So that this doesn't become repetitive: for the reasons behind the answers below, see the Build overview page. The goal of the answers here is to be as helpful and concise as possible.
The version of GHC, as well as which packages can be installed, are specified by
the snapshot. This may be something like lts-22.21, which is from
Stackage. The user's guide discusses
the snapshot in more detail.
The snapshot is determined by finding the relevant project-level configuration
file (stack.yaml, by default) for the directory you're running the command
from. This essentially works by:
- Check for a
STACK_YAMLenvironment variable or the--stack-yamlcommand line argument - If none present, check for a
stack.yamlfile in the current directory or any parents - If no
stack.yamlfile was found, use the implicit global
The implicit global is a shared project used whenever you're outside of another project. It's a sort of "mutable shared state" that you should be aware of when working with Stack.
A frequent request when working with the implicit global is how to move to a more recent LTS snapshot. You can do this using the following command from outside of a project:
stack config set snapshot lts
Stack itself is installed in normal system locations based on the mechanism you
used (see the Install and upgrade page). Stack
installs files in the Stack root and other files in a .stack-work directory
within each project's directory. None of this should affect any existing Haskell
tools at all.
- 'Cabal' can refer to Cabal (the library) or to Cabal (the tool). Cabal (the
library) is used by Stack to build your Haskell code. Cabal (the tool) is
provided by the
cabal-installpackage. - A Cabal file is provided for each package, named
<package_name>.cabal. It defines all package-level metadata, just like it does in the world of Cabal (the tool): modules, executables, test suites, etc. No change at all on this front. - A
stack.yamlfile references one or more packages, and provides information on where dependencies come from. - The
stack initcommand initializes astack.yamlfile from an existing Cabal file. - Stack uses Cabal (the library) via an executable. For
build-type: Simple(the most common case), Stack builds that executable using the version of Cabal which came with the compiler. Stack caches such executables, in the Stack root under directorysetup-exe-cache. - In rare or complex cases, a different version of Cabal to the one that came
with the compiler may be needed.
build-type: Customand asetup-customstanza in the Cabal file, and aSetup.hsfile in the package directory, can be specified. Thestack.yamlfile can then specify the version of Cabal that Stack will use to build the executable (namedsetup) fromSetup.hs. Stack will use Cabal viasetup.
For detail on the differences between a stack.yaml file and a Cabal file, see
stack.yaml vs a Cabal file.
I need to use a different version of a package than what is provided by the LTS Haskell snapshot I'm using, what should I do?
You can make tweaks to a snapshot by modifying the extra-deps configuration
value in your stack.yaml file, e.g.:
snapshot: lts-22.21
packages:
- .
extra-deps:
- text-2.1.1@rev:0I need to use a package (or version of a package) that is not available on Hackage, what should I do?
Add it to the
extra-deps list in your project's
stack.yaml file, specifying the package's source code location relative to the
directory where your stack.yaml file lives, e.g.
snapshot: lts-22.21
packages:
- .
extra-deps:
- third-party/proprietary-dep
- github-version-of/conduit
- patched/diagramsThe above example specifies that the proprietary-dep package is found in the
project's third-party directory, that the conduit package is found in the
project's github-version-of directory, and that the diagrams package is
found in the project's patched directory. This autodetects changes and
reinstalls the package.
To install packages directly from a Git repository, use e.g.:
extra-deps:
- git: https://github.com/githubuser/reponame.git
commit: somecommitIDThose are the targets of the build, and can have one of three formats:
- A package name (e.g.,
my-package) will mean that themy-packagepackage must be built - A package identifier (e.g.,
my-package-1.2.3), which includes a specific version. This is useful for passing tostack installfor getting a specific version from upstream - A directory (e.g.,
./my-package) for including a local directory's package, including any packages in subdirectories
Typically, you will want to get the source for the package and then add it to
your packages list in the stack.yaml file. (See the previous question.)
stack unpack is one approach for getting the source. Another would be to add
the upstream package as a submodule to your project.
