[arch-haskell] What version of cabal2arch should be used to build HABS?

Peter Simons simons at cryp.to
Sat Oct 12 07:37:59 UTC 2013


Hi Magnus,

 > Are [cabal2arch and archlinux] in a state where new versions can be
 > released right now?

only recently, Remy has added the "platform-provides.txt" feature to
these tools, which constitutes a major change in behavior. It's best
illustrated by looking at the diff that results from re-generating the
hasktags package with the new tools:

  --- old/hasktags/PKGBUILD
  +++ new/hasktags/PKGBUILD
  @@ -7,7 +7,7 @@ pkgdesc="Produces ctags \"tags\" and etags \"TAGS\" files for Haskell programs"
   url="http://hackage.haskell.org/package/${_hkgname}"
   license=('custom:BSD3')
   arch=('i686' 'x86_64')
  -makedepends=('ghc' 'haskell-bytestring' 'haskell-haskell98')
  +makedepends=('ghc' 'haskell-bytestring=0.9.1.7' 'haskell-haskell98=1.0.1.1')
   depends=('gmp')
   options=('strip')
   source=(http://hackage.haskell.org/packages/archive/${_hkgname}/${pkgver}/${_hkgname}-${pkgver}.tar.gz)

Basically, there is a list of packages that we consider to be part of a
stable platform:

  https://github.com/archhaskell/cabal2arch/blob/master/data/platform-provides.txt

Every package that's part of this list will have its exact version
number hard-coded into every PKGBUILD that depends on it, meaning that
our builds expect haskell-haskell98=1.0.1.1, and no other version will
do. The packages in that list are those of Haskell Platform 2010.2 and
those provided by GHC 6.12.3.

The current HABS tree has been generated with tools that didn't have
that patch, so if we want to release new versions of these tools, then
we should either re-generate all of HABS, or we should release versions
prior to that extension.

Personally, I am a little scared by this feature, because the generated
files are really quite different from what we used to distribute before.
On the other hand, I can't think of a reason why these of ultra-explicit
dependencies might cause trouble, so I tend to defer to Remy's expertise
in this matter, who clearly though that this is a good idea.

Take care,
Peter





More information about the arch-haskell mailing list