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

A taste of Bazel: build a library, a service and hspec tests

6 May 2020 — by Clément Hurlin

We described several times before how to interface Haskell with other languages. Such interfaces between languages are important because projects rarely use a single language: they are polyglot. So build systems should be polyglot too. It’s better to have one polyglot build system than many single purpose ones. The headline benefit is that a single build system can achieve better caching, better parallelism, and therefore faster builds. It’s also easier to avoid correctness issues.

Bazel, open sourced by Google in 2015, is one such polyglot build system. Using rules_haskell, Bazel supports Haskell. In this post, we’ll show how to get started with Bazel on a small but non-trivial project, featuring a library, a web service and an Hspec test suite.

The Bazel workspace

Bazel has two kinds of files, which both use Python syntax:

  • A unique WORKSPACE file, which identifies a Bazel workspace. The WORKSPACE file is the only place where additional code and data outside of the workspace can be pulled in.
  • BUILD.bazel files containing a specification of the build dependency graph. Bazel is designed for modularity and to scale to very large repositories, so you can break up the dependency graph spec in many BUILD.bazel files scattered across your repository.

First, create a WORKSPACE file at the root of your project using the start script. The created file looks like this:

workspace(name = "your_cool_project_name")

# Import the rule that can download tarballs using HTTP.

    name = "rules_haskell",
    strip_prefix = "rules_haskell-0.12",
    urls = [""],


# Setup rules_haskell.


# Download a GHC binary distribution from
# and register it as an available toolchain.
rules_haskell_toolchains(version = "8.6.5")

The WORKSPACE file is very explicit. There’s a good reason for this. To be fast, Bazel is lazy: Bazel caches and reuses the result of previous builds if possible and loads build files only if required. To do that reliably, Bazel tracks changes of both source files and build files. Tracking build files requires tightly controlling what they include, hence the use of load statements to make symbols available (http_archive is the first such symbol in the snippet above).

Using Stackage

To convert your Stack project to Bazel we will make use of the stack_snapshot rule.

Add what follows to the WORKSPACE file, adapting the list of package according to what your project requires:


    name = "stackage",
    packages = [
        # more packages
    snapshot = "lts-14.11",

The packages attribute lists the packages to pull from The snapshot attribute specifies the desired snapshot version, as in your stack.yaml.

Calling stack_snapshot with name = stackage in the WORKSPACE file extends the Bazel workspace to include third-party code downloaded from Hackage, in an external repository called @stackage. This repository includes a number of targets, like @stackage//:aeson or @stackage//:filepath, also defined by stack_snapshot, based on metadata downloaded from Stackage. Under the hood, Bazel uses Stack to process this metadata.

Compiling sources

Let’s say that your project has the following layout for its Haskell code:

└── haskell
    ├── exe
    ├── src
    └── test
  • library code lives in haskell/src,
  • the executable’s code lives in haskell/exe, and
  • test code lives in haskell/test.

Declaring how to build the library code is as simple as adding the following in haskell/src/BUILD.bazel:

load("@rules_haskell//haskell:defs.bzl", "haskell_library")

    name = "mylib",
    srcs = glob(["**/*.hs"]),  # match all .hs files
    deps = [
    # Make library code available to executable and tests
    visibility = [

Then, declare how the executable code is built with the following haskell/exe/BUILD.bazel:

load("@rules_haskell//haskell:defs.bzl", "haskell_binary")

    name = "server",
    srcs = ["Main.hs"],
    deps = [
        # Depend on the library defined above

And finally for the tests in haskell/test/BUILD.bazel:


    name = "tests",
    # Lists the source files containing tests,
    srcs = ["AppSpec.hs", "Spec.hs"],
    tools = ["@hspec-discover"],
    compiler_flags = [
        "-DHSPEC_DISCOVER=$(location @hspec-discover)"
    deps = [

This rule states that the executable hspec-discover is required, via the tools field. See this snippet for the full WORKSPACE file to see how this binary is obtained via a simple application of the haskell_cabal_binary rule. The rule also makes this executable available to the compiler, using the Make-like variable $(location ...) to find out the runtime path to the executable. This syntax might seem verbose. But rather than grabbing whatever hspec-discover happens to be in the $PATH, what we’re doing here is telling Bazel exactly which binary we want to use for the above target. When the project grows large, other parts of the build could use a different hspec-discover, potentially. If hspec-discover ever changes, Bazel knows exactly what needs to be rebuilt: the :tests target above, since hspec-discover is a dependency, but not :mylib or :server.


Even small projects require important features from a build system:

  • mechanisms to specify exactly what compiler toolchain we want to use, to make the build reproducible,
  • a way to resolve symbolic names to specific package versions on Hackage, using package snapshots,
  • the ability to build preprocessors (like hspec-discover) and tell build targets about their location,

What we have shown is that Bazel today supports doing all of the above. Alternatives like cabal-install and Stack support this too, and for small to medium sized projects, they work just fine and are simpler to handle than the Bazel workhorse. But I hope I’ve given you here a glimpse of what the Bazel way looks like: make all dependencies explicit including binary dependencies, design for cacheability, and infinite extensibility using custom build rules loaded from your own Python-syntax .bzl files like we do in all the build and workspace files above.

The sample files in this blog post have been extracted from a Bazel-based version of servant’s example-servant-minimal. Head there to see the full example.

You may also want to read the rules_haskell documentation, for more advanced use cases.

Then you can move on to build the rest of your polyglot project with Bazel, too. Bazel supports a large number of languages. And when the project gets big and the build times substantial, turn to shared caching among all developers.

About the author

Clément Hurlin

Clément is a Senior Software Engineer that straddles the management/engineer boundary. Clément studied Computer Science at Telecom Nancy and received his PhD from Université Nice Sophia Antipolis, where he proved multithreaded programs using linear logic.

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