Re: [boost] Any-Range containing transformed range fails when compiled with optimizations

Subject: Re: [boost] Any-Range containing transformed range fails when compiled with optimizations
From: Auer, Jens (jens.auer_at_[hidden])
Date: 2016-01-15 08:05:13

Hi,

after some thinking, I have come up with a theory what is happening. I think it is one of the cases where the iterator is not really a standard-conform iterator type because dereferencing it does not return a real reference. The warning message points to the derefernce funtion of any_iterator.

Could it be that any_iterator only works with standard-conform real iterators? If so, it would be nice to prevent combining it with range adaptors with a static if, e.g. by adding
static_assert( std::is_reference< decltype(*m_it) >::value, "Dereferencing must return a reference. Proxy objects are not supported")
to any_single_pass_iterator_wrapper::dereference() in boost/range/detail/any_iterator_wrapper.hpp, l. 302.

Cheers,
  Jens Auer
--
Jens Auer | CGI | Software-Engineer
CGI (Germany) GmbH & Co. KG
Rheinstraße 95 | 64295 Darmstadt | Germany
T: [phone]
jens.auer_at_[hidden]
Unsere Pflichtangaben gemäß § 35a GmbHG / §§ 161, 125a HGB finden Sie unter de.cgi.com/pflichtangaben.

CONFIDENTIALITY NOTICE: Proprietary/Confidential information belonging to CGI Group Inc. and its affiliates may be contained in this message. If you are not a recipient indicated or intended in this message (or responsible for delivery of this message to such person), or you think for any reason that this message may have been addressed to you in error, you may not use or copy or deliver this message to anyone else. In such case, you should destroy this message and are asked to notify the sender by reply e-mail.

________________________________________
Von: Boost [boost-bounces_at_[hidden]]&quot; im Auftrag von &quot;Auer, Jens [jens.auer_at_[hidden]]
Gesendet: Donnerstag, 14. Januar 2016 21:28
An: boost_at_[hidden]
Betreff: [boost] Any-Range containing transformed range fails when compiled with optimizations

Hi,

I have a program which produces different results based on the compiler options. The program uses a boost::any_range containing a transformed range where the transformation function just returns a new object of a simple struct:
#include <vector>
#include <boost/range.hpp>
#include <boost/range/any_range.hpp>
#include <boost/range/adaptor/transformed.hpp>
#include <iostream>

template<typename T1, typename T2> void ASSERT_EQ(T1 x,T2 y) {assert(x == y);};

struct SourceEvent
{
    std::uint32_t event;
    void* ptr;
};

int main()
{
    using boost::adaptors::transformed;

    using R = boost::any_range<SourceEvent, boost::forward_traversal_tag, SourceEvent, std::ptrdiff_t>;
    std::vector<std::int32_t> v(100);
    R r = v | transformed( [](auto const&) {return SourceEvent{0u, nullptr};} );
    auto t = v | transformed( [](auto const&) {return SourceEvent{0u, nullptr};} );

    ASSERT_EQ( 100u, boost::size(r) );
    ASSERT_EQ( 100u, boost::size(t) );

    for (auto const& i: t)
    {
        ASSERT_EQ(0u, i.event );
        ASSERT_EQ(nullptr, i.ptr);
    }

    for (auto const& i: r)
    {
        ASSERT_EQ(0u, i.event );
        ASSERT_EQ(nullptr, i.ptr);
    }

    std::cout << "OK" << std::endl;

    return 0;
}

When compiled without optimization, it just prints ok. The same source compiled with optimizations asserts, and prints some interesting warnings:

In file included from /usr/local/boost-1.60.0/include/boost/range/detail/any_iterator.hpp:22:0,
                 from /usr/local/boost-1.60.0/include/boost/range/any_range.hpp:17,
                 from prog.cc:3:
/usr/local/boost-1.60.0/include/boost/range/detail/any_iterator_wrapper.hpp: In member function 'boost::range_detail::any_forward_iterator_wrapper<WrappedIterator, Reference, Buffer>::reference boost::range_detail::any_forward_iterator_wrapper<WrappedIterator, Reference, Buffer>::dereference() const [with WrappedIterator = boost::iterators::transform_iterator<boost::range_detail::default_constructible_unary_fn_wrapper<main()::<lambda(const auto:1&)>, SourceEvent>, __gnu_cxx::__normal_iterator<int*, std::vector<int> >, boost::iterators::use_default, boost::iterators::use_default>; Reference = SourceEvent; Buffer = boost::any_iterator_buffer<64ul>; boost::range_detail::any_forward_iterator_wrapper<WrappedIterator, Reference, Buffer>::reference = SourceEvent&]':
/usr/local/boost-1.60.0/include/boost/range/detail/any_iterator_wrapper.hpp:302:57: warning: function may return address of local variable [-Wreturn-local-addr]
                 return dereference_cast<reference>(*m_it);
                                                         ^
/usr/local/boost-1.60.0/include/boost/range/detail/any_iterator_wrapper.hpp:302:52: note: declared here
                 return dereference_cast<reference>(*m_it);
                                                    ^


prog.exe: prog.cc:7: void ASSERT_EQ(T1, T2) [with T1 = unsigned int; T2 = unsigned int]: Assertion `x == y' failed.


Aborted

The example is compiled with gcc 5.2 and boost 1.60 (http://melpon.org/wandbox/permlink/QF1qyw5bWHKwyKJ8). Is there an error in my code? I would guess that transforming a range into a range of new objects of another type is quite common, so it should work. I am also curious why the warning is only emitted when compiling with optimizations.

Best wishes,
  Jens

--
Jens Auer | CGI | Software-Engineer
CGI (Germany) GmbH & Co. KG
Rheinstraße 95 | 64295 Darmstadt | Germany
T: [phone]
jens.auer_at_[hidden]<mailto:jens.auer_at_[hidden]>
Unsere Pflichtangaben gemäß § 35a GmbHG / §§ 161, 125a HGB finden Sie unter de.cgi.com/pflichtangaben<http://de.cgi.com/pflichtangaben>.

CONFIDENTIALITY NOTICE: Proprietary/Confidential information belonging to CGI Group Inc. and its affiliates may be contained in this message. If you are not a recipient indicated or intended in this message (or responsible for delivery of this message to such person), or you think for any reason that this message may have been addressed to you in error, you may not use or copy or deliver this message to anyone else. In such case, you should destroy this message and are asked to notify the sender by reply e-mail.

_______________________________________________
Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost