- Boost-users - lists.preview.boost.org

[proto] transform with calling templates
by Felipe Magno de Almeida 21 Oct '12

21 Oct '12

Hello, I had the following transform and function object: struct bind_functions_transform : proto::when < proto::terminal<binding::placeholder::method_value<proto::_,proto::_> > , bind_method_call(proto::_value, proto::_data, proto::_state)> > struct bind_method_call : proto::callable { template <typename T> struct result { typedef typename boost::function_types::parameter_types<T>::type parameter_types; typedef typename boost::mpl::next <typename boost::remove_reference <typename boost::mpl::at_c<parameter_types, 2u>::type>::type>::type type; }; template <typename Sig, typename F, std::size_t I, std::size_t N> boost::mpl::size_t<I+1> operator()(binding::placeholder::method_value<Sig, F>const& method , boost::fusion::vector<Class&, environment, binding::virtual_table<N>&>& data , boost::mpl::size_t<I> v) const { ... } }; So, I needed one more compile-time information which would be equal to all calls and transforms, so I made my transform into a template and also bind_method_call as follows: template <typename T> struct bind_functions_transform : proto::when < proto::terminal<binding::placeholder::method_value<proto::_,proto::_> > , bind_method_call<T>(proto::_value, proto::_data, proto::_state)> > template <typename T> struct bind_method_call : proto::callable [...] And then I received the following error: ../boost_1_50_0/boost/proto/transform/detail/preprocessed/construct_funop.hpp:24:33: error: no matching function for call to ‘jvb::binding::bind_method_call<hello_world>::bind_method_call(const jvb::binding::placeholder::method_value<void(), void (hello_world::*)()>&, boost::fusion::vector<jvb::class_&, jvb::environment, jvb::binding::virtual_table<1ul>&, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_, boost::fusion::void_>&, const mpl_::size_t<0ul>&)’ I've remembered the workaround for msvc with proto::call. Because it seemed that the transform was actually trying to construct instead of calling bind_method_call, so I changed to: template <typename PeerClass> struct bind_functions_transform : proto::when < proto::terminal<binding::placeholder::method_value<proto::_,proto::_> > , proto::call<bind_method_call<PeerClass>(proto::_value, proto::_data, proto::_state)> > And it worked. Well, my question is: is this a known limitation? I didn't find it in proto documentation. I'm using GCC 4.7.1 and x86_64 with Linux Regards, -- Felipe Magno de Almeida

2 1

Boost python, unique_ptr and abstract class
by Maxim Yegorushkin 19 Oct '12

19 Oct '12

I am using boost::python to expose a C++ API into Python. Here is what I am trying to do: #include <boost/python.hpp> #include <memory> struct Subscriber { virtual ~Subscriber() = 0; static std::unique_ptr<Subscriber> create(); }; template<class T> inline T* get_pointer(std::unique_ptr<T> const& p) { return p.get(); } BOOST_PYTHON_MODULE(subscriber) { // Expose the class. boost::python::class_< Subscriber , std::unique_ptr<Subscriber> , boost::noncopyable >("Subscriber"); // Expose its factory function. bp::def("createSubscriber", Subscriber::create); } boost::python::class_<> constructor invocation spits out errors (full compiler log attached): /usr/local/ots/boost-1.51.0/include/boost/python/converter/as_to_python_function.hpp:40:75: error: use of deleted function ‘std::unique_ptr<_Tp, _Dp>::unique_ptr(const std::unique_ptr<_Tp, _Dp>&) [with _Tp = Subscriber; _Dp = std::default_delete<Subscriber>; std::unique_ptr<_Tp, _Dp> = std::unique_ptr<Subscriber>]’ /usr/local/ots/boost-1.51.0/include/boost/python/object/pointer_holder.hpp:194:14: error: cannot allocate an object of abstract type ‘Subscriber’ How do I make boost::python friendly to std::unique_ptr<> and how do I wrap an abstract class please? -- Maxim

