October 2005 - Boost-users - lists.preview.boost.org

[Parameter library]
by Sohail Somani 11 Oct '05

11 Oct '05

Is the prescribed method for the parameter library thread safe? // appears to modify global variables "graph", etc core::depth_first_search(( graph = 'G', visitor = 2, root_vertex = 3.5, index_map = "hello, world", color_map = false)); Thanks

2 1

[iostreams] Problems using filtering_stream and code_converter
by Niklas Wiberg 11 Oct '05

11 Oct '05

Hi, I'm using boost 1.33.0 on windows (VC++6). I would like to use a code_converter to convert internal std::wstring's (which I understand are ucs2 in windows/vstudio, but that shouldn't matter here I guess) to utf-8 when saving to file. Here is the failing code: <code> void main() { typedef boost::iostreams::code_converter<boost::iostreams::file_sink, WideCharToUtf8Cvt> my_sink; std::wstring teststr(L"Blah blah."); boost::iostreams::filtering_wostream out; boost::iostreams::file_sink myFile("out.txt"); out.push(my_sink(myFile)); out << teststr; } </code> WideCharToUtf8Cvt is my own codecvt implementation. The problem is that I get a crash as the stream is being destroyed. The call stack is too deep to include here, but one of the last things are: policy_type& dev() { return **dev_; } in boost.iostreams.detail.code_converter_impl which calls boost::iostreams::detail::optional<boost::iostreams::detail::concept_adapter<boost::iostreams::basic_file_sink<char> > >::operator*() operator*() asserts on initialized_ which is false in this case. I've tried both with vc++6 sp5 and vc++ 7.1 with the same results. Also, there is no difference if I use this typedef for my_sink: typedef boost::iostreams::code_converter<boost::iostreams::file_sink> my_sink; so I don't think my codecvt implementation is to blame... The file out.txt actually gets filled with the expected data. Am I going about this the completely wrong way or am I just missing some important detail? Any help would be greatly appreciated, Niklas Wiberg

2 2

BOOST_STATIC_ASSERT
by Edward Grace 11 Oct '05

11 Oct '05

Dear All, I am encountering a problem with BOOST_STATIC_ASSERT, it is producing compile-time errors that are a little cryptic. When using BOOST_STATIC_ASSERT such as class rtm : public matrix<ElementType> { BOOST_STATIC_ASSERT(::boost::is_float<ElementType>::value || ejg::boost::is_complex<ElementType>::value); public: ... ... .. Instantiation of the rtm class with a non float or non complex type, such as: ejg::oned::rtm<char> M; causes the following compiler error. error: invalid application of sizeof to incomplete type boost::STATIC_ASSERTION_FAILURE<false> Boost 1.32.0 g++ 4.0.1 Obviously generating a compiler error is the correct behaviour, I want to make sure that only complex or float type objects are instantiated. The problem is that the compiler error is rather cryptic. >From the documentation of boost http://www.boost.org/doc/html/boost_staticassert.html, I should be seeing Illegal use of STATIC_ASSERTION_FAILURE<false> Is there any way this can be improved by generating the error or is it too much of a compiler issue? -ed

