Commits on Nov 2, 2010
  1. Revert change introduced accidentally in commit 4c9a0e3.

    (dependencies on base packages are explicitly listed again)
    Rémy Oudompheng committed Nov 2, 2010
Commits on Nov 1, 2010
  1. @peti

    Changed getRepoFromDir to ignore all directory names that begin with …

    …a dot.
    This change means that '.git' will be ignored as well as '.' and '..'.
    peti committed Nov 1, 2010
  2. @peti

    Drop inappropriate cabal2arch attribution.

    The archlinux library should not make claims about who was responsible
    for the generation of a PKGBUILD. It might have been cabal2arch, but it
    might have been something else. This code is generic: we want it to be
    re-used by other tools. Consequently, a "generated by" line should be
    inserted by the tool that uses the library.
    peti committed Nov 1, 2010
  3. @peti
  4. @peti

    Modified getPkgFromDir so that it doesn't lose information from parse…

    …r errors.
    Previously, the function return 'Nothing' in case of a parser error even
    though a more accurate error description was available. Instead, it now
    uses 'fail' to report that error.
    peti committed Nov 1, 2010
  5. @peti

    library-providers.txt: gtk+ refers to 'gtk2'

    This change fixes the build of haskell-gtk.
    peti committed Nov 1, 2010
Commits on Oct 31, 2010
  1. @peti

    drop obsolete comment about 'provides' field

    Our PKGBUILD files feature a comment saying that we provide all package
    dependencies and that the user's front-end to pacman must understand
    'provides'. In fact, this is not true: we no longer specify dependencies
    on packages that are provided by GHC. Instead of that comment, I feel we
    should provide a Maintainer: field, which -- according to namcap -- is
    currently missing.
    peti committed Oct 31, 2010
Commits on Oct 30, 2010
  1. Fix usage of deprecated VersionRange constructors.

    Rémy Oudompheng committed Oct 30, 2010
  2. @peti

    CabalTranslation.hs: dropped useless literal PKGBUILD file in comment…

    …s, probably a left-over from a debugging session
    peti committed Oct 30, 2010
  3. @peti

    Split the more generic version cabal2pkg' out of cabal2pkg.

    cabal2pkg includes a hard-coded decision about the $pkgname of the generated
    PKGBUILD file. The user, however, may want to override that name, so the
    desired package name should to be passed as a parameter. The same applies to
    the value of $pkgrel, which is currently hard-coded to "1".
    The more generic function cabal2pkg' accomplishes this.
    peti committed Oct 30, 2010
  4. @peti
Commits on Oct 29, 2010
  1. @peti

    PkgBuild: drop warning about arbitrary version intervals

    The archlinux library issues warnings while converting Cabal files that
    specify lower and upper bounds for dependencies, like "foo > 1 && < 2",
    because those restrictions cannot be expressed in a PKGBUILD file.
    Now, what are those warnings good for?
    That kind of version restriction is exceedingly common in Cabal files. An
    attempt to convert all of Hackage results in hundreds of warnings, yet
    it's not clear what anyone is supposed to do about them. There is just no
    way to convert those specifications so that Pacman can process them.
    Basically, we ensure that those restrictions are fulfilled long before a
    PKGBUILD is ever created: i.e. by means of the PKGLIST, which specifies
    the exact set of package versions that ought to be converted to ArchLinux.
    peti committed Oct 29, 2010
  2. @peti
  3. @peti

    fixed GHC warnings

    peti committed Oct 29, 2010
  4. @peti
  5. @peti

    reverse_deps.hs: don't print the packages that were requested as thei…

    …r own dependencies
    peti committed Oct 29, 2010
  6. @peti
  7. @peti

    reverse_deps.hs: expect path to ABS repository as first command-line …

    The script no longer has to run in the ABS tree. Also, if the tree cannot be
    read, the binary throws an error (instead of returning "0" silently).
    peti committed Oct 29, 2010
  8. @peti

    SrcRepo.hs: getRepoFromDir shouldn't fail silently

    All PKGBUILD files in the habs tree must to be valid, and if they are
    not, then we want to know!
    peti committed Oct 29, 2010
  9. @peti
  10. @peti
  11. @peti

    CabalTranslation.hs: patch "--force" flag into the generated unregist…

    … script
    If package A depends on B, and B is updated (say because of pkgrel bump) then
    Pacman un-registers the current version of B before registering the new
    version. At this point, ghc-pkg detects that un-registering B would break A,
    and aborts the procedure. The update works anyway, because our generated shell
    script script loses the error code, but we can avoid that error altogether by
    giving ghc-pkg the flag '--force'. Unfortunately, there is no way to tell Cabal
    that --force should be set in the generated "" script, so we have
    to patch that flag into the command line using mad sed(1) magic.
    With that change, an update of a package that's a prerequisite for another one
    will look like this:
     | [...]
     | (1/1) checking for file conflicts
     | unregistering tar- would break the following packages: archlinux-0.3.4 (ignoring)
     | (1/1) upgrading haskell-tar
     | [...]
    Basically, we tell ghc-pkg to shut up and trust that Pacman does the right
    peti committed Oct 29, 2010
  12. @peti

    CabalTranslation.hs: updated generated shell code to take advantage o…

    …f the ${_hkgname} variable
    peti committed Oct 29, 2010
  13. @peti
Commits on Oct 28, 2010
  1. Remove obsolete syntax from install file.

    Rémy Oudompheng committed Oct 28, 2010
Commits on Oct 27, 2010
  1. @peti

    Distribution/ArchLinux/CabalTranslation.hs: build executables with op…

    …timization enabled
    peti committed Oct 26, 2010
Commits on Oct 26, 2010
  1. CabalTranslation: relax constraints on pkglist syntax to allow blank

    lines and comments (with "# ")
    Rémy Oudompheng committed Oct 26, 2010
  2. @peti

    Distribution/ArchLinux/CabalTranslation.hs: removed configure flag --…

    …enable-shared again
    Shared library builds work, alright, but the compiled executables don't
    seem to take advantage of the dynamic objects. Until that's been figured
    out, there's little point in building libraries with 'dyn' enabled. It
    just takes twice as long.
    peti committed Oct 25, 2010
  3. @peti
  4. @peti
Commits on Oct 24, 2010
  1. @peti
  2. @peti

    getSpecifiedCabalsFromTarball: fail fatally if a package descriptor d…

    …oesn't parse
    Previously, Debug.Trace.trace was used to write out a warning, but in
    the plethora of output cabal2arch generates this warnings tends to be
    missed. Now, such an error triggers an exception.
    peti committed Oct 24, 2010
  3. @peti
  4. @peti
Commits on Oct 23, 2010
  1. @peti