In case we've got a compiled Setup binary

Part of GHC trac #3968.

We do not ever want to suggest the use of the BSD4 license.
This will fix a QA check message and the 'cabal init' command.
This change should be safe to push into the stable 1.8 branch.

The MIN_VERSION_<package> macros are useful to test whether we have at
least a given version of a dependency, but we can't extract the actual
version of dependency using this macro. This patch makes the version
of each dependency available to the code of the cabal package, which
can be useful when constructing global names in Template Haskell, for
instance.

So we now both check if the headers can be pre-processed and separately
if they can be compiled. If it fails the pre-processing check then it
is usually (but not always) a missing header. If it fails to compile
then it is definately borked, not missing. Error messages reflect this.

Addresses the issue in tickets #532, #648.
When the all-together test of C libraries and headers fails, use only the
preprocessor to probe the headers individually.  If they all check out,
give a more equivocal error message and suggest -v3.

Resolves ticket #344.

  - ticket #327: require c2hs version 0.15 as minimum
  - ticket #536: pass c2hs include dirs of dependent packages
  - ticket #537: pass c2hs location of gcc -E

Avoid pattern match failure in obscure circumstances
Should fix ticket #383

Probably needed in the case of foreign exports and TH.

 * Not changed for cpp, just code refactored.
 * For c2hs add cpp-options and the sysdefines (-D${os}_BUILD_OS etc)
 * For hsc2hs add the sysdefines

Plan is to switch over from docbook xml.

When using "haddock --optghc-XTemplateHaskell" haddock will write out .o
and .hi files. This is bad because it replaces the ones we previously
built. This results in broken packages later on. Of course haddock
should not do this, it should write temp files elsewhere. The workaround
is to tell haddock to write the files to a temp dir.

This is for the case of intra-package deps where the lib has to be
registered into a local package db. We use "-inplace" suffix for
the local installed package ids (rather than using an ABI hash).

Some ar.exe versions floating around seem to have weird non-posix behaviour.

It does not work as intended and gives inconsistent results between
cabal install and cabal configure. The problem with the approach was
that we were filtering out the dependencies of non-buildable components
at the end. But that does not help much since if one of the deps of
the non-buildable component were not available then we would have
failed earlier with a constraint failure. A proper solution would have
to tackle it from the beginning, not just as a filter at the end.
The meaning of build-depends and buildable: False needs more thought.

It's been deprecated since Cabal 1.1.4 (over 3 years).
The replacement since then has been --destdir=

In the case that the progsdir is not relative to the $prefix

So we can stop pretending that "main-is: foo.hs" is a module name.
Also allows us to deprecate ModuleName.simple which bypasses the
ModuleName type invariant.

Now this like "#if !MIN_VERSION_base(4,2,0)" will work

This is slightly experimental, we'll see how it goes. See ticket #466.

No longer returns Maybe, we can find installed packages for all
the compilers we support (and this feature is a requirement for
support for new compilers).
This is an API change so cannot merge to Cabal-1.8.x branch.

That is, work out which packages are installed for hugs and nhc98.
In both cases there is special support for the core packages.
In future both should use the standard method when they supply
proper installed package info files for the bundled libraries.

This is really hard and rather nasty.