A release of the Nuitka 4.0 project is available, developing a compiler for translating Python scripts into a C representation, which can then be compiled into an executable file using libpython for maximum compatibility with CPython (using native CPython tools for manipulating objects). Provided full compatibility with current releases of Python 2.6, 2.7, 3.4 - 3.13. Compared to CPython, compiled scripts show a 335% performance improvement in pystone tests. The project code is distributed under the Apache license.
In the new version:
- Changes have been made to improve compatibility with Python 3.12, 3.13, and 3.14, such as adding support for lazy annotation processing and new type annotation syntax for generic classes and functions.
- Added experimental support for using the Zig compiler (zig cc) as a C compiler backend in Nuitka.
- Added preliminary support for LTO optimizations when using the Clang compiler.
- The decorator "@nuitka_ignore" is proposed to exclude compilation of functions.
- The following options have been implemented: "--project" for building using the settings from the pyproject.toml file, "--devel-profile-compilation" for profiling at the build stage, "--debug-self-forking" for debugging fork bombs, "--include-windows-runtime-dlls" for controlling the inclusion of C runtime DLLs in Windows, "--qt-debug-plugins" for diagnosing loading of Qt plugins.
- Added support for current versions of the oracledb, win32ctypes, dask, dataparser, puremagic, pygments.lexers, tomli, av, sentry_sdk, jedi, parso, line_profiler, and pandas packages.
Additionally, a progress report on the development of the JIT compiler in CPython is worth noting. The developers have set a goal of increasing JIT performance by 5% by the release of CPython 3.15, and by 10% by the release of 3.16, along with the ability to use the JIT in free-threading mode (without global locking). Currently, using the JIT in alpha builds of CPython 3.15 already demonstrates an average speedup of 11-12% on the platform. macOS AArch64 and 5-6% on the x86_64 platform Linux Compared to the standard interpreter, performance improvements in individual tests with JIT enabled range from over 100% acceleration to 20% slowdown.
Source: opennet.ru
