using directive does not work in template function object.

Discussion in 'C++' started by jimking2000@gmail.com, Apr 9, 2009.

  1. Guest

    Hello everyone

    I can't compile the following code.

    compiler: g++ 4.3.2
    OS: openSUSE 11



    #include <functional>
    #include <algorithm>
    #include <iostream>
    #include <string>
    #include <memory>
    #include <vector>


    using namespace std;


    class test
    {
    public:
    void func(int i) { wcout << i << L": test\n"; }
    };


    namespace
    {
    template<class T>
    class destroyer : public unary_function<T, void>
    {
    public:
    //using typename unary_function<T, void>::result_type;
    using typename unary_function<T, void>::argument_type;


    public:
    typename unary_function<T, void>::result_type
    operator()(/*typename unary_function<T, void>::*/argument_type
    pointer) const // compile error here
    {
    delete pointer;
    }
    };
    }


    int main()
    {
    typedef vector<test *> vectest;
    vectest data;

    for (int i = 0; i < 5; ++i)
    data.push_back(new test);

    for_each(data.begin(), data.end(), bind2nd(mem_fun(&test::func), 0));

    for_each(data.begin(), data.end(), destroyer<vectest::value_type>());

    return 0;
    }


    The error message is:
    make all
    Building file: ../src/Test_Cplusplus.cpp
    Invoking: GCC C++ Compiler
    g++ -O0 -g3 -Wall -c -fmessage-length=0 -MMD -MP -MF"src/
    Test_Cplusplus.d" -MT"src/Test_Cplusplus.d" -o"src/Test_Cplusplus.o"
    "../src/Test_Cplusplus.cpp"
    .../src/Test_Cplusplus.cpp:33: error: ‘argument_type’ is not a type
    /usr/include/c++/4.3/bits/stl_algo.h: In function ‘_Funct std::for_each
    (_IIter, _IIter, _Funct) [with _IIter =
    __gnu_cxx::__normal_iterator<test**, std::vector<test*,
    std::allocator<test*> > >, _Funct = <unnamed>::destroyer<test*>]’:
    .../src/Test_Cplusplus.cpp:51: instantiated from here
    /usr/include/c++/4.3/bits/stl_algo.h:3791: error: invalid conversion
    from ‘test*’ to ‘int’
    /usr/include/c++/4.3/bits/stl_algo.h:3791: error: initializing
    argument 1 of ‘typename std::unary_function<T,
    void>::result_type<unnamed>::destroyer<T>::eek:perator()(int) const [with
    T = test*]’
    .../src/Test_Cplusplus.cpp: In member function ‘typename
    std::unary_function<T,
    void>::result_type<unnamed>::destroyer<T>::eek:perator()(int) const [with
    T = test*]’:
    /usr/include/c++/4.3/bits/stl_algo.h:3791: instantiated from ‘_Funct
    std::for_each(_IIter, _IIter, _Funct) [with _IIter =
    __gnu_cxx::__normal_iterator<test**, std::vector<test*,
    std::allocator<test*> > >, _Funct = <unnamed>::destroyer<test*>]’
    .../src/Test_Cplusplus.cpp:51: instantiated from here
    .../src/Test_Cplusplus.cpp:35: error: type ‘int’ argument given to
    ‘delete’, expected pointer
    make: *** [src/Test_Cplusplus.o] error 1

    So, here, the using directive cannot introduce the argument_type of
    the base template.

    While I can use base::argument_type directly.
    typename unary_function<T, void>::result_type
    operator()(typename unary_function<T, void>::argument_type pointer)
    const // OK

    Does this behavior conform to the standard?

