May 2003 - Boost-users - lists.preview.boost.org

boost::shared_ptr and stl algorithms
by Bill Eidson 15 Apr '05

15 Apr '05

std::vector< boost::shared_ptr<T> > TVec; I want to write a function object to run std::for_each. to do somework against members of T and accumulate results in functor. My attempts have failed in VC7 with memory errors in my operator() method. struct fo { bool operator()( boost::shared_ptr<T>& t ) { return true; } }; Thanks in advance bille

5 9

BGL and reversing / moving edges
by joshrose 17 Dec '03

17 Dec '03

G'day, I was hoping someone might be able to tell me if they've run into any situations using BGL where they needed to move (or reverse (i.e. flip the source and target)) an edge within a graph implemented as an adjacency_list. I'm working on an implementation of an algorithm for max. weight bipartite matching ("the assignment problem"), and one of the assumptions in the algorithm is that non-matched edges are directed from one subgraph, A, to the other subgraph, B, and matched edges are directed the other way. Over the course of growing the matching, a simple way to add and remove edges from the candidate set is just to reverse the edge, thus the need for some way to reverse the edge within BGL. I don't think I can just remove and then re-add the edge in the other direction, primarily because doing so will (I believe) invalidate any edge_descriptors referring to that edge. In particular, there are property maps keeping track of the weight of edges, the predecessors of certain nodes, etc. that use edge descriptors as keys. I've spent a fair bit of time reading the documentation and looking through the implementation of adjacency_lists (incl. edge_list_impl), and I'm not really sure of a way to either move an edge directly within a graph, or even to create a new edge from another s.t. they have the same edge_descriptor (in general a bad thing, probably). Anyone have any thoughts on this? Thanks, -Josh

4 10

Graph Library: Ordered Vertex List?
by dmatt001 01 Aug '03

01 Aug '03

Hi Guys, I would like to use an ordered vertex list (ie based on property type) in an implementation that I am working on such that duplication can be avoided. I have read and understand how to implement an ordered edge list thanks to the previous posts but am finding it hard to get started! Perhaps I have missed the key example? It appears that from the way config::VertList is constructed that it may be a similar process in nature? adj_list_impl::add_vertex appears to discard the boolean return value from the associative insertion. Would such a construction allow me to avoid the duplication I seek to avoid? I am guessing so, but some advice would give me a confidence "boost". ( <-- It's a lame pun: I really do need help ;-) ) My first steps are below rehashing the excellent ordered edges example with a simple search/replace and maching my target container base. I'll keep plugging along with it anyway and it may just compile one day. Suggestions/Ideas/Refs/Ptrs are appreciated. Matt --- first guess follows --- template <class VertDesc> struct order_by_name : public std::binary_function<VertDesc,VertDesc,bool> { bool operator()(const VertDesc& v1, const VertDesc& v2) const { // Order by vertex name. return boost::get(boost::vertex_name, v1) < boost::get(boost::vertex_name, v2); } }; #if !defined BOOST_NO_TEMPLATE_PARTIAL_SPECIALIZATION struct ordered_set_by_nameS { }; namespace boost { template <class ValueType> struct container_gen<ordered_set_by_nameS, ValueType> { typedef std::set<ValueType, order_by_name<ValueType> > type; }; } #else struct ordered_set_by_nameS { template <class T> struct bind_ { typedef std::multiset<T, order_by_name<T> > type; }; }; namespace boost { template <> struct container_selector<ordered_set_by_nameS> { typedef ordered_set_by_nameS type; }; } #endif

3 3

RE: [Boost-Users] Agree totally.
by Jim.Hyslop 15 Jun '03

15 Jun '03

Dan Dimerman [mailto:yg-boost-users@gmane.org] wrote: > Can anyone who successfuly wrote some code with BGL post it > somewhere ? I presume you're asking because you are having problems writing code with the BGL. Rather than ask for someone to post code, why not tell us the problem you are having? Oh, by the way - the subject line really should have something to do with the body of the message. At least, that's the traditional way of doing things and it works quite well ;=) -- Jim

