Difference between revisions of "Haskell package guidelines"

From ArchWiki
Jump to navigation Jump to search
(Haskell Packages)
(Haskell Packages: remove mention of haskell-testing, since it's being merged back to haskell now)
Line 36: Line 36:
Putting [haskell] above [extra] will ensure that the packages from [haskell] take precedence, in case of duplicate packages in the two repositories.
Putting [haskell] above [extra] will ensure that the packages from [haskell] take precedence, in case of duplicate packages in the two repositories.
There is currently also a [haskell-testing] repo for GHC 7.6.
  Server = http://www.kiwilight.com/haskell/testing/$arch     
=== [haskell-extra] ===
=== [haskell-extra] ===

Revision as of 14:48, 8 October 2012

Template:Package Guidelines

Haskell is well supported on Arch Linux, with GHC and other key tools available via the official repositories, a growing number of packages made available by the ArchHaskell group, and a large part of hackage.haskell.org's library database available in the AUR. The AUR packages, especially those owned by archhaskell, are deprecated. It is better to use and contribute to the [haskell] repository.

The community around Haskell on Arch is dying but in the processo of revival, so your help is especially welcome!


All the details on the ArchHaskell group is available on its own page.

Haskell Packages

The core Haskell tools are available in the core system (extra):


Our policy for [extra] is to provide the Haskell platform, and popular Haskell applications.


[community] provides additional packages that are popular and not part of the Haskell platform, such as xmonad. These packages are also provided by the [haskell] repo below. So, if you just want to use a few popular packages, you can just use what is provided by the official Arch repos. But if you want to use Haskell more seriously, read on.


The [haskell] repository is the official repository of packages maintained by the ArchHaskell team. This repository represents the last tier of stability, before resorting to the packages in the AUR, or perhaps building packages yourself with cabal2arch. [haskell] can be accessed by adding the following entry to /etc/pacman.conf:

Server = http://xsounds.org/~haskell/$arch

The set of packages in the [haskell] repository is derived from the habs tree officially located here. A tool called cblrepo is used to keep the habs tree synchronized with the official Haskell packages from Hackage.

Putting [haskell] above [extra] will ensure that the packages from [haskell] take precedence, in case of duplicate packages in the two repositories.


We are moving towards the possibility of distributing the desired contents of [haskell] (that is, all of Hackage) over several repositories (that is, distributing maintenance), while still collecting them all in a single repository. The first test towards this is habs-extra, which is also provided as a repository to be added to /etc/pacman.conf:

Server = http://archhaskell.mynerdside.com/$repo/$arch

This should currently be used in conjunction with [haskell], because it provides additional packages while re-using those already provided by [haskell]. More details are available in this message, and in the README on github.


A huge number (almost 2000) packages built from http://hackage.haskell.org.

These generally improve on installing directly from Hackage as they resolve required C libraries. They can be installed as, for example:

# paktahn -S haskell-csv

Anything not found here can be installed via cabal-install directly from Hackage.

Unfortunately, many of the packages in the AUR are outdated due to a lack of resources. In practice, one could use the cabal2arch program to create PKGBUILDs directly from Hackage. But now it is recommended to use cblrepo and create something like [haskell-extra], which can then be added to the collection of haskell-providing repositories.


Tango-view-refresh-red.pngThis article or section is out of date.Tango-view-refresh-red.png

Reason: this and remaining sections may be out of date given the move to distributed repositories (Discuss in Talk:Haskell package guidelines#)

In almost all cases, cabalised Haskell packages can be automatically translated into Arch packages, via the cabal2arch tool. It is strongly recommended that you use the latest released version of this tool, as it implements the packaging policy for Haskell packages. You can get it in several ways:

  • Add the [haskell] repository to /etc/pacman.conf and use pacman to install the latest release.
  • Download and build the cabal2archAUR package from the AUR.
  • Install directly from Hackage using cabal install cabal2arch.

cabal2arch: an example

This example illustrates how to create a new package with cabal2arch. We will make a new package for the document formatter Pandoc:

First, set the name and email address to be used in the generated PKGBUILD:

export ARCH_HASKELL='My Name <my.name@domain.org>'

Second, find the hackage page for Pandoc, then identify the link to the .cabal file. Use this link as an argument to cabal2arch:

% cd /tmp
% cabal2arch http://hackage.haskell.org/packages/archive/pandoc/
Using /tmp/tmp.D7HAJJx2js/pandoc.cabal
Feeding the PKGBUILD to `makepkg -g`...
==> Retrieving Sources...
  -> Downloading pandoc-
--2011-05-14 07:25:39--  http://hackage.haskell.org/packages/archive/pandoc/
Resolving hackage.haskell.org...
Connecting to hackage.haskell.org||:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 355477 (347K) [application/x-tar]
Saving to: “pandoc-”

