Instead of the ghc-stage and ghc-stage2-package files in a package, we
now have a list of these in ghc.mk.  There are other similar lists (of
boot-packages and non-installable packages), so this is not too bad,
and is simpler.

While poking around in the top-level ghc.mk file I spotted various
opportunities to clean up and re-order some of the cruft that has
accumulated over time.

The point here is that SCCs get in the way of eta
expansion and we must treat them uniformly.

When this ASSERT tripped in CoreToStg it tried to print out
too much, which tripped the asssertion again.  Result: an
infinite loop with no output at all.  Hard to debug!

This is what was giving the "absent entered" messages
See Note [Absent errors] in WwLib.  We now return a 
suitable literal for absent values of unlifted type.

This makes
     \(x:xs) -> e
want when you have -fwarn-incomplete-patterns, which is consistent.

The problem arose with this kind of thing

   x = (,) (scc "blah" Nothing)

Then 'x' is marked NoCafRefs by CoreTidy, becuase it has 
arity 1, and doesn't mention any caffy things.

That in turns means that CorePrep must not float out the
sat binding to give

  sat = scc "blah" Nothing
  x = (,) sat

Rather we must generate

  x = \eta. let sat = scc "blah" Nothing 
            in (,) sat eta

URGH! This Caf stuff is such a mess.

Thorikil ran into this when doing a PPC OS X build. We now also don't
use -m32 on PPC/OSX, but I don't think it should be necessary. We can
add it back if it does turn out to be.

saying

  make TRACE=1

prints most of the macro calls and their arguments.  It's easy to
trace new macros; see rules/trace.mk.

Also add some comments about what extra-packages is doing

The FFI GHCi import lib isn't needed as
compiler/ghci/Linker.lhs + rts/Linker.c link the
interpreted references to FFI to the compiled FFI.
We therefore filter it out so that we don't get
duplicate symbol errors.

The new flag prints out a warning if you have a local,
polymorphic binding that lacks a type signature. It's meant
to help with the transition to the new typechecker, which
discourages local let-generalisation.

At the same time I moved the missing-signature code to TcHsSyn,
where it takes place as part of zonking.  That way the 
types are reported after all typechecking is complete,
thereby fixing Trac #3696.  (It's even more important for
local bindings, which is why I made the change.)

and remove the temporary DOpt class workaround.

We need to directly depend on the stage1 libs. The stage1 compiler lib
doesn't depend on them.

rather than being repeated in the build system