This special edition of the Optech Newsletter summarizes notable developments in Bitcoin during all of 2019. It’s the sequel to our 2018 summary.
This summary is based heavily on our weekly newsletters from the past year for which we reviewed almost 9,000 commits (nearly 2,000 merges), over 1,500 mailing list posts, many thousands of lines of IRC logs, and numerous other public sources. It took us 50 newsletter issues and over 200 printed pages worth of content to summarize all that amazing work originally. Even then, we missed many important contributions, especially from people fixing bugs, writing tests, performing reviews, and providing support—work that’s critical but not necessarily “newsworthy.” In summarizing even further and trying to compress the entire year into this article’s handful of pages, we’ve now also omitted a great many other important contributions.
So, before we continue, we want to extend our heartfelt thanks to everyone who contributed to Bitcoin in 2019. Even if the following summary doesn’t mention you or one of your projects, please know that we at Optech—and probably all Bitcoin users—are more grateful than words can express for all that you’ve done to help Bitcoin.
- Featured summaries
In January, Steven Roose proposed a standardized format for proof of reserves pseudo-transactions that bitcoin custodians can use to generate evidence that they control a certain number of bitcoins. No tool of this type can guarantee that depositors will be able to withdraw their coins from a custodian, but it can make it more difficult for a custodian to conceal the loss or theft of coins. Roose would go on to produce a tool based on Partially Signed Bitcoin Transactions (PSBTs) for creating reserve proofs and would follow through to see the specification published as BIP127.
In February, Bitcoin Core’s master development branch saw the merge of the final set of PRs necessary for using it with the Hardware Wallet Interface (HWI) Python library and command-line tool. HWI would later see its first stable release in March, see Wasabi Wallet add support for it in April, and see BTCPay add support for it via a side package in November. HWI makes it easy for hardware wallets and software wallets to interact using a combination of output script descriptors and Partially Signed Bitcoin Transactions (PSBTs). The increasing support in 2019 for standardized formats and APIs makes it easier for users to choose the right combination of hardware and software solutions for their needs rather than having to choose one solution or another.
Also in February, Pieter Wuille gave a presentation during the Stanford Blockchain Conference on miniscript, a spin-off from his work on output script descriptors. Miniscript provides a structured representation of Bitcoin scripts that simplifies automated analysis by software. The analysis can determine what data a wallet needs to supply in order to satisfy the script (e.g. a signature or a hash preimage), how much transaction data will be used by the script and the data that satisfies it, and whether or not the script passes known consensus rules and popular transaction relay policies.
In addition to miniscript, Wuille, Andrew Poelstra, and Sanket Kanjalkar also provided a composable policy language that compiles down to miniscript (which itself converts to Bitcoin Script). With the policy language, users can easily describe the conditions they want to be fulfilled in order to spend their coins. When multiple users want to share control of a coin, the composability of the policy language makes it easy to combine each user’s own signing policies into a single script.
If widely adopted, miniscript could make it easier for different Bitcoin systems to work together to sign a transaction, significantly reducing the amount of custom code that needs to be written in order to integrate wallet front-ends, LN nodes, coinjoin systems, multisig wallets, consumer hardware wallets, industrial Hardware Signing Modules (HSMs), and other software and hardware.
Wuille and his collaborators continued working on miniscript through the year, subsequently requesting community feedback and opening a PR to add support to Bitcoin Core. Miniscript would also be used by LN developers in December to analyze and optimize several new scripts for upgraded versions of some of their onchain transactions.
In March, Matt Corallo proposed the consensus cleanup soft fork to eliminate potential problems in Bitcoin’s consensus code. If adopted, the fixes would eliminate the time warp attack, lower legacy Script’s worst case CPU usage, make caching transaction validation status more reliable, and eliminate a known (but expensive) attack against lightweight clients.
Although parts of the proposal (such as the time-warp fix) seemed to interest a variety of people, other parts of the proposal (such as fixes for the worst case CPU usage and validity caching) received some criticism. Perhaps it was for that reason that the proposal didn’t make any obvious progress towards implementation in the second half of the year.
