Technical groups
Dropdown arrow
Open source
Consulting services
Technical groups
Dropdown arrow
Open source
Consulting services

Bazel and Nix: A Migration Experience

15 December 2022 — by Ben Radford

I’ve recently been learning how to use Bazel and Nix to build projects. Reading documentation and playing around with codelab exercises is a great way to get started, but when learning a new technology there comes a point where you have to start applying it for real to make further progress. To this end I decided to migrate an old hobby project of mine to use Bazel. This post describes my experience and I hope serves as a useful example for others.

Bazel is a fantastic build system which by itself provides many benefits. However dependency management could be considered an area of weakness for it. This is understandable given its monorepo origins at Google but when not working in a monorepo, what is the best way to provide the dependencies a project needs?

One approach is to simply assume the dependencies exist in the environment (having been put there by apt install or the like), but this is just ignoring the problem. Bazel has the concept of external repositories and is able to fetch and build source code from a variety of places1. Under some circumstances this might be a reasonable solution, but we probably don’t want to spend significant amounts of time building a large dependency tree for a project that itself only takes a few minutes to build.

Instead, Nix can be used to supply the dependencies to Bazel prebuilt. Nix and Bazel are very similar technologies (both build code from source and both place strong emphasis on properties like determinism and hermeticity), but without delving into the details it is perhaps best to think of Nix as a package manager and Bazel as a build system. The advantage of Nix is that while it ostensibly builds everything from source, in practice it usually just downloads artifacts from the Nix binary cache.

The project I migrated is an unfinished game with around fifteen thousand lines of C++ code. It was originally built using some very convoluted makefiles and is structured as follows:

Project structure

This game has a few attributes that make it an interesting case study:

  • It has a non-trivial structure with modules2 shared between client and server executables.

  • It depends on several third party libraries. Previously these were manually built from source. Now with the help of rules_nixpkgs they are provided by Nix. Typically this means they will be fetched prebuilt from the Nix cache, saving the time it would have taken to compile them.

  • In addition to code that needs to be built, the game has assets like 3D meshes and textures that need to be transformed into a format that can be loaded at runtime.

  • One particular transformation is handled by the custom zonebuild tool. If any of the source code that goes into this tool changes and it gets rebuilt, the data transformation it performs also needs to be rerun.

In a more traditional build system it is difficult to express a build graph for the above in a way that gives perfectly correct and incremental builds. Things tend to be more ad-hoc and changes to tools or across language boundaries are usually best dealt with by cleaning and rebuilding from scratch.

Building the source code

Every Bazel project requires a WORKSPACE.bazel file at its root. As the migration advances this file will grow in complexity, but to begin with it simply specifies the workspace name:

workspace(name = "space-game")

A Bazel workspace is organised into packages by BUILD.bazel files. There are many ways you might do this (e.g. a separate package for each module and executable) but for a project of this scale it is reasonable to build all the source code from a single package named //common/src.

As it has no dependencies, building the core module is a good place to start. It just requires a cc_library rule in common/src/BUILD.bazel:

    name = "core",
    hdrs = glob(["core/*.hpp"]),
    srcs = glob(["core/*.hpp", "core/*.cpp"]),
    copts = ["-std=c++17"],
    includes = ["."],
    visibility = ["//visibility:public"],

At this point it is possible to build the core module by running the following command:

bazel build //common/src:core

Bazel does not pollute the source tree with intermediate build files and output files. These are instead created within the outputRoot directory and some convenient symlinks are added to the workspace root. If you follow the bazel-bin symlink you’ll find that cc_library has produced both static and shared libraries:


This is pretty cool. In the old days, linking even a moderately complex project was quite a tricky affair. You had to determine which parts it made sense to link statically and which parts dynamically. Code going into a shared library needed to be compiled with -fPIC because shared libraries might be loaded anywhere in process address space. Getting something wrong usually led to unhelpful linker errors about missing or duplicate symbols. With Bazel all those messy details have been abstracted away.

Next let’s examine how the net module is built. If you check the graph above you’ll see that net depends on core and two third party libraries, boost and enet. As mentioned in the introduction, these libraries will be fetched from Nix. First we need a nixpkgs.nix file:

let nixpkgs = fetchTarball {
  url = "";
  sha256 = "0d643wp3l77hv2pmg2fi7vyxn4rwy0iyr8djcw1h5x72315ck9ik";
in import nixpkgs

This pins the nixpkgs version to git commit ce6aa13, which corresponds to the 22.05 release. You can pick any commit you like but if you are unsure, picking the commit with the latest release tag is a good place to start. The correct sha256 value can be determined by setting it to empty and running bazel build //.... You will get an error telling you what the hash should be:

error: hash mismatch in file downloaded from '':
         specified: sha256:0000000000000000000000000000000000000000000000000000
         got:       sha256:0d643wp3l77hv2pmg2fi7vyxn4rwy0iyr8djcw1h5x72315ck9ik

Next some additions to WORKSPACE.bazel are required:

