Target 0xMiden Plonky3 repo and use AirScriptAir and AirScriptBuilder traits #508

Leo-Besancon · 2025-11-05T11:15:18Z

Describe your changes

The PR introduces:

the AirScriptAir and AirScriptBuilder traits proposed in AirScript integration Plonky3#5 (for now, in the AirScript repo, but will latter be imported from 0xMiden/Plonky3 once our testing is done)
Aux trace constraints for Plonky3 Codegen using the PermutationBuilder trait
An adaptation of the DebugConstraintBuilder struct that generates an aux trace (hardcoded for the only Air that is tested against the aux trace, buses_complex_plonky3::BusesAir)

Checklist before requesting a review

Relevant issues are linked in the PR description.
Documentation/comments updated according to changes.

Leo-Besancon · 2025-11-10T09:31:55Z

@adr1anh here I assumed that the randomness was given by:

            Value::RandomValue(idx) => {
                if *idx == 0 {
                    format!("beta.into()")
                } else {
                    format!("alpha_challenges[{}].into()", idx - 1)
                }
            },

But maybe it should only be alpha_challenges for the constraints, beta being used internally in the prover/verifier?

adr1anh

looks great just left some small comments

air-script/tests/helpers/mod.rs

air-script/tests/helpers/plonky3_traits.rs

codegen/plonky3/src/air/integrity_constraints.rs

Al-Kindi-0

LGTM
Would be great to think of a way to compare the outputs of Winterfell against P3. Haven't spent time thinking about this so not sure how realistic and straightforward this is.

air-script/tests/helpers/plonky3_traits.rs

air-script/src/test_utils/plonky3_traits.rs

air-script/tests/helpers/plonky3_traits.rs

adr1anh · 2025-11-11T10:36:40Z

From the P3 side, we should be able to define an air builder which evaluates the constraints at a set of random points (there's already something super close), but not sure how we can do with with Winterfell

Al-Kindi-0 · 2025-11-11T11:04:46Z

For Winterfell, we might be able to use the verifier constraints evaluator (or a small modification of it) in order to evaluate the constraints on an OOD evaluation frame.

…_utils module

…ness

Copilot

Pull Request Overview

This PR introduces support for the 0xMiden Plonky3 repository by implementing AirScriptAir and AirScriptBuilder traits for improved code generation. The changes add support for auxiliary trace constraints in Plonky3, refactor the code generation to use new traits, and reorganize test structure to use a centralized test utilities module.

Key changes:

Introduced AirScriptAir and AirScriptBuilder traits for Plonky3 code generation
Added auxiliary trace constraint generation support
Reorganized test structure with new test_utils module replacing scattered helper files

Reviewed Changes

Copilot reviewed 128 out of 208 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
scripts/generate_all_e2e_tests.sh	New shell script for generating all end-to-end tests for both Winterfell and Plonky3 backends
codegen/plonky3/src/imports.rs	Updated imports to use new AirScript traits instead of custom helper traits
codegen/plonky3/src/air/mod.rs	Refactored AIR generation to use new trait structure with separate constants, main AIR implementation, and AirScriptAir trait
codegen/plonky3/src/air/integrity_constraints.rs	Added support for auxiliary integrity constraints alongside main constraints
codegen/plonky3/src/air/graph.rs	Enhanced constraint string generation to support extension field types and determine field type requirements
codegen/plonky3/src/air/boundary_constraints.rs	Added auxiliary boundary constraint generation alongside main boundary constraints
air/src/passes/expand_buses.rs	Changed constraint domain from EveryRow to EveryFrame(2) for bus expansion
air-script/tests/* (multiple files)	Removed old test files using deprecated helper structure
air-script/src/tests/* (multiple files)	Added new test files using AirScriptAir trait and reorganized into src/tests directory

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

scripts/generate_all_e2e_tests.sh

Copilot · 2025-11-12T10:55:32Z

codegen/plonky3/src/air/graph.rs

+            Value::Constant(0) => match elem_type {
+                ElemType::Base => format!("AB::Expr::ZERO"),
+                ElemType::Ext => format!("AB::ExprEF::ZERO"),
+            },
+            Value::Constant(1) => match elem_type {
+                ElemType::Base => format!("AB::Expr::ONE"),
+                ElemType::Ext => format!("AB::ExprEF::ONE"),
+            },


The repeated match statements for Value::Constant(0) and Value::Constant(1) could be consolidated using a helper function to reduce duplication and improve maintainability.

