Python is a mature and widely used language and there is a large interest in improving its performance by compiling functions and methods directly to machine code rather than executing instructions in the interpreter. We have already seen a compiler example in Chapter 4, C Performance with Cython, where Python code is enhanced with types, compiled to efficient C code, and the interpreter calls are side-stepped.

In this chapter, we will explore two projects--Numba and PyPy--that approach compilation in a slightly different way. Numba is a library designed to compile small functions on the fly. Instead of transforming Python code to C, Numba analyzes and compiles Python functions directly to machine code. PyPy is a replacement interpreter that works by analyzing the code at runtime and optimizing the slow loops automatically.

These tools are called Just-In-Time (JIT) compilers because the compilation is performed at runtime rather than before running the code (in other cases, the compiler is called ahead-of-time or AOT).

The list of topics to be covered in this chapter is as follows:

• Getting started with Numba
• Implementing fast functions with native mode compilation
• Understanding and implementing universal functions
• JIT classes
• Setting up PyPy
• Running the particle simulator with PyPy
• Other interesting compilers