Upcoming release tasks:

  • Check whether persistent still needs monad-logger; remove dependency if not
  • Check if workaround for still needed in stack.yaml
  • Eventually remove the Ubuntu, Debian, CentOS, Arch packages from our S3 bucket. This was announced with the 1.9.x release, so can do this around time of 1.11.x. Directories, and last Stack version uploaded:
    • s3:// (1.0.0)
    • s3:// (1.5.1)
    • s3:// (1.4.0)
    • s3:// (1.3.0)
    • s3:// (1.5.1)
  • Eventually remove -nopie variants from stack-setup-2.yaml. stack-1.6 was the last release to use them, so wait a few major releases after that, and be sure to announce.
    • Also remove the -nopie variants from etc/scripts/
  • Remove workaround to #4125 from stack.yaml for next major version after 1.8.
  • Explicitly tag release candidates and prereleases in stack --version output.
  • Look through for hints on how to make this document more clear.
  • If store is no longer a dependency, likely can remove from stackage build constraints' expected-test-failures

Iterating on release process

IMPORTANT: all bindists for a given release should be built from a consistent git commit (this can be relaxed a bit for release candidates, but still should be maintained as much as possible).

Since the release process and scripts sometimes need to be iterated on during the process of building all the platforms' binaries, the scripts are all designed so that you can run them from a different directory. This means you can have one source tree for the version of stack being released, and a separate one where you work on the release scripts and process.

To use this way, set the current directory should be the version of stack to be released. You can then call the scripts in another directory to use their version. For example, stack ../stack-release-scripts/etc/scripts/release.hs … or ../stack-release-scripts/etc/scripts/vagrant-releases.hs.

Version scheme

  • Versions with an even second component are development versions (the master branch)
  • Versions with an odd second component are stable versions (the stable branch, or in a vX.Y release candidate branch for not-yet-released versions)
  • Versions with an even third component (e.g. 1.6.2 and 1.7.0) are unreleased versions
  • Versions with an odd third component (e.g. 1.6.1 or 1.7.3) and released versions
  • Pre-release unstable binaries will be released with the date as the fourth component (e.g.
  • Release candidate binaries will be released with an even third component and and odd number as the fourth component (e.g.
  • All branches except release (which matches exactly the most recent release) must have an even third component (development)
  • Branches other than stable, release, and a vX.Y release candidate will always have a 0 third component (e.g. 1.7.0).


  • v1.7.x series pre-release branch (v1.7 branch)
  • release candidate for first release of v1.7.x series (v1.7 branch)
  • continuing development on pre-release branch
  • second release candidate for first release of v1.7.x series (v1.7 branch)
  • 1.7.1: first release of the 1.7.x series (release branch)
  • development for second release of 1.7.x series (stable branch)
  • release candidate for second release of 1.7.x series (stable branch)
  • 1.7.3: second release of 1.7.x series (release branch)
  • 1.8.0: unstable development code (master branch)
  • pre-release snapshot of unstable version (master branch)

Pre-release checks

  • Check that the snapshot in stack.yaml's' GHC version supports building on all required platforms (e.g. GHC 8.4 doesn't support FreeBSD, so it's not a candidate). If not, will have to switch to a different snapshot (or decide that we won't support the platform for this iteration).
  • Check for any P0 and P1 issues that should be dealt with before release
  • Check for un-merged pull requests that should be merged before release
  • Ensure release and stable branches merged to master
  • Ensure no bounds specified in package.yaml unless truly necessary (e.g. in most cases rely on stack.yaml's resolver to manage dependency versions), and ensure base minbound is correct for the supported GHC version(s) (e.g. use stack exec ghc-pkg list base to find out minimum base version)
  • Check compatibility with latest LTS Stackage snapshots
    • stack*.yaml (where * is not nightly), including the ones in subdirectories: bump to use latest LTS minor version (be sure any extra-deps that exist only for custom flags have versions matching the snapshot)
    • Check for any redundant extra-deps
    • Run stack --stack-yaml=stack*.yaml test --pedantic (replace * with the actual file)
  • Check compatibility with latest nightly stackage snapshot:
    • Update stack-nightly.yaml with latest nightly and remove unnecessary extra-deps (be sure any extra-deps that exist only for custom flags have versions matching the snapshot)
    • Run stack --stack-yaml=stack-nightly.yaml test --pedantic
  • Update .travis.yml for any GHC version changes made in above steps.
  • Ensure integration tests pass on a Windows, macOS, and Linux (Linux integration tests are run by Gitlab): stack install --pedantic && stack test --pedantic --flag stack:integration-tests. The actual release script will perform a more thorough test for every platform/variant prior to uploading, so this is just a pre-check

