these resources are potentially vulnerable to attack from other members on the network. And channel jamming is the name of the attack that we’re talking about in the previous item, and also this one, that can target those resources. We’ve gone into the attacks a bit in detail for Optech previously, so look for old podcasts or newsletter entries on channel jamming. We also have a topic on channel jamming for more details.
Luckily, we have researchers, like Clara, who have been working on mitigations to these channel jamming attacks, and one potential mitigation involves reputation systems. And one such system is called HTLC endorsement, and that’s been set forth. This week, we’ve actually added HTLC endorsement to Optech’s Topics wiki. Clara, maybe to start, can you explain HTLC endorsement and its relation to some of the previous local reputation discussions we’ve had previously?
Clara Shikhelman: Sure. Just to give a general overview, the complete solution has two parts. There’s the part that has to do with reputation, which is the thing we’ll focus on in this discussion; and then to complete behaviors that are not solved by reputation but can be still considered as jamming, we’ll probably also use some kind of a fee structure. But when we’re talking about reputation, we want to focus on local reputation. That is, if each node looks at its neighbors and decides whether to allow them to use all of the HTLC slots or all of the liquidity; or do they get just a part of the, say, half of the HTLC slots and half of the liquidity? And we do this because new neighbors, or neighbors that didn’t behave good before, and other reasons, I don’t want to allow them the full capacity of my channel, because if I will allow them, they can easily jam me.
So, to build reputation, we expect the neighbors to send payments, to forward payments, the payments should resolve quickly, the payments should pay fees. There’s a whole list of what we consider a good behavior and doing this, they gain a high reputation and by gaining a high reputation, they unlock more privileges to use more slots and more liquidity. So, this is the first part and this is the general behavior that we want. What we want to avoid is, say, somebody sends me an HTLC and I don’t trust them. I don’t want to forward this HTLC and put my reputation at risk with my neighbor. That’s why we have an endorsement. Endorsement is a way to tell my neighbor, “Here’s an HTLC. [Either] I think it’s fantastic, it’s going to go through, please let me use all of the HTLCs that I’m allowed to use, [or] I cannot endorse it”, and say, “Listen, I don’t know what it is. Don’t allow this HTLC to use the complete slots and the complete liquidity”. So this is the main idea behind endorsement.
Mike Schmidt: So based on an individual node’s interaction with a peer, there’s some sort of algorithm that then scores a particular peer, and HTLCs that come from a peer can then be labeled by that node as endorsed or not. It’s a binary thing, right? And then when you forward that along, you forward that binary flag along as an endorsement or not?
Clara Shikhelman: Yes, exactly. And then when I receive something, I will flag it as endorsed if, first of all, I received it as endorsed, and second of all, if I trust the neighbor.
Mike Schmidt: And so this is all local. So, if a peer or a neighbor is behaving well to me, I may endorse HTLCs coming from them. But if that same neighbor is, according to the algorithm, abusing a different peer, then their HTLCs may not be endorsed. It’s all local, as in the local reputation, and dependent on how you’ve interacted with that particular node, not some shared reputation?
Clara Shikhelman: Yes, so the reputation I give a node only depends on their interaction with me. I don’t know what else they’re doing on the network, I don’t know how they behave with their neighbors, and so on. It’s only based on the specific interaction that the node can observe, that is interactions with them.
Mike Schmidt: And so you have to collect this data in order to do analysis on it to determine the endorsement or not, and is that what we’re talking about here with testing and data collection, is just beginning to collect those pieces of information?
Clara Shikhelman: So the first thing we want to do is to have a general sanity check of the algorithm, of the philosophy, of the sole solution, and this is what we’re aiming to do, so we would like nodes to collect data and to run tests, just to make sure we are on the right track. This is the first phase and then we will see how the endorsements also propagate through the network. But the basic idea is to make sure that things work the way we think they work. Our first priority is, of course, in times of peace as it is right now, that nothing gets harmed. That’s the last thing that we want to see happening. Once we see that no harm is done, we also want to see that, if indeed somebody decided to attack, the results would be as we expect them to be.
Mike Schmidt: And so it looks like you have some buy-in from the different Lightning node implementations, and so they are, I guess, pledging to put in code that will start collecting this local reputation data?
Clara Shikhelman: Yes. I think everybody agrees that we should do something about it before things get messy.
Mike Schmidt: Okay, so assuming that these changes are put into the different implementations, who would be the one collecting the data and contributing to some larger body of data to do the analysis; is it just any node operator that turns on a flag to log this data, or how would that work?
Clara Shikhelman: So each node operator can choose to collect the data. Later on, they can also decide to use it, to share it, but this is really an opt-in thing. I really hope, and here’s a call out to everybody, I do hope you will opt in because the more data we have, the better results and better algorithms we can come up with.
Mike Schmidt: And so, I’m collecting this as a node operator. I’ve opted in.
Clara Shikhelman: Thank you.
Mike Schmidt: Oh, no, in theory! And I now have this pile of data. Who do I send it to, and is that data going to be shared, the aggregated data going to be shared publicly for everybody to analyze, or just researchers, or how will that work?
Clara Shikhelman: So we’re working on the details right now to make sure we’re doing this as carefully as possible. So, of course, we’re involving people whose job is to work with data and figure things out. But the main idea currently is that you will run some local analysis and share with us some aggregated data. And only with researchers, we’re definitely not going to make this data public at any point. There’s also, whichever data would be shared with us, there’s some randomization and fuzzing that will be added on top of it. Yeah, but we’re planning to do this very, very carefully for, I assume, obvious reasons.
Mike Schmidt: Murch or Dave, do you have any comments or questions on the effort here?
Dave Harding: I just think it’s great.
Mike Schmidt: All right. Murch gave the thumbs up. So I guess we’ve already gone through the call to action here, which is node operators, once these flags or these opt-in data collection metrics are available, consider collecting that data and providing it for research purposes. Anything else, Clara?
Clara Shikhelman: No, thank you.
Mike Schmidt: Thanks for taking the time to join us. You’re welcome to stay on or you can drop if you have other things you’re working on.
Proposed changes to Bitcoin Core default relay policy
All right, proposed changes to Bitcoin Core default relay policy is the next item from the newsletter, and Peter Todd started two threads and has two PRs related to these potential policy changes in default relay policy. The first one involves full RBF by default. Peter, maybe you can talk a little bit about how things are now and how you would like things to be with this default being changed?
