Note
This repository is in maintenance mode. Development has moved to https://github.com/argumentcomputer/lurk.
This repository contains the implementation for Lurk Beta, which is backwards compatible with code that ran in Lurk Alpha and is expected to be compatible with Lurk 1.0. However, some low-level data representations are anticipated to change, and we will be refactoring the evaluation model (and consequently its circuit) for efficiency purposes. Also note that since Lurk inherits some security properties from the underlying proving system, those who would rely on Lurk should investigate the security and status of Nova/SuperNova itself. We encourage early adopters to begin writing real applications taking advantage of Lurk so you can begin to familiarize yourself with the programming model. Likewise, we welcome your feedback -- which will help ensure ongoing development meets user need.
For support and discussions, please visit our Zulip forum.
Lurk is a statically scoped dialect of Lisp, influenced by Scheme and Common Lisp. A reference implementation focused on describing and developing the core language can be found in the lurk-lisp repo.
Lurk's distinguishing feature relative to most programming languages is that correct execution of Lurk programs can be directly proved using zk-SNARKs. The resulting proofs are succinct: they are relatively small, can be verified quickly, and they reveal only the information explicitly contained in the statement to be proved.
For more detailed information, refer to the paper: https://eprint.iacr.org/2023/369
Lurk's distinguishing feature relative to most zk-SNARK authoring languages is that Lurk is Turing complete, so arbitrary computational claims can be made and proved (subject to resource limitations, obviously). Because Lurk is a Lisp, its code is simply Lurk data, and any Lurk data can be directly evaluated as a Lurk program. Lurk constructs compound data using SNARK-friendly Poseidon hashes (provided by Neptune), so its data is naturally content-addressable.
Integration with backend proving systems and tooling for proof generation are both still very early. Performance and user experience still have room for significant optimization and improvement, but simple examples can be found in the demo example directory.
- Nova is Lurk's officially-supported IVC backend. It uses Argument's Arecibo fork of the Nova proving system and the Pasta Curves.
- SuperNova is Lurk's in-development NIVC backend. It uses Arecibo's SuperNova extension to the Nova proving system and the Pasta Curves.
- Future work may target Halo2 or other proving systems.
It is an explicit design goal that statements about the evaluation of Lurk programs have identical semantic meaning across backends, with the qualification that Lurk language instances are themselves parameterized on scalar field and hash function. When backends use the same scalar field and hash function, they can be used to generate equivalent proofs. This is because the concrete representation of content-addressed data is fixed.
Lurk backend integration is still immature, so current performance is not representative. As a rough approximation, we estimate that for entirely general computation using Lurk's universal circuit, Nova proving throughput will be on the order of 1,000 iterations per second per GPU. We expect that most compute-heavy applications will use optimized 'coprocessor' circuits, which will dramatically improve performance. Planned improvements to Nova will allow for smaller inner circuits, further improving throughput -- and for full parallelization of reduction proofs.
Lurk's Alpha release has undergone a security audit as of 03/29/2023, performed by Inference.
Please note that the Lurk language and spec will be versioned independently from the crates that implement the spec. This is necessary because semantic versioning implies different requirements for the language and its implementation. For example, Lurk Alpha is released as crate lurk 0.2.0 and Lurk Beta is released as crate lurk 0.3.0. It is our intention for these two versioning systems to coincide at 1.0.
Lurk source files used in tests are in the lurk-lib submodule. You must initialize and update submodules before test will pass:
git submodule update --init --recursiveLurk can be compiled to Wasm with
cargo build --target wasm32-unknown-unknownIf using Nix without a system-wide clang install (e.g. NixOS):
CC=clang cargo build --target wasm32-unknown-unknowncargo run --releaseOr use the wrapper script:
bin/lurkSet the environment variable LURK_FIELD to specify the scalar field of the Lurk language instance:
LURK_FIELD=PALLAS(default): scalar field of PallasLURK_FIELD=VESTA: scalar field of Vesta
➜ lurk-beta ✗ bin/lurk
Finished release [optimized] target(s) in 0.06s
Running `target/release/lurk`
Lurk REPL welcomes you.
