GHC, Alpine, Stack, and Docker

Table of Contents

• Prerequisites
• Building the Docker Images
  • "Quick" Start
  • Overview
• Developing Locally with stack
• TODO

This repository is a small demonstration of the steps required to build an Alpine Linux Docker image with all the tools necessary to compile Haskell programs linked against musl libc.

Such an environment are extremely useful for producing small Docker images for the deployment of services and for creating portable Haskell executables (either statically linked, or bundled with the musl linker and their dynamic dependencies).

At the time of writing, this repository is primarily geared towards showing off a workflow by which one can use stack's Docker integration to seamlessly develop inside an Alpine Linux container. This gives the benefit of stack's build caching and also integrates nicely with development tools like ghcid.

In the future I hope to either update this repository with explicit instructions and/or examples that show how one can create minimal (i.e. tens of MB) Docker images to deploy Haskell applications in production.

Prerequisites

Ensure Docker is installed on your computer and the Docker daemon is running.

Install the Haskell Tool Stack.

Building the Docker Images

At the time of writing (1 Apr. 2019) GHC-HQ doesn't provide a version of GHC that is compatible with musl libc, and the ghc package provided by Alpine is out of date.

This means that the first step in this process is going to be compiling GHC itself and packaging up all the necessary tools and dependencies in an Alpine Linux container.

For convenience, I've combined all of these steps into a single, multi-stage Dockerfile in this repository, and written a Makefile that caches the intermediate build layers so changes to the final environment don't trigger a full rebuild of GHC and the associated tooling.

"Quick" Start

To build the Docker images, navigate to the project root directory and run:

make docker-build-gmp

Keep in mind that this stage compiles GHC, which can take anywhere from 30 mins to upwards of an hour depending on how fast your computer is.

Overview

I've tried to keep the Dockerfile relatively well commented, but the build process can be roughly understood as an impl

• base

  • Base layer used for all the intermediate build images to follow
  • Uses a patched version of the ghcup build script
    • The patch is used to override ghcup's' configuration stage to use the gold linker, which Alpine requires
  • Contains all OS dependencies required to run ghcup and stack

• build-ghc

  • Builds GHC via ghcup
  • Contains all dependencies required to build GHC
  • Contains some (currently unused) logic for selecting between integer-gmp and integer-simple
    • integer-gmp depends on libgmp, which is licensed under the LGPL, so if one wants to distribute a statically linked, closed source binary they will need to build against a version of GHC that uses integer-simple

• build-tooling

  • Downloads stack and verifies the hash

• alpine-haskell

  • Assembles artifacts from the previous layers
  • Copies GHC and stack
  • Installs bash and shadow
    • stack requires usermod/groupmod from shadow as well as bash to run

Developing Locally with stack

Once the images have finished building, this project can be compiled to demonstrate what local development looks like with stack.

For convenience, I've added some targets to the Makefile that run some common development tasks.

First, ensure that ghcid is installed (for example, by running stack install ghcid).

Then, compile this project in Docker by running make build-fast.

Finally, start a ghcid development loop with make ghcid; this loop uses stack --docker to spawn a container in which GHC recompiles everything whenever it detects changes in your code.

Try changing something in executables/Main.hs and see the changes reflected in ghcid!

TODO

• Demonstrate integer-simple support
• Demonstrate static linking
• Create separate Dockerfiles for GHC and combined tooling
  • Automate the creation of these images in CI and host them on Dockerhub