Mailman 3 July 2003 - Cplusplus-sig

RE: [C++-sig] Re: FW: Injected constructors
by Roman Yakovenko July 24, 2003

July 24, 2003

> -----Original Message----- > From: David Abrahams [mailto:dave@boost-consulting.com] > Sent: Thursday, July 24, 2003 12:32 PM > To: c++-sig(a)python.org > Subject: [C++-sig] Re: FW: Injected constructors > > > "Roman Yakovenko" <romany(a)actimize.com> writes: > > > Just a question what is wrong with static member function or global > > function make_xxx? > > Good point. init_factory was going to be a global function, but I > guess I like "make_init" much better than "init_factory". Thanks! I think I was misunderstood. I will explain my self using code. > > The idiom of constructing object and then making additional > > initialization is already used in Boost.Python - borrowed function > > for constructing handle. > > Sure; that's > > .def(init_factory(f)) > > or, I guess: > > .def(make_init(f)) > > right? > > > Also in Python code it will be written explicit what programmer is > > going to do. Did I missed something? > > I don't know. What point are you trying to make with this remark? class MyClass { ... MyClass(); //after constructing object stays in state 1 MyClass( int i ); //after constructing object stays in state 2 MyClass( const YourClass &x); //after constructing object stays in state 3 ... }; Now if I understand right there is a need in other state of the MyClass object. MyClass make_myclass_from_other( const SomeOther &y ); BOOST_PYTHON_MODULE(staticmethod_ext) { class_<MyClass> ... .staticmethod(make_myclass_from_other) ... While in python I'll write obj = MyClass.make_myclass_from_other( third_party_class ) As for me it is much clear then adding an other constructor for the class. At the moment I wrote I thought about class that already has a few constructors. Now I understand that it is not the only case. May be you have nice small class with one constructor and you want to add an other one. I propose those names: make_object / make_instance > -- > Dave Abrahams > Boost Consulting > www.boost-consulting.com > > > _______________________________________________ > C++-sig mailing list > C++-sig(a)python.org > http://mail.python.org/mailman/listinfo/c++-sig >

4 3

Injected constructors
by David Abrahams July 24, 2003

July 24, 2003

It's already possible to inject new methods into a Python class wrapper that are not in the underlying class: #include <vector> struct X { std::vector<int> v; }; int size(X& self) { return self.v.size(); } BOOST_PYTHON_MODULE(test) { class_<X>("X") .def("size", size) ; } >>> x = X() >>> x.size() 0 Currently, there's no easy way to do the same thing for constructors, without intruding on the underlying class or writing a wrapper class which repeats *all* of the base constructors and adds new ones. I am about to add just such a facility: std::auto_ptr<X> x_factory(PyObject* self, int n) { std::auto_ptr<X> x(new X); x->v.resize(n); return x; } BOOST_PYTHON_MODULE(test) { class_<X>("X") .def("size", size) .def("__init__", init_factory(x_factory)) ; } >>> x = X(30) >>> x.size() 30 init_factory creates a python callable object which calls its function argument and injects the resulting object into a holder in the self argument. I'm not in love with the "init_factory" name. Anyone have a better idea? -- Dave Abrahams Boost Consulting www.boost-consulting.com

3 10

GCC type demangling and better error messages
by David Abrahams July 23, 2003

July 23, 2003

Hi, Yesterday I checked in changes which cause types to print in human-readable format with GCC >= 3.1. I have just checked in major changes which provide all the infrastructure for printing readable error messages when no overload matches a call, a project which we are doing under contract to LBL. Expect to see real results in the next couple of days. Major improvements to the health of the CVS state on many different compilers were also made. Greetings, Dave -- Dave Abrahams Boost Consulting www.boost-consulting.com

2 4

Re: is_polymorphic and unions (was: Pyste and STL types)
by spex66＠gmx.net July 23, 2003

