ping?
Paolo Carlini
pcarlini@suse.de
Mon Oct 4 17:49:00 GMT 2004
chris wrote:
> In my current tree I'm redoing the test suite additions to allow for
> keeping track of the number of assignments / swaps / checks / etc.
> performed on elements of a particular container, to allow for a more
> "system dependant" performance check (although obviously raw time is
> still the most important) which is made much easier by the iter_swap
> patch (as I can overload swap without feeling guilty, whereas
> overloading std::iter_swap is someone badly behaved),
Ok, thanks. An observation: looking at your preliminary testcases I
noticed this kind of code:
iter_swap(vec_int.begin(), --vec_int.end());
Some time ago, Martin kindly explained to me privately, that
--vec_int.end() strictly speaking
might be ill-formed: according to 3.10/5, vec_int.end() is an rvalue.
Therefore, in libtsc++-v3
we can invoke operator--() on it only because our vector::iterator is of
class type (3.10/10).
In my opinion, while elegantly improving the testsuite in this area we
should try to avoid this
kind of dependency on implementation details.
Paolo.
More information about the Libstdc++
mailing list