@pycross_image

Bazel starlark rules for building container images from py_binary using @rules_pycross ✨.

Provides:

• load("@pycross_image//bazel/rules:oci.bzl", "py_image")
  • image rule compatible with @rules_oci
• load("@pycross_image//bazel/rules:docker.bzl", "py_image")
  • image rule compatible with @rules_docker

Installation & Usage

See [releases] page for an http_archive of the latest @pycross_image.

Examples:

• @rules_oci example
• @rules_docker example

A few notes about the workspace setup:

• It's divided into "steps" based on load statement dependencies. step1.bzl only depends on things declared in repositories.bzl, step2.bzl depends on things declared in step1.bzl, etc (this pattern is from tensorflow).
• Your workspace may already contain some of these dependencies. Also, selected external workspace names may be differ from yours. Use the example as a study guide rather than canonical reference.
• The examples are only tested with the older WORKSPACE. The rules may not be compatible with bzlmod yet.

How it Works

graph TD;
    pypi[(pypi)]
    DefaultInfo[[DefaultInfo]]
    numpy{{numpy}}
    grpclib{{grpclib}}

    subgraph linux_x86_64
        image.tar-->image;
        image-->app_layer;
        image-->site_packages_layer;
        image-->interpreter_layer;
        app_layer-->DefaultInfo;
        site_packages_layer-->DefaultInfo;
        interpreter_layer-->DefaultInfo;
    end

    DefaultInfo--transition :linux_x86_64-->pycross_binary;
    pycross_binary-->py_binary;
    py_binary-->numpy;
    py_binary-->grpclib;

    numpy-.->numpy-cp310-macosx_arm64.whl
    numpy-.->numpy-cp310-manylinux_x86_64.whl
    grpclib-.->grpclib.whl
    numpy-cp310-macosx_arm64.whl-->pypi
    numpy-cp310-manylinux_x86_64.whl-->pypi

    subgraph zig_toolchain
        grpclib.whl-->grpclib-tar.gz
    end

    grpclib-tar.gz-->pypi

    style pypi fill:#3171b2,stroke:#333
    style numpy fill:#5d97d2,stroke:#3171b2,color:black
    style grpclib fill:#5d97d2,stroke:#3171b2,color:black
    style grpclib.whl stroke:#bc082b

    style numpy-cp310-manylinux_x86_64.whl stroke:#bc082b
    style numpy-cp310-macosx_arm64.whl stroke:#bb22d8
    style linux_x86_64 stroke:#bc082b,fill:none

In this example the py_binary rule has deps on two python wheels:

• numpy has a binary wheel available from pypi for both the darwin and linux platforms.
• grpclib only has a source distribution available.

The pycross_binary rule transitions from the host platform to :linux_x86_64.

• the transition affects how @rules_pycross fetches wheels. If the binary distribution is available, take it.
• if the binary distribution is not available, compile from source using (in this case, with zig and @hermetic_cc_toolchains).

The image is partitioned into three tar layers by matching against filename patterns (see rule implementation for details).