Observed 2026-07-07 during a local full cargo test --locked --workspace run (on a branch consisting of origin/main 5835ebb plus config-parsing changes that do not touch this code path):
---- api::tests::antirollback_accept_and_publish_survives_dropped_join_handle stdout ----
thread 'api::tests::antirollback_accept_and_publish_survives_dropped_join_handle' panicked at crates/registry-notary-server/src/api.rs:10970:17:
assertion `left == right` failed
left: None
right: Some(2)
Flake evidence:
- Fails intermittently only under the full-workspace parallel run (317/318 passed in the failing run).
- Passes 20/20 consecutive runs in isolation (
cargo test -p registry-notary-server --lib antirollback_accept_and_publish_survives_dropped_join_handle).
- The same workspace content passed a full
cargo test --locked --workspace earlier the same day.
The shape (a dropped-JoinHandle survival test asserting a value that a background task publishes, seeing None instead of Some(2)) points at a scheduling race under load rather than a functional regression. Note this is NOT one of the two tests stabilized in 6c7e980 (2026-07-02), so the "full-workspace flakes are fixed" assumption in the dev-gates guidance has a new counterexample.
Observed 2026-07-07 during a local full
cargo test --locked --workspacerun (on a branch consisting of origin/main 5835ebb plus config-parsing changes that do not touch this code path):Flake evidence:
cargo test -p registry-notary-server --lib antirollback_accept_and_publish_survives_dropped_join_handle).cargo test --locked --workspaceearlier the same day.The shape (a dropped-JoinHandle survival test asserting a value that a background task publishes, seeing
Noneinstead ofSome(2)) points at a scheduling race under load rather than a functional regression. Note this is NOT one of the two tests stabilized in 6c7e980 (2026-07-02), so the "full-workspace flakes are fixed" assumption in the dev-gates guidance has a new counterexample.