June 2011 - Boost-users - lists.preview.boost.org

read_graphviz with a simple graph
by David Doria 28 Jun '11

28 Jun '11

If I have a graph with no vertex properties or edge properties: adjacency_list<vecS, vecS, undirectedS> can I read a .dot file into it with read_graphviz? I tried: dynamic_properties dp; Graph g; read_graphviz(fileinput, g, dp); but it throws an exception "Property not found: node_id". I tried to add property_map<Graph, vertex_name_t>::type name = get(vertex_name, g); dp.property("node_id",name); but I get a compiler error "No match for operator<<" Any suggestions? Thanks, David

2 14

Re: [Boost-users] boost::asio and C++0x
by Marat Abrarov 28 Jun '11

28 Jun '11

> It appears that gcc up to 4.5.2 did work; the bug was related to bleeding-edge gcc. > > I am still looking for solutions to this issue. I have successfully built my app with -std=c++0x and > boost::thread, but boost::asio gives me errors. At http://sourceforge.net/projects/asio-samples/ I use qmake to build projects under Linux. I build asio-samples with Boost 1.46.0 + Boost.Asio 1.5.3 + gcc 4.6.0. Feel free to watch at *.pro files describing qmake build dependencies at /build/qmake. Regards, Marat Abrarov.

2 1

boost::asio and C++0x
by Wilde, Donald S 28 Jun '11

28 Jun '11

Hi, All - I am having some challenges mixing boost and c++0x. My configuration: Ubuntu 10.04, gcc 4.4.3, boost 1.46.1 I've updated the libraries and headers from Ubuntu's stock boost 1.40 level, and I've tried bjam with and without 'toolset=gcc cxxflags=-std=c++0x' Either way, my Eclipse reports several errors in the external libraries: Undefined references to boost::system::generic_category() Undefined references to boost::system::system_category() (both from system and asio error headers) Undefined references to pthread_key_create and pthread_key_delete from asio/detail headers { apologies for not giving specific error details but it'd have to come from a screen shot and that's not appreciated on a mailing list } I do have access to gcc 4.5.1, but it's a custom one for our embedded Atom-based SoC. That is the eventual target, but I haven't seen how to use a specific g++ binary yet. I did see a bug tracker post in one of my google searches but I haven't been able to get back to it. I understand that this is a known issue and a moving target but I would appreciate it if someone would summarize the current state for me. Thanks in advance! -- Don Wilde

