Convert-XY: type-safe interchange of C++ and Python containers for NumPy extensions

At compile time, pattern matching is applied to the structure of the C++ type of y. This yields two pieces of information: how to check for compatible structure in the Python type (e.g. a dictionary mapping strings to lists, tuples, arrays or sequences is compatible with map>) and which set of converters is applicable (e.g. sequences vs. dictionaries). Dynamic type checkers and converter functions can be built with this static type information at compile time. Further, the library statically determines when an array buffer pointer can be reused, preventing unnecessary copying which can be very useful for programs operating on large data sets. LIBCVD, a popular computer vision library used in our work, provides a BasicImage class, which can be used to wrap two-dimensional image addressing around a buffer pointer. Convert-XY avoids copying and allocation by reusing the NumPy buffer pointer whenever it can: void example(PyObject *x) { map > y; convert(x, y); }