Release preparation

  • In master branch:

    • package.yaml: bump to next release candidate version (bump second component to next odd number, ensure third component is 0, and add patchlevel 1; e.g. from 1.8.0 to
      • Rename the "Unreleased changes" section to the same version as package.yaml, and mark it clearly as a release candidate (e.g. v1.9.0.1 (release candidate)). Remove any empty sections.
      • Check for any entries that snuck into the previous version's changes due to merges (git diff origin/stable HEAD
  • Cut a release candidate branch vX.Y from master

  • In master branch:

    • package.yaml: bump version to next unstable version (bump to next even second component with .0 third component (e.g. from 1.9.0 to 1.10.0)
    • Changelog: add new "Unreleased changes" section: ``` ## Unreleased changes

    Release notes:

    Major changes:

    Behavior changes:

    Other enhancements:

    Bug fixes: ```

  • In RC branch:

    • Review documentation for any changes that need to be made
      • Ensure all documentation pages listed in mkdocs.yaml (git diff --stat origin/stable..HEAD doc/)
      • Any new documentation pages should have the "may not be correct for the released version of Stack" warning at the top.
      • Search for old Stack version, unstable stack version, and the next "obvious" possible versions in sequence, and UNRELEASED and replace with next release version (X.Y.1, where Y is odd). Note: do not update the Dockerfiles in etc/dockerfiles/stack-build yet; that will come later)
        • Do NOT update templates in .github to point at the new release version yet!
      • Look for any links to "latest" documentation, replace with version tag
    • Update STACK_VERSION in etc/scripts/ to the new release version (X.Y.1)
    • Check if GHC version we're using has bindists using upgraded versions of operating systems (e.g. FreeBSD, Debian) and upgrade relevant Vagrantfiles in etc/vagrant and release shell script (etc/scripts/* to match.
    • Check that for any platform entries that need to be added to (or removed from) releases.yaml,,, and doc/
  • Follow steps in Release process below tagged with [RC] to make a release candidate

  • For subsequent release candidates:

    • Re-do the pre-release checks (above section)
    • package.yaml: bump to next odd patchlevel version (e.g. X.Y.0.3)
    • Rename the "Unreleased changes" section to the new version, clearly marked as a release candidate (e.g. vX.Y.0.3 (release candidate)). Remove any empty sections.
    • Follow steps in Release process below tagged with [RC] to make a release candidate
  • For final release:

    • package.yaml: bump version to odd last component and no patchlevel (e.g. from X.Y.0.2 to X.Y.1).
    • consolidate all the RC changes into a single section for the release version
    • Follow all steps in the Release process section below.

Release process

See stack-release-script's README for requirements to perform the release, and more details about the tool.

  • Create a new draft Github release with tag and name vX.Y.Z (where X.Y.Z matches the version in package.yaml from the previous step), targeting the RC branch. In the case of a release candidate, add (RELEASE CANDIDATE) to the name field. check the This is a pre-release checkbox. [RC]

  • On each machine you'll be releasing from, set environment variables GITHUB_AUTHORIZATION_TOKEN and STACK_RELEASE_GPG_KEY (see stack-release-script's README). [RC]

  • [TODO (for below steps): All the etc/scripts/* should be integrated into etc/scripts/release.hs]

  • On a machine with Vagrant installed: [RC]

    • Run etc/scripts/
  • On macOS: [RC]

    • Run etc/scripts/
  • On Windows: [RC]

    • Use a short path for your working tree (e.g. C:\p\stack-release
    • Ensure that STACK_ROOT, TEMP, and TMP are set to short paths
    • Run etc\scripts\windows-releases.bat (for release candidates, only 64-bit is necessary so feel free to comment out 32-bit)
    • Release Windows installers. See stack-installer README. For release candidates, the windows installers can be skipped. [TODO: this should be integrated into etc/scripts/release.hs]
  • On Linux ARMv7: [RC]

    • Run etc/scripts/
  • On Linux ARM64 (aarch64): [RC]

