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

Newbie question, linking problems
by Mrk 17 Nov '03

17 Nov '03

Hello, I'm trying to get regexp to work under vc6, STL 4.5.3 and boost 1.30.2 but sadly I'm running into some problems. Did anyone ever had this problem with the following methods ? error LNK2001: unresolved external symbol "public: static void __cdecl boost::w32_regex_traits<char>::transform_primary( ... error LNK2001: unresolved external symbol "public: static void __cdecl boost::w32_regex_traits<char>::transform( ... Thanks, Mrk

1 0

Re: [Boost-users] [BGL] Problem with color maps for undirected_dfs_visit
by Jeremy Siek 17 Nov '03

17 Nov '03

Hi Tarjei, On Friday, November 14, 2003, at 05:33 AM, Tarjei Knapstad wrote: > On Thu, 2003-11-13 at 17:09, Jeremy Siek wrote: >> Hi Tarjei, >> >> You need to initialize the edge index property map (only the vertex >> index >> map is automatically initialized, and then only for VertexList=vecS). >> > > Hmm, ok, but I still don't quite understand. I thought that > initializing > the std::vector holding the colors was sufficient...? To my > understanding an iterator_property_map is just a property map which > iterates through some other container (that can provide a > RandomAccessIterator) actually holding the properties? Yes, your edge color map is initialized just fine, but you also need to initialize the edge index map. Here's the reason why: your edge color map uses the edge index map to go from an edge descriptor to an integer offset, which is then used to access into the color vector. Currently, your edge index map is initialized to random garbage, so you are going out-of-bounds when accessing the edge color vector. See the example file exterior_properties.cpp, BTW, you don't need to declare an edge_color property for the adjacency_list because you are using an external edge color map. > > Obviously there's some connection I don't quite get here... I also > don't > quite understand the EColorValue uv_color = get(edge_color, *ei); call > in undir_dfv_impl. It asks for the edge_color at the edge_descriptor > *ei, but from where? > > Anyway, I don't really even need an edge_index property here (or do > I?), > it was more or less to try to get this working. Yes, you need it for the reason given above. BTW, if you switch to using an internal edge color property, then you won't need the edge index property, or the edge color vector. Cheers, Jeremy

2 1

GRAPHVIZ_DIRECTED macro definition?
by Soumitra Chatterjee 17 Nov '03

17 Nov '03

Hi, In the file .../libs/graph/src/graphviz_parser.y we have the following: #ifndef GRAPHVIZ_DIRECTED #error Need to define the GRAPHVIZ_DIRECTED macro to either 0 or 1 #endif I couldn't locate the definition for the above macro in my boost installation. Do we need to define it explicitly, or am I missing some file(s)? Thanks, Soumitra

2 1

Error 112: Missing directory LEDA
by Soumitra Chatterjee 17 Nov '03

17 Nov '03

Hi, I have downloaded boost-1.30.2, and while trying to compile/test it, I am getting the error(s): Error 112: "/TESTC++/Boost-1.30.2/boost/graph/leda_graph.hpp", line 33 # Include file <LEDA/graph.h> not found. #include <LEDA/graph.h> ^^^^^^^^^^^^^^ [...snip...] The directory 'LEDA' seems to be missing from the downloaded package. Have I missed something during the install process, or is there something else to be done after unpacking the installation package? I am using JAM for the build. Thanks, Soumitra