Peter Todd: Yeah, well, so full RBF, to recap, simply means that you apply RBF rules to all transactions, rather than transactions that sets the BIP125 flags. And the number one rationale for this, and really the reason why the mempoolfullrbf flag even exists on Bitcoin Core, is that in multiparty transactions, such as coinjoins, you can cause a lot of trouble if people in your transaction double-spend inputs to it at low free rates. And basically, long story short, the issue that causes is that if that transaction cannot be replaced by the intended transaction, and that low-fee transaction gets to a large percentage of hash power, it can take a very long time to actually make forward progress in your multiparty protocol. Because there’s no easy way for you to determine whether or not a low-fee double-spend exists, particularly for more decentralized coinjoin protocols, where you’re not relying on, say, a central coordinator.
The thing that full RBF does is it just means that in nearly all cases, either the low-fee transaction gets replaced, in which case you make forward progress in the expected amount of time, or the low-fee double-spend isn’t so low fee, it’s actually the higher-fee one, and thus it gets mined in a reasonable amount of time. And while the coinjoin round or other multiparty rounds may fail, at least you’ve made forward progress because you can determine that failure’s happened and you can move forward and use other inputs and so on. So, that’s the clear rationale there.
There’s a lot of other rationales, such as just having BIP125 in general makes life difficult for wallet authors. A lot of wallets lately have been removing the ability to opt out of transaction replacement at all, because it just causes issues. If users send transactions with low fees, their transactions get stuck, well, what do you do? There’s no easy solution to that other than double-spend with a higher fee. So, that’s caused a lot of issues. And of course there’s also privacy concern there, given that BIP125 does create one more way of distinguishing one wallet from another. But the multiparty one I think was the main reason that full RBF got added.
Now, in terms of adoption, I haven’t measured it recently but when I measured it in December of last year, it looks like roughly 20% of newly upgraded nodes had decided to turn on full RBF. And I suspect the number currently is lower than that, but it’s a reasonable amount. And more importantly, a lot of nodes, or at least there’s a core group of nodes that exist that use my full-RBF peering patch, which ensures that full-RBF nodes are connected to each other. And that peering patch means that there is at least some group that propagates full-RBF transactions. And that means that the miners that have turned that on, which appears to be somewhere between maybe 30% to 40% of hashing power, they do mine full-RBF replacements fairly frequently.
This is relatively easy to see by the fact that my OpenTimestamps calendars make full-RBF replacements. And notably, I did change them recently to send the initial transaction with feerates significantly above the minimum mempool limits to ensure that this is an accurate statistic. And Antpool in particular, as well as Binance Pool, they’re mining this stuff all the time. So, I think we have very good evidence that they have in fact turned this on. And I suspect the reason they’ve done this, of course, is because Binance has a habit of doing consolidation transactions with RBF off. And they’ve had issues there where these consolidation transactions get stuck. And people pointed out that it looks like they’ve been double-spending their own consolidation transactions, just to move them along faster. So, there’s an obvious use case for them. And when you have your own mining pool, you can do things like this.
So, I think that’s kind of your background. And as for, “Why would I want to turn this on?” well, obviously you might as well go and get these benefits for everyone, and I don’t think there’s a valid argument against it right now, so I think that sums it up.
Mike Schmidt: In terms of arguments against it, maybe more generally, how would you classify the reception from the community to the PR and the mailing lists posts?
Peter Todd: Well, I think in general, I mean one way I could almost put it is apathy towards this, which is what you’d expect because the reality is, essentially everyone assumes that a transaction is unconfirmed until it confirms. There’s been a small vocal minority of people making a big issue about this, but when you actually ask them to provide examples of merchants that actually depend on the so-called first-seen behavior that Bitcoin Core defaults to, they can’t come up with examples.
We have GAP600, which has made claims that they have a bunch of merchants, but they refuse to provide actual examples of these merchants actually accepting unconfirmed transactions as valid, and this has been going on for months. I keep on pressuring them to provide actual examples and they just refuse to do so. And frankly, my suspicion is that they don’t actually have real examples of this and are pulling some kind of shady business, where they make promises to people that the service works, and then don’t actually deliver.
Mike Schmidt: Murch or Dave, do you have any thoughts on enabling mempoolfullrbf by default in Bitcoin Core?
Mark Erhardt: Yes, many. So, I guess I first have to broadly agree that unconfirmed transactions are unreliable. They are more of a payment promise than a reliable form of settled payment, or anything like that. So, if I could travel back in time, I think I would definitely allow replaceability from the get-go and hopefully we would avoid this whole opt-in business. Or perhaps if we had a flag, at least the default would be replaceability and people could opt out and signal finality. That would have probably reduced a lot of the pinning behavior and other research topics that we’ve been talking about quite a bunch.
But I think in the merging of mempoolfullrbf last year, we also saw and learned that quite a few people regard the signaling of finality as a social signal, as in ,”I don’t intend to make a replacement for this transaction”. And that has tempered my enthusiasm a little bit for pushing mempoolfullrbf very hard. I also don’t really see that many issues in production right now. I don’t know, I think I would probably give it a little more time and wait a little longer to see perhaps some more measurements and if it’s really been adopted sufficiently, then we can still turn it on in a later release, not necessarily this one yet.
Mike Schmidt: Dave?
Peter Todd: Well, correct me if I’m wrong, but we basically are going to have one more release in roughly a month, and then the next one after that follows, you know, what, six, nine months or something like that afterwards? So, I’d say a perfectly reasonable thing to do would be just merge this in like a month and a week, so it would hit the release after the next one.
Mark Erhardt: Yeah, the feature release for the next release is in October. The next release is probably expected in November or early December, and then the one after that would be in May, June next year.
Peter Todd: Sounds good to me.
Dave Harding: I would echo Murch and then Peter, and pretty much anybody reasonable on the idea, that unconfirmed transactions are not final, don’t accept them as final; that’s the full stop there. I’d also kind of like to mention that my enthusiasm for this is tempered. If you had asked me two years ago, I would have been like, “Absolutely, let’s do this, it’s time. Let’s go in”. I am not as convinced as Pete by the argument that this severely affects coinjoin. We’ve had a discussion on the mailing list about this. I think that’s just a difference of opinion there. I should clarify my remark. It absolutely does affect multiparty transactions, such as coinjoins, in a few limited cases. I don’t think the damage is significant, and I think that’s where Peter and I disagree.
