- Boost - lists.boost.org

Constructor-proxy: Template parameter deduction, a question for template masters
by ION_G_M 19 Feb '05

19 Feb '05

Hello to all, while developing Boost Shmem, I wanted to create a constructor proxy to emulate c++ dynamic object allocation syntax when creating shared memory objects using a proxy object in the following way: T * ptr = shmem_segment.create<T>("ObjectName")[size] /*proxy returned here, calls operator() of the object*/ (arg1, arg2, ...argN); Shared memory is allocated and the object is constructed. I want the arguments to be deduced automatically in the proxy object function call to preserve this syntax. I also want to avoid _all_ parameter copying. I mean, if I can deduce that an argument can be passed by value or by referenced using this syntax, great. But, between copying _all_ arguments and passing a reference in _all_, I prefer the second. Trying to find a correct proxy for this, I had many problems when passing temporary arguments and hardcoded constants to the proxy. Right now I am using some ugly casts to avoid the copies, but this has some side-effects. I will explain the problem in the following code: #include <new> //Object to be constructed struct Ctor2Arg { Ctor2Arg(const int &const_ref, int &non_const_ref){} }; //"Clean" proxy template <class T> struct constructor_proxy_t { template<class T0, class T1> void operator()(T0 & t0, T1 & t1) const { char m_mem[sizeof(T)]; //Call constructor and destructor T *ptr = new(m_mem)T(t0, t1); ptr->~T(); } }; //Ugly cast proxy template <class T> struct ugly_constructor_proxy_t { template<class T0, class T1> void operator()(const T0 & t0, const T1 & t1) const { char m_mem[sizeof(T)]; //Convert all refs to non-const refs T0 & t0ref = const_cast<T0 &>(t0); T1 & t1ref = const_cast<T1 &>(t1); //Call constructor and destructor, non-const //refs can be converted to const refs T *ptr = new(m_mem)T(t0ref, t1ref); ptr->~T(); } }; //Main function int main() { //Construct normal and ugly proxies constructor_proxy_t<Ctor2Arg> proxy; ugly_constructor_proxy_t<Ctor2Arg> ugly_proxy; //One const int and one non-const int const int dummy_const = 0; int dummy_non_const = 0; //This is ok, calls: //constructor_proxy_t<Ctor2Arg>::operator() //<int const ,int>(const int & t0=0, int & t1=0) proxy(dummy_const, dummy_non_const); //Both of these calls do not compile in VC7.1 or gcc 3.4.3, //tries to call constructor_proxy_t<Ctor2Arg>::operator() //<int, int>(int & t0=5, int & t1=0) // // gcc 3.4.3 error // //error: no match for call to // `(constructor_proxy_t<Ctor2Arg>) (int, int&)' //candidates are: // void constructor_proxy_t<T>::operator()(T0&, T1&) // [with T0 = int, T1= int, T = Ctor2Arg] // //Obviously, compilers are right, but //it wouldn't be more logical to call //<int const, int> specialization since a //temporary is a constant object? // proxy(dummy_non_const+5, dummy_non_const); // proxy(dummy_const+5, dummy_non_const); //This is ok, calls: //ugly_constructor_proxy_t<Ctor2Arg>::operator() //<int,int>(const int & t0=0, const int & t1=0) ugly_proxy(dummy_const, dummy_non_const); //These are ok, call: //ugly_constructor_proxy_t<Ctor2Arg>::operator() //<int,int>(const int & t0=5, const int & t1=0) ugly_proxy(dummy_non_const+5, dummy_non_const); ugly_proxy(dummy_const+5, dummy_non_const); //This compiles but it should not, since //I am passing a const object as second parameter //so Ctor2Arg obj(dummy_const, dummy_const) //does not compile. This calls: //ugly_constructor_proxy_t<Ctor2Arg>::operator() //<int,int>(const int & t0=0, const int & t1=0) ugly_proxy(dummy_const, dummy_const); return 0; } My question is, has anyone a better way to write the proxy object without ugly casts, no side-effects (I mean, the proxy call should compile only if the constructor compiles), preserving a clean calling syntax, and _no_ object copies at all? And the second question is: when a temporary is passed to a template, why non-const argument is deduced in the function call? Thank you, regards, /Ion

