The only thing you can determine, is, if the type (class) of the last declaration is the one you expected. That means if you know that the first declaration of your function comes in from baseT, you can determine if a deriving class overwrites that decl, and that is exactly what a deriving class needs to do to overload a virtual function. I myself use this pattern to determine, if a type is an XPCOM Interface or not. Perhaps you'd like to check for something similiar. If interested in details, let me know. Mathias Gaunard <mathias.gaunard@ens-lyon.org> Sent by: boost-bounces@lists.boost.org 21.03.2010 19:31 Please respond to boost@lists.boost.org To boost@lists.boost.org cc Subject Re: [boost] [mpl] has_function Lorenzo Caminiti wrote:

Is there a way I can check if a class has a given member function using metaprogramming?

Yes.

What I need is to check at compile-time if a function is overriding a virtual function or not.

This is not the same, and there is no way to do that as far as I know. _______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

The only thing you can determine, is, if the type (class) of the last declaration is the one you expected. That means if you know that the first declaration of your function comes in from baseT, you can determine if a deriving class overwrites

below you find the appropiate code, that checks if a set of functions is declared for a type T, where the declaration should be inherited from nsISupports (or any other interface). So you'd can determine, if the function is overloaded or not. This code is inspired by the following article http://groups.google.com/group/comp.lang.c++.moderated/tree/browse_frm/threa... /** Compile Time Predicate, which checks, if a type is a valid template * parameter for a COMPtr, where only interfaces are valid input parameters. * * The idea is simple : * An interface always inherits from nsISupports and does not overload the * pure virtual declarations of AddREf, Release and QueryInterface. Therefore * the addresses of that function pointers need to be bound to nsISupports. * * Of course, one issue remains unsolved : * What is with abstract classes, which inherit from nsISupports, but already * have their own implementation ? * Currently there is no way, to check, if a class is pure virtual, thus we * neglect that issue. * */ template <typename T> struct IsCOMInterface : public Supports_::BasicTestPredicate<IsCOMInterface<T> > { private: // we are only able to determine this predicate for complete types, a // forward declare (incomplete type) may be a COM interface, but would be // excluded as an error to its incompleteness. Therefore the only thing, // that is always correct is to assert in such conditions. BOOST_MPL_ASSERT_RELATION((sizeof(T)),>,0); static T* Make(); static false_t Test1(...); template <typename _1> static true_t Test1(_1*, equal<nsrefcnt (__stdcall nsISupports::*)(),&_1::AddRef>* = 0); static false_t Test2(...); template <typename _1> static true_t Test2(_1*, equal<nsrefcnt (__stdcall nsISupports::*)(),&_1::Release>* = 0); static false_t Test3(...); template <typename _1> static true_t Test3(_1*, equal<nsresult (__stdcall nsISupports::*)(nsID const&,void**),&_1::QueryInterface>* = 0); public: static const value_type value = sizeof(Test1(Make())) == sizeof (true_t) && sizeof(Test2(Make())) == sizeof(true_t) && sizeof(Test3(Make ())) == sizeof(true_t); }; namespace IsCOMInterface_ { BOOST_MPL_ASSERT((IsCOMInterface<nsISupports>)); BOOST_MPL_ASSERT((IsCOMInterface<nsIWeakReference>)); struct Test1 {}; BOOST_MPL_ASSERT_NOT((IsCOMInterface<Test1>)); struct Test2 { NS_IMETHOD_(nsrefcnt) AddRef(); NS_IMETHOD_(nsrefcnt) Release(); NS_IMETHOD QueryInterface(nsID const&, void**); }; BOOST_MPL_ASSERT_NOT((IsCOMInterface<Test2>)); struct Test3 : public nsISupports { NS_IMETHODIMP_(nsrefcnt) AddRef(); }; BOOST_MPL_ASSERT_NOT((IsCOMInterface<Test3>)); } namespace Supports_ { /** * */ template <typename derivedT> struct BasicTestPredicate { protected: typedef char true_t; struct false_t{true_t dummy[2];}; template <typename V, V> struct equal; public: typedef bool value_type; typedef derivedT type; }; } // namespace Supports_ Lorenzo Caminiti <lorcaminiti@gmail.com> Sent by: boost-bounces@lists.boost.org 22.03.2010 13:33 Please respond to boost@lists.boost.org To boost@lists.boost.org cc Subject Re: [boost] [mpl] has_function On Sun, Mar 21, 2010 at 2:43 PM, Ingo Loehken <Ingo.Loehken@boc-eu.com> wrote: that

decl, and that is exactly what a deriving class needs to do to overload a virtual function.