I was kind of surprised when we had a discussion last year about adding the flag at all to Bitcoin Core by some of the – there were some interesting remarks by Suhas Daftuar, about if we designed relay policy from scratch right now, we might have an interest in non-replaceability for the sake of keeping the implementation simple. Not that it should be the default policy, it would be an opt-in policy for non-replaceability, just because you don’t have to deal with a lot of pinning vectors and stuff if you do that, because your only option then is to CPFP if you want to fee bump. I think it’s a very interesting idea intellectually. In practice, my guess is that we’re eventually going to enable this flag by default, and so maybe we should just rip the Band-Aid off. I don’t know.
Peter Todd: I mean, so for RBF versus CPFP, I take – actually, so here, I’ll start by first of all saying I don’t think we’re actually in disagreement about how bad this is for things like coinjoin. I would rate this as sort of a moderate issue for coinjoins. I don’t think this is something that ends them because if it was something that completely made the idea non-viable, we probably would have seen more attacks already, including deliberate ones, and we just didn’t. This has happened quite a few times, probably by accident, and for the individual coinjoin, it’s a real nuisance, but it’s not something that people have spent a lot of effort attacking. Although, a counterargument to that might have been because we have had miners running full RBF for almost a year now for sure, and probably even a bit longer than that. So, maybe we just didn’t see those attacks because in fact, a small percentage of miners running full RBF is enough to mitigate them. I could be wrong on that too.
But I think in general, I just think we’re closer in agreement on the severity of it. And my main argument is overall, this is beneficial, and I just think the value of having non-replaceability is very low. And where I think we do disagree is RBF versus CPFP. I think CPFP is definitely the inferior way of doing things, and I think RBF is definitely underused in protocols like Lightning, and other situations where it would have been very easy to pre-sign multiple transactions at different feerates and just broadcast them as necessary. The fact is storage and bandwidth is reasonably cheap compared to onchain space, and CPFP is just a more expensive way of doing things. And certainly going into the future, I think it’s going to matter even more, and we’ll probably actually see protocols like Lightning getting redesigned and tweaked in ways to use RBF more aggressively.
Mike Schmidt: I think in the interest of time, we’ll move on to the second of the two threads that you started on the mailing list and associated PRs, which is removing specific limits around OP_RETURN outputs. So right now, Bitcoin Core will not, by default, relay or take into the mempool any transaction that has more than one OP_RETURN output, or any OP_RETURN output that has more than 80 bytes of data associated with it. Peter, maybe you can talk to a little bit about why it’s a good idea to drop this standardness rule or relax it?
Peter Todd: And I’ll point out, there’s also one other thing that you missed there, which is the current rules do not distinguish between operator and outputs that actually carry data and ones that don’t. And even though this standardness rule is about preventing data from being included in transactions, bare OP_RETURN outputs that don’t carry any data at all are also affected. For instance, you can’t have more than one of them in a transaction. And also, if you seemingly turn off OP_RETURN data to zero, it actually prevents non-data-carrying OP_RETURNs as well, which I don’t think really is intended behavior. But anyway, that’s just sort of a quibble.
But yeah, more generally, I mean, so where this came from is Chris Rowland for various identity-related things and so on, he identified a need for expanding, for making use of OP_RETURN in certain scenarios with more than 80 bytes. The 80 bytes is a limit that was set up when the whole OP_RETURN mechanism was adopted to begin with. And, this point of fact, I’m the guy who is responsible for OP_RETURN existing at all. I mean, I identified a need for this many, many years ago and suggested that we add this as a consensus rule, that scriptPubKeys starting with OP_RETURN, which are provably unspendable, because the first thing that happens in the script execution is the OP_RETURN is seen, and then that immediately fails the script. So, we know that they cannot be spent. And unlike other unspendable outputs, OP_RETURN outputs are never even added to the UTXO set to begin with, which certainly has advantages if you do not intend to spend an output.
But then the standardness rule of only allowing 80 bytes’ worth of data, or to be exact, 83 bytes’ worth of script data, including the push opcodes and so on, that’s something that was added by others when this was actually merged to Bitcoin Core. And like I said, Chris Rowland identified a need for more than that for certain types of transactions. He was willing to put up a $1,000 to put in a few hours of work and get the PR done. And I agree that this is sort of a paternalistic limit. I don’t think it really makes sense to have, when there’s so many other ways of publishing data. In certain cases, doing an OP_RETURN makes sense, in other cases it may make sense in other ways, but it just seems silly to have this one remaining paternalistic limit in standardness rules.
Now, I don’t know if you’re aware, but the PR has been closed due to a flood of I think very poorly technically formed commenting about it, so I kind of agree with that, I don’t think it’s going to get merged. But if you have a reason to need this, the thing I would tell people designing applications is you should have fallback mechanisms that don’t use OP_RETURN. And the more these kinds of applications get used, maybe this issue will get revisited in the future. And if not, well, make sure you have fallback mechanisms.
Mike Schmidt: One comment from a developer was saying something along the lines of removing these restrictions that you’re talking about removing could potentially open up additional transaction pinning vectors. Do you have any more details on what was meant there?
Peter Todd: Yeah, well, for the audience, transaction pinning just refers to mechanisms that make it difficult to replace transactions. And where this hypothetically could come from is maybe if your node doesn’t see the OP_RETURN transaction that is pinning the transaction it’s trying to replace, yeah, that can kind of be a concern, but the space of things that allow that is quite broad. So, I don’t think that’s a concern that’s really specific to OP_RETURN, I think that’s just a concern in general. I think you’re better off designing protocols where this really isn’t an issue.
An example where that might be true is, Lightning channel things are an example where you might want to use RBF to replace transactions to get it to make progress faster, right? Suppose you have multiparty channel opening as an example, and your counterparty is trying to lock up your funds by preventing that transaction from confirming, and maybe they accomplish this with an OP_RETURN, or some other mechanism that you just can’t see, to make it hard for your fee bumps to increase. Well, like I say, any difference in relay policy potentially exposes you to this and you probably just want to, in general, design the protocol to be robust to not seeing why transaction isn’t getting replaced. I kind of rate this is relatively low.
