July 2008 - Boost - lists.boost.org

Subversion problem(s)?
by Dean Michael Berris 02 Jul '08

02 Jul '08

I tried doing the following just now... Anything going on? $ svn co http://svn.boost.org/svn/boost/trunk boost-trunk-flux svn: REPORT request failed on '/svn/boost/!svn/vcc/default' svn: REPORT of '/svn/boost/!svn/vcc/default': 400 Bad Request (http://svn.boost.org) -- Dean Michael C. Berris Software Engineer, Friendster, Inc.

2 2

[asio] failure in ~wrapped_handler
by Dean Michael C. Berris 02 Jul '08

02 Jul '08

Hi everyone, I recently ran into a problem with Boost.Asio while using a slightly modified version of the HTTP 2 example server. The debugger trace (attached) shows where a segfault had occurred in the code. The short of it is that when ~wrapped_handler is called, the call to boost::detail::atomic_exchange_and_add fails when boost::detail::sp_counted_base::release is invoked. It occurs when a handler_wrapper is deleted/destructed (~handler_wrapper is called) in the strand_service. I'm just wondering what could be causing this -- it's worth noting that under not-so-heavy load, the problem does not manifest itself; but soon enough when the load on the server goes up (more incoming connections) the server crashes consistently. Any pointers where I should be looking? I'm using Boost.Trunk revision number 43296. -- Dean Michael Berris Software Engineer, Friendster, Inc.

3 2

Re: [boost] [system] boost compilation
by simonvifr＠voila.fr 02 Jul '08

02 Jul '08

Hello, Thank you for your help. The error is now fixed. My configuration was not correctly set. However, it still remains an error when building the C++ program. After including the line "#include <boost/asio.hpp>" the following messages are displayed : Linking console executable: .objs\Menu.o:Menu.cpp:(.text+0x19a3): undefined reference to `boost::system::get_system_category()' .objs\Menu.o:Menu.cpp:(.text+0x19bc): undefined reference to `boost::system::get_posix_category()' .objs\Menu.o:Menu.cpp:(.text+0x19d5): undefined reference to `boost::system::get_posix_category()' .objs\Menu.o:Menu.cpp:(.text+0x19ee): undefined reference to `boost::system::get_system_category()' .objs\Menu.o:Menu.cpp:(.text$_ZN5boost4asio5error19get_system_categoryEv[boost::asio::error::get_system_cathttp://mail3.voila.fr/webmail/fr_FR/write.html# Envoyeregory()]+0x7): undefined reference to `boost::system::get_system_category()' collect2: ld returned 1 exit status Process terminated with status 1 (0 minutes, 0 seconds) I think I need to link the "boost_system-iw-mt-1_35.lib" and/or the "boost_system-iw-mt.lib" libraries but I do not know how. I have tried without success. Thank you for your answer to fix this error. Regards. > Message du 01/07/08 à 16h36 > De : "Beman Dawes" <bdawes(a)acm.org> > A : boost(a)lists.boost.org > Copie à : > Objet : Re: [boost] [system] boost compilation > > > simonvifr(a)voila.fr wrote: > > > I contact you because an error occurs when compiling the "system" > > library with the following command prompt : bjam --with-system > > --toolset=intel --debug-configuration stage > > > > The intel compiler seems to be invoked correctly (see mail below). > > > > The message error is : > > > > error_code.cpp libs\system\src\error_code.cpp(90): error: namespace > > "std" has no member "strerror" const char * c_str = std::strerror( ev > > ); ^ compilation aborted for libs\system\src\error_code.cpp (code 2) > > > > My OS version is Windows XP. > > My boost version is 1.35.0 > > My boost-jam version is 3.1.16-1 (ntx86) > > My intel compiler version is 10.1.021 > > Each of those is exactly identical to the setup on my main development > machine. The Intel compiler is one of the ones I test with regularly. > > Yet the command line you specify about works fine for me. > > One possibility, and it is only a guess, is that your configuration is > finding the wrong set of C++ standard library include files. > > > Thank you for your answer to fix this error. > > HTH, > > --Beman > > _______________________________________________ > Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost > > ____________________________________________________ Sur le mail Voila, vous pouvez personnaliser l’anti-spam ! http://mail.voila.fr

1 0

DocBook -> Boost documentation?
by Boris 02 Jul '08

02 Jul '08

Is there any XSLT file to convert a DocBook to a Boost library documentation (the HTML files everybody uses)? I've been clicking around on the Boost website to find out how a library documentation is created but could only find http://www.boost.org/development/requirements.html#Documentation which refers to http://www.boost.org/doc/libs/1_34_1/more/writingdoc/index.html (which basically says everyone may do what he wants). Boris

