Re: [boost] [flat_set] When to sort?

Subject: Re: [boost] [flat_set] When to sort?
From: Marc Glisse (marc.glisse_at_[hidden])
Date: 2017-04-08 14:07:27

On Sat, 8 Apr 2017, Ion Gaztañaga via Boost wrote:

> On 08/04/2017 9:34, Marc Glisse via Boost wrote:
>
>> That seems more complicated than simply giving (in-place) access to the
>> internal sequence_type. You already had to document the invariants for
>> adopt_sequence(ordered_unique_range_t,*) and more or less promised that
>> extract_sequence takes constant time, so the abstraction doesn't buy you
>> much.
>
> adopt/extract maintain class invariants which can avoid a lot of headaches. 
> Even adopt_sequence(ordered[_unique[_range_t, ...) checks preconditions 
> (ordered [and unique]) in debug mode.

The usual safety vs power. I tend to consider that debug mode should not 
influence design (much), but I understand your wish to maintain invariants 
and isolate the points where "unsafe" operations can be performed, and in 
this case the cost (2 vector moves) is not so bad.

> Could you please elaborate what does "giving access to the internal 
> sequence_type" mean? Something like:
>
> class flat_xxx
> {
>   sequence_type &sequence_ref()
>   {  return m_seq;   }
> };
>
> ?

Yes.

-- 
Marc Glisse