Re: [boost] Proposal: Monotonic Containers

Subject: Re: [boost] Proposal: Monotonic Containers
From: Ross Levine (ross.levine_at_[hidden])
Date: 2009-06-16 03:44:43

The list splice thing can't be helped, because you are splicing from a
destructed storage location. So that's just a note in documentation.
Obviously, there's nothing that can be done.

However, the more pressing issue is the non-static data in
monotonic::allocator (namely, the pointer "storage"). On newer compilers,
there will be no problems. However, the allocator is non-portable, and that
REALLY bothers me. There has to be some way to avoid that.

Luckily it seems c++0x will eliminate that assumption about equal
allocators. However it is a royal pain in the meantime.
I have a few ideas and might play around with them.

On Tue, Jun 16, 2009 at 3:26 AM, Christian Schladetsch <
christian.schladetsch_at_[hidden]> wrote:

> However, you are right that it fails on GCC.
>
> On Tue, Jun 16, 2009 at 6:53 PM, Christian Schladetsch <
> christian.schladetsch_at_[hidden]> wrote:
>
> > BOOST_AUTO_TEST_CASE(test_splice)
> > {
> >     monotonic::storage<100> store1;
> >     std::list<int, monotonic::allocator<int> > list1(store1);
> >     {
> >         monotonic::storage<100> store2;
> >         std::list<int, monotonic::allocator<int> > list2(store2);
> >         list2.push_back(1);
> >         list1.splice(list1.begin(), list2);
> >     }
> >     BOOST_CHECK_EQUAL(*list1.begin(), 1);
> > }
> >
> > Running 5 test cases...
> >
> > *** No errors detected
> >
> >
> > On Tue, Jun 16, 2009 at 6:45 PM, Ross Levine <ross.levine_at_[hidden]>
> wrote:
> >
> >> >
> >> >    monotonic::storage<100> store;
> >> >    {
> >> >        typedef std::vector<char, monotonic::allocator<char> > Vector;
> >> >        Vector vec(store);
> >> >        vec.resize(10);        // on the stack
> >> >        vec.resize(1000);    // now on the heap
> >> >    }
> >>
> >> I am telling you, this code is not portable. Some compilers will create
> >> code
> >> that silently crashes because std::vector is not required to use any
> >> specific monotonic::allocator.
> >>
> >> Please, please remember that it is REQUIRED that all allocators of the
> >> same
> >> type are completely interchangeable in STL containers. That means NO
> >> non-static data. So what you have written below could fail on some
> >> compilers. This type of failure is especially noticable in lists, and
> >> almost
> >> unavoidable when using the splice member function if the compiler makes
> >> the
> >> allocator equality assumption.
> >>
> >> Actually, in general, I would assume this code would fail:
> >>
> >>   monotonic::storage<100> store1;
> >>   boost::monotonic::list<int> list1(store1);
> >>   {
> >>       monotonic::storage<100> store2;
> >>       boost::monotonic::list<int> list2(store2);
> >>       list2.push_back(1);
> >>       list1.splice(list1.begin(), list2);
> >>   }
> >>   std::cout << *list1.begin(); // segfault?
> >>
> >> Though I haven't tested it. Of course, I don't think there's a way to
> fix
> >> that. But it's something to know.
> >>
> >> Also, your allocator's member pointer storage is public. I assume it's
> >> supposed to be private?
> >> _______________________________________________
> >> Unsubscribe & other changes:
> >> http://lists.boost.org/mailman/listinfo.cgi/boost
> >>
> >
> >
> _______________________________________________
> Unsubscribe & other changes:
> http://lists.boost.org/mailman/listinfo.cgi/boost
>