September 2011 - Boost-users - lists.preview.boost.org

[serialization] [units] class_id and class_name to describe quantity dimension in serilization archive
by alfC 25 Sep '11

25 Sep '11

Hi, I am serializing a bunch of object that contains boost.units quantities. Boost.Units provides serialization for quantities already. By default, the output file looks ok but I would like to add a class name to the object description. Current, serializing quantity<si::length> q produces: <q class_id="0" tracking_level="1" version="0" object_id="_0"> <value>5</value> </q> Which doesn't give any information about the quantity at hand, not even for the human reading it. So, I tried to use BOOST_CLASS_EXPORT_GUID to register a name for the serialized class, but it is ignored. Ideally I would like to have something like this, with a class_name description for example: <q class_id="0" class_name = "quantity_si_length" tracking_level="1" version="0" object_id="_0"> <value>5</value> </q> Is something like this possible? Below is the working code: #include <fstream> #include <boost/serialization/export.hpp> #include <boost/archive/text_oarchive.hpp> #include <boost/archive/xml_oarchive.hpp> #include <boost/units/systems/si/io.hpp> using namespace boost::units; BOOST_CLASS_EXPORT_GUID(quantity<si::length>, "quantity_si_length") //ignored int main(int argc, char *argv[]) { std::ofstream ofs("filename.xml"); boost::archive::xml_oarchive oa(ofs); quantity<si::length> q = 5.*si::meter; oa << BOOST_SERIALIZATION_NVP(q); return 0; }

5 7

boost variant question
by petros 25 Sep '11

25 Sep '11

Hi, I am new to the boost variant and have the following issue. I have, two numeric vector-like structures (say a contiguous and a strided – the actual problem has many more), and would like to create a variant of those two to use interchangeably in my code. Then plan is to have different constructors for different hosted structures and play with deriving-from/containing a variant These structures, being stl-like have iterators. Is there a mechanism to export the iterator that corresponds to the structure hosted in the variant and use it? Would I have along-side a variant of the corresponding iterators – if this is possible at all? Is there a –preferrably elegant, nut necessarily efficient- solution to this? Thank you in advance for your help, Kind Regards, Petros

3 6

boost multithread C++ program design for low-latency large-data exchange
by Jack Bryan 25 Sep '11

25 Sep '11

Hi, I am trying to solve a network flow problem by C++ multithreading. Given a network (all nodes are connected by arcs, each arc is connected to 2 and only 2 ending nodes, one is input node and another is output node, each node can have multiple input arcs and output arcs), each node needs to do some computing and then exchange computing result data to its connected input and output nodes. Multiple nodes can be grouped into one task, which is run by one thread. In this way, the whole network computing workload can be partitioned into multiple tasks. All these tasks are pushed into a boost thread pool such that all threads can run the tasks at the same time. But, if a node (in a thread task) needs to do data exchange with another node (in another thread task), there is a synchronization problem. Data receiver needs to wait for data available in the data buffer of the data sender. My proram needs to partition the network such that each thread's task workload is assigned as evenly as possible. If all threads share the one-large data buffer structure, the program parallelism is not good because the critical section is too large. Some threads have to wait the the one-large data buffer structure unlocked even though the part of the data structure (which is useful to them ) has been available for read or write. For example, the one-large data buffer structure has the following buffer cells: cell1 , cell2, cell3 , cell4. When thread 1 is trying to write cell 1, it must lock the whole data buffer structure so that thread 2 cannot read or write cell 2 and so on. So, I want to break the one-large data buffer structure into multiple distinct data cells according to the thread number so that each cell holds the data only needed by one thread task. For example, if we have 2 threads, we create 2 data cells that hold data needed by the 4 thread separately. If we have 4 threads, we create 4 data cells that hold data needed by the 4 thread separately. and so on. My questions are: (1) How to design the data cell ? You can see that its size is based on the number of threads. (2) How to reduce synchronization overhead ? The critical section is small but the overhead of geting and releasing mutex may be very high if the inter-node data exchange frequency is high. (3) When a node's computing is done and data is written to its cell how to notify the data receiver node such that the notification messgae is only received by the waiting thread that run the receiver node computing task. All other unrelated nodes and threads are not impacted. The program is very time-sensitive and the latency of message exchange should be controlled very toughly and reduced as much as possible. Any help is really appreciated. Thanks

1 0

Need a solid dictionary/map/query interface
by Michael Powell 24 Sep '11

24 Sep '11

Hello, A little background, we've got Visual Studio 2008 C++/CLI library through which we'd like to expose a dictionary-mapped data abstraction. We'd like to be able to ask some questions about the dictionary mapping, whether for its keys, values, or some combination thereof. I've got a strong background in C# particularly with regard to its Generic Collections and the Linq area. This is how I'd like to think of the dictionary questions. The rub of course is that the support for C++/CLI Linq is rather slim, if at all. Does boost offer something along these lines comparable to Linq? Thanks... Best regards, Michael

