PyPy

PyPy is an alternative implementation of Python that aims to combine speed and compatibility. It originated in 2003 as a research project by Armin Rigo and others, initially focusing on a flexible and trace-friendly interpreter. The project gained prominence with its RPython meta-tracing framework, which allows the interpreter to be written in a restricted subset of Python and then automatically generate a JIT compiler. This approach, known as 'meta-tracing', enables PyPy to achieve significant performance gains over CPython, the reference implementation, often running pure Python code several times faster.¹

PyPy's architecture is built around RPython (Restricted Python), a statically typed subset of Python. The interpreter is written in RPython, and the PyPy toolchain translates it to C code, adding a tracing JIT compiler. The JIT works by recording traces of frequently executed loops and optimizing them, which can lead to dramatic speedups for compute-intensive tasks. PyPy also includes a garbage collector that is often more efficient than CPython's reference counting, further improving performance. However, PyPy's JIT requires warm-up time, so short-running scripts may not benefit, and it uses more memory in some cases.²

PyPy aims for high compatibility with CPython, supporting most of the Python standard library and many third-party packages. However, some C extensions that rely on CPython's internal APIs may not work without adaptation, as PyPy uses a different C API layer called CPyExt. This has historically limited its use with packages like NumPy, though progress has been made with the 'numpy' package via the 'pypy' wheels. PyPy also supports the Python 3 language specification, with recent versions targeting Python 3.10. For developers, PyPy offers a drop-in replacement for CPython in many scenarios, especially for long-running server processes or CPU-bound applications.³

PyPy's JIT compiler can provide speedups of 2 to 10 times over CPython on pure Python code, with typical benchmarks showing around 4 times faster execution. The JIT is particularly effective for loops and numerical operations, but less so for I/O-bound or short-lived programs. PyPy also offers a 'sandboxing' feature that allows untrusted code to run in a restricted environment, which is unique among Python implementations. Additionally, PyPy's memory usage can be higher than CPython's due to its garbage collector, but it often reduces memory fragmentation. For applications that require predictable low latency, PyPy's JIT warm-up can be a drawback, but for long-running services, it often outperforms CPython.⁴

Beyond its main role, PyPy has several lesser-known features. It includes a 'STM' (Software Transactional Memory) experimental branch that aimed to enable multi-threaded execution without the Global Interpreter Lock (GIL), though it is not actively maintained. PyPy also supports 'PyPy.js', a port to JavaScript via Emscripten, which allows Python to run in web browsers, though it is experimental. The project has a 'PyPy' logo featuring a snake, and its development is funded by donations and grants, including from the Python Software Foundation. Notably, PyPy's JIT is also used in other projects, such as the 'Topaz' Ruby implementation, demonstrating the versatility of the RPython toolchain.⁵

JIT

Just-in-Time compiler, which compiles code at runtime to improve performance.

RPython

Restricted Python, a subset of Python used to write interpreters that can be translated to efficient C code.

CPython

The reference implementation of Python, written in C.

GIL

Global Interpreter Lock, a mutex that protects CPython's internal state, limiting multi-threading.