5 6

[Boost.Build] unable to cross-compile 1_33_0
by BRIDGES Dick 11 Oct '05

11 Oct '05

My apologies in advance if this question is only for the jamboost mailing list but I very much need to cross-compile boost_1_33_0 and I can't seem to make this version work. # bjam --v2 --version Boost.Build V2 (Milestone10) Boost.Jam 03.01.11 (I haven't received any responses on the jamboost mailing list so I assume I'm formulating my question incorrectly. If there is something wrong with the way I'm asing this question, would someone please point it out to me?) I am trying to compile a host (X86, linux) and a target (ARM9) version of the boost_1_33_0 libraries using the "standard" (Core3) gcc and a gcc-arm cross-compiler. Can someone point me to the documentation that is "authoritative" for 1_33_0 and describes how to configure [which files] and invoke bjam for multiple compilers on the same computer? I found (and read) *LOTS* of historical material describing how to configure and invoke bjam for v1 and v2 for prior versions. Vladimir Prus provided an answer (in the form of a user-config.jam mod) for 1_32_0 last April... <extract> I'd use this: using gcc : 3.4 ; using gcc : 3.4arm : /path/one/arm-g++ : <other stuff> ; using gcc : 3.4ppc : /path/two/ppc-g++ : <other stuff> ; and them bjam gcc-3.4 gcc-3.4arm gcc-3.4ppc </extract> ...which worked fine with 1_32.0, but I can't get it to work with the 1_33_0 version. No matter how I invoke bjam, it seems to interpret the compiler as the target. bjam gcc-3.4arm => "don't know how to make gcc-3.4arm" bjam --v2 gcc-3.4arm => "could not find main target gcc-3.4arm" bjam "-sTOOLS=gcc-3.4arm" => "gcc-3.4arm-tools-jam: No such file..." I've been reduced to "trying things." Any advice or suggestions would be greatly appreciated. Regards, "Desperate" Dick Bridges

1 0

Re: [Boost-users] BOOST XML serialisation c++ libraries with C#
by Prasad, Shyam 11 Oct '05

11 Oct '05

Hi Has anyone tried using BOOST XML serialisation with C#? Is that feasible? Basically the scenario is a C# front end client talking to a C++ back end server via XML messages both sides using Boost XML Serialisation / Deserialisaion framework. Regards SP

1 0

[serialization][any]Do we have serialization support for boost::any?
by Li Lirong 11 Oct '05

11 Oct '05

Hi, Do we have serialization support for boost::any? If not, how is it possible to serialize boost::any? Regards, Lirong

6 20

[BGL] Multiple adaptations of graph
by Andreas Brinck 11 Oct '05

11 Oct '05

Hi, I have a simple directed graph class (let's call it MyGraph) which I want to adapt for BGL algorithms. I've defined the necessary traits and functions (in_edges, target etc.) and everything works fine. Problem is I want to be able to treat the graph as an undirected graph as well when using some of the BGL algorithms, but since the graph_traits are only templatized on the graph type I can't define a new set of traits since graph_traits<MyGraph> is already defined. The only solution I've come up with so far (after 10 minutes of thinking :) is to define a new class, MyDirectedGraph, which is an exact copy of MyGraph and define a set of traits for this new class. And then when I want to use the undirected version of the algorithms on my MyGraph object I cast it to a MyDirectedGraph. Somehow I have a feeling there must be a more robust and elegant way to solve this problem. Regards /A.B. ########################################### This message has been scanned by F-Secure Anti-Virus for Microsoft Exchange. For more information, connect to http://www.f-secure.com/

1 0

Bug in boost::regex_match 1.33.0 Win64/cl1400 and Linux/gcc323
by rlop＠gmx.de 11 Oct '05

11 Oct '05

Hello, After upgrading from boost 1.32 to boost 1.33.0, we get an exception in boost::regex_match. The exception is std::runtime_error: Regular expression too big , and only under Win64/cl1400 and RHEL/gcc323 Changing the regex slightly sometimes avoids the exception... Attached a small example, which can be used to reproduce the exception. #include <boost/regex.hpp> int main(void) { boost::regex validValue = boost::regex("([x]{0,64})?"); boost::regex_match("T",validValue); return 0; } best regards Robert Lopuszanski -- 5 GB Mailbox, 50 FreeSMS http://www.gmx.net/de/go/promail +++ GMX - die erste Adresse f�r Mail, Message, More +++

3 3

Arrays in Boost::Python
by Wyatt Greenway 11 Oct '05

11 Oct '05

How do I acomplish this: class A { public: int *Buffer; vector<string> Array; A() { this->Buffer = new int[10]; /* Set items in Buffer */} } /* Boost code */ class_<A>("A") .def("Buffer", &A::Buffer) .def("Array", &A::Array) ; /*-----*/ And in python it would look like so: > MyA = A() > print MyA.Buffer[0] > MyA.Array.append("Hello world!") > print MyA.Array[0] ??????? Thanks in advance for any help

