Up until now we've only used
base and the libraries
with GHC. Because of that we didn't really need to do anything more fancy
runghc to run our program. However, we want to start using
external libraries which are not included with GHC in our programs.
External packages can be downloaded from Hackage - Haskell's central package archive, Stackage - a subset of Hackage packages that are known to work together, or even from remote git repositories. Usually Haskellers use a package manager to download and manage packages for different projects. The most popular package managers for Haskell are cabal and stack.
A major difference between the two tools is their philosophy.
cabal tries to be a more minimalist tool that handles building Haskell projects,
doing package management using the whole of Hackage, and uses complicated algorithms
to make sure packages work together.
stack tries to be a more maximalistic tool that handles installing the right GHC
for each project, provides integration with external tools like hoogle,
and lets the user choose which 'set' of packages (including their versions) they want to use.
If you've installed Haskell using GHCup, you most likely have
If you've installed Haskell using stack, well, you have
Check the haskell.org downloads page if that's not the case.
Using external packages can be done in multiple ways. For quick experimentation, we can just ask stack or cabal to build or even run our program with external packages. But as programs get larger, use more dependencies, and require more functionality, it is better to just create a project description for our programs and even libraries.
Describing packages is done in a cabal file. We can ask cabal or stack
to generate one for use using
cabal init --libandexe or
along with many other files, but we will likely need to edit the file by hand
later so let's just paste some sort of an initial example and then edit it.
cabal-version: 2.4 name: name should match with <name>.cabal version: version should use PvP synopsis: Synopsis will appear in the hackage package listing and search description: The description will appear at the top of a library homepage: Homepage url bug-reports: issue-tracker url license: License name license-file: License file author: Author name maintainer: Maintainer email category: Hackage categories, separated by commas extra-doc-files: README.md common common-settings default-language: Haskell2010 ghc-options: -Wall library import: common-settings hs-source-dirs: src build-depends: base , directory exposed-modules: HsBlog HsBlog.Convert HsBlog.Html HsBlog.Html.Internal HsBlog.Markup -- other-modules: executable hs-blog-gen import: common-settings hs-source-dirs: app main-is: Main.hs build-depends: base , <package-name> ghc-options: -O
The first part should be fairly straightforward from the comments, maybe except for:
cabal-version: Defines which cabal versions can build this project. We've specified 2.4 and above. More info on different versions.
name: The name of your library and package. Must match with the .cabal filename. Usually starts with a lowercase. Check if your package name is already taken on Hackage.
version: Some Haskell packages use semver, most use PvP.
license: Most Haskell packages use BSD-3-Clause. Neil Mitchell blogged about this. You can find more licenses if you'd like at choosealicense.com.
extra-doc-files: Include extra doc files here, such as
Let's fill this with the metadata of our project:
cabal-version: 2.4 name: hs-blog version: 0.1.0.0 synopsis: A custom blog generator from markup files description: This package provides a static blog generator from a custom markup format to HTML. It defines a parser for this custom markup format as well as an html pretty printer EDSL. It is used as the example project in the online book 'Learn Haskell Blog Generator'. See the README for more details. homepage: https://github.com/soupi/learn-haskell-blog-generator bug-reports: https://github.com/soupi/learn-haskell-blog-generator/issues license: BSD-3-Clause license-file: LICENSE.txt author: Gil Mizrahi maintainer: email@example.com category: Learning, Web extra-doc-files: README.md
Cabal package descriptions can include multiple "targets": libraries, executables, and test suites. Since Cabal 2.2, we can use common stanzas to group settings to be shared between different targets, so we don't have to repeat them for each target.
In our case we've created a new common stanza (or block) called
defined the default language (Haskell has two standards, 98 and 2010),
and instructed GHC to compile with
common common-settings default-language: Haskell2010 ghc-options: -Wall
Later, in our targets' descriptions, we can add
import: common-settings ,
and all of these settings will be automatically added.
library target, we define:
- The settings with which to build the library (in this case we just import
- The directory in which the source files can be found
- The packages we require to build the library
- The modules exposed from the library and can be used by others
- The modules not exposed from the library and which cannot be used by others;
these could be any module you don't wish to export, such as an internal utility
In our case we don't have anything like this, so we commented out the
Note that it is common to specify version bounds for packages.