Copilot · 2025-11-12T10:55:32Z

air-script/src/tests/buses/buses_complex.air


    p.insert(2, b) when 1 - s1;
    p.remove(2, a) when 1 - s2;



[nitpick] Lines 37-38 insert the same value a at the same position (3) with different conditions (s3 and s4). Consider adding a comment explaining why this duplication is intentional or if it should be different values/positions.

Suggested change

// Intentionally insert the same value `a` at position 3 under both `s3` and `s4` conditions.

// This ensures that the bus `q` receives `a` regardless of which condition is met.

Co-authored-by: Copilot <[email protected]>

Leo-Besancon · 2025-11-13T14:26:15Z

LGTM Would be great to think of a way to compare the outputs of Winterfell against P3. Haven't spent time thinking about this so not sure how realistic and straightforward this is.

Hi, I've tested a few things out, and was able to adapt Winterfell's evaluate_constraints on one side and Plonky3's VerifierConstraintFolder one the other and evaluate a random frame for both with a simple Air.

It seems to work with a few caveats (for now, the code is specific to the air and a given row, it's a bit messy, and it's on a simple case with no public inputs, no periodic columns, no aux trace.

I think I'll put it in another PR if I find a way to clean it up.

Al-Kindi-0

LGTM, great stuff!
I haven't checked every line of the codegen-ed code but I am assuming it should not be different from the Winterfell one. Also, I believe there is some work going to test consistency between the P3 and Winterfell backends

Al-Kindi-0 · 2025-11-25T13:05:27Z

air-script/src/test_utils/air_tester_macros.rs

        fn $test_name() {
            type Val = Goldilocks;
-            type Challenge = BinomialExtensionField<Val, 3>;
+            type Challenge = BinomialExtensionField<Val, 5>;


nit: it is probably enough to use just degree 2 here making things lighter, though it shouldn't matter really

fixed in 5f2f351

Al-Kindi-0 · 2025-11-25T13:09:38Z

air-script/src/test_utils/plonky3_traits.rs

+    /// Number of beta challenges used in the AIR.
+    fn num_beta_challenges(&self) -> usize {
+        0
+    }


nit: I would add a comment, and change the name of this method though not sure how easy it is, explaining that the number of beta challenge powers is equal to the width of the largest/widest message sent through all buses in an AIR instance

renamed to max_beta_challenge_power and added a comment on 871058c

Al-Kindi-0 · 2025-11-25T13:12:27Z

air-script/src/test_utils/plonky3_traits.rs

+{
+    /// EF evaluations of periodic columns at the AIR’s random point (z). Order defined by
+    /// AirScript.
+    fn periodic_evals(&self) -> &[<Self as ExtensionBuilder>::VarEF];


Q: shouldn't this take z as input?

The comment does seem to insinuate that, but I dont think periodic evals depend on the random point, they're always 1 or 0 for the same row, no matter the input.

Im leaving it as is for now, Ill sync with @Leo-Besancon about this to confirm.

The caller which creates the Builder will have already evaluated the columns at z, to avoid doing it at every row.

Al-Kindi-0 · 2025-11-25T13:13:18Z

air-script/src/test_utils/plonky3_traits.rs

+    /// AirScript.
+    fn periodic_evals(&self) -> &[<Self as ExtensionBuilder>::VarEF];
+
+    /// Global challenges in EF. (We can provide defaults; details not important here.)


I would remove the parenthesis and what is between them

Al-Kindi-0 · 2025-11-25T13:16:44Z

air-script/src/test_utils/plonky3_traits.rs

+    fn beta(&self) -> <Self as ExtensionBuilder>::VarEF;
+    fn beta_powers(&self) -> &[<Self as ExtensionBuilder>::VarEF];
+
+    /// Aux bus boundary values: EF finals, one per aux/bus column, carried in the proof.