Explicit sandboxing on the part of the user is not required by Stack. All builds are automatically isolated into separate package databases without any user interaction. This ensures that you won't accidentally corrupt your installed packages with actions taken in other projects.
With a recent enough version of Cabal (the tool) (1.22 or later), you can. For earlier versions this does not work, due to Cabal issue #1800. Note that even with recent versions, for some commands you may need the following extra level of indirection. Command:
stack exec -- cabal exec -- cabal <command>
However, virtually all cabal commands have an equivalent in Stack, so this
should not be necessary. In particular, users of Cabal (the tool) may be
accustomed to the cabal run command. With Stack, command:
stack build
stack exec <program-name>
Or, if you want to install the binaries in a shared location, command:
stack install <program-name>
assuming your PATH has been set appropriately.
If you have a custom preprocessor, for example, Ruby, you may have a file like:
B.erb
module B where
<% (1..5).each do |i| %>
test<%= i %> :: Int
test<%= i %> = <%= i %>
<% end %>To ensure that Stack picks up changes to this file for rebuilds, add the
following lines to your stack.yaml file:
custom-preprocessor-extensions:
- erb
require-stack-version: ">= 2.6.0"And for backwards compatability with older versions of Stack, also add the following line to your Cabal file:
extra-source-files: B.erb
You could also use the
--custom-preprocessor-extensions
flag.
Yes. In its default configuration, Stack will simply ignore any system GHC
installation and use a sandboxed GHC that it has installed itself. You can find
these sandboxed GHC installations in the ghc-* directories in the
stack path --programs directory.
If you would like Stack to use your system GHC installation, use the
--system-ghc flag or run
stack config set system-ghc --global true to make Stack check your PATH for a
suitable GHC by default.
Stack can only use a system GHC installation if its version is compatible with
the configuration of the current project, particularly the snapshot specified by
the snapshot or
resolver key.
GHC installation doesn't work for all operating systems, so in some cases you
will need to use system-ghc and install GHC yourself.
In its default configuration, Stack determines from the current project which
GHC version, architecture etc it needs. It then looks in the ghc-<version>
subdirectory of the stack path --programs directory for a compatible GHC,
requesting to install one via stack setup if none is found.
If you are using the --system-ghc flag or
have configured system-ghc: true either in the project stack.yaml or the
global config.yaml, Stack will use the first GHC that it finds on your PATH,
falling back on its sandboxed installations only if the found GHC doesn't comply
with the various requirements (version, architecture) that your project needs.
See issue #420 for a
detailed discussion of Stack's behavior when system-ghc is enabled.
Stack will automatically install build tools required by your packages or their dependencies, in particular Alex and Happy.
!!! note
This works when using LTS or nightly snapshots, not with GHC or custom
snapshots. You can manually install build tools by running, e.g.,
`stack build alex happy`.
It checks the two most recent LTS Haskell major versions and the most recent Stackage Nightly for a snapshot that is compatible with all of the version bounds in your Cabal file, favoring the most recent LTS. For more information, see the snapshot auto-detection section in the architecture document.
I'd like to use my installed packages in a different directory. How do I tell Stack where to find my packages?
Set the STACK_YAML environment variable to point to the stack.yaml
configuration file for your project. Then you can run stack exec, stack ghc,
etc., from any directory and still use your packages.
Like all other targets, stack test runs test suites in parallel by default.
This can cause problems with test suites that depend on global resources such
as a database or binding to a fixed port number. A quick hack is to force stack
to run all test suites in sequence, using stack test --jobs=1. For test
suites to run in parallel developers should ensure that their test suites do
not depend on global resources (e.g. by asking the operating system for a random
port to bind to) and where unavoidable, add a lock in order to serialize access
to shared resources.
Yes, see the shell-autocompletion documentation.