Version bounds specify which package versions this library works with.
These can also be generated using cabal with the
cabal gen-bounds command.
library import: common-settings hs-source-dirs: src build-depends: base , directory exposed-modules: HsBlog HsBlog.Convert HsBlog.Html HsBlog.Html.Internal HsBlog.Markup -- other-modules:
Also note that we've added an additional hierarchy for our modules and defined
a different source directory. This means we will need to move the files around
a bit and change the
module name in each file and the
import statements. This is to avoid
conflict with other packages that a user might import.
Do this now.
module HsBlog ( main , process ) where import qualified HsBlog.Markup as Markup import qualified HsBlog.Html as Html import HsBlog.Convert (convert)
module HsBlog.Convert where import qualified HsBlog.Markup as Markup import qualified HsBlog.Html as Html
module HsBlog.Html ... import HsBlog.Html.Internal
module HsBlog.Html.Internal where
We have separated our code into two sections: a library and an executable, why?
First, libraries can be used by others. If we publish our code and someone wants to use it and build upon it, they can. Executables can't be imported by other projects. Second, we can write unit tests for libraries. It is usually benefitical to write most, if not all, of our logic as a library, and provide a thin executable over it.
Executables' descriptions are very similar to libraries, here we define:
- The name of the executable
- Where the source directory for this application is
- Which file is the 'Main' file
- Import our library, which is named
- Add additional flag for GHC:
-Oto compile with optimizations
executable hs-blog-gen import: common-settings hs-source-dirs: app main-is: Main.hs build-depends: base , hs-blog ghc-options: -O
We can write many executables descriptions. In this case we only need one.
Exercise: Add a new file:
app/Main.hs which imports
HsBlog and runs
-- app/Main.hs module Main where import qualified HsBlog main :: IO () main = HsBlog.main
test-suite defines a target for running package tests. We will get back to it
in a later chapter.
cabal-version: 2.4 name: hs-blog version: 0.1.0.0 synopsis: A custom blog generator from markup files description: This package provides a static blog generator from a custom markup format to HTML. It defines a parser for this custom markup format as well as an html pretty printer EDSL. It is used as the example project in the online book 'Learn Haskell Blog Generator'. See the README for more details. homepage: https://github.com/soupi/learn-haskell-blog-generator bug-reports: https://github.com/soupi/learn-haskell-blog-generator/issues license: BSD-3-Clause license-file: LICENSE.txt author: Gil Mizrahi maintainer: firstname.lastname@example.org category: Learning, Web extra-doc-files: README.md common common-settings default-language: Haskell2010 ghc-options: -Wall library import: common-settings hs-source-dirs: src build-depends: base , directory exposed-modules: HsBlog HsBlog.Convert HsBlog.Html HsBlog.Html.Internal HsBlog.Markup -- other-modules: executable hs-blog-gen import: common-settings hs-source-dirs: app main-is: Main.hs build-depends: base , hs-blog ghc-options: -O
We'll also add a
README.md file and a
Just write whatever you want here:
# hs-blog One day it will be a static blog generator. [Read the book](https://soupi.github.io/learn-haskell-blog-generator).
This is BSD-3-Clause with me as the author. Please write your own name for your projects :)
BSD 3-Clause License Copyright (c) 2021, Gil Mizrahi All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. 3. Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
The cabal.project and
files are used by
stack respectively to add additional information on how
to build the package. While
cabal.project isn't necessary to use
is necessary in order to use
stack, so we will cover it briefly.
There are two important fields a
stack.yaml file must have:
resolver: Describes which snapshot to use for packages and ghc version. We will choose the latest (at time of writing) on the
lts-18.9. Visit this link to find out which packages this snapshot includes, what their versions are, and which GHC version is used with this snapshot.
packages: Describes the location of packages we plan to build. In our case we have only one and it can be found in the current directory.
stack.yaml to our project directory:
resolver: lts-18.9 packages: - .