2 1

boost::filesystem help needed
by ericd 06 Jun '03

06 Jun '03

Hi all, I have compiled the boost::filesystem using bjam and that seemed to have gone well. The first question I have concerning this is if it is normal NOT to have a libfilesystem.so. I can only find the libfilesystem.a. The path where this lives is quite deep. /usr/local/boost_1_30_0/boost/libs/filesystem/build/bin/libboost_filesystem.a/gcc/release/runtime-link-dynamic and /usr/local/boost_1_30_0/boost/libs/filesystem/build/bin/libboost_filesystem.a/gcc/debug/runtime-link-dynamic My current setup is an Ultra 80 running Solaris 9. I use both gcc version 3.2.3 for compiling and linking. The version of Boost is 1.30.0. For my second question. Having this built, I have tried to use it but come up with undefined symbols The output follows: g++ -lboost_filesystem -o io main.o Undefined first referenced symbol in file boost::filesystem::path::native_directory_string() constmain.o boost::filesystem::path::native_file_string() constmain.o boost::filesystem::path::path[in-charge](char const*, boost::filesystem::path_fo rmat)main.o boost::filesystem::directory_iterator::directory_iterator[in-charge](boost::file system::path const&)main.o boost::filesystem::path::leaf() constmain.o boost::filesystem::initial_path() main.o boost::filesystem::system_complete(boost::filesystem::path const&)main.o boost::filesystem::directory_iterator::m_inc() main.o boost::filesystem::directory_iterator::directory_iterator[in-charge]()main.o boost::filesystem::is_directory(boost::filesystem::path const&)main.o boost::filesystem::directory_iterator::m_deref() constmain.o boost::filesystem::exists(boost::filesystem::path const&)main.o ld: fatal: Symbol referencing errors. No output written to io Now I am quite sure that the path to the filesystem library is correct and therefore, cannot figure out why this will not link properly. But then again, I might be mistaken. Here is a snip from my makefile. On second thought, here is the whole thing: SHELL = /bin/sh # PROJECTLOCATION = /home/ericd/Projects/FileSystemTest EXEDIR = $(PROJECTLOCATION)/bin # LIBPATH = -L/usr/local/lib # Sources, includes and the likes.. also output names SELFTESTSRC = main.cpp INCLUDE = -I/usr/local/boost_1_30_0/boost SELFTEST = io # ifeq (,$(filter _%,$(notdir $(CURDIR)))) include target.mk else VPATH = $(SRCDIR) # Compiler Linker Defines for test exe CXX = g++ LD = g++ DEBUG = -g CXXSELFTEST = $(DEBUG) $(INCLUDE) -c -O2 -W -Wall -Wno-unused -Wno-multichar -fexceptions LDSELFTESTFLAGS = -lboost_filesystem # Building options all: test test: CXXFLAGS = $(CXXSELFTEST) test: $(SELFTEST) # SELFTESTOBJ = $(SELFTESTSRC:.cpp=.o) # pull in dependency info for *existing* .o files -include $(SELFTESTOBJ:.o=.d) # Compile and generate dependency info %.o: %.cpp $(CXX) $(CXXFLAGS) -MMD $(SRCDIR)/$*.cpp -o $@ @mv -f $*.d $*.d.tmp @sed -e 's|.*:|$*.o:|' < $*.d.tmp > $*.d @sed -e 's/.*://' -e 's/\\$$//' < $*.d.tmp | fmt -1 | sed -e 's/^ *//' -e 's/$$/:/' >> $*.d @rm -f $*.d.tmp # Link all Object Files and copy them over to our bin $(SELFTEST): $(SELFTESTOBJ) $(LD) $(LDSELFTESTFLAGS) -o $(SELFTEST) $(SELFTESTOBJ) cp $(SELFTEST) $(EXEDIR)/$(SELFTEST) # endif I have added in my .cshrc at the LD_LIBRARY_PATH var the location of my filesystem library. Being relatively new to gcc my guess is that something has to be wrong with the flags I give it. I did do an nm -C libfilesystem.a and all missing symbols reported by the link are there. Does anyone have any clue as to why this aint working? Thanks in advance for your input, Eric

3 3

Agree totally.
by Dan Dimerman 03 Jun '03

03 Jun '03

Can anyone who successfuly wrote some code with BGL post it somewhere ?

2 1

RE: [Boost-Users] int64_t problem
by Jim.Hyslop 02 Jun '03

02 Jun '03

Sigis wrote: > I am using OpenInventor and boost 1.29 librariess. Both of > them have defined > int64_t: > > typedef __int64 int64_t; [...] > The only solution I found is to add #undef just before define at > boost/cstdint.hpp: > #undef int64_t > typedef __int64 int64_t; > > But it looks as dirty solution. Maybe it is possilbe to solve > this problme in > more elegant way? First step is to send a complaint (or enhancement request, if you prefer :=) to OpenInventor - any new types they introduce should be wrapped in their own namespace. Also, if #undef really *does* work, then that means either your compiler is buggy (#undef should only work on macros, not typedefs), or OpenInventor are using a #define, not a typedef - double penalty points for that! (well, I suppose if it's a C library, that _might_ be acceptable. Barely.) If it is a #define, you could try wrapping the OpenInventor headers in another header: // Wrapper file for using OpenInventor with Boost [include guard #ifdefs] #include "OpenInventorHeader.h" #undef int64_t #endif // #include guard In any source file that uses both Boost and OpenInventor, include the wrapper header instead of the OpenInventor header. You'll want to check for any include order problems, i.e. you may have to ensure all Boost headers are included after all OpenInventor headers. Subtle issue, though - could this violate the ODR (One Definition Rule)? I think we'd have to see exactly how OpenInventor defines/typedefs int64_t to be sure. Other possible solutions: - Decouple Boost and OpenInventor, by wrapping the OpenInventor objects you are using in proxy objects. The proxies will use an opaque handle to the OpenInventor objects. Use these proxy objects wherever you currently use OpenInventor. This, by the way, has the advantage of decoupling your application from OpenInventor, allowing you the option to move to a different library if you ever decide that should be necessary. - If OpenInventor's int64_t is a typedef, put it inside an OpenInventor namespace (this probably won't work if you do not have the source code to OI in order to rebuild the binaries). If it's not a typedef, then make it a typedef in an OpenInventor namespace (same caveats). - If OpenInventor really does use a typedef, and if the typedef is inside an OI namespace, then hunt down and destroy any using directives (i.e. "using namespace OI;" or "using OI::whatever;") in any headers, and move any using directives inside your source files *after* all #include directives. If you find a "using namespace" directive in any of the OpenInclude directives, then that's triple penalty points against them, and definite cause for a complaint (not an enhancement request :=). - [radical solution, probably unrealistic] change library vendors, and be sure to tell OI why you are dropping them (because they don't play well with everyone else). - <sigh> change library vendors, and be sure to tell Boost why you are dropping them :=( Boost plays nicely - int64_t is a typedef, and it is declared inside namespace boost. -- Jim

