[fcostaoliveira]

The serial runner puts every test on a multiprocessing.Queue and reads them back with get(timeout=0.1), breaking on Empty. put() is flushed by a background feeder thread. If that thread is starved for more than 100 ms on a loaded machine, the runner runs no tests, prints Total Tests Run: 0 and exits 0, which is a silent pass. This was introduced in #148.

Fix: the serial path now iterates the test list directly. In the parallel path, workers do a blocking get() and stop on one None sentinel per worker, instead of relying on a timeout.

Repro: q = Queue(); q.put(1, block=False); q.get(timeout=0.1) raised Empty in 2 of 30 runs at nice 19 under 28 busy loops, and in 0 of 200 idle. Under the same load, RLTest once reported Total Tests Run: 0 with exit 0. The new tests/unit/test_serial_runner_queue.py fails on master (no tests run) and passes with this change.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes core test scheduling for serial and parallel modes; incorrect sentinel or iteration logic could skip or hang tests, though behavior is covered by new unit tests.

Overview
Fixes a silent false pass where serial RLTest runs could report Total Tests Run: 0 and exit 0 when a multiprocessing.Queue feeder thread was slow to flush.

Serial execution no longer round-trips tests through a queue with get(timeout=0.1) and an early break on Empty; it materializes list(self.loader) and runs tests by index on the main thread.

Parallel workers use blocking get() and exit on a per-worker None sentinel (tests + [None] * parallelism) instead of timing out on an apparently empty queue before the feeder has flushed.

Adds tests/unit/test_serial_runner_queue.py to lock in serial behavior under a mocked starved queue and to assert all 20 tests run under parallelism 3.

^{Reviewed by Cursor Bugbot for commit 687bee2. Bugbot is set up for automated code reviews on this repo. Configure here.}