December 2006 - Boost - lists.boost.org

Review request: Boost.FSM
by Andrey Semashev 03 Jan '07

03 Jan '07

Hi, I would like to ask for a formal review of the Boost.FSM library proposed here not long ago. I've uploaded the implementation, test and documentation into Boost.Vault, it is accessible with this link: http://tinyurl.com/yjozfn There were several changes in the library since the proposal: - Integral constants tagged events support added - Transitions maps support improved. Each transition may now decide in run-time to which state to transit. The transition map itself may be broken into several pieces - the common part and state-specific parts. - Added an implementation of thread-locking state machine. It is quite equivalent to the "state_machine" class but adds a tiny synchronization layer between environment and the machine. - The library has been split into several headers and put into its own directory. - The documentation have been improved. -- Best regards, Andrey mailto:andysem@mail.ru PS: Happy New Year everyone!

4 4

Naming of Boost.Graph binary
by David Abrahams 03 Jan '07

03 Jan '07

The binary for Boost.Graph is named differently from all the other library binaries; in particular it is called libbgl-viz-xxx rather than libboost_graph-xxx. This is going to complicate the Getting Started Guide, and although I understand the desire to make it clear that the library is only needed for use with GraphViz, I wonder if this nonuniformity is really worth the price. Can we change the name? We can always install aliases on *nix to smooth the transition. -- Dave Abrahams Boost Consulting www.boost-consulting.com

2 2

Weird behaviour/bug in ptr_map
by Joshua Moore-Oliva 03 Jan '07

03 Jan '07

I was doing what I do regularly with std::maps... a m_plsts.erase( m_plsts.begin(), m_plsts.end() ); on a boost::ptr_map. It failed with an assertion error... Digging around, it came from ptr_container/detail/associative_ptr_container.hpp on line 117. It appears that every single erase (including that done by a key) has an assert for the list not being empty! Why is this the case? As far as I understand, it is valid to attempt to remove a key that does not exist. It follows that the same behaviour would occur whether the map was empty or not. Attempting to remove an element from an empty map should not result in an abort. If I'm unaware of some C++ convention I would love to be informed, however this definitely screams bug to me. Joshua Moore-Oliva

4 10

cw trouble in HEAD
by Gennadiy Rozental 03 Jan '07

03 Jan '07

I am having problem compiling Boost.Test with CW toolset Here an example error message: ### mwcc Compiler: # In: ..\..\..\boost\type_traits\is_member_pointer.hpp # From: ..\src\compiler_log_formatter.cpp # ------------------------------------------ # 50: template< typename T, typename U > struct is_member_pointer< U T::*const > : ::boost::integral_constant<bool,true> { }; # Error: ^ # ',' expected # (included from: # is_pointer.hpp:24 # call_traits.hpp:27 # call_traits.hpp:21 # class_properties.hpp:26 # unit_test_suite_impl.hpp:21 # compiler_log_formatter.ipp:20 # compiler_log_formatter.cpp:16) ### mwcc Compiler: # 51: template< typename T, typename U > struct is_member_pointer< U T::*volatile > : ::boost::integral_constant<bool,true> { }; # Error: ^ # undefined identifier 'is_member_pointer' # (included from: # is_pointer.hpp:24 # call_traits.hpp:27 # call_traits.hpp:21 # class_properties.hpp:26 # unit_test_suite_impl.hpp:21 # compiler_log_formatter.ipp:20 # compiler_log_formatter.cpp:16) Any pointers, Gennadiy

2 1

[BGL] make_filtered_graph() should accept const Graph&
by Vladimir Sykora 02 Jan '07

02 Jan '07

Hi, Can anyone correct the make_filtered_graph() function on boost/graph/filtered_graph.hpp to accept a const Graph& ? Thanks, --vjs

2 2

trivial use of threads AND crt mem check exits with memory leak
by Tom Titchener 02 Jan '07

02 Jan '07

I always get the memory leak trace at the end of this post when I run a trivial application that uses boost threads (source below). I'm compiling with multi-threaded debug dll (/MDd). I'm building a console app. It's on win32/xpsp2 with vc8. I haven't found any thread shutdown code to run in the boost threads documentation. Have I missed another compiler or linker option? I'd really like to use leak detection so I know I'm not losing memory in my own code. Thanks, Tom Titchener #define WIN32_LEAN_AND_MEAN #include <crtdbg.h> #include <stdio.h> #include <iostream> #include <boost/thread.hpp> using namespace std; void CRTSetup(void) { _CrtSetDbgFlag(_CRTDBG_ALLOC_MEM_DF | _CrtSetDbgFlag(_CRTDBG_REPORT_FLAG)); _CrtSetDbgFlag(_CRTDBG_LEAK_CHECK_DF | _CrtSetDbgFlag(_CRTDBG_REPORT_FLAG)); _CrtSetDbgFlag(_CRTDBG_DELAY_FREE_MEM_DF | _CrtSetDbgFlag(_CRTDBG_REPORT_FLAG)); _CrtSetDbgFlag(_CRTDBG_CHECK_ALWAYS_DF | _CrtSetDbgFlag(_CRTDBG_REPORT_FLAG)); _CrtSetReportMode( _CRT_WARN, _CRTDBG_MODE_FILE ); _CrtSetReportFile( _CRT_WARN, _CRTDBG_FILE_STDOUT ); _CrtSetReportMode( _CRT_ERROR, _CRTDBG_MODE_FILE ); _CrtSetReportFile( _CRT_ERROR, _CRTDBG_FILE_STDOUT ); _CrtSetReportMode( _CRT_ASSERT, _CRTDBG_MODE_FILE ); _CrtSetReportFile( _CRT_ASSERT, _CRTDBG_FILE_STDOUT ); } void dummy(void) { cerr << "dummy" << endl; } void main() { CRTSetup(); boost::thread t(dummy); t.join(); } dummy Detected memory leaks! Dumping objects -> {145} normal block at 0x003575D8, 24 bytes long. Data: < < > 00 3C 15 00 FF FF FF FF 00 00 00 00 00 00 00 00 {144} normal block at 0x003575A0, 8 bytes long. Data: < u5 > D8 75 35 00 01 CD CD CD Object dump complete.

