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

Const strangeness and indirect_iterator generators.
by Tim Robertson 08 Jul '03

08 Jul '03

Hi, I've been trying to get indirect_iterator working in what seems (to me) to be a straightforward context -- I have a std::vector of boost::shared_ptr objects, and I need both const and non-const iterators into the vector (I'm using the vector as the sole data member of a container class, and I need the iterators to properly handle const and non-const container types). This looks something like this: class MyClass { // ... }; class MyContainer { public: typedef MyClass value_type; typedef value_type & reference; typedef value_type * pointer; // and so on, for const_reference, const_pointer, etc... // ... private: std::vector<boost::shared_ptr<MyClass> > _data; }; Now, the trouble comes when I attempt to define the iterators for MyContainer using the indirect_iterator generators. If I do the obvious: Approach #1: ------------ typedef indirect_iterator_pair_generator< vector<shared_ptr<MyClass> >::iterator, value_type, reference, const_reference, std::random_access_iterator_tag, pointer, const_pointer>::itgen; typedef typename itgen::iterator my_iterator; typedef typename itgen::const_iterator my_const_iterator; I find that the const_iterator generated is inadequate for use in a container context. Particularly, it produces a my_const_iterator that has the following constructor signature: my_const_iterator(shared_ptr<MyClass> *) Whereas, a const std::vector<shared_ptr<MyClass> > will produce const_iterators that look like this: const shared_ptr<MyClass> * const This causes compilation problems, because inevitably, in this context, you'll want to construct a my_const_iterator from a vector::const_iterator. The problem seems obvious: the indirect_iterator_pair_generator class is considering only non-const base iterators. This led me to try the following: Approach #2: ----------- typedef typename indirect_iterator_generator <typename vector<shared_ptr<MyClass> >::iterator, value_type, reference, pointer, std::random_access_iterator_tag>::type my_iterator; typedef typename indirect_iterator_generator <typename vector<shared_ptr<MyClass> >::const_iterator, value_type, const_reference, const_pointer, std::random_access_iterator_tag>::type my_const_iterator; It seemed that this would solve the problem by independently generating the two types of iterators that are needed (note how I've made the base iterator type ::iterator in one, and ::const_iterator in the other). However, with this method I always obtain compilation errors -- whereever there is a use of the operator->(), the following error is produced (g++ 3.0.3 and 3.2 do the same thing, Linux and Mac OSX): "result of `operator->()' yields non-pointer result" I have yet to figure out what is causing this error. This led me to attempt the following "solution" to the problem: Approach #3: ----------- typedef indirect_iterator_pair_generator <typename vector<shared_ptr<MyClass> >::iterator, value_type,reference,const_reference, std::random_access_iterator_tag, pointer,const_pointer> itgen; typedef typename itgen::iterator my_iterator; typedef indirect_iterator_pair_generator <typename vector<shared_ptr<MyClass> >::const_iterator, value_type,reference,const_reference, std::random_access_iterator_tag, pointer,const_pointer> itgen2; typedef typename itgen2::const_iterator my_const_iterator; This actually works! I can't explain why -- it seems like indirect_iterator_pair_generator and indirect_iterator_generator should behave identically when their template parameters are identical.... The way I see it, there's a bug in at least one of the indirect_iterator generator objects. At the very least, indirect_iterator_pair_generator and indirect_iterator_generator are behaving differently when their parameter types are the same (compare approaches #2 and #3). But there's also the issue of why approach #1 didn't work to begin with -- it *seems* like indirect_iterator_pair_generator should produce a const_iterator type that can be initialized with an iterator from a const vector. This isn't the case. If anyone can explain this to me, and convince me that I'm not going insane, I'd appreciate it greatly ;-) (Standard bug-report disclaimers: Current boost distribution, g++ 3.0.x/3.2.x, both Mac OSX and Linux exhibit identical behaviour.) -Tim

2 1

Ublas >> operators
by Amit Bhatia 08 Jul '03

08 Jul '03

Hello All, I was looking for sowe way to write the matrix to some filestream and read the same way from some file stream. Was just wondering if Ublas has any operators as << and >> defined default? I have been able to find none though. thanks, regards, amit.

2 1

Re: Using BOOST_REQUIRE outside the Test framework?
by Gennadiy E. Rozental 08 Jul '03

08 Jul '03

> Is there some macro/facility/whatever which behaves like assert(): > a) works without the Boost Test Framework > b) *AND* integrates into the BTF in those cases where the BTF is used? > > Regards, > > Joaquín M López Muñoz > Telefónica, Investigación y Desarrollo Jonh Torio developing SMART_ASSERT facility that supposed to satisfy your requirements. Look into main list archive. Gennadiy.

