July 2014 - Boost-users - lists.preview.boost.org

[Boost 1.56 Beta] [Serialization] Compilation problem with classes having special new operators
by Hijok Payne 28 Jul '14

28 Jul '14

I was trying out the new Boost 1.56 beta and some of my serialization code fails to compile (they work fine with previous versions of Boost e.g. Boost 1.55). I am trying to do serialization of Eigen (http://eigen.tuxfamily.org/) matrices with following code: #include <Eigen/Core> #include <fstream> #include <boost/serialization/vector.hpp> #include <boost/archive/binary_oarchive.hpp> #include <boost/archive/binary_iarchive.hpp> //Boost Serialization of Eigen Matrix namespace boost { template<class Archive, typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols> inline void serialize(Archive & ar, Eigen::Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>& t, const unsigned int file_version) { size_t rows = t.rows(), cols = t.cols(); ar & rows; ar & cols; if (rows * cols != t.size()) t.resize(rows, cols); ar & boost::serialization::make_array(t.data(), t.size()); } } //Simple helper for serialization to a file template <typename T> bool serialize(const T& data, const std::string& filename) { std::ofstream ofs(filename.c_str(), std::ios::out | std::ios::binary); if (!ofs.is_open()) return false; { // use scope to ensure archive goes out of scope before stream boost::archive::binary_oarchive oa(ofs); oa << data; } ofs.close(); return true; } //Simple helper for de-serialization from a file template <typename T> bool deSerialize(T& data, const std::string& filename) { std::ifstream ifs(filename.c_str(), std::ios::in | std::ios::binary); if (!ifs.is_open()) return false; { // use scope to ensure archive goes out of scope before stream boost::archive::binary_iarchive ia(ifs); ia >> data; } ifs.close(); return true; } int main(int argc, char *argv[]) { typedef Eigen::Vector3d VectorType; VectorType* vec = new VectorType(1, 2, 3); serialize(vec, "v4d.dat"); std::cout << "v4d =\n" << *vec << std::endl; VectorType* vec_in; deSerialize(vec_in, "v4d.dat"); std::cout << "v4d_in =\n" << *vec_in << std::endl; return 0; } The error in Visual studio 2013 says: boost/archive/detail/iserializer.hpp(237): error C2665: 'Eigen::PlainObjectBase<Eigen::Matrix<double,3,1,0,3,1>>::operator delete' : none of the 3 overloads could convert all the argument types 2> c:\users\abhijit\codes\installs\eigen\eigen-install\include\eigen3\eigen\src/Core/PlainObjectBase.h(132): could be 'void Eigen::PlainObjectBase<Eigen::Matrix<double,3,1,0,3,1>>::operator delete(void *,const std::nothrow_t &) throw()' 2> c:\users\abhijit\codes\installs\eigen\eigen-install\include\eigen3\eigen\src/Core/PlainObjectBase.h(132): or 'void Eigen::PlainObjectBase<Eigen::Matrix<double,3,1,0,3,1>>::operator delete(void *,void *) throw()' 2> while trying to match the argument list '(VectorType *, size_t)' 2> C:\Boost\boost-1_56_0_b1\boost/archive/detail/iserializer.hpp(224) : while compiling class template member function 'void boost::archive::detail::heap_allocation<T>::has_new_operator::invoke_delete(T *)' There is also the following note, but I cannot seem to understand this. // if compilation fails here, the likely cause that the class // T has a class specific new operator but no class specific // delete operator which matches the following signature. Fix // your program to have this. Note that adding operator delete // with only one parameter doesn't seem correct to me since // the standard(3.7.4.2) says " // "If a class T has a member deallocation function named // 'operator delete' with exactly one parameter, then that function // is a usual (non-placement) deallocation function" which I take // to mean that it will call the destructor of type T which we don't // want to do here. I want to understand this issue because this breaks a lot of existing code.

