Re: [Boost-users] bind for_each and a range of functors

Subject: Re: [Boost-users] bind for_each and a range of functors
From: Peter Dimov (pdimov_at_[hidden])
Date: 2008-10-30 08:01:27

Robert Jones:
> On Thu, Oct 16, 2008 at 11:12 PM, Peter Dimov <pdimov_at_[hidden]> wrote:
>
>> With boost::bind, which doesn't evaluate its first argument, you'll need 
>> to
>> use apply<>, as explained in
>>
>> http://www.boost.org/doc/libs/1_36_0/libs/bind/bind.html#nested_binds
>
>
> I understand the "how" of that, but not the "why". Why does bind not
> evaluate if first argument?

With apologies for the late response:

The first argument is not evaluated because this is what's needed most 
often. It's common to use bind as an implementation detail as in

template<class F> void g1( F f )
{
    g2( boost::bind( f, _2, _1 ) );
}

and, since the caller of g1 may well pass boost::bind(...) to it, evaluating 
f would mean more boost::protect's.

The evaluation of the first argument is practically only needed for cases of 
the form bind( _1, _2 ). We could have made it so that placeholders are 
evaluated, but nested binds are not; this would cover nearly all use cases. 
The reason for not doing that in boost::bind is technical (and somewhat 
historical): the object returned by boost::bind always has a fixed 
result_type, and bind( _1, _2 ) cannot have one.

There is no technical reason that prevents bind<R>( _1, _2 ) from being 
handled, though. I just missed this possibility, and nobody has suggested it 
(to the best of my knowledge).

--
Peter Dimov
http://www.pdplayer.com