I would write this as:

Auxiliary buses boundary values, one per bus column, attached to the proof.

or something better

Al-Kindi-0 · 2025-11-25T13:21:40Z

air-script/src/test_utils/plonky3_traits.rs

+/// Used in conjunction with [`check_constraints`] to simulate
+/// an execution trace and verify that the AIR logic enforces all constraints.


nit: alignment of the first line

Al-Kindi-0 · 2025-11-25T13:34:48Z

air-script/src/test_utils/plonky3_traits.rs

+    let logup_current = EF::from(aux_current[1].clone());
+    let logup_next = (a * logup_current + b + c - e) * d.inverse();
+
+    // Dummy implementation for illustration purposes.


Q: Should we create a TODO to remove this dummy implementation with something more solid?

I think this will be removed anyway once we have full support of auxiliary traces, right?

Added a TODO

I think this will be removed anyway once we have full support of auxiliary traces, right?

I think so but not certain, Ill sync with @Leo-Besancon on this

Hello! @Al-Kindi-0 This was used to build the aux trace for the bus complex air (that only worked with this specific air), but in #515 it has already been replaced by a dynamic building of the aux trace that works with any Air (in particular the buses_transitions function added in the Codegen here: https://github.com/0xMiden/air-script/pull/515/files#diff-ed5104fe9ad8f138c0530924ab07f8c45e2ed62332f7475757ec7bf657d8dd77R227-R262)

This way, we now have two ways to build the Aux trace, either with the codegen, or with the manual implementations given by the MidenVM repo.

Just so it is clear, the Plonky3 PRs are:

Add plonky3 backend #461 that outlines the base for the main trace codegen

Target 0xMiden Plonky3 repo and use AirScriptAir and AirScriptBuilder traits #508 (this one), based on Add plonky3 backend #461, that adds manual aux trace handling and periodic columns

Use Plonky3 super trait #515, based on Target 0xMiden Plonky3 repo and use AirScriptAir and AirScriptBuilder traits #508, that uses the super traits and adds automatic aux trace building.

It would be great to merge these PRs in a Plonky3 codegen tracking branch once they are ready :)

Al-Kindi-0 · 2025-11-25T13:39:45Z

air-script/src/test_utils/plonky3_traits.rs

+    let height = main.height();
+
+    let aux_bus_boundary_values: Vec<_> = (0..air.aux_width()).map(|_| EF::GENERATOR).collect();
+    let alpha_f: Vec<F> = (0..5).map(|i| F::from_u64(123456789 * i)).collect(); // Dummy alpha in F


nit: can't we just use EF directly and hence avoid leaking the degree extension (i.e., 5)?

fixed in 1d14475

Al-Kindi-0 · 2025-11-25T13:40:06Z

air-script/src/test_utils/plonky3_traits.rs

+    let aux_bus_boundary_values: Vec<_> = (0..air.aux_width()).map(|_| EF::GENERATOR).collect();
+    let alpha_f: Vec<F> = (0..5).map(|i| F::from_u64(123456789 * i)).collect(); // Dummy alpha in F
+    let alpha = EF::from_basis_coefficients_iter(alpha_f.iter().cloned()).unwrap();
+    let beta = EF::from_u64(987654321);


is this about #508 (comment) ?

Al-Kindi-0 · 2025-11-25T13:59:43Z

air-script/src/tests/computed_indices/test_air_plonky3.rs

+use std::marker::PhantomData;
+
+use p3_challenger::{HashChallenger, SerializingChallenger64};
+use p3_circle::CirclePcs;


Q: why do we need this?

I think this has to do with the macro generate_air_plonky3_test_with_airscript_traits that seems to use it.

Not sure why we're using this one tho, Ill see with Leo so we can use the proper imports.

In any case, the macro should use full path names so we do not need to import those dependencies explicitly.

