Add missing case in parsing of inline pragmas.

raise_info, the info table we use for overwriting thunks with an
exception, had the bogus closure type of FUN.  This was mostly
harmless, but caused some confusion for THUNK_SELECTORs, which rightly
assumed that they shouldn't be pointing at FUNs.

The fix is to change the type to THUNK, which is slightly closer to
being true.

This fixes Alastair Reid's "THUNK_SELECTOR: strange selectee" bug.

Fixing up names to make a reasonable use of Java packages.
This is just one of a number of steps before attacking the Prelude.

Fixing up the output order of the inner classes, making postprocessing easier.

Fix compilation errors when OMIT_NATIVE_CODEGEN is on.

MERGE 4.07

* This fix cures the weird 'ifaceBinds' error that
  Sven and George tripped over.  It was quite obscure!

  Basically, there was a top level binding
	f = x
  lying around, which CoreToStg didn't like.  Why hadn't
  it been substituted away?  Because it had a NOINLINE
  pragma.  Why did it have a NOINLINE pragma?  Because
  it's an always-diverging function, so we never want to
  inline it.

Changed ref to building guide to a ref to the installation guide.

Added a few ids to stop weirdly-named web pages being built for sections
without ids.

Only report shadowed names in source files (merge with 4.07)

Tweaked title not to have   (which is simply omitted in HTML Title
meta-info), and upped version no. to 4.07.

Pls mrg.

Removed a bogus XRef.

Lots of ids added so that all HTML pages generated have sensible names.
Various other little tidies up.

Please merge.

Fix unused-import warning --- again!

MERGE 4.07

* When float outwards (full laziness) remember to
  switch off the demand flag.  Else we wrongly
  can transform
	\x -> let y __D = (...) in y+x
  into
	let y __D = (...)
	in \x -> y+x
  In the latter, y is not necessarily demanded;
  it depends whether the function is called.  We
  should switch off the demand flag.

  The fix is right at the bottom in SetLevels.subst_id_info

MERGE 4.07

* Another wibble on records

don't use underscores in type variable names

don't use underscores in type variable names

Don't try to use getclock on HPUX, gettimeofday is more standard in
any case.

Add ids to the book tags.

don't use underscores in type variable names

disable -fwarn-hi-shadowing again, because we duplicate include paths
in the driver too much.

Tried to clear up some backslash confusion under Windows by always using
forward slashes.

Please merge.

Adding a field to the Method constructor, to allow methods
to say what they might raise. This is needed to actually
compile generated code.

Also, the generated code now imports haskell.runtime.*

verbatim no longer supported; now in rts.tex

de-verbatimed (verbatim is no longer supported).

Don't use commas in an SLIT("..")

zap spurious semicolon wibble

Fix a Real Bug(TM) in the new code for detecting dead threads: we
weren't taking into account the case when a ThreadRelocated is found
on the all_threads list.

MERGE 4.07

* Fix records with polymorphic fields (broken by previous 'fix')
  As a tidy-up I also put a TyCon into a FieldLabel

* Fix a glitch with the result-type-sig change

Add tests for result type sigs

Fix bug in diet_hep_initialise to do with argc/argv handling.

*** MERGE WITH 4.07 (once I've checked it works) ***

* Fix result type signatures.  Note that a consequential change is that
  an ordinary binding with a variable on the left
	f = e
  is now treated as a FunMonoBind, not a PatMonoBind.  This makes
  a few things a bit simpler (eg rnMethodBinds)

* Fix warnings for unused imports.  This meant moving where provenances
  are improved in RnNames.  Move mkExportAvails from RnEnv to RnNames.

* Print module names right (small change in Module.lhs and Rename.lhs)

* Remove a few unused bindings
  
* Add a little hack to let us print info about join points that turn
  out not to be let-no-escaped.  The idea is to call them "$j" and report
  any such variables that are not let-no-escaped.

* Some small things aiming towards -ddump-types (harmless but incomplete)

zu => _ (i.e. sync name decoding with OccName :-)

Synch encoding, decoding, and comments.

Re-instate foreign label and foreign export dynamic support in the NCG
(which both end up emitting a CLitLit into the abstract C) using a new
mkForeignLabel interface to CLabel.

This won't work if the foreign label is in a different DLL yet, but
Julian is on the case.

Z-encode underscores as "zu" (the previous owner of "zu", namely '^',
has been moved to "zc").

Improved NCG note.
Pls mrg

Add test for record selector functions where the field has a
polymorphic type.

Add -fwarn-hi-shadowing to the default set of warnings.

Complain about unrecognised warning options.

Batch finalizers on a per-GC basis.  That is, after a GC a single
thread is created to run the pending finalizers, rather than creating
a thread for each finalizer.

This is almost as fast as having a global thread to run finalizers,
but doesn't require any global state or special treatment by the
scheduler.