snarkmaster wrote: > ... can cover https://github.com/llvm/llvm-project/issues/148380, right?
Above, I mentioned "> `await_suspend` that destroys the coro returns a handle" as one of the cases I don't cover. There's no logical reason this couldn't be supported, but I only needed `void` for short-circuiting coros. I looked through `folly/coro` and found a destroying suspend that returned a handle (see [`DetachedBarrierTask`](https://github.com/facebook/folly/blob/main/folly/coro/detail/BarrierTask.h#L147)). I then made a [quick-and-dirty PoC](https://gist.github.com/snarkmaster/f85ad34dcd0b0f3a810685fc5379948e) of a compiler change to support returning handles from `await_suspend_destroy`. Unfortunately, it didn't make an obvious difference to the code optimization in this case. So, I don't think the current PR should include the extra code to support `await_suspend_destroy` returning handles. We'd need to see some kind of use-case where it makes a meaningful difference to perf, and a reason why making that optimize the normal way is hard. https://github.com/llvm/llvm-project/pull/152623 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