3 2

Re: Intoductory Boost Presentation
by Jorg Brown 16 Nov '03

16 Nov '03

Matt - The corrected code is indeed much better. May I suggest using std::vector instead of vector? The typedef of FooPtr is fine, it turns out; it's the typedef of struct Foo that was the problem. I had declared Foo within main as well, and that blocks the use of boost::shared_ptr on it. As long as Foo is declared at global scope, all is well. As for the correct output, my compiler gives me 3, then 2, then 1. I talked it over with a friend, and I guess the order of the 2 vs 1 is undefined, and depends on your STL's implementation of std::vector. As for the crazy aunt in the basement, the comment was meant for an internal list that only 20-30 people are on, not meant for general consumption, and probably merits an extra thread by itself. Put simply, using boost can generate a surprising amount of code and overhead. This is true of STL as well. There are even some good books describing STL; I happen to like Plauger/Stepanov/Lee/Musser, as well as Jasoutis. But nowhere do I see any reference as to what the downsides are - and nowhere do I see any concise explanation of why an enormous amount of code (660 bytes in an optimized, no-symbols build) is generated when I say something as simple as void Testbed(void) { boost::function<void> SimpleFPtr(&Testbed); } When you say "I was trying to generate interest in these libraries and thus didn't want to focus too much on the negatives", I can understand your point of view, but what I see in real life is that people start using boost::function everywhere they used to use plain old function pointers, with nary a clue that their code size just exploded with no benefit. But like I said, it's worth a whole other thread. Thanks for fixing the code and responding so quickly! jorg -----Original Message----- From: "Matt S Trentini" <matt_s_trentini(a)hotmail.com> Reply-To: Boost Users mailing list <boost-users(a)lists.boost.org> Date: Tue, 11 Nov 2003 23:19:12 +1100 To: boost-users(a)lists.boost.org Subject: [Boost-users] Re: Intoductory Boost Presentation Heya Jon, Ah, the smart_ptr example, I knew that would come back to bite me! ;) I changed this example at the last minute in an effort to simplify but, as your co-worker correctly points out, there are many errors in it. The correct code is below though the graphic requires change (yes, I could have change the list.push_front() to a push_back but I like to see the variety!). int main() { typedef boost::shared_ptr<Foo> FooPtr; vector<FooPtr> foo_vector; list<FooPtr> foo_list; FooPtr foo_ptr( new Foo(1) ); foo_vector.push_back( foo_ptr ); foo_list .push_back( foo_ptr ); foo_vector.push_back ( FooPtr( new Foo(2) ) ); foo_list .push_front( FooPtr( new Foo(3) ) ); } You'll notice that I left the typedef in there. I can't find anything in the standard (or with google) about it being illegal in main, nor does my compiler emit any warnings. Can anyone shed some light on this? And I've got to disagree with your friend about the correct ordering - the expected output is 3, 1, 2 (Remember that foo_ptr gets deleted first). I'll endeavour to fix this example up and upload the changes over the next couple of days. As for these comments: > > The rest of the pages have the classic boost downfall: they > > don't discuss any of the pitfalls, performance problems, or > > code bloat issues, which frankly, seem like the "crazy aunt > > in the basement" of boost. I'll defend myself on two fronts. First, it's an introductory presentation - I was trying to generate interest in these libraries and thus didn't want to focus too much on the negatives. Secondly, I _did_ cover some of those points - like all good presentations not everything is contained in the bullet points. Thanks for the feedback though, it's helping to make this presentation better! Cheers, Matt PS If there are any other problems that you or your co-worker have found please don't hesitate to report them! _______________________________________________ Boost-users mailing list Boost-users(a)lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users ------ End of Forwarded Message

3 2

Intoductory Boost Presentation
by Matthew S Trentini 16 Nov '03

16 Nov '03

Heya Everyone, I recently gave a presentation to my work colleagues on the wonders of the Boost library. Many weren't aware of what it was or what it could do so it just contains an overview and some coverage on nine of the libraries (with short code examples). It went for nearly an hour and was pretty well received, so I thought that maybe someone else out there could use the material. Please feel free to download (it's in a variety of formats) from here: http://www.alphalink.com.au/~mattyt/BoostPresentation1/boost.html Cheers, Matt PS If you want to modify/publish/whatever that's fine... _________________________________________________________________ Hot chart ringtones and polyphonics. Go to http://ninemsn.com.au/mobilemania/default.asp

7 10

Re: [Boost-users] Building Regex++ lib on Pocket PC using EmbeddedVC++4.0
by Reuti＠sdc.canon.co.in 16 Nov '03

