Re: [boost] [poly_collection] Request for comments: fast polymorphic collections

Subject: Re: [boost] [poly_collection] Request for comments: fast polymorphic collections
From: Thorsten Ottosen (tottosen_at_[hidden])
Date: 2016-11-23 12:23:05

On 19-11-2016 10:09, Joaquin M López Muñoz wrote:
> El 18/11/2016 a las 11:56, Thorsten Ottosen escribió:
>> It also seems natural to expect that capacity() - size() for a
>> container to tell us the number of elements I can insert without
>> reallocation will happen, and hence is important if one want to reason
>> about exception-safety. With the exotic definition of capacity() that
>> is no longer the case.
>
> Here I don't see an alternative: please note that c.insert(x) can
> *never* guarantee no memory allocation because x can be of an as-of-yet
> unregistered type, which will force the creation of a new segment etc.

OTOH, if we know the type is registered ... and there is also the 
reverse statement: if capacity is 0, I know the thing can throw.

> So, we can only provide capacity information for *registered* types,
> which is precisely what the exotic definition of capacity() does. Do you
> see any other viable definition?

Hm. Maybe not. We have

cc.capacity()
cc.capacity(index)
cc.capacity<U>()

and the first is specified as

"the minimum return value of capacity for each of the segments of the 
collection, or std::numeric_limits<size_type>::max if there are no 
segments."

I can't see how that is useful for much. The segment specific ones are 
useful, of course. I would still say it should return 0 if there are no 
segments, or that it should not be provided at all. If provided, it 
should follow size() which I assume sums over all segment sizes. That 
is, it should sum over all capacities.

Here are some design questions:

A. Do we want these two states to be independently observable:

- empty, but with empty segments
- empty, but with no segments

B. If I call shrink_to_fit on an container with empty segements, does it 
end up with no segments?

C. Could the segment type be a template argument, or is there good 
reasons for not dong that?