June 2009 - Boost-users - lists.preview.boost.org

Creating a boost thread in a matlab mex file causes matlab to crash
by bjs109 25 Jun '09

25 Jun '09

I'm having some issues getting matlab and boost threads to work together. I'm posting here and a number of other places as I don't think this problem fits into any one domain. Specifically, whenever I try to use boost threads within a mex file, it causes matlab to crash. My test program is: #include <boost/thread/thread.hpp> #include <boost/bind.hpp> #include "mex.h" class TestClass { public: void run(); }; void TestClass::run() { for (int n = 0; n < 5; n++) { sleep(1); } } void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]) { TestClass tc; boost::thread *pThread = new boost::thread(boost::bind(&TestClass::run, &tc)); pThread->join(); } The matlab crash output is as follows Configuration: MATLAB Version: 7.7.0.471 (R2008b) MATLAB License: 166534 Operating System: Linux 2.6.18-128.1.6.el5.centos.plusPAE #1 SMP Thu Apr 2 13:33:12 EDT 2009 i686 GNU C Library: 2.5 stable Processor ID: x86 Family 6 Model 7 Stepping 10, GenuineIntel Virtual Machine: Java 1.6.0_04 with Sun Microsystems Inc. Java HotSpot(TM) Client VM mixed mode Default Encoding: ISO-8859-1 Fault Count: 1 Register State: eax = 00d6bf00 ebx = 003cf590 ecx = 040571cc edx = 040571d0 esi = 04057124 edi = 00000000 ebp = a72ba3b8 esp = a72ba37c eip = 56e58955 flg = 00010246 Stack Trace: [0] 0x56e58955(0x04057124, 2, 2, 2) [1] libpthread.so.0:0x0038949b(0xa72bab90, 0, 0, 0) Running GDB on the application reveals that at the time of crashing, there are 13 threads. The backtrace for the created thread / parent thread is given below: Thread 13 (created thread) #0 0x56e58955 in ?? () #1 0x0031daf2 in thread_proxy (param=0x4241124) at ./boost/function/function_template.hpp:647 #2 0x0067049b in start_thread () from /lib/libpthread.so.0 #3 0x005c742e in clone () from /lib/libc.so.6 Thread 12 (parent thread) #0 0x003da410 in __kernel_vsyscall () #1 0x00674595 in pthread_cond_wait@@GLIBC_2.3.2 () from /lib/libpthread.so.0 #2 0x0031b164 in boost::detail::condition_impl::do_wait (this=0x4241140, pmutex=0x4241124) at libs/thread/src/condition.cpp:371 #3 0x0031dd51 in thread (this=0x96c9c28, threadfunc=@0x42411cc) at ./boost/thread/condition.hpp:150 #4 0x00e9a7c2 in mexFunction (nlhs=0, plhs=0x42418fc, nrhs=0, prhs=0x424195c) at threadtest.cpp:25 #5 0x003e584f in mexRunMexFile () from /usr/local/matlabR2008b/bin/glnx86/../../bin/glnx86/../../bin/glnx86/libmex.so I am using: CentOS version 5.3 Boost version 1.33.1 (the one that comes with CentOS 5.3) Matlab R2008b release GCC version 4.1.2 (again, the one that comes with CentOS 5.3) The line used to compile the code: mex -g threadtest.cpp -lboost_thread Any ideas? -- View this message in context: http://www.nabble.com/Creating-a-boost-thread-in-a-matlab-mex-file-causes-m… Sent from the Boost - Users mailing list archive at Nabble.com.

