Warnings (product) fails on every PR since #7250 — two denied warnings in gc/barrier_arming

[proggeramlug]

The Warnings (product) job in Tests builds with warnings denied. Since #7250 landed it fails on two warnings from the new gc/barrier_arming module, so every PR opened after that commit inherits a red Warnings (product) regardless of its content.

From the job log (Warnings (product), run 30762914997):

error: glob import doesn't reexport anything with visibility `pub(crate)`
       because no imported item is public enough
70 | pub(crate) use barrier_arming::*;
   = help: reduce the glob import's visibility or increase visibility of imported items

error: function `remembered_reconstruct_census` is never used
   --> crates/perry-runtime/src/gc/barrier_arming.rs:118:15

error: could not compile `perry-runtime` (lib) due to 2 previous errors

Both reproduce locally on a plain cargo build --profile perry-dev -p perry -p perry-runtime-static -p perry-stdlib-static (as a warning rather than an error, since that build does not deny).

Why it was not caught

main's last Tests run was c9cd73ba5 at 2026-08-02 07:41, and barrier_arming.rs landed in c7893c4ac at 2026-08-02 12:58 — about five hours later. git merge-base --is-ancestor c7893c4ac c9cd73ba5 is false, so no main run has ever compiled this module under the warnings gate. The most recent main Tests run showing Warnings (product): success predates the code entirely.

This is the "a gate can be unable to fail" family from CLAUDE.md, in its fourth form: the job is genuinely running and genuinely green on main, but its subject never reached it.

Fix sketch

1. gc/mod.rs:70 — either give barrier_arming's intended exports pub(crate) visibility, or drop the pub(crate) use barrier_arming::*; glob if nothing is meant to be re-exported.
2. gc/barrier_arming.rs:118 — remembered_reconstruct_census is dead. Either wire it into the diagnostic path it was written for, gate it behind the relevant cfg, or delete it. Given perf(gc): stop maintaining a remembered set nothing has read yet (#7187 Phase A) #7250's lazily-armed remembered set, a census helper that nothing calls is worth a second look before it is simply silenced — an #[allow(dead_code)] would hide the question rather than answer it.

Notes

• Found while clearing the two lint gates in fix(lint): clear the GC store-site and address-classification gates (#7258, #7259) #7273; both files are untouched by that PR, which is why it is filed rather than fixed there (and gc/ is live territory for gc/perf: classify_heap_generation is 19% of batch.ts (57.4% total GC bookkeeping) with ZERO collections running — write-barrier tower needs its own lever (#5094 evidence) #7187 / gc-matrix: four of the six PR-gating arms run ZERO copying minors corpus-wide (default/verify_evac/cons_scan_off at copy-minor 0/49; header still claims 12/22) #7255).
• Not the same as the gc-ratchet red on the same PR — that is gc-matrix: four of the six PR-gating arms run ZERO copying minors corpus-wide (default/verify_evac/cons_scan_off at copy-minor 0/49; header still claims 12/22) #7255 (zero copying minors corpus-wide), also pre-existing on main for at least six consecutive runs.