1 0

boost_1_30_0: cannot compile red hat gcc
by sebMarc 07 Jul '03

07 Jul '03

I must be freakin stupid. I have the simplest config possible, gcc 2.96 with a red hat system. Downloaded boot1_30_0, unzipped it, tried to compile thread and got: bash-2.05a$ bjam "-sTOOLS=gcc" ...found 341 targets... ...updating 37 targets... MkDir1 ../../../libs/thread/build/bin-stage : command not found MkDir1 ../../../libs/thread/build/bin : command not found mkdir: cannot create directory `\r': File exists mkdir ../../../libs/thread/build/bin ...failed MkDir1 ../../../libs/thread/build/bin... Being jam ignorant, this is it, i am stuck :-(. Sigh.. Tried to run configure and got: bash-2.05a$ chmod +x configure bash-2.05a$ ./configure : bad interpreter: No such file or directory Seb.

3 3

Memory leaks in Solaris
by Lev Assinovsky 07 Jul '03

07 Jul '03

Hi All! I caught a memory leak (24 bytes) when using recursive_mutex class. This class initializes a pthread_mutexattr attribute but never destroys it thus leaving the leak. We use tons of the recursive mutexes in our huge app and that leak leads to significant memory loss. Regards, Lev Assinovsky Aelita Software Corporation O&S Core Division, Programmer ICQ# 165072909

1 0

Using BOOST_REQUIRE outside the Test framework?
by Joaquín Mª López Muñoz 07 Jul '03

07 Jul '03

I'm deploying a library which implements some basic invariant checking when a certain control macro is defined. class A { void foo1(){VERIFY(*invariant conditions *); ...} ... void foon(){VERIFY(*invariant conditions *); ...} }; I use the word VERIFY for reference purposes only. Now, VERIFY is resolved to a call to assert(), which is sensible enough as: * Invariant errors are critical * Invariant checking is only enabled explictly by the user, presumably under testing conditions. On a second thought, it'd be nice if VERIFY could be seamlessly integrated into the Boost Test Framework, i.e. invariant errors are caught by the test execution monitor as other errors defined by BOOST_CHECK, BOOST_REQUIRE etc. Unfortunately, replacing VERIFY with say BOOST_REQUIRE forces the user to link with the Boost Test Framewok object modules. So my question is: Is there some macro/facility/whatever which behaves like assert(): a) works without the Boost Test Framework b) *AND* integrates into the BTF in those cases where the BTF is used? Regards, Joaquín M López Muñoz Telefónica, Investigación y Desarrollo

1 0

multi_array performance
by Marco Segurini 07 Jul '03

07 Jul '03

Hi, I am very new in using Boost library. I start replacing row memory array with 'multi_array' class template but the performace slow down a lot. Running this test program... //begin------------------ #include "windows.h" #include <iostream> #include "boost\multi_array.hpp" using namespace std; using namespace boost; int main() { const size_t DimX = 2000; const size_t DimY = 1500; const size_t DimZ = 1; multi_array<int,3> V3(extents[DimZ][DimY][DimX]); int iVal; unsigned long time1 = GetTickCount(); for (size_t Z=0; Z<DimZ; ++Z) for (size_t Y=0; Y<DimY; ++Y) for (size_t X=0; X<DimX; ++X) iVal = V3[Z][Y][X]; unsigned long time2 = GetTickCount(); cout << "Multi_array: " << time2-time1 << endl; int * pInt = V3.data(); time1 = GetTickCount(); for (size_t Z=0; Z<DimZ; ++Z) for (size_t Y=0; Y<DimY; ++Y) for (size_t X=0; X<DimX; ++X) iVal = pInt[(Z*DimY+Y)*DimX+X]; time2 = GetTickCount(); cout << "Pointer....: " << time2-time1 << endl; return 0; } //end------------------ ... the output (using vc++ in release mode with no optimization) is Multi_array: 1663 Pointer....:70 So, where am I wrong? Thanks. Marco.

6 6

[BGL] dijkstra_shortest_paths and the single-pair problem
by Richard Jepps 04 Jul '03