Users of Cabal (the tool) are used to running cabal update regularly. You can
do the same with Stack by running stack update. But generally, it's not
necessary: if the package index is missing, or if a snapshot refers to
package/version that isn't available, Stack will automatically update and then
try again. If you run into a situation where Stack doesn't automatically do the
update for you, please report it as a bug.
No, Stack is very explicit about which packages it's going to build for you.
There are three sources of information to tell it which packages to install:
the selected snapshot, the extra-deps configuration value, and your local
packages. The only way to get Stack to change its build plan is to modify one
of those three. Updating the index will have no impact on Stack's behavior.
You can configure this in your project-level configuration file (stack.yaml,
by default). See YAML configuration.
You can create multiple YAML configuration files for your project, one for each build plan. For example, you might set up your project directory like so:
myproject/
stack-ghc-9.0.2.yaml
stack-ghc-9.2.4.yaml
stack.yaml --> symlink to stack-ghc-9.2.4.yaml
myproject.cabal
src/
...
When you run stack build, you can set the STACK_YAML environment variable to
indicate which build plan to use. On Unix-like operating systems command:
stack build # builds using the default stack.yaml
STACK_YAML=stack-ghc-7.10.yaml
stack build # builds using the given yaml fileOn Windows (with PowerShell) command:
$Env:STACK_YAML='stack-ghc-9.0.2.yaml'
stack buildYes, Stack supports using Docker with images that contain preinstalled Stackage packages and the tools. See Docker integration for details.
The way that Stack itself builds statically-linked Stack executables for Linux is as follows:
-
In the Cabal file, the following
ld-optionsare set:-staticand-pthread. -
The Stack command is run in a Docker container based on Alpine Linux. The relevant Docker image repository is set out in Stack's
stack.yamlfile. See also Olivier Benz's GHC musl project. -
Stack's configuration includes:
extra-include-dirs: - /usr/include extra-lib-dirs: - /lib - /usr/lib
-
The build command is
stack build --docker --system-ghc --no-install-ghc(on x86_64) orstack build --docker --docker-stack-exe=image --system-ghc --no-install-ghc(on AArch64; the host Stack and the image Stack must have the same version number).
See the Travis CI instructions
See the Azure CI instructions
Currently, on Windows, GHC produces binaries linked statically with
GNU Multiple Precision Arithmetic Library (GMP), which is
used by integer-gmp library
to provide big integer implementation for Haskell. Contrary to the majority of
Haskell code licensed under permissive BSD3 license, GMP library is licensed
under LGPL, which means resulting binaries
have to be provided with source code or object files.
That may or may not be acceptable for your situation. Current workaround is to
use GHC built with alternative big integer implementation called
integer-simple, which is free from LGPL limitations as it's pure Haskell and
does not use GMP. Unfortunately it has yet to be available out of the box with
Stack. See issue #399
for the ongoing effort and information on workarounds.
When executing a binary after building with stack build (e.g. for target
"foo"), the command foo.exe might complain about missing runtime libraries
(whereas stack exec foo works).
Windows is not able to find the necessary C++ libraries from the standard
prompt because they're not in the PATH environment variable. stack exec works
because it's modifying PATH to include extra things.
Those libraries are shipped with GHC (and, theoretically in some cases, MSYS2).
The easiest way to find them is stack exec which. For example, command:
stack exec -- which libstdc++-6.dll
/c/Users/Michael/AppData/Local/Programs/stack/i386-windows/ghc-7.8.4/mingw/bin/libstdc++-6.dll
A quick workaround is adding this path to the PATH environment variable or copying the files somewhere Windows finds them (see https://msdn.microsoft.com/de-de/library/7d83bc18.aspx).
See issue #425.
Another issue that may arise with building on Windows is as follows. The default
location of Stack's programs folder is %LOCALAPPDATA\Programs\stack. If there
is a space character in the %LOCALAPPDATA% path this may, in some
circumstances, cause problems with building packages that make use of the GNU
project's autoconf package and configure shell script files. It may be
necessary to override the default location of Stack's programs folder. See the
local-programs-path option for more
information.