For things where it really matters, like a good example would be the Lightning commitment transactions, where you’re trying to get a state in an adversarial situation, I kind of got asked, “Well, surely we should be using things like OP_CHECKSEQUENCEVERIFY (CSV), etc, to make pinning not an issue to begin with”. I think that’s a much better approach and using that in conjunction with, say, RBF where you’ve pre-signed a bunch of transactions at different feerates to cover all eventualities, I’d say that’s a better approach than hoping that pinning is not an issue and trying to design mechanisms more related to that. So, that’s kind of my view on it, but I know other people disagree. Also, my part, I mean I’ve got to go write up more about my viewpoint on RBF versus CPFP there, so that’s on me too.
Mike Schmidt: Dave, Murch, aren’t we just putting data in the witness these days? What do you think about relaxing the OP_RETURN limits?
Mark Erhardt: Yeah, I’ve been thinking about this a bit this week. So, I oppose this change actually, and my opposition comes from two observations, or maybe three. One is, I don’t think that we should encourage further use of the Bitcoin blockchain as a data publication mechanism. We do have enough room for a hash. If people want to publish something and timestamp it into the blockchain, they can hash it, publish it elsewhere, and per the hash in the blockchain, prove that it existed. I mean, that’s basically what Peter does with OpenTimestamps, right?
Peter Todd: No, it’s not. I mean, you’re talking about different applications there, though. Timestamping and data publishing are very different things. And one does not replace the other.
Mark Erhardt: You’ve talked a lot already. I can talk now, yeah? So, I don’t think that the blockchain is good for publishing data, I don’t think that we need to encourage that further. It makes it cheaper to publish data up to 135 bytes; it makes it widely more accessible to publish data; I think that the arguments that I have seen what it should be used for don’t really seem that huge of a benefit to me personally. You can disagree with that. And therefore, just because other mechanisms have been discovered by which you can publish data, doesn’t mean that we should open up more avenues to do so, (a); and (b) for example, we can skip witness data during the Initial Block Download (IBD) if we’re running a pruned node anyway already, which we can never do for OP_RETURN data.
So, there are ways that witness data does not affect us that OP_RETURN data does. And so, I just don’t really see huge benefits. I do see some small downsides and we’ve had this rule forever, it doesn’t seem to bother a lot of people, so I think it’s fine to keep it. Now you can respond if you want.
Peter Todd: Well, I mean so I just want to make clear with this. It really bothers me how people present timestamping as a solution to these kind of problems. Because, I mean you’re welcome to go say that Bitcoin should not be used for publication, but it really annoys me to see people mislead people by saying, “No, timestamping is good enough”. Now, the answer could very well be that, “Well, we just don’t want your use case to work and ideally we would make your use case not work. Let’s go tell people that something else provides the same benefits”, is I think just wrong and we shouldn’t mislead people that way. We should be honest to them about what Bitcoin is or isn’t capable of doing.
Now, it is also I think interesting how I think you’re the only person, well, I mean maybe not in general, but in the GitHub discussion around that PR, no one actually brought up what you just mentioned about IBD with witness data versus non-witness data. So, I think it kind of says something about the low quality of discussion we had, that that obvious point never actually got mentioned. And I personally don’t agree with that, but you certainly are being smart to go bring that up, and it’s just something that never happened in that PR.
But when you look at it from the point of view of the people using this, I mean like I said before, I think it just makes more sense for them to design things that have alternate mechanisms of publishing. And you can reap most of the benefits of OP_RETURN, which tends to come down to Simplified Payment Verification (SPV) proofs, because going through witness data involves going through coinbase transactions, which have unboundedly large size, which is quite a problem for certain types of publication mechanisms. This is actually why OpenTimestamps does not timestamp data via witnesses, because pre-taproot, there wasn’t a good way to go do that that didn’t involve going through the coinbase transaction, which can be arbitrarily big. Post-taproot, there is one method that can do this, although it’s a bit risky, but pre-taproot that wasn’t really possible.
But anyway, I mean I think the bigger picture is, this PR was closed because it’s not likely to get merged. And I don’t think it’s good for people designing these types of things to assume standardness rules will be set in any way beneficial to them.
Mike Schmidt: Dave, any commentary on this? Okay. Well, if any of the listeners are interested in this type of discussion around policy, like these two points that Peter has brought up today, there’s some philosophical and other discussion about relay policy and mempool in our Waiting for confirmation series, which is aggregated as a blog post on the bitcoinops.org website, so dig into that if some of this discussion is perking your interest. Peter, thanks for joining us. You’re welcome to get back to your beautiful scenery. I’m glad your connection held up and you sounded pretty good. Thanks for joining us.
Peter Todd: Thank you, and credit goes to silent.link for the digital whatever SIM card, or whatever that actually made all this work. So thanks to them and thank you for having me on.
Silent Payments: Implement BIP352
Mike Schmidt: Absolutely. Next section from the newsletter involves a select Bitcoin Core PR Review Club PR. And this month we’ve selected Silent Payments: Implement BIP 352, which is a PR by Josie, who has joined us. Josie, you missed the introduction, so maybe you can do a quick introduction for folks, and then we can jump into some of the details here.
Josie Baker: Yeah, sure. Thanks for having me too. So for introductions, I’m Josie. I work on Bitcoin Core pretty much full-time as a contributor. I’ve done stuff around the wallet and some other things, and for the past eight months or so, I’ve been working on silent payments with Ruben Somsen. Silent payments was a proposal he came up with, I think around March 2022. And then I heard him give a presentation on it, got really excited about it, and decided to help write the BIP. So, we worked on that for several months. And then I started working on an implementation for Core, which was based off a PR that a pseudonymous developer, w0xlt had opened. And so I’d been working on that for a while to kind of keep it up to date with the BIP.
So, the PR Review Club we just did is kind of the ready-for-primetime version of that PR. So, we’ve broken it up into the first part, which is what we reviewed on the PR Review Club, which just implements the protocol from the BIP; and then I’ve also opened a sending and a receiving PR that are both based on that PR. Sorry, that was longer than just an introduction.
Mike Schmidt: No, that’s good, we’re jumping into it. Okay, so there’s sort of this parent or tracking PR for silent payments progress. And I guess I should note that that is specific to Bitcoin Core. And I guess there would ideally also be work for other wallets who wanted to implement silent payments as well. It looks like you’re focusing first right now on the Bitcoin Core implementation and then potentially in the future, other wallets?
