wibble

mingw32: Use C runtime's version of system()

Reinstate old signature for 'makeForeignObj', i.e.

  makeForeignObj :: Addr -> Addr -> IO ForeignObj

combining the primitive 'makeForeignObj' and addForeignFinalizer(+).

The primitive ForeignObj constructor is now exported as 'mkForeignObj'

Added:

   writeStablePtrOffAddr   :: Addr -> Int -> StablePtr a -> IO ()
   readStablePtrOffAddr    :: Addr -> Int -> IO (StablePtr a)
   writeForeignObjOffAddr  :: Addr -> Int -> ForeignObj -> IO ()

- added showListWith, non-overloaded version of showList.

Fix a bug in weak pointer support: if finalize is called on a weak
pointer, then a DEAD_WEAK object could appear on the weak pointer
list.

To avoid needing to double-link this list, add a link field to
DEAD_WEAK objects, and remove them from the list at garbage collection
time.

Allow the memo table itself to be collected when the function becomes
unreachabl.

	- individual finalizers refer back to the memo table
	  via a weak pointer.

	- a finalizer for the whole table walks through each
	  bucket calling 'finalize' on every weak pointer.

Back out accidental commits.

Fix small stats bug in the -G1 case.

Change the semantics of weak pointers - now weak pointer objects
aren't automatically considered reachable, so a weak pointer to a weak
pointer object is now meaningful.

Fix stable pointer bug:

   - the 'keep' flag is now a pointer to the stable name object.

   - makeStableName always returns an existing stable name object
     if one exists.

   - re-use stable name table entries iff the stable name object
     becomes unreachable (takes into account generational GC now),
     and the reference count is 0.

Fix bug in finalizeWeakPointersNow - some weak ptrs could be missed.

Add memo table test.

Fix run queue corruption bug.

Remove bogus TICK_ALLOC_HEAP.

- clean up a little
- add some code to avoid promoting thunks (#ifdefed out)

hClose: don't raise an exception when passed a handle that's already closed,
i.e.,

    h <- openFile "a" ReadMode >> hClose h >> hClose h

will now succeed (provided 'a' exists and is readable, of course.)
Repeated hClose's are just redundant.

Include GC bytes copied in stats string.

Allow eager promotion to be turned off by a compile-time option.

Don't promote blackholes from generation 0 if possible.

- Fix a couple of uninitialized variables
- Fill the nursery with junk after GC if sanity checking is on

Updates on the Win32 front

- If -H is larger than -M, set -M to -H
- Fix a couple of bugs in calcLive().

- Fix a bug in 2-space (-G1) collection
- Calculate an estimate of the number of bytes copied during GC
  for GC stats.

Print time spent collecting each generation.

Integer fixes.

Couple of Integer fixes.

- Fix off-by-one in __encodeFloat;
- Tidy up __encodeDouble a bit.

Tweaks to make docs build cleanly

Fixed waitQSemN bug

Added field 'StgWord stgexpr' to struct StgBCO, to help with debugging
the interpreter.  This field (if nonzero) points into the Hugs heap, at
the STG tree from which the BCO was derived.

Some test updates.

Fix negate (toInteger (minBound :: Int)).

Hack around a couple of explicit uses of the Integer rep.

Eliminate an unnecessary pattern match on J#.

Allow completely unbound tyvars of non-type kind

Added test that caused 3.03 and 4.01 to enter a blackhole

Allow -f flag to override existing symlink tree.

Make decoding of RtsFlags.DebugFlags work on big-endian architectures.

Add two new operations to StgPrimFloat.c:

	__int_encodeFloat
	__int_encodeDouble

for encoding floats/doubles from small integers.  This avoids having
to convert small integers to large ones before an encodeFloat
operation, and fixes the two cases of slowdown in nofib after the
small integer changes.

Also:
	- remove encodeFloat and decodeFloat as primops
	- use foreign import for encode{Float,Double} and
	  the various isNaN etc. ccalls in PrelNumExtra.