- Boost-users - lists.boost.org

[thread] Out of order problem signalling a condition and waiting on a predicate
by Kirit Sælensminde 09 May '07

09 May '07

I'm not entirely sure what's going on, but I seem to be having some trouble with the predicate wait() where the predicate is true but the waiting thread is still blocked. I'm sure I'm doing something daft so here is a description with the relevant code in the hope that somebody can point it out to me. I'm using MSVC 7.1 and Boost 1.34.0 on a dual-core computer. What I've done is to build on the basic boost::thread primitives to make a pool of worker objects that can have tasks allocated to them. A good starting point is the queuing of work done through this code (a member of class Worker): boost::shared_ptr< Join< void > > operator()( boost::function0< void > f ) { boost::shared_ptr< Join< void > > join( new Join< void > ); queue( join, f ); return join; } Join< void > is really a future for a task that has no return value. Join< T > (where a T is returned by the task) is implemented in terms of this primitive. The queue() function adds the job to a queue and then signals its worker thread that there is something to do: void FSLib::Worker::queue( boost::shared_ptr< Join< void > > join, boost::function0< void > f ) { boost::mutex::scoped_lock lock( m_mutex ); m_queue.push_back( std::make_pair( join, f ) ); m_control.notify_all(); } The boost::thread is initialised with a static function which waits for items to appear in its queue, performs the function and then notifies the Join object of the results. The function looks like this: void FSLib::Worker::execute() { FSLib::Exceptions::StructuredHandler handler; do { t_queue job; { // Find a job to perform boost::mutex::scoped_lock lock( m_mutex ); if ( m_queue.empty() ) m_control.wait( lock ); if ( !m_terminate && !m_queue.empty() ) job.swap( m_queue ); } for ( t_queue::const_iterator j( job.begin() ); j != job.end() && !m_terminate; ++j ) { // Execute job try { j->second(); } catch ( FSLib::Exceptions::Exception &e ) { boost::mutex::scoped_lock lock( m_mutex ); j->first->m_exception = toString( e ); } catch ( ... ) { boost::mutex::scoped_lock lock( m_mutex ); j->first->m_exception = L"An unknown exception was caught"; } // Notify joins boost::mutex::scoped_lock lock( m_mutex ); j->first->m_completed = true; j->first->m_control.notify_all(); } } while ( !m_terminate ); } (I suspect that I shouldn't be catching all exceptions, but in any case no exceptions are thrown when the problem occurs.) The critical part is at the end. It sets a flag on the Join object and then notifies it's condition object. The Join< void > class looks like this: template<> class F3UTIL_DECLSPEC Join< void > { protected: Join(); public: virtual ~Join(); void wait(); Nullable< wstring > exception(); private: volatile bool m_completed; Nullable< wstring > m_exception; boost::mutex m_mutex; boost::condition m_control; friend void Worker::execute(); friend boost::shared_ptr< Join< void > > Worker::operator()( boost::function0< void > f ); }; The flag is marked volatile which according to Microsoft's documentation ensures that writes to memory happen immediately and the flag will be read from memory every time it is accessed. When the result of the operation is needed then wait() or exception() is called (wait() simply calls exception() but throws a new exception if one was caught). Here is exception(): Nullable< wstring > FSLib::Worker::Join< void >::exception() { boost::mutex::scoped_lock lock( m_mutex ); m_control.wait( lock, boost::lambda::bind( &Join< void >::m_completed, this ) ); return m_exception; } What happens is that once in every few thousand operations it looks like the signals get confused. The waiting thread is stuck inside the condition's wait member even though the predicate is true (it doesn't make any difference if I re-write the exception() member with an explicit while loop). All the worker threads are at the beginning of execute() with empty queues. Because the condition object has already been signalled by the worker thread the Join object gets stuck forever. I think I've got all of the mutexes in the right places, but I think I must be missing something important. I guess there is also a small chance that this is a bug in the boost::thread library (I'm using 1.34.0). I could work around it by sharing a condition between all of the joins. This would massively reduce the probability of this hanging the application (as happens now), but I don't think it would fundamentally solve the problem. Another workaround would be to use the timed wait() functions, but I can't see from the documentation how I'd actually manage to set an xtime up such that it would wake up after a few hundred milliseconds. Another workaround would involve a busy wait - checking maybe every hundred milliseconds, but I'm really loathe to do that. As I say, I'm sure I've missed something, but can't for the life of me work out what it is. K