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

# Import the rules_nixpkgs repository.
    name = "io_tweag_rules_nixpkgs",
    sha256 = "b01f170580f646ee3cde1ea4c117d00e561afaf3c59eda604cf09194a824ff10",
    strip_prefix = "rules_nixpkgs-0.9.0",
    urls = [""],

# Import the transitive dependencies of rules_nixpkgs.
load("@io_tweag_rules_nixpkgs//nixpkgs:repositories.bzl", "rules_nixpkgs_dependencies")

# Import a repository for the version of nixpkgs we pinned in nixpkgs.nix above.
load("@io_tweag_rules_nixpkgs//nixpkgs:nixpkgs.bzl", "nixpkgs_local_repository")
    name = "nixpkgs",
    nix_file = "//:nixpkgs.nix",

# Configure a toolchain from the nixpkgs repository above. This means we will
# use a fixed version of the C++ compiler, which helps with reproducible builds.
load("@io_tweag_rules_nixpkgs//nixpkgs:nixpkgs.bzl", "nixpkgs_cc_configure")
    name = "nixpkgs_config_cc",
    repository = "@nixpkgs",

load("@io_tweag_rules_nixpkgs//nixpkgs:nixpkgs.bzl", "nixpkgs_package")

The nixpkgs_package rule loaded by the last line can be used to import a Nix package as a Bazel repository. It has many optional parameters, allowing its behaviour to be customised precisely as needed (though this requires some familiarity with Nix and Bazel).

The attribute_path parameter indicates the particular Nix package to be imported. You can find attribute paths by searching for a Nix package with the nix-env command:

$ nix-env -f nixpkgs.nix -qaP '.*enet.*'  # Note, the last argument is a regex.
nixpkgs.enet                                           enet-1.3.17
nixpkgs.freenet                                        freenet-build01480

If attribute_path is not given, it defaults to the value given for the name parameter. So for enet it can be left out. Once you’ve found a Nix package you can examine what files it contains:

$ tree $(nix-build nixpkgs.nix -A enet)
|-- include
|   `-- enet
|       |-- callbacks.h
|       |-- enet.h
|       |-- list.h
|       |-- protocol.h
|       |-- time.h
|       |-- types.h
|       |-- unix.h
|       |-- utility.h
|       `-- win32.h
`-- lib
    |-- ->
    |-- ->
    `-- pkgconfig
        `-- libenet.pc

This information will be useful when crafting build_file_content. The value given for this parameter is written to a BUILD.bazel file in the root of the @enet repository. Again in WORKSPACE.bazel:

    name = "enet",
    repository = "@nixpkgs",
    build_file_content = """\
    name = "enet",
    hdrs = glob(["include/**/*.h"]),
    srcs = ["lib/"],
    includes = ["include"],
    visibility = ["//visibility:public"],

The cc_library rule is quite versatile. It is not obvious from the name but its srcs parameter can take prebuilt libraries in addition to source files. Note that there are multiple symlinks for the shared library but you cannot choose one arbitrarily3. You will get a runtime error if you choose wrong, but fortunately the error will indicate the correct choice.

This general pattern is used for all the third party dependencies (including boost) so let’s see how the net module is built:

    name = "net",
    hdrs = glob(["net/*.hpp"]),
    srcs = glob(["net/*.hpp", "net/*.cpp"]),
    copts = ["-std=c++17"],
    includes = ["."],
    deps = [
    visibility = ["//visibility:public"],

Note that the @enet dependency is syntactic sugar for @enet//:enet and this target is just the cc_library(name="enet", ...) rule defined a little earlier. The remaining modules (math, physics and script) are built in a similar fashion.

The final piece of the puzzle is to build the executables. Taking client as an example:

    name = "client",
    srcs = glob(["client/*.hpp", "client/*.cpp"]),
    copts = ["-std=c++17"],
    includes = ["."],
    deps = [
    visibility = ["//visibility:public"],

Apart from using cc_binary, this looks much the same as the rules we have already seen. It depends on some internal modules and also on cegui, ogre and ois. Like enet, these are external repositories provided by rules_nixpkgs.

The asset transformation pipeline

Having seen how the code is built, let’s turn to the game assets. An asset could be a polygon mesh or a texture for a 3D model. It could be a music track, a voice recording or a pre-rendered video. These assets are usually authored and stored in a high resolution lossless format. They must be transformed into a format more suitable for distribution and optimised for runtime use. For proper games the asset pipeline can be quite complicated and might take hours to run in full, but for this project it is simple:

Asset transformations

Files with no incoming arrows are checked into source control while the rest are produced by Bazel during the build. A brief description of each follows:

  • geometry.xml: List of the triangles and faces comprising a mesh.
  • texture.tga: UV mapped texture for the mesh.
  • material.script: A high level representation of a shader program.
  • mesh.bin: An efficient binary format for meshes used by Ogre3D.
  • Contains all the assets needed to render a 3D model.
  • collision.dat: Stores map geometry in a k-d tree for fast collision detection.

