added LaunchPool

docs/LaunchPool.md

@@ -0,0 +1,110 @@
+# LaunchPool
+
+`LaunchPool` is a bounded-concurrency task pool: you submit lambdas with `launch`, and the pool runs them using a fixed number of worker coroutines.
+
+## Constructor
+
+```
+LaunchPool(maxWorkers, maxQueueSize = Channel.UNLIMITED)
+```
+
+| Parameter | Description |
+|-----------|-------------|
+| `maxWorkers` | Maximum number of tasks that run in parallel. |
+| `maxQueueSize` | Maximum number of tasks that may wait in the queue. When the queue is full, `launch` suspends the caller until space becomes available. Defaults to `Channel.UNLIMITED` (no bound). |
+
+## Methods
+
+### `launch(lambda): Deferred`
+
+Schedules `lambda` for execution and returns a `Deferred` for its result.
+
+- Suspends if the queue is full (`maxQueueSize` reached).
+- Throws `IllegalStateException` if the pool is already closed or cancelled.
+- Any exception thrown by `lambda` is captured in the returned `Deferred` and **does not escape the pool**.
+
+```lyng
+val pool = LaunchPool(4)
+val d1 = pool.launch { computeSomething() }
+val d2 = pool.launch { computeOther() }
+pool.closeAndJoin()
+println(d1.await())
+println(d2.await())
+```
+
+### `closeAndJoin()`
+
+Stops accepting new tasks and suspends until all queued and running tasks complete normally. After this call, any further `launch` throws `IllegalStateException`. Idempotent — safe to call multiple times.
+
+### `cancel()`
+
+Immediately closes the queue and cancels all worker coroutines. Queued but unstarted tasks are discarded. After this call, `launch` throws `IllegalStateException`. Idempotent.
+
+### `cancelAndJoin()`
+
+Like `cancel()`, but also suspends until all worker coroutines have stopped. Useful when you need to be sure no worker code is still running before proceeding. Idempotent.
+
+## Exception handling
+
+Exceptions from submitted lambdas are captured per-task in the returned `Deferred`. The pool itself continues running after a task failure:
+
+```lyng
+val pool = LaunchPool(2)
+val good = pool.launch { 42 }
+val bad  = pool.launch { throw IllegalArgumentException("boom") }
+pool.closeAndJoin()
+
+assertEquals(42, good.await())
+assertThrows(IllegalArgumentException) { bad.await() }
+```
+
+## Bounded queue / back-pressure
+
+When `maxQueueSize` is set, the producer suspends if the queue fills up, providing automatic back-pressure:
+
+```lyng
+// 1 worker, queue of 2 — producer can be at most 2 tasks ahead of what's running
+val pool = LaunchPool(1, 2)
+val d1 = pool.launch { delay(10); "a" }
+val d2 = pool.launch { delay(10); "b" }
+val d3 = pool.launch { delay(10); "c" }   // suspends until d1 is picked up by the worker
+pool.closeAndJoin()
+```
+
+## Collecting all results
+
+`launch` returns a `Deferred`, so you can collect results via `map`:
+
+```lyng
+val pool = LaunchPool(4)
+val jobs = (1..10).map { n -> pool.launch { n * n } }
+pool.closeAndJoin()
+val results = jobs.map { (it as Deferred).await() }
+// results == [1, 4, 9, 16, 25, 36, 49, 64, 81, 100]
+```
+
+## Concurrency limit in practice
+
+With `maxWorkers = 2`, at most 2 tasks run simultaneously regardless of how many are queued:
+
+```lyng
+val mu = Mutex()
+var active = 0
+var maxSeen = 0
+
+val pool = LaunchPool(2)
+(1..8).map {
+    pool.launch {
+        mu.withLock { active++; if (active > maxSeen) maxSeen = active }
+        delay(5)
+        mu.withLock { active-- }
+    }
+}
+pool.closeAndJoin()
+assert(maxSeen <= 2)
+```
+
+## See also
+
+- [parallelism.md](parallelism.md) — `launch`, `Deferred`, `Mutex`, `Channel`, and coroutine basics
+- [Channel.md](Channel.md) — the underlying channel primitive used by `LaunchPool`

docs/parallelism.md

@@ -257,6 +257,27 @@ Channels can also be buffered so the producer can run ahead:
 
 For the full API — including `tryReceive`, `Channel.UNLIMITED`, and the fan-out / ping-pong patterns — see the [Channel] reference page.
 
+## LaunchPool
+
+When you need **bounded concurrency** — run at most *N* tasks at the same time without spawning a new coroutine per task — use [LaunchPool]:
+
+```lyng
+val pool = LaunchPool(4)          // at most 4 tasks run in parallel
+
+val jobs = (1..20).map { n ->
+    pool.launch { expensiveCompute(n) }
+}
+pool.closeAndJoin()               // wait for all tasks to complete
+
+val results = jobs.map { (it as Deferred).await() }
+```
+
+Exceptions thrown inside a submitted lambda are captured in the returned `Deferred` and do not crash the pool, so other tasks continue running normally.
+
+See [LaunchPool] for the full API including bounded queues and cancellation.
+
+[LaunchPool]: LaunchPool.md
+
 | | Flow | Channel |
 |---|---|---|
 | **temperature** | cold (lazy) | hot (eager) |