1 1

Geometry: modifiable?
by Jeremy Murphy 28 Jul '14

28 Jul '14

Is there a generic way to modify geometry structurally? Say if I have a ring and I want to insert a new point? Or would I need to reach into the underlying container to make that kind of change? Thanks, cheers. Jeremy

2 3

[SmartPtr] enable_shared_from_raw
by Gavin Lambert 27 Jul '14

27 Jul '14

Hi all, Firstly, any chance this could get some documentation? Is the author still around? I'm thinking about using this at the moment for two separate reasons (and ways in which enable_shared_from_this seems to fall short): 1. shared_from_this can't be used from the constructor (requiring factory-method-init shenanigans); reportedly (https://groups.google.com/forum/#!topic/boost-list/8JWFZ56ry6k) shared_from_raw can, although I can't find specific details on the caveats mentioned. 2. What I *really* want to be able to do is to create a child object (most commonly a callback functor) that contains a weak_ptr, and it seems silly (and badly performant) to use the usual factory-init pattern to call shared_from_this just to degrade it back to a weak_ptr. And it should be possible to obtain that weak_ptr before any shared_ptr exists, and "fix" it later to point at the real shared_ptr, especially if it is known that the weak_ptr will never actually be used until that real shared_ptr exists. enable_shared_from_raw almost seems to fit the bill (according to http://stackoverflow.com/questions/22184772/why-is-boostenable-shared-from-…) but I encountered the same problem as in this post (http://lists.boost.org/boost-users/2013/05/78833.php) it works with shared_from_raw but not weak_from_raw. However this change: https://github.com/boostorg/smart_ptr/pull/8 fixes that. It looks like the two calls went down different code paths for some reason I can't fathom. Making this change (or the similar-but-better-style one mentioned in the PR discussion) makes the above scenario work as expected. I'm not sure if there was some specific reason to not do this before or if it was just an oversight. Any chance this can make it into 1.56? Or is it too late for that? Regards, Gavin Lambert

3 4

[mpi] potentially blocking send - require blocking
by Dave Clarke 25 Jul '14

25 Jul '14

The boost::mpi documentation for send() is: "This routine executes a potentially blocking send" This differs from MPI_Send() "This routine will block until the message is sent to the destination." I am developing an application which requires the code not to progress until the communication is complete. In which situations does "potentially blocking" apply? And, is there a way of guaranteeing it is complete, the equivalent of MPI_Wait() for MPI_Isend()? Regards, Dave

3 2

Compiling boost on Windows
by Freddy Martinez Garcia 24 Jul '14

24 Jul '14

Hi guys… I need compile the boost for windows but I’m not a windows user… I had made some compilation but the files generated by gcc (from MinGW) has .a extension and not .lib or .dll extension… how can I change that ?? best regard Freddy Martinez Garcia Lic. Ciencia de la Computacion C++ and Qt Senior Developer freddy311082(a)gmail.com

2 1

unsubscribe
by Richard Greene 23 Jul '14

23 Jul '14

1 0

[mpl] BOOST_MPL_ASSERT_MSG for static_assert
by Michael Powell 23 Jul '14

23 Jul '14

Hello, I would like to add some diagnostic messages during compile time, potentially enumerating one or more types as well. static_assert is great for simple messages but falls a bit short due to inability to list types and such. I wonder if BOOST_MPL_ASSERT_MSG is the right answer, however, I am getting the following errors: BOOST_MPL_ASSERT_MSG( std::is_base_of<xnode, Node>::value , "requested type is not a node" , (types<node_type>) ); 1>i:\source\kingdom software\cppxml\xml\domain\parts\xnode.h(111): warning C4002: too many actual parameters for macro 'BOOST_MPL_ASSERT_MSG' It's an error message, but not the one(s) I am expecting. Do I need to enable something, preprocessor definition, something like that? Best regards, Michael Powell

3 3

libboost_thread.a Cross-compile Linaro (ARM-Linux)
by Joseph Longhi 23 Jul '14

23 Jul '14

Hi, I have been trying to get a static library for boost_thread to be built but have so far been unable to. When running: ./b2 toolset=gcc-arm target-os=linux --with-threads The results show 2 skipped which are libboost_thread.a and libboost_system.a. I have all the shared objects built but can't seem to get around this problem. Any help would be greatly appreciated.