July 23, 2003

Hi Nicodemus Hi Dave I've read all your threads belonging to pyste / boost / union actually I've a project with a lot of such unions as you described. The project is from www.sedris.org for a standardised api for geometric & georeferenced data... and I want to port them as quick as possible to python :) so I'm using pyste--AllFromHeader('xxx.h') approach 36 hours from starting yesterday... I'm stuck with a sort of problems (1) error C1204:Compiler limit:internal structure overflow (as described in boost/faq) (2) error C2664: problem with call_policy (?) (3) error C2504: in ..../type_traits/is_polymorphic.hpp (79) ---> problem of a union solution to (1) is obviously splitting... sigh -> any chance to got this as a feature from pyste, Nicodemus? solution to (2) is pending... detailled description is following (also for (3)): ::::::::::::::::::::::::::::::::::: PYSTE::>> sedris.cpp def("SRM_ChangeCoordinateSRF", &SRM_ChangeCoordinateSRF); ::::::::::::::::::::::::::::::::::: ERROR::>> boost/python/make_function.hpp(77) : error C2664: 'struct boost::mpl::list4<short, struct SRM_SRFParamsPair *, struct SRM_Coordinate const *, struct SRM_Coordinate * > __cdecl boost::python::detail::get_signature( short(__cdecl *)(struct SRM_SRFParamsPair * ,const SRM_Coordinate *,SRM_Coordinate *))' : can not convert parameter 1 from' short (__cdecl *)(struct SRM_SRFParamsPair *const ,const SRM_Coordinate *,SRM_Coordinate *)' in short (__cdecl *)(struct SRM_SRFParamsPair * ,const SRM_Coordinate *,SRM_Coordinate *)' This conversion requires a reinterpret_cast, a C-style cast or function-style cast ::::::::::::::::::::::::::::::::::: SOURCE::>> for (2): **snippet from file:: srm.h & srm_types.h** EXPORT_DLL extern SRM_Status_Code SRM_ChangeCoordinateSRF ( const SRM_SRF_Parameters_Pair coord_op_params_ptr, const SRM_Coordinate *source_coord_ptr, SRM_Coordinate *dest_coord_ptr ); typedef struct SRM_SRFParamsPair *SRM_SRF_Parameters_Pair; typedef struct { SRM_Dimensionality dimensionality; union { SRM_Coordinate_2D two_d; SRM_Coordinate_3D three_d; } coordinate; } SRM_Coordinate; **end of snippet from file:: srm.h & srm_types.h** ::::::::::::::::::::::::::::::::::: PYSTE::>>sedris.cpp class_< SRM_SRF_Parameters >("SRM_SRF_Parameters", init< >()) .def(init< const SRM_SRF_Parameters & >()) .def_readwrite("dimensionality", &SRM_SRF_Parameters::dimensionality) .def_readwrite("parameters", &SRM_SRF_Parameters::parameters) ; ::::::::::::::::::::::::::::::::::: ERROR::>> error C2504: in ..../type_traits/is_polymorphic.hpp (79) ---> problem of a union ::::::::::::::::::::::::::::::::::: SOURCE::>>for (3): **snippet from file:: srm_types.h** typedef struct { SRM_Dimensionality dimensionality; union { SRM_SRF_Parameters_2D two_d; SRM_SRF_Parameters_3D three_d; } parameters; } SRM_SRF_Parameters; **end of snippet from file:: srm_types.h** TRY&ERROR::>> I've tried to adopt Dave's tip from http://mail.python.org/pipermail/c++-sig/2003-March/003607.html > If so, then we have a bug in is_polymorphic. If not, we should > change all uses of is_polymorphic<T> in > > boost/python/class.hpp > boost/python/object/class_converters.hpp > boost/python/object/inheritance.hpp > > to: > mpl::and_<is_class<T>, is_polymorphic<T> > putting #include for mpl::_and but boost wan't compile (if you need the error-trace, I will post it on demand) so I'm stuck with UNION, and have no C/C++ wrapper-idea cause of a massive lack of C-knowledge... sorry hot-blood pythonista ;) any hints to direct me to a path to the light is very appreciated! thanks for your patience Peter (=PA=)

3 2

exceptions and multiple modules
by greg Landrum July 23, 2003

July 23, 2003

[boost 1.30, g++ 3.2, redhat 8.0, python 2.2] We've been having strange order dependence problems when using exceptions and multiple extension modules. I've attached an extremely simplified example that demonstrates the problem. The example includes two small extension modules, moda and modb, each of which defines a single function that throws an IndexError. If I do: >>> import modb >>> import moda >>> modb.tossitb() everything works fine (I get the expected IndexError) If, however, I reverse the import error: >>> import moda >>> import modb >>> modb.tossitb() I get the dreaded "RuntimeError: unidentifiable C++ exception" Any ideas as to what might be going on here? "Jamsegment" includes the lines I added to the $Boost/libs/python/test Jamfile to build things. Thanks -greg