March also saw Kalle Alm request initial feedback on signet, which would eventually become BIP325. The signet protocol allows creating testnets where all valid new blocks must be signed by a centralized party. Although this centralization would be antithetical to Bitcoin, it’s ideal for a testnet where testers sometimes want to create a disruptive scenario (such as a chain reorganization) and other times just want a stable platform to use for testing software interoperation. On Bitcoin’s existing testnet, reorgs and other disruptions can occur frequently and for prolonged lengths of time, making regular testing impractical.
Signet would mature throughout the year and eventually be integrated into software such as C-Lightning as well as used for a demonstration of eltoo. A pull request adding support to Bitcoin Core remains open.
Additionally in March, Lightning Labs announced Lightning Loop, providing a non-custodial solution for users who want to withdraw some of their funds from a LN channel to an onchain UTXO without closing the channel. In June, they would upgrade Loop to also allow users to spend a UTXO into an existing channel. Loop uses Hash Time Locked Contracts (HTLCs) similar to those used by regular offchain LN transactions, ensuring that a user’s funds are either transferred as expected or that the user receives a refund of all costs except for any onchain transaction fees. This makes Loop almost completely trustless.
Major releases of popular infrastructure projects
C-Lightning 0.7 released in March added a plugin system that would see heavy use by the end of the year. It was also the first C-Lightning release supporting reproducible builds for increased safety through improved auditability.
LND 0.6-beta released in April included support for Static Channel Backups (SCBs) that help users recover any funds settled in their LN channels even if they’ve lost their recent channel state. The release also featured an improved autopilot to help users open new channels, plus built-in compatibility with Lightning Loop for moving funds onchain without closing a channel or using a custodian.
Bitcoin Core 0.18 released in May improved Partially Signed Bitcoin Transaction (PSBT) support and added support for output script descriptors. The combination of those two features allowed it to be used with the first released version of the Hardware Wallet Interface (HWI).
Eclair 0.3 released in May improved backup safety, added support for plugins, and made it possible to run as a Tor hidden service.
LND 0.8-beta released in October added support for a more extensible onion format, improved backup safety, and improved the watchtower support.
Bitcoin Core 0.19 released in November implemented the new CPFP carve-out mempool policy, added initial support for BIP158-style compact block filters (currently RPC only), improved security by disabling protocols such as BIP37 bloom filters and BIP70 payment requests by default. It also switches GUI users to bech32 addresses by default.
C-Lightning 0.8 released in December added support for multipath payments and switched its default network to mainnet from testnet. It was also the first major C-Lightning release to support alternative databases, with postgresql support available in addition to the default sqlite support.
In April, James O’Beirne proposed AssumeUTXO, a method for allowing full nodes to defer verification of old block chain history by downloading and temporarily using a trusted copy of the recent UTXO set. This would allow wallets and other software using the full node to start receiving and sending transactions within minutes of the node being started instead of having to wait hours or days, as is the case now for a newly started node. AssumeUTXO proposes that the node download and verify the old block chain history in the background until it eventually verified its initial UTXO state, allowing it to ultimately obtain the same trustless security as a node that doesn’t use AssumeUTXO. O’Beirne would continue working on the project throughout the year, incrementally adding new features and refactoring existing code on the path towards a goal of ultimately adding AssumeUTXO to Bitcoin Core.
Also in April, Pierre-Marie Padiou proposed the idea of trampoline payments, a method for allowing lightweight LN nodes to outsource pathfinding to heavyweight routing nodes. A lightweight node, such as a mobile app, might not keep track of the full LN routing graph, making it unable to find routes to other nodes. Padiou’s proposal would allow the lightweight node to route the payment to a nearby node and then have that node calculate the rest of the path. In essence, the payment would bounce off the trampoline node on the way to its destination. To add privacy, the original spender might require the payment bounce off several trampoline nodes in sequence so that none of them know whether or not it was routing the payment to the final recipient or just another trampoline node.
In May, Pieter Wuille proposed a taproot soft fork consisting of bip-taproot and bip-tapscript (which both depend on last year’s bip-schnorr proposal). If implemented, this change will allow single-sig, multisig, and many contracts to all use the same style of scriptPubKeys. Many spends from multisigs and complex contracts will also look identical to each other and single-sig spends. This can significantly improve user privacy and coin fungibility while also reducing the amount of block chain space used by multisig and contract use cases.
