[byRandom]

Proposal #5 from the architecture review: one work pool for the app's background threads.

Before

• TileCache and StructureCache each had their own copy of the same machinery:
  • a Mutex+Condvar queue of pending keys and a set of keys in progress;
  • workers that take the most urgent key and keep per-seed state;
  • the result sent before the key leaves the in-progress set;
  • a call to egui's request_repaint.
• Spawn and Strongholds each had their own copy of a one-shot background computation.
• All of it took an egui::Context, so none of it could be tested.

After

New crates/chunkscope-app/src/work.rs:

• WorkPool<K, V>:
  • new(name, workers, wake, compute(&mut state, key));
  • request(wanted, done), which returns the number of keys still to come;
  • receive().
• Job<T>: spawn(name, wake, compute), is_loading(), get().
• The wake is a plain function; the app passes request_repaint.
• Dropping a pool now also stops its idle workers.

TileCache and StructureCache keep their caches and become thin users of the pool. Spawn and Strongholds wrap a Job. Net change in the existing files: 99 lines added, 261 removed, not counting the new module.

Tests

• Seven new unit tests, with no egui involved:
  • the most urgent key is computed first;
  • a key in progress is not queued again;
  • a new request drops stale work, even one of the same length;
  • keys already done are skipped;
  • each worker keeps its state;
  • the wake fires after each result;
  • idle workers stop when the pool is dropped;
  • a job delivers its value once.
• The app's 11 tests passed in 20 runs in a row.
• Mutation testing: every rule was broken on purpose and a test failed for each.
• The app shows the same map and markers as before (checked under a virtual display).

No behaviour change, so no release.

---

Generated by Claude Code