README.md revision 14299
113511Sgabeblack@google.com 213511Sgabeblack@google.com 313511Sgabeblack@google.com# pybind11 — Seamless operability between C++11 and Python 413511Sgabeblack@google.com 513511Sgabeblack@google.com[](http://pybind11.readthedocs.org/en/master/?badge=master) 613511Sgabeblack@google.com[](http://pybind11.readthedocs.org/en/stable/?badge=stable) 713511Sgabeblack@google.com[](https://gitter.im/pybind/Lobby) 813511Sgabeblack@google.com[](https://travis-ci.org/pybind/pybind11) 913511Sgabeblack@google.com[](https://ci.appveyor.com/project/wjakob/pybind11) 1013511Sgabeblack@google.com 1113511Sgabeblack@google.com**pybind11** is a lightweight header-only library that exposes C++ types in Python 1213511Sgabeblack@google.comand vice versa, mainly to create Python bindings of existing C++ code. Its 1313511Sgabeblack@google.comgoals and syntax are similar to the excellent 1413511Sgabeblack@google.com[Boost.Python](http://www.boost.org/doc/libs/1_58_0/libs/python/doc/) library 1513511Sgabeblack@google.comby David Abrahams: to minimize boilerplate code in traditional extension 1613511Sgabeblack@google.commodules by inferring type information using compile-time introspection. 1713511Sgabeblack@google.com 1813511Sgabeblack@google.comThe main issue with Boost.Python—and the reason for creating such a similar 1913511Sgabeblack@google.comproject—is Boost. Boost is an enormously large and complex suite of utility 2013511Sgabeblack@google.comlibraries that works with almost every C++ compiler in existence. This 2113511Sgabeblack@google.comcompatibility has its cost: arcane template tricks and workarounds are 2213511Sgabeblack@google.comnecessary to support the oldest and buggiest of compiler specimens. Now that 2313511Sgabeblack@google.comC++11-compatible compilers are widely available, this heavy machinery has 2413511Sgabeblack@google.combecome an excessively large and unnecessary dependency. 2513511Sgabeblack@google.com 2613511Sgabeblack@google.comThink of this library as a tiny self-contained version of Boost.Python with 2713511Sgabeblack@google.comeverything stripped away that isn't relevant for binding generation. Without 2813511Sgabeblack@google.comcomments, the core header files only require ~4K lines of code and depend on 2913511Sgabeblack@google.comPython (2.7 or 3.x, or PyPy2.7 >= 5.7) and the C++ standard library. This 3013511Sgabeblack@google.comcompact implementation was possible thanks to some of the new C++11 language 3113511Sgabeblack@google.comfeatures (specifically: tuples, lambda functions and variadic templates). Since 3213511Sgabeblack@google.comits creation, this library has grown beyond Boost.Python in many ways, leading 3313511Sgabeblack@google.comto dramatically simpler binding code in many common situations. 3413511Sgabeblack@google.com 3513511Sgabeblack@google.comTutorial and reference documentation is provided at 3613511Sgabeblack@google.com[http://pybind11.readthedocs.org/en/master](http://pybind11.readthedocs.org/en/master). 3713511Sgabeblack@google.comA PDF version of the manual is available 3813511Sgabeblack@google.com[here](https://media.readthedocs.org/pdf/pybind11/master/pybind11.pdf). 3913511Sgabeblack@google.com 4013511Sgabeblack@google.com## Core features 4113511Sgabeblack@google.compybind11 can map the following core C++ features to Python 4213511Sgabeblack@google.com 4313511Sgabeblack@google.com- Functions accepting and returning custom data structures per value, reference, or pointer 4413511Sgabeblack@google.com- Instance methods and static methods 4513511Sgabeblack@google.com- Overloaded functions 4613511Sgabeblack@google.com- Instance attributes and static attributes 4713511Sgabeblack@google.com- Arbitrary exception types 4813511Sgabeblack@google.com- Enumerations 4913511Sgabeblack@google.com- Callbacks 