2 1

boost::any Question
by Piyush Kapadia 10 Oct '05

10 Oct '05

I am new user of Boost, I tried to use boost::variant first, but in comparison to boost::any performance of boost::variant appears to be around 4 times slower, so settled for boost::any. Also boost::any has cleaner interface but I don't know how to query current type information for boost::any ? My code looks like following map<unsigned int, boost::any> result_table; result_table[1] = 543.434; vector<int> vec; .. Some vec population code; result_table[2] = vec; Now I need to develop a stringify function for result_table; I can fetch the results back as boost::any value = result_table[2]; But I need to know the type of value, so that I can treat it accordingly to stringify; I know there exists type_traits, but very limited examples of type_traits is provided on Boost documentation and particularly, I am clueless about how to use type_traits in above context to find out what is the current type returned by result_table[int] Also I want to know if Boost has any Hash map implementation that can be superior in performance compare to regular std::map ?? Thanks, Piyush

2 6

[BGL] Simple question about accessing vertex property
by Giulio Veronesi 10 Oct '05

10 Oct '05

Hi, How do I access the vertex_index property of a vertex? Is it necessary to use the get function? More precisaly, let us consider the following: typedef adjacency_list<vecS, vecS, bidirectionalS, no_property, property<edge_weight_t, int> > TGraph; typedef graph_traits<TGraph>::vertex_descriptor TNode; Suppose that 'vertex' is a TNode. Is it correct to do int idx = vertex; to get in idx the vertex_index of vertex? Thanks in adavnce, Giulio

3 2

Boost Serialization Question.
by Piyush Kapadia 10 Oct '05

10 Oct '05

I am able to use Boost XML and TEXT serialization but Binary Serialization fails to read file, though it can write binary file. Is Binary Serialization fully Supported and if so does it work with STD container classes ?

3 2

Re: [Boost-users] __stdcall bind-s with boost::bind andboost::lambda
by Bulygin, Sergey 10 Oct '05

10 Oct '05

boost::lambda::bind<RetType>(..) works like a charm. :) Thanks a lot, Peter! Sergey Peter Dimov wrote: > To bind to __stdcall calling convention function with boost::bind, > there is a macro to define: BOOST_BIND_ENABLE_STDCALL. How can I do > the same (bind to __stdcall-type function) with boost::lambda::bind? You may want to try specifying the return type explicitly as in bind<void>(my_stdcall_function, _1, 0) and see if it helps.

1 0

__stdcall bind-s with boost::bind and boost::lambda
by Bulygin, Sergey 10 Oct '05

10 Oct '05

To bind to __stdcall calling convention function with boost::bind, there is a macro to define: BOOST_BIND_ENABLE_STDCALL. How can I do the same (bind to __stdcall-type function) with boost::lambda::bind? Thank you, Sergey

2 1

Type List Iteration?
by Duane Murphy 10 Oct '05

10 Oct '05

How do I iterate over a type list and call a function? MPL? I have a base class BASE that has several subclasses A, B, C that have special capabilities, but no way to identify those capabilities aside from the type. I would like to determine if an object of type BASE is also of one of the special subclasses: ie void f( BASE* base) { if ( dynamic_cast< A* >( base ) != NULL ) { some_function( dynamic_cast< A* >( base ) ); } else if ( dynamic_cast< B* >( base ) != NULL ) { some_function( dynamic_cast< B* >( base ) ); } else if ( dynamic_cast< C* >( base ) != NULL ) { some_function( dynamic_cast< C* >( base ) ); } } I think you can see why I'd like a type list iterator solution to this. I have arranged for the some_function to be a template as well to make things easier. What I am looking for is the equivalent of find_if() but for types and done at run time. I looked over MPL, but MPL seems to apply at compile time without a door to the run time world. :-) Thanks for your helpful tips! ...Duane