3 2

singleton report (lifetimes)
by Jason Hise 19 Feb '05

19 Feb '05

Just thought I would release a progress report on the singleton. I have added a new lifetime policy called static_lifetime, which derives from the longevity lifetime and just hides the longevity part from client code, setting it to zero. The result: singletons using the static_lifetime policy are guaranteed to be destroyed in reverse order of creation, if they are created. I am thinking that this should be the default lifetime for the singleton, however I am open to suggestions. Here is a brief summary of each lifetime in the library. If there are any ideas for other lifetimes I would love to hear them. dependency_lifetime - Uses pointers to the singleton as reference counted dependencies. The singleton, if created, is destroyed upon the destruction of the last pointer. A static or global pointer must exist to ensure that the singleton doesn't get destroyed and recreated as the program executes. Singletons can own dependencies to other singletons to ensure proper order of creation and destruction. If a circular reference exists, the singletons owning pointers to each other will never be destroyed. This was my former choice for the default policy. longevity_lifetime - Upon first creation of a singleton using this policy, the singleton instance is registered with a hidden singleton, the longevity_registry. The longevity_registry has a static pointer to itself in its constructor, ensuring it lives at least until main exits. Upon destruction, it destroys the registered singletons in order. Singletons with lower longevities are destroyed first, and singletons with the same longevity are destroyed in reverse order of creation. static_lifetime - Deriving from the longevity_lifetime, this just hides the longevity template parameter and uses zero. As a result, singletons using this policy are destroyed in reverse order of creation, and are destroyed before any longevity based singletons using a longevity greater than one. As the longevity is an unsigned int, this means most of them. This is my current choice for a default. leaky_lifetime - This never destroys the singleton instance. Because of the extensive framework that it provides in terms of member functions and classes, all of the other lifetimes derive from it. This way the other lifetimes can add the specialized destruction behavior without duplicating all the little details, like creation behavior and member functions of the pointer class. The leaky lifetime is used internally by a hidden singleton used in the multiton implementation. This singleton stores multiton instance holders in a map (ideally this should be an assoc_vector if one is ever added to boost). It can never be destroyed because the singletons it stores instance holders to could potentially be destroyed later, or even be recreated by other singletons in the process of being destroyed. Fortunately the map only needs to be destroyed if it needs to free the memory, and this memory which will be automatically cleaned up by the OS anyway. I would like to point out that the method of creation is the same regardless of the lifetime used. An enumerated trait of each lifetime controls when automatic creation can occur. This can be set to create the singleton upon creation of a pointer, upon dereferencing a pointer, both, or neither. If this is set to not automatically create the singleton at all, client code can explicitly call a create member function from any singleton pointer. The leaky_lifetime implements all of these details so that the other lifetimes which derive from it do not have to worry about it. Phoenix re-creation ability works automatically for all of these. To achieve a non-phoenix lifetime, a one_time_creator can be specified which uses a real creator the first time, and a creator designed to fail the second time. Numerous failure creators are provided: null_creator - create simply returns null throw_creator - create throws an exception (which can optionally be specified) assert_creator - causes a boost assertion to fail, and then uses some other failure creator (defaults to throw_creator) if execution gets past the assertion The real creators provided include: create_using_new - client code can customize with a boolean trait whether to make it use no throw new create_using_malloc - client code can make this throw or no throw as well create_statically - uses a static aligned_storage buffer to hold the instance, cannot be used with a multiton create_using_std_allocator - allocates memory with a standard allocator (which can be customized) The one_time_creator can be chained to use a different creator for the first, second, and third creations, ad infinitum. It makes sure to destroy each instance with the internal creator used to create it. I am diligently working on the html documentation, which is going to be quite extensive. It seems like the more documentation I write, the more ideas I get for improving the library. When I get the documentation into a coherent state I will upload it. However, I have made so many changes now that I think an outside perspective would be a good thing before I start setting class and function names in stone. Thus, an up to date stable version has been uploaded to the sandbox. What would be really helpful is if someone could try to write a test program in a multi threaded context. My background in threading is very shaky, and I want to be sure that I am using the locking policy the right way. Is anyone willing to give this a try? -Jason

1 0

[Reminder] Mailing list and lists.boost.org outage Saturday
by Douglas Gregor 19 Feb '05

19 Feb '05

This is a reminder that the Boost mailing lists and lists.boost.org will go down at 12:00pm Saturday Eastern Standard Time. We expect that all will be running again by 8pm at the absolute latest. Doug