2 1

No more than 10 types in boost::tuple?
by Tarjei Knapstad 02 Jun '03

02 Jun '03

I tried constructing a tuple with 12 types today without much luck. I've read through the headers to see if there was any obvious way of getting around the 10 limit but couldn't find one (I half expected to find something along the way of vectorxx<>, in MPL). Am I being stupid, or is there really no way of constructing a tuple with more than 10 elements? Also, are there any plans of integrating and/or making boost::tuple dependant on the MPL? I realize that tuple predates the MPL, but there's no use in duplicating efforts. boost::tuple has it's own compile time IF<> and also provides it's own typelist class which are both readily available in the MPL. Cheers, -- Tarjei

2 3

How to compile jam.exe with Visual Studio .NET (Win2000) where to download it?
by rasasenkuviene 01 Jun '03

01 Jun '03

I was trying to build boost-jam and boost-build executable with build.bat in Visual Studio .NET (Windows 2000) from downloaded sources. However, ended with single non-functional jam0.exe and truncated to zero length my own compiler cl.exe from Visual Studio .NET. Nmake coul'nt be used, because makefiles were not included. What is going on? How to compile or obtain needed boost-jam or boost-build executables under Windows2000? Or this BOOST is intended for UNIX only?

3 2

Newbee Questions on Directory Structure of Boost Libs.
by Adalbert Perbandt 01 Jun '03

01 Jun '03

Hi there, I'm new to Boost and have some problems understanding the way Boost libs. are built and installed. I downloaded boost_1_30_0. unpacked the archive and built the test library (I had installed bjam previously). I did the build step on two different platforms and with different compilers: - On Linux (RedHat 7.3) with gcc/3.2.2 and - on Sun SPARC Solaris 2.7 with Sun's Forte 6.2 compiler. I created an installation directory /opt/boost_1.30.0. Into this directory I copied the directory tree below boost/test and libs/test: /opt/boost_1.30.0/boost/test/... /opt/boost_1.30.0/libs/test/... I think so far everything is correct, isn't it? I do not completely understand the directory structure generated during the build process: ../boost_1_30_0/libs/test/build/bin/libboost_prg_exec_monitor.a /libboost_prg_exec_monitor.so /libboost_test_exec_monitor.a /libboost_test_exec_monitor.so /libboost_unit_test_framework.a /libboost_unit_test_framework.so Below the directories given above a specific subdirectory tree was created. Example: .../libboost_prg_exec_monitor.a/ /sunpro/ /debug/ /runtime-link-dynamic/ /runtime-link-static/ /threading-multi/libboost_prg_exec_monitor.a /threading-single/libboost_prg_exec_monitor.a >From this I conclude: 1. The builds are separated from each other in compiler specific subdirectory trees, e.g. the "sunpro" subtree contains the build results from $ bjam "-sTOOLS=sunpro" i.e. the sun forte 6.2 compiler and the "gcc" subtree contains the build results from $ bjam "-sTOOLS=gcc" i.e. the gcc 3.2.2 compiler on Linux. Obviously there is no separation by OS platform (Solaris or Linux). What would I have to do to build the lib. with gcc on Solaris? Where would the products be installed? Furthermore there is no separation by compiler version (e.g. gcc 2.95.2 is not separated from gcc 3.2.2). Is this considered unnecessary? 2. The libs were built with debug symbols (-g option) only), correct? I'm not familiar with jam and therefore don't know what to do to get them built with optimization active. It seems to me as if this should be rather simple because in ../boost_1_30_0/tools/build/sunpro-tools.jam obviously two variants of the CFLAGS variable for optimization are defined: flags sunpro CFLAGS <optimization>speed : -fast -O4 ; flags sunpro CFLAGS <optimization>space : -O2 -xspace ; Any hints are welcome. Thanks, Adalbert Perbandt.

3 3