- Boost - lists.boost.org

RE: [boost] [shared_ptr] InterlockedIncrement rocks!
by Ben Hutchings 22 Sep '04

22 Sep '04

Tomas Puverle <Tomas.Puverle(a)morganstanley.com> wrote: > Ben, > > > Atomic operations are highly platform-dependent, with > > different platforms having different capabilities and many > > requiring use of assembly language rather than providing an > > API for atomic operations. > > Would that be a problem? It means that it's hard to write portable lock-free algorithms. However I don't think this is a problem for shared_ptr because it already has a portable mutex-based implementation of the counter operations which can be used as a fallback where a lock-free implementation is impracticable. <snip> > > Second, shared_ptr has two counters and updating them safely > > without using a lock requires some subtlety. > > On ia32, ia64 and amd64 you can always use DCAS. That isn't available on all IA32; it was introduced with the Pentium. Worse, on a 386SX even 32-bit atomic operations are unavailable. (This is why Windows 95 requires a 386DX and why there have been some binary compatibility problems with glibc and libstdc++ built for 386+ vs 486+.) > On Ultrasparc you could restrict the counter to 32-bits and use CAS > (64-bit). > You can add other platforms as the implementation becomes available, > but use a lock by default. Right. > > It's something > > I meant to implement but hadn't quite got round to. Perhaps > > I could make it work on g++ (and hence most Linux > > configurations) by using libstdc++'s atomic primitives? I > > No. This is not portable enough. > The problem is that libstdc++'s atomic ops come with libstdc++ which > precludes you from using other standard library implementations. Perhaps I could put #define BOOST_STDLIB_LIBSTDCPP3 in <boost/config/stdlib/libstdcpp3.hpp> and then guard any use of <bits/atomicity.h> with #ifdef BOOST_STDLIB_LIBSTDCPP3. <snip> > > don't think they are really public though, so this might be a > > bad idea. > > I think this would be a great idea, and it would make shared_ptr a lot > more acceptable at our company. I meant that using libstdc++'s atomic operations might be a bad idea. Making shared_ptr lock-free wherever possible is indeed a great idea and will hopefully become a great reality!

3 2

platform specific structures
by Matt Hurd 22 Sep '04

22 Sep '04

I'm interested in getting something like "needs_lock" below into the normal traits, perhaps as part of boost thread. It informs you whether of not an operation on such a contiguous block needs a lock to be atomic. For example, on ia32 32 bit aligned ops are atomic, on ia64 it is 64 bit. doubles on ia32 need locking to be atomic, on ia64 they don't. sizeof(void*) works generically for these two platforms as is included in the code below. A default, safety first implementation might return true always for needs lock, or perhaps, true for all sizeof's greater than a byte. "Ifdef" hell for platforms can read to readability problems and big config.hpp equivalents. An alternative is to redirect headers to generic, where possible, or platform specific headers, where necessary, but this may lead to a bit of a maintenance nightmare due to supporting similar code for different platforms even though the code is typically much more readable. Another often intertwined approach is to include a platform independent api/function layer that is then used by the library, e.g. ACE's os.h. Any suggestions on best practice for this? Regards, Matt Hurd matthurd(a)acm.org www.hurd.com.au /*_____________________________________________________________________ created: 2004-6-7 16:20 filename: needs_lock.hpp author: Matt Hurd _______________________________________________________________________*/ #ifndef NEEDS_LOCK_HPP_200467 #define NEEDS_LOCK_HPP_200467 #include <boost/type_traits.hpp> #include "boost/type_traits/is_integral.hpp" #include "boost/type_traits/is_float.hpp" #include "boost/type_traits/detail/ice_or.hpp" #include "boost/config.hpp" // should be the last #include #include "boost/type_traits/detail/bool_trait_def.hpp" namespace boost { namespace detail { template< typename T > struct needs_lock_impl { BOOST_STATIC_CONSTANT(bool, value = (sizeof(T) > sizeof( void *)) ); }; } // namespace detail BOOST_TT_AUX_BOOL_TRAIT_DEF1(needs_lock,T,::boost::detail::needs_lock_impl<T>::value) } // namespace boost #include "boost/type_traits/detail/bool_trait_undef.hpp" #endif

5 17

RE: [boost] [shared_ptr] InterlockedIncrement rocks!
by Puverle, Tomas (IT) 22 Sep '04

