[Patch] libstdc++/23425

[Paolo Carlini]

Paolo Carlini wrote:

>Ok. No confusions, anyway, I was really asking whether the current
>optimizers are basically able to deal perfectly with the
>trivial_destructor case, eventually producing the same code that could
>be obtained if we had a perfect type_traits to begin with. In case of
>vector, to be sure, not deque.
>  
>
To clarify a bit more as I see the issue, my point is that if the above
is true - if the optimizers are able to deal very well with trivial
destructors in vector - then the real value of an accurate
has_trivial_destructor trait would be:

1- For deque.
2- For vector, but *only* in the sense of guiding the choice between an
inline and an out of line version of _M_erase_at_end.

I think the perspective, thanks to the progress of the loop optimizer,
is not the same implicit in the original design of the STL that we
inherited.

Paolo.