GHC.loadModule compiles a module after it has been parsed and
typechecked explicity. If we are compiling to object code and there is
a valid object file already on disk, then we can skip the compilation
step. This is useful in Haddock, when processing a package that uses
Template Haskell and hence needs actual compilation, and the package
has already been compiled.

As usual, the recomp avoidance can be disabled with -fforce-recomp.

Partly this is cleaner as we only have to preprocess the source files
once, but also it is necessary to avoid Haddock recompiling source
files when Template Haskell is in use, saving some time in validate
and fixing a problem whereby when HADDOCK_DOCS=YES, make always
re-haddocks the DPH packages.  This also needs an additional fix to
GHC.

HsColour support still uses Cabal, and hence preprocesses the source
files again. We could move this into the build system too, but there
is a version dependency that would mean adding extra autoconf stuff.

It was making incorrect URLs due to a shell script error.

This is an implementation matter really (the key word is "apparently"!).  
See Note [Specialising in no-dict case] in DsBinds.

It showed up when compiling GHC.Float.

More on Trac #597

Fixes some sub-items of Trac #597

The -fspec-inline-join-point thing was a gross hack intended to help
Roman play around, but he's not using it and it was a terribly blunt
instrument so I've nuked it.  

Instead I've re-instated the let-binding specialiser. 
See Note [Local let bindings]

It now says something like

  SpecConstr
    Function `happyDoAction'
      has 17 call pattterns, but the limit is 3
    Use -fspec-constr-count=n to set the bound
    Use -dppr-debug to see specialisations

Previously it just silently did no specialisation

So that you can write

   f (v1 -> v2 -> pat)

See the long Note [INLINE and default methods].  

This patch changes a couple of data types, with a knock-on effect on
the format of interface files.  A lot of files get touched, but is a
relatively minor change.  The main tiresome bit is the extra plumbing
to communicate default methods between the type checker and the
desugarer.

From Maxime Henrion <mhenrion@gmail.com>

The main idea is that I'm now treating a single-method dictionary very
much like a multi-method dictionary.  In particular, it respond to
exprIsConApp_maybe, even though newtypes aren't *really* proper
constructors.

See long comments with Note [Single-method classes] for why
this slight hack is justified.

If we see

   foreign export ccall foo :: ...blah...

we now use the "foreign" to suggest -XForeignFunctionInterface

This patch mainly concerns the plumbing for running
the passes and printing intermediate output

* I was debugging so I added some call-site info
  (that touches a lot of code)

* I used substExpr a bit less in Simplify, hoping to
  make the simplifier a little faster and cleaner

I was experimenting with making coercions opaque to
arity.  I think this is ultimately the right thing to do
but I've left the functionality unchanged for now.

See Note [INLINE for small functions]

  class C a b
  instance Int `C` Bool

This was accidentally disallowed before.

The Stg Lint failure reported in Trac #3789 were bogus.
This patch fixes STG Lint, which must have been unused
for ages.