Mailman 3 February 2002 - Cplusplus-sig

to_python part two
by Robert Andre Feb. 27, 2002

Feb. 27, 2002

Lets see if this makes more sense. I am including only the relevant parts of the code below. I have a C++ class which is meant to act like an array and defined as follows, ------------- uarray.h ------------ typedef valarray<size_t> Shape ; // set of indices, e.g. (3,5) class uarray { public: uarray(const Shape&) ; // construction double operator[](const Shape&) ; // return a single double uarray operator[](size_t) ; // return a slice } ----------------------------------- and I have a file for mapping my Shape class to python tuples, ---------- shape.h ------------ BOOST_PYTHON_BEGIN_CONVERSION_NAMESPACE BOOST_PYTHON_DECL PyObject* to_python(const Shape&) ; BOOST_PYTHON_DECL Shape from_python(PyObject*, boost::python::type<Shape>) ; Shape from_python(PyObject*, boost::python::type<const Shape&>) ; BOOST_PYTHON_END_CONVERSION_NAMESPACE ------------------------------- and a separate wrapping code for boost, ------------- arraymod.cc ----------------- BOOST_PYTHON_MODULE_INIT(arraymod) { try { python::module_builder arraymod("arraymod") ; // --- class builders --- python::class_builder<uarray> uarray_class(arraymod, "uarray") ; uarray_class.def((double (uarray::*)(const Shape&) const) &uarray::operator[], "__getitem__") ; } catch(...) { // python::handle_exception() ; throw ; } ; } ------------------------------------------- All of this works great so that I can do the following >> u = uarray((5,10)) >> print u[(2,3)] 0.0 HOWEVER, what I can't do now but I would like to be able to do is the following in python >> u = uarray((5,10)) >> print u[(2,3)] # same as before 0.0 >> print u[3][2] # <--NEW-- slice first then return [2] of slice 0.0 where u[3] returns a rank one uarray using the uarray::operator[](size_t) method and u[3][2] returns the double value. This requires that the method uarray::operator[](size_t) return a double for rank 1 arrays or a uarray if the rank>1. C++ doesn't do multiple return types for the same argument type but python does so my solution (I hope) is to create a helper function which deals directly with the python argument and return types, -------------- uarray_helper.cc ------------------ # include <boost/python/detail/config.hpp> # include <boost/python/detail/wrap_python.hpp> # include <boost/python/detail/none.hpp> # include <boost/python/detail/signatures.hpp> # include <boost/python/detail/extension_class.hpp> # include <boost/smart_ptr.hpp> # include <boost/python/errors.hpp> PyObject* uarray_getitem(const uarray& u, PyObject* p) { PyObject* result = 0 ; ... code ... uarray uslice = u[k] ; // find slice, int k extracted from p result = to_python(uslice) ; // <-- COMPILER COMPLAINS -- convert uarray to PyObject* ... more code ... return result ; } ; -------------------------------------------------- and wrap it instead of operator[]() ------------- arraymod.cc ----------------- BOOST_PYTHON_MODULE_INIT(arraymod) { try { python::module_builder arraymod("arraymod") ; // --- class builders --- python::class_builder<uarray> uarray_class(arraymod, "uarray") ; uarray_class.def(&uarray_helper, "__getitem__") ; // <----- CHANGE ---- } catch(...) { // python::handle_exception() ; throw ; } ; } ------------------------------------------- My problem is in compiling uarray_helper.cc. to_python(uarray) is known from extension_class.hpp but it complains about py_extension_class_converters. I am unfortunately at my limit of C++ knowledge. g++ -I/usr/local/include/python2.2 -I/usr/local/include -c -g -ftemplate-depth-32 -o uarray_helper.o uarray_helper.cc /usr/local/include/boost/python/detail/extension_class.hpp: In function `PyObject* boost::python::to_python(const T&) [with T = uarray]': uarray_helper.cc:41: instantiated from here /usr/local/include/boost/python/detail/extension_class.hpp:393: `py_extension_class_converters' undeclared (first use this function) /usr/local/include/boost/python/detail/extension_class.hpp:393: (Each undeclared identifier is reported only once for each function it appears in.) Dave, thanks for any help you can provide. Boost python is really slick and I am hoping to be able to use it here at the princeton plasma physics lab. rob andre

