Skip to main content

API

Overview

Technically, the API fawkes-crypto provides for describing circuits for use in R1CS and Plonk backends are different. Each is enabled using r1cs and plonk Cargo features. In this section we start with high-level overview of their common principles and then we dive into the concrete code examples of using each in the later sections.

Fawkes-crypto API uses the following vocabulary for describing Constraint Systems.

caution

The code examples we give here are have some technicalities (like borrowing and wrapper types) simplified in them to convey the intuition — this is best read as pseudocode. We will look at the formally correct, runnable examples in the next sections.

  • trait CS describes the whole Constraint System. It has methods for allocating new variables in pub\texttt{pub} and sec\texttt{sec} as well as adding new constraints to the system.

    The field over which the CS constraints work is given by the associated type CS::Fr.

    Two main implementors of this trait are struct BuildCS and struct DebugCS. The former one is the format which proving and verifying functions accept, while the latter is useful for the circuit.

  • struct CNum<CS> is a reference to a specific variable in the CS. For example, if you allocate a new variable with

    fn CS::alloc(self, value: Option<CS::Fr>) -> CNum<CS>

    you get a return value of type CNum<CS>. If you're Prover, you should pass value = Some(x) to set this new variable's value to x. If you're Verifier, you pass value = None since you may not know the value of this variable (and it's not your job to compute it).

    If you want to enforce a constraint on some variables of your CS, you will also refer to the variables using CNum<CS>. For example, given a, b, c : CNum<CS>, to add the constraint a×b=ca \times b = c to your CS you may call a function like

    // a * b == c
    fn CS::enforce_mul(a: CNum<CS>, b: CNum<CS>, c: CNum<CS>);

    If you have a x: CNum<CS> that refers to a secret variable, i.e. one from sec\texttt{sec}, and you want to make it public moving it to pub\texttt{pub} you call

    fn CS::inputize(n: CNum<CS>);

    You can also do regular arithmetic operations on CNums. This applies operations to the wires in the circuit model of our CS, creating new wires (and leaving input wires available for further use, of course). For example, if we have a, b: CNum<CS>, we can do let c = a * b. This will alloc a new variable for c (using the correct value derived as the product of values of a and b), and then do enforce_mul(a, b, c).

  • struct CBool<CS> is just a regular CNum<CS> which was restricted (using a constraint) to hold only values 0 or 1 (in CS::Fr). It can be used for if-then-else/switch constructs which we will see shortly.

  • trait Signal<CS> generalizes CNum<CS>, CBool<CS> as well as fixed-size vectors and pairs of other Signals. You can also implement Signal trait for your own types (such types will typically store some number of CNums and CBools in them).

    A Signal<CS> is something you can copy, test for equality with a Signal<CS> of the same type, allocate, inputize — all with a single call to the corresponding method.

    Since Signal<CS> is a generic type that circuit operates on, it can be passed in the "then" and "else" branches of the conditional "switch" operation:

    // Returns self if bit is true, if_else otherwise
    fn Signal::switch(&self, bit: CBool<C>, if_else: Self) -> Self

    You can do let (x, y) = (a,b).switch(bit, (c, d)) to implement conditional assignment

    (x,y)={(a,b) if bit=1(c,d) otherwise.(x, y) = \begin{cases} (a, b) & \text{ if $\text{bit} = 1$} \\ (c, d) & \text{ otherwise.} \end{cases}

    in your CS. This works because (CNum<CS>, CNum<CS>) is an instance of Signal<CS>.

  • Num<Fp> is a wrapper type for Fp field. We use it to implement traits for Fp. It's a rather technical detail, it's safe to view Num::from(x) as x.

Now we will look at more concrete (runnable) code examples that use R1CS and Plonk constraints. They are enabled using either r1cs or plonk Cargo flags. Once you enable one of the two flags, the corresponding definitions of CS, CNum, CBool will appear in cicuit::{cs, num, bool}.

Plonk Constraints

In Fawkes-crypto circuits are described as functions of type

circuit: Fn(Signal<BuildCS<Fr>>, Signal<BuildCS<Fr>>)

The two arguments are pub and sec. Their concrete types can be anything that implements Signal, e.g. CNum<_>, CBool<_>, fixed-size vectors of CNum or CBool. The circuit doesn't accept the cs: BuildCS<Fr> itself, because Signal<_> values store smart reference to the CS value they correspond to. Both pub and sec must, of course, belong to the same CS.

tip

The pub and sec here can have different types. For example, you could make pub: (CNum<_>, CNum<_>) and sec: (CNum<_>, CNum<_>, CNum<_>) to have 2 public inputs and 3 secret ones.

To verify a circuit using Plonk constraints (Halo2 backend with KZG10 commitments), one does the following steps.

  1. Pick public parameters and generate Verifier and Prover keys.

    use fawkes_crypto::backend::plonk::{
    engines::Bn256,
    setup::setup,
    Parameters
    };

    // Generate parameters.
    //
    // The k=10 here is a parameter of KZG10 commitment scheme. See its paper
    // for details. And Bn256 is one of the available "engines" that
    // fawkes-crypto implements for Plonk; for details, see
    // backend::plonk::engines module.
    let parameters = Parameters::<Bn256>::setup(10);

    // Sample keys. The vk goes to Verifier, pk goes to Prover
    let (vk, pk) = setup(&parameters, circuit);
    caution

    This step is also called trusted setup. It must be executed either by a trusted third party or in a Secure Multi-Party Computation protocol, Prover and Verifier must not see the intermediate values from setup computation.

  2. Generate the proof. This step is done by the Prover using the public inputs pub and secret input sec.

    use fawkes_crypto::backend::plonk::prover;

    let (inputs, proof) = prover::prove(
    &parameters,
    &pk,
    &pub, // actual value of pub to pass to circuit
    &sec, // actual value of sec to pass to circuit
    circuit,
    );

    Prover communicates both inputs and proof to the verifier. The proof is the proof\texttt{proof} we've introduced before. And inputs is the list of all public inputs of the circuit, it will include both pub and all the other public inputs that circuit will create and assign values to duing its execution. It may contain some computation results that the Prover wants to reveal to the Verifier, for example.

  3. Verify the proof. This is done by the Verifier.

    use fawkes_crypto::backend::plonk::verifier;

    let result: bool = verifier::verify(
    &parameters,
    &vk,
    &proof,
    &inputs,
    );

    This uses proof and inputs that Verifier got from the Prover.

    caution

    Depending on the application, the Verifier may need to perform some checks on inputs value before using it in verify. For example, make sure that pub\texttt{pub} values included in inputs are the ones that Verifier expects.

    This is highly application specific, therefore we don't show this in code above.

Both setup and prove will call the circuit function that you pass to them to build the circuit and extract the information they need from it.

R1CS Constraints

R1CS constraints in fawkes-crypto are described with a function of the same type as in Plonk (presented in the previous section):

circuit: Fn(Signal<BuildCS<Fr>>, Signal<BuildCS<Fr>>)

To have your circuit proven, you perform the following steps.

  1. Generate setup parameters. This is trusted setup, it must be performed by a trusted party or in an MPC protocol.

    use fawkes_crypto::backend::bellman_groth16::{
    engines::Bn256,
    setup::setup
    }
    // The Bn256 is one of the engines fawkes-crypto provides. See
    // fawkes_crypto::backend::bellman_groth16 for more options.
    let params = setup::<Bn256, _, _, _>(circuit);
  2. Prove the circuit.

    use fawkes_crypto::backend::bellman_groth16::prover;
    let (inputs, proof) = prover::prove(&params, &pub, &sec, circuit);
  3. Verify the circuit.

    use fawkes_crypto::backend::bellman_groth16::verifier;
    let res = verifier::verify(&params.get_vk(), &snark_proof, &inputs);
caution

Like in Plonk which we describe in the previous section, setup from step 1 shown above must be performed by a trusted party. Neither Prover nor Verifier should be allowed to tamper with it. And in step 3, Verifier must validate the contents of inputs (application-dependent).

The API for R1CS constraints is quite similar to Plonk. The main differences are:

  • Setup is called differently. The available setup parameters and the list of engines are different due to R1CS and Plonk working over different backends and polynomial commitment schemes (which imposes its own restrictions on the fields over which your constraints are applied).

  • The methods of trait CS are slightly different. Plonk's CS provides enforce_mul and enforce_add while R1CS has only enforce for enforcing multiplications (equivalent to enforce_mul). This is due to R1CS allowing linear operations for free, with no constraint overhead, so you can just create fresh CNums whenever you need additions. In other words, when you do

    let a: CNum<CS> = CNum<_>::alloc(rcs, …);
    let b: CNum<CS> = CNum<_>::alloc(rcs, …);
    let c: CNum<CS> = a + b;

    in Plonk, this causes a call to rcs.borrow_mut().enforce_add(&a, &b, &c) creating an extra constraint in rcs. In R1CS, the same code will not modify rcs in any way — all the information about c's relation to a and b will be stored inside c itself.

Cicuits Library

Fawkes-crypto has a few useful circuits defined for the programmer to import when building their Constraint Systems.

  • circuit::bitify defines circuits for bit-decomposition of CNum<CS> values into Vec<CBool<CS>> as well as circuits for comparing CNum<CS> values for inequality with << and >>.

  • circuit::ecc defines curcuits for elliptic-curve operations.

  • circuit::poseidon and circuit::eddsaposeidon implement the Poseidon hash.

  • circuit::mux3 implements Pedersen hash.