Fix non-contiguous circuit input diagnostics

[zakazaka95]

Summary

• detect malformed core::circuit::Circuit types while they still have a source location
• report the first missing transitive input index at each malformed use instead of reaching Sierra specialization
• cover gaps, non-zero starts, output subsets, and valid reuse of one input

---

Type of change

• Bug fix (fixes incorrect behavior)
• New feature
• Performance improvement
• Documentation change with concrete technical impact
• Style, wording, formatting, or typo-only change

---

Why is this change needed?

Sierra circuit specialization assumes that the input indices reachable from a circuit's outputs are exactly 0..N. A gap currently reaches an infallible specialization path and crashes the compiler instead of producing an actionable source diagnostic.

---

What was the behavior or documentation before?

Circuits using inputs such as {0, 2}, starting at a non-zero index, or exposing only input 1 as an output caused an internal compiler panic.

---

What is the behavior or documentation after?

Semantic analysis emits E2204 at the circuit expression, identifying the expected and actual index. Reusing a contiguous input remains valid.

---

Related issue or discussion (if any)

Closes #10138.

---

Additional context

Latest helper refactor (636760d): the focused circuit regression, strict all-target/all-feature Clippy, workspace formatting and diff checks pass.

Earlier implementation validation (September 3):

• focused expr_diagnostics::circuit regression: 1 passed
• cargo test -p cairo-lang-semantic: 111 passed
• strict all-target/all-feature Clippy for cairo-lang-semantic
• nightly Rust formatting check and git diff --check

[cursor]

PR Summary

Low Risk
Scoped semantic validation and diagnostics for circuit types; no changes to auth, runtime, or Sierra codegen paths beyond failing earlier with a source error.

Overview
Fixes compiler internal panics when circuit code uses malformed CircuitInput index sets by validating during semantic analysis instead of at Sierra specialization.

add_value_type_based_diagnostics now walks nested types for core::circuit::Circuit, collects CircuitInput indices reachable from the circuit’s output type, and reports E2204 when that set is not exactly 0..n (gaps, starting above zero, or outputs that only use a non-zero input). Reusing a single contiguous input stays allowed.

Adds ModuleHelper::extern_type_id, wires the new CircuitInputIndicesNotContiguous diagnostic kind, and extends expr diagnostic tests under circuit for gap, non-zero start, output-subset, and valid reuse cases.

^{Reviewed by Cursor Bugbot for commit 636760d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[reviewable-StarkWare]

This change is 

[orizi]

@orizi made 7 comments.
Reviewable status: 0 of 4 files reviewed, 7 unresolved discussions (waiting on zakazaka95).

---

crates/cairo-lang-semantic/src/diagnostic.rs line 75 at r1 (raw file):

    /// Reports a non-contiguous circuit input set once per concrete circuit type.
    pub(crate) fn report_circuit_input_indices(

no reason for this to exist as a function - write it inline at callsite.

---

crates/cairo-lang-semantic/src/types.rs line 966 at r1 (raw file):

}

/// The first gap in a circuit's sorted input indices.

move this section to after the array violation section for a more consistent ordering.

---

crates/cairo-lang-semantic/src/types.rs line 984 at r1 (raw file):

    ty: TypeId<'db>,
) -> Option<CircuitInputIndexViolation<'db>> {
    let circuit_module = core_submodule(db, SmolStrId::from(db, "circuit"));

this would require adding extern_type_id at the helper - but this still is more sensible.

