yield_completion in fibers possibly destroys busy mutex

[Thadeu Luiz Barbosa Dias <thadeuluiz@…>]

Im using something similar to this page:

http://www.boost.org/doc/libs/1_62_0/libs/fiber/doc/html/fiber/callbacks/then_there_s____boost_asio__.html

in my project but it contains a bug. yield_completion::wait() does not reacquire the lock before returning, therefore async_result_base::get() and async_result<T>::get() might return before the notifying thread(calling async_handler::operator() releases the lock. with that, async_result<T> might be destroyed while the lock is being held by the notifying thread causing a crash.

I assumed that the lock gets reacquired on fiber resume, but that is not the case, lk.owns_lock() is false after the suspend statement. the solution is to simply call lk.lock() after the suspend call on the context.

[olli]

should be fixed in #99 (​https://github.com/boostorg/fiber/issues/99)