2 2

strange nested bind behavior
by Zachary Turner 24 Jun '09

24 Jun '09

I have a couple of classes that all have a few functions with identical names and signatures. Some of these common functions are template functions where one of the arguments to the function is a callback. For example. class Foo { template<typename Handler> void func(int x, Handler callback) }; class Bar { template<typename Handler> void func(int x, Handler callback) }; Somewhere else, I have a class such as: class Test { variant<Foo&, Bar&> variant_obj; void callback(double d, float f, boost::system::error_code ec, size_t s); void test(); }; and i want to invoke func on the object represented by the variant, with the callback going to this->callback, bind the values of d and f from the place I invoke the visitor, and allowing the implementation of func to specify the values of ec and s. Hopefully this makes sense. I've done the following: struct func_visitor : public boost::static_visitor<> { template<typename T, typename Handler> void operator()(T& t, int x, Handler h) const { t.func(x, h); } }; //... void Test::test() { boost::apply_visitor( boost::bind( func_visitor(), _1, 7, boost::protect(boost::bind( &Test::callback, this, 3.0, 5.0f, _1, _2))), variant_obj); } So now, func should expect that Handler has the signature void (error_code, size_t) I compiled it with empty definitions for both occurences of func() just to make sure everything was ok, then I implement one copy of func as follows: template<typename Handler> void func(int x, Handler callback) { callback(make_error_code(success), (size_t)0); } and now I get compilation error pointing to the line above that invokes callback, with the error along the lines of (it's on a different machine with no network access so i can't paste it) void boost::_bi::protected_bind_t<F>::operator()<boost::system::error_code, size_t>(A1&,A2&) const' : cannot convert parameter 2 from 'size_t' to 'size_t &' If I change the above function to the following: template<typename Handler> void func(int x, Handler callback) { size_t temp = 0; callback(make_error_code(success), temp); } then everything works. Is this normal, or is this some kind of bug in either the library or my compiler? All of my function signatures take arguments by value, so I'm not sure why it would want to convert it to a reference first.

3 2

Recursive boost::variant and/or movability?
by Matthew L. Creech 24 Jun '09

24 Jun '09

Hi, I'm using boost::variant<> while parsing an input file to create a AST. The example at: http://www.boost.org/doc/libs/1_39_0/doc/html/variant/tutorial.html#variant… does something very similar. However, it seems that the example usage of boost::recursive_wrapper<> is inefficient, not because of the wrapper itself, but because the creation of a tree of variants becomes more and more costly as that tree grows. The attached example is similar to the tutorial. Creating a nested tree 10,000 deep (with no branching) causes more than 100,000,000 copy-constructions. I.e. since there's no move-awareness, for each single leaf that gets added to the tree we end up creating a new node, adding the leaf to it, then recursively copying over the entire sub-tree up to that point. Creating a move-constructor for the variant type didn't do any good (it was never invoked). But several discussions online (including notes from the original review) imply that boost::variant<> is move-aware. Is this the case? If so, does anyone know how I can get move semantics working using boost::recursive_wrapper<>? I also tried changing the variant type(s) to std::unique_ptr<> rather than boost::recursive_wrapper<>, but that didn't work either (build errors due to non-copyability, specifically in "variant/variant.hpp":404). This is Boost 1.39, and I'm building with GCC 4.4 in C++0x mode. Thanks for the help! -- Matthew L. Creech

1 1

Re: [Boost-users] [Proto] fatal error C1001: An internal error has occurred in the compiler.
by christophe henry 24 Jun '09

24 Jun '09

Thanks to Eric and Joel, it now works. Just for info, the solution: // grammar forbidding address of for terminals struct state_grammar : proto::not_<proto::address_of<proto::_> > {}; // Forward-declare an expression wrapper template<typename Expr> struct make_state; struct sm_domain : proto::domain< proto::generator<make_state>, state_grammar > {}; template<typename Expr> struct make_state : proto::extends<Expr, make_state<Expr>, sm_domain> { typedef proto::extends<Expr, make_state<Expr>, sm_domain> base_type; // Needs a constructor make_state(Expr const &e = Expr()) : base_type(e) {} // Unhide Proto's overloaded assignment operator using base_type::operator=; }; Then, to define a terminal: struct s1: make_state<proto::terminal<state_tag>::type>{}; Thanks a lot! Regards, Christophe _________________________________________________________________ Invite your mail contacts to join your friends list with Windows Live Spaces. It's easy! http://spaces.live.com/spacesapi.aspx?wx_action=create&wx_url=/friends.aspx…

1 0

Assertion `!pthread_mutex_destroy(&m)' failed
by nmittal 24 Jun '09

24 Jun '09

Hi, I have a small program that receives messages asynchronously (in their own threads). I am using boost::mutexs to synchronize access.. Please refer to the attached code. boost::mutex mtx_ticker_add; int main(int argc, char *argv[]) { //Handler Ctrl-C signal(2, ctrl_c_handler); while (bRun) { sleep(1); } } void ctrl_c_handler(int sig) { cout << "Received SIGTERM" << endl; bRun=false; } void onMessage( void *msg ){ mtx_ticker_add.lock(); //Do Stuff mtx_ticker_add.unlock(); } the program works fine and responds to messages. however I get the following error when i press CTRL-C mysql_dealer: /usr/local/include/boost-1_39/boost/thread/pthread/mutex.hpp:45: boost::mutex::~mutex(): Assertion `!pthread_mutex_destroy(&m)' failed. any help would be much appreciated. thanks Nishant -- View this message in context: http://www.nabble.com/Assertion-%60%21pthread_mutex_destroy%28-m%29%27-fail… Sent from the Boost - Users mailing list archive at Nabble.com.

