Remove bogus code: the mutable list does not need to be rebuilt during
compaction.  It just needs to be updated, which is already happening.

(fixes crashes in compacting collection; I think a different fix is
required for STABLE, though)

Update to match ClosureTypes.h

remove gaps in the closure type sequence, and add a big warning comment

modInfoExports, TypecheckedSource and ParsedSource are exported.
Added modInfoPrintUnqualified function.

- DriverPipeline.compile: report errors in GHC_OPTIONS pragmas using the
  Message callback, and give them a proper line number.

- GHC.checkModule: read the GHC_OPTIONS pragma, and report errors
  appropriately.

main_thread double-linkage bug

In the event of a complete deadlock, we should handle the case when
the main thread is BlockedOnSTM (bug #1183455, the strange error
message only occurs on Windows because on Unix we end up waiting for
^C instead of deadlocking).

Turn off splitting on x86_64

Pass -ignore-dot-ghci to GHC

Now build.mk.sample does *nothing* by default

Haddock can't handle recursive modules currently, so we disable it for now.

Haddock fixes, "-- *" should be used with care...

Significant clean-up of the handling of hi-boot files. 
Previously, when compling A.hs, we loaded A.hi-boot, and
it went into the External Package Table.  It was strange
but it worked.  This tidy up stops it going anywhere;
it's just read in, and typechecked into a ModDetails.

All this was on the way to improving the handling of
instances in hs-boot files, something Chris Ryder wanted.
I think they work quite sensibly now.  

If I've got all this right (have not had a chance to
fully test it) we can merge it into STABLE.

Improve kind error msg; may make some tests change their output

Improve location info for kind errors; may make some tests change their output

And yesterday's "I messed up DocBook XML, too!" award goes to...

   *drum roll*

   ... Simon Peyton-Jones!

Congratulations! :-)

Rejig handling of environments in arrow notation: instead of the
proc_level stuff, we just record the environment of the proc, and
use that on the left side of -< and the head of (|...|).

This also makes the arrow1 test yield a compile error, as it should,
but the error message is uninformative.

back out Directory.hs:1.14 and directory.c:1.7; neither meant for HEAD

LIBRARY, GHCI_LIBRARY: permit settings of these to be overridden in Makefiles; Merge to STABLE

Update the CgCmmBranch (ghc directory) up to the HEAD ghc-6-4-branch-point

Be consistent about quotes in error messages; MERGE TO STABLE

Update comments on hslibs to make it clear that that they are (a) deprecated,
and (b) not required for building GHC.

	Merge to STABLE

Initialise foreign exports from GNU C __attribute__((constructor)) functions
in the stub C file, rather than from __stginit_ functions.
For non-profiling ways, leave out __stginit_ alltogether.

Better reporting of WAIT_FAILEDs from WaitFor*Object() calls.

Merge to STABLE.

fix typo (Sourceforge bug #1183229)

(for STABLE)

Fix TH handling of guards

Fix previous commit

Typo in comment only.

Make the status messages from ghc --make display the number of modules
to be compiled, as in:

[3 of 9] Compiling Foo.hs     ( Foo.hs, Foo.o )

Fix breakage in i386 floating-pt comparisons

preprocessFile: return the source filename rather than "<buffer>"

stopTaskManager: don't complain too loudly if we can't stop all the
tasks.  The IO manager thread turns out to be an offender here;
perhaps we should start sending signals to threads if they don't stop
when they're told to.

Add :

  loadMsg :: Session -> LoadHowMuch
	  -> (Messages-> IO ()) -> IO SuccessFlag

ie. just like load, except it takes a message callback and plumbs it
down to hscMain.  checkModule now uses loadMsgs.  Better name
suggestions welcome.

- checkModule is back, and now returns a ModuleInfo

- added:

  modInfoTopLevelScope :: [Name]
  modInfoExports       :: [Name]

- in order to implement modInfoExports, ModDetails now contains
  md_exports::NameSet.

Fix assertion

tidy up prev commit; mingw doesn't support these two symbols

Dynamic Linking:
On non-Win32, we can store cross-dylib pointers in static data, so disable
a Win32-specific hack on the other platforms, to prevent code bloat.

Few changes to GHC API:

  * The Instance type is removed. Each instance is represented as dictonary Id.
Now there is:

  isDictonaryId :: Id -> Bool

predicate which allows to distinguish them.

  * The full list of all available predicates of type (Id -> Bool) is exported
  * Few predicates for TyCon are exported
  * checkModule is removed. It isn't used anymore. Instead the full load is made.
The hook API will be required at some time in order to make available the
parsed/typechecked source.

make 'mkdependC' happy

initDynLinker is invoked from unload now. This makes safe to call it even when
then linker is still not initialized.