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

[Serialization] Boost segfaults in archive/basic_xml_iarchive.hpp
by Mccall, Kurt E. (JSC-EG411) 22 Jun '15

22 Jun '15

If someone could help me diagnose this segmentation fault, I'd appreciate it greatly. It has stumped me for a couple of days. I'm deserializing a non-polymorphic derived class, and Boost crashes when it attempts to process the base class: Here is the pseudocode: struct Base { int id_; std::string long_name_; template<Archive> void serialize(Archive & ar, const unsigned int file_version) { ar & BOOST_SERIALIZATION_NVP(id_); ar & BOOST_SERIALIZATION_NVP(long_name_); // SEGFAULT OCCURS HERE } } struct Derived : public Base { template<Archive> void serialize(Archive & ar, const unsigned int file_version) { ar & BOOST_SERIALIZATION_BASE_OBJECT_NVP(Base); … } } I tried the methods described at http://stackoverflow.com/questions/3396330/where-to-put-boost-class-export-… for exporting the derived class using BOOST_CLASS_EXPORT, but none of them affected the segfault. I guess that it isn't really necessary to export non-polymorphic derived classes. Anyway, the code crashes in archive/basic_xml_iarchive.hpp in this code at line 92: // Anything not an attribute - see below - should be a name value // pair and be processed here typedef detail::common_iarchive<Archive> detail_common_iarchive; template<class T> void load_override( #ifndef BOOST_NO_FUNCTION_TEMPLATE_ORDERING const #endif boost::serialization::nvp< T > & t, int ){ this->This()->load_start(t.name()); this->detail_common_iarchive::load_override(t.value(), 0); // SEGFAULT OCCURS HERE this->This()->load_end(t.name()); } In GDB, "this" has a valid address ( 0x7fffffffdb30) as does "t" (0x7ffffff487b0). In GDB, executing t.name() yielded the correct name "long_name_", but when executing t.value(), the debugger complained about an invalid address: RuntimeError: Cannot access memory at address 0xffffffffffffffe8 That error message comes from GDB's Python utility for pretty-printing strings (the variable being deserialized is a std::string), which is invoked when I print t.value(), but regardless something is wrong in t.value(). Does any of this suggest what I might be doing wrong? Thanks for any help.

2 2

[Serialization] Non-default constructor and template class
by Mccall, Kurt E. (JSC-EG411) 22 Jun '15

22 Jun '15

I am attempting to serialize a template class without a default constructor, so I am overriding boost::serialization::save_construct_data(). I'm not sure of the correct syntax to declare that function. The line marked below fails to compile with a "too many template-parameter-lists" error. // Forward declaration of my class template<class Key, class Value, class Hasher> class HashMap; namespace boost { namespace serialization { template<class Archive> template<class Key, class Value, class Hasher> void save_construct_data(Archive &ar, const HashMap< Key, Value, Hasher > *t, const unsigned int file_version); // error here }} Thanks for any help, Kurt

4 4

[Variant]Maximum Limit of types in variant and fusion vector
by Elizabeta 21 Jun '15

21 Jun '15

Hi I want to use boost::variant with types more than 50. Until now I was using boost::variant with types up to 50 like this : #define BOOST_MPL_CFG_NO_PREPROCESSED_HEADERS #define BOOST_MPL_LIMIT_LIST_SIZE 50 #define BOOST_MPL_LIMIT_VECTOR_SIZE 50 #include <boost/variant.hpp> However , now if I change BOOST_MPL_LIMIT_LIST_SIZE and BOOST_MPL_LIMIT_VECTOR_SIZE to 100 it gives me compiler errors: Cannot open include file: 'boost/mpl/list/list100.hpp': No such file or directory So is 50 the maximum limit of types that can be put in the variant ? Also it seems that the maximum limit of types for vector fusion is 50, and can't increase that, and I want to have fusion vector for more than 50. Is this possible. Any suggestion -- View this message in context: http://boost.2283326.n4.nabble.com/Variant-Maximum-Limit-of-types-in-varian… Sent from the Boost - Users mailing list archive at Nabble.com.

