Is it safe, like in the case of std::mutex
for a std::promise<T>
to be made mutable
, or does it depend on T
? As in:
using Data = std::tuple<bool, int, int>;
struct X {
std::future<Data> prepare() const {
return m_promise.get_future();
}
void asyncHandler(int a, int b) const {
m_promise.set_value({true, a, b});
}
void cancel() const {
m_promise.set_value({false, 0, 0});
}
mutable std::promise<Data> m_promise; // Is this safe?
};
void performAsyncOp(const X& x) {
std::future<Data> fut = x.prepare();
dispatch(x);
std::future_status result = fut.wait_for(std::chrono::milliseconds(150));
if (result == std::future_status::timeout) {
x.cancel();
}
handleResult(fut.get());
}
Let's have a detailed look at the API:
// retrieving the result
future<R> get_future();
// setting the result
void set_value(see below);
void set_exception(exception_ptr p);
// setting the result with deferred notification
void set_value_at_thread_exit(see below);
void set_exception_at_thread_exit(exception_ptr p);
None of the methods is marked const
, so we can't infer any knowledge about the constness from just this. However, the standard mandates thread-safety on the following methods (c.f. 33.6.6.2): set_value
, set_exception
, set_value_at_thread_exit
, and set_exception_at_thread_exit
.
This leaves get_future
unspecified with respect to thread-safety. However, get_future
throws an exception if called more than once, c.f. 33.6.6.14.1. So calling get_future
from multiple threads doesn't really make sense from a practical point of view.
There is no guarantee for thread-safety when calling get_future
and any of the set
methods and get_future
(no matter if it will throw or not) simultaneously, as far as I can see.