Re: [boost] [containers] Are there flat_map/set and stable_vector proposals at work?

Subject: Re: [boost] [containers] Are there flat_map/set and stable_vector proposals at work?
From: Rob Stewart (robertstewart_at_[hidden])
Date: 2013-10-28 05:08:01

On Oct 25, 2013, at 3:49 PM, Klaim - Joël Lamotte <mjklaim_at_[hidden]> wrote:

> On Fri, Oct 25, 2013 at 9:44 PM, Rob Stewart <robertstewart_at_[hidden]>wrote:
> 
>> On Oct 25, 2013, at 2:00 PM, Nevin Liber <nevin_at_[hidden]> wrote:
>> 
>>> On 24 October 2013 17:10, Rob Stewart <robertstewart_at_[hidden]> wrote:
>>> 
>>>> I meant that stack allocated memory could be used if std::array's interface satisfies the needs of flat_*. std::array can be one choice among
>>>> many.
>>> 
>>> I'm not seeing how a container of exactly N elements would be useful as a backing store for flat_xxx.  Could you show an example?
>> 
>> I was just thinking of a use case in which a map is needed, but memory is constrained or preallocated. I was also just showing how permitting
>> user-specified containers opens possibilities.
>> 
>> A flat_map, backed by std::array, is quite possibly no faster than a normal C++11 map, with a suitable allocator. A C++03 map won't use its
>> allocator for nodes, but IIRC, a C++11 map will.
> 
> But then it would make the flat_map semantic changes because it have runtime-fixed-size and it makes it not inter-changeable, wouldn't it? That being said it would be the same as to have a pool allocator I guess.

The semantics would be the same: std::bad_alloc when memory is exhausted. The difference is that one gets a prescribed amount of memory a priori and the other can try to use all available application memory.

___
Rob

(Sent from my portable computation engine)