Suggestion:

    let circuit_module = ModuleHelper::core(db).submodule("circuit").extern_type_id("Circuit");
    let circuit_extern_id = circuit_module.extern_type_id("Circuit");
    let circuit_input_extern_id = circuit_module.extern_type_id("CircuitInput");

---

crates/cairo-lang-semantic/src/types.rs line 1006 at r1 (raw file):

                    GenericArgumentId::Type(ty) => Some(*ty),
                    _ => None,
                }));

Suggestion:

                let ConcreteExternTypeLongId { extern_type_id, generic_args } = extrn.long(db);
                if extern_type_id != circuit_extern_id {
                        stack.extend(generic_args.iter().filter_map(|arg| match arg {
                        GenericArgumentId::Type(ty) => Some(*ty),
                        _ => None,
                    }));
                } else if let [GenericArgumentId::Type(outputs)] = generic_args[..]
                    && let Some((expected, actual)) =
                        non_contiguous_circuit_input(db, outputs, circuit_module)
                {
                    return Some(CircuitInputIndexViolation { circuit: ty, expected, actual });
                }

---

crates/cairo-lang-semantic/src/types.rs line 1033 at r1 (raw file):

    db: &'db dyn Database,
    outputs: TypeId<'db>,
    circuit_module: ModuleId<'db>,

Suggestion:

    circuit_input_extern_id: ModuleId<'db>,

---

crates/cairo-lang-semantic/src/types.rs line 1048 at r1 (raw file):

                if extern_type_id.parent_module(db) == circuit_module
                    && extern_type_id.name(db).long(db).as_str() == "CircuitInput"
                {

Suggestion:

                let ConcreteExternTypeLongId { extern_type_id, generic_args } = extrn.long(db);
                if extern_type_id == circuit_input_extern_id {

---

crates/cairo-lang-semantic/src/types.rs line 1054 at r1 (raw file):

                    {
                        input_indices.insert(index);
                    }

so that we would also catch very large values if relevant.

Suggestion:

                    if let [GenericArgumentId::Constant(index)] = generic_args[..]
                        && let Some(index) = index.to_int(db)
                        && let Some(index) = index.to_usize()
                    {
                        input_indices.insert(index);
                    }

[zakazaka95]

Addressed all seven review points in 239afb2: reporting is inlined, circuit validation now follows array validation, ModuleHelper::extern_type_id resolves the exact circuit extern types, traversal receives the CircuitInput ID directly, and constants use to_int(db).to_usize(). The focused regression and full cairo-lang-semantic suite pass (111/111), and strict clippy passes.

[orizi]

@orizi reviewed 2 files and all commit messages, made 2 comments, and resolved 6 discussions.
Reviewable status: 2 of 5 files reviewed, 3 unresolved discussions (waiting on zakazaka95).

---

crates/cairo-lang-semantic/src/diagnostic.rs line 46 at r2 (raw file):

    builder: DiagnosticsBuilder<'db, SemanticDiagnostic<'db>>,
    context_module: ModuleId<'db>,
    pub(crate) diagnosed_circuits: OrderedHashSet<semantic::TypeId<'db>>,

this shouldn't appear here.
if we have the same problematic type twice, we can report it twice.

Code quote:

    pub(crate) diagnosed_circuits: OrderedHashSet<semantic::TypeId<'db>>,

---

crates/cairo-lang-semantic/src/types.rs line 962 at r2 (raw file):

    } else if let Some(CircuitInputIndexViolation { circuit, expected, actual }) =
        circuit_input_index_violation(db, ty)
        && diagnostics.diagnosed_circuits.insert(circuit)

this shouldn't be required.

Code quote:

diagnostics.diagnosed_circuits.insert(circuit)

[orizi]

reply to reviews on https://reviewable.io/reviews/starkware-libs/cairo/10366

@orizi made 1 comment.
Reviewable status: 2 of 5 files reviewed, 3 unresolved discussions (waiting on zakazaka95).

[zakazaka95]

Acknowledged. Addressed the latest review in f3b083d. I removed the cross-use deduplication so repeated malformed uses can each report. Focused regression 1/1 and full cairo-lang-semantic 111/111 pass, along with doctests, root formatting, strict scoped Clippy, and diff checks.

@zakazaka95 made 4 comments.
Reviewable status: 2 of 5 files reviewed, 3 unresolved discussions (waiting on orizi).

---

crates/cairo-lang-semantic/src/diagnostic.rs line 75 at r1 (raw file):

Previously, orizi wrote…

no reason for this to exist as a function - write it inline at callsite.

