April 2008 - Boost-users - lists.preview.boost.org

[python] Expose memory-aligned data types?
by Dan Fike 22 Apr '08

22 Apr '08

Using Boost.Python, is it possible to expose a C++ memory-aligned data type to Python? For example, I want to expose this data type. class __declspec(align(16)) foo { public: foo() {}; bool isAligned() { if ((int)this % 16 == 0) { return true; } else { return false; } } private: int a, b, c, d; }; I've written the following code to expose this class to Python: BOOST_PYTHON_MODULE(myModule) { boost::python::class_<foo>("foo", boost::python::init<>()) .def("isAligned", &bar.isAligned) ; } When compiling this code (MSVC 8.0, VS2005, Win32), I get the following error (entire error listed at the bottom): d:\...\boost\python\converter\as_to_python_function.hpp(21) : error C2719: 'unnamed-parameter': formal parameter with __declspec(align('16')) won't be aligned Note that I used to get a STATIC_ASSERTION_FAILURE in the type_with_alignment class definition when compiling with Boost 1.34. I found adding a 16-byte aligned struct type to the BOOST_TT_ALIGNMENT_BASE_TYPES list in type_with_alignment.hpp made the ASSERT go away, but introduced the above error. In an attempt to fix it, I downloaded Boost revision 43984 from early April, 2008, which included a patch that I thought would help me. See http://svn.boost.org/trac/boost/changeset/43984. While the type_with_alignment.hpp included in that revision did correct the STATIC_ASSERTION_FAILURE, I am still hitting the above error. Now, if I try to build the same code in x64 instead of Win32 (again, from revision 43984), I do not encounter the error. Not only does it compile, but I have verified that any objects 16-byte aligned objects I instanciate in Python are actually aligned correctly. Unsurprisingly, however, the x64 build produces the same error when I try to expose a 32-byte aligned data type. On the other hand, the Win32 build still complains if I try it with an 8-byte aligned data type. I'm hoping there is a piece of Boost.Python code missing from the class exposition that I don't know about. I've scoured the documentation and internet (and code) for such a thing, but I may have overlooked it. Any advice? -Dan Fike Here is the entire error, as printed by VS2005... d:\...\boost\python\converter\as_to_python_function.hpp(21) : error C2719: 'unnamed-parameter': formal parameter with __declspec(align('16')) won't be aligned d:\...\boost\python\to_python_converter.hpp(88) : see reference to class template instantiation 'boost::python::converter::as_to_python_function' being compiled with [ T=foo, ToPython=boost::python::objects::class_cref_wrapper>> ] d:\...\boost\python\to_python_converter.hpp(82) : while compiling class template member function 'boost::python::to_python_converter::to_python_converter(void)' with [ T=foo, Conversion=boost::python::objects::class_cref_wrapper>>, has_get_pytype=true ] d:\...\boost\python\object\class_wrapper.hpp(26) : see reference to class template instantiation 'boost::python::to_python_converter' being compiled with [ T=foo, Conversion=boost::python::objects::class_cref_wrapper>>, has_get_pytype=true ] d:\...\boost\python\object\class_metadata.hpp(272) : see reference to class template instantiation 'boost::python::objects::class_cref_wrapper' being compiled with [ Src=foo, MakeInstance=boost::python::objects::make_instance> ] d:\...\boost\python\object\class_metadata.hpp(229) : see reference to function template instantiation 'void boost::python::objects::class_metadata::maybe_register_class_to_python(T2 *,boost::mpl::false_)' being compiled with [ T=foo, X1=boost::python::detail::not_specified, X2=boost::python::detail::not_specified, X3=boost::python::detail::not_specified, T2=foo ] d:\...\boost\python\object\class_metadata.hpp(219) : see reference to function template instantiation 'void boost::python::objects::class_metadata::register_aux2>(T2 *,Callback)' being compiled with [ T=foo, X1=boost::python::detail::not_specified, X2=boost::python::detail::not_specified, X3=boost::python::detail::not_specified, T2=foo, Callback=boost::integral_constant ] d:\...\boost\python\object\class_metadata.hpp(217) : while compiling class template member function 'void boost::python::objects::class_metadata::register_aux(void *)' with [ T=foo, X1=boost::python::detail::not_specified, X2=boost::python::detail::not_specified, X3=boost::python::detail::not_specified ] d:\...\boost\python\class.hpp(174) : see reference to class template instantiation 'boost::python::objects::class_metadata' being compiled with [ T=foo, X1=boost::python::detail::not_specified, X2=boost::python::detail::not_specified, X3=boost::python::detail::not_specified ] d:\...\boost\python\class.hpp(205) : see reference to class template instantiation 'boost::python::class_::id_vector' being compiled with [ W=foo ] d:\...\MyProject\my_project_python_modules.h(77) : see reference to function template instantiation 'boost::python::class_::class_>(const char *,const boost::python::init_base &)' being compiled with [ W=foo, DerivedT=boost::python::init<> ]

