Context
Scheduled mutation run 29495843050 (2026-07-16, first green run after the baseline fix in #140) exercised 22 mutants across crates/comenqd/src/listener.rs, crates/comenqd/src/logging.rs, and crates/comenqd/src/worker.rs: 4 caught, 3 missed, 4 timed out, 11 unviable.
Two of the three missed mutants are negations of the same predicate in WorkerHooks::notify_drained_if_empty:
crates/comenqd/src/worker.rs:94:25: delete ! in WorkerHooks::notify_drained_if_empty
crates/comenqd/src/worker.rs:96:26: delete ! in WorkerHooks::notify_drained_if_empty
#[cfg(any(test, feature = "test-support"))]
fn notify_drained_if_empty(&self, queue_path: &Path) -> std::io::Result<()> {
if let Some(n) = &self.drained {
// Ignore sentinel files left by the queue implementation and
// consider the directory empty when no other files remain.
let empty = !stdfs::read_dir(queue_path)?
.filter_map(Result::ok)
.any(|e| !is_metadata_file(e.file_name()));
if empty {
n.notify_one();
}
}
...
}Analysis
This helper is cfg-gated to test and test-support builds, and exists specifically so integration tests can wait for the queue to drain. Because it is test synchronisation machinery, an inverted empty calculation would not fail functionally — it would either notify on every poll (masking a real drain) or never notify (causing waiters to hang or fall back to a timeout), which is exactly the kind of failure that manifests as flaky or slow tests rather than a clean assertion failure. That is consistent with it going undetected by the current suite: nothing exercises the predicate directly, only indirectly through tests that use the notification as one of several synchronisation signals.
Proposed next step
Add a direct unit test for notify_drained_if_empty that covers: an empty directory (should notify), a directory containing only sentinel/metadata files (should notify), and a directory containing at least one non-metadata file (should not notify). This directly exercises the !...any(...) predicate and its negation, independent of the higher-level integration tests that currently rely on it only incidentally.
Context
Scheduled mutation run 29495843050 (2026-07-16, first green run after the baseline fix in #140) exercised 22 mutants across
crates/comenqd/src/listener.rs,crates/comenqd/src/logging.rs, andcrates/comenqd/src/worker.rs: 4 caught, 3 missed, 4 timed out, 11 unviable.Two of the three missed mutants are negations of the same predicate in
WorkerHooks::notify_drained_if_empty:Analysis
This helper is
cfg-gated to test andtest-supportbuilds, and exists specifically so integration tests can wait for the queue to drain. Because it is test synchronisation machinery, an invertedemptycalculation would not fail functionally — it would either notify on every poll (masking a real drain) or never notify (causing waiters to hang or fall back to a timeout), which is exactly the kind of failure that manifests as flaky or slow tests rather than a clean assertion failure. That is consistent with it going undetected by the current suite: nothing exercises the predicate directly, only indirectly through tests that use the notification as one of several synchronisation signals.Proposed next step
Add a direct unit test for
notify_drained_if_emptythat covers: an empty directory (should notify), a directory containing only sentinel/metadata files (should notify), and a directory containing at least one non-metadata file (should not notify). This directly exercises the!...any(...)predicate and its negation, independent of the higher-level integration tests that currently rely on it only incidentally.