Done. Inlined at the callsite in 239afb2; the standalone helper no longer exists.

---

crates/cairo-lang-semantic/src/diagnostic.rs line 46 at r2 (raw file):

Previously, orizi wrote…

this shouldn't appear here.
if we have the same problematic type twice, we can report it twice.

Done. Removed in f3b083d. SemanticDiagnostics no longer keeps diagnosed_circuits state, so the same problematic type can be reported at each use.

---

crates/cairo-lang-semantic/src/types.rs line 962 at r2 (raw file):

Previously, orizi wrote…

this shouldn't be required.

Done. Removed in f3b083d. The diagnosed_circuits insertion guard and the now-unused carried circuit value are gone.

[orizi]

@orizi reviewed 2 files and all commit messages, made 1 comment, and resolved 3 discussions.
Reviewable status: 4 of 5 files reviewed, 1 unresolved discussion (waiting on zakazaka95).

---

crates/cairo-lang-semantic/src/types.rs line 1132 at r3 (raw file):

                        GenericArgumentId::Type(ty) => Some(*ty),
                        _ => None,
                    }));

Suggestion:

                    stack.extend(
                        generic_args.iter().filter_map(|arg| try_extract_matches!(arg, GenericArgumentId::Type))
                    );

[zakazaka95]

@zakazaka95 made 1 comment.
Reviewable status: 4 of 5 files reviewed, 1 unresolved discussion (waiting on orizi).

---

crates/cairo-lang-semantic/src/types.rs line 1132 at r3 (raw file):

                        GenericArgumentId::Type(ty) => Some(*ty),
                        _ => None,
                    }));

Done in 636760d. The circuit regression passes, along with strict scoped Clippy and workspace formatting.

[orizi]

@orizi reviewed 1 file and all commit messages, made 2 comments, and resolved 1 discussion.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on zakazaka95).

---

a discussion (no related file):
@TomerStarkware @eytan-starkware for 2nd eye.

[eytan-starkware]

@eytan-starkware+AGNT made 2 comments.
Reviewable status: all files reviewed, 2 unresolved discussions (waiting on zakazaka95).

---

crates/cairo-lang-semantic/src/types.rs line 1188 at r4 (raw file):

                    continue;
                }
                stack.extend(generic_args.iter().filter_map(|arg| match arg {

Same extraction as in circuit_input_index_violation above (already switched to the macro in r4) - use it here too for consistency.

Suggestion:

                stack.extend(
                    generic_args.iter().filter_map(|arg| try_extract_matches!(arg, GenericArgumentId::Type)),
                );

---

crates/cairo-lang-semantic/src/types.rs line 1194 at r4 (raw file):

            }
            TypeLongId::Tuple(types) => stack.extend(types.iter().copied()),
            _ => {}

A Missing type here means a diagnostic was already reported and the input set is unknown, so the contiguity result below is a cascade of that error. E.g. with a typo in one input type:

use core::circuit::{CircuitElement, CircuitInput, CircuitInputs, circuit_add};
fn main() {
    let a = CircuitElement::<CircuitInput<0>> {};
    let b = CircuitElement::<CircuitInpt<1>> {}; // typo
    let c = CircuitElement::<CircuitInput<2>> {};
    let sum = circuit_add(a, b);
    let _acc = (sum, c).new_inputs();
}

this PR reports, on top of the E0006: Type not found at the typo:

error[E2204]: Circuit input indices must be contiguous and start at 0. Expected index 1, found 2.
 --> lib.cairo:7:16
    let _acc = (sum, c).new_inputs();
               ^^^^^^^^^^^^^^^^^^^^^

Fixing the typo makes it disappear, and it points at the wrong place. Bail out on Missing, as type_size_info does:

Suggestion:

            TypeLongId::Tuple(types) => stack.extend(types.iter().copied()),
            TypeLongId::Missing(_) => return None,
            _ => {}

Please also add the snippet above as a case in test_data/circuit, expecting only the E0006.