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

Specifying an alternate 'operator>>' for istream_iterator
by remi.chateauneu＠gmx.de 18 Jul '05

18 Jul '05

For istream_iterator or ostream_iterator, the operator '<<' and '>>' are used to serialize the data. Have anyone ever needed another operator (typically for serializing to another format, for example binary) ? To implement this, it seems possible to add the serializing operator to istream_iterator, as an extra template argument, which would by default use the existing operator>>. In other words : template<typename _Tp, typename _CharT = char, typename _Traits = char_traits<_CharT>, typename _Dist = ptrdiff_t> class istream_iterator { ... ... would be replaced by something like : template<typename _Tp, typename _CharT = char, typename _Traits = char_traits<_CharT>, typename _Dist = ptrdiff_t, typename _SerializingOperator = XXXXX > class istream_iterator { ... XXXXX would represent the existing operator>> for this data type. On the other hand, 'operator>>' is not a method of class _Tp, but a friend method. So, how to specify it are a default template argument ?

2 1

documentation lacuna in boost::unit_test v1.33?
by Serge Skorokhodov 18 Jul '05

18 Jul '05

Hi, I run into a problem switching from unit_test 1.32 to 1.33. The code for test log configuration must be changed from something like using namespace boost::unit_test; unit_test_log::instance().set_log_threshold_level(...) etc. to something like unit_test_log.instance().set_threshold_level(...) etc; I believe that it worth mentioning in release notes;) Another question: it is still impossible to link main implementation automatically in the lib rather than pull it in directly with #include "boost/test/included/unit_test_framework.hpp"? -- Serge