2 1

to_python
by Robert Andre Feb. 27, 2002

Feb. 27, 2002

I have a c++ array type called uarray which I would like to wrap so that the python return type might be a double or a new uarray. I am trying to create a helper function such as PyObject* uarray_getitem(const uarray&, PyObject* p) ; to wrap as __getitem__. But I am having trouble because I want to call to_python(my_uarray) when the return type should be a uarray but the compiler complains about not having a definition for the to_python(uarray) function: /usr/local/include/boost/python/detail/extension_class.hpp: In function `PyObject* boost::python::to_python(const T&) [with T = PPPL_RGA::uarray]': array_boost.cc:41: instantiated from here /usr/local/include/boost/python/detail/extension_class.hpp:393: `py_extension_class_converters' undeclared (first use this function) /usr/local/include/boost/python/detail/extension_class.hpp:393: (Each undeclared identifier is reported only once for each function it appears in.) So, a couple of questions: 1) Is creating a helper function with PyObjects the right approach? 2) Is there a way to declare to_python() functions for user types explicitly in a header outside of the BOOST_PYTHON_MODULE_INIT block? thanks, rob andre

2 1

Q: wrap an abstract class in v2
by Min Xu Feb. 23, 2002

Feb. 23, 2002

I am wondering how can I wrap an abstract class in v2. the technique for v1 fails. Can anyone show me an example using boost.python v2? The abstract class has a undefined virtual function. Thanks. Min

2 1

Help with Boost 1.27.0
by Winkelman, Jeff Feb. 22, 2002

Feb. 22, 2002

I was wondering if someone could help me out. Several months ago I wrote a Python extension using Boost, and most recently had to add some features to it. After upgrading to version 1.27.0, and figuring out how to get the new build system operational, I rebuilt my project. After successfully building, I then imported my module in python and attempted to use some of the exposed classes. I have the following problem. After creating an object, any method on that object that I try to execute fails with the following error: >>> x.enableDebug(1) Traceback (most recent call last): File "<stdin>", line 1, in ? RuntimeError: Access violation - no RTTI data! I am running Windows 2000, using Visual C++ as my development environment, and I am also running Python 2.1. Originally I was trying this with Python 2.2. Should Boost still work for 2.2 if I'm not trying to use any of the new typing / features in 2.2? I re-built against 2.1 and still have the same problem. If any one has any ideas, I'd appreciate your help. Thanks, Jeff

2 1

Lifetime of Python object for a pointer argument
by Min Xu Feb. 20, 2002

Feb. 20, 2002

Hi, I am a newbie of boost python. In the process of converting a SWIG wrapping of a C++ library to boost python, I encounter the following problem: // C++ part class grid { public: grid() { ... } ... }; class geometry { public: geometry(const grid * gd) : gd(gd) {} const grid* getGrid() const { return gd; } private: const grid* gd; }; // Wrapper #include <boost/python/class_builder.hpp> namespace python = boost::python; PyObject* to_python(const grid* g) { return to_python(*g); // convert const grid* in terms of const grid& } BOOST_PYTHON_MODULE_INIT(ngf2) { try { python::module_builder this_module("ngf2"); python::class_builder<grid> grid_class(this_module, "grid"); grid_class.def(python::constructor<>()); python::class_builder<geometry> geometry_class(this_module, "geometry"); geometry_class.def(python::constructor<const grid*>()); } ... } The compilation works well (the newest CVS version of boost python). In python: >>> from ngf2 import * >>> a=geometry (grid()) >>> a.getGrid ().a() 8.6723632812500586 # corrupted as grid() is deleted automatically? >>> g=grid() >>> b=geometry (g) >>> b.getGrid ().a() 1.0 # expected. >>> del g >>> b.getGrid ().a() 1.3285421879284814e-269 # corrupted I understand it may be a ref counting problem. Is there any way to make the first idiom work? Thanks! Min Xu

3 15

RE: [C++-sig] First time pointers, first crash...
by Stefan Franke Feb. 19, 2002

Feb. 19, 2002

> -----Original Message----- > From: c++-sig-admin(a)python.org [mailto:c++-sig-admin@python.org]On > Behalf Of David Abrahams > Sent: Tuesday, February 19, 2002 4:24 PM > To: c++-sig(a)python.org > Subject: Re: [C++-sig] First time pointers, first crash... [...] > What does "supposed to" mean? > What does "generically" mean? Excuse me if it's not clear what I meant. English is not my native language. I wanted to ask is just if I have to learn jam to use those parts of Boost that consist not only of header files. The docs for Boost.Python, for example, specify the commands to build the lib and to build & run the test suite. That's all. So as a newbie I don't learn how to - build a single example or other test project - pass configuration options or defines to the build process - what the relevant files are And the perspective of having to learn again "something like make" is pure horror -- for an outsider, perhaps jam is really easy to use & learn. Maybe a "quick intro to jam for Boost users" could help here. Best regards, Stefan

