fix: ultra num public inputs derivation #19950

iakovenkos · 2026-01-26T19:31:41Z

Don't use get_value() on vk member, derive from proof size, make sure it matches the value from the vk
Added new bb::ProofLength gadget responsible for all Chonk and Honk proof/public inputs lengths computations

iakovenkos · 2026-01-27T11:04:45Z

barretenberg/cpp/src/barretenberg/chonk/chonk_proof.cpp

    // ECCVM proof
    start_idx = end_idx;
-    end_idx += static_cast<std::ptrdiff_t>(ECCVMFlavor::PROOF_LENGTH_WITHOUT_PUB_INPUTS);
+    end_idx += static_cast<std::ptrdiff_t>(ECCVMFlavor::PROOF_LENGTH);


It was somewhat misleading to call those w/o pub inputs in VMs

iakovenkos · 2026-01-27T11:07:27Z

barretenberg/cpp/src/barretenberg/dsl/acir_format/mock_verifier_inputs.cpp

-    // Populate mock witness accumulator commitments
-    populate_field_elements_for_mock_commitments(proof, Flavor::NUM_WITNESS_ENTITIES / 2);
+    // Populate mock accumulator commitments (non_shifted + shifted)
+    populate_field_elements_for_mock_commitments(proof, Flavor::NUM_ACCUMULATOR_COMMITMENTS);


Realized that MLB Flavor is quite confusing due to copy-pasted structures. Simplified those as a byproducr

iakovenkos · 2026-01-27T11:08:36Z

barretenberg/cpp/src/barretenberg/dsl/acir_format/mock_verifier_inputs.test.cpp

-
-TYPED_TEST_SUITE(MockVerifierInputsTest, FlavorTypes);
+// Static assertions for proof length constants
+static_assert(MERGE_PROOF_SIZE == 42, "Merge proof size changed");


Should be faster to modify now

iakovenkos · 2026-01-27T11:10:39Z

barretenberg/cpp/src/barretenberg/flavor/multilinear_batching_flavor.hpp

-        DEFINE_FLAVOR_MEMBERS(DataType,
-                              batched_unshifted_accumulator, // column 0: batched unshifted poly for accumulator
-                              batched_unshifted_instance,    // column 1: batched unshifted poly for instance
-                              eq_accumulator,                // column 2: eq(u, r_acc) - selects accumulator eval point


This class was essentially unused, + eq's are not witnesses.

iakovenkos · 2026-01-27T11:17:40Z

barretenberg/cpp/src/barretenberg/honk/proof_length.hpp

@@ -0,0 +1,181 @@
+// === AUDIT STATUS ===


So, my goal was to always derive num_public_inputs from the proof size. But the methods used to compute the "fixed" sizes were flavor-linked. Instead I wanted to have something context-dependent like HypernovaInstanceToAccum length that still re-uses lots from Honk proof building blocks. As a result, we have all honk/chonk proof length computations in one place + it's easy to deduce num_public inputs

iakovenkos · 2026-01-27T16:37:56Z

barretenberg/cpp/src/barretenberg/ultra_honk/oink_verifier.cpp

        verifier_instance->vk_and_hash->hash.assert_equal(vk_hash);
+
+        // Assert that the provided num_public_inputs matches VK's value (in-circuit constraint)
+        vk->num_public_inputs.assert_equal(FF(num_public_inputs), "OinkVerifier: num_public_inputs mismatch with VK");


note that previously we would use .get_value(). though substituting the real witness value with something else would have created a different circuit, feels more robust when such discrepancy immediately leads to a failed constraint

Interesting, so the idea is that leads to a better error in the circuit failure case?

I was concerned about vk->num_public_inputs.get_value() in general, as then we would need to analyze whether a malicious actor can ignore this value + the binding of the vk field with the proof was very implicit, namely through the loop for receiving the public inputs that uses the underlying native value, changing which would be enough to change this circuit's VK and cause a verification failure against a correctly computed VK, but now it's obvious that this loop uses the same value that's in the vk

iakovenkos · 2026-01-27T16:41:10Z

barretenberg/cpp/src/barretenberg/ultra_honk/ultra_verifier.cpp

 */
-template <typename Flavor, class IO>
-typename UltraVerifier_<Flavor, IO>::PaddingData UltraVerifier_<Flavor, IO>::process_padding() const
+template <typename Flavor, class IO> size_t UltraVerifier_<Flavor, IO>::compute_log_n() const


Had to split process_padding method, cause integer log_n is needed before oink, and padding indicator array can only be computed after oink -- vk member log_circuit_size gets hashed and untagged as a free witness

…i/pub-inputs-clean-up

ledwards2225

LGTM - thanks for the improvements

AztecBot · 2026-01-27T19:42:10Z

Flakey Tests

🤖 says: This CI run detected 4 tests that failed, but were tolerated due to a .test_patterns.yml entry.

