The old approach caused that the last awaiter to co_await a coroutine function has overwritten the reference to the previous (so that any previous awaiter would not be resumed when the coroutine function has finished) but at the same time the await_suspend() would return false, so that the awaiter would get resume again immediatelly. However, since the last awaiter has also been registered as the only awaiting coroutine, once the awaited coroutine has finished, the awaiter was resumed for the second time, even if not suspended, causing crashes or memory leaks.
This change makes TaskAwaiterBase::await_suspend() a void function, and stores awaiters as a list, so that they can all get resumed when the coroutine finishes.
The old approach caused that the last awaiter to co_await a coroutine function has overwritten the reference to the previous (so that any previous awaiter would not be resumed when the coroutine function has finished) but at the same time the
await_suspend()
would returnfalse
, so that the awaiter would get resume again immediatelly. However, since the last awaiter has also been registered as the only awaiting coroutine, once the awaited coroutine has finished, the awaiter was resumed for the second time, even if not suspended, causing crashes or memory leaks.This change makes
TaskAwaiterBase::await_suspend()
a void function, and stores awaiters as a list, so that they can all get resumed when the coroutine finishes.