3 4

Anyone interested in a generic loop (and loop unrolling) library ?
by Hui Li 01 Jul '08

01 Jul '08

Loops, in particular loop-unrolling, can be made generic, easy to write, and accessible to everybody. With the help of Boost.Lambda (and a few other boost libraries), we can easily construct arbitrary loops that are unrolled at compile-time. I haven't implemented the full library yet, in fact I've implemented only a few very preliminary functionalities (hardly flawless), it serves only to illustrate the ideas. From what they can do, a possible Boost.Loop library looks very promising. However I feel that I need much help on the design of such a library. The implementation could be straightforward (with the help of so many good libraries in Boost), once a good design is found. Hopefully there are interested people here so I won't have to work on my own. (If there already exists such a library, please let me know.) In the example, you can see how easy it is to construct arbitrary (limited only by Boost.Lamda) loops that are generically unrolled at compile time. With an optimizing compiler, it can be as efficient as hand-written codes. Not only that, but also can we make (almost) any linear-algebra libraries work together without difficulty. In the example, we don't need any extra coding to make a c_array, std::vector, or even a column_view of a ublas::matrix to work together. In the following I give the full example code to illustrate what it's capable of and how to use it. The header file, loop.hpp, is attached at the end of this post. The example has been tested with apple-gcc4.0.1 . Thanks, Hui /**********************************************************************************************************/ #include <vector> #include <iostream> #include <iterator> #include <functional> #include <boost/array.hpp> #include <boost/numeric/ublas/matrix.hpp> #include <boost/numeric/ublas/matrix_proxy.hpp> #include <boost/numeric/ublas/io.hpp> #include "loop.hpp" using namespace std; using namespace boost; using namespace boost::lambda; using namespace boost::numeric; using namespace boost::numeric::ublas; using namespace loop; int idt(int i) { return i; } int main(int argc, char* argv[]) { const size_t N = 8, M = 3; boost::array<double,N> v; std::vector<double> w(N); matrix<double> m(N,M); // All loops are unrolled at compile time // indtrans acts like an array whose individual element equals its index index_transform<double(int)> indtrans(&idt); // equivalent to for(i=0;i<N;i+=2){ v[i]=w[i]=1; } loop_over< linear_range<0,N,2> >( _1 = _2 = 1, v, w ); // equivalent to for(i=1;i<N;i+=2){ v[i]=i; } loop_over< linear_range<1,N,2> >( _1 = _2, v, indtrans ); // equivalent to for(i=1;i<N;i+=2){ w[i]=i; } loop_over< linear_range<1,N,2> >( _1 = _2, w, indtrans ); // loop_over< linear_range<1,N,2> >( _1 = _2 = _3, v, w, indtrans ); // I wonder why this line doesn't compile with gcc4?? m.clear(); // set m to a zero matrix // equivalent to for(i=1;i<N;++i){ column(m,2)[i]=1+i*2; } loop_over< linear_range<0,N,1> >( _1 = 1+_2*2, loop::ref(column(m,2)), indtrans ); // equivalent to for(i=1;i<2;++i){ v[i]*=-2; } followed by for(i=N-3;i<N;++i){ v[i]*=-2; } loop_over< link_range<linear_range<0,2,1>,linear_range<N-3,N,1> > >( _1 *= -2, v); // vertically print out v and w, side by side loop_over< linear_range<0,N,1> >( cout << _1 << constant('\t') << _2 << constant('\n'), v, w ); cout<<endl; // print out m cout << m << endl << endl; double s = 0; // calculate the inner-product of w and column(m,2) (when the column is viewed as a vector/array) loop_over< linear_range<0,N,1> >( _1 += _2*_3, loop::ref(loop::scalar(s)), w, loop::ref(column(m,2)) ); cout << s << endl << endl; return 0; } /**********************************************************************************************************/ Here is the loop.hpp /********************************************************************************************************** * loop.hpp * Created by Hui Li on 6/26/08. ******************************************************************************************************** */ #ifndef LOOP_H #define LOOP_H #include <boost/static_assert.hpp> #include <boost/ref.hpp> #include <boost/lambda/lambda.hpp> #include <boost/function.hpp> #include <boost/function_types/function_arity.hpp> #include <boost/mpl/if.hpp> #include <boost/mpl/bool.hpp> #include <boost/mpl/assert.hpp> namespace loop { // the empty range struct empty_range {}; // a range starting from Start, ending at End, with stride Stride template < int Start, int End, int Stride = 1 > struct linear_range { BOOST_MPL_ASSERT(( boost::mpl::bool_<(Start*Stride<End*Stride)> )); static const int first = Start; typedef typename boost::mpl::if_ < boost::mpl::bool_< ((Start+Stride)*Stride < End*Stride) >, linear_range < Start+Stride, End, Stride >, empty_range >::type next_range; }; // link two arbitrary ranges template < typename LeftRange, typename RightRange > struct link_range { static const int first = LeftRange::first; typedef typename boost::mpl::if_ < typename boost::is_same < typename LeftRange::next_range, empty_range >::type, RightRange, link_range < typename LeftRange::next_range, RightRange > >::type next_range; }; // implementation of the unrolled loop over a range template < typename LambdaExpr, typename Range > struct loop_Impl { template < typename V1 > void operator()(LambdaExpr& expr, V1& v1) { expr(v1[Range::first]); loop_Impl < LambdaExpr, typename Range::next_range >()(expr,v1); } template < typename V1, typename V2 > void operator()(LambdaExpr& expr, V1& v1, V2& v2) { expr(v1[Range::first],v2[Range::first]); loop_Impl < LambdaExpr, typename Range::next_range >()(expr,v1,v2); } template < typename V1, typename V2, typename V3 > void operator()(LambdaExpr& expr, V1& v1, V2& v2, V3& v3) { expr(v1[Range::first],v2[Range::first],v3[Range::first]); loop_Impl < LambdaExpr, typename Range::next_range >()(expr,v1,v2,v3); } }; // looping over an empty range does nothing template < typename LambdaExpr > struct loop_Impl < LambdaExpr, empty_range > { template < typename V1 > void operator()(LambdaExpr&, V1){} template < typename V1, typename V2 > void operator()(LambdaExpr&, V1&, V2&){} template < typename V1, typename V2, typename V3 > void operator()(LambdaExpr&, V1&, V2&, V3&){} }; // helper functions template < typename Range, typename LambdaExpr, typename V1 > void loop_over(LambdaExpr& expr, V1& v1) { loop_Impl<LambdaExpr,Range>()(expr,v1); } template < typename Range, typename LambdaExpr, typename V1, typename V2 > void loop_over(LambdaExpr& expr, V1& v1, V2& v2) { loop_Impl<LambdaExpr,Range>()(expr,v1,v2); } template < typename Range, typename LambdaExpr, typename V1, typename V2, typename V3 > void loop_over(LambdaExpr& expr, V1& v1, V2& v2, V3& v3) { loop_Impl<LambdaExpr,Range>()(expr,v1,v2,v3); } // map a call to operator[] to a unary function all template < typename Function > struct index_transform { BOOST_STATIC_ASSERT(( boost::function_types::function_arity<Function>::value == 1 )); typedef boost::function<Function> func_type; typename func_type::result_type operator[](typename func_type::argument_type x) { return func(x); } template < typename FuncTp > index_transform(FuncTp f):func(f){} private: func_type func; }; // force removal of const-ness of a reference // not so elegant but should be okay template < typename T > T& ref(const T& t) { return const_cast<T&>(t); } // independent of whatever the index might be, it always returns the reference to the same variable template < typename T > struct scalar_Impl { T& operator[](int) { return ref; } scalar_Impl(T& t):ref(t){} private: T& ref; }; // scalar helper template < typename T > scalar_Impl<T> scalar(T& t) { return scalar_Impl<T>(t); } } // namespace loop #endif /**********************************************************************************************************/