3 3

Weird behaviour/bug in ptr_map
by Joshua Moore-Oliva 02 Jan '07

02 Jan '07

Sorry if you get this twice, it didn't post for me on the first try. I was doing what I do regularly with std::maps... a m_plsts.erase( m_plsts.begin(), m_plsts.end() ); on a boost::ptr_map. It failed with an assertion error... Digging around, it came from ptr_container/detail/associative_ptr_container.hpp on line 117. It appears that every single erase (including that done by a key) has an assert for the map not being empty! Why is this the case? As far as I understand, it is valid to attempt to remove a key that does not exist. It follows that the same behaviour would occur whether the map was empty or not. Attempting to remove an element from an empty map should not result in an abort. If I'm unaware of some C++ convention I would love to be informed, however this definitely screams bug to me. Joshua Moore-Oliva

2 1

[any] interest in boost::variant style visitors for boost::any?
by Korcan Hussein 02 Jan '07

02 Jan '07

I propose 3 changes/additions made to boost::any: 1. Support for custom allocator types. 2. Support for in-place factories. 3. Last but not least boost::variant style visitors. I'll elaborate on the last one, i've written primarily support for boost::variant style visitors for boost::any. The method is based on the acyclic visitor pattern and works as follows: struct printer_fn : boost::dynamic_visitor<int, std::string> { void operator()(int) const { std::cout << "int\n" } void operator(const std::string&) cosnt { std::cout << "string\n" } } const printer; ...... boost::any a = 1; boost::apply_visitor(a, printer); a = std::string(); boost::apply_visitor(a, printer); I'm looking to also provide the ability to extend their visitors (and override if need be) to support more types as well as provide "catch-all" (i.e (const) boost::any&) implementation which defaults to throwing an exception. I also want this to be constant-correct in the way of const_iterator and iterator, i need to look into this carefully since there are 4-cases to deal with (instead of 2). Is there any interest for these changes/additions? Thanks. Korcan Hussein. _________________________________________________________________ Be the first to hear what's new at MSN - sign up to our free newsletters! http://www.msn.co.uk/newsletters

2 1

fusion and forwarding problem
by Christian Holmquist 01 Jan '07

01 Jan '07

First of all a great thank-you (!!!) to Joel and the rest behind the development of fusion. It's solved so many tasks for me and made me write so little code the past months I probably shouldn't even get paid. database-abstraction, remote function invocation with async handlers, console/web interface invocation, etc.. It's changed my perception of effective programming to a complete new level. What before was a struggle with mpl, is now clean and easy. Ok, so to my humble problem. I'm constantly finding myself writing those overloaded functions like: ------------------ template<class A0, class A1, ....> void operator()(const A0& a0, const A1& a1, ...) const { do_something_very_clever(fusion_sequence_type(a0, a1, ...)); } ------------------ This is cumbersome and boring, and usually I just add the overloads when someone else needs it.. It's not always the case I can know beforehand to overload const T& or just T& also, but it depends on the use case. Anyways, my idea is to generate the correct overloaded operator()(...) by specializing on the size of the fusion sequence, instantiate the sequence with the parameters when invoked by client code, and then use CRTP downcast to invoke a function with the fusion sequence. This way I can do the job with my neat fusion sequence without client-code needing to mess with boost::fusion::tie and boost::ref etc, and I don't have to bother with the overloads. I guess an example would explain better than the above noise. Here's what a trivial multi-dispatch function would look like. I use function_types here to build my fusion sequence, but it could just as well have been some preprocessor enumeration dito instead. The overloading class selecting the overload is named to pack_args (attached). template<class Func> struct simple_signal : public pack_args < simple_signal const, typename boost::fusion::result_of::as_vector<boost::function_types::parameter_types<Func> >::type, void > { typedef boost::function1<void, const sequence_type> function; template<class F> struct invoke { invoke(const F& f) : f_(f) {} void operator()(const sequence_type& seq) const { boost::fusion::unpack_args(f_, seq); } F f_; }; template<class F> void connect(const F& f) { m_functions.push_back(invoke<F>(f)); } // callback from pack_args. I agree this convention looks ugly, but it'll do for now.. void on_packed_args(const sequence_type& seq) const { for(std::list<function>::const_iterator i = m_functions.begin(); i != m_functions.end(); ++i) { (*i)(seq); } } private: std::list<function> m_functions; }; I can now use the above with simple_signal<void(int const, int const) > sig; sig(24, 54); // call all connected functions here.. If I have overlooked anything in fusion that would give me this functionality, please point me in the right direction and I'll be forever grateful :) Regards, Christian

4 10

[thread][string-algp][1.34] Doc build
by John Maddock 01 Jan '07

01 Jan '07

There are a couple of warnings from both the string-algo and thread lib doc builds that look like they might be errors: Citation to nonexistent publication abbrev: Citation to nonexistent publication abbrev: Any ideas? Presumably something common to both libs? John.

2 6