Even in cases where multisig and contract spends can’t take full advantage of taproot’s privacy and space savings, they still may only need to put a subset of their code onchain, giving them more privacy and space savings than they have today. In addition to taproot, tapscript brings small refinements to Bitcoin’s scripting capabilities, mainly by making it easier and cleaner to add new opcodes in the future.
The proposals received significant discussion and review throughout the rest of the year, including through a series of group review sessions organized by Anthony Towns that had more than 150 people sign up to help review.
Towns also proposed in May two new signature hashes to be used in
combination with tapscript,
SIGHASH_ANYPREVOUTANYSCRIPT. A signature hash (sighash) is the hash
of a transaction’s fields and related data to which a signature commits. Different
sighashes in Bitcoin commit to different parts of a transaction, allowing
signers to optionally let other people make certain modifications to
their transactions. The two new proposed sighashes function similar to
BIP118’s SIGHASH_NOINPUT by deliberately not
identifying which UTXO they spend, allowing the signature
to spend any UTXO whose script it can fulfill (e.g. that uses the same
The primary suggested use for noinput-style sighashes is to enable the previously proposed eltoo update layer for LN. Eltoo can simplify several aspects of channel construction and management; it’s especially desirable for simplifying channels involving more than two participants that can significantly reduce onchain channel costs.
A third soft fork proposed this month came from Jeremy Rubin, who
described a new opcode now called
(CTV). This would allow a limited form of covenant
where an output of one transaction would require a subsequent transaction
spending it to contain certain other outputs. A suggested use for this
would be committed future payments where a spender pays a single small
output that can only be spent using a transaction (or a tree of
transactions) that later pays dozens, hundreds, or even thousands of
different receivers. This could enable new techniques to enhance
coinjoin-style privacy, support security-enhancing vaults, or
manage spender costs when transaction feerates spike.
Rubin would continue working on CTV for the remainder of the year, including opening PRs (1, 2) for improvements to parts of Bitcoin Core where optimizations could make a deployed version of CTV more effective.
Notable technical conferences and other events
- Stanford Blockchain Conference, January, Stanford University
- MIT Bitcoin Expo, March, MIT
- Optech Executive Briefing, May, New York City
- Magical Crypto Friends (technical track), May, New York City
- Breaking Bitcoin, June, Amsterdam
- Bitcoin Core developers meetup, June, Amsterdam
- Edge Dev++, September, Tel Aviv
- Scaling Bitcoin, September, Tel Aviv
- Cryptoeconomic Systems Summit, October, MIT
Gleb Naumenko, Pieter Wuille, Gregory Maxwell, Sasha Fedorova, and Ivan Beschastnikh published a paper about erlay, a protocol for relaying unconfirmed transaction announcements between nodes that makes use of libminisketch-based set reconciliation to produce an estimated 84% reduction in announcement bandwidth. The paper also demonstrates that erlay would make it much more practical to significantly increase the default number of outbound connections that nodes make. This could improve each node’s resistance to eclipse attacks that can trick it into accepting blocks not on the most proof-of-work block chain. More outbound connections also improves node resistance against other attacks that could be used to track or delay payments originating from the node. Work on erlay would continue through the year with additional research and the proposal of BIP330 for the set reconciliation protocol.
Other improvements made in P2P relay this year included Bitcoin Core’s privacy improvements for transaction relay (eliminating a problem described in the TxProbe paper by Sergi Delgado-Segura and others) and the addition of two extra outbound connections used only for the relay of new blocks, improving resistance against eclipse attacks.
After a significant amount of prior work, June also saw the merge of altruist LN watchtowers into LND. Altruist watchtowers don’t receive any reward via the protocol for helping to secure their client’s channels, so a user needs to run their own watchtower or depend on the charity of a watchtower operator, but this is enough to demonstrate that watchtowers can reliably send penalty transactions on behalf of other users—ensuring that users who go offline for significant amounts of time don’t lose any money.
Altruist watchtowers would eventually be released in LND 0.7.0-beta and would see additional development through the remainder of the year, including a proposed specification and discussion about how they could be combined with next-generation payment channels such as eltoo.