1 0

Re: [Boost-users] [odeint] optional data
by Erik Schultheis 22 Jul '14

22 Jul '14

Thanks for the fast response. > > Hi > > On 07/22/2014 01:24 AM, Erik Schultheis wrote: > > Dear all, > > > > i am simulating particle trajectories, and depending on the data i want > > to generate i may or may not have to track additional data [boost.ublas > > c_matrix] along those trajectories. Right now i enable/disable that via > > a template parameter (so, using std::tuple(vec, vec) or tuple(vec, vec, > > c_matrix) and fusion_algebra), which has a few drawbacks. > > The algorithm is already templated to support varying dimensions via a > > template parameter, so the additional data practically doubles the > > number of template instantiations for system and all possible observers. > > This actually got so bad that gcc (mingw gcc 4.8) refuses to compile > > with debugging enabled. Therefore, i would like to make the switch > > between tracking the matrix and ignoring it a runtime choice instead of > > a template parameter. > > I don't understand fully what you exactly want to do. Can you show us > some more code? Can you put the tracking into an observer? I mean, do > you need to solve an ODE to track the additional data? > Unfortunatly, the tracking is an ode itself, so the change depends on the position and the current matrix value, so it cannot be done inside of an observer. As for the current implementation: I have a class BasicTracer<S, I>, where S is the template parameter that decides the state type (currently one of std::array<vec<2>, 2>, std::array<vec<3>, 2>, std::tuple<vec<2>, <vec<2>, matrix<4>> and std::tuple<vec<3>, vec<3>, matrix<6>>) and I the integration startegy (either Integrator, that only operates on the vectors, or MonodromyIntegrator, that also operates on the matrix). The reason for this is that I have two different use cases for the program: 1) a human-viewable visualisation of the trajectories that does not need matrices, and 2) the actual data i want to generate. Since 1) needs far more trajectories to be calculated, it is really helpful that the computation time in this case is lower. With the mentioned limitation (gcc refuses to compile with -g, "tracer.o: too many sections (50892)" and "ccQF7l7o.s: Fatal error: can't close .obj\Debug\src\tracer.o: File too big") the code works as it is now. For now i have only worked with small problem sizes, where the wait time for a result is in the order of a few minutes, however, this is going to change and therefore i want to parallelize the code. The differentiation between he use cases via templates make the code quite complex at times, and i hope to reduce that complexity before i add the parallelization. Therefore, I want to transform from a templated approach to a runtime approach. I am not sure if it is possible to make the code select the dimension at runtime without significant performance impact, but at least the second parameter should be removable without increasing runtime significantly. I am not really sure which parts of the code to post to significantly inprove my description. The tracer itself is a type like template class BasicTracer<std::tuple<vec<3>, vec<3>, monodromy_matrix<3>>, MonodromyIntegrator<std::tuple<vec<3>, vec<3>, monodromy_matrix<3>>>>; or template class BasicTracer<std::array<vec<2>, 2>, Integrator<std::array<vec<2>, 2>>>; the integration is performed like double starte = mEnergy(p); try { mMasterObserver.startTrajectory(p); perform_integration(mIntegrator, p, 2.0, 1.0 / mPotential.getSize(), mMasterObserver, mAbsErrorBound, mRelErrorBound); } catch(int& i) {}; /// \todo this is the only reason for getLastState's existence mEnergyError += std::pow( starte - mEnergy( mMasterObserver.getLastState() ), 2 ); where perform integration just sets parameters to the odeint function: template<typename System, typename State, typename Observer> std::size_t perform_integration(const System& system, State& start, double t_end, double dt, Observer& obs, double abs_err, double rel_err) { using namespace boost::numeric::odeint; // when range access is possible, try range algebra, otherwise, try fusion using algebra_type = typename std::conditional<boost::has_range_iterator<State>::value, range_algebra, fusion_algebra>::type; typedef runge_kutta_cash_karp54<State, double, State, double, algebra_type> error_stepper_type; typedef controlled_runge_kutta<error_stepper_type> controlled_stepper_type; typedef default_error_checker<typename controlled_stepper_type::value_type, typename controlled_stepper_type::algebra_type> error_checker_type; controlled_stepper_type stepper(error_checker_type(abs_err, rel_err)); return