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

bjam and msvc PDB files destination directory?
by Martin B. 03 Nov '10

03 Nov '10

Hi! Is it possible to tell bjam to put the generated PDB files (MSVC) alongside the dll/lib files into the stage directory? Here's what I'm using so far: (Boost 1.44.0) bjam.exe --build-type=complete runtime-link=shared variant=release toolset=msvc-8.0 threading=multi debug-symbols=on debug-store=database define=_SECURE_SCL=0 --with-regex Is there an additional option that I can pass on the command line to have the (correctly generated) PDB files also copied into the stage directory? thanks! - Martin

1 0

[proto] : Using proto with NVIDIA CUDA nvcc (redux)
by Manjunath Kudlur 03 Nov '10

03 Nov '10

I remember there were postings in this mailing list before about trying to using proto with NVIDIA's CUDA compiler, and Joel Falcou has contributed some code to make proto work with nvcc. Well, I was trying few experiments with the most recent CUDA compiler release (3.2). I believe one of the exciting things about using proto with CUDA is the ability to write mini programs in different DSELs using proto and ship the mini programs over to the "device" side and execute it there. This idiom will look as follows : template<typename F> __global__ void kernel(F f) { f(...); } kernel<<<blocksize, gridsize>>>(a+b /* a big proto expression */); I tried to write the simplest form of the above use case : #include <boost/proto/proto.hpp> namespace proto=boost::proto; template<typename> struct expr; struct domain : proto::domain<proto::pod_generator<expr> > {}; template<typename E> struct expr { BOOST_PROTO_BASIC_EXTENDS(E, expr<E>, domain) }; template<typename F> __global__ void kernel(F f) { } struct term {}; expr<proto::terminal<term>::type> a = {{{}}}; expr<proto::terminal<term>::type> b = {{{}}}; expr<proto::terminal<term>::type> c = {{{}}}; int main() { kernel<<<1,1>>>(proto::deep_copy(b+c)); return 0; } Note the deep_copy when passing proto expression to kernel functions. Since proto captures terminals by reference, the host side pointers that are held in proto expressions won't make sense on the device side, so at the very least, we have to make copies of terminals before passing to the device. Now, compiling the above program with nvcc (CUDA 3.2, host compiler gcc 4.4.3, Ubuntu 10.04) results in the following error (besides spewing a bunch of warnings) : $ nvcc -v -keep t.cu -I $HOME/boost/boost-trunk/ ... t.cudafe1.stub.c: In function ‘void __device_stub__Z6kernelI4exprIN5boost5proto7exprns_10basic_exprINS2_3tag4plusENS2_7argsns_5list2IS0_INS4_IS6_NS8_IS0_INS4_INS5_8terminalENS7_4termI4termEELl0EEEESE_EELl2EEEES0_INS4_IS9_NSA_IiEELl0EEEEEELl2EEEEEvT_(_Z4exprIN5boost5proto7exprns_10basic_exprINS1_3tag4plusENS1_7argsns_5list2IS_INS3_IS5_NS7_IS_INS3_INS4_8terminalENS6_4termI4termEELl0EEEESD_EELl2EEEES_INS3_IS8_NS9_IiEELl0EEEEEELl2EEEE&)’: t.cudafe1.stub.c:8: error: invalid cast from type ‘const expr<boost::proto::exprns_::basic_expr<boost::proto::tag::address_of, boost::proto::argsns_::list1<expr<boost::proto::exprns_::basic_expr<boost::proto::tag::plus, boost::proto::argsns_::list2<expr<boost::proto::exprns_::basic_expr<boost::proto::tag::plus, boost::proto::argsns_::list2<expr<boost::proto::exprns_::basic_expr<boost::proto::tag::terminal, boost::proto::argsns_::term<term>, 0l> >, expr<boost::proto::exprns_::basic_expr<boost::proto::tag::terminal, boost::proto::argsns_::term<term>, 0l> > >, 2l> >, expr<boost::proto::exprns_::basic_expr<boost::proto::tag::terminal, boost::proto::argsns_::term<int>, 0l> > >, 