Farid Zaripov wrote:
[...]
Are you suggesting to loosen the test so as not to exercise this
requirement?
I suggest only not to exercise iterators if one deque is empty.
But that would only hide the bug. swapping empty deques needs
to work (and not invalidate iterators) just as well as
swapping ones with elements.
But in current implementation iterator not invalidated. Yes, it's
mebmers
has changed, but it's still valid "end iterator".
But it doesn't point to the right container. Unless I'm still
missing something this needs to pass:
#include <assert.h>
#include <deque>
int main ()
{
std::deque<int> x, y;
std::deque<int>::iterator i = x.begin (), j = x.end ();
std::deque<int>::iterator k = y.begin (), l = y.end ();
std::swap (x, y);
assert (i == y.begin () && j == y.end ());
assert (k == y.begin () && l == y.end ());
}
If we decide to set deque<>._C_end._C_node = 0 for end iterator of
empty
deque It would be a big change (we should change the all places where
_C_end._C_node member dereferenced).
Couldn't we just repoint _C_end._C_node in swap()? I thought
that's what we were doing anyway so I guess I still don't fully
understand the problem.
Martin