     0K .......... .......... .......... .......... .......... 14%  210K 1s
    50K .......... .......... .......... .......... .......... 28%  393K 1s
   100K .......... .......... .......... .......... .......... 43%  338K 1s
   150K .......... .......... .......... .......... .......... 57%  419K 0s
   200K .......... .......... .......... .......... .......... 72%  404K 0s
   250K .......... .......... .......... .......... .......... 86%  554K 0s
   300K .......... .......... .......... .......... .......   100%  506K=0.9s

2011-05-14 07:25:40 (369 KB/s) - “pandoc-” saved [355477/355477]

==> Generating checksums for source files...

Checking what was created:

% ls
% cd haskell-pandoc
% ls
haskell-pandoc.install	PKGBUILD

You can now inspect the PKGBUILD and install script for the library:

# Maintainer: 
pkgdesc="Conversion between markup formats"
arch=('i686' 'x86_64')
depends=('ghc' 'haskell-http=4000.1.1' 'haskell-bytestring=' 'haskell-containers=' 'haskell-directory=' 'haskell-extensible-exceptions=' 'haskell-filepath=' 'haskell-mtl=' 'haskell-network=' 'haskell-old-time=' 'haskell-parsec=3.1.1' 'haskell-pretty=' 'haskell-process=' 'haskell-random=' 'haskell-syb=0.3' 'haskell-texmath<0.5' 'haskell-utf8-string>=0.3' 'haskell-xhtml=3000.2.0.1' 'haskell-xml<1.4' 'haskell-zip-archive<0.2')
build() {
    cd ${srcdir}/${_hkgname}-${pkgver}
    runhaskell Setup configure -O ${PKGBUILD_HASKELL_ENABLE_PROFILING:+-p } --enable-split-objs --enable-shared \
       --prefix=/usr --docdir=/usr/share/doc/${pkgname} --libsubdir=\$compiler/site-local/\$pkgid
    runhaskell Setup build
    runhaskell Setup haddock
    runhaskell Setup register   --gen-script
    runhaskell Setup unregister --gen-script
    sed -i -r -e "s|ghc-pkg.*unregister[^ ]* |&'--force' |" unregister.sh
package() {
    cd ${srcdir}/${_hkgname}-${pkgver}
    install -D -m744 register.sh   ${pkgdir}/usr/share/haskell/${pkgname}/register.sh
    install    -m744 unregister.sh ${pkgdir}/usr/share/haskell/${pkgname}/unregister.sh
    install -d -m755 ${pkgdir}/usr/share/doc/ghc/html/libraries
    ln -s /usr/share/doc/${pkgname}/html ${pkgdir}/usr/share/doc/ghc/html/libraries/${_hkgname}
    runhaskell Setup copy --destdir=${pkgdir}

It follows the conventions for Haskell packages:

  • Libraries are prefixed with haskell-
  • All libraries that the package depend on are listed (libraries shipped with GHC are dealt with by having the ghc package provide them)
  • It uses cabal to generate a post-install register/unregister script, with a standard name.
  • We use haddock to build the documentation.

All Haskell libraries should follow these naming conventions, and using the latest release of cabal2arch will ensure this is the case.

Note: Beginning with cabal2archAUR 1.1-2, a new environment variable, PKGBUILD_HASKELL_ENABLE_PROFILING, is generated into the PKGBUILD. If this variable is of non-zero length, such as "1" or "true", then profiling builds will occur. Thus, if a user desires profiling, then it is advised to export this environment variable in a file such as ~/.bashrc or ~/.zshrc.

Guidelines for Libraries

In general, each .cabal file should map to one PKGBUILD. The following conventions hold:

  • libraries have their cabal names prefixed with haskell-
  • all libraries have a dependency on ghc
  • all libraries that are depended on must be listed in the depends array in the PKGBUILD
  • be careful about dependencies from gtk2hs: cairo, svg, glib, gtk. These are all provided by the gtk2hs package, not , e.g. "haskell-cairo"

Registering Haskell libraries is done via a register hook, see above.

Guidelines for Programs

  • Have their normal name. Examples: hmp3, xmonad, ghc, cabal-install
  • Be careful about dynamically linked run-time dependencies on C. For example, all GHC-produced binaries have a run-time dependency on 'gmp'. OpenGL or GtT-based binaries will have additional 'depends'. cabal2arch will attempt to work out the C dependencies, but there may be others implied by Haskell dependencies that are missed.
  • Use executable stripping, --enable-executable-stripping. cabal2arch will do this automatically.

Haskell Build Order

  • haskell-x11 update: haskell-x11-xft -> xmonad -> xmonad-contrib