3 6

Multi_array and rotated views
by petros 28 Jun '11

28 Jun '11

Hi, I am coming back to the list with a question on multi_array views. Some time ago, I asked if I could have a view that would offer a “rotated” multi_array. That is, if I have a 3D MA (=multi_array) of dimensions (shape) {L,M,N}), I would like to obtain a reference to it which would look like {M,N,L} i.e. cyclically rotated (I would actually wish to be able to perform this rotation many times). The operation should be performed at compile time (no need/wish to do it at run time and pay the possible overhead in speed). A couple of friends suggested that I used a storage structure something like this (files are attached for convenience of inspection). namespace boost { class cyclic_storage_order { typedef detail::multi_array::size_type size_type; public: cyclic_storage_order( const std::size_t shift ) :shift_(shift) {} template <std::size_t NumDims> operator general_storage_order<NumDims>() const { boost::array<size_type,NumDims> ordering; boost::array<bool,NumDims> ascending; for (size_type i=0; i != NumDims; ++i) { ordering[i] = (i+shift_)%NumDims; ascending[i] = true; } return general_storage_order<NumDims>(ordering.begin(), ascending.begin()); } private: std::size_t shift_; }; } Now, I am trying to use this is the following example (again the file is attached): #define _SCL_SECURE_NO_WARNINGS #include <iostream> using std::cout; using std::endl; #include <boost/multi_array.hpp> using boost::multi_array; using boost::multi_array_ref; using boost::const_multi_array_ref; #include "MultiArrayRotation.h" int main(){ const unsigned short dimensionality = 3; typedef multi_array< int, dimensionality > IntArray; typedef IntArray::size_type size_type; typedef const_multi_array_ref< int, dimensionality > IntArrayCRef; boost::array<size_type,dimensionality> sizes = { { 2,3,4 } }; IntArray a( sizes ); boost::cyclic_storage_order cyclic_storage(1); boost::array<size_type,dimensionality> sizesRotated = { { 3, 4, 2 } }; IntArrayCRef rotated( a.data(), sizesRotated, cyclic_storage ); //initialize: for ( size_t i = 0 ; i != sizes[0]; ++i ) for ( size_t j = 0 ; j != sizes[1]; ++j ) for ( size_t k = 0 ; k != sizes[2]; ++k ) a[i][j][k] = (i+1) + 10 * (j+1) + 100*(k+1); typedef IntArray::const_iterator iterator3; typedef boost::subarray_gen<IntArray,2>::type::const_iterator iterator2; typedef boost::subarray_gen<IntArray,1>::type::const_iterator iterator1; for ( iterator3 it3 = a.begin(); it3 != a.end(); ++ it3 ) { cout << "in first iterator:"; boost::detail::multi_array::const_sub_array<int,2> & d2 = *it3; cout << (d2.shape())[0] << "x" << (d2.shape())[1] << endl; for ( iterator2 it2 = it3->begin(); it2 != it3->end(); ++ it2 ) { // cout << "\t" << it2->shape()[0] << endl; for ( iterator1 it1 = it2->begin(); it1 != it2->end(); ++ it1 ) cout << "\t\t" << (*it1) << endl; } } //for a rotated system: typedef IntArrayCRef::const_iterator iterator3CRef; typedef boost::subarray_gen<IntArrayCRef,2>::type::const_iterator iterator2CRef; typedef boost::subarray_gen<IntArrayCRef,1>::type::const_iterator iterator1CRef; cout << endl << "In Rotated System" << endl; for ( iterator3CRef it3 = rotated.begin(); it3 != rotated.end(); ++ it3 ) { cout << "in first iterator:"; //boost::const_multi_array_ref::const_sub_array<int,2> & d2 = *it3; cout << (it3->shape())[0] << "x" << (it3->shape())[1] << endl; for ( iterator2CRef it2 = it3->begin(); it2 != it3->end(); ++ it2 ) { // cout << "\t" << it2->shape()[0] << endl; for ( iterator1CRef it1 = it2->begin(); it1 != it2->end(); ++ it1 ) cout << "\t\t" << (*it1) << endl; } } return 1; } This gives me undesired results, though. What I have done here, is define a 2x3x4 main array with entries like abc (i.e. 3 digits). >From the value of abc I can infer the coordinates of the element in the main array. Following the cascade of iterators to sub_arrays, I see that, in the end, I obtain for each value of the first coordinate, the corresponding slices. Then I do a rotation (once) and descend down the iterators. I would expect now 3 slices (instead of 2, before), organized so that each one has the same middle digit (in the abc “representation”. This, however, is not happenning! I attach the code which I have built successfully w/msvc2010 on win7. I use boost1.45. A friend (Larry Evans) , who had offered a lot of help in the past and whom I contacted for suggestion let me know that, the program did not compile in his environment (gcc 4.6.0 and boost v1.46). Any suggestion, help will be greatly appreciated. Kind Regards, Petros

2 3

filesystem hidden files
by Kraus Philipp 28 Jun '11

28 Jun '11

Hello, I use the Boost Filesystem to browser the data within a directory. On unix systems there are file with a dot on begin, the hidden files. Is there a method for checking the path variable if the file is hidden) like path p("~/.myhiddenfile"); if (is_regular_file(p) && is_hidden(p)) I would like to exclude hidden files of my browsing list. Thanks Phil

1 0

making boost build generic library names (without -mgw45-mt-1_46_1)?
by Ákos Maróy 28 Jun '11

28 Jun '11

Hi, I wonder how can one make the boost build process create libraries with generic boost library names (like libboost_math_c99f.a) on Windows using mingw / gcc, as opposed to library names like libboost_math_c99f-mgw45-mt-1_46_1.a? Although I understand this makes sense in some contexts, a lot of times this is not convenient. For example, I'd have to update my dependencies every time I change the boost version, or the underlying gcc version changes. moreover, on other platforms (Linux, MacOS), the libraries built have the generic name (libboost_math_c99f.a) How to achieve the same effect on Windows? Akos

2 2

Does Boost use a platform-configurable [[ noreturn]] attribute ?
by John M. Dlugosz 27 Jun '11

27 Jun '11

Does the Boost 1.46 library have a macro or something that I can use to decorate a function as "noreturn" in a platform independent manner? Is there documentation on platform-independence tools that I might use in general, as Boost itself uses them? —John