I myself use this pattern to determine, if a type is an XPCOM Interface or not. Perhaps you'd like to check for something similiar.

If interested in details, let me know.

Yes, I am interested. Can you please point me to some code that programs this pattern? Thanks a lot. Lorenzo _______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

just to throw another couple of ideas what about such essential things like data compression and data encription? i think these two will perfectly fit into boost as policy driven (template?) libraries at least there would be a kind of standard interface to these facilities -- Pavel

On Mar 22, 2010, at 12:51 PM, Scott McMurray wrote:

On 22 March 2010 15:27, DE <satan66613@yandex.ru> wrote:

...

what about such essential things like data compression and data encription?

i think these two will perfectly fit into boost as policy driven (template?) libraries

at least there would be a kind of standard interface to these facilities

How about, as a prerequisite to a larger encryption library, more cryptographic hashes?

For instance, UUID has an SHA-1, but would rather not maintain its own. Also, there are MD5-based UUIDs that the library can't provide yet, since it lacks an MD5 hasher.

Except for the scope (I think this might be a big small), I think this would make a good SOC project. MD5, SHA-1, SHA-2, with a place for the current NIST contest winner when it is decided. If there is time, the concepts can be used to implement some symmetrical encryption algorithms (RC4, RC5, AES, etc). -- Marshall

I'll summarize extra ideas. So, it will be a container library with this list of features supported with template parametres: 1 - Mutithreading 2 - Exception safety 3 - Different internal realizations Also it will contain support of transformation algorithms, such as archiving, cryptography and others. After the first sight I'd like to offer realiation of that kind. We provide a special class which will implement stream operations being able to work with stl streams. There will be a template argument specifing an exact transform algorithm. It will extend the queue container to take use of it adjustability. Also it will give us an appotunity co create transformation chains. -- Biin Ilya MIPT

Hi Biin,

This sounds like a very ambitious project, and I'm not convinced that there is a huge demand for such a library in the Boost community. Also, building new Boost libraries may be a commitment of several years. Maybe you can revise your ideas to address the needs of Boost existing or upcoming Boost libraries?

Andrew Sutton andrew.n.sutton@gmail.com _______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

There is a boost-crypto library in the vault, I think it will be possible and logical to implement this idea

Also it will contain support of transformation algorithms, such as archiving, cryptography and others. After the first sight I'd like to offer realiation of that kind. We provide a special class which will implement stream operations being able to work with stl streams. There will be a template argument specifing an exact transform algorithm. It will extend the queue container to take use of it adjustability. Also it will give us an appotunity co create transformation chains.

on top of it. Though there is a lot of algorithms there it's not ready for usage. -- Ilya Biin MIPT

Lorenzo Caminiti wrote:

It turned out that I only needed to check if a struct is declared within one of the base classes or not. This seems simpler that my original goal for has_function and I was able to adapt Ingo's code to get the job done -- see below. However, I am getting the following compiler error:

$ g++ -Wall -Werror test/noinherit/05.cpp test/noinherit/05.cpp: In instantiation of ‘const bool z::has_contract_f_<x>::value’: test/noinherit/05.cpp:34: instantiated from here test/noinherit/05.cpp:31: error: the default argument for parameter 1 of ‘static char z::has_contract_f_<T>::check(C*, typename C::contract_f_<0>*) [with C = x, T = x]’ has not yet been parsed

Do you know how I can get around this error?

// File test/noinherit/05.cpp

#include <iostream>

struct x { virtual void f() {} template<int Z> struct contract_f_ {}; };

struct y { void g() {} template<int Z> struct contract_g_ {}; };

struct z: x, y { void f() {} template<int Z> struct contract_f_ {};

template<typename T> struct has_contract_f_ { private: typedef char true_t; struct false_t { true_t dummy[2]; }; static T* make();

static false_t check(...); template<typename C> static true_t check(C*, typename C::template contract_f_<0>* = 0); public: static const bool value = sizeof(check(make())) == sizeof(true_t); // line 31 };

static const bool bx = has_contract_f_<x>::value; // line 34 static const bool by = has_contract_f_<y>::value; };

int main() { std::cout << z::bx << std::endl; std::cout << z::by << std::endl; return 0; }

If has_contract_f_ is declared outside struct z than there is no error -- the code compiles and works. However, I need has_contract_f_ to be a member struct of z, because of the way I use it within my library, and that generates the compiler error reported above.

Thanks a lot. Lorenzo

Compiles fine for me (MSVC9, Vista). - Jeff

On Sat, Mar 27, 2010 at 11:56 PM, Steven Watanabe <watanabesj@gmail.com> wrote:

...