For additional options and configurations, please consult the relevant user guides.
Now, instead of manually running
runghc Main.hs, we will use either
cabal to build and run our program and package (I mostly use stack, but it's up to you).
Building the project - on the first run, cabal will download the package dependencies and use the GHC on PATH to build the project.
Cabal caches packages between projects, so if a new project uses the same packages
with the same versions (and the same flag settings) they will not need to be reinstalled.
cabal commands are usually prefixed with
v2- to note that we want to use the new
build system implementation.
In older version of cabal, packages could be installed either globally, or in sandboxes. In each sandbox (and globally) there could only be one version of a package installed, and users would usually create different sandboxes for different projects, without caching packages between projects.
With the new build system implementation, multiple versions of the same package can be installed globally, and for each project cabal will (try to) choose a specific version for each package dependency such that they all work together, without needing sandboxing. This change helps us increase sharing of built packages while avoiding conflicts and manual handling of sandboxes.
A few important commands we should be familiar with:
fetches information from remote package repositories (specifically Hackage unless specified otherwise)
and updates the local package index which includes various information about available packages such as
their names, versions and dependencies.
Usually the first command to run before fetching package dependencies.
compiles the various targets (such as
It will also fetch and install the package dependencies when they're not already installed.
cabal v2-run hs-blog-gen -- <program arguments>
Can be used to compile and then run a target (in our case our
executable which we named
We separate arguments passed to
cabal and arguments passed to our target program with
cabal v2-repl hs-blog
ghci in the context of the target (in our case our
library which we named
it will load the target's package dependencies and modules to be available in
Deletes the build artifacts that we built.
There are more interesting commands we could use, such as
cabal v2-freeze to generate
a file which records the packages versions and flags we used to build this project,
cabal v2-sdist to bundle the project source to a package tarball which can be
uploaded to Hackage. If you'd like to learn more visit the
Cabal user guide.
Building the project - on the first run, stack will install the right GHC for this project
which is specified by the
resolver field in the
download the package dependencies, and compile the project.
Stack caches these installations between projects that use the same resolver, so future projects with the same resolver and future runs of this project won't require reinstallation. This approach is kind of a middle ground between full packages sharing and sandboxes.
Let's look at the (somewhat) equivalent commands for Stack:
will compile the project as described above - installing GHC and package dependencies if they are not
stack exec hs-blog-gen -- <program arguments>
will run the executable passing the program arguments to our executable.
stack ghci hs-blog
ghci in the context of our library
hs-blog - loading the library modules
cleans up build artifacts.
The Stack user guide contains more information about how stack works and how to use it effectively.
Both stack and cabal create build artifacts that we will not want to track using
our version control. These build artifacts are found in the
.stack-work directories. We can add these to a
(or similar for other version control programs) to ignore them:
dist dist-newstyle .stack-work
My suggestion is:
- Search for a tutorial on something you'd like to do, and see which packages come up
- Use the download amount on Hackage as an indication of package popularity
- Use Stackage package synopses to locate a relevant package
- Check social network channels for recommendations, but know that sometimes people tend to recommend inappropriate solutions and packages that might be too complicated or still experimental
It's also important to note the amount of dependencies a package has. Adding many dependencies will affect compilation time and code size. And it can sometimes be a good thing to consider when comparing packages, or considering whether a package is needed at all.
We've created a package description for our library and used
to build our program. In future chapters we'll start adding external packages,
we'll only have to add them to the
build-depends section in the cabal file and
our package manager will download and install the required package for us!
We've made some change to our project directory, and it should now look like this:
. ├── app │ └── Main.hs ├── hs-blog.cabal ├── LICENSE.txt ├── README.md ├── src │ ├── HsBlog │ │ ├── Convert.hs │ │ ├── Html │ │ │ └── Internal.hs │ │ ├── Html.hs │ │ └── Markup.hs │ └── HsBlog.hs └── stack.yaml 4 directories, 10 files
Note that this package format is something we could release on Hackage for other Haskell developers to use!