Josie Baker: Yeah, maybe it’s worth saying a few things about that. So we’ve been working on two things concurrently, which is, one, the BIP. And maybe for those that are less familiar, this is an application layer BIP, meaning it doesn’t have anything to do with consensus, it’s just a technical spec for how wallets could implement this protocol, which is silent payments. I think we spoke about the BIP itself several months back when we first opened that PR to the BIPs repo, here on the Optech Recap. And since then, the BIP has gone through several rounds of very thorough review. And just, I’ll take a second here to give a huge shoutout to all the people who have contributed to that review, it’s been super-helpful.
There’s not been really many changes on the protocol itself. We’ve made a few refinements since then, like changing the format of the address and a few other tweaks. But I would say I’m feeling pretty confident that the BIP is settled. So, if you wanted to go out and start implementing silent payments today in your wallet, you could do that reasonably safe. And we’ve split it up so that sending is much easier to implement than receiving. So, my hope is that as the BIP settles and we refine the wording and add more test vectors, people can start implementing sending and receiving support in their wallets. And then, because I’m more familiar with Core, that’s where I’m going to focus, getting it into the Bitcoin Core wallet.
What that’ll look like is, if and when those PRs get merged, you’ll be able to get a silent payment address for your Bitcoin Core wallet, which means that you’ll be able to send to other silent payment addresses and you’ll also be able to receive silent payments to the Bitcoin Core wallet. But yeah, very much this is a wallet protocol, so it becomes more useful the more wallets that implement it. And we tried to learn from what happened with taproot, where it’s really important for adoption that people can send to and parse silent payment addresses. You might not care about receiving silent payments, but if I want to receive them and I post my silent payment address but nobody’s wallet can send to it, it’s not going to be that useful. So, we put a lot of time into making sure that sending is really, really simple. And so my hope is that even while we’re still working on this PR for Core, which implements the full thing, that other wallets can pick up and start implementing sending support.
Mike Schmidt: You mentioned our discussion previously. In the Recap #255 Newsletter, we had Ruben Somsen and Josie on discussing the draft BIP for silent payments. So, if you’re curious about a higher-level discussion, maybe jump back and listen to that segment. I know we’re a little bit off of this Core PR that we’ve selected for the newsletter, but I do have a quick question since we’re on the topic of other wallets and libraries. Your focus on Bitcoin Core, getting some Bitcoin Core functionality and also in the wallet, is there a plan in the future to have something like we covered recently, a payjoin development kit that is separate from the BDK for folks to integrate payjoin into their wallet; is that something that would be planned in the future, or would that be a different team taking initiative there?
Josie Baker: Yeah, I mean, absolutely. I think the right way to encourage wallet adoption is make it as easy as possible for people. I mean, we’re all busy, some of us are just doing this for the love of the ecosystem, so it’s very difficult to – I would never want to be in the position where I’m going to some open-source project and being like, “Hey, we came up this new thing. Can you take time off from what you’re busy with to implement my thing?” And so, after the Bitcoin Core PR, I’d really like to focus on what are the popular wallets, Software Development Kits (SDKs), that people are using, and how can we make it easy to integrate silent payments there?
Once we have it in Bitcoin core, I think that’s the right time to start marketing and advocating for it and talking to wallet developers. And like I said earlier, so far I’ve been just really humbled and blown away by how people have approached us and shown interest in this, because of course I think silent payments are cool, and Ruben thinks it’s cool, and you spend a lot of time working on something and then you put it out there and you’re like, “I hope other people think this is cool”. And so far, it seems like there’s been some enthusiasm about it. So I’m pretty hopeful about wallet uptake and adoption.
Mike Schmidt: Back to the PR in question for the PR Review Club, which is #28122, that implements the logic for BIP352, which is the silent payments BIP, without any wallet code. So Josie, what is the non-wallet logic code that needs to be in Core before you can put in the other two PRs that involve actual sending and receiving?
Josie Baker: Yeah, before I jump into that question, maybe it’s worth mentioning, this PR Review Club was the first in our new monthly format. So PR Review Club used to be once a week, now it’s going to be once a month, where instead of one meeting, we do two meetings. And the cool thing about that is it gives us more time to prepare the notes, it gives the people who want to attend the club more time to read and prepare. And then doing two days instead of just one, so two hours in total, it really allowed us to dig into the PR. So that’s kind of a shameless plug for PR Review Club.
To the specific question, what parts of silent payments don’t involve wallet code? So, I’d say that really, silent payments is, at the base of it, a protocol for how a sender and receiver can deterministically create a shared secret. And so, the sender creates a shared secret and then the receiver uses information already in the blockchain, which is the pubkeys present in the inputs to a transaction, to arrive at the same shared secret, and that’s how they can find this address that is unique to them.
So, that first PR for Core isn’t really talking about how we do it in the wallet, how we parse the address, and it’s more just, how do we make sure the right primitives are in Core so that we can implement this protocol? So, for example, with silent payments, we use the private keys of the UTXOs that you want to spend to do the Diffie-Hellman step. So, we needed to add some functions in Bitcoin Core so that we could add private keys together. For the receiver, they look at a transaction and they look at the public keys and then they want to add those public keys together to do the Diffie-Hellman step. So, we needed to add some functions to Bitcoin Core for adding public keys.
Then, we added some functions that do the silent payments protocol. And then, those functions will then be used in later PRs in the actual wallet code. And my reasoning for wanting to do it this way is, to do it all at once is going to be a massive and complex PR, so we want to break it up into small chunks that people can reason about. And doing it this way also made it easier to test the Bitcoin Core implementation against the test vectors on the BIP. Because now, I don’t really need to worry about modeling things as UTXOs or transaction inputs and outputs, I just have this interface that accepts public and private keys and then spits out public and private keys. And then we can very easily test that Core logic and then have more confidence in the later PRs, when we build the wallet logic on top of it, that we’re implementing the protocol correctly.
Mike Schmidt: Okay, so this PR is for the non-wallet logic code, and this is referencing one of the Review Club questions. So, why is this PR adding a bunch of code to the wallet directory then?
Josie Baker: Yeah, great question. I think at the end of the day, like I mentioned earlier, this is an application layer BIP, and it’s a protocol for wallets to communicate with each other. So, when we’re talking about code separation and organization of Bitcoin Core, we like to keep these things as modular as possible, so it’s easy to reason about what’s going on in the code base, the safety, etc. And so, since silent payments is just going to be strictly used for wallets, like you generate an address, you want to send to that address, it’s all happening in wallet software, we wanted to organize that code under the wallet directory. We had some good discussion about that too, of like, “Well, is that really the best scenario, or is that really the best place for the code? And can you think of an example where we might want to use silent payments logic outside of the wallet context?”