3 7

[mpl] nested lambdas?
by Geoffrey Romer 10 Oct '05

10 Oct '05

Suppose I'm tying to build a nested lambda expression, like so (typename...::type omitted for simplicity): lambda<higher_order_func<lambda<foo<_1> >, _2> > - Is this legal? - Does the _1 refer to the first argument of the inner lambda, or of the outer lambda? - Can the above be overridden? I.e. what if I want to refer to the arguments of both lambdas within the inner lambda?

2 3

[iostreams] Closing a filtering_ostream
by Paul Giaccone 10 Oct '05

10 Oct '05

How do you close a filtering_ostream object once you have finished writing to it? In my program below, I would expect the file "hello.bz2" to have been written once the close() function has been called, but it is not the case. The contents are not actually written until I exit the scope in which the filtering_ostream object is defined (at which point the destructor presumably does the job). (The file "some_text.txt" contains just the text "hello world!" so I would expect it not to be written until the file is closed.) I would like to be able to do this explicitly and not have to rely on the destruction of the object - could you explain how can this be done? Another question: what should the second parameter to close() be? I've used BOOST_IOS::trunc here, but is there some other more appropriate value? Thanks, Paul Giaccone #include <fstream> #include <iostream> #include <boost/iostreams/copy.hpp> #include <boost/iostreams/filter/bzip2.hpp> #include <boost/iostreams/device/file.hpp> #include <boost/iostreams/filtering_stream.hpp> int main(void) { { boost::iostreams::filtering_ostream out; out.push(boost::iostreams::bzip2_compressor()); out.push(boost::iostreams::file_sink("hello.bz2", BOOST_IOS::trunc)); std::ifstream input_stream("some_text.txt"); boost::iostreams::copy(input_stream, out); input_stream.close(); if (!boost::iostreams::flush(out)) { std::cerr << "failed to write file" << std::endl; } boost::iostreams::close(out, BOOST_IOS::_Dummy_enum_val); } //exiting scope closes the stream; close() seems not to boost::iostreams::filtering_istream in; in.push(boost::iostreams::bzip2_decompressor()); in.push(boost::iostreams::file_source("hello.bz2")); boost::iostreams::copy(in, std::cout); std::cout.flush(); return 0; }

3 10

[mpl] possible bug in insert<set, x>
by Geoffrey Romer 10 Oct '05

10 Oct '05

I'm having trouble with insert<> as called on a set. Specifically, it doesn't seem to be possible to iterate over sets built with insert<>. The following code, for example, works correctly (or at least, compiles without errors): --- typedef set<double, char> m_set; typedef transform<m_set, lambda<identity<_1> > >::type result; --- but the following (which as I understand it should be equivalent) fails to compile: --- typedef insert<insert<set<>, double>::type, char>::type m_set; typedef transform<m_set, lambda<identity<_1> > >::type result; --- The error consists basically of the compiler complaining there's no member type next_ in set<mpl_::na, mpl_::na ...>. I've provided the full error log in an attachment (I'm using g++ 4.0.2 and Boost 1.33.0). The bug isn't specific to transform; I've encountered it when using other algorithms when they iterate over a set built with inserts. It's possible this problem relates to the problem discussed in this message: (http://thread.gmane.org/gmane.comp.lib.boost.devel/116988) which also concerns non-identical behavior for sets built with the two methods. I hope I'm not doing something stupid, I'm starting to feel like a broken record here...