In July, the Bitcoin Core project merged Carl Dong’s PR adding
support for reproducible builds of Bitcoin Core’s Linux binaries using
GNU Guix (pronounced “geeks”). Although Bitcoin Core has long provided
support for reproducible builds using the Gitian system, it can be
difficult to set up and it depends on the security of several hundred
Ubuntu packages. By comparison, Guix can be much easier to install and
run, and builds of Bitcoin Core using it currently depend on a much
smaller number of packages. In the long term, contributors to Guix are
also working on eliminating the trusting trust problem to make it
easy for users to verify that binaries such as
bitcoind are derived
solely from auditable source code.
Work continued on Guix build support throughout the year, with some contributors hopeful that Guix will be used for the first major version of Bitcoin Core released in 2020 (perhaps in parallel with the older Gitian-based mechanism).
In August, Bryan Bishop described a method for implementing vaults on Bitcoin without using covenants. Vaults is a term used to describe a script that limits an attacker’s ability to steal funds even if they obtain a user’s normal private key. A covenant is a script that can only be spent to certain other scripts. There’s no known way to create covenants using the current Bitcoin Script language, but it turns out that they’re not necessary if users are willing to run code that performs a few extra steps when depositing their money into the vault contract.
Perhaps more notably, Bishop described a new weakness in previous vault proposals as well as a mitigation for the weakness that would limit the maximum amount of funds that could be stolen from a vault by an attacker. The development of practical vaults could be useful for both individual users and large custodial organizations such as exchanges.
In Optech’s second year, we signed up six new member companies, held an executive briefing during NYC block chain week, published a 24-week series promoting bech32 sending support, added a wallet and services compatibility matrix to our website, published 51 weekly newsletters, saw several of our newsletters and blog posts translated into languages such as Japanese and Spanish, created a topics index, added a chapter to our Scalability Workbook, hosted two schnorr/taproot workshops with publicly released jupyter notebooks, and published field reports from BTSE and BRD.
Adam Gibson proposed a novel form of non-interactive coinjoin for the existing Bitcoin system. The protocol, called SNICKER, involves a user selecting one of their UTXOs and a randomly-selected UTXO from the global UTXO set to both be spent in the same transaction. The proposing user signs their part of this transaction and uploads it in the Partially Signed Bitcoin Transaction (PSBT) format to a public server. If the other user checks the server and sees the PSBT, they can download it, sign it, and broadcast it—completing the coinjoin without both users needing to be online at the same time. The proposing user can create and upload as many PSBTs as they want using their same UTXO until some other user accepts the coinjoin.
SNICKER’s major advantages over other coinjoin approaches are that it doesn’t require the users be online at the same time and that it should be easy to add support for it to any wallet that already has BIP174 PSBT support, which is an increasing number of wallets.
Also in September, the maintainers of C-Lightning, Eclair, and LND disclosed a vulnerability that affected previous versions of their software. It appeared that, in some cases, each of the implementations failed to confirm that channel funding transactions paid the correct script or the correct amount (or both). If exploited, this could result in channel payments being impossible to confirm onchain, making it possible for nodes to lose money by relaying payments from an invalid channel to a valid channel. Optech is unaware of any users who lost money before the first public announcements of the vulnerability. The LN specification was updated to help future implementers avoid this problem and there’s an expectation that other proposed changes to LN’s communication protocol will help avoid other failures of this type.
LN developers made significant progress in October and November towards addressing a long-standing concern about ensuring that users can always close their channels without excessive delays. If a user decides that they want to close one of their channels and they’re unable to contact their remote peer, they broadcast the latest commitment transaction for that channel—a pre-signed transaction that spends the channel’s funds onchain to each party according to the latest version of their offchain contract. A potential problem with this arrangement is that the commitment transaction was potentially created days or weeks earlier when transaction fees were lower, so it may not pay a high enough fee to confirm quickly before any security-essential time locks expire.
It’s always been known that the solution to this problem is to make it possible to fee bump commitment transactions. Unfortunately, nodes such as Bitcoin Core have to limit the use of fee bumping in order to prevent Denial of Service (DoS) attacks that waste their bandwidth and CPU. In trustless multi-user protocols like LN, your counterparty might be an attacker who could deliberately trigger the anti-DoS policy in order to delay the confirmation of your LN commitment transaction, an attack sometimes called transaction pinning. A pinned transaction may not confirm before its time locks expire, allowing an attacking counterparty to steal funds from you.