One example that was brought up in the PR Review Club is, “Well, yeah, I could have a Bitcoin Core node that is scanning for silent payments, but not actually using them in the wallet. And they’re scanning for silent payments in order to hand them off to a light client. So, the light client doesn’t have access to the blockchain, but the light client wants to receive silent payments. They can have a Bitcoin Core node doing the scanning for them, collecting that tweak data that they need, and then parsing it off to the client to do the actual protocol and drive the outputs”. And so in that case, yeah, maybe you do want some silent payments code to live outside the wallet so that I could compile Bitcoin Core, don’t compile it with the wallet, but just have it running and scanning for silent payments.
Mike Schmidt: There were some questions in the PR review club about versioning and compatibility. I invite everybody to look at all the questions, including the ones we highlighted in the newsletter, and not just the ones that I’m choosing to ask here. But maybe for folks to wrap their head around versioning, maybe you could give an example of what would a potential example improvement, version improvement to silent payments be, so we can wrap our minds around what versioning of silent payments even means?
Josie Baker: Yeah, I think versioning is just this really interesting question of, like, versioning is a beast, right? And I don’t know if you caught this part, but it basically boils down to how can we futureproof and predict the future? So, one example that comes up, in silent payments, we pick a subset of scriptPubKey types that will be used for this shared secret derivation step. But let’s imagine in the future we soft fork in some new output type, entroot. Now there’s taproot and bech32, now there’s this new output type, entroot, and we want to also include that in this shared secret derivation step.
What we would do is we would release a silent payment v1, because we’re on v0 now, so we have a silent payments versions v1. And v1 would say, “If you’re sending to a v1 address and you want to spend these entroot outputs, it’s fine to include them”. And the receiver will recognize that this is a V1 transaction and they’ll include those entroot outputs in the shared secret derivation. So, that’s one example that we thought of that I think helps us be pretty resilient to soft forks in the future, without really needing to change stuff.
Another example would be, maybe we want to add more data to the silent payment address, some metadata. Kind of a contrived example I thought of was with silent payments on the BIP, when we’re talking about like clients. In theory, if we assumed this perfect, private P2P messaging protocol, you can imagine that the receiver, instead of scanning for silent payments, could just get a notification from you, the sender. You, the sender, crack the silent payment, and then you encrypt it, a message with the txid, or something, and you encrypt it to their silent payment address using the pubkeys, and then you send them this notification, they decrypt it, and they’re like, “Oh, yeah, someone sent me a silent payment”, and then they get that without having to scan the chain. And this is okay because they can always fall back to scanning the chain if the notifications are missed or censored or unreliable.
So, one idea I had was, okay, well yeah, this perfect P2P messaging protocol doesn’t exist today, but if it does, then maybe the receiver wants to put a hint in their silent payment address that says, “By the way, I accept notifications, and I want them over this protocol”, like let’s say Simplex or something, and so they put some little byte or two in the address. And so that would be an example of a future version, where if I’m a v0 silent payment sender, I’d read their address, and I don’t understand anything about notifications because I’m still using the v0 portion of the protocol, so I just ignore those extra bytes, send them a silent payment, and they would find it by scanning the chain. But if I’m a v1 and I do support it, notifications, I read that address, I’m like, “Oh, this receiver accepts notifications”, so I’m going to go ahead and send them a notification as well.
So, that gives an example of one where the shared secret derivation part might change, and then there’s an example where just the protocol might change.
Mike Schmidt: Thanks for walking through that, Josie. I think I heard one potential call to action for the audience here, which is folks who are technically capable and potentially working on wallet or library software and are interested in pursuing silent payments in their wallet or service, they can reach out to you and coordinate on any questions that they may have in their implementation. Is there anything else that you’d call for the audience to do, other than review these PRs that are currently open?
Josie Baker: Yeah, I would say if you’re someone who works on and reviews Bitcoin Core, we’d love to have your review on the PRs that are open. So, that PR that we went over in the Review Club, I think that’s the one that I would love to have the most attention on right now, because I think it’s the most ready. And the other two that I’ve written for sending and receiving will also depend on it. And then like you said, if you’re in the broader ecosystem and you work with wallets or you’re contributing to and involved with wallet projects, and this is a protocol that you’re interested and want to implement, please reach out. I’d love to help in any way I can. And also, having these people, like these bleeding-edge first movers, who come and want to start implementing this, has really helped us a lot because they’ve had great commentary.
If you go to the BIP, you can see a ton of comments from a few people who’ve started to implement it. And they’re like, “Hey, by the way, think about this test; and I found this edge case and maybe you want to think about that”. And I mean, code is what matters. So, people implementing this is really what gives us confidence that we have a good, solid and robust protocol. So yeah, feel free to reach out. I’d love to see people just tackling sending, you know, like pretty much every mobile wallet today could implement sending fairly trivially, and that would do a lot to make the case for adoption. So, if that’s something that you’re interested in, I’d love to chat.
Mike Schmidt: Thanks for joining us, Josie. You’re welcome to drop if you have other things you do, and you can hang out with us.
Josie Baker: Yeah, thanks for having me. Always fun to join these and have a chat.
Mike Schmidt: Next segment from the newsletter is Releases and release candidates, of which we have one, BDK 0.28.1. It includes a bug fix and also adds a taproot descriptor template, BIP86. And so with that feature change, users can now, if you’re a BDK user and you upgrade to 0.28.1, you can now create taproot descriptors with templates.
Bitcoin Core #27746
Bitcoin Core #27746, simplifying the relationship between block storage and the chainstate objects. Murch, I think you’re better equipped for explaining this one than I am. Do you want to take a crack at it?
Mark Erhardt: I’ll take a crack at it. So, assumeUTXO of course is a project where we try to be able to serve a snapshot of the UTXO set, and then in the background have a full node, on the one hand start synchronizing the rest of the chain tip from the UTXO snapshot, and in the background also to verify that the UTXO snapshot was accurate by syncing the whole blockchain in the background. And from my understanding, this PR lays the groundwork for being able to have two UTXO states in the Bitcoin Core, where one of course would be the one starting from the snapshot and heading towards the chain tip, and the other one is starting from the Genesis block and heading to the UTXO snapshot. And yeah, that’s pretty much what I know about it.
Mike Schmidt: So, some foundational assumeUTXO work. Dave?