See issue #4726.
Stack downloads and extracts files to $STACK_ROOT/programs on most platforms,
which defaults to ~/.stack/programs. On Windows $LOCALAPPDATA\Programs\stack
is used. If there is not enough free space in this directory, Stack may fail.
For instance, stack setup with a GHC installation requires roughly 1GB free.
If this is an issue, you can set local-programs-path in your
~/.stack/config.yaml to a directory on a file system with more free space.
If you use Stack with Nix integration, be aware that Nix uses a TMPDIR
variable, and if it is not set Nix sets it to some subdirectory of /run, which
on most Linuxes is a Ramdir. Nix will run the builds in TMPDIR, therefore if
you don't have enough RAM you will get errors about disk space. If this happens
to you, please manually set TMPDIR before launching Stack to some directory
on the disk.
Because GHC options often only affect optimization levels and warning behavior,
Stack doesn't recompile when it detects an option change by default. This
behavior can be changed though by setting the
rebuild-ghc-options option to
true.
To force recompilation manually, use the --force-dirty flag. If this still
doesn't lead to a rebuild, add the -fforce-recomp flag to your
--ghc-options.
By default, Stack applies command line GHC options only to project packages. For an explanation of this choice see this discussion on issue #827.
If you still want to set specific GHC options for a dependency, use the
ghc-options option in your
YAML configuration file.
To change the set of packages that command line GHC options apply to, use the apply-ghc-options option.
stack setup on a Windows system only tells me to add certain paths to the PATH variable instead of doing it
With PowerShell, it is easy to automate even that step. Command:
$Env:Path = ( stack setup | %{ $_ -replace '[^ ]+ ', ''} ), $Env:Path -join ";"The first thing to remove is project-specific .stack-work directory within
the project's directory. Next, remove the Stack root directory overall. You may
have errors if you remove the latter but leave the former. Removing Stack
itself will relate to how it was installed, and if you used GHC installed
outside of Stack, that would need to be removed separately.
See issue #644 for more details.
(Updated in January 2019)
This is related to more recent versions of Linux distributions that have GCC with PIE enabled by default. The continuously-updated distros like Arch, in particular, had been in flux with this change and the upgrading libtinfo6/ncurses6, and there were some workarounds attempted in Stack that ended up causing trouble as these distros evolved.
GHC added official support for this setup in 8.0.2, so if you are using an
older version your best bet is to upgrade. You may be able to work around it
for older versions by editing ~/.stack/programs/x86_64-osx/ghc-VER/lib/ghc- VER/settings (replace VER with the GHC version) and adding -no-pie (or
--no-pie in the case of Gentoo, at least as of December 2017) to the C
compiler link flags.
If stack setup complains that there is no linuxNN-*-nopie bindist available,
try adding ghc-build: * (replacing the * with the actual value that
precedes -nopie, which may be empty) to your ~/.stack/config.yaml (this
will no longer be necessary for stack >= 1.7).
If you are experiencing this with GHC >= 8.0.2, try running
stack setup --reinstall if you've upgraded your Linux distribution or you set
up GHC before late December 2017.
If GHC doesn't recognize your C compiler as being able to use -no-pie, this
can happen even with GCC and Clang, it might be necessary to enable this feature
manually. To do this, just change
("C compiler supports -no-pie", "NO"), to
("C compiler supports -no-pie", "YES"),
in the file ~/.stack/programs/x86_64-osx/ghc-VER/lib/ghc-VER/settings.
If you are still having trouble after trying the above, check the following for more possible workarounds:
These are written to *.dump-* files inside the package's .stack-work
directory. Specifically, they will be available at
PKG-DIR/$(stack path --dist-dir)/build/SOURCE-PATH, where SOURCE-PATH is the
path to the source file, relative to the location of the Cabal file. When
building named components such as test-suites, SOURCE-PATH will also include
COMPONENT/COMPONENT-tmp, where COMPONENT is the name of the component.