4 5

bimap : Ordering
by Norman Goldstein 20 Jun '15

20 Jun '15

When iterating though a bimap with a left iterator, I notice that for the pairs having the same left member e.g. (L, R1) (L,R2), (L,R3), ... the pairs are not in increasing R, but in the insertion order of the pairs. Is this an artefact of the design, or an intention? The actual bimap I am using is typedef boost::bimap< boost::bimaps::multiset_of< unsigned int >, boost::bimaps::multiset_of< unsigned int >, boost::bimaps::set_of_relation<>, boost::bimaps::with_info< int > > BMAP; I set it up so that info stores the insertion order index. Thanks.

1 0

[Test] Small Enhancement Suggestion
by Merrill Cornish 19 Jun '15

19 Jun '15

I couldn't find the official way of suggesting an enhancement, but here goes: When a Boost Test currently runs in console mode, it logs the message lines "Running X test cases..." at the start and "*** Y failure detected in test suite 'My Test Suite' " at the end. That's all good. However, if <ahem> the test goes bad wrong and never reaches the end, there is no record in the log of which test did it. Suggestion: Change the initial log message to include the test suite name. For example, "Running X test cases from 'My Test Suite'" Merrill Cornish

2 1

Re: [Boost-users] [IPC Named Mutex]
by Lorenzo Trivelli 19 Jun '15

19 Jun '15

> >One thing you could try...in interprocess's win32_api.hpp, there is another implementation of get_last_bootup_time() which uses WMI instead of the event log (there were warnings in the source about WMI not supporting hibernation/clock changes correctly, so beware). If you define >BOOST_INTERPROCESS_BOOTSTAMP_IS_LASTBOOTUPTIME >it should use that other version...give that a try. (in visual studio Project Properties -> C/C++ -> Preprocessor -> Preprocessor Definitions: add a semi colon ';' then t Thanks Tom, I'll try to fallback to the "old" WMI version (which BTW worked in the 1.44 version I we used before) I'd also be glad to fix it by myself and so help boost developement, , but I do not completely understand the usage of the system boot time... If boost needs in some way a tmestamp to differentiate things, in my opinion there are plenty of them to use instead of the last boot time. Obviously it has been done for a reason, but I cannot understand why :-D Best regards Lorenzo

6 11

Linking Boost Log
by Florian Lindner 19 Jun '15

19 Jun '15

Hello, I try to compile/link a boost log example: http://www.boost.org/doc/libs/1_58_0/libs/log/example/doc/tutorial_file.cpp Using: g++ boostlog.cpp -lboost_log -lpthread with g++ 5.1.0 and boost 1.58 on arch linux. The trivial examples worked fine when I put #define BOOST_LOG_DYN_LINK 1 at the very first line. But with this example I get % g++ boostlog.cpp -lboost_log -lpthread /usr/bin/ld: /tmp/ccT2pz5x.o: undefined reference to symbol '_ZN5boost6detail12set_tss_dataEPKvNS_10shared_ptrINS0_20tss_cleanup_functionEEEPvb' /usr/lib/libboost_thread.so.1.58.0: error adding symbols: DSO missing from command line Probably a missing library but which one? I've not really found any linking documentation and I was guessing so far... Thanks, Florian

1 0

[Serialization]: compiler error when registering derived class
by Mccall, Kurt E. (JSC-EG411) 18 Jun '15

18 Jun '15

I'm embarrassed to to ask this, but maybe someone could interpret this strange compiler error for me: I am registering a derived class everywhere that it is archived through a base class pointer: class MyDerivedClass : public MyBaseClass { … }; class AnotherClass { MyBaseClass *p; … void serialize(Archive &ar, unsigned int) { ar.register_type(static_cast<MyDerivedClass *>(NULL)); // compiler error here … ar & p; } At the line above, the error is error: no matching function for call to ‘boost::archive::xml_oarchive::register_type(AnotherClass:serialize (Archive&, unsigned int) [with Archive = boost::archive::xml_oarchive]::MyDerivedClass*)’ Does the complier think that MyDerivedClass is a member of boost::archive::xml_oarchive? Im using Boost 1.57 and gcc 4.4.7. Thanks.