04 Jul '03

I would like to use the dijkstra_shortest_paths algorithm to find the shortest path from one specified vertex to another. The standard Boost implementation will find this, but it will continue until it has found the shortest paths to all nodes in the graph (i.e. solved the single source problem). The BGL book (Section 5.1, pp76) says: "It turns out that there are no algorithms for solving the single-pair problem that are asymptotically faster than the algorithms that solve the single-source problem". I think this is because the single-source algorithms all have the property that it is possible to determine whether the shortest path to any given node has been found during the execution of the algorithm, therefore you can use the same algorithm to solve both types of problem provided that the termination criterion is appropriate to the problem. Consider this problem: "find the shortest path from a source vertex u to a nearby destination vertex v in an infinitely large directed weighted graph G in finite time." The single-source problem can only be solved in infinite time, but the single-pair problem can be solved relatively quickly (because the two vertices are near each other). I think that the termination criterion is quite straightforward using a DijkstraVisitor - it is when vis.finish_vertex(u, g) is called and u == destination_node. Unfortunately I can't see how to stop the algorithm in a reasonable way. An unreasonable solution is to provide a custom graph and turn off all its edges when the termination criterion is met. This is unreasonable because the responsibility for termination belongs to the algorithm, not the data structure - I shouldn't have to modify the data structure to terminate the algorithm. It's also unreasonable because I have to customise both the graph and the algorithm to do something very simple. Is there a better way? Thanks Richard

3 5

function object with lambda
by Matt Hurd 04 Jul '03

04 Jul '03

Not sure how to force a for_each with a lambda to invoke the random function object successively. In the for_each version below get_long is called just once. Not sure how to use var, std_function, bind or something else to overcome this. I'm sure the answer is straight forward... anyone? Regards, Matt. #include <iostream> #include <boost/random.hpp> #include <vector> #include <boost/lambda/lambda.hpp> #include <boost/lambda/bind.hpp> typedef boost::mt19937 base_generator_type; base_generator_type generator( 100123u ); boost::uniform_int<base_generator_type,long> gen_long(generator,-100,100); using namespace std; using namespace boost::lambda; int main(int argc, char *argv[]) { vector<long> v(10); for_each(v.begin(),v.end(), _1 = gen_long() ); for_each(v.begin(), v.end(), cout << _1 << ' '); cout << "\n"; for( vector<>::iterator i = v.begin(); i != v.end(); ++i ) { *i = gen_long(); } for_each(v.begin(), v.end(), cout << _1 << ' '); cout << "\n"; system("PAUSE"); return 0; }

2 4

Re: [Boost-Users] [BGL] dijkstra_shortest_paths and the single-pair problem
by JOAQUIN LOPEZ MU?Z 03 Jul '03

03 Jul '03

Throw an exception? Joaquín M López Muñoz Telefónica, Investigación y Desarrollo ----- Mensaje Original ----- De: Richard Jepps <yg-boost-users(a)m.gmane.org> Fecha: Martes, Julio 1, 2003 3:00 pm Asunto: [Boost-Users] [BGL] dijkstra_shortest_paths and the single-pair problem > I would like to use the dijkstra_shortest_paths algorithm to find the > shortest path from one specified vertex to another. > > The standard Boost implementation will find this, but it will > continue until > it has found the shortest paths to all nodes in the graph (i.e. > solved the > single source problem). > > The BGL book (Section 5.1, pp76) says: "It turns out that there > are no > algorithms for solving the single-pair problem that are asymptotically > faster than the algorithms that solve the single-source problem". > I think this is because the single-source algorithms all have the > propertythat it is possible to determine whether the shortest path > to any given node > has been found during the execution of the algorithm, therefore > you can use > the same algorithm to solve both types of problem provided that the > termination criterion is appropriate to the problem. > > Consider this problem: "find the shortest path from a source > vertex u to a > nearby destination vertex v in an infinitely large directed > weighted graph G > in finite time." > The single-source problem can only be solved in infinite time, but the > single-pair problem can be solved relatively quickly (because the two > vertices are near each other). > > I think that the termination criterion is quite straightforward > using a > DijkstraVisitor - it is when vis.finish_vertex(u, g) is called and > u == > destination_node. Unfortunately I can't see how to stop the > algorithm in a > reasonable way. > > An unreasonable solution is to provide a custom graph and turn off > all its > edges when the termination criterion is met. > This is unreasonable because the responsibility for termination > belongs to > the algorithm, not the data structure - I shouldn't have to modify > the data > structure to terminate the algorithm. > It's also unreasonable because I have to customise both the graph > and the > algorithm to do something very simple. > > Is there a better way? > > Thanks > > Richard > > > > > > > > ------------------------ Yahoo! Groups Sponsor -------------------- > -~--> > Get A Free Psychic Reading! Your Online Answer To Life's Important > Questions.http://us.click.yahoo.com/Lj3uPC/Me7FAA/ySSFAA/EbFolB/TM > ------------------------------------------------------------------- > --~-> > > Info: <" target="l">http://www.boost.org> > Wiki: <" target="l">http://www.crystalclearsoftware.com/cgi- > bin/boost_wiki/wiki.pl>Unsubscribe: <')" >mailto:boost-users- > unsubscribe(a)yahoogroups.com> > > Your use of Yahoo! Groups is subject to > http://docs.yahoo.com/info/terms/ > > >

5 5