For the past few release of GHC (I think 7.10.2 and 7.10.3), I've tried to help out with the testing effort by using Stackage Nightly as a "stress test." Stackage Nightly provides a large (1810 at last count) collection of real-world Haskell packages to test compatibility. For minor version upgrades of GHC, this stress test is especially good, since virtually no packages should stop working with a minor version bump (besides things like conflicting identifier or module names). The test does end up providing quite a few false positives for a new major GHC release, but it still quite informative.
Until now, each time I ran the test was a manual procedure. It didn't take that much time to set up, but (1) did take some non-zero part of my time, and (2) made it impossible for others to run this test (or even automate it) without my help. I got this set up as the ghc-rc-stackage project on Github, and I wanted to put up a post explaining how to use it in case GHC developers or testers want to play around with this.
From a Linux machine with Docker installed, run the following from any directory:
docker pull fpco/ghc-rc-stackage docker run --rm -it \ -v `pwd`/build:/build -v `pwd`/fake-home:/fake-home \ -e USERID=`id -u` \ fpco/ghc-rc-stackage \ /stackage/build.sh
This will populate a
build subdirectory with various content, include binary
artifacts, generated documentation, and most important, build logs. Those will
be present in build/logs/nightly. Pay attention to the console output to see
which packages failed so you know which logs to pay attention to.
This script will try to build the most recent Stackage Nightly snapshot with the current GHC release candidate selected for the repo (at time of writing, this is GHC 8.0.1 rc2).
In order to perform these builds, we have a Docker image that contains:
stackage-curatorbinaries necessary for running a Stackage build
/stackage/build.sh) that downloads the YAML configuration for the most recent Stackage Nightly and calls the
stackage-curatorexecutable with appropriate command line arguments to kick off the build.
docker run line above does some magic to bind-mount appropriate
directories and set the USERID environment variable used by that script, which
will in turn be used to set appropriate permissions.
Making local modifications to this setup is easy, just:
docker build --tag fpco/ghc-rc-stackage .
docker runcommand from above
These steps are covered in the project's README.md
This is intended to provide basic functionality. The only planned enhancements I have for this right now are bumping the links for future release candidates in the future. As usual, if you have ideas for improvement, pull requests are very much welcome!