Re: [boost] [devector] optimum growing/insertion policy

Subject: Re: [boost] [devector] optimum growing/insertion policy
From: Thorsten Ottosen (tottosen_at_[hidden])
Date: 2017-10-17 13:25:19

Den 17-10-2017 kl. 14:23 skrev Joaquin M López Muñoz via Boost:
> El 16/10/2017 a las 10:45, Thorsten Ottosen via Boost escribió:

>>>> I think in the case where the user has called reserve, it’s a little
>>>> problematic that insert may allocate anyway. That is, for many uses 
>>>> there is
>>>> no infinitely growing sequence .
>>
>> And this one? I'm still not sure we want to allocate on insert before 
>> the buffer is full.
> 
> If we must guarantee that the buffer is full before reallocation, the 
> resulting
> policy won't be optimal wrt number of movements.

Correct. If you want that for insert, you would need to allocate twice 
the needed space.

> This is the tradeoff. I 
> guess
> your reservation stems from the desire that capacity()-size() indicates the
> number of insertions before reallocation, to that I just answered Ion in 
> another post.
> My (current) opinion is that capacity() should not be provided.

My reservation is not really with capacity(), but with the cost of 
allocation for small flat containers. Last I checked, you could copy 
hundreds of objects for the price of one allocation (and the 
deallocation that is implied).

kind regards

Thorsten