> (let ((square (lambda (x) (* x x)))) (square 8))
[9 iterations] => 64
>Or enable info log-level for a trace of reduction frames:
➜ lurk-beta ✗ RUST_LOG=lurk::lem::eval=info bin/lurk
Finished release [optimized] target(s) in 0.05s
Running `target/release/lurk`
Lurk REPL welcomes you.
user> (let ((square (lambda (x) (* x x)))) (square 8))
2023-12-10T15:58:21.902414Z INFO lurk::lem::eval: Frame: 0
Expr: (let ((.lurk.user.square (lambda (.lurk.user.x) (* .lurk.user.x .lurk.user.x)))) (.lurk.user.square 8))
Env: nil
Cont: Outermost
at src/lem/eval.rs:99
2023-12-10T15:58:21.902943Z INFO lurk::lem::eval: Frame: 1
Expr: (lambda (.lurk.user.x) (* .lurk.user.x .lurk.user.x))
Env: nil
Cont: Let{ var: .lurk.user.square, saved_env: nil, body: (.lurk.user.square 8), continuation: Outermost }
at src/lem/eval.rs:107
2023-12-10T15:58:21.903170Z INFO lurk::lem::eval: Frame: 2
Expr: (.lurk.user.square 8)
Env: ((.lurk.user.square . <FUNCTION (.lurk.user.x) (* .lurk.user.x .lurk.user.x)>))
Cont: Tail{ saved_env: nil, continuation: Outermost }
at src/lem/eval.rs:107
2023-12-10T15:58:21.903649Z INFO lurk::lem::eval: Frame: 3
Expr: .lurk.user.square
Env: ((.lurk.user.square . <FUNCTION (.lurk.user.x) (* .lurk.user.x .lurk.user.x)>))
Cont: Call{ unevaled_arg: 8, saved_env: ((.lurk.user.square . <FUNCTION (.lurk.user.x) (* .lurk.user.x .lurk.user.x)>)), continuation: Tail{ saved_env: nil, continuation: Outermost } }
at src/lem/eval.rs:107
2023-12-10T15:58:21.903678Z INFO lurk::lem::eval: Frame: 4
Expr: 8
Env: ((.lurk.user.square . <FUNCTION (.lurk.user.x) (* .lurk.user.x .lurk.user.x)>))
Cont: Call2{ function: <FUNCTION (.lurk.user.x) (* .lurk.user.x .lurk.user.x)>, saved_env: ((.lurk.user.square . <FUNCTION (.lurk.user.x) (* .lurk.user.x .lurk.user.x)>)), continuation: Tail{ saved_env: nil, continuation: Outermost } }
at src/lem/eval.rs:107
2023-12-10T15:58:21.903696Z INFO lurk::lem::eval: Frame: 5
Expr: (* .lurk.user.x .lurk.user.x)
Env: ((.lurk.user.x . 8))
Cont: Tail{ saved_env: nil, continuation: Outermost }
at src/lem/eval.rs:107
2023-12-10T15:58:21.903772Z INFO lurk::lem::eval: Frame: 6
Expr: .lurk.user.x
Env: ((.lurk.user.x . 8))
Cont: Binop{ operator: product#, saved_env: ((.lurk.user.x . 8)), unevaled_args: (.lurk.user.x), continuation: Tail{ saved_env: nil, continuation: Outermost } }
at src/lem/eval.rs:107
2023-12-10T15:58:21.903844Z INFO lurk::lem::eval: Frame: 7
Expr: .lurk.user.x
Env: ((.lurk.user.x . 8))
Cont: Binop2{ operator: product#, evaled_arg: 8, continuation: Tail{ saved_env: nil, continuation: Outermost } }
at src/lem/eval.rs:107
2023-12-10T15:58:21.903866Z INFO lurk::lem::eval: Frame: 8
Expr: Thunk{ value: 64 => cont: Outermost }
Env: nil
Cont: Dummy
at src/lem/eval.rs:107
2023-12-10T15:58:21.903878Z INFO lurk::lem::eval: Frame: 9
Expr: 64
Env: nil
Cont: Terminal
at src/lem/eval.rs:107
[9 iterations] => 64
user> You can install the lurk Repl on your machine with
$ cargo install --path .Install Nix and enable Nix flakes. Then, you can enter into a Nix devshell with the appropriate dependencies for Lurk with
$ nix developor
$ direnv allowAnd then build with Cargo as usual:
$ cargo buildMIT or Apache 2.0