Python Bindings with pybind11: Build and Package C++ Extensions for Python with CMake and NumPy
Reseña del libro "Python Bindings with pybind11: Build and Package C++ Extensions for Python with CMake and NumPy"
Fast C++ Is Only the Beginning A native extension is not production-ready simply because it compiles. If NumPy silently copies data, ownership rules are wrong, the GIL is mishandled, or the wheel fails on another machine, impressive benchmark numbers quickly become support problems. You do not have to choose between Python’s ease of use and C++ performance. Python Bindings with pybind11 shows you how to move demanding work into modern C++ while preserving the clean, familiar Python experience your users expect. You will follow the complete path from identifying a worthwhile bottleneck to distributing a tested extension that others can install and use. Along the way, you will learn to cross the Python–C++ boundary without introducing hidden copies, memory errors, awkward APIs, fragile builds, or packages that work only on your computer. With this practical guide, you will learn how to: Find the Python bottlenecks that are genuinely worth accelerating, so you invest effort where it produces measurable value. Expose C++ functions, classes, inheritance, exceptions, and object models through APIs that feel natural to Python users. Process NumPy 2.x arrays efficiently while controlling dtypes, shapes, strides, memory layout, copying, and ownership. Prevent dangling references and hard-to-diagnose crashes by choosing suitable holders, return policies, and lifetime rules. Build and package reproducibly with CMake, Ninja, pyproject.toml, scikit-build-core, source distributions, and cross-platform wheels. Diagnose numerical failures, configuration errors, import problems, crashes, and memory bugs across both Python and C++. Handle the GIL and native concurrency safely, measure real costs, and optimize the work that affects users most. Complete, progressive examples show how the pieces fit together. A production-focused NumPy capstone then takes you through API design, zero-copy processing, safe parallel execution, testing, packaging, continuous integration, benchmarking, and maintenance. Whether you are a Python developer who needs more speed, a C++ developer bringing an existing library to Python, or a package maintainer replacing a fragile extension workflow, this book gives you a clear path from native code to a dependable Python package. You do not have to choose between Python usability and C++ performance. Build both, start creating fast, Python-friendly extensions you can test, package, and ship with confidence.
