GitLab

Remove local copies. Also fixes a bug recently introduced
in the writeFileAtomic function, spotted by Peter Robinson.

The new cabal init stuff needs some work.

It works by ignoring the possibility that there could be multiple
installed packages sharing the same source package Id. We just pick
the "top most" one which is usually ok. We make no attempt to check
that we are using consistent installed packages.

ghc ticket #3602

We were doing it wrong, but Linux realpath() C function was letting
us get away with it. The Solaris realpath() is stricter.
The new implementation is also simpler, relying on the fact that
the canonicalizePath function will resolve symlinks.

Needs recent api changes.

Previously the --haddock-index=template flag controled both the
template used and whether it's used at all. When no path was set
then it was not used. The problem with that is that since we are
not enabling this feature by default then the default is blank.
That is the default config file would look like:
-- haddock-index:
which doesn't help anyone discover what it means or what a
sensible setting would be. By having a separate toggle to     
enable/disable we can have a default for the index file which
makes it easy to discover in the config file:
-- documentation: False
-- doc-index-file: $datadir/doc/index.html
All the user has to do is uncomment the first line and use True.

Also update the code somewhat following the changes in
the Cabal API for path templates and substitutions.

If the haddock-index flag is set it keeps an index 
of the haddock documentation of the packages in 
the global and user databases

Now supporting explicit --user or --global switches in bootstrap.sh with usage feedback for bad args
This change was adapted from work submitted to the cabal-devel mailing list by Jason Dusek.

Fix handling of base dir in tar file creation.

That is, use the "preferred-versions" mechanism when deciding
whether there is a new version available. This would allow us to
upload a new version without everyone immediately being told to
get it and try it out.

The code to do it was already there but we were checking for the
mime type text/plain using just (==) when in fact the server reports  
  text/plain; charset="ISO-8859-1"
so we have to parse the field a bit better (still a bit of a hack).

Fixes ticket #485. This means that for constraints like:
    build-depends: base >= 3 && < 5
we will pick version 4. However we will continue to apply the
version 3 preference for things like:
    build-depends: base >= 3
Where there is no upper bound on the version. Note that we now
also ignore preferences for base given on the command line.
We should implement #483 to split prefs from shims.

Make it clear that it's the specification of the package name that
is at fault rather than the package to which the name refers.

So checking for parsec 2 fails. We don't strictly need parsec, it's
just a dependency of network, so remove the check.

Part of ticket #369

Update to development version of Cabal