2 2

Boost Ref Counting & Smart Pointers
by Piyush Kapadia 10 Oct '05

10 Oct '05

Does Boost Library internally does Reference counting for most of its internal objects ? Does it also use it's shared_ptr kind of Smart pointer implementation internally ? Sorry for my ignorance, but isn't there any cleaner Smart Pointer implementation where we don't have to use new every time initiating object and smart pointer becomes part of the class rather that separate implementation like following - shared_ptr sp(new int(5));

4 3

A error to link boost program_options(mingw)
by Huifeng Shen 10 Oct '05

10 Oct '05

I try to compile the first.cpp example of boost program_options library I use the command g++ first.cpp -lboost_program_options-mt. It fails when linking. But I also try use the same command to compile simple_ls.cpp of filesystem, boost_back_inserter_example.cpp of iostream, days_alive.cpp of date_time(The library link is also different). It compiles and works as expected. I use depends.exe of VS.Net 2003 to prove the exe is depended on the dll file. The error of program_options is as below Info: resolving vtable for boost::program_options::variables_mapby linking to __imp___ZTVN5boost15program_options13variables_mapE (auto-import) Info: resolving vtable for boost::program_options::value_semantic_codecvt_helper<char>by linking to __imp___ZTVN5boost15program_options29value_semantic_codecvt_helperIcEE (auto-import) Info: resolving boost::program_options::arg by linking to __imp___ZN5boost15program_options3argE (auto-import) Info: resolving vtable for boost::program_options::validation_errorby linking to __imp___ZTVN5boost15program_options16validation_errorE (auto-import) C:\DOCUME~1\ADMINI~1\LOCALS~1\Temp/cckbcaaa.o(.text$_ZN5boost15program_optio ns13variables_mapD1Ev[boost::program_options::variables_map::~variables_map( )]+0x3a):first.cpp: variable 'vtable for boost::program_options::variables_map' can't be auto-imported. Please read the documentation for ld's --enable-auto-import for details. C:\DOCUME~1\ADMINI~1\LOCALS~1\Temp/cckbcaaa.o(.text$_ZN5boost15program_optio ns29value_semantic_codecvt_helperIcED2Ev[boost::program_options::value_seman tic_codecvt_helper<char>::~value_semantic_codecvt_helper()]+0xb):first.cpp: variable 'vtable for boost::program_options::value_semantic_codecvt_helper<char>' can't be auto-imported. Please read the documentation for ld's --enable-auto-import for details. C:\DOCUME~1\ADMINI~1\LOCALS~1\Temp/cckbcaaa.o(.text$_ZN5boost15program_optio ns29value_semantic_codecvt_helperIcEC2Ev[boost::program_options::value_seman tic_codecvt_helper<char>::value_semantic_codecvt_helper()]+0x16):first.cpp: variable 'vtable for boost::program_options::value_semantic_codecvt_helper<char>' can't be auto-imported. Please read the documentation for ld's --enable-auto-import for details. C:\DOCUME~1\ADMINI~1\LOCALS~1\Temp/cckbcaaa.o(.text$_ZN5boost15program_optio ns16validation_errorD2Ev[boost::program_options::validation_error::~validati on_error()]+0x3a):first.cpp: variable 'vtable for boost::program_options::validation_error' can't be auto-imported. Please read the documentation for ld's --enable-auto-import for details. C:\DOCUME~1\ADMINI~1\LOCALS~1\Temp/cckbcaaa.o(.text$_ZN5boost15program_optio ns16validation_errorD1Ev[boost::program_options::validation_error::~validati on_error()]+0x3a):first.cpp: variable 'vtable for boost::program_options::validation_error' can't be auto-imported. Please read the documentation for ld's --enable-auto-import for details. C:\DOCUME~1\ADMINI~1\LOCALS~1\Temp/cckbcaaa.o(.text$_ZN5boost15program_optio ns16validation_errorC1ERKSs[boost::program_options::validation_error::valida tion_error(std::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)]+0x53):first.cpp: variable 'vtable for boost::program_options::validation_error' can't be auto-imported. Please read the documentation for ld's --enable-auto-import for details. C:\DOCUME~1\ADMINI~1\LOCALS~1\Temp/cckbcaaa.o(.text$_ZN5boost15program_optio ns16validation_errorC2ERKS1_[boost::program_options::validation_error::valid ation_error(boost::program_options::validation_error const&)]+0x53):first.cpp: variable 'vtable for boost::program_options::validation_error' can't be auto-imported. Please read the documentation for ld's --enable-auto-import for details. collect2: ld returned 1 exit status I want to know how to link the dynamic library of program_options