2 1

RE: [C++-sig] First time pointers, first crash...
by Stefan Franke Feb. 19, 2002

Feb. 19, 2002

> > My question: Can you point my to the position where the new operator > > for the wrapper object is called. Then I could put a > breakpoint there > > and see which allocator is actually used. > > line 334 of boost/python/detail/extension_class.hpp > Thank you very much. That helped a lot. It seems my problem is that I have false allocator/deallocator pairs because the destructors in <new> are inlined, whereas there's a global operator new which is used because it is linked to the app (even if not declared in any header file). In out app the latter is usually superposed by another (inlined) operator new, Boost.Python of course doesn't see it. Strange, because I'm building Boost.Python as a DLL project. I have to do more research...

1 0

Wrapping already refcounted C++ types
by Stefan Franke Feb. 19, 2002

Feb. 19, 2002

Dear group, I would like to wrap a couple of C++ classes in a library that already implements its own reference counting. The instances are never copied by value but referenced either (a) by plain C pointers or (b) by their respective smart pointer class So I have to write to_python() and from_python() conversions for these pointer types. The behaviour I'd like to implement for case (a) is that the wrapper object increments the original instance's refcount when created and releases the reference on destruction. My questions are: - Where are the right places to put the respective incref/decrefs? - What to do in case (b) Best regards Stefan Franke

2 1

RE: [C++-sig] First time pointers, first crash...
by Stefan Franke Feb. 19, 2002

Feb. 19, 2002

> -----Original Message----- > From: c++-sig-admin(a)python.org [mailto:c++-sig-admin@python.org]On > Behalf Of David Abrahams > Sent: Monday, February 18, 2002 10:27 PM > To: c++-sig(a)python.org > Subject: Re: [C++-sig] First time pointers, first crash... [...] > Yes. I presume you know that BOOST_DEBUG_PYTHON requires _DEBUG and a > specially-compiled version of Python with reference-count > tracking enabled. > I built and ran your example with Boost.Build, which takes care of all > configuration issues for me. > > > If so it's very strange and I have no clue what's going on here. > > If you're also using Boost.Build, then neither do I. No I'm not. I'm building Boost.Python as part of an MSVC workspace that contains 65 other subprojects -- our app is rather big. My reasons for not using Boost.Build are - I don't know how to use jam. I'm able to build & run the demos like shown in the docs, but I don't even know how the set up and run this little test example like you did. Maybe you can give me a few hints. - Our app is cross-platform (Win, Mac OS, OS-X). Jam is too, of course, but I don't have an idea how to manage all compile & link options for the various platforms within jam. - I'm doing something experimental here. My co-workers are quite sceptical about "this Python thing" anyway, and expecting them (espacially or Mac guys) to use yet another build tool for a simgle subproject is pretty much. One general question here: Are users of Boost in general supposed to learn jam? Since I consider my case pretty "real-world", I hope there is - and ever will be - a way and some documentation how to build the libs generically. Stefan

3 2

RE: [C++-sig] First time pointers, first crash...
by Stefan Franke Feb. 19, 2002

Feb. 19, 2002

As far as I understand the problem, the crash occurs because a non-matching pair of allocator / deallocator is used and the debug heap manager complains about an illegal address. HEAP[boosttest.exe]: Invalid Address specified to RtlValidateHeap( b60000, 1be9ac8 ) Of course, I thought, this is because our app uses an own global new operator so I took this out (I had tried this even before my first posting). But unluckily this does not change anything. As can be seen from the stack trace in my previous post, the problem occurs at the following point: [classes.hpp: line 275] template <class T> void class_t<T>::delete_instance(PyObject* obj) const { delete downcast<T>(obj); } I think this is the location where the wrapper object itself is deleted (which shouldn't affect the wrapped object - does it?). In my example, the wrapper object is created by Boost.Python for the return value of the wrapped GetFoo() function. foo *GetFoo() { return myFoo; } [...] this_module.def(GetFoo, "getfoo"); My question: Can you point my to the position where the new operator for the wrapper object is called. Then I could put a breakpoint there and see which allocator is actually used. Thanks, Stefan

2 1