On 21 January 2013 11:19, Andrew Hundt <athundt@gmail.com> wrote:
On Mon, Jan 21, 2013 at 10:45 AM, Olaf van der Spek <ml@vdspek.org> wrote:
What happened to hybrid array/vector?
Strictly speaking it's not a proper superset since there are some fundamental interface differences between a container with a hard size limit of N vs one with a more flexible size limit.
I see no fundamental difference between your solution and a hybrid vector with a null allocator. The limitations of static_vector also provides some of its strengths.
For instance, an individual call to push back() will never be an O(N) operation due to reallocation which is important when latency is critical.
Neither is there one in a hybrid vector with a null allocator.
Also, for real time systems allocation is not desirable and hybrid_vector would require much more care to avoid allocations than static_vector will.
I need correctness first. If I can live with a fixed upper bound, I can use a null allocator.
There are also some properties of hybrid_vector that place it outside the scope of static_vector. For instance, hybrid_vector's swap() could sometimes be O(1) for large N, and O(N) for small N, while static_vector always occupies the small N case.
And you see this as an *advantage* for your solution? For swap, the hybrid vector is never any worse and sometimes better than your solution. Ultimately, I think static_vector and hybrid_vector make the most
sense as two separate classes.
Which is what it boils down to. The rest is just rationalization. I see your solution as a subset of hybrid vector, and if hybrid vector exists, I don't see any reason to use yours over it. -- Nevin ":-)" Liber <mailto:nevin@eviloverlord.com> (847) 691-1404