Pyo3-pack

Build and publish crates with pyo3, rust-cpython and cffi bindings as well as rust binaries as python packages.

This project was meant as a zero configuration replacement for setuptools-rust. It supports building wheels for python 2.7 and 3.5+ on windows, linux and mac and can upload them to pypi.

Usage

You can either download binaries from the latest release or install it from source:

cargo install pyo3-pack

You can also install pyo3-pack with pip:

pip install pyo3-pack

There are three main subsommands:

• publish builds the crate into python packages and publishes them to pypi.
• build builds the wheels and stores them in a folder (target/wheels by default), but doesn't upload them.
• develop builds the crate and install it's as a python module directly in the current virtualenv

pyo3 and rust-cpython bindings are automatically detected, for cffi or binaries you need to pass -b cffi or -b bin. pyo3-pack needs no extra configuration files, and also doesn't clash with an existing setuptools-rust or milksnake configuration. You can even integrate it with testing tools such as tox (see get-fourtytwo for an example).

The name of the package will be the name of the cargo project, i.e. the name field in the [package] section of Cargo.toml. The name of the module, which you are using when importing, will be the name value in the [lib] section (which defaults to the name of the package). For binaries it's simply the name of the binary generated by cargo.

Pip allows adding so called console scripts, which are shell commands that execute some function in you program. You can add console scripts in a section [package.metadata.pyo3-pack.scripts]. The keys are the script names while the values are the path to the function in the format some.module.path:class.function, where the class part is optional. The function is called with no arguments. Example:

[package.metadata.pyo3-pack.scripts]
get_42 = "get_fourtytwo:DummyClass.get_42"

For pyo3 and rust-cpython, pyo3-pack can only build packages for installed python versions, so you might want to use pyenv, deadsnakes or docker for building. If you don't set your own interpreters with -i, a heuristic is used to search for python installations. You can get a list all found versions with the list-python subcommand.

Build

USAGE:
    pyo3-pack build [FLAGS] [OPTIONS]

FLAGS:
    -d, --debug              Do a debug build (don't pass --release to cargo)
    -h, --help               Prints help information
        --skip-auditwheel    Don't check for manylinux compliance
    -V, --version            Prints version information

OPTIONS:
    -b, --bindings-crate <bindings>
            The crate providing the python bindings. pyo3, rust-cpython and cffi are supported

        --cargo-extra-args <cargo_extra_args>...
            Extra arguments that will be passed to cargo as `cargo rustc [...] [arg1] [arg2] --`

    -i, --interpreter <interpreter>...
            The python versions to build wheels for, given as the names of the interpreters. Uses autodiscovery if not
            explicitly set.
    -m, --manifest-path <manifest_path>             The path to the Cargo.toml [default: Cargo.toml]
    -o, --out <out>
            The directory to store the built wheels in. Defaults to a new "wheels" directory in the project's target
            directory
        --rustc-extra-args <rustc_extra_args>...
            Extra arguments that will be passed to rustc as `cargo rustc [...] -- [arg1] [arg2]`

Publish

USAGE:
    pyo3-pack publish [FLAGS] [OPTIONS]

FLAGS:
    -d, --debug              Do a debug build (don't pass --release to cargo)
    -h, --help               Prints help information
        --skip-auditwheel    Don't check for manylinux compliance
    -V, --version            Prints version information

OPTIONS:
    -b, --bindings-crate <bindings>
            The crate providing the python bindings. pyo3, rust-cpython and cffi are supported

        --cargo-extra-args <cargo_extra_args>...
            Extra arguments that will be passed to cargo as `cargo rustc [...] [arg1] [arg2] --`

    -i, --interpreter <interpreter>...
            The python versions to build wheels for, given as the names of the interpreters. Uses autodiscovery if not
            explicitly set.
    -m, --manifest-path <manifest_path>             The path to the Cargo.toml [default: Cargo.toml]
    -o, --out <out>
            The directory to store the built wheels in. Defaults to a new "wheels" directory in the project's target
            directory
    -p, --password <password>
            Password for pypi or your custom registry. Note that you can also pass the password through
            PYO3_PACK_PASSWORD
    -r, --repository-url <registry>
            The url of registry where the wheels are uploaded to [default: https://upload.pypi.org/legacy/]

        --rustc-extra-args <rustc_extra_args>...
            Extra arguments that will be passed to rustc as `cargo rustc [...] -- [arg1] [arg2]`

    -u, --username <username>                       Username for pypi or your custom registry

Develop

USAGE:
    pyo3-pack develop [FLAGS] [OPTIONS]

FLAGS:
    -h, --help       Prints help information
        --release    Compile in release mode. This is useful e.g. for benchmarking
    -V, --version    Prints version information

OPTIONS:
    -b, --bindings-crate <binding_crate>
            The crate providing the python bindings. pyo3, rust-cpython and cffi are supported

        --cargo-extra-args <cargo_extra_args>...
            Extra arguments that will be passed to cargo as `cargo rustc [...] [arg1] [arg2] --`

    -m, --manifest-path <manifest_path>             The path to the Cargo.toml [default: Cargo.toml]
        --rustc-extra-args <rustc_extra_args>...
            Extra arguments that will be passed to rustc as `cargo rustc [...] -- [arg1] [arg2]`

Cffi

Cffi wheels are compatible with all python versions, but they need to have cffi installed to build (pip install cffi). Until eqrion/cbdingen#203 is resolved, you also need to use a build script that writes c headers to a file called target/header.h

extern crate cbindgen;

use std::env;
use std::path::Path;

fn main() {
    let crate_dir = env::var("CARGO_MANIFEST_DIR").unwrap();

    let mut config: cbindgen::Config = Default::default();
    config.language = cbindgen::Language::C;
    cbindgen::generate_with_config(&crate_dir, config)
        .expect("Unable to generate bindings")
        .write_to_file(Path::new("target").join("header.h"));
}

Manylinux and auditwheel

For portability reasons, native python modules on linux must only dynamically link a set of very few libraries which are installed basically everywhere, hence the name manylinux. The pypa offers a special docker container and a tool called auditwheel to ensure compliance with the manylinux rules. pyo3-pack contains a reimplementation of the most important part of auditwheel that checks the generated library, so there's no need to use external tools. If you want to disable the manylinux compliance checks for some reason, use the --skip-auditwheel flag.

To ship a completely static binary with musl, you can use pyo3-pack build -b bin --cargo-extra-args="--target=x86_64-unknown-linux-musl".

Note that the pyo3-pack pip package is not manylinux compliant (A compliant package, which you get with --no-default-features --features auditwheel, can neither upload nor use the keyring)

Code

The main part is the pyo3-pack library, which is completely documented and should be well integratable. The accompanying main.rs takes care username and password for the pypi upload and otherwise calls into the library.

Without the upload and password-storage features, pyo3-pack itself is manylinux compliant (and has much less dependencies).

There are three different examples, which are also used for integration testing: get_fourtytwo with pyo3 bindings, points crate with cffi bindings and hello-world as a binary. The sysconfig folder contains the output of python -m sysconfig for different python versions and platform, which is helpful during development.

You need to install virtualenv and cffi (pip install virtualenv cffi) to run the tests.

You might want to have look into my blog post which explains the intricacies of building native python packages.