22 Sep '04

Ben, > Atomic operations are highly platform-dependent, with > different platforms having different capabilities and many > requiring use of assembly language rather than providing an > API for atomic operations. Would that be a problem? I have implemented this for our internal managed pointer class, and our code needs to run on a few different platforms (At the moment I am discussing the posibility of contributing to boost with our legal dept - note that it wasn't until recently that I was allowed to even post to an external mail group. Before that I used to follow boost quite closely) > Second, shared_ptr has two counters and updating them safely > without using a lock requires some subtlety. On ia32, ia64 and amd64 you can always use DCAS. On Ultrasparc you could restrict the counter to 32-bits and use CAS (64-bit). You can add other platforms as the implementation becomes available, but use a lock by default. > It's something > I meant to implement but hadn't quite got round to. Perhaps > I could make it work on g++ (and hence most Linux > configurations) by using libstdc++'s atomic primitives? I No. This is not portable enough. The problem is that libstdc++'s atomic ops come with libstdc++ which precludes you from using other standard library implementations. The ones that come with Linux in sys/atomic.h tend to be quite inconsistent, not to mention that many are implemented in such a way that they don't return the new/old value. On Solaris, sys/atomic.h ops are not available in user space anyway, not to mention that they are declared as extern - so they can't be inlined. > don't think they are really public though, so this might be a > bad idea. I think this would be a great idea, and it would make shared_ptr a lot more acceptable at our company. Tom -------------------------------------------------------- NOTICE: If received in error, please destroy and notify sender. Sender does not waive confidentiality or privilege, and use is prohibited.

1 0

RE: [boost] Re: immutable string
by Ben Hutchings 22 Sep '04

22 Sep '04

Maxim Yegorushkin <e-maxim(a)yandex.ru> wrote: > Ben Hutchings <ben.hutchings(a)businesswebsoftware.com> wrote: > > [] > > > There's one possible problem I noticed, which is that an allocated > > buffer will be aligned properly for boost::detail::atomic_count but > > perhaps might not be aligned properly for char_t (or to whatever > > buffer_storage_alignment specifies). Let me apply the James Kanze > > test [*]: is it safe to create a const_string<double> on a SPARC? > > > > [*] See <http://gcc.gnu.org/bugzilla/show_bug.cgi?id=8670> if you > > don't know what I'm talking about. > > I fixed it using the union trick. > > But I think there might be another solution. The standard states in > 5.3.4.10 that arrays of char's allocated with new expression are > always properly aligned for types whose size is no greater than the > size of the array. Does allocator<char>::allocate() have the same > requirement? I could not find it in my copy of the standard. The default allocator's allocate() member function uses ::operator new(std::size_t) so any specialisation could be used to allocate memory aligned for any type. However, the requirements for allocators in general (table 32 in subclause 20.1.5) say only that "[m]emory is allocated for n objects of type T" by the allocate() function.

2 1

RE: [boost] [shared_ptr] InterlockedIncrement rocks!
by Ben Hutchings 22 Sep '04

22 Sep '04

Tomas Puverle <Tomas.Puverle(a)morganstanley.com> wrote: > > Also, why not including windows.h directly, like this: > > #define WIN32_LEAN_AND_MEAN > > #include <windows.h> > > I believe WIN32_EXTRA_LEAN may be even better. > > > <side-note> > > From my tests it seems your version as is now fails to link > > for gcc (unresolved external > > _InterlockedIncrement/_InterlockedDecrement). > > However, if #including <windows.h> and use > > InterlockedIncrement/Decrement - they work. > > </side-note> > > That is not suprising - Interlocked* are parts of the Win32 API. > > In fact, it has been a complete mistery to me as to why boost doesn't > use atomic operations for its ref counting. The mutex on the counter > was one of the big reasons why people here decided not to use > shared_ptr. Atomic operations are highly platform-dependent, with different platforms having different capabilities and many requiring use of assembly language rather than providing an API for atomic operations. Second, shared_ptr has two counters and updating them safely without using a lock requires some subtlety. It's something I meant to implement but hadn't quite got round to. Perhaps I could make it work on g++ (and hence most Linux configurations) by using libstdc++'s atomic primitives? I don't think they are really public though, so this might be a bad idea.