16 Nov '03

Thanx. With latest version I could build but I could not create import library corressponding to dll created. Is there some macro for this? ===== Reuti "John Maddock" <john(a)johnmaddock.co.uk> To: "Boost Users mailing list" <boost-users(a)lists.boost.org> Sent by: cc: boost-users-bounces@list Subject: Re: [Boost-users] Building Regex++ lib on Pocket PC s.boost.org using EmbeddedVC++4.0 11/13/2003 05:32 PM Please respond to Boost Users mailing list > Has anyone tried building regex++ lib for PPC using EVC++? I've never tried with that compiler - I think it's even worse than VC6 for conformance? > I can build it using VC++6.0 using the .mak files supplied with it.But I am > getting errors and warnings. > Error is regarding the use of typename keyword in lib source and warning is > about C++ Exceptions unwinding capability. Defining BOOST_NO_EXCEPTIONS should fix that - in fact it should get automatically set for vc with no eh support, so there could be a config bug there. > I replaced C++ try, catch by Win32 __try , __except but don't know what to > do with template error? What is there error: you may be able to just remove the typename keyword to fix the problem? John. _______________________________________________ Boost-users mailing list Boost-users(a)lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users

2 1

Preprocessor Metaprogramming Problem
by Mitch 16 Nov '03

16 Nov '03

I'm trying to figure out the longest sequence in a sequence. Here's the MACRO I tried. #define OP(size, state, elem) BOOST_PP_MAX(state, BOOST_PP_SEQ_SIZE(elem)) #define MACRO(seq) BOOST_PP_SEQ_FOLD_LEFT(OP, 0, seq) If I process ((a)(b))((c)) I get 2 back as expected. However, the extra parenthesis seems awkward. I'd like to process ((a)(b))(c) instead, but can't. The '(c)' gets coverted to plain 'c' as the elem parameter to OP and that's not a valid sequence. Any ideas? Mitch