3 4

boost::timer is not a standard sleep
by Yunior Peralta González 09 May '07

09 May '07

I don't know why but the boost::timer does work well. Please, look at the following example bellow and if any idea let me know. #include <boost/timer.hpp> #include <iostream> #include <unistd.h> int main() { boost::timer timer; sleep(1); /* block until one second has elapsed */ std::cout << timer.elapsed() << std::endl; /* show the elapsed time (one second expected) */ return 0; } Though I am not sure I thing the timer just have in mind the time the processor was busy and not the time the application was running. Has the boost library any standard sleep(), I have heard boost::asio has a very good one. Thanks in advance.

1 0

member_of: a pointer to member trait
by Bernabe Cantos 09 May '07

09 May '07

Hello, I'm using BOOST libraries in my developments. I need a trait returning the class type from a pointer to member type. An example: member_of<int(A::*)(void)>::type // this equals to 'A' member_of<void(B::*)(int) const>::type // this equals to 'B const' member_of<void(void)>::type // this equals to an undefined struct I don't know if the functionality I need exists in BOOST. I hope so. If not, should it be difficult to provide? Maybe doing some modifications to is_member_function_pointer. Thanks in advance

1 1

Building Boost 1.34.0 libraries with bjam
by Richard Hadsell 09 May '07

09 May '07

Options for building the Boost libraries with bjam seem to have changed from 1.33 to 1.34. At least, I can't be sure that the result will be the same. I looked at the 1.34.0 beta docs and the common.jam and gcc.jam files, but I am still unsure of what to do. I built the 1.33.x libraries this way: setenv PYTHON_VERSION 2.3 bjam -sTOOLS=gcc -sGCC_ROOT_DIRECTORY=/opt/gcc344 -sGXX="g++344 -m64" -sGCC="gcc344 -m64" --with-python-root=/usr --stagedir=stageFC3 -sBUILD="debug release <threading>multi <shared-linkable>true <instruction-set>opteron" stage Most of the options make it obvious what I want. The '<shared-linkable>true' option was necessary to add '-fPIC', which our binaries need in order to dynamically link (plug-in) shared libraries. I will appreciate any help you can offer in making the transition to 1.34.0 easier. BTW, it would be nice to see a list of available options for each toolset that comes with Boost bjam. -- Dick Hadsell 914-259-6320 Fax: 914-259-6499 Reply-to: hadsell(a)blueskystudios.com Blue Sky Studios http://www.blueskystudios.com 44 South Broadway, White Plains, NY 10601

7 10

BoostMultiArray & polymorphic access
by Terence Wilson 09 May '07

09 May '07

Hello, We have an image processing application that needs to process N-D data of various numeric types & sizes. Boost MultiArray looks like a pretty good match for our requirements, but I'm struggling to understand how I should structure my code. I would like to have a single type, an Image class say, that can host MultiArray's of all our supported types at runtime i.e. it would not be a class template, however, this would require switch statements to drill-down on the template type of the MultiArray before access. Is there a cleaner way? Thanks, Terence

2 1

2007 Boost Conference is Next Week!
by David Abrahams 08 May '07

08 May '07

Hi Everyone, As you probably know, Boost holds its first annual conference next week (http://boostcon.com) Since we first announced the conference, there have been a couple of recent additions to the already stellar program (http://www.boostcon.com/program) most notably a ConceptGCC InstallFest, so every attendee can have a toolchain with some of the most important C++0x features already implemented (http://www.boostcon.com/program/sessions#gregor-conceptgcc-installfest) There are still a few places left; if you have been thinking of coming, now is the time to take action! Onsite registrations will be accepted Sunday May 13 from 4-6pm in the big entry lounge of Aspen Meadows. If you're planning to register onsite, you should still book a hotel room ASAP (http://www.boostcon.com/location/lodging#reservations) Please send any questions to boostcon-plan-AT-lists.boost-consulting.com See you in Aspen! -- Dave Abrahams Boost Consulting http://www.boost-consulting.com Don't Miss BoostCon 2007! ==> http://www.boostcon.com