2 2

boost::variant usage question
by Zachary Turner 24 Jun '09

24 Jun '09

Suppose I've got a bunch of different classes that all support a common compile-time interface. In other words, a bunch of different classes, each of which has N methods with identical names and signatures but no common base class. I want to make a variant out of all these types, and in addition I want to allow visititation to all of the common methods. Currently I have defined a separate static_visitor derived class for every common method, but this is a little bit annoying. For example, I have currently something like this: struct Foo1 { void f(int); void g(double); void h(string); }; struct Foo2 { void f(int); //Common method void g(double); //Common method void h(float); //Not a common method }; struct visit_f : public boost::static_visitor<> { template<typename T> void operator()(T& t, int i) const { t.f(i); } }; struct visit_g : public boost::static_visitor<> { template<typename T> void operator()(T& t, double d) const { t.g(d); } }; What would be the suggested way of making this more generic, so that I need not have a separate visitor for every method?

2 1

Threads example question
by Anders Wallin 24 Jun '09

24 Jun '09

I'm trying to learn how to use Boost.Threads with this example: http://pastebin.ca/1472832 On Ubuntu 8.04 with Boost 1.34 I'm compiling with a makefile like this: t1: t1.o g++ t1.o /usr/lib/libboost_thread-mt.so -o t1 t1.o: t1.cpp g++ -I /usr/include/boost/ -c t1.cpp However I get very similar execution times for the single-threaded version and the one which creates two threads. What am I doing wrong? thanks, Anders W

3 4

[Accumulators] Questions about the library
by Adi Shavit 24 Jun '09

24 Jun '09

Hi, I've been trying out boost::accumulators and it's quite an amazing library! I'd like to use the library in an embedded setting where floating point operations are very expensive, but accuracy is not crucial (though numerical stability is). I have a few question regarding this library. 1. I sometimes need to clear the contents of an accumulator. What is the best way to do this? 2. If my platform does not support HW floating point ops, and all my sample values are, e.g. ints, is there a way to specify that the *computation *and *return value itself* (i.e. result_type, e.g. the mean) will be done in ints (naturally, at the cost of precision)? 3. Is there any info about numerical stability of the algorithms used? For example, there are several ways<http://en.wikipedia.org/wiki/Algorithms_for_calculating_variance>to calculate the variance, not all being equal. The same goes for algorithmic efficiency and memory requirements. 4. On the same note, is there anywhere that explains/describes the actual algorithms used? 5. Finally, is there an implementation of a rolling (sliding window) incremental variance computation? Thanks for the great library. Adi

1 0

[multi_array] Slices along the N'th dimension of a multi_array in a generic function
by Jesse Perla 24 Jun '09

24 Jun '09

I have a generic function which will take in a boost::multi_array of dimension size N. I need to take slices along the M'th < N dimension (known at compile time). I can't figure out how to do this with the multi_array library and associated ranges/array_views. Is there another way to do this rather than building a boost::indices object directly? Besides the sliced dimension, I will always want the entire ranges for the other ones. I believe the problem is that the index_gen object needs different types as we build the indices in parts. The following code shows what a hardcoded example would look like for N = 4 and M = 3 const std::size_t n_1 = 3; const std::size_t n_2 = 4; const std::size_t n_3 = 5; const std::size_t n_4 = 6; typedef boost::multi_array<double, 4> array_type; array_type vals(boost::extents[n_1][n_2][n_3][n_4]); //...fill the array... typedef boost::multi_array_types::index_range range; typedef boost::multi_array_types::index_gen index_gen_type; //Trying to do this dynamically instead, but in pieces. Note that you need the different autos at each stage since you can't build an index directly? auto slice_indices2 = boost::indices[range(0,n_1)][range(0,n_2)]; //first dimensions auto slice_indices3 = slice_indices2[1]; //Slice dimension.... for the 2nd index. auto slice_indices4 = slice_indices3[range(0,n_4)]; //last dimensions auto vals_slice3 = vals[ slice_indices4 ]; //Take slices with the final object. This is a proper multi_array view cout << vals_slice3[0][0][0] << endl; Thanks, Jesse

2 1

[accumulators] serialization
by er 24 Jun '09

24 Jun '09

Hi All, I'm wondering if there is/will be a way to do serialization with accumulators? Thanks.

2 1