2 1

Trouble with deriving from boost::archive
by Haskins, Gregory 09 Oct '05

09 Oct '05

I posted this earlier, but I don't think it went through all the way. I apologize if this is a duplicate: ------------------------------------------------------------------------ -------------------- Hi All, Boost newbie here: I am having problems with getting serialization to work with my app. I have a top-level "environment" context that objects in my app need to be associated with. I thought about various ways to accomplish this and I thought the best way might be to derive an Archive class that maintains a reference to the context and use custom save_construct_data/load_construct_data overrides. However, I am hitting a snag and I don't quite understand what is happening. I derive an input and output archive from text_iarchive/text_oarchive respectively and I use this archive context. However I get the following error in compilation: db.hh: In function `void boost::serialization::load_construct_data(Archive&, FlatFileDb::Db*, unsigned int) [with Archive = boost::archive::text_iarchive]': /ccase/cca/vendor/boost/boost/serialization/serialization.hpp:170: instantiated from `void boost::serialization::load_construct_data_adl(Archive&, T*, unsigned int) [with Archive = boost::archive::text_iarchive, T = FlatFileDb::Db]' /ccase/cca/vendor/boost/boost/archive/detail/iserializer.hpp:303: instantiated from `void boost::archive::detail::pointer_iserializer<T, Archive>::load_object_ptr(boost::archive::detail::basic_iarchive&, void*&, unsigned int) const [with T = FlatFileDb::Db, Archive = boost::archive::text_iarchive]' /ccase/cca/vendor/boost/boost/archive/detail/iserializer.hpp:209: instantiated from here db.hh:84: error: 'class boost::archive::text_iarchive' has no member named ' m_env' Somehow the system is assigning <class Archive> to 'class boost::archive::text_iarchive' even though I am creating the archive as something derived from text_iarchive. The weird thing is that this happens even if I comment out the code the invokes the actual archiving. Its almost like there is some default value of Archive that gets assigned once I include the text_iarchive.hpp file. Any ideas on how to stop this from happening? Is there a better way to accomplish my goal? Thanks in advance! -Greg PS: Here are some code snippets of what I am doing. I have a header file called "archive.hh" with ----------------------------------------------------------- #include <boost/archive/text_iarchive.hpp> #include <boost/archive/text_oarchive.hpp> namespace FlatFileDb { class Env; class OArchive : public boost::archive::text_oarchive_impl<OArchive> { public: OArchive(Env &env, std::ostream & os, unsigned int flags = 0) : boost::archive::text_oarchive_impl<OArchive>(os, flags), m_env(env) {} ~OArchive(){} Env &m_env; }; class IArchive : public boost::archive::text_iarchive_impl<IArchive> { public: IArchive(Env &env, std::istream & is, unsigned int flags = 0) : boost::archive::text_iarchive_impl<IArchive>(is, flags), m_env(env) {} ~IArchive(){} Env &m_env; }; }; -------------------------------------------------- and then I have another header file (db.hh) with: -------------------------------------------------- <.../snip...> namespace boost { namespace serialization { template<class Archive> inline void save_construct_data( Archive & ar, const FlatFileDb::Db *t, const unsigned int file_version ){ // save data required to construct instance ar << t->m_name; } template<class Archive> inline void load_construct_data( Archive & ar, FlatFileDb::Db* t, const unsigned int file_version ){ // retrieve data from archive required to construct new instance std::string name; ar >> name; // invoke inplace constructor to initialize instance of my_class ::new(t)FlatFileDb::Db(ar.m_env, name); <-- !!! blows up here !!!y } }} // namespace ...

