A guide to the best practice for creating a new Haskell project or program.
Almost all new Haskell projects use the following tools. Each is intrinsically useful, but using a set of common tools also benefits everyone by increasing productivity, and you're more likely to get patches.
Use darcs, unless you have a specific reason not to, in which case use git. If you don't like git, go back and look at darcs. It's written in Haskell, and it's used by many Haskell developers. See the wikibook Understanding darcs to get started.
For libraries, use Haddock. We recommend using recent versions of haddock (2.8 or above, as of December 2010).
To get started, try Haskell/Testing. For a slightly more advanced introduction, Simple Unit Testing in Haskell is a blog article about creating a testing framework for QuickCheck using some Template Haskell.
Structure of a simple project
The basic structure of a new Haskell project can be adopted from HNop, the minimal Haskell project. It consists of the following files, for the mythical project "haq".
- Haq.hs -- the main haskell source file
- haq.cabal -- the cabal build description
- Setup.hs -- build script itself
- _darcs or .git -- revision control
- README -- info
- LICENSE -- license
You can of course elaborate on this, with subdirectories and multiple modules.
Here is a transcript on how you'd create a minimal darcs-using and cabalised Haskell project, for the cool new Haskell program "haq", build it, install it and release.
The new tool 'mkcabal' automates all this for you, but it's important that you understand all the parts first.
We will now walk through the creation of the infrastructure for a simple Haskell executable. Advice for libraries follows after.
Create a directory
Create somewhere for the source:
$ mkdir haq $ cd haq
Write some Haskell source
Write your program:
$ cat > Haq.hs -- -- Copyright (c) 2006 Don Stewart - http://www.cse.unsw.edu.au/~dons -- GPL version 2 or later (see http://www.gnu.org/copyleft/gpl.html) -- import System.Environment -- 'main' runs the main program main :: IO () main = getArgs >>= print . haqify . head haqify s = "Haq! " ++ s
Stick it in darcs
Place the source under revision control:
$ darcs init $ darcs add Haq.hs $ darcs record addfile ./Haq.hs Shall I record this change? (1/?) [ynWsfqadjkc], or ? for help: y hunk ./Haq.hs 1 +-- +-- Copyright (c) 2006 Don Stewart - http://www.cse.unsw.edu.au/~dons +-- GPL version 2 or later (see http://www.gnu.org/copyleft/gpl.html) +-- +import System.Environment + +-- | 'main' runs the main program +main :: IO () +main = getArgs >>= print . haqify . head + +haqify s = "Haq! " ++ s Shall I record this change? (2/?) [ynWsfqadjkc], or ? for help: y What is the patch name? Import haq source Do you want to add a long comment? [yn]n Finished recording patch 'Import haq source'
And we can see that darcs is now running the show:
$ ls Haq.hs _darcs
$ git config --global user.name "John Doe" $ git config --global user.email firstname.lastname@example.org $ git init $ git add * $ git commit -m 'Import haq source' $ ls -A .git Haq.hs
Add a build system
Create a .cabal file describing how to build your project:
$ cat > haq.cabal Name: haq Version: 0.0 Synopsis: Super cool mega lambdas Description: My super cool, indeed, even mega lambdas will demonstrate a basic project. You will marvel. License: GPL License-file: LICENSE Author: Don Stewart Maintainer: Don Stewart <email@example.com> Build-Depends: base Executable: haq Main-is: Haq.hs
(If your package uses other packages, e.g. array, you'll need to add them to the Build-Depends: field.) Add a Setup.lhs that will actually do the building:
$ cat > Setup.lhs #! /usr/bin/env runhaskell > import Distribution.Simple > main = defaultMain
Cabal allows either Setup.hs or Setup.lhs; as long as the format is appropriate, it doesn't matter which one you choose. But it's a good idea to always include the
#! /usr/bin/env runhaskell line; because it follows the shebang convention, you could execute the Setup.hs directly in a Unix shell instead of always manually calling runhaskell (assuming the Setup file is marked executable, of course).
