GitLab

This required changing the PkgInfo type from an alias for GenericPackageDescription 
to a record with a Repo and a GenericPackageDescription.
ResolvedPackage and ResolvedDependency also had to be changed to
keep a Repo rather than a tarball URL.

Use the package-url field in the GenericPackageDescription to record the tarball url for each package.

This adds a dependency on zlib, to unzip the downloaded 00-index.tar.gz.
Once unzipped, a small pure haskell tar reader is used to read
the index contents.
Servers in the server list now have names which are used 
as local cache directory names.
The --with-server command line flag has been removed since it
did not allow specifying a server name.

This is very untested, but should make cabal-install able to resolve
dependencies again after the move to Cabal 1.2.0. The package descriptions
are stored in a very hacky format atm the moment.

Refactoring: add listInstalledPackages, and use that everywhere instead of getInstalledPackages with lots of arguments that all come from ConfigFlags.

Lemmih, Paolo Martini are the primary authors of cabal-install but were
missing from the list in LICENCE.

But it does not work yet as it need to be updated to take configurations into
account. At the moment it thinks no packages have any dependencies since
it does not finalise the configuration for the package.

Fixes a bug in my previous filepath patch

Follow directory layout of HackageDB data also when calculating URLs to download from, not just the local paths.

So it "Just Works"tm without needing extra options or sudo fiddling.
If you care you can still do sudo fiddling and --global-install

We can now install purely using cabal, no Makefile needed.
The install instrcutions previously said to install the data
file manually, but cabal does that for us now.

Fix the license, it's BSD3 not BSD4 (I double checked with Issac)
Have cabal install the serv.list for us as a data file.
Add HTTP and filepath as dependencies, rather than bundling them.

Enabled at -v2

So that at ordinary -v1 we do not get lots of silly warnings for outdated
.cabal files, like warnings about hs-src-dir vs hs-src-dirs.

This is relevant since we pass on this default verbosity when we call cabal,
so it's much better if they match up.

This should make it all "Just Work"tm. We no longer need a Makefile
to do the installation. The point is, don't bother trying to use
/etc/cabal-install/srve.list as the global list. That's hard to do
because it involves installing that file, and we only want to do that
for global installs, so it's even more complex. Instead we should just
use cabal's facility to install data files. That way it ends up in
/usr/local/share/cabal-install-x.y/
or somewhere similar for a user install. But either way, cabal handles it.
The per-user config sill overrides this one, so it's still possible to edit
the list even if /usr/local/share/ is read-only. Also, we always use
$HOME/.cabal-install/ for the per-user config and package cache.
The one downside is that if an admin wants to use /etc/ and /var rather
than /root/.cabal-install/ then this is a tad harder. But the point is
at the moment it'll at least work for most people without fiddling,
including windows users.