1 0

Multimethods Design
by Jesse Jones 22 Sep '04

22 Sep '04

There seems to be some interest in my multimethod library so I thought I'd post a message describing the design in more detail. In this message I'll focus on the API and ignore the implementation except when it affects the API. I may post another message later focusing on the implementation. The vault is still full, so if you want to look at the code go to <https://sourceforge.net/projects/multimethods/>. The draft proposal to add multimethods as a core feature is also interesting: <http://std.dkuug.dk/jtc1/sc22/wg21/docs/papers/2003/n1529.html#1.1.1>. Also <http://www.op59.net/accu-2003-multimethods.html> has some info on why multimethods are a Good Thing. The design is actually quite simple. Function objects (aka methods) may be added to a generic function object. Then the generic function can be called with a set of actual arguments. It will then dispatch to the most specialized method using the dynamic type of the actual arguments. One way to think of this is as a form of runtime overloading. The API for generic functions looks like this (using an arity 2 function as an example): template <typename Result, typename Formal1, typename Formal2> class generic_function2 : public detail::base_generic_function< Result > { When this is instantiated it creates a function object. Currently the arity is encoded in the class name, but presumably boost::preprocessor could be used to allow users to instantiate the classes in the way that boost:;function does. base_generic_function is a helper class that does all of the heavy lifting. Method formal arguments and actual argument types must be compatible with Formal1 and Formal2 (this is checked at compile time). The formal argument types may also be any_type which means that the formal argument in the methods for that position may be of any type. The API is as follows. First there are some standard typedefs: typedef Result result_type; typedef typename Formal1 first_argument_type; typedef typename Formal2 second_argument_type; A default ctor (and compiler provided copy ctor/assignment op): generic_function2(); Members to add methods: template <typename R, typename T1, typename T2> generic_function2& operator<<(R (*method)(T1, T2)); template <typename Functor> generic_function2& operator<<(Functor method); If the method has the same arguments as an existing method a std::runtime_error exception is thrown. If the Functor version is used the functor must have result_type, first_argument_type, and second_argument_type typedefs. A member used to invoke the generic function: template <typename T1, typename T2> RESULT operator()(const T1& arg1, const T2& arg2) const; This works by building a list of the applicable methods and then sorting this list from most to least specific using the dynamic type of the actual arguments. A method A is more specific than another method B if, for a given set of actual argument, for each argument, A has a better match for the actual argument than B in at least one slot and the other arguments are no worse than B's (this is similar to the rule used for overloaded functions). Argument matching follows the usual C++ rules so derived types are better matches than base types and promotions are better matches than conversions. If no applicable methods or a most specific method cannot be found a runtime_error exception is thrown. When a method is called it may call the next most specific method using one of the following methods: RESULT next_method(); template <typename T1, typename T2> RESULT next_method(const T1& arg1, const T2& arg2); The second overload calls the next method using a new set of actual arguments, but the types of those arguments don't affect which method is called. The first overload calls the next method with the last set of actual arguments that were used. If there are no more methods or the next method is ambiguous a runtime_error exception is thrown. The only missing piece is that clients of the library must register polymorphic types if they want them to be treated polymorphically. This is done with a macro that looks like this: #define DECLARE_SUBTYPE(type, base) This seems like a simple yet flexible API, but there's probably room for improvement. For example: 0) The names "generic function" and "method" seem potentially confusing in a C++ context (they're from Dylan and (I think) CLOS). 1) There's no support for removing methods. This seems like a good idea, but it's not possible to compare arbitrary function objects so we'd probably have to add an add_method member which returned a cookie that a remove_method member would take to determine which method to remove (similar to the way boost::signal works). 2) Policy decisions like type compatibility and ordering are hardwired into the code. It might make sense to refactor these into policy classes although I'm not sure what they'd be used for. Maybe special casing smart pointers. 3) The library allows lossy conversions like floats to chars. This was done so that it world work as much like normal C++ code as possible, but it's a pretty evil thing. Maybe an option could be added that would reject such conversions. 4) boost::bind is a bit annoying because it doesn't provide the typedefs that the library requires. This is a known issue with bind and, as far as I know, isn't fixable. The workaround is to assign the result of boost:;bind to a boost:;function and add that to the generic function. 5) Method formal arguments can't be non-const reference types. The problem here is that typeid(const T&) == typeid(T&) == typeid(T). It might be possible to work around this, but the type code is already full of hacks working around typeid and type_info limitations... 6) It might be a good idea to add a member that will return a function pointer or boost:;function to the most specific method for a given set of types. This would allow users to skip the overhead of dispatching when using multimethods inside loops in some cases. -- Jesse

