Pythran

0.18.1 · active · verified Sat Apr 11

Pythran is an Ahead-of-Time (AOT) compiler that translates a subset of Python, primarily focused on scientific computing and numerical kernels, into native C++14 modules. It aims to provide C++/Fortran-like performance for Python code, leveraging multi-cores and SIMD instructions. The current stable version is 0.18.1, and it actively maintains a release cadence of several updates per year, supporting Python 3.7 and upward.

Warnings

breaking Pythran dropped support for Python 2.7 after version 0.9.5. It now exclusively supports Python 3.7 and upward.
Fix: Ensure your project uses Python 3.7 or newer.
gotcha Pythran only supports a subset of the Python language, specifically targeting numerical computing. Features like classes, polymorphic variables (with some exceptions), and heterogeneous containers (except tuples) are not supported.
Fix: Design your functions to adhere to Pythran's supported Python subset, typically pure numerical functions operating on primitive types and NumPy arrays.
gotcha A C++14 enabled compiler (e.g., g++ >= 5, clang >= 3.5) is a prerequisite for Pythran. Older compilers may lead to compilation errors.
Fix: Update your system's C++ compiler to a version that supports C++14 standards.
gotcha For optimal performance, users often need to explicitly specify compiler optimization flags (e.g., `-O3`, `-fopenmp`, `-march=native`) either via the `pythran` CLI or in a `~/.pythranrc` configuration file.
Fix: Consult Pythran's documentation for recommended optimization flags and consider setting them in `~/.pythranrc` or passing them directly to the `pythran` command.
deprecated Support for `numpy.distutils` with Python 3.9 was obsoleted in Pythran 0.18.0. Users integrating Pythran into larger build systems might need to adapt.
Fix: Migrate build systems away from `numpy.distutils` if targeting Python 3.9+ with Pythran 0.18.0+.
gotcha Windows support for Pythran is ongoing and primarily targets Visual Studio 2017 or clang-cl. Specific environment variables (`CXX`, `CC`) might be required for proper setup.
Fix: Refer to the Pythran documentation for detailed Windows installation and configuration instructions, especially regarding compiler setup.

Install

pip install pythran Install Pythran

Imports

compiled_module
```
import compiled_module
```
Pythran compiles a Python file (e.g., `my_module.py`) into a native shared library (e.g., `my_module.so` or `my_module.pyd`) which can then be imported directly as a standard Python module.
pythran.magic
```
%load_ext pythran.magic
```
For IPython/Jupyter integration, load the magic function to compile code directly in cells using `%%pythran`.

Quickstart

Create a Python file with numerical kernels annotated for export. Compile it using the `pythran` command-line tool, then import the resulting native module. This example defines a dot product function.

# dprod.py
""" Naive dotproduct! Pythran supports numpy.dot """
#pythran export dprod(int list, int list)
def dprod(l0, l1):
    """WoW, generator expression, zip and sum."""
    return sum(x * y for x, y in zip(l0, l1))

# To compile the Python file to a native module:
# pythran dprod.py

# Then, import and use the compiled module:
# python -c 'import dprod; print(dprod.dprod([1,2,3], [4,5,6]))'

view raw JSON →