If you are on Mac OS X 10.11 ("El Capitan") or later, there is a GHC issue
#11617 which prevents the
DYLD_LIBRARY_PATH environment variable from being passed to GHC (see issue
#1161) when System
Integrity Protection (a.k.a. "rootless") is enabled. There are two known
workarounds:
- Known to work in all cases: disable System Integrity Protection. WARNING: Disabling SIP will severely reduce the security of your system, so only do this if absolutely necessary!
- Experimental: modify GHC's shell script wrappers to use a shell outside the protected directories (see issue #1161).
Some versions of Windows PowerShell don't pass the -- to programs (see issue
#813). The workaround
is to quote the "--". For example, command:
stack exec "--" cabal --versionThis is known to be a problem on Windows 7, but seems to be fixed on Windows 10.
No, this is currently out of the scope of Stack's target set of features. Instead of attempting to automate the installation of 3rd party dependencies, we have the following approaches for handling system dependencies:
-
Nix and docker help make your build and execution environment deterministic and predictable. This way, you can install system dependencies into a container, and share this container with all developers.
-
If you have installed some libraries into a non-standard location, use the
extra-lib-dirsoption or theextra-include-dirsoption to specify it.
In the future, Stack might give operating system-specific suggestions for how to install system libraries.
In principle, you can use the following command to add a certificate to your system certificate keychain:
sudo security add-trusted-cert -d -r trustRoot -k /Library/Keychains/System.keychain <certificate>Some users have reported issues with this approach, see issue #907 for more information.
Use the SYSTEM_CERTIFICATE_PATH environment variable to point at the directory
where you keep your SSL certificates.
Add ghc-options: -vN to the Cabal file or pass it via
stack build --ghc-options="-v".
Yes:
- If a package directory contains an Hpack
package.yamlfile, then Stack will use it to generate a Cabal file when building the package. - You can run
stack initto initialize astack.yamlfile regardless of whether your packages are declared with Cabal files or with Hpackpackage.yamlfiles. - You can use the
with-hpackYAML configuration or command line option to specify an Hpack executable to use instead of Stack's in-built Hpack functionality.
This is likely to be caused by having both a LLVM installation and default Apple Clang compiler on the PATH. The symptom of this issue is a linker error "bad relocation (Invalid pointer diff)". The compiler picks up inconsistent versions of binaries and the mysterious error occurs.
The workaround is to remove LLVM binaries from the PATH.
clang: warning: argument unused during compilation: '-nopie'
[-Wunused-command-line-argument]This warning is shown when compiler support of -no-pie is expected but
unavailable. It's possible to bypass the warning for a specific version of GHC
by modifying a global setting:
# ~/.stack/programs/x86_64-osx/ghc-8.2.2/lib/ghc-8.2.2/settings
-- ("C compiler supports -no-pie", "YES"),
++ ("C compiler supports -no-pie", "NO"),Note that we're fixing ghc-8.2.2 in this case; repeat for other versions as necessary.
You should apply this fix for the version of GHC that matches your snapshot.
Issue #4009 goes into further detail.
How do I install GHC in Stack when it fails with the error: Missing ghc bindist for "linux64-ncurses6"?
Example Error:
No setup information found for ghc-8.6.4 on your platform.
This probably means a GHC bindist has not yet been added for OS key 'linux64-ncurses6'.
Supported versions: ghc-7.10.3, ghc-8.0.1, ghc-8.0.2, ghc-8.2.1, ghc-8.2.2
Most Linux distributions have standardized on providing libtinfo.so.6 (either directly or as a symlink to libncursesw.so.6). As such, there aren't GHC 8.6.* bindists that link to libncursesw.so.6 available.
So creating a symlink to libncursesw.so.6 as libtinfo.so.6 can prevent this error (root privileges might be required). Command:
ln -s /usr/lib/libncursesw.so.6 /usr/lib/libtinfo.so.6