1 0

Occasional blocking with tcp::socket::receive
by Stuart Siegel 19 Oct '12

19 Oct '12

The call to receive below will occasionally hang. All interactions are synchronous and single threaded. Things work perfectly if I use the alternative <sys/socket.h> (POSIX) implementation under the #else. Since I'm pretty new to boost::asio, I must be doing something wrong but after many hours of trying different things, I'm still stuck. Any advice would be appreciated. #if 1 namespace bas = boost::asio; remaining = X //(something typically about 5000 bytes of char data) bas::streambuf b; bas::streambuf::mutable_buffers_type bufs = b.prepare(1024); while (remaining > 0) { size_t n = sock.receive(bufs); //<-------------this blocks every so often remaining -= n; for ( bas::buffers_iterator<bas::streambuf::mutable_buffers_type, char> it = bas::buffers_begin(bufs); it < (bas::buffers_begin(bufs) + n); it++ ) qba.append(*it); } #else //this works perfectly in the 250000 or so test calls I've hit it with char buf[1024]; while (remaining > 0) { n = read( sock, buf, min( sizeof(buf), remaining ) ); remaining -= n; for (int i = 0 ; i < n; i++ ) qba.append(buf); } #endif

1 0

[Boost.Functional.OverloadedFunction][C++03] Number of overloads in boost::overload_function<...>
by Felipe Magno de Almeida 19 Oct '12

19 Oct '12

Hello, Is there a way to know how many overloads are there in a overloaded_function without resorting to preprocessor metaprogramming in C++03? It doesn't seem to be a MPL sequence, for example. Regards, -- Felipe Magno de Almeida

3 6

boost solution for transforming stl algorithms on proxy iterators?
by Oswin Krause 19 Oct '12

19 Oct '12

Hi, I often have problems which look a bit like this: //example code, only for problem statement //std::vector<bool> chosen only because it is no container :). If this is too silly for you, //think about a matrix interpreted as range of row-vectors. std::vector<bool> myContainer;//does only have proxy iterators. std::vector<double> myDistances; //stores for every element in myContainer a distance //now i want to to sort the elements of both ranges in place //as if they were in a hypothetic range std::vector<std::pair<bool, double> > //and sort them by the second part: boost::sort(make_key_value_range(myContainer,myDistances));//bam, not allowed I omited the definition of make_key_value_range but it should be clear that this returns a proxy range with writable iterators and reference type similar to std::pair<std::vector<bool>::reference,double&> for which swap is overloaded to do the right thing. This is not allowed by the standard as boost::sort calls std::sort which in turn requires random_access_iterators - and they disallow proxy references. This issue is AFAIK resolved with C++11 but some of us still need to support C++03 - and I am really annoyed by this problem. The thing is: this will compile on most compilers. One could just lie about the iterator category, saying it is a random_access_iterator and even VC would be happy with it. However it could just break randomly on a new compiler without warning or has severe performance issus (for example gcc works, but completely ignores my nicely defined swap and instead does a triangle copy :-( ). So how can one resolve this issue? I have the problem, that i need to do this in-place, since copying would result in std::bad_alloc almost surely. Also the above single line of code just looks that nice, that i just want to have it this way. So, how do you resolve this issue? is there maybe some boost library capable of getting this done? e.g. boost.betterSTL? :-). Greetings, Oswin

5 10

Using Adaptable Generator concept with generate_n and boost smart pointers
by Marco Craveiro 18 Oct '12

18 Oct '12

Hi boosters, I'm trying to create a class that models the concept of adaptable generator [1]. The idea is that it generates pointers to test data objects which I'd like to then put inside of smart pointers such as shared_ptr or scoped_ptr (or even std::shared_ptr) via generate_n, etc. This is a (non-compilable) example of my intentions: #include <algorithm> #include <iostream> #include <memory> #include <vector> #include <iterator> struct base { int i; }; struct derived : public base { int z; }; struct generator { typedef base result_type; result_type* operator()() { return new derived(); } }; int main(void) { std::vector<std::shared_ptr<base> > z; std::generate_n(std::back_inserter(z), 5, generator()); return 0; } The std::shared_ptr is incidental; it could have been boost::shared_ptr, etc - its up to client code to decide. I get this to work fine when result_type is say boost::shared_ptr<base> but what I really like is for it to be a raw pointer; is there some boost magic (or standard C++ 11 magic) I can apply to convert it to the value_type of the container I'm trying to generate? Many thanks for your time, Marco -- So young, and already so unknown -- Pauli blog: http://mcraveiro.blogspot.com [1] http://www.sgi.com/tech/stl/AdaptableGenerator.html

1 0

Generically connecting a signal to a member function
by Chris Stankevitz 18 Oct '12

