search

vacp2p / zerokit

A set of Zero Knowledge modules, written in Rust and designed to be used in other system programming environments.
Apache License 2.0
130 stars 7 forks source link

Unable to see constraint traces #3

Open oskarth opened 2 years ago

oskarth commented 2 years ago

Problem

When debugging circuits, it is very useful to see what constraints haven't been met and why.

Details

This should be possible with ConstraintLayer but I haven't managed to get to work properly:

In poseidon_tornado with

RUST_LOG="poseidon_tornado=TRACE,ark_relations=TRACE" cargo run

We get

Constraint trace requires enabling `ConstraintLayer`
thread 'main' panicked at 'assertion failed: {\n    use ark_relations::r1cs::{ConstraintSynthesizer, ConstraintSystem};\n    let cs = ConstraintSystem::<E::Fr>::new_ref();\n    circom.clone().generate_constraints(cs.clone()).unwrap();\n    cs.is_satisfied().unwrap()\n}', /home/oskarth/repos/github.com/gakonst/ark-circom/src/circom/builder.rs:82:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

The panic happens because constraints aren't met, which is expected. This then panics because of the debug_assert here https://github.com/gakonst/ark-circom/blob/master/src/circom/builder.rs#L82-L87

The problem is the lack of tracing, even with https://github.com/vacp2p/zerokit/commit/9b1fa85e46fc1d37f2f3e201ea6624940582925a

Tracing instrumentation based on example here https://github.com/arkworks-rs/r1cs-tutorial/blob/main/merkle-tree-example/src/constraints.rs#L109-L113

Same when trying to do tracing instrumentation in various ways directly in ark-circom etc.

Error from https://github.com/arkworks-rs/snark/blob/master/relations/src/r1cs/constraint_system.rs#L609

Probably missing something simple here but haven't been to figure out what's wrong.

Acceptance criteria

Failing constraints should produce some form of useful debugging information re which constraints are failing.

oskarth commented 2 years ago

This is actually quite a big problem, e.g. a typo "nullifer" vs "nullifier" had me stumped for quite a while. Right now there's zero useful feedback here. Some kind of type assertions would be useful here? Probably requires looking into proof/witness WASM format a bit more.