\033FLAKED\033 (8;;http://ci.aztec-labs.com/47bbe7f15962c754�47bbe7f15962c7548;;�):  yarn-project/end-to-end/scripts/run_test.sh simple src/e2e_epochs/epochs_first_slot.test.ts (230s) (code: 1) group:e2e-p2p-epoch-flakes (\033iakovenkos\033: upd vks hash)
\033FLAKED\033 (8;;http://ci.aztec-labs.com/48187d2442d1e742�48187d2442d1e7428;;�):  yarn-project/end-to-end/scripts/run_test.sh simple src/e2e_epochs/epochs_high_tps_block_building.test.ts (279s) (code: 1) group:e2e-p2p-epoch-flakes (\033iakovenkos\033: upd vks hash)
\033FLAKED\033 (8;;http://ci.aztec-labs.com/d224bff8670c857e�d224bff8670c857e8;;�):  yarn-project/end-to-end/scripts/run_test.sh simple src/e2e_p2p/reqresp.test.ts (368s) (code: 1) group:e2e-p2p-epoch-flakes (\033iakovenkos\033: upd vks hash)
\033FLAKED\033 (8;;http://ci.aztec-labs.com/dc8d8eac9e38bcf9�dc8d8eac9e38bcf98;;�):  yarn-project/end-to-end/scripts/run_test.sh simple src/e2e_p2p/valid_epoch_pruned_slash.test.ts (404s) (code: 1) group:e2e-p2p-epoch-flakes (\033iakovenkos\033: upd vks hash)

BEGIN_COMMIT_OVERRIDE chore: use powers of gamma in logup (#19881) fix: ultra num public inputs derivation (#19950) END_COMMIT_OVERRIDE

iakovenkos added 6 commits January 26, 2026 19:15

don't use get_value to compute_num_public inputs

a6cc51b

mlb flavor -> num entities clean up

507eb70

deduplicate honk flavor proof size derivation

c0d1a24

rename

eaf2acc

simplify rollup proof splitting

b08fb76

move decl to class level

4bf13e6

iakovenkos marked this pull request as ready for review January 27, 2026 09:48

clean up

197a250

iakovenkos commented Jan 27, 2026

View reviewed changes

iakovenkos added 2 commits January 27, 2026 12:23

ipa again

6022a56

fix tests

9d1005d

iakovenkos commented Jan 27, 2026

View reviewed changes

iakovenkos added 2 commits January 27, 2026 16:43

more clean up

efee288

Merge remote-tracking branch 'origin/merge-train/barretenberg' into s…

f1742ac

…i/pub-inputs-clean-up

iakovenkos self-assigned this Jan 27, 2026

ledwards2225 approved these changes Jan 27, 2026

View reviewed changes

upd vk hashes

2adff9f

iakovenkos added the ci-full Run all master checks. label Jan 27, 2026

upd vks hash

22b11f5

iakovenkos merged commit 198c4de into merge-train/barretenberg Jan 27, 2026
8 checks passed

iakovenkos deleted the si/pub-inputs-clean-up branch January 27, 2026 20:14

AztecBot mentioned this pull request Jan 27, 2026

feat: merge-train/barretenberg #19973

Merged

github-merge-queue bot pushed a commit that referenced this pull request Jan 28, 2026

feat: merge-train/barretenberg (#19973)

c9a3c06

BEGIN_COMMIT_OVERRIDE chore: use powers of gamma in logup (#19881) fix: ultra num public inputs derivation (#19950) END_COMMIT_OVERRIDE

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: ultra num public inputs derivation #19950

fix: ultra num public inputs derivation #19950

iakovenkos commented Jan 26, 2026 •

edited

Loading

Uh oh!

iakovenkos Jan 27, 2026 •

edited

Loading

Uh oh!

iakovenkos Jan 27, 2026

Uh oh!

iakovenkos Jan 27, 2026

Uh oh!

iakovenkos Jan 27, 2026

Uh oh!

iakovenkos Jan 27, 2026

Uh oh!

iakovenkos Jan 27, 2026

Uh oh!

ledwards2225 Jan 27, 2026

Uh oh!

iakovenkos Jan 27, 2026

Uh oh!

iakovenkos Jan 27, 2026

Uh oh!

ledwards2225 left a comment

Uh oh!

AztecBot commented Jan 27, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

fix: ultra num public inputs derivation #19950

fix: ultra num public inputs derivation #19950

Conversation

iakovenkos commented Jan 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

iakovenkos Jan 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

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!

ledwards2225 left a comment

Choose a reason for hiding this comment

Uh oh!

AztecBot commented Jan 27, 2026

Flakey Tests

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

iakovenkos commented Jan 26, 2026 •

edited

Loading

iakovenkos Jan 27, 2026 •

edited

Loading