Re: [boost] [utility] new auto_buffer class --- RFC

Subject: Re: [boost] [utility] new auto_buffer class --- RFC
From: David Abrahams (dave_at_[hidden])
Date: 2009-03-14 00:08:33

On Mar 3, 2009, at 1:45 AM, Thorsten Ottosen wrote:

> Steven Watanabe skrev:
>> AMDG
>> Thorsten Ottosen wrote:
>>>> Certainly swap should be provided somehow; a fast-as-possible  
>>>> swap is
>>>> not otherwise possible since the user can't get at the buffer  
>>>> pointer.
>>>> boost::array supports swap, and yours should be at least as fast as
>>>> that.  On that precedent I'd say you should support swap() under  
>>>> that
>>>> name.  Has anyone ever complained about the deceptive slowness of
>>>> boost::array's swap?
>>>
>>> I don't know. I just stay clear of it. The point is that we don't  
>>> want
>>> inexperienced users to use an O(n) swap accidently.
>> Please define n.
>
> n would be c.size()

I'm sorry, I know this will sound harsh, but I don't know how else to  
say it: we went down this road with ptr_container, and should not  
repeat that mistake.  The instinct to protect naive users from  
performance pitfalls by leaving out fundamental operations (swap in  
this case; assignment and copy in the case of ptr_container) is just  
wrong.  That's especially true when the implementations would NOT  
violate the efficiency guarantees required by the concepts.

Let me also point out that std::array will support swap for C++0x (by  
swapping individual elements).

--
David Abrahams
BoostPro Computing
http://boostpro.com