Using AppVeyor for Haskell+Windows CI

UPDATE August 2018 As pointed out by Mark Wotton, the instructions below were out of date (see Sed: A Debugging Story for more information). I've updated the contents below, but for future readers: please see the more up-to-date AppVeyor configuration on Stack's Github repo.

I don't think I ever documented this before, so just a quick post to get this out there. Many of us working on open source Haskell libraries already use Travis CI for doing continuous integration builds of our software. Some time ago they added support for OS X, making it possible to cover Linux and OS X with multiple configurations on their systems. For any project with a stack.yaml file, this can be easily achieved using the Stack recommended Travis configuration.

Unfortunately, this leaves Windows testing out, which is unfortunate, because Windows is likely to be the most common build to fail. Fortunately, AppVeyor provides a similar experience to Travis, but for Windows. In order to get set up, just:

1. Sign in with their web interface and add your Github repo
2. Add an appveyor.yml to your project

Here's a simple file I've used on a few projects with succeess:

build: off

before_test:
# http://help.appveyor.com/discussions/problems/6312-curl-command-not-found
- set PATH=C:\Program Files\Git\mingw64\bin;%PATH%

- curl -sS -ostack.zip -L --insecure http://www.stackage.org/stack/windows-i386
- 7z x stack.zip stack.exe

clone_folder: "c:\\stack"
environment:
  global:
    STACK_ROOT: "c:\\sr"

    # Override the temp directory to avoid sed escaping issues
    # See https://github.com/haskell/cabal/issues/5386
    TMP: "c:\\tmp"

test_script:
- stack setup > nul
# The ugly echo "" hack is to avoid complaints about 0 being an invalid file
# descriptor
- echo "" | stack --no-terminal test

All this does is:

• Downloads the Stack zip file
• Unpacks the stack.exe executable
• Changes the STACK_ROOT to deal with Windows long path issues
• Run stack setup to get a toolchain
• Run stack --no-terminal test to build your package and run the test suites

You're free to modify this in any way you want, e.g., add in --bench to build benchmarks, add --pedantic to fail on warnings, etc. If you have more system library dependencies, you'll need to consult the AppVeyor docs to see how to install them. And in our use cases for Stack, we found that using the AppVeyor caching functionality made builds unreliable (due to the large size of the cache). You may want to experiment with turning it back on, since this setup is slow (it downloads and installs a full GHC toolchain and builds all library dependencies each time).