    Thanks
    Jim
     
    , Apr 9, 2009
    #1
    1. Advertising

  2. * :
    > Hello everyone
    >
    > I can't compile the following code.
    >
    > compiler: g++ 4.3.2
    > OS: openSUSE 11
    >
    >
    >
    > #include <functional>
    > #include <algorithm>
    > #include <iostream>
    > #include <string>
    > #include <memory>
    > #include <vector>
    >
    >
    > using namespace std;
    >
    >
    > class test
    > {
    > public:
    > void func(int i) { wcout << i << L": test\n"; }
    > };
    >
    >
    > namespace
    > {
    > template<class T>
    > class destroyer : public unary_function<T, void>
    > {
    > public:
    > //using typename unary_function<T, void>::result_type;
    > using typename unary_function<T, void>::argument_type;
    >
    >
    > public:
    > typename unary_function<T, void>::result_type
    > operator()(/*typename unary_function<T, void>::*/argument_type
    > pointer) const // compile error here
    > {
    > delete pointer;
    > }
    > };
    > }
    >
    >
    > int main()
    > {
    > typedef vector<test *> vectest;
    > vectest data;
    >
    > for (int i = 0; i < 5; ++i)
    > data.push_back(new test);
    >
    > for_each(data.begin(), data.end(), bind2nd(mem_fun(&test::func), 0));
    >
    > for_each(data.begin(), data.end(), destroyer<vectest::value_type>());
    >
    > return 0;
    > }
    >
    >
    > The error message is:
    > make all
    > Building file: ../src/Test_Cplusplus.cpp
    > Invoking: GCC C++ Compiler
    > g++ -O0 -g3 -Wall -c -fmessage-length=0 -MMD -MP -MF"src/
    > Test_Cplusplus.d" -MT"src/Test_Cplusplus.d" -o"src/Test_Cplusplus.o"
    > "../src/Test_Cplusplus.cpp"
    > ../src/Test_Cplusplus.cpp:33: error: ‘argument_type’ is not a type
    > /usr/include/c++/4.3/bits/stl_algo.h: In function ‘_Funct std::for_each


    [snip]

    > Does this behavior conform to the standard?


    Not as far as I know. E.g. the code compiles fine with Comeau Online, and I
    don't know any rule of the standard that would make it ill-formed. However, the
    common technique for making such code work with g++ is to use a 'typedef'
    instead of a 'using'-declaration; it's so common that I actually /believed/ it
    was probably required by the standard, but checking I can find no such rule.

    You might consider defining a macro like

    #define USING_BASE_TYPE( name ) typedef Base::name name;

    I just wish there was some less visually imposing convention for macro names.

    It might at first glance seem as if this macro would be of little value, for
    what about multiple inheritance?

    However, in practice the types you need mainly come from a single "main" base class.

    The practical upshot is that with regard to template handling the standard is
    not a document to be relied on, because every compiler, except possibly Comeau,
    has its own special quirks, sort of like Internet Explorer wrt. HTML.

    So the only way to do things portably is to test, test, test, with different
    compilers, and adopt coding conventions like the macro above that work
    cross-compiler.

    By the way, regarding cross-compiler portable code, note that std::wcout is not
    supported by MinGW g++ for Windows... That is, with a default build of that
    compiler, such as the binary that most people install. Given that the standard
    library's wide streams are just so much baggage, of negative real value
    considering the complexity they add via templating of stream functionality, I
    think that's entirely reasonable -- so, Just Say No to the wide streams! :)


    Cheers & hth.,

    - Alf

    --
    Due to hosting requirements I need visits to <url: http://alfps.izfree.com/>.
    No ads, and there is some C++ stuff! :) Just going there is good. Linking
    to it is even better! Thanks in advance!
     