1 0

boost::shared_ptr performance
by Matthew Wilson 15 Nov '03

15 Nov '03

[I posted this in boost.devel, but got no response. Presumably that was the wrong group. :)] Hi Boosters I'm just doing some research on reference-counting, and contrasting self-counting types with externally counted types, the latter effected with the rather excellent shared pointer technique. [As a side question, is it true that Nicolai Josuttis was the inventor of the technique?] Naturally, one of the shared pointer's I've used is boost::shared_ptr. The measurements have demonstrated some surprising performance characteristics for shared_ptr in particular, and I want to solicit the opinion of the expert's here before I make any bold claims, as some results are just too surprising for me to trust them. There are five types tested. 1. A self-counting type, using an ancient Synesis (my company) template ref-counting bolt-in class 2. A self-counting type, using a custom-written (for this test) template ref-counting bolt-in class 3. Externally counting, using boost::shared_ptr 4. Externally counting, using a custom-written (for this test) shared pointer type 5. Externally counting, using a custom-written (for this test) shared pointer type, which has a custom allocator for the shared count members There were three scenarios investigated, and the same algorithms were applied to the above five types A: Saving refs vector<smart-ptr> ptrs; loop (100,000 times) { smart-ptr sp1(new Thing); sp1 = sp1; smart-ptr p2; p2 = p1; smart-ptr p3(p1); smart-ptr p4(p3); smart-ptr p5(p4); smart-ptr p6(p5); process_Thing(p6.get()); // A function to do something (actually it just returns the result of a call on Thing) to prevent elision by optimisation ptrs.push_back(p5); } The whole is executed within a single thread, and the program is tested in both single and multithreaded forms. B: Discarding refs This is identical to A ("Saving refs") except that saving them to the vector of ptrs is omitted. Again, this is done in one thread, and single and multi-threaded forms tested C: Multi-threaded sharing 20 smart-pointers are created, containing 20 Thing instances, in the main thread. Then 20 threads, using the following thread-proc, are created, and executed. Two times are recorded for each scenario: the total time, and the sum of the individual thread-times (obtained by the WinSTL threadtimes_counter, based on the Win32 API function GetThreadTimes()) std::vector<ptr_type> ptrs(ITERATIONS / 10); for(size_t i = 0; i < 10; ++i) { for(size_t j = 0; j < ITERATIONS / 10; ++j) { ptrs.push_back(things[i % things.size()]); } } ITERATIONS is 100,000. things is the shared shared_ptr array The purpose of this one is to really exercise the sharing between threads. Results (all results are expressed in terms of the most costly pointer) ============================================= [Note, all were compiled with Intel 7.0 with very aggressive speed optimisations (-O2 -Ox -QaxMKW /Qvc7 -EHsc), and executed in a single session on an otherwise quiescent 512MB 2GHz XP box.] A (single-threaded) -------------------- 1. Synesis self-counter: 65% 2. Custom self-counter: 58% 3. boost::shared_ptr: 100% 4. custom SharedPtr: 53% 5. custom SharedPtr + rc pool): 48% I was not surprised that the crusty old Synesis type was slower than the custom self-counter. As expected, the custom shared-ptr with rc pool was the fastest. I was a little surprised that the Boost shared_ptr was up to twice as slow. However, they're all within 100% of each other, so it's no big result I guess. A (multi-threaded) -------------------- 1. Synesis self-counter: 90% 2. Custom self-counter: 46% 3. boost::shared_ptr: 100% 4. custom SharedPtr: 28% 5. custom SharedPtr + rc pool): 27% This was surprising. I expected the use of thread-safe operations (atomic increments, etc.) to flatten out the curve, but the reverse has happened. Interestingly, the two real-world types (Synesis::RefCounter and boost::shared_ptr) fair less well. Again, not what I expected. B (single-threaded) -------------------- 1. Synesis self-counter: 65% 2. Custom self-counter: 55% 3. boost::shared_ptr: 100% 4. custom SharedPtr: 59% 5. custom SharedPtr + rc pool): 53% This scenario has a lot more reference-increments, and a lot less deallocation of the reference-counted instance. Nonetheless, the results are virtually identical, in relative terms (the actual times taken are about 50% across the board). B (multi-threaded) -------------------- 1. Synesis self-counter: 71% 2. Custom self-counter: 41% 3. boost::shared_ptr: 100% 4. custom SharedPtr: 32% 5. custom SharedPtr + rc pool): 31% Pretty much the same as with B (single-threaded). The results are the same, in relative terms Up to this point, the results are mildly interesting, but nothing earth shattering. C - (Times are total time, and thread times, respectively) 1. Synesis self-counter: 2.9% 39% 2. Custom self-counter: 2.7% 35% 3. boost::shared_ptr: 100% 100% 4. custom SharedPtr: 1.0% 13% 5. custom SharedPtr + rc pool): 1.2% 15% This scenario tells quite a different tail indeed. One thing to note is that because of the degree of reference-sharing between the ptr-storing vectors in the 20 threads, the ref-count-pool becomes a pessimisation as a result of the number of actual smart pointer instances being created exceeding the arbitrary 1024 cached ref-counts available. It's likely that this number would not be exceeded in real-world scenarios, so the pool is probably still worth having. Naturally, the very big question is: what's happening to boost::shared_ptr? In thread times, it performs 2-3 times worse than in single threaded scenarios (irrespective of whether they are built for single or multiple threads). However, in total time, it performs extremely poorly, up to 100 times slower than some of the others! Clearly, there is some serious contention involved. There are three possibilities I have thought of: (i) It performs very poorly in multi-threaded sharing scenarios. If this is the case, can any people more expert in Boost than me proffer an explanation as to precisely the problem, and is any such person motivated to try and address the issue? (ii) The particular sharing scenario is well away from any real-world scenario in which boost::shared_ptr is likely to be used. If this is the case, why do four different implementations, representing two very different reference-counting strategies, all maintain a consistent relative performance? What are the scenarios for which boost::shared_ptr is optimised? (iii) There are environment settings for which boost::shared_ptr will perform much better. If that's so, can someone help me out with suggestions of what defines to make? Why are these settings not the default. I'd be very interested in your thoughts. Many thanks in advance. -- Matthew Wilson STLSoft moderator (http://www.stlsoft.org) Contributing editor, C/C++ Users Journal (www.synesis.com.au/articles.html#columns) "You can tell a Yorkshireman, but you can't tell him much!" -- Uncle Michael ---------------------------------------------------------------------------- --- _______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

1 0

Property map deletion
by David Ohlemacher 14 Nov '03

14 Nov '03

Hi all. Partway through my app's execution, I find that I no longer need the values in a custom property map (edge). The map potentially may be quite large. What is the best way to empty it? Thanks, -d

3 7