4 3

[system] boost compilation
by simonvifr＠voila.fr 01 Jul '08

01 Jul '08

Hello, I contact you because an error occurs when compiling the "system" library with the following command prompt : bjam --with-system --toolset=intel --debug-configuration stage The intel compiler seems to be invoked correctly (see mail below). The message error is : error_code.cpp libs\system\src\error_code.cpp(90): error: namespace "std" has no member "strerror" const char * c_str = std::strerror( ev ); ^ compilation aborted for libs\system\src\error_code.cpp (code 2) My OS version is Windows XP. My boost version is 1.35.0 My boost-jam version is 3.1.16-1 (ntx86) My intel compiler version is 10.1.021 Thank you for your answer to fix this error. Regards. > Message du 29/06/08 à 16h30 > De : "Vladimir Prus" <ghost(a)cs.msu.su> > A : simonvifr(a)voila.fr > Copie à : boost-build(a)lists.boost.org > Objet : Re: [Boost-build] boost compilation > > > On Sunday 29 June 2008 18:14:43 simonvifr(a)voila.fr wrote: > > > > Hello, > > > > It is now ok for running the icl.exe in the same command prompt. > > However the following command prompt generate errors : bjam --with-system --toolset=intel stage > > I have then added "--debug-configuration" but I have still errors when executing the command prompt : bjam --with-system --toolset=intel --debug-configuration stage > > > > The error messages are : > > > > error_code.cpp > > libs\system\src\error_code.cpp(90): error: namespace "std" has no member "strerror" > > const char * c_str = std::strerror( ev ); > > ^ > > compilation aborted for libs\system\src\error_code.cpp (code 2) > > Okay, so it appears that the intel compiler is invoked correctly, and then reports > an error of the source code itself. At this point, I don't think it's Boost.Build > issue, but either issue with the Boost.System library, or the Intel compiler -- I > suggest you ask at boost(a)lists.boost.org, making sure "[system]" is in email subject. > > > An other error message is : > > > > notice: [cmdline-cfg] Detected command-line request for intel: toolset= intel version= > > notice: [cmdline-cfg] toolset intel not previously configured; attempting to auto-configure now > > What makes you think this is an error message -- it is not. > > - Volodya > ____________________________________________________ Sur le mail Voila, vous pouvez personnaliser l’anti-spam ! http://mail.voila.fr

2 1

[1.36.0] Bug fix schedule change
by Beman Dawes 01 Jul '08

01 Jul '08

The original 1.36.0 schedule called for branches/release to be closed for routine bug fixes on Thursday, July 3. Because of the Fourth of July holiday in the US, that closing date is being moved to Monday, July 7. Thus developers can still commit routine bug fixes through the weekend. --Beman

1 0

[system] LWG issue 805 name changes applied
by Beman Dawes 01 Jul '08

01 Jul '08

The C++ committee's LWG issue 805 has been applied to Boost.System in trunk, using a namespace and enum as a workaround for non-availability of C++ scoped enums. This changes names: posix_errno -> errc posix_category -> generic_category get_posix_category -> get_generic_category "POSIX" -> "GENERIC" The old names are still supported as deprecated synonyms, so virtually no existing code is expected to break. The changes tested OK locally on Windows against Boost.Filesystem and Boost.ASIO. Let me know if any problems arise. Thanks, --Beman

3 2

error_code and error_category problem
by Berserker 01 Jul '08

01 Jul '08

In my project (using asio) I have something like this: void Foo::myCallback(const boost::system::error_code &e) { if(e == boost::asio::error::operation_aborted) { // bla bla... } } This call is made across a shared dll and I notice that for a "strange" reason e == boost::asio::error::operation_aborted was always false. I discover that when the temporary object boost::system::error_code is constructed from the constants boost::asio::error::operation_aborted the error_code's costructor take a pointer to a "error_category" and the comparison operator is (m_cat is an error_category pointer): inline friend bool operator==( const error_code & lhs, const error_code & rhs ) { return lhs.m_cat == rhs.m_cat && lhs.m_val == rhs.m_val; } In my situation (using shared modules) lhs.m_cat == rhs.m_cat points to differents memory regions (even if are of the same type "system_category") because of the static initialization in the function "get_system_category". The problem exist in the comparison operator of error_category too: bool operator==(const error_category & rhs) const { return this == &rhs; } I suggest to change the comparison criteria (for example using an enumerated value) _________________________________________________________________ Sai che con Messenger guardi la TV con i tuoi amici? http://www.messenger.it/messengertv.html

2 1

[interprocess] Bug in os_file_functions.hpp
by Lars Hagström 01 Jul '08

01 Jul '08

Hi, I've just tried to use the file_locks from the boost trunk, and stumbled across a bug (well, two related bugs, really) when using the try_lock* metods. The problem occurs on linux systems only. The try_acquire_file_lock_sharable and try_acquire_file_lock functions both have the following code in them: int ret = ::fcntl(hnd, F_SETLK, &lock); if(ret == -1){ return (errno != EAGAIN && errno != EACCES) ? acquired = false, true : false; } I believe this should be changed into int ret = ::fcntl(hnd, F_SETLK, &lock); if(ret == -1){ return (errno == EAGAIN || errno == EACCES) ? acquired = false, true : false; } (its the errno checking line that is changed) Without my fix the result of trying to lock a lock that someone else has is an exception. Cheers Lars

2 2