This post walks through how to hermetically build a python interpreter in bazel and set up the toolchain.


When sharing a bazel workspace with multiple collaborators, I want to make sure everyone is using the same python version. This is particularly important when some the pip packages we use are sensitive to python minor version (e.g. pysopg2-binary).

One solution is to check-in pre-compiled python interpreter binaries as suggested in [1] and [2]. However, I still need to deal with developers on different operating systems (macOS vs Ubuntu) and the thought of having binary blobs next to source code makes me sad.

My Solution

The idea is to download and build the python interpreter from source and add the binary to bazel’s toolchain. You can see an example here:

Step 1: Build Python Interpreter

In WORKSPACE, I use an http_archive rule to create an external repository where the python interpreter source code is fetched and built. I configured the installation to store the artifacts in the external workspace and exposed a symlink to the resulting binary via exports_files.

Note: you can achieve the same result with a custom repository rule.


load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")

# Special logic for building python interpreter with OpenSSL from homebrew.
# See
_py_configure = """
if [[ "$OSTYPE" == "darwin"* ]]; then
    ./configure --prefix=$(pwd)/bazel_install --with-openssl=$(brew --prefix openssl)
    ./configure --prefix=$(pwd)/bazel_install

    name = "python_interpreter",
    urls = [""],
    sha256 = "dfab5ec723c218082fe3d5d7ae17ecbdebffa9a1aea4d64aa3a2ecdd2e795864",
    strip_prefix = "Python-3.8.3",
    patch_cmds = [
        "mkdir $(pwd)/bazel_install",
        "make install",
        "ln -s bazel_install/bin/python3 python_bin",
    build_file_content = """
    name = "files",
    srcs = glob(["bazel_install/**"], exclude = ["**/* *"]),
    visibility = ["//visibility:public"],

Step 2: Set Up Python Toolchain

Nothing special here, just need to set up py_runtime and toolchain.

# BUILD.bazel

    name = "python3_runtime",
    files = ["@python_interpreter//:files"],
    interpreter = "@python_interpreter//:python_bin",
    python_version = "PY3",
    visibility = ["//visibility:public"],

    name = "my_py_runtime_pair",
    py2_runtime = None,
    py3_runtime = ":python3_runtime",

    name = "my_py_toolchain",
    toolchain = ":my_py_runtime_pair",
    toolchain_type = "@bazel_tools//tools/python:toolchain_type",



At this point, any py_binary rules you have should run with the custome interpreter we’ve just built!

Bonus: Getting pip_import to work

Note: This is a work in progress with an open PR

In order for pip_import to install packages with the correct python version, we need to specify the custom interpreter.


    name = "py_deps",
    requirements = "//:requirements.txt",
    python_interpreter_target = "@python_interpreter//:python_bin",