<snip>    template<typename T>    struct has_contract_f_ {    private:        typedef char true_t;        struct false_t { true_t dummy[2]; };        static T* make();

       static false_t check(...);        template<typename C> static true_t check(C*,                typename C::template contract_f_<0>* = 0);    public:        static const bool value =                sizeof(check(make())) == sizeof(true_t); // line 31    };

<snip>

Does it work if you pass 0 explicitly instead of using a default argument.

Unfortunately, it does not work (that is the first thing I tried...). If I change the `value` evaluation to: sizeof(check(make(), 0)) == sizeof(true_t); // line 31 Then `check(...)` is always called and `value` is always false. I _think_ because in this context 0 is always interpreted as a `int` instead of a pointer to `C::contract_f_<0>`. On Sat, Mar 27, 2010 at 10:55 PM, Jeffrey Hellrung <jhellrung@ucla.edu> wrote:

Compiles fine for me (MSVC9, Vista).

I have not tried the code on MSVC yet -- it's good it works on MSVC. I get the error with GCC/Ubuntu (what should the C++ ISO standard behavior be? error or not?).

On 28 March 2010 15:51, Mathieu - <ptr.jetable@gmail.com> wrote:

...

I have not tried the code on MSVC yet -- it's good it works on MSVC. I get the error with GCC/Ubuntu (what should the C++ ISO standard behavior be? error or not?).

What is your version of g++? It compiles fine for me on g++ 4.4.3.

My bad... Dont pay attention to my previous message it doesn't compile. (I copy / pasted the wrong code...)

AMDG Lorenzo Caminiti wrote:

Unfortunately, it does not work (that is the first thing I tried...). If I change the `value` evaluation to:

sizeof(check(make(), 0)) == sizeof(true_t); // line 31

Then `check(...)` is always called and `value` is always false. I _think_ because in this context 0 is always interpreted as a `int` instead of a pointer to `C::contract_f_<0>`.

Okay. You might look at how https://svn.boost.org/trac/boost/ticket/861 solves the problem. In Christ, Steven Watanabe

On Sun, Mar 28, 2010 at 10:44 AM, Steven Watanabe <watanabesj@gmail.com> wrote:

Okay.  You might look at how https://svn.boost.org/trac/boost/ticket/861 solves the problem.

Indeed BOOST_MPL_HAS_TEMPLATE_XXX_TRAIT_DEF() does what I need but I still get the same "default parameter not yet parsed" error if I use this macro at *class* scope -- see below. The Boost.MPL doc indicates this macro should be used at namespace scope (and that works just fine). However, I need to use it to declare the metafunction at class scope -- is there a way to do that? For simplicity, here I am using BOOST_MPL_HAS_XXX_TRAIT_DEF() and I have remove the `Z` template parameter of `contract_f_`. If the macro was to work at class scope than I would use `BOOST_MPL_HAS_TEMPLATE_XXX_TRAIT_DEF()` from the Boost.MPL path indicated by Steven in order to support the `Z` template parameter. $ g++ -Wall -Werror test/noinherit/07.cpp test/noinherit/07.cpp: In instantiation of ‘const bool z::has_contract_f_<x, mpl_::bool_<false> >::value’: test/noinherit/07.cpp:17: instantiated from ‘z::has_contract_f_<x, mpl_::bool_<false> >’ test/noinherit/07.cpp:22: instantiated from here test/noinherit/07.cpp:17: error: the default argument for parameter 1 of ‘static char (& z::has_contract_f_<T, fallback_>::gcc_3_2_wknd::test(const volatile boost::mpl::aux::type_wrapper<U>*, boost::mpl::aux::type_wrapper<typename U::contract_f_>*))[2] [with U = x, T = x, fallback_ = mpl_::bool_<false>]’ has not yet been parsed // File test/noinherit/07.cpp #include <boost/mpl/has_xxx.hpp> #include <iostream> struct x { virtual void f() {} struct contract_f_ {}; }; struct y { void g() {} struct contract_g_ {}; }; struct z: x, y { BOOST_MPL_HAS_XXX_TRAIT_DEF(contract_f_) void f() {} struct contract_f_ {}; static const bool bx = has_contract_f_<x>::value; static const bool by = has_contract_f_<y>::value; }; int main() { std::cout << z::bx << std::endl; std::cout << z::by << std::endl; return 0; } Thanks, Lorenzo

On Sun, Mar 21, 2010 at 2:30 PM, Mathias Gaunard <mathias.gaunard@ens-lyon.org> wrote:

Lorenzo Caminiti wrote:

...

Is there a way I can check if a class has a given member function using metaprogramming?

Yes.

Can you please point me to some code that shows how to do this? Thanks a lot. Lorenzo