Remove more usage-analysis stuff

Wibble to TcRules (fixes HEAD breakage)

-------------------------------------
  Fix parsing of floating-point constants in External Core
	-------------------------------------

This fix accidentally made it into the previous (unrelated) commit,
so it's really the *previous* change to LexCore you should look
at.

The fix updates LexCore so that it can parse literals in scientific
notation (e.g. 4.3e-3)

Type sigs and comments only

Remove usageSP directory

-------------------------------------
	Fix a name-capture bug in Ext-Core
	-------------------------------------

Don't expand newtypes (even non-recursive ones) when going to External Core.
Reason: the expansion was performed *after* Tidying; the expansion performs
type substitution, which is only done right if you take account of the Uniques.
But since it's post-tidying, we got capture of occurence names.

I hope the lack of newtype expansion doesn't hurt anyone; I doubt it will.
If so, we can think again.

Thanks to Tobias Gedell for this one.

-------------------------------------
	Remove all vestiges of usage analysis
	-------------------------------------

This commit removes a large blob of usage-analysis-related code, almost
all of which was commented out.

Sadly, it doesn't look as if Keith is going to have enough time to polish it
up, and in any case the actual performance benefits (so far as we can measure
them) turned out to be pretty modest (a few percent).

So, with regret, I'm chopping it all out.  It's still there in the repository
if anyone wants go hack on it.  And Tobias Gedell at Chalmers is implementing
a different analysis, via External Core.

---------------------------------------------------
			External Core fix
	output implicit bindings in correct dependency order
	---------------------------------------------------

In coreSyn/MkExternalCore, output constructor wrappers before the
other implicit bindings, because the latter may use the former.

Thanks to Tobias Gedell for this one.

Make utils/genprimopcode recognise the type ().
    It was previously written 'Unit', which is easily
    confused with the type 'Unit' (used for generic
    derived instances).

---------------------------------------------------
	Template-Haskell fix to make the global environment
		      more side-effect-ful
	---------------------------------------------------

Consider

	f = $(...foldl...) $(...foldl...)

The first splice sucks in the type sig for foldl, which the second
splice needs.  That means that the second splice is going to have to
consult the persistent compiler state to see the effect of imports
by the first one.

We used to cache the global type environment in the TcGblEnv, but
this commit switches to the obvious thing: consult the persistent
state on every global lookup.  After all, reading a MutVar is no
big deal; and it's a benign, ever-growing cache of type signatures,
so the side effect is fine.

On the way I tidied up the knot-tying in TcIfaceSig a bit more.
Previously, I think the setUnfoldingInfo was being strict in the
unfolding, which forced it to be type-checked.  Now it's lazy.
That could mean a lot less typechecking overall, for things whose
unfolding isn't looked at.  I hope I havn't broken it, though.

---------------------------------------------------
	Important fix to the handling of class methods that
	      mention their own class type variable
	---------------------------------------------------

[NB: I'm not 100% certain that this commit is independent of the
     Template-Haskell-related commit I'm doing at the same time.
     I've tried to separate them but may not have succeeded totally.]

This bug gives utterly bogus (detected by Core Lint) programs.
Isaac Jones discovered it.  Here's an example, now enshrined as tc165.

    class C a where
	f :: (Eq a) => a

    instance C () where
	f = f

The instance decl was translated as

    dfC() = MkC (let f = \dEq -> f in f)

which is utterly wrong.  Reason: the 'f' on the left was being treated
as an available Inst, but it doesn't obey INVARIANT 2 for Insts, which
is that they are applied to all their dictionaries.  (See the data type
decl for Inst.)

Solution: don't include such class methods in the available Insts.

Use nameIsLocalOrFrom instead of open code

Add type sig

Modify error message

Initialize stp->n_to_blocks as 0.  Add function for MinGW32 in Signals.c.

PowerPC bugfix: Don't forget to save GPR13
(Thanks to Ashley Yakeley for reporting this)

MERGE TO STABLE

Quick fix for profiling after I broke it in rev. 1.68.

The profiling system needs some initialisation after module
registration (hs_add_root()).  Since module registration now happens
later, the profiling system was getting initialised too early before
any cost centers had been registered.  This is a quick fix; we still
can't handle multiple hs_add_root()s with profiling.

gcc 2.96 miscompiles LDV_recordDead_FILL_SLOP_DYNAMIC() when DEBUG is
on, and I think I must have debugged this crash about three times
already.  So I'm inserting a #error to stop me doing it again.

Multi-init protection.

Multiple inits now don't crash, but they still don't do anything
sensible because the finalizers have been run during the first
hs_exit().

- re-instate setProgArgv, it is used in System.Environment (bah, could
  have sworn I grepped for it and found nothing...)

- Remove init_stack symbol from the Linker's symbol table; this is
  now static.

I moved startTime here, from RtsStartup.c where it seemed out of placed

Make it multi-init-safe

- Remove setProgArgv
- Belatedly add startupHaskell/shutdownHaskell

Make it multi-init-safe

Update the FFI docs to use hs_init()/hs_exit() rather than
startupHaskell()/shutdownHaskell().

Flesh out support for hs_init() and hs_exit() according to the latest
FFI spec.