3 4

boost::bind function with its input argument error
by Jack Bryan 24 Sep '11

24 Sep '11

Hi, I am trying to do multithreading by boost::bind. But, I got error: src/model.cpp:225: instantiated from here /boost_1_45_0v/include/boost/bind/mem_fn.hpp:333: error: pointer to member type void (Model::)(taskDataSPType&) incompatible with object type taskDataSPT1ype make: * [model.o] Error 1 Here is my code: void Model::runTask(taskDataSPType& myTask) {} taskDataSPType myTask; fifo_pool tp(thread_num); for (int i = 0 ; i < totalTaskNum ; +i) { taskQueue.wait_and_pop(myTask, i); myTask.taskID = i ; // run the typical iterations tp.schedule(boost::bind(&Model::runTask, &myTask)); } tp.wait(); In another header file, I have : typedef struct taskDataSPT1ype taskDataSPType; struct taskDataSPT1ype { int taskID; int startNode; int endNode; };

2 2

boost::bind with derivated class
by Kraus Philipp 24 Sep '11

24 Sep '11

Hello, I'm using boost::bind for binding parameters to a method for calling a thread group: myclass::mymethod( const mysuperclass& myobj ) { bost::threadgroup mythreads; mythreads.create_thread( boost::bind( &myclass::mythreadmethod, this, boost::ref(myobj) ) ); } myclass::mythreadmethod( const mysuperclass myobj ) {} // I need a copy of the object, so I don't set the reference operator & mysuperclass is a pure virtual class. Now I create a derivate class with class myderivate : public mysuperclass {} and call myderivate y; myclass x; x.mymethod( y ); The compiler breaks with the error, that myobj is a class with pure virtual methods. How I can call the bind methods with the correct object? myobj is an object of the derivated class so I would "send them" to the methods. Thanks Phil

2 2

[Enable-if] Selectively instantiating methods.
by Robert Jones 24 Sep '11

24 Sep '11

Hi All This doesn't work. template<typename T> struct S { typename disable_if<is_const<T>, int>::type mutating_method(); }; I understand why not, although only after the compiler threw it out, but obviously (in hindsight), the method itself is not a template, so can't be selectively instantiated. The question is whether and how I can achieve the effect I'm after, ie., that the method is not present if the struct is instantiated with a const type? Thx, Rob.

5 5

using AM and PM with boost::posix_time newbie
by Adam Zedan 23 Sep '11

23 Sep '11

the following code works properly std::string ts("2002-01-20 10:27:30"); boost::posix_time::ptime t(boost::posix_time::time_from_string(ts)); std::cout << t.time_of_day() << "\n"; 1) Now i want to actually change the string to std::string ts("2002-01-20 10:27:30 AM"); but that wont work ?? Why any suggestions would be appreciated?? 2) also to display time i am using boost::posix_time::ptime pc(boost::posix_time::second_clock::local_time()); std::cout << pc.time_of_day() << "\n"; //output is 10:55:30 how could i make it display am/pm in the output ??

1 0

Building boost.log
by Sillence, Martin (GE Oil & Gas) 23 Sep '11

23 Sep '11

Hi, We're using boost 1.47 and are interested in also using boost.log, I see some promising posts about using this in production: http://thread.gmane.org/gmane.comp.lib.boost.devel/221431 How are people building this, when we try (with the tunk from svn) it depends on boost.move which clashes with some of the functionality within boost 1.47 Is it possible to co-exist with 1.47 or are people using it with an older library? Thanks, M

1 0

Boost::object_pool - Memory allocation problems
by Christian Berg 23 Sep '11

23 Sep '11

#include <boost/pool/object_pool.hpp> class test { test(void){}; public: std::vector<const int*> m_a1; int* m_b1; int* m_b2; const int* m_c1; const int* m_c2; const std::string* m_d1; unsigned int m_e1; unsigned int m_e2; unsigned char m_f1; unsigned char m_f2; unsigned char m_f3; unsigned char m_f4; }; static boost::object_pool<test> s_my_pool; int main(int argc, char* argv[]) { int max = 2097120; for(int i = 0; i < max; i++) { test* tst_c = s_my_pool.construct(test()); } Consider the code above. Compiled with Visual Studio 2010. The purpose of the class is not important. It is its size (48 bytes) that matters. The goal is to use object_pool to minimize memory usage. When running this code the memory rises with just a few bytes more than the needed 48 bytes * 2.097.120 = 100.661.760 bytes, hence the object_pool seems to be working fine. But if I increase the number of elements created by one (to 2097121) the memory use rises to a massive 200.000.000+ bytes. It seems that the object_pool has reached a threshold and that the strategy then is to double the allocated memory. But this can't be right, can it? If so the object_pool is useless because it's using more memory than a simple test* tst = new test(); would have done. Why is object_pool acting like this and is there a way to change the behavior? PS! The threshold value is not exactly the same in another application I have made, but it is pretty close. Best Regards Christian Berg

2 1