Re: [Boost-users] future<>.then and promises

Subject: Re: [Boost-users] future<>.then and promises
From: Vicente J. Botet Escriba (vicente.botet_at_[hidden])
Date: 2013-06-28 13:48:14

Le 28/06/13 01:45, Klaim - Joël Lamotte a écrit :
>
> On Thu, Jun 27, 2013 at 9:01 PM, Vicente J. Botet Escriba 
> <vicente.botet_at_[hidden] <mailto:vicente.botet_at_[hidden]>> wrote:
>
>     I guess you need to call fut.then() before the value is set. I
>     will need to check if this works when the future is not created
>     with async().
>
>
> If the callback isn't called when set using then() if the value is 
> already acquired, I don't see the point of then().
> Between the time you retrieve a future from any function returning it 
> and the time you call then(), the value could have been set.
> So I expect ft.then( task ) to call immediately task if the value have 
> already been set.
Sorry, I was wrong and there is an issue with the implementation.

When I fix it, the continuation would not be called immediately but as 
the doc states it should

"- If the parent was created with |promise<<| or with a 
|packaged_task<>| (has no associated launch policy), the continuation 
behaves the same as the third overload with a policy argument of 
|launch::async | launch::deferred| and the same argument for func.
"

> Or maybe I didn't understood correctly what you meant?
>
> On a related topic: I just upgraded to the latest SVN and forced Boost 
> to use decltype as result_of, which fixed a problem already mentionned.
> Just in case it is not already known:
>
> 1.
>
>    ft.then( []( future<T> f ) { return 42; } );
>
> This will not compile (VS2012U3) because there is an attempt to copy 
> future<T> (which looks like a bug I reported where returning a future 
> would
> trigger copy instead of moving the future, bug fixed since then so I'm 
> a bit surprise about this one)
>
Currently the prototype of a continuation takes a future by reference, 
but I could change it to by value (which need to move the future.
> 2.
>   ft.then( []( future<T>& f ) { return 42; } );
>
> This trigger the following warning:
>
> 1>e:\projects\sdk\boost\boost\include\boost-1_54\boost\thread\future.hpp(3956): 
> warning C4715: 
> 'boost::future<aosd::backend::Id<aosd::backend::Project> 
> >::then<<lambda_f0fec0502a034b6d05aaa9610f9b705e> >' : not all control 
> paths return a value
>
> Indeed the then() implementation have a series of if-else ending with 
> an else with assert false forced.
> The compiler can't catch the assertion I suppose something have to be 
> written there after the assertion
> to silence the warning.
>
This is associate to the issue of the current implementation. Instead of 
asserting the code should be

       return 
boost::detail::make_future_async_continuation_shared_state<BOOST_THREAD_FUTURE<R>, 
future_type, F>(
           lock, *this, boost::forward<F>(func)
       );

Best,
Vcente