5013511Sgabeblack@google.com- Iterators and ranges 5113511Sgabeblack@google.com- Custom operators 5213511Sgabeblack@google.com- Single and multiple inheritance 5313511Sgabeblack@google.com- STL data structures 5413511Sgabeblack@google.com- Smart pointers with reference counting like ``std::shared_ptr`` 5513511Sgabeblack@google.com- Internal references with correct reference counting 5613511Sgabeblack@google.com- C++ classes with virtual (and pure virtual) methods can be extended in Python 5713511Sgabeblack@google.com 5813511Sgabeblack@google.com## Goodies 5913511Sgabeblack@google.comIn addition to the core functionality, pybind11 provides some extra goodies: 6013511Sgabeblack@google.com 6113511Sgabeblack@google.com- Python 2.7, 3.x, and PyPy (PyPy2.7 >= 5.7) are supported with an 6213511Sgabeblack@google.com implementation-agnostic interface. 6313511Sgabeblack@google.com 6413511Sgabeblack@google.com- It is possible to bind C++11 lambda functions with captured variables. The 6513511Sgabeblack@google.com lambda capture data is stored inside the resulting Python function object. 6613511Sgabeblack@google.com 6713511Sgabeblack@google.com- pybind11 uses C++11 move constructors and move assignment operators whenever 6813511Sgabeblack@google.com possible to efficiently transfer custom data types. 6913511Sgabeblack@google.com 7013511Sgabeblack@google.com- It's easy to expose the internal storage of custom data types through 7113511Sgabeblack@google.com Pythons' buffer protocols. This is handy e.g. for fast conversion between 7213511Sgabeblack@google.com C++ matrix classes like Eigen and NumPy without expensive copy operations. 7313511Sgabeblack@google.com 7413511Sgabeblack@google.com- pybind11 can automatically vectorize functions so that they are transparently 7513511Sgabeblack@google.com applied to all entries of one or more NumPy array arguments. 7613511Sgabeblack@google.com 7713511Sgabeblack@google.com- Python's slice-based access and assignment operations can be supported with 78 just a few lines of code. 79 80- Everything is contained in just a few header files; there is no need to link 81 against any additional libraries. 82 83- Binaries are generally smaller by a factor of at least 2 compared to 84 equivalent bindings generated by Boost.Python. A recent pybind11 conversion 85 of PyRosetta, an enormous Boost.Python binding project, 86 [reported](http://graylab.jhu.edu/RosettaCon2016/PyRosetta-4.pdf) a binary 87 size reduction of **5.4x** and compile time reduction by **5.8x**. 88 89- Function signatures are precomputed at compile time (using ``constexpr``), 90 leading to smaller binaries. 91 92- With little extra effort, C++ types can be pickled and unpickled similar to 93 regular Python objects. 94 95## Supported compilers 96 971. Clang/LLVM 3.3 or newer (for Apple Xcode's clang, this is 5.0.0 or newer) 982. GCC 4.8 or newer 993. Microsoft Visual Studio 2015 Update 3 or newer 1004. Intel C++ compiler 17 or newer (16 with pybind11 v2.0 and 15 with pybind11 v2.0 and a [workaround](https://github.com/pybind/pybind11/issues/276)) 1015. Cygwin/GCC (tested on 2.5.1) 102 103## About 104 105This project was created by [Wenzel Jakob](http://rgl.epfl.ch/people/wjakob). 106Significant features and/or improvements to the code were contributed by 107Jonas Adler, 108Lori A. Burns, 109Sylvain Corlay, 110Trent Houliston, 111Axel Huebl, 112@hulucc, 113Sergey Lyskov 114Johan Mabille, 115Tomasz Miąsko, 116Dean Moldovan, 117Ben Pritchard, 118Jason Rhinelander, 119Boris Schäling, 120Pim Schellart, 121Henry Schreiner, 122Ivan Smirnov, and 123Patrick Stewart. 124 125### License 126 127pybind11 is provided under a BSD-style license that can be found in the 128``LICENSE`` file. By using, distributing, or contributing to this project, 129you agree to the terms and conditions of this license. 130