2l> >&>, 1l> >’ to type ‘char*’ t.cudafe1.stub.c:8: error: invalid cast from type ‘const expr<boost::proto::exprns_::basic_expr<boost::proto::tag::address_of, boost::proto::argsns_::list1<expr<boost::proto::exprns_::basic_expr<boost::proto::tag::plus, boost::proto::argsns_::list2<expr<boost::proto::exprns_::basic_expr<boost::proto::tag::plus, boost::proto::argsns_::list2<expr<boost::proto::exprns_::basic_expr<boost::proto::tag::terminal, boost::proto::argsns_::term<term>, 0l> >, expr<boost::proto::exprns_::basic_expr<boost::proto::tag::terminal, boost::proto::argsns_::term<term>, 0l> > >, 2l> >, expr<boost::proto::exprns_::basic_expr<boost::proto::tag::terminal, boost::proto::argsns_::term<int>, 0l> > >, 2l> >&>, 1l> >’ to type ‘size_t’ # --error 0x1 -- The CUDA compiler works by splitting the original C++ code into "host" code and "device" code. The device code is completely lowered to C. So it seems like the above error is when passing a proto expression to the C-lowered version of device code. The proto expression is a simple POD type, so I am perplexed by the above error. (BTW, when the host compiler is gcc 4.1.3, the above program causes a segmentation fault in gcc, but that's for another forum). This might well be CUDA compiler's problem, but I dug a little deeper. I transformed the proto expression into a simpler POD type as follows : #include <boost/proto/proto.hpp> namespace proto=boost::proto; struct filler {}; template<typename tag, typename C0, typename C1=filler> struct myexpr { C0 c0; C1 c1; }; struct myadd {}; struct myterm {}; struct xform_callable : proto::callable { template<typename> struct result; template<typename This, typename LHS> struct result<This(LHS&)> { typedef myexpr<myterm, LHS> type; }; template<typename LHS> typename result<xform_callable(LHS&)>::type operator()(LHS &l) { myexpr<myterm, LHS> e = {l}; return e; } template<typename This, typename tag, typename LHS, typename RHS> struct result<This(tag, LHS, RHS)> { typedef myexpr<tag, LHS, RHS> type; }; template<typename tag, typename LHS, typename RHS> typename result<xform_callable(tag, LHS, RHS)>::type operator()(tag, LHS l, RHS r) { myexpr<tag, LHS, RHS> e = {l, r}; return e; } }; struct xform : proto::or_< proto::when<proto::plus<proto::_, proto::_>, xform_callable(myadd(), xform(proto::_left), xform(proto::_right))> , proto::when<proto::terminal<proto::_>, xform_callable(proto::_value)> > {}; template<typename> struct expr; struct domain : proto::domain<proto::pod_generator<expr> > {}; template<typename E> struct expr { BOOST_PROTO_BASIC_EXTENDS(E, expr<E>, domain) }; template<typename F> __global__ void kernel(F f) { } struct term {}; expr<proto::terminal<term>::type> a = {{{}}}; expr<proto::terminal<term>::type> b = {{{}}}; expr<proto::terminal<term>::type> c = {{{}}}; int main() { kernel<<<1,1>>>(xform()(b+c)); return 0; } The above program compiles fine with CUDA 3.2. That leaves me with the question : what does a proto expression contain more than myexpr above, that causes nvcc to fail? Again, this might be CUDA compiler's problem and they are the ones that should fix it. However, if someone in this forum can think of a simple fix to proto (or some other boost library that is actually causing the problem), that would be great too. The uniform template expansion across device and host code in CUDA programs is a powerful feature, and I envision some cool libraries that can be written using proto with CUDA. So making proto work with CUDA is a big win. Manjunath

5 12

[fusion] Are Sequence stream IO operators recursive?
by Hugh Dickinson 03 Nov '10

