From 95a37dd34092fc42bf83ddc1ddad6296f54219a4 Mon Sep 17 00:00:00 2001 From: Laytan Laats Date: Sun, 23 Nov 2025 19:48:03 +0100 Subject: [PATCH] remove terminate from demo This was causing deadlocks because: 1. The main thread would `thread.terminate` a running thread 2. `thread.terminate` does not mean termination happens immediately 3. The thread that was terminated would see that the main thread released `print_mutex` and acquire it 4. The worker would execute `fmt.printf` which is a cancellation point 5. Cancellation point reached, cancelled 6. Deadlock because the thread was holding `print_mutex` You would usually solve this with `pthread_cleanup_push` that would release the mutex in case it is cancelled, or use `pthread_setcancelstate` to disable cancellation while the mutex is held. But the real fix is just not using forced termination and using a flag or other mechanism to indicate to the thread it should stop. `thread.terminate` shouldn't even be a thing IMO. It is way to dangerous and if somebody knows what they are doing they can use the core:sys procedures to achieve it. And we certainly shouldn't be using it in the demo as an example. --- examples/demo/demo.odin | 13 ------------- 1 file changed, 13 deletions(-) diff --git a/examples/demo/demo.odin b/examples/demo/demo.odin index c559d4771..161d48acb 100644 --- a/examples/demo/demo.odin +++ b/examples/demo/demo.odin @@ -1215,19 +1215,6 @@ threading_example :: proc() { } thread.pool_start(&pool) - - { - // Wait a moment before we cancel a thread - time.sleep(5 * time.Millisecond) - - // Allow one thread to print at a time. - for !did_acquire(&print_mutex) { thread.yield() } - - thread.terminate(pool.threads[N - 1], 0) - fmt.println("Canceled last thread") - print_mutex = false - } - thread.pool_finish(&pool) } }