URK!  when scanning the heap we might find a DEAD_WEAK that either (a)
is the size of a WEAK, because we just overwrote its info table, or
(b) is the size of a DEAD_WEAK, because its been around at least one
GC.  The code herein only accounted for case (a).

Fix: check the word at location p+sizeof(DEAD_WEAK), see whether it
looks like an info table ptr, and if so we're in case (b).  This is
guaranteed to work, because we know that in case (a) the appropriate
word will be a closure pointer.

Clean ups:

   - reduce the namespace pollution of StgTypes.h, it doesn't define
     the shorthand versions any more (W_, I_ etc.).  These are moved into
     Stg.h.  StgTypes.h also defines StgClosure as an "opaque" struct.

   - RtsAPI.h is now standalone, and includes HsFFI.h and thereby
     config.h & StgTypes.h.  Now we don't need to #include "Stg.h" in
     *_stub.c.

   - all the rts_mkXXXX and rts_getXXXX functions are defined in terms
     of the HsXXXX types rather than random C types (this fixes some
     potential bugs in our foreign export support).

   - added HsWord type, to match StgWord.  The Haskell version of this
     type isn't "documented", but perhaps it should be.

small fixes

wibble

Fix compilation with 4.08.1

conditionalise imports of StgInterp etc.

add hsc2hs, missed in the merge

This commit completes the merge of compiler part
	of the HEAD with the before-ghci-branch to
        before-ghci-branch-merged.

add -optc-DGHCI

revert 1.109 (ghci isn't included in DIRS unless we're bootstrapping)

Wave goodbye to FLOATS_AS_DOUBLES, it was a somewhat misconceived idea
which will cause trouble with the FFI on 64-bit machines.

More small changes

merge before-ghci -> before-ghci-branch-merged into the ghc
(non-compiler) parts of the tree.

kill this file once and for all

Merge before-ghci -> before-ghci-branch-merged in the non-ghc parts of
the tree.

Updated to reflect reality and be a bit more helpful.

build the rts *after* the compiler

Change the way source distributions are built.

Now we do a 'make boot', 'make distclean', build a link tree to filter
some files out (eg. .cvsignore), and finally tar up the result.  The
process isn't bulletproof, so it makes sense to compare the manifest
against previous versions.

Now you don't have to be in a linked build tree to make a source
distribution.

.depend should be in DIST_CLEAN, not MAINTAINER_CLEAN

sgmlverb.c is in maintainer-clean

config.status should be in distclean

Remove FunDeps hi-boot files

Remove SpecEnv.hi-boot-5; hasnt been used for ages

enhance the clean target for doc subdirs

clean more files: %.s for each %.hs/%.lhs, *_stub.{c,h}, *.raw_s.

Dealing with instance-decl imports; and removing unnecessary imports

More renamer... not in a working state I fear

clean up Parser.info in MAINTAINER_CLEAN.

"make clean" fixes

"make clean" fixes

Remove unused files.

maintainer-clean should include distclean.

"make clean" in here now works as advertised.

Finally get CompManager to compile.  Also rm some redundant imports.

stgSyn/StgInterp_HC_OPTS ---> ghci/StgInterp_HC_OPTS

don't generate heimdall.summary unless -heimdall is on.

Prepend "./" to the command to run only if the file exists in the
current directory, rather than based on whether the command begins
with a ''/'.  This means that 'make HC=ghc-4.08.1' in ghc/tests works
as expected.

remind me that -nohi doesn't do anything

remind me what this test does

Found a better way to do the arch test for $(GhcWithInterpreter), used
it for $(GhcWithNativeCodeGen) too and removed one $(shell...).