On 04/18/2012 01:57 PM, Christoph Rupp wrote:
Hi,

i have a C++ class that i'm wrapping with boost::python:

class Foo {
   // ...
   const char *get_buffer();
   size_t get_size() const;
   // ...
};

I want to send this class through a thrift interface (which uses the
write() method to send the data to a file or a socket). For a few days now
i'm trying to implement this. By now i understand that i have to implement
the buffer protocol, and i know that i can create a buffer with
boost::python (read-only access is sufficient):

inline object as_buffer(const Foo&f) {
   return object (handle<>  (PyBuffer_FromMemory ((void*)f.get_buffer(),
f.get_size())));
}

(This is based on an email from 2009 that i found in this mailing list's
archive.)

But i do not know how to glue it together to form a full module. This is
how the module looks right now:

BOOST_PYTHON_MODULE(Foo)
{
   class_<Foo, boost::noncopyable>("Foo",
           init<int32_t, bool>())
     .def("add", afn)
     .def("finalize",&Foo::finalize)
     .def("empty",&Foo::empty)
     .def("clear",&Foo::clear)
   ;
}

How can i include that buffer conversion?


Sorry about the long delay in a reply, but I'm afraid that right now it looks *really* tricky (possibly impossible) to add the buffer protocol to a Boost.Python class. Python doesn't have a special named methods for the buffer protocol:

http://stackoverflow.com/questions/2079272/which-methods-implement-the-buffer-interface-in-python

And Boost.Python doesn't give you access to the PyTypeObject structure where you'd be able to add the PyBufferProcs slots.

I think you have two options:

- Wrap your buffer-getter function with another name and find a way to get your serialization code to use that.

- Wrap this class class using the raw Python C-API and use Boost.Python custom lvalue converters to make it work with any Boost.Python-wrapped code you have. You could use Boost.Python to wrap all the member functions, for instance, and add them to the Python C-API class's dict.

Neither option is great, but hopefully one will allow you to move forward.

HTH!

Jim
_______________________________________________
Cplusplus-sig mailing list
Cplusplus-sig@python.org
http://mail.python.org/mailman/listinfo/cplusplus-sig

Reply via email to