    • Run etc/scripts/
  • Build a Linux static bindist [RC]

    • Follow directions in the Build Linux static binary distribution with Nix section below.
  • Build sdist using stack sdist ., and upload it to the Github release with a name like stack-X.Y.Z-sdist-0.tar.gz. Also upload GPG signature and checksums. [RC]

TODO: did this last time by copying to _release and then using release.hs to upload sigs and checksum -- should add all this logic to release.hs itself. e.g.:

mv /home/vagrant/stack-release/.stack-work/dist/x86_64-linux/Cabal- _release/stack- stack ../stack/etc/scripts/release.hs _release/stack- _release/stack- _release/stack-

  • Use etc/scripts/ to generate a Cabal spec and sdist with dependency bounds. [RC] [TODO: should add this logic to release.hs itself]

  • Upload _release/stack-X.Y.Z-sdist-1.tar.gz to the Github release similarly to the sdist-0 above. [RC]

  • For any GPG key used to sign an uploaded bindist, ensure that signs their key and uploads to keyserver:

gpg --sign-key -u 0x575159689BEFB442 <OTHER-KEY-ID> gpg --send-keys <OTHER-KEY-ID>

  • Publish Github release. Include the changelog and in the description and use e.g. git shortlog -s origin/release..HEAD|sed $'s/^[0-9 \t]*/* /'|LC_ALL=C sort -f to get the list of contributors (contributors not necessary for release candidates). See previous releases for example formatting and extra info (such as link to website for install instructions). [RC]

  • Push signed Git tag, matching Github release tag name, e.g.: git tag -d vX.Y.Z; git tag -s -m vX.Y.Z vX.Y.Z && git push -f origin vX.Y.Z. [RC]

  • Upload package to Hackage: stack upload . (note that this uploads revision 0 without bounds, which is best for stackage since we have told it to ignore revisions for stack)

  • Edit Cabal file metadata on Hackage and paste in the contents of _release/stack-X.Y.Z_bounds.cabal to add bounds (this helps non-stackage users build stack). You will have to add the x-revision: 1 field.

  • Reset the release branch to the released commit, e.g.: git checkout release && git merge --ff-only vX.Y.Z && git push origin release

  • Update the stable branch similarly

  • In the stable or, in the case of a release candidate, vX.Y branch: [RC]

    • package.yaml: bump the version number even third component (e.g. from 1.6.1 to 1.6.2) or, in the case of a release candidate even fourth component (e.g. from to
    • Add an "Unreleased changes" section:


      Unreleased changes

      Release notes:

      Major changes:

      Behavior changes:

      Other enhancements:

      Bug fixes: ```

  • Activate version for new release tag (or, in the case of release candidates, the vX.Y branch), on, and ensure that stable documentation has updated. [RC]

  • Deactivate version for release candidate on

  • Update /stable rewrite rules (be sure to change both places) to new released version, and update production cluster.

  • Delete the RC branch (locally and on origin). E.g. git branch -d vX.Y; git push origin :vX.Y.

  • Update fpco/stack-build Docker images with new version

    • Add etc/dockerfiles/stack-build/lts-X.Y/Dockerfile (where X.Y is the latest stackage LTS version), containing (note where X.Z is the previous LTS version, and X.Y.Z is the newly released stack version)

    FROM fpco/stack-build:lts-X.Z ARG STACK_VERSION=X.Y.Z RUN wget -qO-$STACK_VERSION/stack-$STACK_ VERSION-linux-x86_64.tar.gz | tar xz --wildcards --strip-components=1 -C /usr/local/bin '*/stack'

    • Run the appropriate job for the LTS major version in Gitlab pipelines.

    • Check that the newly build Docker image has the new Stack version

  • Merge any changes made in the RC/release/stable branches to master (be careful about version and changelog). [RC]

  • master branch: update templates in .github to point at the new release version (X.Y.1).

  • Announce to,, mailing lists, subject ANN: stack-X.Y.Z (or ANN: stack-X.Y release candidate), containing the markdown for the release description from Github. [RC]

  • Add back to Stackage nightly if fallen out (be sure to have a < 9.9.9 constraint to avoid the accidentally uploaded stack-9.9.9 from being used).

Build Linux static binary distribution with Nix

TODO: script this process and/or integrate with etc/scripts/release.hs

TODO: run integration tests against static binary

These instructions are tested on Ubuntu 16.04, but theoretically should work on any Linux distribution.

