- Boost-users - lists.boost.org

Re: [Boost-users] [boost] Convert function pointer type to function type
by Peter Soetens 04 Jul '06

04 Jul '06

Quoting Peter Dimov: > Peter Soetens wrote: >> Hi, >> >> I'm desperately seeking for doing this conversion using the boost >> libraries. User code often provides a function pointer, while library >> code often works with function types. [...] > > One obvious way is > > template<class F> boost::function<F> function_wrapper( F * pf ) > { > return boost::function<F>( pf ); > } > > but a more interesting question is why do you need a function_wrapper at > all. Everything that you could do with the returned function<F> (call it) is > also possible with pf itself. Looking at the simplicity of that answer, this was clearly a boost-users question :-]. Sorry. This was a minimalistic example, removing the context. The big picture uses boost::function<F> to store partially bound objects, since this is a way to store the result of boost::bind for later use in a not templated class. I got confused by the boost::function_traits<T>, which only takes a "function type" and not a "function pointer type." I assumed one was not convertible to the other. I tried using boost::remove_pointer<F>, which didn't work and confirmed back then my assumption (confusingly, is_pointer<F> did work, but I could not remove it using type_traits...) My books write about function pointer types but do not mention 'function types'. These concepts appear to me underdocumented in standard C++ works. *Thanks a lot*. Peter -- www.fmtc.be ---------------------------------------------------------------- This message was sent using IMP, the Internet Messaging Program.

1 0

Re: [Boost-users] [BGL] All Possible Matchings (Hossein Haeri)
by Grégoire Dooms 04 Jul '06

04 Jul '06

boost-users-request(a)lists.boost.org wrote: Message: 9 Date: Mon, 3 Jul 2006 16:46:55 +0100 (BST) From: Hossein Haeri <powerprogman(a)yahoo.com> Subject: [Boost-users] [BGL] All Possible Matchings To: boost-users(a)lists.boost.org Message-ID: <20060703154655.35182.qmail(a)web53405.mail.yahoo.com> Content-Type: text/plain; charset=iso-8859-1 > Dear all, > > I have a bipartite graph, and want to generate all its > possible matchings. Any suggestions? > You should have a look at Constraint Programming and the Alldifferent constraint. In CP, the task of filtering for a constraint over some integer variables amounts to partitioning, for each of its variables, the potential values into three categories: the values participating in all solutions (obviously at most one per variable), the values participating in some solutions, and the values participating in no solutions. In the case of the Alldifferent constraint, it can be reinterpreted in graph matching terms: Build a bipartite variable-value graph. A solution to the Alldifferent constraint is a maximum matching covering all variables in this graph. Hence filtering can be done by identifying edges which belong to some/all/no max matching. In a nutshell, the following property of matchings is used [Peterson] [Berge] : M1 (+) M2 = P M1 (+) P = M2 where P is an alternating path (for instance, look at the LEDA book for a good exposition of this property and matching algorithms). This property is used in matching algorithms with P an augmenting alternating path. If you consider "maintaining" paths (eg. non-augmenting and non-decreasing -- even length, starting at a free node and ending at a covered node, or an alternating cycle) and apply them on a max matching you get another max matching. Here are a few references: Ref. to the original paper : http://citeseer.ist.psu.edu/context/210815/0 A survey on this subject: http://citeseer.ist.psu.edu/579391.html (or http://www.cs.cornell.edu/~vanhoeve/papers/alldiff.pdf for a more recent/in progress version) Slides by JC Regin: http://www.constraint-programming.com/people/regin/papers/GraphAndCP.pdf Finally, an open-source implementation: http://www.gecode.org/gecode-doc-latest/distinct_2dom_8icc-source.html Hope this helps, -- Grégoire Dooms.

1 0

When to use thread.join()
by Simon Reye 04 Jul '06

04 Jul '06

My apologies if this has been covered. I am new to boost and have been working with the boost thread class. I assume that the join() function waits for the thread to finish execution. But what if the thread has already finished executing? Can I still call join() and have it not wait for infinity? Simon

2 2

[Spirit] puts namespace std into the global scope
by Johan Råde 04 Jul '06

04 Jul '06

It seems that the header boost/spirit.hpp puts the entire namespace std into the global scope. That is not very nice! --- Example 1 -------- #include <string> #include <sstream> #include <fstream> #include <vector> #include <boost/spirit.hpp> void f() { string s; istringstream iss; ofstream ofs; vector<int> v; } ---------------------- Example 1 should not compile (undefined symbols), but does compile. If the line #include <boost/spirit.hpp> is removed, then Example 1 does not compile. --- Example 2 -------- #include <boost/filesystem/path.hpp> #include <boost/filesystem/fstream.hpp> #include <boost/spirit.hpp> using boost::filesystem::path; using boost::filesystem::ifstream; void g() { path p; ifstream ifs(p); } ---------------------- Example 2 should compile but does not. The compiler complains that the symbol ifstream is ambigous. If the line #include <boost/spirit.hpp> is removed, then Example 2 does compile. I´m using Boost 1.33.1 and MSVC 7.1 on a Windows XP computer. --Johan Råde rade(a)maths.lth.se