1 0

Re: [Boost-users] tuple
by Bjorn.Karlsson＠readsoft.com 17 Jul '05

17 Jul '05

Hello Jim, > "Jim Pritts" <jpritts(a)sdf.lonestar.org> wrote in message > > Is it possible to do something like the following: > > typedef boost::tuple<double, int, int> element; > > std::vector<element> d; > > std::vector<element>::iterator i = std::max_element(d.begin(), > d.end(), > > bind(std::max<double>,bind(get<0>,_2),_1)); Good question! The above won't compile, in part because get() requires three template arguments, so you can't get it's address using &get<0>. Furthermore, it's not trivial to explicitly instantiate get() correctly or to bind to tuple's member function get(), which is why I'd suggest creating a helper class--a function object--for selecting one of the elements of a tuple. To use it with Boost.Bind without explicitly specifying the return type, you'd need to do this: template <typename T,int i> class select { public: typedef typename boost::tuples::element<i,T>::type result_type; template <typename U> result_type operator()(const U& u) const { // Select the ith element of the tuple T and return it return u.get(); } }; The required typedef, result_type, that enables Boost.Bind to deduce the return type of the function call operator, uses a helper class from Boost.Tuples, the class template element, which retrieves the type of a tuple element using its index. With a class template like select, you could use Boost.Bind with max_element like so: int main() { typedef boost::tuple<double,int,int> element; std::vector<element> d; // Code for adding elements omitted std::vector<element>::iterator i= std::max_element( d.begin(), d.end(), boost::bind(std::less<double>(), boost::bind(select<element,0>(),_1), boost::bind(select<element,0>(),_2))); } You'll also note that I've adjusted the binders from the example you gave. There are three binders in the call to max_element; the two innermost binders extract the first element of the tuples _1 and _2 (to be substituted with the actual arguments), and the third binder binds std::less<double>. The result of all these binders is a binary predicate, which is what std::max_element expects. Note that in your example, you had used std::max(), which does not work well as a binary predicate for max_element(). Although the above version of select works, it's rather inflexible and sometimes inefficient, because the function call operator always returns a copy of the ith element of the tuple. What if you need to assign a value to that element? To enable that, you'll need a new version of select, a class template that is only parameterized on the element index, and with two parameterized function call operators that return a copy of the element or a reference to the element, depending on the constness of the tuple. Using Boost.Type_traits, it's easy to add a reference to a type without risking forming a reference to reference (those are currently illegal in C++): template <int i> class select { public: template <typename T> typename boost::tuples::element<i,T>::type operator()(const T& t) const { // Select the ith element of the tuple t and return it return t.get(); } template <typename T> typename boost::add_reference<typename boost::tuples::element<i,T>::type>::type operator()(T& t) const { // Select the ith element of the tuple t and return it return t.get(); } }; This version of select works fine, but Boost.Bind is no longer able to deduce the return type of the function call operator(s), so you'll need to provide that information explicitly when using select: std::vector<element>::iterator i= std::max_element( d.begin(), d.end(), boost::bind(std::less<double>(), boost::bind<double>(select<0>(),_1), boost::bind<double>(select<0>(),_2))); If you think we're all done here, think again. ;-) When using Boost.Lambda, which also offers binders, there's an extended system for deducing return types. It comes in the form of a nested class template called sig that is responsible for providing the correct return type depending on which function call operator is being called, i.e., depending on the actual arguments to the function call operator(s). That is precisely the kind of mechanism that you need to solve the problem with the multiple function call operators of select. The following version of select is the most user-friendly version, but it is a little more complex. It uses Boost.Mpl for a compile-time if, Boost.Type_traits for adding a reference to the element type, Boost.Lambda's return type deduction system, and of course Boost.Tuples to operate on the tuple and retrieve the ith element's type: template <int i> class select { public: // Provide the return type template <typename Arg> class sig { typedef typename boost::tuples::element<1,Arg>::type argument_type; typedef typename boost::tuples::element<i,argument_type>::type element_type; public: // Select the correct return type; depends on the typedef typename boost::mpl::if_<boost::is_const<element_type>, element_type, typename boost::add_reference<element_type>::type>::type type; }; template <typename T> typename boost::tuples::element<i,T>::type operator()(const T& t) const { // Select the ith element of the tuple t and return it return t.get(); } template <typename T> typename boost::add_reference<typename boost::tuples::element<i,T>::type>::type operator()(T& t) const { // Select the ith element of the tuple t and return a reference to it return t.get(); } }; Using this version of select and Boost.Lambda's binders, you'll have both an easy-to-use interface and a powerful way of selecting tuple elements, both for lvalues and rvalues. Here's a complete program to demonstrate your example again. Note that the class template select requires you to include the headers "boost/tuple/tuple.hpp", "boost/type_traits.hpp", and "boost/mpl/if.hpp". For Boost.Lambda's bind(), you'll need to include "boost/lambda/bind.hpp". #include <iostream> #include <vector> #include <algorithm> #include <functional> #include "boost/tuple/tuple.hpp" #include "boost/type_traits.hpp" #include "boost/mpl/if.hpp" //#include "boost/lambda/bind.hpp" template <int i> class select { public: // Provide the return type template <typename Arg> class sig { typedef typename boost::tuples::element<1,Arg>::type argument_type; typedef typename boost::tuples::element<i,argument_type>::type element_type; public: // Select the correct return type; depends on the typedef typename boost::mpl::if_<boost::is_const<element_type>, element_type, typename boost::add_reference<element_type>::type>::type type; }; template <typename T> typename boost::tuples::element<i,T>::type operator()(const T& t) const { // Select the ith element of the tuple t and return it return t.get(); } template <typename T> typename boost::add_reference<typename boost::tuples::element<i,T>::type>::type operator()(T& t) const { // Select the ith element of the tuple t and return a reference to it return t.get(); } }; int main() { typedef boost::tuple<double,int,int> element; std::vector<element> d; // Code for adding elements omitted using namespace boost::lambda; std::vector<element>::iterator i= std::max_element( d.begin(), d.end(), bind(std::less<double>(), bind(select<0>(),_1), bind(select<0>(),_2))); std::cout << "max element: " << (*i).get<0>(); } Regards, Bjorn Karlsson

1 0

Re: [Boost-users] [variant] not copy-constructable
by Bjorn.Karlsson＠readsoft.com 17 Jul '05

17 Jul '05

> From: Felipe Magno de Almeida [mailto:felipe.m.almeida@gmail.com] > > What I wanted is a vector of variant<object1, object2> and I dont want > object1 and object2 to be copied because it is expensive. But Im not > figuring out how to do this... tried boost::ref either with no > success... :/ Boost.Smart_ptr (http://www.boost.org/libs/smart_ptr/smart_ptr.htm) can help: use boost::shared_ptr, which is a reference-counted smart pointer. typedef boost::variant<object1,object2> variant_type; std::vector<boost::shared_ptr<variant_type> > vec; boost::shared_ptr<variant_type> v(new variant_type(object1())); vec.push_back(v); // No copying of the variant Bjorn Karlsson