Dave Harding: Actually, I just want to add some clarification here. I think we actually have, for about a year-and-a-half now, actually had the ability to have two separate chainstates in Bitcoin Core for the assumeUTXO project. What this PR does, as I understand it, is it changes the logic related to how we store blocks. So currently, the decision made to store blocks by Bitcoin Core on disk, which is a DoS factor, like if somebody can just send you random junk data that looks like blocks and you store it to your disk, you can eventually run out of space; so, what this PR does is removes some of the chainstate logic that was in that decision loop to store blocks.
So previously, the decision made to store blocks, you had to check the chainstate to see, for example, if that block connected to a previous block. And now a lot of that logic has been removed, so you don’t actually need as much chainstate in order to decide whether to store a block. And that’s just really good for separation of concerns in the code, so that you’re not pulling that in. It’s really good if you have these two separate chainstates, that you can do this logic separately. That’s all it does.
Core Lightning #6376 and #6475
Mike Schmidt: Thanks for clarifying, Dave. Next two PRs are to the Core Lightning repository, #6376 and #6475, which implement a plugin called renepay. And luckily, we have an author that contributed to these PRs here, Eduardo. Eduardo, do you want to introduce yourself and then you can explain renepay?
Eduardo Quintana: Okay. Hello everyone, can you hear me?
Mike Schmidt: Yeah.
Eduardo Quintana: All right. So, I started working on renepay from a theoretical point of view in 2022. It was an idea from a paper by René Pickhardt of 2021. The idea is to, let’s say, propose a new payment algorithm for the LN. Just to give you an idea of what’s the current state, right now, when you want to try a payment on the network, you will search for a path with a minimum cost or fees in the network in order to forward the payment from the source to the destination. But usually, these kind of payments will fail because of this lack of liquidity in the channels along this path.
So, with renepay we have a model of the network in which we define the state as some kind of a Bayesian model. So, the state of the network is not a certain value of the liquidity in each channel but a probability solution on the liquidity. And as a first approximation, we use the probability distribution which is uniform. So for example, if there is a channel which does not belong to us, it will belong to some other nodes, we don’t know where the liquidity is but it has a capacity of 3 million sats, then we say that initially, this probability distribution would be a uniform distribution from zero to 3 million sats. And once we send the payment that includes that channel, the payment goes through, then we know that the channel has had, when we send it, at least the capacity, the liquidity, for us to forward the payment on. We update the knowledge of the state of that channel. So, that’s why it’s called Bayesian because we update this knowledge every time we interact with the network.
So at the end, we make use of these multipart payments. And if you think of it, if you make many, you divide your payments into many parts, you are decreasing the probability of your payment going because you are exposing yourself to more risk of finding no channels with less liquidity than the other. But at the same time, you are diminishing the amount of liquidity that you are asking for each of these paths. So, the algorithm tries to find a sweet spot between making many, many paths and with a low liquidity, so in order to increase the reliability of each payment trial. But as a matter of fact, this was just a simplistic way of putting it. At the end, there is a more robust way to solve this problem, which is the use of minimum cost flows algorithms, in which we consider a cost function that is related to the probability of a forwarded payment, given the amount that we want to forward, and the cost in fees that this payment is going to cost for the person who’s sending the payment.
So right now, this plugin is already merged into CLN. It surely has a lot of bugs because it’s just in its infancy. This was the result of three months of work, my first contribution to CLN as a matter of fact, and there are many things to try. There is the possibility that this plugin is very inefficient because we haven’t highlighted many different min cost flow solver, but at least this is the first trial and it’s a candidate for the future payment algorithm for the LN. Something more to add, I guess, what promises is that with the usual payment algorithm, you have sometimes many, many failures to try to deliver the payment and after a certain time you will just stop trying. But with this renepay, you can put a limit on how many trials you want to have or how much time you want to wait for it, but theoretically you can deliver because this algorithm, once you start running it, it by itself starts discovering more where the liquidity is allocated along the network and from the source to the destination. So, here you can forward as much amount of sats as the actual max flow you can actually deliver. And well, that’s pretty much the summary of what renepay is. If you have any questions, I’m glad to answer.
Mike Schmidt: Well, it sounds like there’s some work to be done, potential bug fixes, maybe performance improvements. So, I’m sure you have your hands full there. Are you aware of any of the other Lightning implementations working on something like renepay?
Eduardo Quintana: Okay, so I think it was six to seven months ago, I know of Carsten Otto was working on his own, it’s like a sort of a plugin for LND, it was called manageJ, I don’t remember exactly, but there is a discussion on the Linux forum. That’s the only implementation that I know that precedes the one in CLN, so we are right at the beginning.
Mike Schmidt: Dave or Murch, do you have any questions or comments about renepay?
Dave Harding: I just wanted to answer your previous question that LDK also merged an implementation very close to René and Richter’s original description. I did that a few months ago; we covered that in the newsletter, I believe.
Mike Schmidt: Okay, great.
Mark Erhardt: I kind of wanted to repeat something that I’m not sure was super-clear. So, the way that this payment planning algorithm works is that it tries, as Eduardo said, to find a sweet spot between the count of parts on the multipath and the amount per path. And then whenever an attempt fails, it will learn and update its model of the network for where the liquidity is available and not available in the Lightning chart. So with each attempt, for example, if it cannot route X amount in a channel, it knows that X is too much, so it can change the assumed channel liquidity for that channel to zero to X, the statistical – I think it’s the average or you can later correct me! Anyway, the sweet spot between zero and X would then be picked as for the model, whereas previously it would have first assumed that about half of the capacity is on one side.
So, with each attempt it learns, it’s a Bayesian learning algorithm where it updates its model with each attempt, and that’s why it very quickly finds an optimal route for the updated model.
Eduardo Quintana: Okay, so just let me clarify that. Once we try payment along the path and there is a channel, but one of the channels that fails because it cannot forward these X amounts, then we know for sure that every channel before that channel was able to forward it, so we state this in our knowledge. And that channel that failed, we also state in our knowledge the fact that this liquidity is below X. So, if the liquidity distribution is a uniform function, let’s say at the beginning it will be zero to the capacity, so it could be anywhere between zero and the capacity, and there is a probability distribution function to describe this. And once we know that the liquidity is less than X, X-minus-1, then we update the knowledge by changing this probability distribution from zero to X-minus-1.
