Implement tagToEnum# for the bytecode system.  Blargh.  We spot tail-calls
   tagToEnum# <type> arg
and emit code to push the arg, then do a bytecode test-sequence to
determine what value it is, push the relevant constructor, and merge
control flow again, at a label which does the normal tail-call
sequence: slide the constructor down to the sequel and enter it.

Blargyle, or as some would say, barferama.

RTS support for the ugly tagToEnum# hack.  Actually a very general
thing -- just a bytecode unconditional jump, so we can do more general
control-flow in BCOs.

SuSE's Version 7.1 once again broke DocBook, so as a last resort we
use a home-made CATALOG.   >:-(

play quoting games to hopefully get this to build with both ghc-4.08.2
and ghc-4.11+.

- add index terms
- elaborate the description of the 'it' variable

markup fix.

Split the primitives section out into a separate file, and add a
paragraph discouraging its use.

- update documentation for SPECIALISE pragma
- change upper-case markup to lower-case

put the GHCi chapter before the "using GHC" chapter.

update (no -optCrts -Rgc-timing, -no-link-chk)

update (no more -optCrts)

remove unnecessary parens

Document new profiling options.

-------------------------------------
	Be less keen about inlining primops
	-------------------------------------

Marcin pointed out that it's bad to take

	let x = a +# (b *# c)
	in
	x + f x + g x

and inline x to get

	(a +# (b *# c)) +
	f (a +# (b *# c)) +
	g (a +# (b *# c))

Yet that's just what we were doing.  The offending clause is
in PrimOp.primOpIsCheap, which regards some primop applications
as 'cheap' (and hence freely duplicable).

For now, I've made it return False.  Let's see what that does.

Fix orphan calculation (again)

add note about installing DocBook toolchain on FreeBSD

a couple of micro-optimisations

fix and elaborate the documentation for <literal>extra_libraries</literal>

GHCi test update

put the NUL terminator in the right place in stringToStringBuffer.

Don't create an installation-time Makefile (target binary-dist).

Basic support for the simplified installation story.

Comments only

---------------------------------------------
	Be more eager to inline for an applied coerce
	---------------------------------------------

Given this:

	(coerce T f) x y

SimplUtils.interestingCallContext was returning False (uninteresting
call context).  This is wrong: it's a very intersting context.  

Darn those coerces!  They are scheduled for destruction, but meanwhile,
this makes matters a bit better.

-------------------------------------------------------
	Be more careful about floating out from  INLINE pragmas
	-------------------------------------------------------

Given this:

	x = __inline__ (f (g y))

we were floating the (g y) out as a MFE, thus:

	lvl = g y
	x = __inline__ (f lvl)

This is bad.  The (g y) redex gets outside the __inline__ envelope, 
and may never get inlined. 

The solution involved a bit of fiddling in SetLevels.

Fix a problem with the 'it' variable in GHCi.  New bindings for 'it'
were getting confused with old bindings, because we always used the
same 'it' name.

Now, we generate a new unique for 'it' each time around.

Also, make sure that any existing variables shadowed by new
command-line bindings are correctly removed from the environments to
avoid space leaks.

------------------------------------
	Be more gung-ho about INLINE pragmas
	------------------------------------

When we see 

	{-# INLINE f #-}
	f = g $ h

should we inline f inside a lambda?  Previously we said 'no'
because f looks like a redex, but that's a great mistake sometimes.
So this commit is more gung-ho: if you say INLINE that's what you get.

The changes are to exprIsDupable and exprIsCheap.

exprIsAtom shouldn't return True for an SCC.

-------------------------------
PS to:	Improve orphan-module resolution
	-------------------------------

{This comment got missed out of the previous commit.}

I also moved mkFinalIface from HscMain to MkIface, where it is much tidier.
(This is related, because it's mkFinalIface that now sets mi_orphan.)

Reorder the calls to
	tcSimplifyTop
and
	tcCheckMain

in TcModule.  We must to tcCheckMain first, else we get a bogus 
complaint when compiling

	main = return ()

(Something about ambiguous monad.)

-------------------------------
	Improve orphan-module resolution
	-------------------------------

Consider the following rule (and there are lots of these in
the Prelude):

	fromIntegral T = fromIntegral_T

where T is defined in the module being compiled.

is an orphan.  Of course it isn't, an declaring it an orphan would
make the whole module an orphan module, which is bad.

This commit arranges to determine orphan rules, and the orphan-hood
of a module, much later than before.  (Before mi_orphan was set by
the renamer, now it is set by MkIface.)

Update NOTES

PutFullMVar went away

Use strerror instead of our own errno descriptions.

Re-instate filtering of the CAF refs for recursive bindings.  This may
be the cause of GC being real slow on a bootstrapped compiler right now.

prelude/PrimOp_HC_OPTS += -H80m at least until we figure out why
ghc-5.0 is leaking space so badly.

remove 4 old primops

wibbles

add -package lang

remove bogus import