4 10

Re: is_polymorphic and unions
by spex66＠gmx.net July 23, 2003

July 23, 2003

Hi Dave, thanks for your fast helping hand :) (don't know how to respond in thread from a mail-client?!?) >> solution to (1) is obviously splitting... sigh -> any chance to got this as >> a feature from pyste, Nicodemus? > >The current CVS includes support for "--multiple": >http://tinyurl.com/hp3j as far as I could see, its a splitting at header-file level... that's not enough in my case. Maybe a hardcore classlevel would be a nice option, too? >> This conversion requires a reinterpret_cast, a C-style cast or >> function-style cast > >This is an msvc bug which preserves top-level constness of arguments >in function (pointer) types. If you can redeclare the offending >function as: > > short SRM_ChangeCoordinateSRF(SRM_SRF_Parameters_Pair, const SRM_Coordinate *,SRM_Coordinate *); I'm a little bit challenged with C++ actually :/ Do you mean 'redeclare' in its original position in the header-file (srm.h in this case)? If so... in exactly the way you give the example, or the original EXPORT_DLL extern SRM_Status_Code SRM_ChangeCoordinateSRF ( const SRM_SRF_Parameters_Pair coord_op_params_ptr, const SRM_Coordinate *source_coord_ptr, SRM_Coordinate *dest_coord_ptr ); style (deleting first occur of const)? Or have I to redeclare it in the mSedris.cpp (build from pyste)... dumb as I am (belonging to C++ of course), I've tried to include your line in front of BOOST_PYTHON_MODULE(mSedris) but my VC6 told me that's not the right solution sorry for that kind of question-level... I'm happy to make my first steps with boost.python.object... the examples looks terrific!!! > >you can still make the first parameter const in the function >*definition* without upsetting the compiler. If you can't change the >declaration of 'whatever' you'll have to resort to forcibly casting >the offending function pointer. > >I suppose Pyste could be patched to do this for you... This would be great :) >> ::::::::::::::::::::::::::::::::::: >> ERROR::>> >> error C2504: in ..../type_traits/is_polymorphic.hpp (79) ---> problem of a >> union > >What's the complete error message for that line? here it comes (followed by matching pyste-result, and original header-file-snippets) parts translated from german VC6 messages into matching english one: ..\..\..\boost/type_traits/is_polymorphic.hpp(26) : error C2569: 'type' : enum/union cannot be used as a base class C:\arwa\dev\sedris_sdk_3.1.2\include\srm_types.h(5165) : see reference to class template instantiation 'type' being compiled ..\..\..\boost/type_traits/is_polymorphic.hpp(79) : see reference to class template instantiation 'boost::detail::is_polymorphic_imp1<union SRM_SRF_Parameters::type>' being compiled ..\..\..\boost/type_traits/is_polymorphic.hpp(84) : see reference to class template instantiation 'boost::detail::is_polymorphic_imp<union SRM_SRF_Parameters::type>' being compiled ..\..\..\boost/python/object/make_ptr_instance.hpp(37) : see reference to class template instantiation 'boost::is_polymorphic<union SRM_SRF_Parameters::type>' being compiled ..\..\..\boost/python/object/make_ptr_instance.hpp(30) : see reference to class template instantiation 'struct _typeobject *__cdecl boost::python::objects::make_ptr_instance<union SRM_SRF_Parameters::type,struct boost::python::objects::pointer_holder<union SRM_SRF_Parameters::type *,union SRM_SRF_Parameters::type> >::get_class_object_impl(volatile const SRM_SRF_Parameters::type *)' being compiled PYSTE-AGAIN::>> class_< SRM_SRF_Parameters >("SRM_SRF_Parameters", init< >()) .def(init< const SRM_SRF_Parameters & >()) .def_readwrite("dimensionality", &SRM_SRF_Parameters::dimensionality) .def_readwrite("parameters", &SRM_SRF_Parameters::parameters) ; SOURCE-AGAIN::>> typedef struct { SRM_Dimensionality dimensionality; union { SRM_SRF_Parameters_2D two_d; SRM_SRF_Parameters_3D three_d; } parameters; } SRM_SRF_Parameters; I have not reproduce the >> mpl::and_<is_class<T>, is_polymorphic<T> > stuff. But from my memory, trying this last time... do you mean to replace occurences only from is_polymorphic<T> or also is_polymorphic<Target> // is_polymorphic<Source> // is_polymorphic<Base> in the same style? Thats what I have tried last time (plus # include <boost/mpl/and.hpp>) :::::::::::::::::::: And a new kind of compiler-error, maybe time to use boost.array? :::::::::::::::::::::::::::::::::: PYSTE::>> class_< SRM_Matrix_3x3 >("SRM_Matrix_3x3", init< >()) .def(init< const SRM_Matrix_3x3 & >()) .def_readwrite("mat", &SRM_Matrix_3x3::mat) ; :::::::::::::::::::::::::::::::::: ERROR::>> ..\..\..\boost/python/data_members.hpp(68) : error C2440: '=' : 'const double [3][3]' cannot convert in 'double [3][3]' Es gibt keinen Kontext, in dem diese Konvertierung moeglich ist ..\..\..\boost/python/data_members.hpp(55) : Bei der Kompilierung der Member-Funktion 'struct _object *__cdecl boost::python::detail::member< double [3][3], struct SRM_Matrix_3x3, struct boost::python::default_call_policies >::set(double (SRM_Matrix_3x3::*)[3][3], struct _object *, struct _object *, const struct boost::python::default_call_policies &)' der Klassenvorlage :::::::::::::::::::::::::::::::::: SOURCE::>> typedef struct { SRM_Long_Float mat[3][3]; } SRM_Matrix_3x3; thanks again for your help and your time, but as it is in software-project life... time is the REAL bottleneck, *sigh* but building wonders and miracles in zero-time is the real thrill :) Peter (=PA=)