Record your changes:
$ darcs add haq.cabal Setup.lhs $ darcs record --all What is the patch name? Add a build system Do you want to add a long comment? [yn]n Finished recording patch 'Add a build system'
$ git add haq.cabal Setup.lhs $ git commit -m 'Add a build system'
Build your project
Now build it!
$ runhaskell Setup.lhs configure --prefix=$HOME --user $ runhaskell Setup.lhs build $ runhaskell Setup.lhs install
And now you can run your cool project:
$ haq me "Haq! me"
You can also run it in-place, avoiding the install phase:
$ dist/build/haq/haq you "Haq! you"
Build some haddock documentation
Generate some API documentation into dist/doc/*
$ runhaskell Setup.lhs haddock
which generates files in dist/doc/ including:
$ w3m -dump dist/doc/html/haq/Main.html haq Contents Index Main Synopsis main :: IO () Documentation main :: IO () main runs the main program Produced by Haddock version 0.7
No output? Make sure you have actually installed haddock. It is a separate program, not something that comes with the Haskell compiler, like Cabal.
Add some automated testing: QuickCheck
We'll use QuickCheck to specify a simple property of our Haq.hs code. Create a tests module, Tests.hs, with some QuickCheck boilerplate:
$ cat > Tests.hs import Char import List import Test.QuickCheck import Text.Printf main = mapM_ (\(s,a) -> printf "%-25s: " s >> a) tests instance Arbitrary Char where arbitrary = choose ('\0', '\128') coarbitrary c = variant (ord c `rem` 4)
Now let's write a simple property:
$ cat >> Tests.hs -- reversing twice a finite list, is the same as identity prop_reversereverse s = (reverse . reverse) s == id s where _ = s :: [Int] -- and add this to the tests list tests = [("reverse.reverse/id", test prop_reversereverse)]
We can now run this test, and have QuickCheck generate the test data:
$ runhaskell Tests.hs reverse.reverse/id : OK, passed 100 tests.
Let's add a test for the 'haqify' function:
-- Dropping the "Haq! " string is the same as identity prop_haq s = drop (length "Haq! ") (haqify s) == id s where haqify s = "Haq! " ++ s tests = [("reverse.reverse/id", test prop_reversereverse) ,("drop.haq/id", test prop_haq)]
and let's test that:
$ runhaskell Tests.hs reverse.reverse/id : OK, passed 100 tests. drop.haq/id : OK, passed 100 tests.
Running the test suite from darcs
We can arrange for darcs to run the test suite on every commit:
$ darcs setpref test "runhaskell Tests.hs" Changing value of test from '' to 'runhaskell Tests.hs'
will run the full set of QuickChecks. (If your test requires it you may need to ensure other things are built too e.g.: darcs setpref test "alex Tokens.x;happy Grammar.y;runhaskell Tests.hs").
Let's commit a new patch:
$ darcs add Tests.hs $ darcs record --all What is the patch name? Add testsuite Do you want to add a long comment? [yn]n Running test... reverse.reverse/id : OK, passed 100 tests. drop.haq/id : OK, passed 100 tests. Test ran successfully. Looks like a good patch. Finished recording patch 'Add testsuite'
Excellent, now patches must pass the test suite before they can be committed.
Tag the stable version, create a tarball, and sell it!
Tag the stable version:
$ darcs tag What is the version name? 0.0 Finished tagging patch 'TAG 0.0'
Advanced Darcs functionality: lazy get
As your repositories accumulate patches, new users can become annoyed at how long it takes to accomplish the initial darcs get. (Some projects, like yi or GHC, can have thousands of patches.) Darcs is quick enough, but downloading thousands of individual patches can still take a while. Isn't there some way to make things more efficient?
Darcs provides the --lazy option to darcs get. This enables to download only the latest version of the repository. Patches are later downloaded on demand if needed.