    Alf P. Steinbach, Apr 9, 2009
    #2
    1. Advertising

  3. * Alf P. Steinbach:
    > * :
    >> Hello everyone
    >>
    >> I can't compile the following code.
    >>
    >> compiler: g++ 4.3.2
    >> OS: openSUSE 11
    >>
    >>
    >>
    >> #include <functional>
    >> #include <algorithm>
    >> #include <iostream>
    >> #include <string>
    >> #include <memory>
    >> #include <vector>
    >>
    >>
    >> using namespace std;
    >>
    >>
    >> class test
    >> {
    >> public:
    >> void func(int i) { wcout << i << L": test\n"; }
    >> };
    >>
    >>
    >> namespace
    >> {
    >> template<class T>
    >> class destroyer : public unary_function<T, void>
    >> {
    >> public:
    >> //using typename unary_function<T, void>::result_type;
    >> using typename unary_function<T, void>::argument_type;
    >>
    >>
    >> public:
    >> typename unary_function<T, void>::result_type
    >> operator()(/*typename unary_function<T, void>::*/argument_type
    >> pointer) const // compile error here
    >> {
    >> delete pointer;
    >> }
    >> };
    >> }
    >>
    >>
    >> int main()
    >> {
    >> typedef vector<test *> vectest;
    >> vectest data;
    >>
    >> for (int i = 0; i < 5; ++i)
    >> data.push_back(new test);
    >>
    >> for_each(data.begin(), data.end(), bind2nd(mem_fun(&test::func), 0));
    >>
    >> for_each(data.begin(), data.end(), destroyer<vectest::value_type>());
    >>
    >> return 0;
    >> }
    >>
    >>
    >> The error message is:
    >> make all
    >> Building file: ../src/Test_Cplusplus.cpp
    >> Invoking: GCC C++ Compiler
    >> g++ -O0 -g3 -Wall -c -fmessage-length=0 -MMD -MP -MF"src/
    >> Test_Cplusplus.d" -MT"src/Test_Cplusplus.d" -o"src/Test_Cplusplus.o"
    >> "../src/Test_Cplusplus.cpp"
    >> ../src/Test_Cplusplus.cpp:33: error: ‘argument_type’ is not a type
    >> /usr/include/c++/4.3/bits/stl_algo.h: In function ‘_Funct std::for_each

    >
    > [snip]
    >
    >> Does this behavior conform to the standard?

    >
    > Not as far as I know. E.g. the code compiles fine with Comeau Online,


    That should of course be "As far as I know".

    Grr.


    > and I don't know any rule of the standard that would make it ill-formed.
    > However, the common technique for making such code work with g++ is to
    > use a 'typedef' instead of a 'using'-declaration; it's so common that I
    > actually /believed/ it was probably required by the standard, but
    > checking I can find no such rule.
    >
    > You might consider defining a macro like
    >
    > #define USING_BASE_TYPE( name ) typedef Base::name name;
    >
    > I just wish there was some less visually imposing convention for macro
    > names.
    >
    > It might at first glance seem as if this macro would be of little value,
    > for what about multiple inheritance?
    >
    > However, in practice the types you need mainly come from a single "main"
    > base class.
    >
    > The practical upshot is that with regard to template handling the
    > standard is not a document to be relied on, because every compiler,
    > except possibly Comeau, has its own special quirks, sort of like
    > Internet Explorer wrt. HTML.
    >
    > So the only way to do things portably is to test, test, test, with
    > different compilers, and adopt coding conventions like the macro above
    > that work cross-compiler.
    >
    > By the way, regarding cross-compiler portable code, note that std::wcout
    > is not supported by MinGW g++ for Windows... That is, with a default
    > build of that compiler, such as the binary that most people install.
    > Given that the standard library's wide streams are just so much baggage,
    > of negative real value considering the complexity they add via
    > templating of stream functionality, I think that's entirely reasonable
    > -- so, Just Say No to the wide streams! :)
    >
    >
    > Cheers & hth.,
    >
    > - Alf
    >



    --
    Due to hosting requirements I need visits to <url: http://alfps.izfree.com/>.
    No ads, and there is some C++ stuff! :) Just going there is good. Linking
    to it is even better! Thanks in advance!
     
    Alf P. Steinbach, Apr 9, 2009
    #3
  4. Guest

    On Apr 9, 12:05 pm, "Alf P. Steinbach" <> wrote:
    > * Alf P. Steinbach:
    >
    >
    >
    > > * :
    > >> Hello everyone

    >
    > >> I can't compile the following code.

    >
    > >> compiler: g++ 4.3.2
    > >> OS: openSUSE 11

    >
    > >> #include <functional>
    > >> #include <algorithm>
    > >> #include <iostream>
    > >> #include <string>
    > >> #include <memory>
    > >> #include <vector>

    >
    > >> using namespace std;

    >
    > >> class test
    > >> {
    > >> public:
    > >>  void func(int i) { wcout << i << L": test\n"; }
    > >> };

    >
    > >> namespace
    > >> {
    > >>  template<class T>
    > >>  class destroyer : public unary_function<T, void>
    > >>  {
    > >>  public:
    > >>      //using typename unary_function<T, void>::result_type;
    > >>   using typename unary_function<T, void>::argument_type;

    >
    > >>  public:
    > >>      typename unary_function<T, void>::result_type
    > >>      operator()(/*typename unary_function<T, void>::*/argument_type
    > >> pointer) const // compile error here
    > >>   {
    > >>    delete pointer;
    > >>   }
    > >>  };
    > >> }