3 2

filesystem.dll built with BOOST_ALL_DYN_LINK
by Larry E. Ramey 16 Jul '05

16 Jul '05

Hello all, I just pulled boost from the CVS repository and built the filesystem library. bjam -d2 --with-filesystem This works and links great on my linux system. When I go over to windows with MSVC 7.1 and try to link I get stuck trying to autolink to the static lib. So It looks to me like the proper solution is to uncomment #define BOOST_ALL_DYN_LINK in the boost/config/user.hpp file. So I do that. Now we are trying to link to the .dll. Good. Only problem is I have two undefined symbols. (demangled) boost::filesystem::path::iterator::increment boost::filesytem::path::iterator::decrement So what did I do wrong? I can't believe I'm the first person to have this problem, but I couldn't find anything in the FAQ or mailing list archives. -- This e-mail and any attachments may contain confidential and privileged information. If you are not the intended recipient, please notify the sender immediately by return e-mail, delete this e-mail and destroy any copies. Any dissemination or use of this information by a person other than the intended recipient is unauthorized and may be illegal. Unless otherwise stated, opinions expressed in this e-mail are those of the author and are not endorsed by the author's employer.

3 3

gcc 4.0 and none_t
by Vincent Torri 16 Jul '05

16 Jul '05

Hello, I have installed boost 1.33, compiled with gcc 4.0. In a program, I use none_t like that: third(subContextBlock.localCtxt) = val( boost::detail::none_t() ) and gcc reports this error: ../../src/parser/grammar/statement.cpp:143: error: 'none_t' is not a member of 'boost::detail' I don't understand at all... Does someone see any problem ? Also, I don't understand the '*' in the definition of none_t. Could someone explain this Thank you Vincent Torri

1 0

Regex Compile Problem
by Chris Coleman 16 Jul '05

16 Jul '05

Hi, While working on a project of ours I have come accross this problem using boost-1.32 on Mac OS X 10.4: It appears to be related to the regex library. <snip> Making all in mantrad source='mantra_request_admin.cpp' object='mantra_request_admin.o' libtool=no \ depfile='.deps/mantra_request_admin.Po' tmpdepfile='.deps/mantra_request_admin.TPo' \ depmode=gcc3 /bin/sh ../../depcomp \ g++ -DHAVE_CONFIG_H -I. -I. -I../.. -I.. -I../libs/sockets -I../libs/sigaware -I../libs/threadpool -Wall -g -D_THREAD_SAFE -c -o mantra_request_admin.o `test -f 'mantra_request_admin.cpp' || echo './'`mantra_request_admin.cpp /opt/local/include/boost/regex/v4/char_regex_traits.hpp: In member function 'uint_fast32_t boost::deprecated::char_regex_traits_i<wchar_t>::lookup_classname(const wchar_t*, const wchar_t*) const': /opt/local/include/boost/regex/v4/char_regex_traits.hpp:68: error: 'boost::re_detail::c_traits_base::<anonymous enum>' is/uses anonymous type /opt/local/include/boost/regex/v4/char_regex_traits.hpp:68: error: trying to instantiate 'template<class T> struct boost::is_integral' make[3]: *** [mantra_request_admin.o] Error 1 make[2]: *** [all-recursive] Error 1 make[1]: *** [all-recursive] Error 1 make: *** [all] Error 2 </snip> The thing is I have not had this problem on any other platform, or indeed 2 other Mac's running an (as far as I'm aware) identical setup. Does anyone have any ideas as to the possible cause of this? Many Thanks Chris This message has been checked for viruses but the contents of an attachment may still contain software viruses, which could damage your computer system: you are advised to perform your own checks. Email communications with the University of Nottingham may be monitored as permitted by UK legislation.

