chore: resolve `field_t` issues from external audit by iakovenkos · Pull Request #16210 · AztecProtocol/aztec-packages

iakovenkos · 2025-08-05T10:40:45Z

`field_t` Issues and Resolutions Tracker

For more details, see the comments below.

ID	Title	Status
17	`test_slice_random` not covering the entire input space	`slice` is replaced with a simpler method `split_at`
16	Unexpected random value in `test_slice_random`	`slice` is replaced with a simpler method `split_at`
15	[opt] Duplicate constraint in `invert`	✅
14	`set_public` on constant `field_t` passes `IS_CONSTANT` as a wire index	added the assertion that the input is not constant
13	[opt] Multiplication by zero in `field_t` not canonicalized to constant	❌ (breaks bigfield assertions + no efficiency gains in real circuits)
12	[opt] `decompose_into_bits` with `num_bits = 253` does unnecessary aliasing check	deleted
11	[doc] Incorrect documentation of `slice` method	`slice` is replaced with a simpler method `split_at`
10	Missing builder context validation in field operations	✅
9	[doc] Confusing comment in `accumulate`	✅
8	[opt] `assert_equal` optimization	✅
7	Unnecessary witness multiplied by 0 in `normalize`	✅
6	[opt] `madd` could be optimized when 2 inputs are constant	✅
5	Inconsistent naming of PLONK gate terms	Made the comments consistent with the naming in gate-related structs
4	`bool_t` has a constructor that takes context and value	✅
3	Use `get_value` to get a boolean value	✅
2	No restriction on `num_bits` in `ranged_less_than`	✅
1	[doc] Assumption over `ranged_less_than` is overly restrictive, method comment has mistakes	✅

iakovenkos · 2025-08-05T10:41:46Z

 {
    if (other.is_constant()) {
-        additive_constant = (other.witness_bool ^ other.witness_inverted) ? bb::fr::one() : bb::fr::zero();
+        additive_constant = other.get_value();


issue 3

use get_value() on a bool_t

iakovenkos · 2025-08-05T10:43:40Z

                 true,
                 "Attempting to create a bool_t from a witness_t not satisfying x^2 - x = 0");
-    result.witness_bool = (witness == bb::fr::one());
+    bool_t result(context, witness == bb::fr::one());


issue 4

use an appropriate field_t constructor

iakovenkos · 2025-08-05T10:44:31Z

+}
+
+// Variadic version: compare first with the rest
+template <typename T, typename... Ts> T* validate_context(T* first, Ts*... rest)


issue 10

the contexts are validated for all basic arithmetic ops

iakovenkos · 2025-08-05T11:37:45Z

+    return first;
+}
+
+template <typename T, typename Container> T* validate_context(const Container& elements)


issue 10

the contexts are validated for all entries of an accumulate input

iakovenkos · 2025-08-05T11:38:46Z


        // Let q = (a < b)
-        // Assume both a and b are < K where K = 2^{num_bits} - 1
+        // Assume both a and b are < K where K = 2^{num_bits}


iakovenkos · 2025-08-05T11:40:38Z


    context->create_add_gate({ .a = witness_index,
-                               .b = witness_index,
+                               .b = context->zero_idx,


iakovenkos · 2025-08-05T11:41:33Z


    // Step 2: compute output value
    size_t num_elements = accumulator.size();
-    // If `input` contains non-constant `field_t` elements, add the accumulated constant value to the output value,


issue 9

removed confusing comment

iakovenkos · 2025-08-05T13:03:22Z

-        field_t left = lhs.normalize();
-        field_t right = rhs.normalize();
-        ctx->assert_equal(left.witness_index, right.witness_index, msg);
+        if (lhs.is_normalized() || rhs.is_normalized()) {


issue 8

although the optimization can be applied in other cases, I think it only makes sense when both arguments are non-normalized witnesses, cause in other cases it would only remove a copy constraint when a witness argument is not normalized

iakovenkos · 2025-08-05T14:22:09Z

 template <typename Builder> field_t<Builder> field_t<Builder>::add_two(const field_t& add_b, const field_t& add_c) const
 {
-    if ((add_b.is_constant()) && (add_c.is_constant()) && (is_constant())) {
+    int num_const =


issue 6

optimized add_two: it does not create gates when at least two inputs are constant

iakovenkos · 2025-08-05T14:23:03Z

+    };
+    uint32_t set_public() const
+    {
+        ASSERT(!is_constant());


iakovenkos · 2025-08-06T13:47:38Z

-    //   = a.v * b.v * [a.mul * b.mul] + a.v * [a.mul * b.add] + b.v * [b.mul + a.add] + c.v * [c.mul] +
-    //     [a.add * b.add + c.add]
-    //   = a.v * b.v * [     q_m     ] + a.v * [     q_1     ] + b.v * [     q_2     ] + c.v * [ q_3 ] + [ q_c ]
+    //   = a.v * b.v * [a.mul * b.mul] + a.v * [a.mul * b.add] + b.v * [b.mul + a.add] + c.v * [c.mul]