1 1

[iterators] NULL pointer initialization and comparison using iterator_facade
by Lang Stefan 22 Apr '08

22 Apr '08

Hello list, I have created a generic iterator template class for arrays that also doubles as container-wrapper for the array in question. To achieve this I have based my class on boost::iterator_facade . I have defined all the neccessary functions as described in the documentation and mostly the iterators work as intended. Here's a cut-down version: #include <boost/type_traits/is_convertible.hpp> #include <boost/utility/enable_if.hpp> #include <boost/iterator/iterator_facade.hpp> template<class Value> class cArrayIterator : public boost::iterator_facade< cArrayIterator<Value> , Value , boost::random_access_traversal_tag > { private: struct enabler {}; public: cArrayIterator() : first(0),size(0),position(0) {} cArrayIterator(Value* _first, difference_type _size, difference_type _position = 0) : first(_first),size(_size),position(_position) {} template<class OtherValue> cArrayIterator( const cArrayIterator<OtherValue>& other , typename boost::enable_if< boost::is_convertible<OtherValue*,Value*> , enabler >::type = enabler() ) : first(other.first),size(other.size),position(other.position) {} private: Value* first; difference_type position; difference_type size; friend class boost::iterator_core_access; void advance(difference_type n){ position += n; } void decrement(){ --position; } Value& dereference() const{ return first[position]; } template<class OtherValue> difference_type distance_to(const cArrayIterator<OtherValue>& z) const{ return z.position - this->position; } template<class OtherValue> bool equal(const cArrayIterator<OtherValue>& other) const{ return this->first + this->position == other.first + other.position; } void increment(){ ++position; } }; I have a big problem though: The project I am working on is quite big and I intend to replace a particular typedef (currently simply a pointer to a struct) with this iterator class. There are two kinds of instructions though that my iterator class currently can't cope with: 1. Initializing a 'pointer' with NULL (in other words, I need a constructor with one argument of type int, but only in case that argument's value is 0) 2. Comparing a 'pointer' to NULL (again, comparison to int argument, but only makes sense for the value 0) I did add a constructor with only one argument (of type pointer to the underlying object's type) and could eliminate problem 1. However I feel it's not quite right, since for my array iterator it doesn't make any sense to initialize it with just one pointer and no further information (it requires pointer to first element, position, and size of array), so in cases where the pointer argument is not NULL this constructor shouldn't even be available! Here's the code for the new constructor: cArrayIterator(Value* _first) // workaround : first(_first),position(0) { if (NULL == _first) size = 0; else size = 1; // not really sensible, but the best I can think of atm } Moreover, I hoped this constructor would also eliminate problem 2 by implicitely converting the int value of 0 to a NULL iterator, but it didn't. I tried to provide my own comparison operators, but so far failed to meet the syntax the compiler would recognize as a fit for a comparison int to iterator. This is what I got so far: template <class Value> bool operator==(cArrayIterator<const Value> const left, cArrayIterator<Value> const right) { return false;// I'll take care of that once the function gets actually called :/ } How do I correctly define a comparison operator that compares a 'NULL pointer' to a class representing a pointer? Or does iterator_facade already provide a way to deal with this? Stefan

2 1

[thread] how to use thread::timed_join
by Johan Råde 22 Apr '08

22 Apr '08

I want to do a timed join, with a relative time of 0.1s. I can not figure out from the docs how to do this. I find the prototypes somewhat mystifying: bool timed_join(const system_time& wait_until); template<typename TimeDuration> bool timed_join(TimeDuration const& rel_time); What is system_time? What should TimeDuration be? I'm using Windows. Thank you, Johan Råde

5 7

Edges list
by Chartier Nicolas 22 Apr '08

22 Apr '08

I'd like to know how I could retrieve a weighted edges list from a graph, without forgetting vertices and weight, in a list for example. Any suggestion welcome ;) NC ------- Chartier Nicolas +33 (0)6 78 32 94 33

1 0

Re: [Boost-users] Mingw cross compilation
by James Choa 22 Apr '08

22 Apr '08

> Hi James, > > I managed to cross-compile static release builds of serialization, > filesystem, system, date_time, and program_options from boost 1.35.0 > with i386-mingw32 gcc 4.3.0 (http://mingw-cross.sourceforge.net/) under > Fedora 8 using a three-step procedure: > - create a user-config.jam file with content: > using gcc : : i386-mingw32-gcc : ; > - run bjam (assuming you have compiled it already): > bjam --user-config=user-config.jam --toolset=gcc link=static release ... > - manually run i386-mingw32-ranlib on the generated libraries > > There might be a better way to do this, but at least the above worked > for me. > /Mikko Thanks! I'll give it a try

1 1

[iostreams] Obfuscator
by Robert Bielik 22 Apr '08

22 Apr '08

Hi all, I'm trying to make filter to obfuscate some data so its not that obvious what it is. Problem is (I think) that in some cases, the obfuscator produces a null character as its output, and I'm not sure, but it seems that upon reading back the data, the null character is interpreted as EOF. At least, that is what I get from boost::iostreams::get(src) at the position where the null character was in the output. Can I somehow tell the stream that it is binary and not care about null characters? TIA /Rob