1 0

[Boost] Visual Studio 2013 Debugger Visualizers
by dariomt＠gmail.com 18 Jun '15

18 Jun '15

Hi, FYI, I just found VS2013 debugger visualizers for boost here: https://visualstudiogallery.msdn.microsoft.com/61f12e7a-bc62-4b2c-b02e-d660… Just download and double-click to install. So far works like a charm! I didn't find them listed in the wiki at https://svn.boost.org/trac/boost/wiki/DebuggerVisualizers It may be a good idea to update the wiki page with the VS2013 info. Regards, Dario

3 2

[Pyhton] boost::shared_ptr problem between python modules
by Tamas.Ruszkai＠leica-geosystems.com 17 Jun '15

17 Jun '15

Hi All, I use - Visual Studio 2013 - boost 1.58 with NuGet - 64 bit Intel I have a class (called Log::FileSink) that I want to instantiate via a boost::shared_ptr. This class is derived from an abstract interface (Log::ISink) which is wrapped (Log::ISinkRealizer) boost::python::class_<Log::ISinkRealizer, boost::shared_ptr<Log:: ISinkRealizer>, boost::noncopyable>("ISink", boost::python::no_init) .def("AddEntry", boost::python::pure_virtual(&Log:: ISinkRealizer::AddEntry)) .def("ShallLog", boost::python::pure_virtual(&Log:: ISinkRealizer::ShallLog)); boost::python::class_<Log::FileSink, boost::shared_ptr<Log:: FileSink>, boost::python::bases<Log::ISink>, boost::noncopyable> ("FileSink", boost::python::init<const std::string&>()) .def("AddDomain", &Log::FileSink::AddDomain): When I create a python instance from the FileSink and I pass it to the constructor of another class inside the same module then it works fine. >>> import DiagnsoticsPy >>> logFile = DiagnsoticsPy.FileSink("system.log"); >>> logFile.AddDomain("Sensor.SensorController", DiagnsoticsPy.Severity.VerboseAndAbove); >>> logFile.AddDomain("Python.DI", DiagnsoticsPy.Severity.VerboseAndAbove); >>> logSource = DiagnsoticsPy.Source("Python.DI", logFile) boost::python::class_<Log::Source, boost::shared_ptr<Log::Source >>("Source", boost::python::init<const std::string&>()) .def(boost::python::init<const std::string&, boost:: shared_ptr<Log::ISink>>()); class DIAGNOSTICS_API Source { public: Source(const std::string& name, boost::shared_ptr<Log:: ISink> sink); }; However when I want to pass the same "logFile" instance to a method which is defined in another python module then the conversion between python and C++ fails. boost::python::class_<Sensor::SensorController, boost::shared_ptr <Sensor::SensorController>, boost::noncopyable>("SensorController", boost::python::init<boost::shared_ptr<Log::ISink>>()); class SENSORCONTROL_API SensorController : public boost:: enable_shared_from_this<SensorController> { public: explicit SensorController(boost::shared_ptr<Log::ISink> logSink); }; >>> import SensorControlPy >>> sensorController = SensorControlPy.SensorController(logFile); Python argument types in SensorController.__init__(SensorController, FileSink) did not match C++ signature: __init__(struct _object * __ptr64, class boost::shared_ptr<class Log::ISink>) If I put the all the Python bindings (also the SensorController) into the same python module then the call to the SensorController constructor works.

1 0