1 0

RE: [boost] Re: Formal review of "Output Formatters" library beginstoday
by Reece Dunn 22 Sep '04

22 Sep '04

Rozental, Gennadiy wrote: >Here some more details: > > > 1. What is your evaluation of the design? > >Library trying to solve in fact trivial problem. Would it be presented as >trivial function > >Template<typename C> >Foo( C const& c, std::string open, std::string close, std::string sep ); > >I may understand it. This would mean that you would duplicate code for rendreing container types. Also, you are missing special formatting, e.g. wrapping open/close values around elements, e.g.: std::cout << formatob( vec, containerfmt( wrappedfmt().format( "<e>", "</e>" )).format( "<l>", "</l>" )); // output: <l><e>a</e><e>b</e></l> the indentation functionality that Volodya has expressed an interest in, and position-based output, e.g.: [0] John, [1] James, [2] Marina Also, the library should be usable on a standard stream. This is one of the key design aims. Can you do: std::cout << Foo( vec, "< ", " >", " :: " ); ? This requires returning an object that has << implemented. The mechanism for this is outlined in the standard by using *manipulators*. Thus, you would need that class to defer processing to the Foo function if you want to use your approach. >As to presented design - I see it as unnecessary complicated. And in many >cases simply incorrect. >Library does not follow standard practice as to wide char support. I am addressing these issues. See my other discussions for details. >Interfaces seem confusing and error prone. >Decoration model seems too simple to warrant so much trouble as reusable >library. Maybe more powerful/flexible solution should be considered. I have been reshaping the decoration model and have commented on it elsewhere. > > 2. What is your evaluation of the implementation? > >In most part I either don't understand or don't like it. IMO simple task >library trying to solve shouldn't require that much code. Simple >overloading >base solution should've done the trick. This does not allow for flexibility to add indenting, wrapped data, etc. > > 4. What is your evaluation of the potential usefulness of the library? > >Now to the most important: why would I ever want to use library like this? > >We have already: >std::ostream >boost::format >boost::serailization > >In what scenario I would use this library? But you can't do: std::vector< int > vec; std::cout << vec; // oops! no operator << defined for vectors. >I most definitely wouldn't want to repeat all this formatting every time I >need to output my vector. If you want [ a, b, c, d ] style output, you don't. The extra machinery is used to customize this. Also, you can save a format object, e.g.: // old style containerfmt_t< char, pairfmt_t< char > > myformat = containerfmt( pairfmt().format( " = " )).format( "<< ", " >>" ); std::cout << formatob( vec, myformat ); // << ( a = 5 ), ( b = 7 ), ... >> std::cout << formatob( lst, myformat ); // << ( i = 2.132 ), ( j = 3.14159 ), ... >> std::cin >> formatob( vec, myformat ); >If I write output operation rarely - I would use explicit loop -it's more >flexible anyway. Depends what you are outputting. If the type contains nested containers, it can be more complex. >I wouldn't use this library for filtering - we will have iostreams library >for that (If I am not mistaken). This library does not provide filtering capabilities. What I was saying is that if you want to pretty print the data, it might be worth using Jonathan's library. >I wouldn't use this library for XML printing - why would I want to mention >tag name twice? You can use a wrapper format object: std::cout << formatob( vec, containerfmt( wrappedfmt().format( "<e>", "</e>" )).format( "<l>", "</l>" )); // output: <l><e>a</e><e>b</e></l> Regards, Reece _________________________________________________________________ Want to block unwanted pop-ups? Download the free MSN Toolbar now! http://toolbar.msn.co.uk/