I know it sounds very complicated, but it is just a uniform probability distribution, so it can be described by two numbers, A and B, which are the bounds, the lower and the upper bounds of where the liquidity is. It’s just two parameters for each channel. At the beginning, A is zero and B is the capacity. And once we start learning things about the channels, we update these two numbers. There is one detail, which is the fact that as time goes by, channels that do not belong to us, they have movements in their liquidity because the network is dynamic and we don’t know about other payments that are going out of there. So, I propose a way to forget the knowledge of where the liquidity is with time. And by now it’s linear, but it can be something else. It’s just a proposal, but it’s also something to think about.
Mike Schmidt: Well, Eduardo, thank you for joining us.
Eduardo Quintana: Okay, thanks for having me. Yeah, thanks for walking us through that and thanks for your work on renepay.
Core Lightning #6466 and #6473
Next PR from the newsletter is Core Lightning #6466 and #6473. And this adds support for backing up and restoring a wallet in CLN using codex32. So, there’s actually an RPC called getcodexsecret that will print out the formatted Hardware Security Model (HSM) secret in BIP93 format, and BIP93 is the codex32 BIP, which is the standard for backing up and restoring a master seed from an HD wallet using Shamir’s secret sharing. And these changes also allow for restoring the LN node and CLN from a similar 32-byte secret encoded at that same codex32 secret. And both of these are within the Lightning-hsmtool, which is the Core Lightning’s way of interacting with HSM secrets in Lightningd or CLN. Murch or Dave, anything to add there?
Dave Harding: Well, I mean, about this specific PR, I just have nothing really to add. I do want to say that I think codex32 is really cool and I’m really excited to see wallets starting to add support for it. It’s going to be a long time until we have a lot of wallets supporting it, but I’m glad to see CLN taking the first step. I also got a message from René Pickhardt correcting my earlier comment about LDK implementing his system, so I just wanted to correct that. They implemented part of it. They implemented a probabilistic model of liquidity, but not the actual pathfinding algorithm or payment planning algorithm, as I believe René likes to call it, so I wanted to correct that, sorry.
Core Lightning #6253 and #5675
Mike Schmidt: Thanks for clarifying, René and Dave. We have another pair of PRs to CLN, Core Lightning #6253 and #5675, adding experimental support for BOLTs #863, which is the draft spec for splicing. We had Dusty Daemon on last week when we went through a bunch of other CLN PRs, and he spoke a bit about splicing and his work on it in our #262 recap podcast. So if you’re curious about some of the details there, check out that podcast. We’ve also talked about splicing a few different times in the newsletter and on our recap here. So, Murch, I’m not sure how much you want to get into it, other than a round of applause for rolling out more splicing-related work. Thumbs up from Murch.
Rust Bitcoin #1945
Next PR is to the Rust Bitcoin repository, #1945. This one’s a little bit different. It modifies the project’s policies for how much review that a PR requires before it’s merged, if that PR is just a refactoring PR. And they changed their contributing markdown file details and added a section that they call refactor carve output, in the contributing file. And the relevant text there is, “This repository is going through a heavy refactoring and ‘trivial’ API redesign as we push towards API stabilization. As such, reviewers are either bored or overloaded with notifications. Hence, we have created a carve out to the 2-ACK rule”.
So, I guess traditionally they’ve had a two-act rule for the Rust Bitcoin project, and they have a carve out here that states, “A PR may be considered for merge if it has a single ACK and has sat open for at least two weeks with no comments, questions, or NACKs”. So, I think that was the relevant segment from the change in their policy. And we noted in the newsletter that other projects with challenges getting refactors may want to investigate Rust Bitcoin’s new policy. Murch or Dave, any thoughts?
Dave Harding: I wanted to add that when I was a maintainer of a project, I really liked merge deadlines, pretty similar to what they’re doing here, which was that once something got a little bit of review, or even if the PR was opened by a maintainer who deemed the PR to be minor, they would say, “This is going to get merged on such-and-such date, unless somebody complains”. And again, you still need to have somebody who knows the project and can say that this is safe. But as long as that’s in place, I think a lot of this minor stuff can be dealt with in a more simple way. And the time-based deadline just makes it clear to everybody what’s going to happen and when it’s going to happen so there’s no surprises. So, that’s what you really want to be as a maintainer, is not a surprise the people who are contributing to your project. So, I thought this was just a good idea and I was glad to see this project doing that.
Mike Schmidt: Last PR this week is to the BOLTs Repository, #759, adding support for onion messages to the LN spec. And onion messages allow sending one-way messages across the LN using encryption and blinded paths. Onion messages are also used by the BOLT12 offers protocol. Dave, maybe a question for you. It seems like we’ve been talking a lot about onion messages for the last year or plus in the newsletter, and also on these recaps, but it wasn’t even merged into the spec. Do you have insight into why that took so long?
Dave Harding: I think they were waiting for – there was some holdout there, I think, from LND. So, Olaoluwa Osuntokun, roasbeef, or Laolu, he was concerned, as we’ve covered in previous newsletters, that onion messages are subject to abuse. They’re free to send, they don’t involve any money, any lockup of capital. You can technically send them even if you don’t have a channel with a node, although most implementations only allow you to send that to your direct peer. So, it’s kind of just like an extra onion routing layer over the internet. It’s basically what Tor provides, and Tor has its own problems.
So, he was concerned about message spam. Now, in previous newsletters that we’ve addressed, I think I might link it here in the, yeah, we have a link in this newsletter to a previous newsletter, #207, about some discussion about trying to solve that spam issue, that DoS abuse, where people can just send messages to your node for free, using up your bandwidth and your peers’ bandwidth. But the guys said there was some concern there, and so they wanted to make sure it was safe. Everybody wanted to discuss it to the point where everybody was relatively okay. And I think that they finally got to that point. But before that point, it needed to be implemented, and people have been building stuff on top of it, most notably BOLT12, the offers protocol. So, I’m excited to see this merged, and I’m really excited to see offers.
Mike Schmidt: For those curious about some of the tech that we just talked about, we have Optech topic entries for onion messages, blinded paths, and offers on the Optech Topics page for more detail. All right, fairly beefy recap this week. Thanks to all our guests for joining. Thanks for Dave, Eduardo, Murch, my co-host as always, Clara, Josie, Peter Todd. We had a great set of contributors this week. Thank you all for your time.
Mark Erhardt: Thanks, hear you soon.
Mike Schmidt: Cheers, bye.
Eduardo Quintana: Thank you.