Rename and reorder the internals for unique ids etc. Also fixed a couple
of THSyntax names.

The Great Renaming. I hope I've kept everything in sync - and all the tests
pass. Now datatypes follow the

data Foo = <a kind of Foo>Foo
         | <another kind of Foo>Foo

convention and the smart constructors can be uniformly created by
lowercasing the first letter of the constructor.

Solaris2 needs _POSIX_PTHREAD_SEMANTICS for the getpw*_r() prototypes.
Make libraries/unix/Makefile use a new variable unix_SRC_HSC2HS_OPTS
which we configure in mk/config.mk.

Add a microscopic amount of documentation for functional dependencies; who would like to volunteer to write more?

The 6.0 release notes.

Call strerror() instead of sys_errlist[errno] (deprecated & smells bad)
(Broke Solaris-build here)

Restore the correct partitioning of command-line arguments into
objects and compilation-manager inputs for --make mode, and restore a
comment explaining the behaviour.

Rev. 1.120 made some subtle changes to the semantics, in particular in
cases where a file given on the command line is neither a source nor
an object file.

I believe the behaviour for one-shot mode has not changed.  The
behaviour for GHCi mode is now the same as --make mode (previous to
rev. 1.120 it was subtly broken in this respect).

revert rev. 1.16; looks like this was an experiment that got
accidentally committed.

Put the call to saveDynFlags back in one place.

Improved greencard-related definitions

Added support for newtypes to TH and altered a test for them.

When --make is used we need to save the initial dynamic flags too

Fix building on 5.02

Generate
  int main (int argc, char *argv [])
instead of
  int main (void)

Suggested by: Nicolas Oury

Skeleton 6.0 release notes.

defaultsHook is now ghc_rts_opts.

Update version numbering policy

Version 6.0

--with-gcc=<path>: for mingw32 plat, canonicalise path using cygpath

---------------------------------
	Improve the dynamic-linking story
	---------------------------------

Arrange proper initialisation for the dynamic linker.  Whenever the dynamic linker does
anything (Linker.linkExpr), it first initialises itself, arranging to only do so once
of course.

"Initialising itself" includes loading any .o files, libraries, and packages specified
on the command line.  The main effect of all this is to fix a Template Haskell problem,
which happened when a TH link needed some C library that it couldn't link.  Now it does.



While I was at it, I tidied up main/Main.hs quite a bit.  This is a delicate area (handling
the command line arguments), but I don't believe I broke anything!  All the libraries
build, and lots of tests run.

Wolfgang: I also jiggled the darwin_TARGET_OS stuff a little in Linker.lhs, but again
I think the net effect is zero. You might want to check.

Comments only

--------------------------
	Minor Template Haskell bug
	--------------------------

This bug meant that spliced-in class declarations yielded a 'op not in scope',
where op was the class operation.  Thanks to Andre Pang for spotting this.

Some consequential tidying up in parsing too.

- rearrange the file a little

- build a binary distribution from stage 2 by default, unless
  overriden from the command line.

Remove obsolete references to DriFT, DtdToHaskell and Xtract.

tryM again: make it work with GHC 5.02.x (I hope).  Seems we changed
the representation of UserErrors several times in the last few
releases...

Fix a pair of off-by-one errors that meant that BCOs were normally two
words larger than they needed to be.

Eeek!  intsToReverseBitmap was borked for bitmaps with more than
wORD_SIZE_IN_BITS entries.

More Win32 help

Add checks for ghc-pkg and green-card 3.00 (used in x11 lib)

Rename per-package configuration files from $(PACKAGE).conf.* to
package.conf.*, making them easier to find (since each package is
in a separate directory anyway).

Make retainer profiling work again after the recent SRT changes.

heapCensus should grok IND_OLDGEN objects, because compacting GC
doesn't always eliminate them (perhaps it should).

Now that we have auto packages, it makes sense to keep all the
interfaces for hierarchical libraries in the same directory tree.  So
now, instead of putting interfaces for package P in $libdir/imports/P,
we put them all in $libdir/imports.

Interfaces for old non-auto non-hierarchical packages now go in
$libdir/hslibs-imports/P for package P.

Doc the -L ghc-pkg option.

Add -L (--list-packages-local) flag which lists packages in the
topmost configuration file only.

update, including a short note on Nhc98.

Should have been committed with recent SRT changes (see rev. 1.39 of StgSyn.lhs).

Change the way SRTs are represented:

Previously, the SRT associated with a function or thunk would be a
sub-list of the enclosing top-level function's SRT.  But this approach
can lead to lots of duplication: if a CAF is referenced in several
different thunks, then it may appear several times in the SRT.
Let-no-escapes compound the problem, because the occurrence of a
let-no-escape-bound variable would expand to all the CAFs referred to
by the let-no-escape.

The new way is to describe the SRT associated with a function or thunk
as a (pointer+offset,bitmap) pair, where the pointer+offset points
into some SRT table (the enclosing function's SRT), and the bitmap
indicates which entries in this table are "live" for this closure.
The bitmap is stored in the 16 bits previously used for the length
field, but this rarely overflows.  When it does overflow, we store the
bitmap externally in a new "SRT descriptor".

Now the enclosing SRT can be a set, hence eliminating the duplicates.

Also, we now have one SRT per top-level function in a recursive group,
where previously we used to have one SRT for the whole group.  This
helps keep the size of SRTs down.

Bottom line: very little difference most of the time.  GHC itself got
slightly smaller.  One bad case of a module in GHC which had a huge
SRT has gone away.

While I was in the area:

  - Several parts of the back-end require bitmaps.  Functions for
    creating bitmaps are now centralised in the Bitmap module.

  - We were trying to be independent of word-size in a couple of
    places in the back end, but we've now abandoned that strategy so I
    simplified things a bit.

Comments only

Calling mmap() with a size > 3Gig results in EINVAL on Linux.  Catch
this case and report it as an out of memory condition instead of a
panic.