... and a *draft* patch!

[Paolo Carlini]

> 2- Change the __uninitialized_*, or, in other terms, add a three
>     arguments _Construct overload for use there.
>
> Everything considered I prefer 2-, for this reason: when the resolution
> of 438 will be in effect, it will suffice to simply remove all the
> _Construct (both two and three arguments) and call __alloc::construct
> directly from the __uninitialized_*. In the meanwhile we would end
> up with a consistent set of _Construct / _Destroy and a consistent set
> of uses from the __uninitialized; all the changes concentrated in
> stl_uninitialized and stl_construct.

... of course missing comments, testcases, etc., etc., only to concretely
explain my take.

Paolo.

P.S. The failure of vector/capacity/2.cc for a preliminary version, pointed
out that we can often spare unnecessary copy-constructions.

////////////////
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch_DRAFT
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20041017/b1d57e06/attachment.ksh>