Extension

Python’s flexibility comes at a performance cost. When you need speed for numerical computing, system interfaces, or wrapping existing C/C++ libraries, native extensions bridge the gap. This section covers multiple approaches:

• ctypes - Standard library FFI for calling C functions without compilation

• Python C API - Traditional approach with maximum control (legacy)

• pybind11 (recommended for C++) - Clean C++11 syntax with automatic type conversions, used by PyTorch, TensorFlow, and SciPy

• cffi - Cleaner alternative to ctypes with PyPy compatibility

• Cython - Python-like syntax that compiles to C for gradual optimization

For most new projects wrapping C++ code, pybind11 is the recommended choice. For calling existing C libraries without a build step, use ctypes or cffi.

• ctypes
• Python C API
• Modern C/C++ Extensions