1 0

assert
by Jason Hise 19 Feb '05

19 Feb '05

My version of assert.hpp has the line: #include <assert.h> Shouldn't this be: #include <cassert> ? -Jason

6 6

[bug]: optional<T> fails on self-assignment
by Joe Gottman 18 Feb '05

18 Feb '05

I just discovered that optional<T> is not safe under self-assignment. Consider the following code fragment: boost::optional<long> x(1); //At this point x is initialized and *x == 1 x = x; //x should be unchanged, but instead x is now uninitialized Optional's assignment operator looks like optional& operator= ( optional const& rhs ) { this->assign( rhs ) ; return *this ; } and assign is defined by void assign ( argument_type val ) { destroy(); construct(val); } Thus, if (this == &rhs) in operator=(), any value held by rhs will be destroyed. I can see two possible fixes for this. the first is to put an "if (this != &rhs) " around the call to assign in operator=(). This is simplest, but it slows the code down in order to protect against a very rare situation. The other is to rewrite assign as follows: void assign(argument_type val) { if (m_initialized) { get() = val; // call assignment operator } else { construct(val); //call copy constructor } return *this; } This makes optional self-assignment safe. It has the added advantage that is T's assignment operator has the strong exception safety guarantee, so does optional<T>'s assignment operator. It has the disadvantage of making optional depend on both assignment operator and copy constructor to copy, but very few classes have a copy constructor but don't have an assignment operator. Joe Gottman

7 31

New profiling library update
by christopher diggins 18 Feb '05

18 Feb '05

I have updated the profiling library and written a few tests. I still need to rewrite the documentation. the latest version is at http://www.ootl.org/profiler/toc.htm . Christopher Diggins Object Oriented Template Library (OOTL) http://www.ootl.org