03 Nov '10

I'm trying to output an adapted fusion sequence (PMT::QE::DataSet<PMT::QE::Measurement>) containing other adapted fusion sequences to std::cout. I had hoped for recursive behaviour, such that the contents of each sub-sequence are printed in turn. Having looked at the code in $BOOST_ROOT/boost/fusion/sequence/io/detail/out.hpp it looks like: ---- PMT::QE::DataSet<PMT::QE::Measurement> output; ... Fill output ... std::cout << output << std::endl; ---- should work. Unfortunately, I'm getting compilation errors like: no match for ‘operator<<’ in ‘os << boost::fusion::operator* [with Iterator = boost::fusion::basic_iterator<boost::fusion::struct_iterator_tag, boost::fusion::assoc_struct_category, const PMT::QE::DataSet<PMT::QE::Measurement>, 0>](((const boost::fusion::iterator_base<boost::fusion::basic_iterator<boost::fusion::struct_iterator_tag, boost::fusion::assoc_struct_category, const PMT::QE::DataSet<PMT::QE::Measurement>, 0> >&)((const boost::fusion::iterator_base<boost::fusion::basic_iterator<boost::fusion::struct_iterator_tag, boost::fusion::assoc_struct_category, const PMT::QE::DataSet<PMT::QE::Measurement>, 0> >*)((const boost::fusion::basic_iterator<boost::fusion::struct_iterator_tag, boost::fusion::assoc_struct_category, const PMT::QE::DataSet<PMT::QE::Measurement>, 0>*)first))))’ The other errors simply replace the index 0 with 1 and 2 (PMT::QE::DataSet has 3 elements). On the other hand, the following does work: ---- PMT::QE::DataSet<PMT::QE::Measurement> output; ... Fill output ... std::cout << at_c<0>(output) << std::endl; std::cout << at_c<1>(output) << std::endl; std::cout << at_c<2>(output) << std::endl; ---- Is this the expected behaviour? What am I missing? Is there a way to achieve the desired recursive behaviour? Many thanks, Hugh

4 5

boost threads without exceptions...
by Dhirendra Pal Singh 03 Nov '10

03 Nov '10

Hi Folks, If this question has been asked already, I apologise, but I have to go back to my group and present the answer. So please bear with me. In my current project we are using boost extensively, with exceptions disabled, until last week, when I started using boost threads. I get the following error when compiling... *************** deepsin@deepsin-lnx01:~/code/boost_thread/boost_threads_eg1$ g++ -fno-exceptions -g -I /usr/local/include/boost/ t1.cpp -o t1 -lboost_thread In file included from /usr/local/include/boost/exception_ptr.hpp:9, from /usr/local/include/boost/thread/future.hpp:14, from /usr/local/include/boost/thread.hpp:24, from t1.cpp:3: /usr/local/include/boost/exception/detail/exception_ptr.hpp:17:2: error: #error This header requires exception handling to be enabled. ************** This means that any header file that includes, thread.hpp can't be compiled with -fno-exceptions. I have couple of questions if someone could help me with, I would appreciate. Q1). https://svn.boost.org/trac/boost/ticket/2100, actually says that threads can be compiled without exceptions. Unfortunately its not the case, as I looked at top-of-trunk codebase and the problem still exists. Is this true? I mean threads require exceptions for a fact that is?. If not then can someone help me how to get it to compile without exceptions, and discard the reset of my email please. Q2). If the answer to Q1 is "yes boost threads require exceptions", then was the decision conscious, given a ticked, was opened and closed specifically for this purpose. Q3). Also any idea what is needed to make boost threads work without exception. I mean in terms of man-hours. If my group is okay perhaps I may be willing to work on that change, submit it back to the community, but would need some sort of understanding of the work required and time needed. Any help would be much appreciated.... Thanks in advance... /dev/deep

