Problem:
cargo test --locked --offline does not compile on the promoted da51a48 source. The benchmark binaries still build, but the repo's #[cfg(test)] target is stale.
Evidence:
On a clean worktree at origin/main / da51a48:
cargo fetch --locked
cargo test --locked --offline
The test build fails before running tests. The first errors are unresolved direct-centered symbols in src/point_add/mod.rs, for example:
error[E0425]: cannot find value `DIRECT_CENTERED_LOW_BRANCH_META_BITS` in this scope
--> src/point_add/mod.rs:3967:31
error[E0425]: cannot find value `DIRECT_CENTERED_BRANCH_SIDECAR_COMPONENT_SCRATCH_BITS` in this scope
--> src/point_add/mod.rs:3982:33
The same build also hits stale simulator API calls in test code, for example:
error[E0599]: no method named `apply` found for struct `sim::Simulator` in the current scope
--> src/point_add/venting.rs:1539:17
help: there is a method `apply_iter` with a similar name
The benchmark build path is separate and still compiles:
cargo build --release --locked --offline --bin build_circuit --bin eval_circuit
Effect:
Solver-side regression checks cannot use cargo test, so stale research helpers and old API assumptions can accumulate unnoticed. It also makes it harder for contestants to distinguish a benchmark failure from a dead test harness.
Smallest useful fix:
Either restore/update the direct-centered helper constants/functions and replace old Simulator::apply / global_phase test calls with the current API, or gate/remove the stale test modules until they are maintained again.
Gate:
cargo test --locked --offline compiles on origin/main; if the full suite is intentionally too slow, at least cargo test --locked --offline --no-run should pass.
Problem:
cargo test --locked --offlinedoes not compile on the promotedda51a48source. The benchmark binaries still build, but the repo's#[cfg(test)]target is stale.Evidence:
On a clean worktree at
origin/main/da51a48:cargo fetch --locked cargo test --locked --offlineThe test build fails before running tests. The first errors are unresolved direct-centered symbols in
src/point_add/mod.rs, for example:The same build also hits stale simulator API calls in test code, for example:
The benchmark build path is separate and still compiles:
Effect:
Solver-side regression checks cannot use
cargo test, so stale research helpers and old API assumptions can accumulate unnoticed. It also makes it harder for contestants to distinguish a benchmark failure from a dead test harness.Smallest useful fix:
Either restore/update the direct-centered helper constants/functions and replace old
Simulator::apply/global_phasetest calls with the current API, or gate/remove the stale test modules until they are maintained again.Gate:
cargo test --locked --offlinecompiles onorigin/main; if the full suite is intentionally too slow, at leastcargo test --locked --offline --no-runshould pass.