When distributing your Haskell program, you have roughly three options:
- distributing via a Darcs repository
- distributing a tarball
- a Darcs tarball
- a Cabal tarball
With a Darcs repository, if it is public, than you are done. However: perhaps you don't have a server with Darcs, or perhaps your computer isn't set up for people to darcs pull from it. In which case you'll need to distribute the source via tarball.
Tarballs via darcs
Darcs provides a command where it will make a compressed tarball, and it will place a copy of all the files it manages into it. (Note that nothing in _darcs will be included - it'll just be your source files, no revision history.)
$ darcs dist -d haq-0.0 Created dist as haq-0.0.tar.gz
And you're all set up!
Tarballs via Cabal
Since our code is cabalised, we can create a tarball with Cabal directly:
$ runhaskell Setup.lhs sdist Building source dist for haq-0.0... Source tarball created: dist/haq-0.0.tar.gz
This has advantages and disadvantages compared to a Darcs-produced tarball. The primary advantage is that Cabal will do more checking of our repository, and more importantly, it'll ensure that the tarball has the structure needed by HackageDB and cabal-install.
However, it does have a disadvantage: it packages up only the files needed to build the project. It will deliberately fail to include other files in the repository, even if they turn out to be necessary at some point. To include other files (such as Test.hs in the above example), we need to add lines to the cabal file like:
If we had them, we could make sure files like AUTHORS or the README get included as well:
data-files: AUTHORS, README
The following files were created:
$ ls Haq.hs Tests.hs dist haq.cabal Setup.lhs _darcs haq-0.0.tar.gz
The process for creating a Haskell library is almost identical. The differences are as follows, for the hypothetical "ltree" library:
The source should live under a directory path that fits into the existing module layout guide. So we would create the following directory structure, for the module Data.LTree:
$ mkdir Data $ cat > Data/LTree.hs module Data.LTree where
So our Data.LTree module lives in Data/LTree.hs
The Cabal file
Cabal files for libraries list the publically visible modules, and have no executable section:
$ cat ltree.cabal Name: ltree Version: 0.1 Description: Lambda tree implementation License: BSD3 License-file: LICENSE Author: Don Stewart Maintainer: firstname.lastname@example.org Build-Depends: base Exposed-modules: Data.LTree
We can thus build our library:
$ runhaskell Setup.lhs configure --prefix=$HOME --user $ runhaskell Setup.lhs build Preprocessing library ltree-0.1... Building ltree-0.1... [1 of 1] Compiling Data.LTree ( Data/LTree.hs, dist/build/Data/LTree.o ) /usr/bin/ar: creating dist/build/libHSltree-0.1.a
and our library has been created as a object archive. On *nix systems, you should probably add the --user flag to the configure step (this means you want to update your local package database during installation). Now install it:
$ runhaskell Setup.lhs install Installing: /home/dons/lib/ltree-0.1/ghc-6.6 & /home/dons/bin ltree-0.1... Registering ltree-0.1... Reading package info from ".installed-pkg-config" ... done. Saving old package config file... done. Writing new package config file... done.
And we're done! You can use your new library from, for example, ghci:
$ ghci -package ltree Prelude> :m + Data.LTree Prelude Data.LTree>
The new library is in scope, and ready to go.
More complex build systems
For larger projects it is useful to have source trees stored in subdirectories. This can be done simply by creating a directory, for example, "src", into which you will put your src tree.
To have Cabal find this code, you add the following line to your Cabal file:
Cabal can set up to also run configure scripts, along with a range of other features. For more information consult the Cabal documentation.
If your library uses internal modules that are not exposed, do not forget to list them in the other-modules field:
Failing to do so (as of GHC 6.8.3) may lead to your library deceptively building without errors but actually being unusable from applications, which would fail at build time with a linker error.
A package management tool for Haskell called cabal-install provides a command line tool to help developers create a simple cabal project. Just run and answer all the questions. Default values are provided for each.