  • Install nix (tested with v2.0.4 and v2.1.2, but should work with any)

curl | sh

  • Install and authenticate cachix (first two steps at after signing up)

  • Add nh2's cache:

cachix use static-haskell-nix

NOTE: to clear cache index, use rm $HOME/.cache/nix/binary-cache-v5.sqlite* (useful if someone else uploads new stuff to the cache and you want to use it right away). The recent narinfo-cache-positive/negative-ttl options might also help.

  • Check out stack commit to be released to ~/stack-release (or elsewhere, in which case adjust following instructions)

  • rm -f ~/stack-release/stack.cabal, to ensure it's regenerated

  • clone (commit 6bd86c02ed11cc34f209430e23761cd14461c067)

  • in static-stack directory, run (from

$(nix-build --no-out-link -A stack2nix-script) ~/stack-release nix-build --no-out-link default.nix -A static_stack --arg release true

  • Copy the binary built above (in /nix/store/XXX-stack-X.Y.Z/bin/stack) to ~/.stack-release/_release/bin/stack-X.Y.Z-linux-x86_64-static/stack (replace X.Y.Z with the version, and the /nix/store/* path with that output at the end of the previous command)

  • Package, sign, and upload to Github using stack's release script in the stack directory:

cd ~/stack-release stack etc/scripts/release.hs --no-test-haddocks --binary-variant=static --build-args=--dry-run upload

(adding --build-args=--dry-run ensures the binary you copied will be used rather than building a new one)

  • Download the bindist from github and double check that the stack in it is actually static (use ldd /path/to/stack) and that --version reports correctly (and not dirty).

Setting up a Windows VM for releases

These instructions are a bit rough, but has the steps to get the Windows machine set up.

Using Virtualbox

  1. Download Virtualbox VM image:

  2. Launch the VM using Virtualbox and the image downloaded

  3. Adjust settings:

    • Number of CPUs: at least half the host's
    • Memory: at least 3 GB
    • Video RAM: the minimum recommended by Virtualbox
    • Enable 3D and 2D accelerated mode (this makes programs with lots of console output much faster)
    • Enabled shared clipboard (in VM window, Devices->Shared Clipboard->Both Directions)

Now continue to the General Windows setup subsection below.

Using ESXi

  1. Download the MSEdge on Win10 VM for VMWare (Windows, Mac).
  2. Unzip the file downloaded file
  3. Upload the VMDK file to the ESXi datastore
  4. SSH into ESXi CLI and run:
    • vmkfstools -i /vmfs/volumes/datastore1/win10-msedge/MSEdge-Win10-VMWare-disk1-ORIG.vmdk /vmfs/volumes/datastore1/win10-msedge/MSEdge-Win10-VMWare-disk1.vmdk -d thin. This converts the disk to a format that is compatible with ESXi. You may have to run esxcli system module load -m multiextent first (see
    • vmkfstools -X 80G /vmfs/volumes/datastore1/win10-msedge/MSEdge-Win10-VMWare-disk1.vmdk. This makes the disk twice as large, which helps avoid running out of disk space.
  5. In the ESXi web UI:
    • Create a new VM
      • Give is 8192 MB of memory
      • Give it 4 virtual CPUs
      • Remove the default hard disk
      • Add an Existing hard disk
        • Select /datastore1/win10-msedge/MSEdge-Win10-VMWare-disk1.vmdk
    • Power on the VM
    • In Windows settings:
      • Search for "disk management"
        • Extend the partition to take the whole disk.
      • In all likelyhood, you will want to search for "remote desktop" and enable remote desktop. Then you can connect to the VM using Microsoft Remote Desktop instead of using it from within the ESXi web UI.

Now continue to the General Windows setup subsection below.

General Windows setup

  1. In Settings->Update & Security->Windows Update->Advanced options:

    • Change Choose how updates are installed to Notify to schedule restart
    • Check Defer upgrades (this avoids rebooting in the middle of the stack build)
  2. In Settings->System->Power & sleep

    • Disable turning off the screen or going to sleep when plugged in
  3. Install msysgit:

  4. Install TortoiseHG:

  5. Install nsis-2.46.5-Unicode-setup.exe from

  6. Install Stack using the Windows 64-bit installer

    a. Restart any command prompts to ensure they get new %STACK_ROOT% value.