1 0

RE: [boost] OutFmt library implementation questions
by Reece Dunn 22 Sep '04

22 Sep '04

Rozental, Gennadiy wrote: > > >4. How does this all machinery will work with wchar_t? > > > > std::wcout << formatob( vec, containerfmtex< wchar_t * > > >()).format( L" - > > " ); > > > > Use the ex< DelimiterType > variant on the innermost construct. > >This is inacceptable and does not follow standard practice. The new version templates that are templatized on character type are also templatized on the format object that they use (to render their elements). This means that in order to provide c/w-style interface, you need: template< typename CharT, class ElementRenderer = basic_default< CharT > > class basic_container{ ... }; basic_container< char > ccontainer(){ return( basic_container< char >()); } basic_container< char, ElementRenderer > ccontainer( const ElementRenderer & er ){ return( basic_container< char >( er )); } basic_container< wchar_t > wcontainer(){ return( basic_container< wchar_t >()); } basic_container< wchar_t, ElementRenderer > wcontainer( const ElementRenderer & er ){ return( basic_container< wchar_t >( er )); } // what about custom character type (e.g. unicode?) basic_container< char > my_format = ccontainer().decorate( " : " ); std::cout << io::object( vec, my_format ); std::cout << io::object( vec, ccontainer().decorate( "< ", " >" )); > > This is a VC 6.x workaround. > >But you don't list this compiler amoung supported. At the moment there are a few problems with it. I am looking to resolve these if possible. > > >6. Why the same entity you sometime call format_type and sometime > > >DelimeterType? > > > > format_type is the typedefed version (should be delimiter_type) and > > DelimiterType is the template parameter. > >So is it delimeter type or forat type - it can't be both. delimiter type. In the new version it is char_type since the decorators (formatters in the review implementation) now take a CharT parameter instead of the storage type. > > >7. What deducer is supposed to deduce? I couldn't get > > through all the > > >details, but couldn't the same be achieved using partial > > specialization > > >and/or partial ordering. Cause it look way too complex and > > unnecessary. > > > > The deduction mechanism is used to evaluate what type T is > > (e.g. std::pair< > > char, int >), what format object is needed to render it > > (including nested > > format objects) and to construct the appropriate format object. > > > > It could be possible to use partial specialization, but the > > type traits are > > used elsewhere to redirect formatting behaviour (e.g. when > > reading in a > > container do diffreent things if it is a sequential, > > associative or set > > container). > >IMO all that you needed is function overloading. I don't see a place for >all >these deduce/trats staff. Consider the pair format object (pair renderer): how do you render each pair type. Using function overloading, you would have: class basic_pair { public: OutStream & write( OutStream & os, const std::pair< T1, T2 > & p ){ ... } OutStream & write( OutStream & os, const boost::compact_pair< T1, T2 > & p ){ ... } OutStream & write( OutStream & os, const std::complex< T > & p ){ ... } OutStream & write( OutStream & os, const boost::math::interval< T > & p ){ ... } }; But this would add dependencies on those libraries. Also, it is not extensible: how do you add support for boost::rational< T >, for instance? Using the deduction mechanism, I implement the output using boost::io::detail::getval< n >( pair_type ). Input is more complex because you need to distinguish between types that you can get the elements separately (e.g. std::pair) and types that you need to set both values together (e.g. boost::math::rational): the former uses refval< n >( type ) and the latter uses assignval( type, first, second ). How would function overloading solve this? (Note, function overloading is used for getval, refval and assignval). For containers, you need to use a different function/function arguments to insert an element depending on whether the container is sequential, associative or a set. My library provides a description of the types that it can render (e.g. separable pairs, 4-ary types, associative containers) without saying what these types are. The stl and boost directories in my library then tell my library that a std::vector is a sequential container. My library then knows how to handle a std::vector. Doing this, I can separate the dependencies on external libraries and allow it to be extended to support other sequential containers, pairs, etc. > > >11. Need for all this hierarchies: > > >Formatter->formater_t->openclose_formatter_t > > > > This is to ensure that the return type of 'format' in Formatter > > [FormatObject] is Formatter and not openclose_formatter_t. Thus: > > formatob( vec, containerfmt().format( " / " )); > > will work as intended without overriding 'format' in > > containerfmt_t (created > > by containerfmt()) to return the correct type. > > > > >openclose_formatter->openclose_formatter_t > > > > > >are completely unclear. Explanations presented in docs does not help > > >since format methods seems to be returning *this anyway. > > > > They return: > > *static_cast< ReturnType * >( this ) > > to allow 'format' to return the correct type when inherited > > in the format > > objects. > >Formatob_t::format return *this. You did not answer my question. I still >believe it's all unnecessary. formatob_t::format returns *this because the FormatObject (Renderer) that it inherits returns a value of type FormatObject. Thus, if you didn't have these overloads, then: std::cout << formatob( vec ).format( " + " ); would not work, since it will try to output a containerfmt() type. > > >Why couldn't we just use something trivial like > > > > > >template<char Ch> > > >struct formatter > > >{ > > > std::basic_string<Ch> m_open; > > > std::basic_string<Ch> m_close; > > > std::basic_string<Ch> m_sep; > > >}; > > > > > >Is beyond me. If any member is empty it means it missing. > > > > This is the direction I am going in the development version. The only > > difference is that I am using io::decoration< CharT, int id > > > instead of > > std::basic_string< CharT >. The decoration class uses > > std::basic_string to > > store the values, while providing support for looking up the > > default value > > on a stream and an external function for checking for a match > > on an input > > stream. > >You presented library call *Output* formatter. Why do we talk about input >at >all. The *output* designation is a "legacy" from before the library supported input facilities. People expressed an interest in inputting as well, so I added it; output stuck. Regards, Reece _________________________________________________________________ Use MSN Messenger to send music and pics to your friends http://www.msn.co.uk/messenger

