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

[serialization] bad cast for shared library compiled by g++ with visibility=hidden
by Pavol Juhas 15 Aug '11

15 Aug '11

Hello, I am working on a C++ shared library (DLL) that should have a hierarchy of serializable classes using fixed archive format. The serialize template methods of the classes are all instantiated in the DLL, the idea is the library clients should be able to load/save derived objects using shared_ptr<Base>, without an explicit knowledge of the Derived classes. Everything seems to work fine if I build the shared library with g++ using the default visibility option (i.e., every function in the DLL is exported). However, if I compile the code with "-fvisibility=hidden" the test program crashes with bad_cast exception, whenever it tries to serialize a derived object stored as a shared_ptr<Base>. Strangely, if I build the library with NDEBUG defined, the code works fine. The difference seems to be in serialization/smart_cast.hpp which uses different cast code with the NDEBUG macro defined. Any idea why is this happening? Is serialization supported for the shared libraries build with g++ -fvisibility=hidden option? I am using g++ 4.4.3 and the Boost libraries 1.40. Attached is a minimum example where the testlib.hpp and testlib.cpp are the shared library and testprg.cpp is the test program. The code can be built with the attached Makefile as follows: (a) make clean runtest - default g++ visibility, test program works (b) make clean runtest VISIBILITY=1 - builds the library with -fvisibility=hidden, test program fails with the bad_cast exception (c) make clean runtest VISIBILITY=1 NDEBUG=1 - builds the library with -fvisibility=hidden and NDEBUG defined. test program works. Thank you for any pointers, Pavol

1 0

Complier errors using property_tree 5 minute tutorial
by Bill Moo 14 Aug '11

14 Aug '11

Hello Users, I am using boost 1.47.0's property_tree on Windows 7 x64 using the Intel Compiler v 12.0.1029.2010 and having followed the 5 Minute Tutorial to the letter I get errors at complie time. This is my header file : #ifndef INFOSTRUCT_HPP #define INFOSTRUCT_HPP #include <string> struct INFO { unsigned m_Type ; std::string m_Desc ; std::string m_User ; std::string m_Pass ; std::string m_Server ; void load(const std::string & fileName) ; void save(const std::string & fileName) ; } ; #endif // INFOSTRUCT_HPP and this is the .cpp file: #include <boost/property_tree/ptree.hpp> #include <boost/property_tree/xml_parser.hpp> #include "infoStruct.hpp" void INFO::load(const std::string &fileName) { boost::property_tree::ptree pTree ; boost::property_tree::xml_parser::read_xml(fileName, pTree) ; m_Type = pTree.get("Database.Driver", 0) ; m_Pass = pTree.get < std::string>("Database.Pass") ; } void INFO::save(const std::string &fileName) { return ; } But when I compile this I get lots and lots of compiler errors this most significant of those being: ..\..\Includes\boost/property_tree/detail/ptree_implementation.hpp(52): error #501: argument of type "const std::string std::_Pair_base<const std::string, boost::property_tree::basic_ptree<std::string, std::string, std::less<std::string>>>::*" is incompatible with template parameter of type "const std::string std::pair<const std::string, boost::property_tree::basic_ptree<std::string, std::string, std::less<std::string>>>::*" &value_type::first>, ^ detected during: instantiation of class "boost::property_tree::basic_ptree<Key, Data, KeyCompare>::subs [with Key=std::string, Data=std::string, KeyCompare=std::less<std::string>]" at line 182 instantiation of "boost::property_tree::basic_ptree<Key, Data, KeyCompare>::basic_ptree() [with Key=std::string, Data=std::string, KeyCompare=std::less<std::string>]" at line 8 of "..\DataFactory\infoStruct.cpp" With the line in question being : boost::property_tree::ptree pTree ; I honestly cannot see where or what I am doing wrong, so I am hoping someone here can help me. -- Bill

2 4

[mpl][fusion] Issues in the mpl/fusion interface
by Gabriel Redner 14 Aug '11

14 Aug '11

Hi all, I am writing some code which uses MPL and fusion together, and in general I am happy with their interoperability. However, I have stumbled on a few areas in which functionality I expected was missing, or where expressions I expected to work did not. All of these issues have workarounds, but I'd appreciate some input on whether the problem is with the libraries, or with my expectations. 1. mpl::pair and fusion::pair Fusion sequences are conforming MPL sequences, and MPL sequences can be transmuted into fusion sequences. However, the same is not true of their pair classes. I feel that fusion::pair ought to conform to some "MPL pair" concept such that: - I can access its types by the same 'first' and 'second' members as in mpl::pair - mpl::first and mpl::second metafunctions should work against it For instance, I'd like to be able to extract a sequence of key types from a map: // OK! typedef mpl::map<mpl::pair<int, char>, mpl::pair<bool, short> > m1; typedef mpl::transform< m1, mpl::first<mpl::placeholders::_1>, mpl::back_inserter<mpl::vector<> > >::type keys1; // Not OK! typedef fusion::map<fusion::pair<int, char>, fusion::pair<bool, short> > m2; typedef mpl::transform< m2, mpl::first<mpl::placeholders::_1>, mpl::back_inserter<mpl::vector<> > >::type keys2; // OK, but requires different code for mpl and fusion maps typedef mpl::transform< m2, fusion::result_of::first<mpl::placeholders::_1>, mpl::back_inserter<mpl::vector<> > >::type keys3; Also, I feel there should be a canned way to translate mpl::pair into fusion::pair. fusion::result_of::as_pair? 2. mpl::map and fusion::map This code works nicely: typedef mpl::vector<int, char, bool> v1; typedef fusion::result_of::as_vector<v1>::type v2; v2 v(1, 'a', false); However, this does not: typedef mpl::map<mpl::pair<int, char>, mpl::pair<bool, short> > m1; typedef fusion::result_of::as_map<m1>::type m2; m2 m('a', 3); The type 'm2' ends up as fusion::map<mpl::pair<int, char>, mpl::pair<bool, short> >, which clearly is no good. I feel that as_map should automatically translate mpl::pair to fusion::pair so that the result is a usable container. Please let me know if I am barking up the wrong tree, or whether these seem like useful features to anyone else. If there is consensus that they are useful, I would be interested in developing patches for them. Thanks, -Gabe

