Pcons

A modern open-source cross-platform zero-install Python-based build system. Builds anything that requires a repeatable workflow, using a dependency graph. Easy to use, reliable and quick. Uses Ninja (or Makefile, XCode, or MSVS) to do the builds. Optimized for C/C++, Fortran, CUDA, wasm etc. but should work for anything that needs building.

Overview

Pcons is inspired by SCons and CMake, taking a few of the best ideas from each:

• From SCons: Environments, Tools, dependency tracking, Python as the configuration language
• From CMake: Generator architecture (configure once, build fast), usage requirements that propagate through dependencies

Key design principles:

• Configuration, not execution: Pcons generates Ninja files; Ninja executes the build
• Python is the language: No custom DSL—build scripts are real Python with full IDE support
• Language-agnostic: Build C++, Rust, LaTeX, protobuf, or anything else
• Explicit over implicit: Dependencies are discoverable and traceable
• Extensible: Add-on modules for domain-specific tasks (plugin bundles, SDK configuration, etc.)

Language and OS Support

Pcons currently supports C and C++, ObjC and C++, Swift, Fortran, CUDA, assembly, WASM (Emscripten/WASI), LaTeX, Python native extensions, and Rust (via integration). The list is always growing, and it's easy to create your own tools. All built-in tools work on Mac, Windows and Linux. It has cross-compile presets for Android NDK and iOS.

Pcons can build not just executables, but also tar/zip archives, python packages, Windows installers, and MacOS installer packages.

Why another software build tool?

I was one of the original developers of SCons, and helped maintain it for many years. I love that python is the config language; that makes build descriptions incredibly flexible and powerful. Recently I've been using CMake for more projects, and despite the deeply painful configuration language, I've come to appreciate its power: conan integration, the separation between describing the build andrunning it, and dependency propagation, among other things. I feel that SCons hasn't kept up with modern python; like any very widely used mature project, it has a lot of accumulated wisdom but also a bit ossified ways of doing things.

I've been thinking for years now about rearchitecting SCons onto a modern python stack with Path and decorators and all the other wonderful stuff python has been doing, and fixing some of the pain points at the same time (substitution/quoting, extensibility, tracing, separation between description and building, and more), but I've never had the time to dig into it. But recently as I've been using a lot more of Claude Code as a programming assistant, and it has gotten significantly better, it seemed like the right time to try this as a collaborative project. So, meet pcons!

Here's a comparison between pcons and other common modern build tools. I think pcons fills a real need, for a general-purpose broadly applicable extensible software build tool using a modern well-known language to describe builds and tools.

Status

✅ Ready for small-scale production use - and still under active development, so feedback is very welcome. It's working in several medium-sized projects.

Core functionality is working and well tested: C/C++/Fortran compilation, static and shared libraries, programs, install targets, installers (Win/Mac), and mixed-language builds. See ARCHITECTURE.md for design details.

Rust / Go interop

pcons links Rust crates into C/C++ projects: project.CargoBuild() drives cargo build and links the resulting library, with optional cbindgen header generation. See examples/43_rust_cxx_hybrid (hand-written FFI header) and examples/44_rust_cxx_cbindgen (cbindgen-generated header). Go support is straightforward to add on the same pattern — open an issue if you'd like it.

Quick Example

# pcons-build.py
from pcons import Project

project = Project("myapp")
env = project.Environment(toolchain="c")
env.cc.flags.extend(["-Wall"])

# Build a static library
lib = project.StaticLibrary("core", env, sources=["src/core.c"])
lib.public.include_dirs.append("include")

# Build a program that links the library
app = project.Program("myapp", env, sources=["src/main.c"])
app.private.link_libs.append(lib)

uvx pcons # generate build.ninja and run it, producing build/myapp (or build/myapp.exe)

Installation

No installation needed, if you have uv; just use uvx pcons to configure and build. uvx pcons --help for more info. If you want to install it, though:

# Install as a CLI tool (recommended)
uv tool install pcons
pcons ...

# Or add to a project's dependencies
uv add pcons

# Or with pip
pip install pcons

Verifying Release Signatures

Release artifacts on the GitHub Releases page are signed with Sigstore using short-lived certificates issued via GitHub Actions OIDC; transparency-log records are stored at rekor.sigstore.dev. Each .tar.gz and .whl has a matching .sigstore.json bundle. To verify with cosign:

cosign verify-blob \
  pcons-x.y.z.tar.gz \
  --bundle pcons-x.y.z.tar.gz.sigstore.json \
  --new-bundle-format \
  --certificate-identity-regexp='https://github.com/DarkStarSystems/pcons/.*' \
  --certificate-oidc-issuer='https://token.actions.githubusercontent.com'

Documentation

• User Guide is at ReadTheDocs
• Porting from CMake - Side-by-side guide for migrating CMake projects to pcons
• ARCHITECTURE.md - Design document and implementation status
• CONTRIBUTING.md - How to contribute

Development

# Run tests
uv run pytest

# Run linter
make lint

# Format code
make fmt

# Or use uv directly
uv run ruff check pcons/
uvx ty check pcons/ examples/

This Project is AI-Assisted

PCons is my long-term vision for a modern build tool. I've used Claude Code extensively to assist in creating this project, mostly Claude Opus 4.6. It has been a huge help in realizing the vision I've had for a long time. If you reflexively or morally reject all AI-generated or AI-assisted code, pcons is not for you. That said, I've reviewed every decision and nearly every line, and this code reflects my vision, my architecture, my goals and my priorities. I take full responsibility for it, and as a professional software engineer with 40+ years of C/C++/python experience I stand behind it. I also intend to support it long-term.

One of my sub-goals has been to make sure the documentation and source organization is clear; not just for humans but for browsing by AI agents. I want to make it easy for a human or an AI agent to create a best-practices pcons-build.py for your project quickly and easily. Using AI to auto-generate doc and making sure APIs are clean and consistent helps with that goal.

License

MIT License - see LICENSE