1 0

Re: [Boost-users] boost::bind, C++ member function and C callback function
by Li, PingShan (Kansas City) 08 May '07

08 May '07

Changing the OpenSSL API is not an option for me. Let's assume that the callback function accepts a void* parameter, can you show me an example that I can bind C++ member function to a C callback function? Thank you > I got the following error under VC++ 7.1: > > error C2664: 'CRYPTO_set_id_callback' : cannot convert > parameter 1 from 'boost::_bi::bind_t<R,F,L>' to > 'unsigned long (__cdecl *)(void)' > with > [ > R=unsigned long, > F=boost::_mfi::mf0<unsigned long,Test>, > L=boost::_bi::list1<boost::_bi::list_av_1<Test *>::B1> > ] > No user-defined-conversion operator available that > can perform this conversion, or the > operator cannot be called It doesn't even take in void *???? Maybe the fact that it doesn't take in state means something. You should look up that specific callback in the openssl documentation > Is it possible to use boost::bind to bind a C++ member > function as a callback function for C > interface? Its easy given that said C interface allows the user to pass in "other data" a la void *.

2 1

Re: [Boost-users] Undefined reference
by Carl Wainwright 08 May '07

08 May '07

Hi there, Surely if that was the case I would have received a compilation error. As it is my code compiles without any errors/warnings which makes me think it's a linker error or something else. Additionally the headers for boost are in /usr/include/boost so I don't believe I need to specify this include directive in my makefile Cheers Carl |-----Original Message----- |From: boost-users-bounces(a)lists.boost.org |[mailto:boost-users-bounces@lists.boost.org] On Behalf Of Xu, Peng |Sent: 08 May 2007 18:01 |To: boost-users(a)lists.boost.org |Subject: Re: [Boost-users] Undefined reference | |You need to specify where boost headers are located, |-I/path-to-boost-headers. | |-----Original Message----- |From: boost-users-bounces(a)lists.boost.org |[mailto:boost-users-bounces@lists.boost.org] On Behalf Of Carl |Wainwright |Sent: Tuesday, May 08, 2007 11:56 AM |To: boost-users(a)lists.boost.org |Subject: [Boost-users] Undefined reference | |Hi all, | |I've started just recently to use boost in my C++ application |to manage some threads. However, when I link my application I |get strange undefined references | |g++ -g -O -o test_exe -lcppunit -ldl -lpthread -lboost_thread Main.o |test.o |test.o(.text+0x874): In function `threadTestCases::spawnThread()': |: undefined reference to `boost::thread::thread(boost::function0<void, |std::allocator<boost::function_base> > const&)' |test.o(.text+0x8b7): In function `threadTestCases::spawnThread()': |: undefined reference to `boost::thread::join()' |test.o(.text+0x8ce): In function `threadTestCases::spawnThread()': |: undefined reference to `boost::thread::~thread()' |test.o(.text+0x8eb): In function `threadTestCases::spawnThread()': |: undefined reference to `boost::thread::~thread()' |collect2: ld returned 1 exit status |make: *** [test_exe] Error 1 | |As you can see I have linked in the boost_thread and pthread |but still I get the problem. My LD_LIBRARY_PATH is correctly |setup to define the path where boost is installed (/usr/lib/) | |The code is as follows and is used in a CPPUNIT framework. The |boost code is in the spawnThread member function. | |//--------------------------------------------------------------------- |#include "test.h" |#include <ostream> |#include <pthread.h> | |#include <boost/thread/thread.hpp> |#include <boost/bind.hpp> | |using namespace std; |using namespace boost; | |CPPUNIT_TEST_SUITE_REGISTRATION( threadTestCases ); | |void threadTestCases::setUp() |{ | cout << "\nthreadTestCases::setUp\n" << endl ; } | |void threadTestCases::tearDown() |{ | cout << "\nthreadTestCases::tearDown\n" << endl ; } | |void threadTestCases::testThread() |{ | cout << "Thread [testThread] spawned\n" << endl; | | try | { | CPPUNIT_ASSERT(1 == 2); | cout << "Thread end" << endl; | } | catch(...) | { | cout << "Exception caught in testThread, continue\n" << endl; | } | | pthread_exit(0); |} | |void *my_thread(void *) |{ | cout << "Thread [my_thread] spawned\n" << endl ; | | try | { | CPPUNIT_ASSERT(1 == 2); | } | catch(...) | { | cout << "Exception caught in my_thread, continue\n" << endl; | } | | pthread_exit(0); |} | |void threadTestCases::spawnThread() |{ |pthread_t thread[2]; |int ret; | | cout << "\nSpawning new thread\n" << endl; | | // Create thread via boost | boost::thread mfThread(boost::bind(&threadTestCases::testThread, |this)); | | // Start execution of thread | mfThread.join(); |} | | |Is there anything I have done wrong in the code or perhaps in |the build environment which is using Boost library v1.32.0-1 |for Linux RedHat with |g++ (GCC 3.4.6) | |Any help would be appreciated as this has been bugging me for |several hours today. | |Cheers |Carl |_______________________________________________ |Boost-users mailing list |Boost-users(a)lists.boost.org |http://lists.boost.org/mailman/listinfo.cgi/boost-users |_______________________________________________ |Boost-users mailing list |Boost-users(a)lists.boost.org |http://lists.boost.org/mailman/listinfo.cgi/boost-users |