3 2

libraries and source install required to use Boost python?
by Dave Reed July 22, 2003

July 22, 2003

I've just tried boost last week so I'm not up on all the details yet. First, thanks to all the developers - I'm very impressed and looking forward to using it more. I finally got it to work in a separate directory using a Jamfile with based on the "project" example. My C++ code uses OpenGL so by looking at examples, it seems I need to add: <gcc><*><find-library>GLU <gcc><*><find-library>GL Is this the correct way to do it (it appears to work fine)? What if the library is in a non-standard place - do you specify -I and -L compiler flags somehow? I didn't see much documentation on creating Jamfiles. Is there specific documentation on these type of things? I'm also uncertain what the comments in the Jamrules file are saying: # Establish the root of the boost installation. Most targets will want # $(BOOST_ROOT) in their #include path. $(gTOP) is the name of the variable # containing the path from the invocation directory to the project root. path-global BOOST_ROOT : $($(gTOP))/../../.. ; What include path is it talking about? Should I get setting CFLAGS to -I//usr/local/src/boost_1_30_0 Also, I just downloaded the latest Red Hat beta and it now includes boost for the first time (or at least part of it). On my other system where I installed boost from scratch, I've got a boost-build.jam that points to where boost is installed. boost-build.jam # Specify path to the boost build system BOOST_ROOT = /usr/local/src/boost_1_30_0 ; boost-build /usr/local/src/boost_1_30_0/tools/build ; I'm not certain what to do on the Red Hat beta system since there's no source. Here's the RPMS they've split it into: boost-1.30.0-3 boost-devel-1.30.0-3 boost-doc-1.30.0-3 boost-jam-3.1.4-2 boost-python-1.30.0-3 boost-python-devel-1.30.0-3 I can post a list of the files (rpm -ql for each rpm) from the boost RPMS if anyone wants or put them on a web site. Thanks, Dave

2 1

importing module
by Mattheos Kakaris July 21, 2003

July 21, 2003

while running bjam in the boost_1_30_0/libs/python/example/tutorial folder for the hello.cpp example it creates the following directory structure: /bin/hello.so/gcc/debug/runtime-link-dynamic/shared-linkable-true When running Python and trying to import module hello, it rightfully gives the message: ImportError: No module named hello What should I do so that python will always know were to look for the module? Kind regards Mat

3 2

VC++ 7.1 (VStudio.NET 2003) linking problem
by Mike Rovner July 21, 2003

July 21, 2003

I'm trying to move my project from VC++ 7.0 to 7.1 There was no problem under 7.0 I have several classes to wrap separated to several source files, i.e. file1.cpp #include <boost/python.hpp> ... f(){ class_<C1>("C1")...; } file2: #include <boost/python.hpp> ... class_<C0>("C0")...; f(); When linking I got the following: error LNK2005: "class boost::arg<3> `anonymous namespace'::_3" (?_3@?A0x2264d27d@@3V?$arg@$02@boost@@A) already defined in file1.obj repeating for every holder from _1 to _9. Googling and boost archive searching were unsuccessful. boost was taken from latest CVS. Any clues and/or links are very appreciated. Regards, Mike

5 15

Explaining the Magic
by Brett Calcott July 20, 2003

July 20, 2003

Hi David, I am in Aussie now, and though still homeless, I at least have an office. I thought we could make a start on our plans to document some of the boost python implementation. I brushed up on reStructuredText so, as discussed I'll try and construct a document from our conversations. As suggested I have had a look at the luabind conversation, and though I understand bits here and there I need something to tie it all too. Here's two starting questions: 1) What happens when you do the following: struct boring {}; ...etc... class_<boring>("boring") ; There seems to be a fair bit going on. - Python needs a new ClassType to be registered. - We need to construct a new type that can hold our boring struct. - Inward and outward converters need to be registered for the type. Can you gesture in the general direction where these things are done. 2) Can you give a brief overview of the data structures that are present in the registry and an overview of the process that happens as a type makes its way from c++ to python and back again. In both of these cases, I'm quite capable of reading code - but the thing I don't get from scanning the source is a sense of the architecture, both structurally, and temporally (er, I mean in what order things go on). Cheers, Brett (Kiwi in Oz) ps. Take your time to answer these -- I am trying to write a paper, and get some foundation boost_python simulation framework written, so I am happy if we progress slowly...

2 2