18 Oct '12

Hello, I wrote a function called "Connect" which generically connects a signal to a member function. The pseudo-code for this function is: void Connect(Signal, MemberFunction, pClass) { Signal.connect(Signal::slot_type(MemberFunction, pClass)); } Problem: I had to create "overloads" for different number of signal arguments. This was somewhat expected and we see it in boost with things like boost::bind. What I did not expect was that each of my overloads had to have a different name. This is because all overloads have the same function signature: one signal, one member function, one class pointer. Question: Is there a way to write a generic "Connect" that handles signals of different number of arguments? If overloading is involved, can it be done like bind such that all overloads have the same name? Below is my solution that exhibits the "problem" I described above. Thank you for your help, Chris === #include <boost/bind.hpp> #include <boost/signals2.hpp> #include <iostream> //------------------------------------- // Connect0 // // Generically connect a member // function to a signal producing zero // arguments. //------------------------------------- template < class TASignal, class TAMemberFunction, class TAClass> void Connect0( TASignal& Signal, TAMemberFunction MemberFunction, TAClass* pClass) { typedef typename TASignal::slot_type TDSlot; TDSlot Slot(MemberFunction, pClass); Signal.connect(Slot); } //------------------------------------- // Connect0 // // Generically connect a member // function to a signal producing one // arguments. //------------------------------------- template < class TASignal, class TAMemberFunction, class TAClass> void Connect1( TASignal& Signal, TAMemberFunction MemberFunction, TAClass* pClass) { typedef typename TASignal::slot_type TDSlot; TDSlot Slot(MemberFunction, pClass, _1); Signal.connect(Slot); } //------------------------------------- // Connect2 // // Generically connect a member // function to a signal producing two // arguments. //------------------------------------- template < class TASignal, class TAMemberFunction, class TAClass> void Connect2( TASignal& Signal, TAMemberFunction MemberFunction, TAClass* pClass) { typedef typename TASignal::slot_type TDSlot; TDSlot Slot(MemberFunction, pClass, _1, _2); Signal.connect(Slot); } //------------------------------------- //------------------------------------- struct TCClass { void f0() {} void f1(int) {} void f2(int, int) {} }; //------------------------------------- //------------------------------------- int main() { boost::signals2::signal<void()> Signal0; boost::signals2::signal<void(int)> Signal1; boost::signals2::signal<void(int, int)> Signal2; TCClass Class; Connect0(Signal0, &TCClass::f0, &Class); Connect1(Signal1, &TCClass::f1, &Class); Connect2(Signal2, &TCClass::f2, &Class); Signal0(); Signal1(1); Signal2(2, 2); return 0; }

1 0

Intrsuive multisets and stable sorting
by Dan Posluns 18 Oct '12

18 Oct '12

Hi there, I'm attempting to use intrusive associative containers, but it's essential to the application that as I add "equivalent" objects to the multiset that their order is maintained, ie. that the sort is stable. I'm looking at using avl_multiset for its fast retrieve times. I ran an experiment several times over where I added 1000 entries of random numbers from 0-9 that also contained ordering information to a multiset, and that ordering information wound up being preserved, even when I removed some items and added new ones to the initial tree. So that looks promising. If someone were able to confirm the results of my experiment it would be better, though. Best would be documentation that lists which multiset implementations are stable and which aren't, but I can't find such a thing in the docs. Thanks, Dan.

3 5

[fusion] making an inner struct boost fusion compatible?
by Oswin Krause 17 Oct '12

17 Oct '12

Hi list, I have the following problem: template<class T, class U, class V> struct MyType{ struct nested{ typename MetaFunc<T>::type a; typename MetaFunc<U>::type b; typename MetaFunc<V>::type c; }; }; now this inner struct must be a boost fusion compatible random access sequence. The examples cover only structs on namespace lvel, how can i handle this way? Later these classes will be generated by a macro automatically so even dirty hacks are allowed :). But i don't want to clatter the namespace by pulling nested outside of the class. Is there an easy way or do i have to write the necessary mtafunctions to make this possible? Greetings, Oswin

3 8

undefined reference to boost::hash_range
by Michael Dickey 17 Oct '12

17 Oct '12

I ran into an odd error with 1.51.0 when compiling the following code using gcc 4.4.6 on rhel6/64 with -fkeep-inline-functions (using this flag for debug builds): #include <boost/filesystem/path.hpp> int main(void) { boost::filesystem::path a_path; return 0; } The linker complains about an undefined reference to boost::hash_range<> referenced from boost/filesystem/path.hpp. Looking at path.hpp I can see it includes hash_fwd.hpp not hash.hpp. This is despite the fact that it's actually using various hash methods inside the header file (including hash_range). When I change it to use hash.hpp instead, the linker error goes away. Should path.hpp really use hash.hpp instead, or is this some weird compiler issue, or something else entirely? I figured someone here would know =) Thanks, -Mike

1 0