1 0

Spirit on Intel 8.0/Linux
by Aleksey Gurtovoy 22 Sep '04

22 Sep '04

Folks, Is this a known problem that somebody can comment on (from http://tinyurl.com/5hhne) icc: error: Fatal error in /opt/intel_cc_80/bin/mcpcom, terminated by segmentation violation compilation aborted for ../libs/spirit/test/./ast_calc_tests.cpp (code 1) . /opt/intel_cc_80/bin/iccvars.sh icc -c -w1 -g -O0 -I"../bin/boost/libs/spirit/test" -I"../libs/spirit" -I".." -I"/usr/include" -I"/boost/head-regression/boost" -o "../bin/boost/libs/spirit/test/ast_calc_tests.test/intel-8.0-linux/debug/ast_calc_tests.o" "../libs/spirit/test/./ast_calc_tests.cpp" ? This obviously looks like a compiler issue, but before I mark it up as such I'd like to hear if anybody has to say anything on it. There are more Spirit tests failing in a similar manner on this configuration -- see http://tinyurl.com/5dtbz. -- Aleksey Gurtovoy MetaCommunications Engineering

1 0

RE: [boost] outputformatters - using just one decorator string
by Reece Dunn 22 Sep '04

22 Sep '04

John Torjo wrote: >Hi Reece, > >This just popped into my head. A lot of confusion seems to come from how >you pass decorators. What I'm trying to say is that it can be very >difficult to get the decorator strings right (especially for more complex >things like XML) from the first time. > >I would suggest every formatter takes only one string which parses and >finds all decorators itself. >Each formatter will have one or more escape sequences which it uses to find >the decorators. > >Something like: > >// in this case, '%' is used for identifying an element >// '*' is to mean "leave default" (don't change) > >std::vector<int> v; >// equivalent to your "[ ", ", ", " ]" >std::cout << formatob(v, "[ %, % ]"); This looks good :) On the implementation side it would require splitting the string into the 2/3 components and constructing the decorators accordingly. Maybe defining an "extract_decorators" function internally, so wrapper_decorators and sequence_decorators can process this internally. How about: std::cout << io::object( v )( "[ %, % ]" ); I am sure this would be feasible. >// eq. to "<< ", " | ", " >>" >std::cout << formatob(v, "<< % | % >>"); > >// set only open/close decorators >std::cout << formatob(v, "<< %*% >>"); Here, you would need special handling of string == "*". This should also be implementable. >// sets only the separator >std::cout << formatob(v, "*% | %*"); > >// for pair - you have "%1" and "%2" >std::vector< std::pair<int,long> > vp; >// write XML >std::cout << formatob(vp, > containerfmt("<elem> % </elem><elem> % </elem>", > pairfrt("<int>%1</int><long>%2</long>"))); Here, you don't need %1 and %2. You just do: "<int>%</int><long>%</long>". >How's that? In terms of implementability, I don't see a problem. The two issues I have are: [1] What are the performance penalties for this? [2] What if you want '%' or '*' in your format string, e.g. "[ a * b * c ]"? You would thus need to implement some sort of escaping capability. NOTE: This could sit along side using the 'decorate' function and Volodya's idea for expressions, e.g.: io::object( v, "<<" + fmt::container( fmt::pair()( "<a>%</a>%" )) + " >>" ).decorate( " : " ); // output: << <a>2.75</a>1 : <a>3.14159</a>7 >> Regards, Reece _________________________________________________________________ Use MSN Messenger to send music and pics to your friends http://www.msn.co.uk/messenger

2 1