  7. Visit in Edge to ensure system has correct CA certificates

  8. Run in command prompt:

    md C:\p
    md C:\tmp
    cd /d C:\p
  9. Create C:\p\env.bat:

    SET TEMP=C:\tmp
    SET TMP=C:\tmp
    SET PATH=C:\Users\IEUser\AppData\Roaming\local\bin;"c:\Program Files\Git\usr\bin";"C:\Program Files\Microsoft SDKs\Windows\v7.1\Bin";%PATH%
  10. Run C:\p\env.bat (do this every time you open a new command prompt)

  11. stack exec -- gpg --import, and paste in the your GPG secret key (must be done using stack exec because that uses the right keyring for the embedded msys GPG; you can get the key from another machine with gpg --export-secret-keys --armor <KEY ID>)

  12. Run in command prompt (adjust the and settings):

    git config --global
    git config --global "Emanuel Borsboom"
    git config --global push.default simple
    git config --global core.autocrlf true
    git clone
    git clone -b stable --reference C:\p\stack-release stack-release
    cd stack-release
    stack install cabal-install

Setting up an ARM VM for releases

  1. Use Scaleway to start ARMv7 and ARM64 VMs.

  2. Select Ubuntu Xenial as the operating system

  3. Install the correct version of LLVM: sudo apt-get install -y llvm-3.9 (appropriate for GHC 8.2, might need different version for other GHCs)

  4. Symlink opt-3.X to opt: sudo ln -s opt-3.9 /usr/bin/opt (adjust the version if you installed a different one above)

  5. Switch to gold linker:

    update-alternatives --install "/usr/bin/ld" "ld" "/usr/bin/" 20 update-alternatives --install "/usr/bin/ld" "ld" "/usr/bin/ld.bfd" 10 update-alternatives --config ld

  6. Add swap space:

    dd if=/dev/zero of=/swapfile1 bs=1024 count=4194304 mkswap /swapfile1 swapon /swapfile1 echo '/swapfile1 none swap sw 0 0' >>/etc/fstab

  7. Install additional tools:

    apt-get update && apt-get install -y unzip gpg

  8. Import your GPG key (gpg --import and paste the private key)

  9. Git settings (adjust for your preferences/email/name)

    git config --global push.default simple git config --global "" git config --global "Emanuel Borsboom"

  10. Install build tools and dependencies packages

    sudo apt-get install -y g++ gcc libc6-dev libffi-dev libgmp-dev make xz-utils zlib1g-dev git gnupg

  11. Install clang+llvm

    NOTE: the Debian jessie llvm package does not work (executables built with it just exit with "schedule: re-entered unsafely.").

    The version of LLVM needed depends on the version of GHC you need.

    • GHC 8.0.2 (the standard for building Stack)

    wget && \ sudo tar xvf clang+llvm-3.7.1-armv7a-linux-gnueabihf.tar.xz -C /opt

    Run this now and add it to the .profile:

    export PATH="$HOME/.local/bin:/opt/clang+llvm-3.7.1-armv7a-linux-gnueabihf/bin:$PATH"

    • GHC 7.10.3

    wget && \ sudo tar xvf clang+llvm-3.5.2-armv7a-linux-gnueabihf.tar.xz -C /opt

    Run this now and add it to the .profile:

    export PATH="$HOME/.local/bin:/opt/clang+llvm-3.5.2-armv7a-linux-gnueabihf/bin:$PATH"

  12. Install Stack

    Binary: get an existing stack binary and put it in ~/.local/bin.

    From source (using cabal-install):

    wget && \ tar xvf ghc-7.10.3-armv7-deb8-linux.tar.xz && \ cd ghc-7.10.3 && \ ./configure --prefix=/opt/ghc-7.10.3 && \ sudo make install && \ cd .. export PATH="/opt/ghc-7.10.3/bin:$PATH" wget &&&&& \ tar xvf cabal-install- && \ cd cabal-install- && \ EXTRA_CONFIGURE_OPTS="" ./ && \ cd .. && \ export PATH="$HOME/.cabal/bin:$PATH" && \ cabal update

    Edit ~/.cabal/config, and set executable-stripping: False and library-stripping: False.

    cabal unpack stack && \ cd stack-* && \ cabal install && \ mv ~/.cabal/bin/stack ~/.local/bin