2 2

Re: [Boost-users] Applying Boost license to non boost projects
by Robert Marion 22 Apr '08

22 Apr '08

You may wish to consider applying the liberal BSD (http://www.opensource.org/licenses/bsd-license.php) or MIT (http://www.opensource.org/licenses/mit-license.php) licenses. These are both in widespread use for open source, and impose no restrictions, aside from attribution, on the user. Putting the license text at the top of all headers is not strictly necessary, but not a bad idea. Robert Marion http://www.palamida.com > I am interested in adopting the Boost Software license for an open source C++ > library (Castor: http://www.mpprogramming.com/2.html) > > [ snip ] > - Does this same guideline apply to non boost projects ? The need for stating > Boost.org there might confuse some users of non boost prj to think that the > library is supplied by Boost. > - Do all header files in my prj (which is a pure header library) need to > include this text at the top/somewhere in order to adopt the license. > > -Roshan

2 1

[BGL] implicit astar_search weirdness, bug in mutable_queue?
by Oliver Seiler 21 Apr '08

21 Apr '08

Hi there -- long-time reader, first-time poster! I've been trying to use astar_search from the graph library for a little problem I'm working on. Specifically, I wanted to use it to perform a search on an implicit graph (more specifically, an implicit tree), as touched on briefly in the docs for astar_search. In this case, I'm using boost 1.33.1 Unfortunately, this doesn't appear to be well supported "out of the box". Google didn't turn up much on the subject, except for a paper done by Kris Beevers and Jufeng Peng (www.cs.rpi.edu/~beevek/research/astar_bgl04.pdf) which mentions the need for a non-const version of the breadth-first search code in BGL, and a message to the list from Ian Downes, Feb 28th, asking about doing an implicit search as well (couldn't find any followups). Using Beevers paper as a template, I started bumping into some serious memory corruptions problems with my test-cases. Finally today I ran the tests under valgrind and managed to track it down to mutable_queue (boost/pending/mutable_queue.hpp), which is used by the astar_search code to manage the set of partial solutions. mutable_queue is constructed with a vector of integers, index_array, sized to the initial size of the implicit graph (1 vertex in my case). Unfortunately, this vector is never resized, so calls to mutable_queue::push wind up walking off the end of the index_array vector as the graph gets expanded (by my astar_visitor::examine_vertex callback), with various undefined (but generally bad) side-effects. So, after some mucking around, I've wound up with a solution to this where I specialize mutable_queue for my graph's vertex type, leaving astar_search and the standard mutable_queue definitions alone. The specialization is basically a copy of mutable_queue.hpp, with a change to the push method to ensure that index_array is sized correctly before indexing into it. Problem solved, everything working like a charm. Now, is this a bug in mutable_queue? Reading the two-paragraph description in mutable_queue.hpp, I'd guess this is how it is supposed to act. This would mean that implicit search with astar_search (and presumably any other search) isn't supported, because the queue that it uses for partial results can't get bigger as the graph itself grows. The documentation probably shouldn't mention implicit search if this is the case, or maybe there is need for a non-const version of astar_search that can use a more appropriate queue type... My solution, though it seems to work fine, is a bit of a bodge... As a side-note, I initially thought that the memory corruption that I was seeing was a result of vertex iterators becoming invalidated as the graph was expanded. I attempted a switch over to using listS for the VertexList parameter of the adjacency_list template (from vecS), but this causes all sorts of compiler errors under gcc 4.1.2. I tried making a similar change to the astar_search example program (astar-cities.cpp) and this too fails to compile (with 1.33.1 and 1.34.1). Shouldn't listS and vecS for the VertexList parameter be interchangeable? Cheers, Oliver

1 1

Installed library names
by Yang Zhang 21 Apr '08

21 Apr '08

Hi, I'm working with a project whose build system passes -lboost_serialization as a flag to gcc. As far as I can tell from the configure options and from the Getting Started documentation, boost's installer only ever uses fully qualified names (libboost_serialization-gcc41-mt-1_35.so.1.35.0 on my system). Could the installer have possibly provided a libboost_serialization.so as well? Or did the authors of the project I'm trying to build probably manually create symlinks (or have their package manager create these) and assume that I would have the same? Thanks!

2 1

Re: [Boost-users] [multi_index] custom template class for handle multi_index map
by damny＠web.de 21 Apr '08

21 Apr '08

hi joaquin, you are a genius :o) the absent default copy ctor and assignment operator were the reason for my problem ... can't see the wood for the trees. thank you daniel _________________________________________________________________________ In 5 Schritten zur eigenen Homepage. Jetzt Domain sichern und gestalten! Nur 3,99 EUR/Monat! http://www.maildomain.web.de/?mc=021114

1 0