p2w - Python to WebAssembly Compiler

p2w compiles a substantial subset of Python to WebAssembly, leveraging WASM GC for automatic memory management.

Features

Supported Python Features

• Data types: integers (arbitrary precision), floats, booleans, strings, bytes, lists, tuples, dicts, sets
• Control flow: if/elif/else, for/while loops, break/continue, match statements
• Functions: definitions, default arguments, *args/**kwargs, closures, lambdas, decorators
• Classes: inheritance, properties, static/class methods, special methods (__init__, __str__, etc.)
• Comprehensions: list, dict, set, generator expressions
• Exception handling: try/except/finally, raise, exception chaining
• Context managers: with statement
• Generators: yield, generator functions
• Other: f-strings, type annotations (ignored at runtime), walrus operator, unpacking

JavaScript Interop

p2w provides seamless JavaScript interoperability via the js module:

import js

canvas = js.document.getElementById("chart")
ctx = canvas.getContext("2d")
ctx.fillRect(0, 0, 100, 100)
js.console.log("Hello from Python!")

What Works

Browser Demos

The demos/ directory contains two browser demos:

• data-dashboard: Bar chart visualization using Canvas API
• simulation: Physics simulation with real-time rendering

NB: They are not working yet. They were ported from https://github.com/abilian/prescrypt/tree/main/demos but never finished.

Golden Programs

The programs/internal/ directory contains 104 test programs covering the supported Python subset. These serve as both regression tests and documentation of working features, including:

• Classes with inheritance, properties, and special methods
• Generators and comprehensions
• Exception handling with chaining
• Context managers
• Match statements
• F-strings and string operations
• Collection types and methods

Benchmarks

Two benchmark suites validate correctness and measure performance:

programs/benchmarks/ - Classic benchmarks adapted for p2w:

• fibonacci, primes, sieve, matmul
• fannkuch, binarytrees, nbody
• mandelbrot, spectralnorm, fasta
• pystone

programs/benchmarks-alioth/ - Benchmarks from the Debian Benchmark Game with GCC baseline comparison:

• binarytrees, nbody, spectralnorm, mandelbrot, fannkuchredux
• Includes a runner script for automated comparison against GCC (-O3 -ffast-math)

Installation

# Clone the repository
git clone https://git.sr.ht/~sfermigier/p2w
cd p2w

# Install with uv
uv sync

Usage

Quick Start

Given a Python source file fib.py:

def fib(n):
    a, b = 0, 1
    for _ in range(n):
        a, b = b, a + b
    return a

print(fib(30))

The simplest way to compile and run it:

uv run p2w -r fib.py
# Output: 832040

Step by Step

If you want to inspect or keep the intermediate files:

# 1. Compile Python to WAT (WebAssembly Text format)
uv run p2w fib.py -o fib.wat

# 2. Convert WAT to WASM binary (requires wasm-tools)
wasm-tools parse fib.wat -o fib.wasm

The generated WASM cannot run standalone — it imports host functions for I/O (write_char, write_i32, etc.). The p2w -r flag handles this automatically by generating a Node.js loader that provides these imports and runs the module. For browser execution, see the demos in demos/.

CLI Options

p2w [-h] [-o OUTPUT] [-r] [-v] [-d] [-V] source

  -o, --output FILE   write WAT to file instead of stdout
  -r, --run           compile and run immediately (requires wasm-tools + Node.js)
  -v, --verbose       show compilation details on stderr
  -d, --debug         dump AST and debug info to stderr
  -V, --version       show version

As a Library

from p2w import compile_to_wat

wat_code = compile_to_wat('print("hello")')

Examples

Fibonacci

def fib(n: int) -> int:
    a, b = 0, 1
    for i in range(n):
        a, b = b, a + b
    return a

print(f"fib(30) = {fib(30)}")

Browser Demo

The demos/ directory contains browser examples:

• data-dashboard: Interactive bar chart visualization
• simulation: Physics simulation

To run a demo:

cd demos/data-dashboard
make  # Builds app.wasm
# Serve with any HTTP server and open index.html

Development

# Run tests
make test

# Run linting and type checking
make lint

# Format code
make format

# Run tests with coverage
make test-cov

Test Structure

• tests/a_unit/ - Unit tests
• tests/b_integration/ - Integration tests
• tests/c_e2e/ - End-to-end tests

Architecture

p2w follows a 7-phase compilation pipeline:

1. Parse: Python source → AST (using Python's ast module)
2. Inline: Small functions inlined at call sites
3. Analyze: Scope analysis, variable collection, generator detection
4. Infer types: Forward-flow type inference, escape analysis, native type eligibility
5. Generate code: AST → WAT via singledispatch visitor pattern
6. Emit: WAT output combining user code with runtime library (types, builtins, helpers)
7. Assemble: WAT → WASM (via wasm-tools)

The generated WASM uses GC extensions for automatic memory management. Execution requires a JavaScript host (Node.js or browser) that provides I/O and runtime imports.

See docs/architecture.md for full details including the type system, runtime architecture, calling convention, and JavaScript interop.

Requirements

• Python 3.12+
• wasm-tools (for WAT to WASM conversion)
• Node.js 22+ (for running compiled WASM via p2w -r)
• For browser execution: any browser with WASM GC support (recent Chrome/Firefox)

Prior Art and References

It's too long at this point to cite every influence and/or alternative. Here are a few that stand out:

• Compiling Scheme to WebAssembly - Huge influence on unstucking the project
• Compylo - A previous attempt, started by Ethan Zouzoulkowsky (then student at EPITA) in 2023
• Prescrypt - A Python->JS compiler, itself based on PScript

License

MIT