[libcxx-commits] [libcxx] [libc++] fix `counting_semaphore` lost wakeups (PR #79265)

[Jan Kokemüller via libcxx-commits]

================
@@ -94,20 +95,25 @@ public:
     auto __old = __a_.fetch_add(__update, memory_order_release);
     _LIBCPP_ASSERT_ARGUMENT_WITHIN_DOMAIN(
         __update <= _LIBCPP_SEMAPHORE_MAX - __old, "update is greater than the expected value");
-
-    if (__old > 0) {
-      // Nothing to do
-    } else if (__update > 1)
+    (void)__old;
+    if (__update > 1) {
       __a_.notify_all();
-    else
+    } else {
       __a_.notify_one();
+    }
   }
   _LIBCPP_AVAILABILITY_SYNC _LIBCPP_HIDE_FROM_ABI void acquire() {
     auto const __test_fn = [this]() -> bool {
       auto __old = __a_.load(memory_order_relaxed);
       return (__old != 0) && __a_.compare_exchange_strong(__old, __old - 1, memory_order_acquire, memory_order_relaxed);
     };
-    __cxx_atomic_wait(&__a_.__a_, __test_fn);
+    auto const __test_with_old = [this](__cxx_contention_t& __monitor) -> bool {
----------------
jiixyj wrote:


> but I still think
> 
> ```c++
> __a_.wait(0, memory_order_acquire);
> ```
> 
> is simpler and easier to understand.

I absolutely agree -- in this case it is easier to understand on a first glance. On the other hand, in general, you don't want to even spend one nanosecond thinking about your `wait()` method. You just want to take your `try_wait()` and your eventcount abstraction (i.e. `__cxx_atomic_wait()`), compose them, and get your `wait()` for free.

So even if that code would be longer, it is simpler from an architectural viewpoint. Ideally, reviewers would see `wait()` and go "ha, this is just `try_wait()` made blocking by an eventcount". I do think though that the current code in `semaphore`/`latch` could be made clearer to express that design intent, and a few comments wouldn't have hurt, either...

https://github.com/llvm/llvm-project/pull/79265