Back out yesterday's change - needs more thought.

All Simon's recent tuning changes.  Rough summary follows:

* Fix Kevin Atkinson's cant-find-instance bug.  Turns out that Rename.slurpSourceRefs
  needs to repeatedly call getImportedInstDecls, and then go back to slurping
  source-refs.  Comments with Rename.slurpSourceRefs.

* Add a case to Simplify.mkDupableAlt for the quite-common case where there's
  a very simple alternative, in which case there's no point in creating a
  join-point binding.

* Fix CoreUtils.exprOkForSpeculation so that it returns True of (==# a# b#).
  This lack meant that
	case ==# a# b# of { True -> x; False -> x }
  was not simplifying

* Make float-out dump bindings at the top of a function argument, as
  at the top of a let(rec) rhs.  See notes with FloatOut.floatRhs

* Make the ArgOf case of mkDupableAlt generate a OneShot lambda.
  This gave a noticeable boost to spectral/boyer2


* Reduce the number of coerces, using worker/wrapper stuff.
  The main idea is in WwLib.mkWWcoerce.  The gloss is that we must do
  the w/w split even for small non-recursive things.  See notes with
  WorkWrap.tryWw.

* This further complicated getWorkerId, so I finally bit the bullet and
  make the workerInfo field of the IdInfo work properly, including
  under substitutions.  Death to getWorkerId.  Kevin Glynn will be happy.

* Make all lambdas over realWorldStatePrimTy
  into one-shot lambdas.  This is a GROSS HACK.

* Also make the occurrence analyser aware of one-shot lambdas.

* Make various Prelude things into INLINE, so that foldr doesn't
  get inlined in their body, so that the caller gets the benefit
  of fusion.  Notably in PrelArr.lhs.

Assembler/Disassembler: handle and print calls to compiled code
Evaluator: return to scheduler when entering unknown closure
StgCRun: debugging trace in miniinterpreter (temporary)
Updates: fix normal and vectored returns to Hugs

Make vectored returns to Hugs work, and make IS_CODE_PTR etc work
for dynamically loaded objects.

Should really call OnExitHook() in shutdownHaskellAndExit().

Mods to enable interworking with simple compiled code.  Supports fns and
data decls.  Classes, instances, primops, don't work yet.
Unregisterised, mininterpreted x86-ELF is the supported object format.
GC appears to work correctly.

we're on to 4.04 now.

Mods to enable interworking with simple compiled code.  Supports fns and
data decls.  Classes, instances, primops, don't work yet.
Unregisterised, mininterpreted x86-ELF is the supported object format.
GC appears to work correctly.

Add a note about @raiseInThread@.

-ddump-rdr is now -ddump-parsed

only pop explicit layout contexts with '}'.

print the compiler version on the same line as the compilee version;
looks a bit tidier.

Header file for an alternative bignum implementation, which is in
interpreter/sainteger.c.  Currently used by STGHugs.
Temporary -- until we resolve GMP licensing issues.

Re-jig some stuff in here so it doesn't look like you need so many
tools to build GHC.

Added maxPrecedence

Added needed includes of SchedAPI.h

Whoops, add index terms too.

Add -ddump-most and -ddump-all, with the obvious meanings.

Redo previous commit to cut down on the use of COMPILING_RTS where
possible - SchedAPI.h is now an RTS internal header file which
RtsAPI.h no longer includes.

Don't include RtsAPI.h when compiling RTS.

Temporary workaround for problem which caused the following program

  main = putStrLn ("aa" ++ IOExts.trace "bb" "cc")

to deadlock - when the (untouched) stderr is evaluated (which
IOExts.trace forces), it will touch stdout (see code for details of
why), but it has already been locked by putStrLn. Boom - game over.

This temporary 'fix' is, to put it kindly, in the fancy-footwork
category as it doesn't solve the problem, but merely turns it on its
head. Instead of stderr depending on stdout, stdout now depends on
stderr, so the following program will deadlock

  main = hPutStrLn stderr ("aa" ++ myTrace "bb" "cc")

  myTrace msg v = unsafePerformIO $ do
     putStrLn msg
     return v

The 'theory' is that this is far less likely to occur in practice than
the other way around.

The next step / real solution would be to give up the lock on an
output Handle while filling up its output buffer. However, that
requires ripping out / re-org'ing a fair bit of buffer management
code, which I'll delay doing.

Remove -#include "RtsAPI.h" - you now get it for free.

Stg.h contains all the defs and protos that are visible when compiling
.hc code (==> .hs code). Augment this set with the externally visible
RTS API.

Tidied up the handling of the code that emits the precedence level
predicates that gets used in applications of showParen and readParen.

Two new Dark Corner regression tests:

- testing the derived Show&Read instances for constructors with
  a field label that is a varsym rather than a varid.
- check that the default fixity & assoc of an operator is 'infixl 9'.

New option, -fail, to indicate failure (whatever the return code).

Make sure that instance gates have their home modules
loaded by RnIfaces.getImportedInstDecls.  This was causing
Kevin Atkinson's missing-instance bug.

* If a field label is a 'varsym', wrap parens around it when
  Show'ing and Read'ing it back in.

* If there's no fixity decl for a 'consym', the default is
  for it to be left-assoc.

Don't use WinSock under cygwin

Don't compile in WinSock support under cygwin

exitWith: don't call exit(), but use the RTS' shutdownHaskellAndExit()

New RTS entry point, shutdownHaskellAndExit(), which does what the name
implies - used when you want to exit from within Haskell code (e.g.,
System.exitWith.)

Remove final fflush() - it seems to be incompatible with recent
releases of Linux's libc implementation, and we don't use stdio in any
case.

The allocator for mega groups now checks if consecutive megablocks on
the free list are contiguous in memory. The omission of this check
caused all kinds of funny runtime errors and took away at least five
happy years of my life...  :-{

8-bit clarity

Make this more Haskell 1.3-insensitive

Allow the following (not strictly H98) construct:

    do
	blah
	e $ do
	blah

This doesn't break any existing code, just allows more programs
through.  Only the semantics for layout after a 'do' have changed.

Add a test for deriving Show (f (f a)); makes Hugs hang

Make the r.u.n message happen only with -DDEBUG

Back out a change that creeped in by mistake.