3 2

cannot open shared object file
by Mike Douglass 03 Nov '10

03 Nov '10

Downloaded the trunk ( Today Nov 2 ) Did: ./bjam --with-libraries=graph,regex ./bjam compiled OK. mv boost-trunk/stage/lib/* /usr/local/lib mv boost-trunk/boost /usr/include my program plc.cpp compiled OK with the following makefile. g++ -ggdb -O3 plc.cpp -o plc -I./Includes -L/usr/local/lib -lboost_graph -lboost_regex But running plc gave the following error ./plc < graph.xml > out.xml ./plc: error while loading shared libraries: libboost_graph.so.1.45.0: cannot open shared object file: No such file or directory But clearly libboost_graph.so.1.45.0 is there. [george@localhost lib]$ ls -lrt /usr/local/lib total 6768 -rwxr-xr-x 1 root root 1155683 Nov 2 20:57 libboost_regex.so.1.45.0 -rwxr-xr-x 1 root root 1155683 Nov 2 20:57 libboost_regex.so -rw-r--r-- 1 root root 2563346 Nov 2 20:57 libboost_regex.a -rwxr-xr-x 1 root root 534275 Nov 2 20:57 libboost_graph.so.1.45.0 -rwxr-xr-x 1 root root 534275 Nov 2 20:57 libboost_graph.so -rw-r--r-- 1 root root 923492 Nov 2 20:57 libboost_graph.a [george@localhost lib]$ Whats wrong? Running Centos 5.5 [george@localhost Program]$ uname -a Linux localhost.localdomain 2.6.18-194.el5 #1 SMP Fri Apr 2 14:58:14 EDT 2010 x86_64 x86_64 x86_64 GNU/Linux Thanks

1 0

Multi file test module in Boost.Test
by Guillaume Dumont 03 Nov '10

03 Nov '10

Hi, I have been trying to get a multi file test module to execute without success. I have the following setup: // main.cpp #define BOOST_TEST_MODULE example #include <boost/test/unit_test.hpp> // suite1.cpp #include <boost/test/unit_test.hpp> BOOST_AUTO_TEST_SUITE(suite1); BOOST_AUTO_TEST_CASE(test1) { ... } BOOST_TEST_SUITE_END(); // suite2.cpp #include <boost/test/unit_test.hpp> BOOST_AUTO_TEST_SUITE(suite2); BOOST_AUTO_TEST_CASE(test2) { ... } BOOST_TEST_SUITE_END(); But only one of the two test cases runs. If I put the suites into the same .cpp file all the test cases are run. Is it possible to link against the static library variant of boost test and have a multi file test module? From what I understand it should be, but I can just find how. Thank you. -- Guillaume Dumont ========================= dumont.guillaume(a)gmail.com

1 0

[proto] vec3 sample error
by Kamil Zubair 02 Nov '10

02 Nov '10

I found a strange error when I try the vec3 sample from proto, my code is like this: struct Vec3SubscriptCtx : proto::callable_context< Vec3SubscriptCtx const > { typedef int result_type; Vec3SubscriptCtx(int i) : i_(i) {} // Index array terminals with our subscript. Everything // else will be handled by the default evaluation context. int operator ()(proto::tag::terminal, int const (&arr)[3]) const { return arr[this->i_]; } int i_; }; struct Vec3 : proto::extends<proto::terminal<int[3]>::type, Vec3> { explicit Vec3(int i=0, int j=0, int k=0) { (*this)[0] = i; (*this)[1] = j; (*this)[2] = k; } int &operator [](int i) { return proto::value(*this)[i]; } int const &operator [](int i) const { return proto::value(*this)[i]; } // Here we define a operator = for Vec3 terminals that // takes a Vec3 expression. template< typename Expr > Vec3 &operator =(Expr const & expr) { typedef Vec3SubscriptCtx const CVec3SubscriptCtx; (*this)[0] = proto::eval(proto::as_expr(expr), CVec3SubscriptCtx(0)); (*this)[1] = proto::eval(proto::as_expr(expr), CVec3SubscriptCtx(1)); (*this)[2] = proto::eval(proto::as_expr(expr), CVec3SubscriptCtx(2)); return *this; } template< typename Expr > Vec3 (Expr const & expr) { typedef Vec3SubscriptCtx const CVec3SubscriptCtx; (*this)[0] = proto::eval(proto::as_expr(expr), CVec3SubscriptCtx(0)); (*this)[1] = proto::eval(proto::as_expr(expr), CVec3SubscriptCtx(1)); (*this)[2] = proto::eval(proto::as_expr(expr), CVec3SubscriptCtx(2)); } void print() const { std::cout << '{' << (*this)[0] << ", " << (*this)[1] << ", " << (*this)[2] << '}' << std::endl; } }; int main(int argc, const char* argv[]) { Vec3 a(1, 2, 3); Vec3 b(3, 2, 1); Vec3 c; c = b; c.print(); return 0; } I expected that when c is printed it will display 3, 2, 1 ( the value of b ) but instead it print 0,0,0 like b is never assigned to c. I try this using vs 2010 with boost 1.44 in windows 7 64.