In #515 (53eb542), I refactored the macro to import the dependencies explicitly and to use TwoAdicFriPcs instead of CirclePcs

Leo-Besancon added 3 commits November 4, 2025 16:42

feat: target latest commit on 0xMiden/Plonky3 main branch

94350e8

fix: AirBuilder::Var is now Clone, not Copy

69626dd

feat: Update to target AirScriptAir and AirScriptBuilder traits

4138e94

Leo-Besancon added the codegen label Nov 5, 2025

Leo-Besancon added 4 commits November 6, 2025 08:39

refactor: Make AirScriptAir dyn compatible, clean code

2b44fbf

feat: handle aux trace

cbea42d

tests: update winterfell E2E test following change

30e4b07

chore: fix tests

0633eeb

Leo-Besancon marked this pull request as ready for review November 6, 2025 13:48

adr1anh self-requested a review November 10, 2025 10:15

adr1anh approved these changes Nov 10, 2025

View reviewed changes

Al-Kindi-0 approved these changes Nov 11, 2025

View reviewed changes

air-script/tests/helpers/plonky3_traits.rs Outdated Show resolved Hide resolved

air-script/src/test_utils/plonky3_traits.rs Outdated Show resolved Hide resolved

air-script/tests/helpers/plonky3_traits.rs Outdated Show resolved Hide resolved

Leo-Besancon mentioned this pull request Nov 12, 2025

[Plonky3] Use prove / verify in E2E tests for full workflow #512

Closed

Leo-Besancon added 5 commits November 12, 2025 08:29

chore(tests): move air-script/tests codegen and helpers into src/test…

1c8c54a

…_utils module

tests: add codegen script to help update all e2e tests

de9088a

refactor: review comments

d81248b

chore: fmt + test fixes

1e71f68

refactor: reverse the roles of alpha and beta, fix permutation_random…

e2f86e9

…ness

adr1anh requested review from adr1anh and Copilot November 12, 2025 10:54

adr1anh approved these changes Nov 12, 2025

View reviewed changes

Copilot AI reviewed Nov 12, 2025

View reviewed changes

Leo-Besancon and others added 3 commits November 12, 2025 13:04

fix: Use slash for unix path in scripts/generate_all_e2e_tests.sh

94df522

Co-authored-by: Copilot <[email protected]>

fix: Change all the backslash..

b4c3cda

chore: remove unneeded print

d09e0ff

Al-Kindi-0 approved these changes Nov 25, 2025

View reviewed changes

Soulthym and others added 7 commits November 27, 2025 15:56

tweak: Only use extension field of degree 2 for testing

5f2f351

docs: comment num_beta_challenges

06cd71e

refactor: rename num_beta_challenges -> max_beta_challenge_power

871058c

docs: comment tweak + added TODOs

7f02a2a

refactor: avoid leaking extension field degree

1d14475

chores: make lint

bdcd45d

Use Plonky3 super trait (#515)

a2bcabd

Al-Kindi-0 merged commit b3a441e into add_plonky3_backend Dec 4, 2025
9 checks passed

Al-Kindi-0 deleted the target_plonky3_fork branch December 4, 2025 15:10



	// Intentionally insert the same value `a` at position 3 under both `s3` and `s4` conditions.
	// This ensures that the bus `q` receives `a` regardless of which condition is met.

		/// Used in conjunction with [`check_constraints`] to simulate
		/// an execution trace and verify that the AIR logic enforces all constraints.

Target 0xMiden Plonky3 repo and use AirScriptAir and AirScriptBuilder traits #508

Target 0xMiden Plonky3 repo and use AirScriptAir and AirScriptBuilder traits #508

Uh oh!

Conversation

Leo-Besancon commented Nov 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Describe your changes

Checklist before requesting a review

Uh oh!

Leo-Besancon commented Nov 10, 2025

Uh oh!

adr1anh left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Al-Kindi-0 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

adr1anh commented Nov 11, 2025

Uh oh!

Al-Kindi-0 commented Nov 11, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Uh oh!

Copilot AI Nov 12, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 12, 2025

Choose a reason for hiding this comment

Uh oh!

Leo-Besancon commented Nov 13, 2025

Uh oh!

Al-Kindi-0 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Leo-Besancon commented Nov 5, 2025 •

edited

Loading