    >
    > >> int main()
    > >> {
    > >>     typedef vector<test *> vectest;
    > >>     vectest data;

    >
    > >>  for (int i = 0; i < 5; ++i)
    > >>   data.push_back(new test);

    >
    > >>  for_each(data.begin(), data.end(), bind2nd(mem_fun(&test::func), 0));

    >
    > >>  for_each(data.begin(), data.end(), destroyer<vectest::value_type>());

    >
    > >>     return 0;
    > >> }

    >
    > >> The error message is:
    > >> make all
    > >> Building file: ../src/Test_Cplusplus.cpp
    > >> Invoking: GCC C++ Compiler
    > >> g++ -O0 -g3 -Wall -c -fmessage-length=0 -MMD -MP -MF"src/
    > >> Test_Cplusplus.d" -MT"src/Test_Cplusplus.d" -o"src/Test_Cplusplus.o"
    > >> "../src/Test_Cplusplus.cpp"
    > >> ../src/Test_Cplusplus.cpp:33: error: argument_type is not a type
    > >> /usr/include/c++/4.3/bits/stl_algo.h: In function _Funct std::for_each

    >
    > > [snip]

    >
    > >> Does this behavior conform to the standard?

    >
    > > Not as far as I know. E.g. the code compiles fine with Comeau Online,

    >
    > That should of course be "As far as I know".
    >
    > Grr.
    >
    >
    >
    > > and I don't know any rule of the standard that would make it ill-formed..
    > > However, the common technique for making such code work with g++ is to
    > > use a 'typedef' instead of a 'using'-declaration; it's so common that I
    > > actually /believed/ it was probably required by the standard, but
    > > checking I can find no such rule.

    >
    > > You might consider defining a macro like

    >
    > >   #define USING_BASE_TYPE( name ) typedef Base::name name;

    >
    > > I just wish there was some less visually imposing convention for macro
    > > names.

    >
    > > It might at first glance seem as if this macro would be of little value,
    > > for what about multiple inheritance?

    >
    > > However, in practice the types you need mainly come from a single "main"
    > > base class.

    >
    > > The practical upshot is that with regard to template handling the
    > > standard is not a document to be relied on, because every compiler,
    > > except possibly Comeau, has its own special quirks, sort of like
    > > Internet Explorer wrt. HTML.

    >
    > > So the only way to do things portably is to test, test, test, with
    > > different compilers, and adopt coding conventions like the macro above
    > > that work cross-compiler.

    >
    > > By the way, regarding cross-compiler portable code, note that std::wcout
    > > is not supported by MinGW g++ for Windows... That is, with a default
    > > build of that compiler, such as the binary that most people install.
    > > Given that the standard library's wide streams are just so much baggage,
    > > of negative real value considering the complexity they add via
    > > templating of stream functionality, I think that's entirely reasonable  
    > > --  so, Just Say No to the wide streams! :)

    >
    > > Cheers & hth.,

    >
    > > - Alf

    >
    > --
    > Due to hosting requirements I need visits to <url:http://alfps.izfree.com/>.
    > No ads, and there is some C++ stuff! :) Just going there is good. Linking
    > to it is even better! Thanks in advance!


    Hi Alf

    Thanks for your reply.

    I'd like to write code like this:

    class destroyer : public unary_function<T, void>
    {
    public:
    result_type operator()(argument_type pointer) const
    {
    ...
    }
    };

    instead of this:

    class destroyer : public unary_function<T, void>
    {
    public:
    void operator()(T pointer) const
    {
    ...
    }
    };

    Because if I want to change unary_function<T, void> to
    unary_function<T, int>, I need revise only one place.

    But the customized function object cannot be a template, since the
    template rule is complex and odd. Until I know the using directive, I
    can keep my coding style with template. What baffles me is it is still
    not portable!!!

    So maybe it's time to change my coding style. Althought I wish gcc
    provides the functionality to let the first test case run.

    Regards,
    Jim King
     
    , Apr 9, 2009
    #4
    1. Advertising

Want to reply to this thread or ask your own question?

It takes just 2 minutes to sign up (and it's free!). Just click the sign up button to choose a username and then you can ask your own questions on the forum.

Share This Page