2 2

Re: [Boost-users] adding elements to a std::container in a
by Martin Elzen 14 Aug '11

14 Aug '11

> Subject: [Boost-users] adding elements to a std::container in a single > expression > Message-ID: <j2787j$bj9$1(a)dough.gmane.org> > Content-Type: text/plain; format=flowed; charset="iso-8859-1"; > reply-type=original > > until now I've usually written my own template to do this: > > > template<typename T> > struct push_back_container:T > { push_back_container &_push_back(const typename T::value_type &_r) > { push_back(_r); > return *this; > } > }; > > one can use this like this: > > push_back_container<std::vector<something> > >()._push_back(something1)._push_back(_something2) > > Is there some more elegant way to do this? Been there, and had the same issues. I've always wondered why std::vector gives you so many ways to assign values to it, but so few to append items, as personally I want to append items much more often than totally change a vector's contents. For what you want to do there is also a standard assign-with-back-inserter whatchamacallit : include vector include algorithm // assign include iterator // back_inserter std::copy( source.begin(), source.end(), std::back_inserter( targetSequenceWithpush_back_memberFn ) );

1 0

Chrono - setting system time
by asif saeed 14 Aug '11

14 Aug '11

Hi, Can Chrono be used to set system time? I have a piece of Windows code using SYSTEMTIME struct for setting all the fields and then calling SetLocalTime(). Is there any facility in boost::chrono to do the same portably? Best regards, Asif

2 1

Issues with brent_find_minima
by Frank Astier 14 Aug '11

14 Aug '11

I must be missing something: the following program doesn't seem to find the right minimum (boost 1.47.0). double f(double x) { return x * cos(x); } int main(int argc, char** argv) { cout << brent_find_minima(f, 3.0, 4.0, 1e-8) << endl; return 0; } I get: 4 -2.61457 But I think the minimum is -3.28837 at x = 3.42562. Did I miss something in calling brent_find_minima?? Also, this function should be called "brent_find_minimum" since it finds a single minimum, IMHO. The singular is minimum in Latin, minima is the plural. Or can the function be called repetitively and each time return a different minimum?? Frank fastier(a)yahoo.com

3 2

[serialization] unregistered class exception when using BOOST_CLASS_EXPORT_GUID
by Ron Hashimshony 14 Aug '11

14 Aug '11

Ron Hashimshony wrote: > Hi all boost users, > We have been using boost.serialization for years. We want to move > from 1.33 to 1.47 and encountered a problem. > Many of our classes are abstract, and the derived are defined with > BOOST_CLASS_EXPORT_GUID. > At one point in time, a new class (B) was added that derives from a > class that was defined with the BOOST_CLASS_EXPORT_GUID macro (A). > It all worked well in 1.33. In 1.47, when we get to de-serialize > class A, we get an exception of unregistered class. > We traced it down, and found out that if we remove the derived class > (B), it > works fine, but this is not an option, and then the derived class (B) > cannot be de-serialized from the file... > If there any way to bypass this problem? was there a change made that > broke this > functionality? > Many Thanks! > Ron > I believe this is a problem introduced in version 140 which is under investigation. Which type of archive are you using? Robert Ramey Hi Robert, We are using XML archives only at the moment. we want to use the portable binaries as well. Any idea when this problem will be solved? Thanks.

1 0

adding elements to a std::container in a single expression
by Peter Foelsche 14 Aug '11

14 Aug '11

until now I've usually written my own template to do this: template<typename T> struct push_back_container:T { push_back_container &_push_back(const typename T::value_type &_r) { push_back(_r); return *this; } }; one can use this like this: push_back_container<std::vector<something> >()._push_back(something1)._push_back(_something2) Is there some more elegant way to do this? Peter

2 1

[C++0x] ISO/IEC approves C++0x standard!
by Beman Dawes 13 Aug '11

13 Aug '11

The C++0x standard has been approved by ISO/IEC. The vote was unanimous; all 21 National Bodies voting were in favor. That means the dozen or so Boost libraries that were in TR1 are now part of the standard itself. Good work, Boosters! --Beman

2 1

[serialization] unregistered class exception when using BOOST_CLASS_EXPORT_GUID
by Ron Hashimshony 13 Aug '11

13 Aug '11

Hi all boost users, We have been using boost.serialization for years. We want to move from 1.33 to 1.47 and encountered a problem. Many of our classes are abstract, and the derived are defined with BOOST_CLASS_EXPORT_GUID. At one point in time, a new class (B) was added that derives from a class that was defined with the BOOST_CLASS_EXPORT_GUID macro (A). It all worked well in 1.33. In 1.47, when we get to de-serialize class A, we get an exception of unregistered class. We traced it down, and found out that if we remove the derived class (B), it works fine, but this is not an option, and then the derived class (B) cannot be de-serialized from the file... If there any way to bypass this problem? was there a change made that broke this functionality? Many Thanks! Ron

2 1