3 3

auto-diff expression templates-> MPL
by Russell L. Carter 03 Jul '06

03 Jul '06

Greetings, This is a beginner question about the MPL. I'm building a general purpose simulation program which will have an optimization module as well as an internal ODE solver which requires derivatives of the implemented algorithm in various forms. Around 1998-2002 there was a lot of enthusiasm for generating the derivatives automatically using expression templates. I'm not quite through with an exhaustive review of the literature but it appears the various efforts ran to ground on the compilation memory and cpu requirements that evaluating expression templates in their original form required. So 1st question, has anybody done something like this with mpl/fusion sequences and algorithms? Has there been an improvement in compile time performance? Do you have any pointers to trivial examples? I've read Alexandrescu, and Abrahams & Gurtovoy and it's not clear to me how to map the expression templates as described in the earlier efforts onto mpl sequences and algorithms. That's probably enough for now. Thanks for any help. Russell

1 0

[shmem] and kernel 2.4 / linuxthreads
by Sascha Lumma 03 Jul '06

03 Jul '06

Hi all, I noticed that kernel 2.6 is a requirement for shmem on linux. Is it possible to get shmem to run - maybe with constraints - together with kernel 2.4 and the older pthread version (linuxthreads)? A start on a kernel 2.4 box produces a segmentation fault in shared_memory.hpp. To reach most users as possible, it would be nice to be compatible or if not possible to get as a minimum a normal termination. Sascha Lumma

2 2

[BGL] python bindings build graph from edge list
by John Reid 03 Jul '06

03 Jul '06

Hi, I was trying to build a graph from an edge list and ran into some problems. The following code throws an exception on write_graphviz(). I think it is because there's no property map called node_id in the graph. import boost.graph as bgl edges = [ ( 'a', 'b' ), ( 'b', 'c' ), ( 'c', 'a' ) ] g = bgl.Graph( edges ) g.write_graphviz( 'bind-copy.dot' ) I get the following: Traceback (most recent call last): File "<stdin>", line 1, in ? KeyError: 'node_id' I couldn't manage to generate the node_id property map automatically. So after you have built the graph with the edge list constructor how do you know which vertex is which? I ended up using the following code to create the graph: def build( edges ): # Create a graph to build g = bgl.Graph() # Construct some property maps node_ids = g.vertex_properties[ 'node_id' ] = g.vertex_property_map( 'string' ) color = g.vertex_properties[ 'color' ] = g.vertex_property_map( 'string' ) centrality = g.vertex_properties[ 'centrality' ] = g.vertex_property_map( 'float' ) weight = g.edge_properties[ 'weight' ] = g.edge_property_map( 'float' ) length = g.edge_properties[ 'len' ] = g.edge_property_map( 'float' ) # create the edges and vertices name_2_vertex = { } for ( name1, name2 ) in edges: if not name_2_vertex.has_key( name1 ): v1 = name_2_vertex[ name1 ] = g.add_vertex() node_ids[ v1 ] = name1 else: v1 = name_2_vertex[ name1 ] if not name_2_vertex.has_key( name2 ): v2 = name_2_vertex[ name2 ] = g.add_vertex() node_ids[ v2 ] = name2 else: v2 = name_2_vertex[ name2 ] e = g.add_edge( v1, v2 ) weight[ e ] = 0.1 length[ e ] = 4 # Color articulation points red for v in bgl.biconnected_components( g ): color[ v ] = 'red' # Assign centrality values to property map bgl.brandes_betweenness_centrality( g, centrality ) return g This code seems to works fine. So I thought I'd post this just to see if I was missing something obvious and/or in the hope someone else might find it useful. Best, John.

3 6

[BGL] All Possible Matchings
by Hossein Haeri 03 Jul '06

03 Jul '06

Dear all, I have a bipartite graph, and want to generate all its possible matchings. Any suggestions? TIA, --Hossein ___________________________________________________________ Yahoo! Messenger - with free PC-PC calling and photo sharing. http://uk.messenger.yahoo.com

1 0

[BGL] Bipartite Graphs
by Hossein Haeri 03 Jul '06

03 Jul '06

Dear all, Any special facility there is in BGL for bipartite graphs? TIA, --Hossein ___________________________________________________________ All New Yahoo! Mail Tired of Vi@gr@! come-ons? Let our SpamGuard protect you. http://uk.docs.yahoo.com/nowyoucan.html

2 2

[BGL]Customising the Adjacency List Storage
by Louis Lavery 03 Jul '06

03 Jul '06

According to the book [BGL p232], in order to customise the EdgeList I need to define a container generator and parallel edge traits class, and over load the push() and erase() functions for the custom container [BGL p234]. My custom container is user::PushBackList<T>... namespace user { template<typename ValueType> struct PushBackList { //... }; struct PushBackListS { }; } ...and I've put the (specialised) container_gen and parallel_edge_traits classes in namespace boost... namespace boost { template<typename ValueType> struct container_gen<user::PushBackListS,ValueType> { typedef user::PushBackList<ValueType> type; }; template<> struct parallel_edge_traits<user::PushBackListS> { typedef allow_parallel_edge_tag type; }; } ...now all I need do is overload the push() and erase() functions. The problem is they are in nested namespace graph_detail and are called with full qualification (i.e. boost::graph_detail::push(...)) from adjacency_list class. That means I need to put my overloads in boost::graph_detail. Is that right? Thanks, Louis.

2 1