Bazel has no built-in rules for these transformations but genrule can be used to run arbitrary Bash commands. To avoid duplication I decided to define a couple of helper macros in a resources.bzl extension file. In Bazel a macro is a function that can instantiate rules. The assemble_assets macro is used in //common/data/ships like this:

load("//common/data:resources.bzl", "assemble_assets")


The version of assemble_assets shown below has been simplified a little for clarity:

load("@rules_pkg//pkg:zip.bzl", "pkg_zip")

def assemble_assets(name):
    # Convert xml mesh data to ogre binary format.
    mesh_converter = "@ogre//:bin/OgreXMLConverter"
        name = "convert_{name}_mesh",
        srcs = [":{name}-geometry.xml"],
        outs = ["{name}-mesh.bin"],
        tools = [mesh_converter],
        cmd = """$(execpath {mesh_converter}) \
            $(execpath {name}-geometry.xml) \
            $(execpath {name}-mesh.bin)

    # Package the assets into a zip.
        name = "zip_{name}_assets",
        out = "{name}",
        srcs = [

The most interesting thing here is that the OgreXMLConverter tool is being used to transform mesh data. This tool comes as part of the ogre Nix package and it must be explicitly exported from BUILD.bazel. It also needs to be passed to the tools parameter of the genrule or else Bazel will complain that it has not been declared as a prerequisite.

Ships are treated as spherical objects by the physics engine so they do not need their geometry to be transformed into a k-d tree for collision detection, but for maps this extra step is needed:

def assemble_map(name):

    # Build k-d tree for collision detection.
        name = "{name}_collision",
        srcs = [":{name}-geometry.xml"],
        tools = ["//common/src:zonebuild"],
        cmd = """$(execpath //common/src:zonebuild) \
            $(execpath {name}-geometry.xml) \
            $(execpath {name}-collision.dat)
        outs = ["{name}-collision.dat"],

Maps have meshes, textures and materials and the transformation of those is delegated to assemble_assets. Then the k-d tree is built by invoking zonebuild from another genrule. This is essentially the same as invoking OgreXMLConverter except this time the tool is built in the main repository instead of coming prebuilt from an external repository.

Beyond what we have seen above there are some pkg_zip rules for various other game resources like background textures, configuration files, fonts, etc.

Running the game

With both the code and data building we now turn to actually running the game. For the server executable this is relatively straightforward. Its only data dependency is the map collision data.

However the graphics engine used by client requires all assets to be listed in a resources.cfg file. It also needs to dynamically load various plugins with dlopen. This is facilitated by a wrapper script which enumerates all the resources and sets the plugin folder appropriately. To preserve the locality of the logic, it is convenient to inline such scripts directly in BUILD.bazel using a write_file rule:

    name = "write-run-client-sh",
    out = "",
    content = [
        "sed -i \"s|^PluginFolder=.*|PluginFolder=$1|\" plugins.cfg",
        "scripts/ . >resources.cfg",

This causes Bazel to generate the script file as needed and avoids the hassle of creating and committing another file to version control. The script is invoked by a sh_binary rule:

    name = "run-client",
    srcs = [""],
    data = [
    args = [
        "$(rootpath @ogre//:lib/OGRE)",

Bazel ensures that all the data dependencies are available in the run-client.runfiles directory where they will be found by the script. The full path to the plugin directory is obtained by using rootpath to resolve the @ogre//:lib/OGRE label. This path is then passed as an argument to and then substituted into plugins.cfg.

Now the entire game can be built and run with just two commands:

bazel run //common:run-server
bazel run //common:run-client


It took around three days to complete the migration from start to finish. However the majority of that time was spent updating the code to work with C++17 and fixing various issues that arose from upgrading to newer versions of the third party libraries. The Bazel migration itself was fairly easy and wouldn’t have taken more than a day if the code had been otherwise up-to-date. Of course this was a pretty small project so your mileage may vary.

I was initially concerned that there might not be Nix packages for some of the more obscure dependencies but I was pleasantly surprised. Every single one of the libraries I had chosen for this project in 2009 can now be found in Nix. Having easy access to over 80,000 packages is reason enough to pair Bazel with Nix, but if reproducible builds matter to you it makes even more sense.

Correct and incremental builds are awesome. Bazel is very good at managing complex dependency graphs where some build outputs (e.g. tools) are themselves used as part of the build. When you change something Bazel will rebuild exactly what is necessary and it will do so in the correct order.

Clearly, not building more than necessary saves time, but so does not building less. When a build system neglects to rebuild something it should have, the result is confusion and wasted time because runtime behaviour does not match expectations. After being bitten by this a few times developers will tend to clean the build more frequently, incurring the cost of a full rebuild each time. So the confidence Bazel provides here is great for developer productivity.

  1. For example, elsewhere on the filesystem or from a remote Git repository.
  2. The term module is used here informally to refer to a grouping of related source files in a directory under //common/src. These are not C++20 modules.
  3. The srcs parameter should exactly match the shared library name in the dynamic section of the executable. You can use readelf -d EXECUTABLE to check this.

About the author

Ben Radford

A software engineer who is passionate about DevOps and automation.

If you enjoyed this article, you might be interested in joining the Tweag team.

This article is licensed under a Creative Commons Attribution 4.0 International license.


AboutOpen SourceCareersContact Us

Connect with us

© 2024 Modus Create, LLC

Privacy PolicySitemap