1 0

Compile-time conditional function invocation
by Jeff Mirwaisi 18 Feb '05

18 Feb '05

Below I have outlined my need and implementation for a compile-time conditional function invocation utility. I thought it might be generally useful to others using simmilar metaprogramming techniques. If suitable solutions can be found for some of the outstanding issues perhaps it will prove suitable for inclusion in boost. The following quick documentation and a snapshot of the implementation can be found at http://dump.qxxy.com/execute_if/ and (intermitantly) http://66.1.253.185:8080/src/execute_if/index.html Any feedback would be appreciated. Thanks Jeff Mirwaisi Introduction The following is a simple compile-time conditional function invocation and binding utility type. Generic code often needs to handle a subset, of all the types, in a slightly different way than the rest. template <typename T> void f(T t) { t.a(); t.b(); ... if ( has_member_function_c<T>::value ) t.c(); } This solution obviously will not work if type T has no member function c. The compiler expects all the types used with the template function f to have a member function c even though its clear the conditional expression is either true or false at compile time. One possible workaround for this situation is to use SFINAE and enable_if to overload template function f for types which have a member function c and those that do not. template <typename T> void f(T t, typename enable_if<has_member_function_c<>,T>::type* SFINAE = 0) { t.a(); t.b(); ... t.c(); } template <typename T> void f(T t, typename enable_if<not<has_member_function_c<T> >,T>::type* SFINAE = 0) { t.a(); t.b(); ... } This solution works! Unfortunately it forces us to repeat the majority of the implementation of the two template function overloads f, and it does not scale. If we add a second conditional invocation, the number of overloads jump to four, the same is true of conditional callers, e.g. a second group of two template functions overloads g if theres a single conditional invocation, four if there are two conditional calls etc. To combat this overload explosion we could isolate the conditional call to member function c with a helper type. template <bool> struct call_member_function_c { static inline void invoke(T& t) {t.c();} }; template <> struct call_member_function_c<false> { static inline void invoke(T& t) {} }; template <typename T> void f(T t) { t.a(); t.b(); call_member_function_c< has_member_function_c<T> >::invoke(t); }; This solution improves on the previous one, it scales well as we add conditional member function invocations (call_member_function_d,call_member_function_e,...) and as we increase the number of callers(f,g,h,...). But it has introduced an explicit no-op function for each call_member_function_X helper type, and a partial specialization for each helper type to select whether to generate a call or a no-op. This "helper" type can be formalized and we can provide a generic no-op implementing replacement for all "helpers" which reduces our problem to the simplest solution allowed by C++. Target Holders Unfortunately the limitations of the current language wont allow us to create a generic "helper" type which can be loaded with different template function targets. Instead the user is expected to provide a minimal holder type which delegates to the target template function. Note that the following example is also a late template argument binding function object. This allows the user to either invoke the conditionally executed target immediately or produce a function object which can be stored and called at a later point. template <typename T> void template_fn_target(T) {} struct template_fn_target_holder { typedef void result_type; template <typename Arg> inline result_type operator()(Arg arg) {return template_fn_target(arg);} template <typename Arg> static inline result_type invoke(Arg arg) {return template_fn_target(arg);} }; The target holder type provides the result type of the operation to the user, the function object usage expects a function call overload that can handle the arguments specified, if the call is to be made instantly than we can simplify the use by calling a static member function of the target holder type named invoke. execute_if template <typename Condition > struct execute_if { template< typename TargetHolder > static inline typename TargetHolder::result_type invoke( arg1,...,argN ); template< typename TargetHolder > static inline TargetHolder bind(); }; When the condition value is true the subsequent invoke or bind operation are performed on the target type, if the value is false a no-op invoke or bind are generated. execute_if_c template < bool > struct execute_if_c { template< typename TargetHolder > static inline typename TargetHolder::result_type invoke( arg1,...,argN ); template< typename TargetHolder > static inline TargetHolder bind(); }; The execute_if_c variation on the above execute_if type provides a less generic version of enable_if which uses a boolean value as the condition upon which to select either the specified target or a no-op. execute_if_else template <typename Condition > struct execute_if_else { template< typename TargetHolderA,typename TargetHolderB > static inline typename TargetHolder?::result_type invoke( arg1,...,argN ); template< typename TargetHolderA,typename TargetHolderB > static inline TargetHolder? bind(); }; When the condition value is true the subsequent invoke or bind operation are performed on target holder A, if the value is false target holder B will be used execute_if_else_c template < bool > struct execute_if_else_c { template< typename TargetHolderA,typename TargetHolderB > static inline typename TargetHolder? ::result_type invoke( arg1,...,argN ); template< typename TargetHolderA,typename TargetHolderB > static inline TargetHolder? bind(); }; The execute_if_else_c variation on the above execute_if_else type provides a less generic version of enable_if which uses a boolean value as the condition upon which to select either target holder A or target holder B. To-Do * Currently manually written nullary,unary,and binary execute_if and proxy functions, needs to be rewritten to use the boost preprocessor lib * Possible move of the return type template parameter into the execute_ifxxx template parameter list (out of the bind and invoke template functions) * Investigate using the preprocessor to automatically generate the template function holder types, and other techniques to reduce the users responsibility * Document the lack of a means of representing a template function in the c++ type system, no function equivalent of template template classes * The current code also implements a set of invoke_fn and bind_fn overloads which operate on function references and function object references, we must determine wether these should be provided at all, and if so they should probably be documented

2 1

gui library
by Matthew Johnson 18 Feb '05

18 Feb '05

I spent some time reading the archives regarding the gui project that was started towards the end of last year. Is that project still underway? What is the best way of keeping up with the status of it? Thanks, Matthew

3 2

New article on BOOST_FOREACH
by Eric Niebler 18 Feb '05

18 Feb '05

For anybody interested in peeking under the hood of BOOST_FOREACH, there is a new article about it at The C++ Source (http://www.artima.com/cppsource/index.jsp) It focuses mainly on the conditional operator and how to use it to write macrco's that do not reevaluate their arguments. -- Eric Niebler Boost Consulting www.boost-consulting.com

1 0

program options documentation
by Rutger ter Borg 18 Feb '05

18 Feb '05

Hi, I noticed that the documentation of the Boost.Program_options library has a small inconsistency regarding the add_options method. In the tutorial section and in the Options Descriptions Component subsection of the Library Overview, add_options is documented in a form that is probably meant as the correct one: desc.add_options() ("compression", po::value<int>(), "set compression level"); while on the other hand, under Library Overview/Options Description Component/Syntactic information it is written as desc.add_options() ("compression", "compression level", value<string>()); , i.e., the 2nd and 3rd arguments are exemplified in a different (likely wrong) order. Best regards, Rutger

2 1