Last year, Matt Corallo suggested carving out a special exemption from the part of Bitcoin Core’s transaction relay policy related to Child-Pays-For-Parent (CPFP) fee bumping. This limited exemption ensures that two-party contract protocols (such as current-generation LN) can guarantee each party the ability to create their own fee bump. Corallo’s idea was named CPFP carve-out and his implementation of it was released as part of Bitcoin Core 0.19. Even before that release, other LN developers worked on the revisions to the LN scripts and protocol messages necessary to start using the change. As of this writing, those specification changes are awaiting final implementation and acceptance before seeing deployment on the network.
New open source infrastructure solutions
Hardware Wallet Interface released in March makes it easy for a wallet already compatible with Partially Signed Bitcoin Transactions (PSBTs) and output script descriptors to use several different models of hardware wallets for secure key storage and signing.
Lightning Loop released in March (with loop-in support added in June) provides a non-custodial service that allows users to add or remove funds from their LN channels without closing existing channels or opening new channels.
Discussion in November about using bech32 addresses for taproot payments brought additional attention to an issue discovered in May. According to BIP173,
mis-copied bech32 strings are supposed to have a worst-case failure rate
of about 1-in-a-billion. However, it was discovered that bech32 strings
ending with a
p could have any number of preceding
added or removed. This doesn’t practically affect bech32 addresses for
segwit P2WPKH or P2WSH addresses, as at least 19 consecutive
characters would need to be added or removed in order to transform one
address type into another—and any other length change for v0 segwit
addresses would be invalid.
But that’s not the case for v1+ segwit addresses, such as those proposed for taproot, where a
single added or removed
q character in a vulnerable address could lead
to a loss of funds. BIP173 co-author Pieter Wuille performed
additional analysis and found that this was the only
deviation from bech32’s expected error correction ability, so he
proposed limiting the use of BIP173 addresses in Bitcoin to only 20
byte or 32 byte witness programs. This will ensure that v1 and subsequent segwit
address versions provide the same reliable error correction as v0 segwit
addresses. He also described a small tweak to the bech32 algorithm that
will allow other applications using bech32, as well as next-generation
Bitcoin address formats, to use BCH error detection without this
Also in November, Bitcoin Core removed its dependency on OpenSSL, which had been part of its codebase since the original 2009 release of Bitcoin 0.1. OpenSSL was the cause of consensus vulnerabilities, remote memory leaks (potential private key leaks), other bugs, and poor performance. It’s hoped that its removal will reduce the frequency of future vulnerabilities.
As part of the OpenSSL removal, Bitcoin Core deprecated its support for the BIP70 payment protocol in version 0.18, and later disabled support by default in version 0.19. This decision was supported by the CEO of one of the few companies that continued to use BIP70 in 2019.
In December, LN developers achieved one of their major goals from last year’s planning meeting: the implementation of basic multipath payments. These are payments that can be split into several parts, with each part being routed separately through different channels. This allows users to spend or receive money using more than one of their channels at a time, making it possible to spend their full offchain balance or receive up to their full capacity in a single payment (within the limitations of certain safety restrictions). It’s expected that this will make LN significantly more user-friendly by eliminating the need for spenders to worry about the balances of specific channels.
In the summary above, we see no revolutionary proposals or improvements. Instead, we see a flurry of incremental improvements—solutions that take cases where Bitcoin and LN are already successful and build on them to make the system even better. We see developers working to make hardware wallets more accessible (HWI), generalize communication between wallets for multisig and contract use cases (descriptors, PSBTs, miniscript), strengthen consensus security (cleanup soft fork), simplify testing (signet), eliminate unnecessary custody (loop), make it easier to start running a node (assumeutxo), improve privacy and save block space (taproot), simplify LN enforcement (anyprevout), better manage feerate spikes (CTV), reduce node bandwidth (erlay), keep LN users safe when offline (watchtowers), reduce the need for trust (reproducible builds), prevent thefts (vaults), make privacy more accessible (SNICKER), better manage onchain fees for LN users (anchor outputs), and make LN payments automatically work more often (multipath payments).
(And those are just the highlights for the year!)
We can only guess what Bitcoin contributors will accomplish next year, but we suspect it will be more of the same—dozens of modest changes that each make the system better without breaking it for anyone who’s already satisfied.
The Optech newsletter will return to its regular Wednesday publication schedule on January 8th.