1 0

Boost.Format to a string
by John M. Dlugosz 27 Jun '11

27 Jun '11

I brought this up a while back, and I ran into it again yesterday. Using boost::format to produce a string, as opposed to sending it directly to cout, is awkward. boost::format ("blah blah blah") % exp1 % exp2 %exp3 is fine in and of itself. But then what? You need to wrap the whole thing in extra parens to use str: std::string s= (boost::format ("blah blah blah") % exp1 % exp2 %exp3).str(); WHY NOT allow implicit conversion to string? Or, some other mechanism that doesn't complicate the expression and bury the interesting part? —John

2 2

[msm] how to write a generic class for all state machine back-end.
by Jerry Chen 27 Jun '11

27 Jun '11

In our project, we need a generic interface to refer to different state machine (back-end). For example, in the following codes, we want to use SMBase* to refer to any concrete state machine. SMBase should have similar interface as state machine. struct User { A(SMBase* s) : sm(s) (); SMBase* sm; ... // User can use sm->start(), sm->process_event() etc } struct SMBase { virtual void start() = 0; ... // how to design the rest ??? } // a wrapper implement SMBase template < typename SM > class SMWrapper : public SMBase{ SMWrapper(SM & s): sm(s){}; virtual void start() { sm.start(); } ... SM sm; } main() { typedef boost::msm::back::state_machine SM1; typedef boost::msm::back::state_machine SM2; User a1(new SMWrapper(SM1()) ); // Wrap the concrete state machine and pass it to User. User a2(new SMWrapper(SM2()) ); // Pass a different state machine. ... } The problem of designing SMBase is the function "process_event". We want it to to be virtual function so SMWrapper can implement it. But it also needs to be a template according to state machine interface. But we cannot have a virtual template function. How can we achieve/design this generic state machine interface? Many thanks. Jerry

4 4

Re: [Boost-users] [boost] [review][assign] Formal review of Assign v2 ends today
by Lars Viklund 27 Jun '11

27 Jun '11

On Fri, Jun 24, 2011 at 11:56:14PM +0100, John Bytheway wrote: > The formal review of the Assign v2 library upgrade by Erwann Rogard is > scheduled to end today. > > Despite the end of the review, please feel free to discuss the library > further, make any final comments, or even submit a late review. I will > certainly not declare the review request before next Monday, and will > incorporate any comments made before then. Hopefully this is not too late to be counted as a proper mini-review. The test suite for the library fails miserably on VisualAge C++ 11.1, as can be seen in the logs at [1], which doesn't bode well for a newly written library aimed for inclusion in Boost. The test suite fails similarly on SunStudio 12.1 on Solaris 10, with the results seen in [2]. As a comparison, the tests for assign_v1 passes with flying colours with VisualAge. Having the replacement library support way fewer platforms in Boost sounds very strange to me, and is alone grounds for not including the library. (I was not able to complete building the v1 tests on the Solaris box I had, as the linker seems slightly broken, something that doesn't affect the v2 tests, as none of them can compile.) The performance metrics seem to lack a very important metric - namely compile-time costs in memory usage and compile time increases. The documentation mentions compile-time costs briefly in lamenting that the test suite takes several minutes to build, which isn't horribly encouraging. A library that is intended to be easily dropped into any TU you might need it in should be very cheap to use. I currently avoid using Karma/Qi even for things where they would simplify things, solely due to the cost of including them. Regarding the code, I'm a bit concerned about the namespace layout. The current setup will cause any users of 1.0 that also happen to include 2.0 indirectly to end up with namespaces 'v2' and 'ref' injected wherever they previously did 'using namespace boost::assign;'. As for naming of the functions and types, I find them both opaque and overly generic. 'put()' has a strong risk of being used already in user code, and things like 'csv()' are utterly meaningless without having read the documentation, especially with regard to the non-type template argument. The use of asdf_aux namespaces seems a bit off, considering that the canonical naming for implementation detail is just that, 'detail'. This library feels misguided, trying to be something the name doesn't indicate it should be. I found assign_v1 to be an excellent library for quick and dirty initialization of sequences. The whole transformation pipeline that v2 introduces feels out of place and most probably belongs in a quite different library. The documentation even mentions Boost.Range's sequence adaptors - which makes me wonder if they're not better off integrated with Range? As for the tutorial, which is the closest thing to examples I find, they seem overly terse and seems to assume that the concepts used in the library are already known. My vote for this library is: NO. [1] http://www.acc.umu.se/~zao/assign_v2/vacpp-test-20110627-1014.txt [2] http://www.acc.umu.se/~zao/assign_v2/suncc-test-20110627-1044.txt -- Lars Viklund | zao(a)acc.umu.se

2 2