$ cabal init Package name [default "test"]? Package version [default "0.1"]? Please choose a license: ...
mkcabal is a tool that existed before cabal init, which also automatically populates a new cabal project :
N.B. This tool does not work in Windows. The Windows version of GHC does not include the readline package that this tool needs.
$ mkcabal Project name: haq What license ["GPL","LGPL","BSD3","BSD4","PublicDomain","AllRightsReserved"] ["BSD3"]: What kind of project [Executable,Library] [Executable]: Is this your name? - "Don Stewart " [Y/n]: Is this your email address? - "<email@example.com>" [Y/n]: Created Setup.lhs and haq.cabal $ ls Haq.hs LICENSE Setup.lhs _darcs dist haq.cabal
which will fill out some stub Cabal files for the project 'haq'.
To create an entirely new project tree:
$ mkcabal --init-project Project name: haq What license ["GPL","LGPL","BSD3","BSD4","PublicDomain","AllRightsReserved"] ["BSD3"]: What kind of project [Executable,Library] [Executable]: Is this your name? - "Don Stewart " [Y/n]: Is this your email address? - "<firstname.lastname@example.org>" [Y/n]: Created new project directory: haq $ cd haq $ ls Haq.hs LICENSE README Setup.lhs haq.cabal
Code for the common base library package must be BSD licensed or something more Free/Open. Otherwise, it is entirely up to you as the author.
Choose a licence (inspired by this). Check the licences of things you use, both other Haskell packages and C libraries, since these may impose conditions you must follow.
Use the same licence as related projects, where possible. The Haskell community is split into 2 camps, roughly, those who release everything under BSD or public domain, and the GPL/LGPLers (this split roughly mirrors the copyleft/noncopyleft divide in Free software communities). Some Haskellers recommend specifically avoiding the LGPL, due to cross module optimisation issues. Like many licensing questions, this advice is controversial. Several Haskell projects (wxHaskell, HaXml, etc.) use the LGPL with an extra permissive clause to avoid the cross-module optimisation problem.
It's important to release your code as stable, tagged tarballs. Don't just rely on darcs for distribution.
- darcs dist generates tarballs directly from a darcs repository
$ cd fps $ ls Data LICENSE README Setup.hs TODO _darcs cbits dist fps.cabal tests $ darcs dist -d fps-0.8 Created dist as fps-0.8.tar.gz
You can now just post your fps-0.8.tar.gz
You can also have darcs do the equivalent of 'daily snapshots' for you by using a post-hook.
put the following in _darcs/prefs/defaults:
apply posthook darcs dist apply run-posthook
- Tag each release using darcs tag. For example:
$ darcs tag 0.8 Finished tagging patch 'TAG 0.8'
Then people can darcs get --lazy --tag 0.8, to get just the tagged version (and not the entire history).
You can host public and private Darcs repositories on http://patch-tag.com/ for free. Otherwise, a Darcs repository can be published simply by making it available from a web page. Another option is to host on the Haskell Community Server at http://code.haskell.org/. You can request an account via http://community.haskell.org/admin/. You can also use https://github.com/ for Git hosting.
A complete example of writing, packaging and releasing a new Haskell library under this process has been documented.
||At least part of this page was imported from the Haskell wiki article How to write a Haskell program, in accordance to its Simple Permissive License. If you wish to modify this page and if your changes will also be useful on that wiki, you might consider modifying that source page instead of this one, as changes from that page may propagate here, but not the other way around. Alternately, you can explicitly dual license your contributions under the Simple Permissive License. Note also that the original tutorial contains extra information about announcing your software and joining the Haskell community, which may be of interest to you.|
- This is actually a good thing, since it allows us to do things like create an elaborate test suite which doesn't get included in the tarball, so users aren't bothered by it. It also can reveal hidden assumptions and omissions in our code - perhaps your code was only building and running because of a file accidentally generated.