integrate_const(stepper, system, start, 0.0, t_end, dt, std::ref(obs) ); } and the operator() of the integrators look like template<class S> void Integrator<S>::operator()( const S& state, S& deriv, double/*t*/ ) const { using namespace system_state; auto position = get_position(state); auto p = position * mPotential.getSize(); double x = p[0]; double y = p[1]; // prevent out of bounds access if(y < 0 || y > mPotential.getSize()-1 || x < 0 || x > mPotential.getSize()-1) { throw(1); }; // calculate acceleration auto& acceleration = get_velocity(deriv); for(int i = 0; i < DIMENSION; ++i) { /// the compiler should be able to ompimize this std::array<int, DIMENSION> didx; didx.fill(0); didx[i] = 1; acceleration[i] = -linearInterpolate( mPotential.getDerivative(didx), p ); } // change in position = current velocity get_position( deriv ) = get_velocity( state ); } template<class S> void MonodromyIntegrator<S>::operator()( const S& state, S& deriv, double/*t*/ ) const { using namespace system_state; auto position = get_position(state); auto p = position * mPotential.getSize(); double x = p[0]; double y = p[1]; // prevent out of bounds access if(y < 0 || y > mPotential.getSize()-1 || x < 0 || x > mPotential.getSize()-1) { throw(1); }; // calculate acceleration auto& acceleration = get_velocity(deriv); for(int i = 0; i < DIMENSION; ++i) { std::array<int, DIMENSION> didx; didx.fill(0); didx[i] = 1; acceleration[i] = -linearInterpolate( mPotential.getDerivative(didx), p ); } // change in position = current velocity get_position( deriv ) = get_velocity( state ); // change in monodromy boost::numeric::ublas::axpy_prod(getMonodromyCoeff( mPotential, p ), get_data(state), get_data(deriv), true); } If there is some other part of the code that would be helpful, please let me know. > > However, the cost of tracking that matrix is quite high (that is why it > > is optional), but simply disableing the d matrix / dt calculation in the > > system does not speed up the calculation much (86 s -> 58 s vs 27 s > > without matrix). A lot of time is spent in adding and multiplying > > numbers that are not used anywhere, or actually even copying them (e.g > > std::swap(c_matrix, c_matrix) shows up quite high in gprof (even though > > i am not sure how reliable the -pg option is in conjunction with -O3)). > > > > So what i am asking here is for suggestions how to best let odeint treat > > the c_matrix as uninitialized memory and never perform any operations on > > that depending on a runtime parameter. > > Options i thought of were: > > * deriving a class from c_matrix and providing my own versions of > > operator overloads that check a flag: > > as far as i know, boost.ublas matrices use some template magic to > > let operators only return expression templates to optimize evaluation if > > operations are chained, and i would not be sure about how to actually > > retain that behaviour in case the matrix is tracked. > > * creating my own algebra class: > > from what i could tell from a few quick glances into the algebra > > files, this would probably be conceptually easier than the other option, > > but would require a lot of repetetive coding. > > > > Does anyone have suggestion for a more efficient implementation strategy? > > Have you considered to use a fusion view which only points to tuple( vec > , vec ). This could be used as the state type you use. Nevertheless, you > need to instantiate the stepper with a real resizable state_type which > you could be again a tuple( vec , vec ). > No, I didn't know about fusion view up until now :), so i am going to take a look at it. Why would I need a resizable state_type? Whether I track the matrix or not is decided before the call into the integrate function. > > >From the options you showed here, an own algebra looks most promising. > This algebra you also be implemented in terms of the existing > fusion_algebra and redirect all calls to fusion_algebra.

2 1

[Statechart] How to decide between post_event and process_event atrun time ?
by Philippe DAVID 22 Jul '14

22 Jul '14

Hi, I have a method in which I want to post/process an event. The thing is, this method can be called from outside the state machine (so I need to call process_event), or indirectly by a reaction of the state machine (in this case I must call post_event since process_event is not reentrant). I don't want to write 2 distinct methods to do this (one that calls post, and the other process), I would much rather prefer adding a 'if' in my method that tells me if I have to call process_event or post_event (if I'm being called indirectly by the state machine or not). Is there a way to do this with statechart ? regards, Philippe

3 3