fix: add tests

[Copilot]

Review of the journal/WAL design revealed a data race: write_index is bumped via CAS before data is copied into the ring buffer slot, so the WAL worker can read stale data. Additionally, the busy-wait spin loop caused thread starvation on low-core machines, and all existing tests shared a single ./wal/ directory causing parallel test failures.

Data race fix: filled_index

Added a separate filled_index atomic to JournalInner. Writers reserve a slot via write_index CAS, copy data, then advance filled_index in-order with Release ordering. The WAL worker reads filled_index (not write_index) with Acquire to determine safe-to-read entries.

// After writing data to the reserved slot:
while self.inner.filled_index.compare_exchange_weak(
    current_written,
    current_written.wrapping_add(1),
    Ordering::Release,
    Ordering::Relaxed,
).is_err() {
    std::hint::spin_loop();
}

WAL worker changes

• Removed busy-wait spin loop — worker now parks immediately when idle, woken via unpark() on commit
• Drains all available data before parking (removed IO_BATCH_LIMIT cap) — a burst of entries is fully persisted in one flush/sync_all pass
• Added Drop for WalInner — sets shutdown flag and unparks worker to prevent orphaned spinning threads

Test infrastructure

• Added tempfile dev-dep; each test gets an isolated temp directory via test_journal() helper
• All existing tests updated to use isolated dirs and explicit shutdown() for cleanup
• Fixed bogus matches!(err, JournalBusy) assertion (was always true — matched any value into a new binding)

New WAL tests (21 tests, 31 total)

• Segment helpers: file naming, creation, scanning (sorted, non-wal files, invalid names)
• Recovery: empty dir, data replay into ring buffer, respects consumed checkpoint, partial last segment truncation, incomplete non-last segment detection, consumed-ahead-of-written detection, cross-segment recovery
• Runtime: end-to-end persistence verification, segment rotation, shutdown rejects commits, shutdown idempotent, commit future resolves after persistence, persisted_index advancement
• Cleanup: old segment truncation

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.