2 1

[autolink] Debug + Multi Threaded DLL
by Bruno Martínez 08 Oct '05

08 Oct '05

Hi. I have a project that uses boost.threads and sdl, under vc7.1. Because of sdl, I have to link with the non-debug multi threaded dll version of the runtime libraries, even on debug builds. However, autolinking picks the debug version of boost.threads so I get conflicts. Is it right to disable autolinking and force a non debug version of boost.threads? Any ohter idea? Thanks, Bruno

2 1

Re: [Boost-users] ld warnings when building boost on Fedora Core 4 with gcc 3.4.4
by Foster, Gareth 08 Oct '05

08 Oct '05

> So what compiler does work with boost on Fedora Core 4? Just the gcc 4.0 series? Have you investigated the gcc-compat packages? I know that if you want to install the binary version of realplayer from real.com, you have to install these *-compat packages. It may be the case that these packages cover development as well as just the running of old binaries. Worth a look to see what is there at any rate. Gaz

3 4

Re: [Boost-users] aligned_storage / alignment_of / type_with_alignment
by Gottlob Frege 07 Oct '05

07 Oct '05

> > Date: Fri, 07 Oct 2005 08:32:34 +0200 > From: Joaqu?n M? L?pez Mu?oz <joaquin(a)tid.es> > Subject: Re: [Boost-users] aligned_storage / alignment_of / > type_with_alignment > To: boost-users(a)lists.boost.org > Message-ID: <43461682.7A1C50F0(a)tid.es> > Content-Type: text/plain; charset=iso-8859-1 > > Gottlob Frege ha escrito: > > This is *almost* correct. There's a typo in 1.33 Boost.TypeTraits docs > (corrected > in 1.33.1) by which the explanations on type_with_alignment and > aligned_storage > are swapped. So, what you want is > > template <typename T> > struct uninitted > { > aligned_storage<sizeof(T), alignment_of<T>::value>::type myT; > > void init_later(some_params) > { > new (&myT) T(some_params); > } > }; > > Apart from this, your construct is OK and works as you expect. Actually, > this is > what aligned_storage is meant to be used for. Best, > > Joaqu?n M L?pez Mu?oz > Telef?nica, Investigaci?n y Desarrollo > > OK, thanks. By its name, I thought aligned_storage sounded like the right thing (and thus I included it in the subject line) but I wasn't sure how it fit in. I think aligned_storage<sizeof(T), alignment_of<T>::value>::type deserves its own template - maybe 'storage_for' ? Anyhow, I think I have everything I need, so away I go.... Tony

1 0

[mpl] trouble with conditionals
by Geoffrey Romer 07 Oct '05

07 Oct '05

I'm struggling with a problem of how to correctly use conditionals in MPL. In a nutshell, the problem is that the compiler greedily evaluates all parts of a conditional, even the branch that isn't followed. This is a problem not just for performance, but because sometimes the unused branch might not compile. I know that eval_if is supposed to address this problem, but unless I'm missing something, it doesn't seem to be a complete solution. It delays evaluating the branches until one of them is selected, but the compiler still evaluates the *arguments* to both branches. If the arguments themselves don't compile, I'm screwed, despite the fact that the parts of my code that are actually relevant (the branches that get used) are perfectly correct. How do I resolve this problem? Here's a toy example, a metafunction which, given a map, finds the value associated with the key "char" and, if it exists, increments the associated value (which we assume is numeric): template <typename m_map> class toy_metafunction { typedef typename if_<typename has_key<m_map, char>::type, typename insert<m_map, pair<char, typename plus<int_<1>, at<m_map, char> >::type > >::type, m_map >::type type; }; typedef typename toy_metafunction<map< > >::type result; This does not compile, because the compiler attempts to evaluate plus<int_<1>, void_>. Notice that replacing if_ with eval_if does not help, because it's not the evaluation of insert that's the problem. Thanks in advance for any help.

4 12