https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100127
--- Comment #4 from Iain Sandoe <iains at gcc dot gnu.org> ---
(In reply to Riccardo Brugo from comment #0)
> Created attachment 50621 [details]
Of course, we should not ICE - but ...
> struct promise_type {
> std::optional<value_type> _value = std::nullopt;
>
> future get_return_object() {
> return future{handle_type::from_promise(*this)};
> }
> void return_value(value_type val) {
> _value = static_cast<value_type &&>(val);
> }
> auto initial_suspend() noexcept {
> class awaiter {
> std::optional<value_type> & value;
> public:
> explicit awaiter(std::optional<value_type> & val) noexcept :
> value{val} {}
> bool await_ready() noexcept { return value.has_value(); }
> void await_suspend(handle_type) noexcept { }
> value_type & await_resume() noexcept { return *value; }
> };
>
> return awaiter{_value};
> }
The ICE is caused by the code handling a reference from the co_await - where it
was not expecting any value to be returned from the initial or final
await_resume() functions.
---
How do you expect to use the value returned by value_type & await_resume()
noexcept { return *value; } in this?
see: https://eel.is/c++draft/dcl.fct.def.coroutine#5
essentially the value(s) of the initial and final co_await expressions are
discarded. So of course, you can have any side-effect you like from the
evaluation of the expressions - but there is no way for you to see the result
of the await_resume().
Ergo it might as well be void...
however, let's fix the ICE anyway.
