Re: [boost] Opinion and help with implementing futures

From: Edd Dawson (lists_at_[hidden])
Date: 2008-08-30 05:45:59

On 30 Aug 2008, at 01:23, Bram Neijt wrote:

> Dear reader,
>
> I've been wanting to implement a future in C++ for some time, and
> combined some code to come up with the attached. The idea is that the
> future is a function wrapper around a function call which will call  
> the
> function in a second thread upon creation. Then at a later moment, the
> future can yield the return value of the wrapped function (or block
> until this result is know).
>
> I'm happy with the result I got so far. It's a very rough draft, and I
> would like your help to clean up some details:
> - I would like to have the Future wrapper work with the  
> boost::function<
> int (argument, argument)> syntax, but I can't figure out how to get  
> the
> return type from that syntax to allow for the return value to be  
> defined
> (the d_result member).

Is this really necessary? Once you have a future, you only care about  
its result, surely?

>
> - What would be the best way to handle function arguments? Binding all
> the parameters before passing the resulting bound function object  
> seems
> like the solution, but is it really the best solution?

I think so; by the time the underlying functor is called in the other  
thread, the arguments may have disappeared, making it too late to do a  
copy.

Here's my attempt:
http://www.mr-edd.co.uk/?page_id=58

I found that with my mechanism at least, bind had to create a very  
large number of copies, so I actually ended up making a  
shared_ptr<tuple> containing copies of all the arguments and passing  
that to the thread that makes the eventual call.

This way the arguments were only copied once, but the shared_ptr  
multiple times, which isn't a big deal.

> And last but not least: has this already been done and what is the
> general opinion within the boost community on having a class like  
> this?

I have seen many others floating around, too. I'm sure others will  
chip-in in due course with more info.

Kind regards,

Edd