Re: [boost] [range] Problem with adaptors::map_keys

Subject: Re: [boost] [range] Problem with adaptors::map_keys
From: Jeffrey Lee Hellrung, Jr. (jhellrung_at_[hidden])
Date: 2010-11-29 13:15:59

On 11/28/2010 11:44 PM, Alexander Fokin wrote:
>> Three things:
>> - value_type should just be pair<  Key, T>  (no reference or const
>> qualifications on Key or T)
>> - reference should be pair<  Key const&, T&  >, with T possibly
>> const-qualified (as you've done)
>> - you can (if you want) use iterator_adaptor instead of iterator_facade to
>> automatically get a correct equal, increment, and decrement.
> I have modified the code as you have suggested and now it compiles OK.
> The modified code is available at http://codepad.org/AJFIVbdY.
>
> However, there is one thing I'm concerned about. Consider the example
> at http://codepad.org/plyTCBpx. The following lines do not call the
> copy constructor of class A:
> BOOST_FOREACH(auto pair, m) {
>      const A&key = pair.first;
> }

Good.

> However, when I use range adaptors, the copy constructor is called.
> That was exactly what I was trying to avoid by using references in
> value_type.

It's certainly a reasonable thing to avoid.  I'm not sure why A's copy 
constructor would be called.  Perhaps you should run it through a 
debugger and set a break point in A's copy constructor.  I don't have Qt 
installed so it would take me a little more effort to help you more than 
that ;)

It could be the case that the iterator for the range produced by (m | 
map_keys) returns by value upon dereferencing, rather than by reference. 
  Might be worth checking...

> I guess I'll try to fix this by myself and then file a patch.

Good luck!

- Jeff