2 1

[serialization] Class version in binary archive
by Guy Prémont 02 Nov '10

02 Nov '10

Hi all, I have a whole hierarchy of classes that is serialized without problem using XML archive. When I wanted to use binary archive (that should be faster and smaller), i met an little problem. A bit of history, to make it easier to spot related changes in class versions, we decided not to go with serial class version numbers (1, 2, 3...) but with meaningful version numbers: internal branch number + date. We end up with class versions like 420100714 or 520101102. In XML or text, no problem at all. In binary, it took me a while to find out that class version numbers are serialized as unsigned char! However, the documentation here <http://www.boost.org/doc/libs/1_44_0/libs/serialization/doc/archive_referen ce.html> says unsigned int. It is a bit surprising, since the type passed to the serialization method is int. I understand that I need to create my own binary archive to implement save_override and load_override for version_type. However, doing the override means that i am preventing the use of standard boost binary archive. Is that a known bug, or is it even considered a bug? Thanks -- Guy Prémont, D.Sc. Architecte logiciel senior / Senior software architect CM Labs Simulations Inc. <blocked::http://www.cm-labs.com/> http://www.cm-labs.com/ Tel. 514-287-1166 ext. 237

1 0

Re: [Boost-users] [1.44 MSM meta state machine] accessing objects fromactions
by Christophe Henry 02 Nov '10

02 Nov '10

> Currently my queue object is a member of the state machine itself, so I > can write > > struct Send_ : public msm::front::state_machine_def<Send_> > { > string m_next_line_to_send; > socket m_socket; > string get_next_send_line(); > }; > > and in my a_send action > void operator()(Event &const &evt, Fsm &fsm, S&, T&) > { > fsm.m_next_line_to_send=fsm.get_next_send_line(); > fsm.m_socket.async_write(fsm.m_next_line_to_send); > } > > This works well as long as I am not using sub-statemachines. > > Here is my problem: > I would like to use a substate machine for the actual sending. The main > state machine has the states Disconnected, Working, Error, Shutdown. The > Working state should actually be a substate machine. > > The problem is, when a_send gets called by the substate machine Working, > the fsm object is of type Working. Correct. There is no "fsm call stack" yet. This gets complicated if you have a submachine of a submachine of ... For the moment I don't have this. > How can I access the member variables of the parent of a substate > machine? Or is there a better way of accessing the objects the state > machine should work with from actions? For the moment, all we can do is adding a pointer to the main fsm's data in the submachine. It is less elegant with the 1.44 than the coming 1.45: With 1.44 (for example inside the on_entry method of Send_): Send_::Working& w = fsm.get_state<Send_::Working&>(); w.data = &fsm.data; // w.data is a pointer True, this is not really nice. With the upcoming 1.45 (or trunk for the moment), you could use the new constructors: Send myfsm (msm::back::states_ << Working(data) ); //data being externally defined and Working_ having a corresponding constructor But this won't work for your case where data is contained in Send_ (and therefore does not exist at the time of construct). You can however reuse the same syntax with the new set_states method: Replace the version with get_state by: fsm.set_states(msm::back::states_ << Working(fsm.data)); //data being defined in Send_ and Working_ having a corresponding constructor. The whole explanation is there (http://svn.boost.org/svn/boost/trunk/libs/msm/doc/HTML/ch03s05.html#backend…) > The tutorial explains state machines very nicely with the example of a > CD player. But all the actions are empty and nowhere is explained, how > you would actually access a CD player member variable or other such > object so that you could actually manipulate the player to do something. Seems that I need to improve my doc here. It's too late for the 1.45 but I'll do it for the 1.46. I agree that this is still not perfect and it's on my list but the list is still pretty long so I hope this solution will be ok for the time being. Regards, Christophe

2 3

iostreams zip header
by Kraus Philipp 02 Nov '10

02 Nov '10

Hi, I'm using the gzip and bzip2 compression for compress data. That works fine, but can I remove the zip header in the output stream? The compressor is pushed into a filtering_streambuf and output stream, but I don't need the the zip header (only the compressed data). Is there any method to remove the zip header or should I write my own code at the gzip / bzip2 specifiation? Thanks Phil

1 0