For GHC, I also added:

  hs_add_root( void (*fn)(void) );

which is used to specify the root module.  This *must* be called prior
to invoking any Haskell functions.

The previous way of doing things still works:

  startupHaskell( argc, argv, root );

but the right way to do this is now

  hs_init( &argc, &argv );
  hs_add_root( root );

It is possible to invoke hs_add_root() multiple times with different
roots.

- setProgArgv() has been removed; it was unused and looks like it was
  there to support STG Hugs.

Initialise prog_argc, prog_argv and rts_argc to sensible zero values
just in case hs_init() is called without passing any arguments in.

- Add headers
- Make initThreadLabels multi-init-safe

Add #ifdef around an assertion that only holds for the threaded RTS.

Fixes to make previous revision compatible with autoconf 2.13.

Un-break the non-threaded RTS
Sorry :-/

Synched OpenGL detection macro with HOpenGL-1.04.
TODO: Modularization of the macro and caching.

This commit fixes many bugs and limitations in the threaded RTS.
There are still some issues remaining, though.

The following bugs should have been fixed:

- [+] "safe" calls could cause crashes
- [+] yieldToReturningWorker/grabReturnCapability
    -     It used to deadlock.
- [+] couldn't wake blocked workers
    -     Calls into the RTS could go unanswered for a long time, and
          that includes ordinary callbacks in some circumstances.
- [+] couldn't block on an MVar and expect to be woken up by a signal
      handler
    -     Depending on the exact situation, the RTS shut down or
          blocked forever and ignored the signal.
- [+] The locking scheme in RtsAPI.c didn't work
- [+] run_thread label in wrong place (schedule())
- [+] Deadlock in GHC.Handle
    -     if a signal arrived at the wrong time, an mvar was never
          filled again
- [+] Signals delivered to the "wrong" thread were ignored or handled
      too late.

Issues:
*) If GC can move TSO objects (I don't know - can it?), then ghci
will occasionally crash when calling foreign functions, because the
parameters are stored on the TSO stack.

*) There is still a race condition lurking in the code
(both threaded and non-threaded RTS are affected):
If a signal arrives after the check for pending signals in
schedule(), but before the call to select() in awaitEvent(),
select() will be called anyway. The signal handler will be
executed much later than expected.

*) For Win32, GHC doesn't yet support non-blocking IO, so while a
thread is waiting for IO, no call-ins can happen. If the RTS is
blocked in awaitEvent, it uses a polling loop on Win32, so call-ins
should work (although the polling loop looks ugly).

*) Deadlock detection is disabled for the threaded rts, because I
don't know how to do it properly in the presence of foreign call-ins
from foreign threads.
This causes the tests conc031, conc033 and conc034 to fail.

*) "safe" is currently treated as "threadsafe". Implementing "safe" in
a way that blocks other Haskell threads is more difficult than was
thought at first. I think it could be done with a few additional lines
of code, but personally, I'm strongly in favour of abolishing the
distinction.

*) Running finalizers at program termination is inefficient - there
are two OS threads passing messages back and forth for every finalizer
that is run. Also (just as in the non-threaded case) the finalizers
are run in parallel to any remaining haskell threads and to any
foreign call-ins that might still happen.

remove setting of $(way_) and $(_way); these are now set in boilerplate.mk

- Generalise seq to allow an unlifted type in its second argument.  This
  works because seq is *always* inlined and replaced by a case.

- Remove getTag, a wired-in Id with an unfolding, with a definition
  in GHC.Base:

	getTag x = x `seq` dataToTag# x

  this is why we required the above generalisation to seq (dataToTag#
  returns an Int#).  See the comments in GHC.Base for more details.

- As a side-effect, this fixes a bug in the interpreter, where the
  compiler optimised away the evaluation of the argument to dataToTag#,
  but the interpreter ended up passing it an unevaluated thunk (nullary
  constructors aren't always evaluated in GHCi, but the simplifier
  assumes they are).  Now, in the interpreter, getTag won't be inlined
  so the compiler can't optimise away the evaluation, and we're saved.

  The real bug here is either (a) dataToTag# requires an evaluated
  argument or (b) the interpreter doesn't supply it with one, take your
  pick.

- Reverse the code for workers and wrappers for nullary constructors.
  For some reason it was the wrong way around, but the effects were
  harmless since they both evaluate to the same thing.

- When passing a nullary constructor as an argument, we should pass
  the name of the worker rather than the wrapper.  Again, this is
  mostly harmless, but it enables some small simplification in
  pushAtom.

- Rearrange/cleanup pushAtom.

Perform 'tidying' on the implicit bindings before emitting
	External Core.  We were getting silly bindings like
		\ tpl -> case tpl of tpl -> (tpl,tpl) -> tpl

	Maybe we should add these implicit bindings in CoreTidy,
	rather than in both MkExternalCore and CorePrep?

Fix compilation with DEBUG

We have to desugar polymorphic FFI declarations properly, too.
(companion commit to rev. 1.84 of TcType.lhs).

Allow FFI declarations to have polymorphic types. eg:

foreign import ccall unsafe "stdlib.h &free" pFree :: FunPtr (Ptr a -> IO ())

(the change actually removed a line, the extra lines are because I
reformatted an import declaration).