3 2

Re: [Boost-users] Container/Algorithm question
by JOAQUIN LOPEZ MU?Z 16 Jul '05

16 Jul '05

----- Mensaje original ----- De: Elisha Berns <e.berns(a)computer.org> Fecha: Viernes, Julio 15, 2005 7:40 pm Asunto: Re: [Boost-users] Container/Algorithm question > Thanks Joaquin, > > Your code worked perfectly and probably sped up the application by a > factor of 3. Such a large speedup hints at an improvement on algorithmic complexity. My proposal has O(log n) insertion and O(log n) lookup. > One thing though, out of curiosity, what kind of > containerdoes Boose.MultiIndex use internally? Strictly speaking, Boost.MultiIndex does not rely on any external container. A multi_index_container, though, is composed of several indices, each of which is roughly modelled after some std:: container. Ordered indices, which are the ones used in this case, behave like std::set, so the proposal is equivalent to a hand made container using two std::sets --actually, it should be more efficient than the hand made option. > And is there any > way to further > optimize Boost.MultiIndex by specifying what type of container is used > internally? > Yes. Starting with Boost 1.33. Boost.MultiIndex provides hashed indices, which, if properly used, can yield better performance both at insertion and lookup wrt ordered indices. I'm compiler-disabled during weekends, so let me come back to you next Monday with alternatives to my first proposal taking advantage of hashed indices. Have a nice weekend, Joaquín M López Muñoz Telefónica, Investigación y Desarrollo PS: I'm curious, in which context do you need this kind of rather odd container?

2 1

Boost.Spirit bug report [Was: Bug report on Boost 1.32 lib]
by Aleksey Gurtovoy 15 Jul '05

15 Jul '05

Ulzii Luvsanbat <batul(a)microsoft.com> writes: > Hi Aleksey, Hi Ulzii, > > I got this email address from boost.org, if you're not the right person > please forward this to the appropriate folks. The most appropriate place for bug reports would be either of our two main mailing lists (http://www.boost.org/more/mailing_lists.htm) to one of which I'm cc-ing. Please see http://www.boost.org/more/bugs.htm for the more details on the Boost bug reporting policy. > We believe we ran into a > bug in Boost 1.32 library. Here are the details: > > Basically boost_1_32_0\libs\spirit\test\bug_fixes.cpp test is failing > for us with Visual C++ Compiler. > > In range_run.ipp file: > > template <typename CharT> > inline bool > range_run<CharT>::test(CharT v) const > { > if (!run.empty()) > { > const_iterator iter = > std::lower_bound( > run.begin(), run.end(), v, > range_char_compare<CharT>() > ); > > if (iter != run.end() && iter->includes(v)) > return true; > > if (iter != run.begin()) > return (--iter)->includes(v); > } > > return false; > > } > > but range_char_compare::operator() in > boost\spirit\utility\impl\chset\range_run.hpp does not accept the same > types as arguments: > > template <typename CharT> > struct range_char_compare { > > bool operator()(range<CharT> const& x, const CharT y) const > { return x.first < y; } > > bool operator()(const CharT x, range<CharT> const& y) const > { return x < y.first; } > }; > > Now, in our lower_bound implementation (in Debug) we > assume/enforce that Predicate is a valid ordering on the (First, Last) > interval. > > The C++ Standard says: > > 25.3.3 Binary search [lib.alg.binary.search] > > 1 All of the algorithms in this section are versions of > binary search and assume that the sequence being searched is in > order according to the implied or explicit comparison function. > > The explicit comparison function provided to lower_bound in the > repro case cannot be applied on the sequence being searched. Thus, we > cannot check the assumption of 25.3.3.1. > > The test passes if we add one more overload in range_char_compare, > something like: > > template <typename CharT> > struct range_char_compare { > bool operator()(range<CharT> const& x, const CharT y) const > { return x.first < y; } > > bool operator()(const CharT x, range<CharT> const& y) const > { return x < y.first; } > > bool operator()(range<CharT> const& x, range<CharT> const& y) const > { return x.first < y.first; } > > }; > > Can you please confirm this? I believe http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2001/n1313.html covers this exact issue in detail. HTH, -- Aleksey Gurtovoy MetaCommunications Engineering

2 1

very basic boost::filesystem::path question
by Peter Koch Larsen 15 Jul '05

15 Jul '05

Hi boosters I have a very basic question. Why does path("s:/dbgdata/testdata.2") throw the exception boost::filesystem::filesystem_error with the errormessage "boost::filesystem::path: invalid name "s:" in path: "s:/dbgdata/testdata.2"? It certainly looks as if I've overlooked something quite fundamental. I'm using MSVC 8-0 Beta 2 and boost version 1.32.0. Kind regards Peter Koch Larsen

8 8