1 0

filtering ostream problem... pushing zlib_compressor works in 1_33, dies in 1_34_0
by Jeff 08 May '07

08 May '07

Hello, The code below works fine in 1_33, but throws a bad_alloc exception after some 10-20K iterations in 1_34_0. Am I doing something wrong? Did the API or usage requirements change? Thanks, Jeff --------------------------------------------------------------------------------------------------------------------------------------- #include <iostream> #include <fstream> #include <sstream> #include <string> #include <boost/iostreams/filtering_stream.hpp> #include <boost/iostreams/copy.hpp> #include <boost/iostreams/filter/zlib.hpp> #include <boost/iostreams/device/array.hpp> #include <boost/iostreams/stream.hpp> #include <boost/iostreams/filtering_streambuf.hpp> bool binaryFile2String(const std::string& filename, std::string& out) { std::ifstream is(filename.c_str(), std::ios_base::in|std::ios_base::binary); if ( is.bad() || is.fail() ) return false; is.seekg(0,std::ios::end); const std::ios::pos_type len(is.tellg()); if ( len > out.max_size() ) return false; out.reserve(len); is.seekg(0, std::ios::beg); std::ostringstream os; os << is.rdbuf(); if ( ! is && ! is.eof() ) return false; out = os.str(); is.close(); return true; } int main(int argc, char** argv) { const char* inFileName = "/tmp/car.xml"; // put your data file here int loops(20000); if ( argc > 1 ) loops = atoi(argv[1]); std::string stringData; if ( ! binaryFile2String(inFileName,stringData) ) { std::cerr<<"Failed reading file "<<inFileName<<" into string"<<std::endl; exit(1); } register int i(0); try { for (i=loops; i > 0; --i) { if ( 0 == (i % 1000) ) std::cerr<<' '<<i; std::istringstream ifs(stringData); std::ostringstream toFS; boost::iostreams::filtering_ostream ofs; // this line causing the bad_alloc exception ofs.push(boost::iostreams::zlib_compressor()); ofs.push(toFS); boost::iostreams::copy(ifs,ofs); ofs.reset(); } } catch(const std::exception& e) { std::cerr<<"\nCaught std::exception on iteration "<<i<<": "<<e.what()<<std::endl; } if ( 0 == i ) { std::cerr<<"\nTEST OK on "<<loops<<" iterations"<<std::endl; } else { std::cerr<<"\nTEST FAILED on iteration "<<i<<std::endl; } return 0; }

2 1

Pointer in function parameter?
by Meryl Silverburgh 08 May '07

08 May '07

Hi, I have a vector <A*> and I want to switch boost's ref count pointer like this: vector< boost::shared_ptr<A> > aVector; aFunc(aVector[0]) ; // just call a My question is should I change the function parameter of aFuncion from this: void aFunc(A* aPtr) { // access aPtr } to this: void aFunc(shared_ptr<A> aPtr) { // access aPtr } Thank you.

3 7