Add netbsd to the list of platforms supported.

ToDo: add info to the User's Guide once the Shiny New Documentation is
committed.

Add extended version of previous commit message as a comment.

Further major improvements in interface processing, mostly in the
handling of types.

Interfaces can contain references to unboxed types, and these need to
be handled carefully.  The following is a summary of how the interface
loader now works.  It is applied to groups of interfaces simultaneously,
viz, the entire Prelude at once:

1.  Throw away any entity not mentioned in the export lists.

2.  Delete type (not data or newtype) definitions which refer to
    unknown types in their right hand sides.  Because Hugs doesn't
    know of any unboxed types, this has the side effect of removing
    all type defns referring to unboxed types.  Repeat step 2 until
    a fixed point is reached.

3.  Make abstract all data/newtype defns which refer to an unknown
    type.  eg, data Word = MkW Word# becomes data Word, because
    Word# is unknown.  Hugs is happy to know about abstract boxed
    Words, but not about Word#s.

4.  Step 2 could delete types referred to by values, instances and
    classes.  So filter all entities, and delete those referring to
    unknown types _or_ classes.  This could cause other entities
    to become invalid, so iterate step 4 to a fixed point.

    After step 4, the interfaces no longer contain anything
    unpalatable to Hugs.

5.  Steps 1-4 operate purely on the iface syntax trees.  We now start
    creating symbol table entries.  First, create a module table
    entry for each interface, and locate and read in the corresponding
    object file.

6.  Traverse all interfaces.  For each entity, create an entry in
    the name, tycon, class or instance table, and fill in relevant
    fields, but do not attempt to link tycon/class/instance/name uses
    to their symbol table entries.

7.  Revisit all symbol table entries created in step 6.  We should
    now be able to replace all references to tycons/classes/instances/
    names with the relevant symbol table entries.

8.  Traverse all interfaces.  For each iface, examine the export lists
    and use it to build export lists in the module table.  Do the
    implicit 'import Prelude' thing if necessary.  Finally, resolve
    references in the object code for this module.

I'm sure the number of passes could be reduced.  For the moment,
understandability is of much higher priority.

Hugs can now complete stages 1 through 8 for the whole GHC Prelude,
excepting doing the object linking, which needs further work.

Add test displaying bug in fixity resolution on lhs of a function
binding.  We currently fail on this example, but at least putting it
in the test suite will make sure we don't forget about it :)

Add NetBSD/x86 support.

Don't set O_NONBLOCK on stdout and stderr.  This is a workaround for a
combination of bizarre Unix semantics and shells which don't reset the
nonblocking flag after running a program.

Include hslibs in a binary-dist.

Add a few minBound::Int tests.

Fix bug in abs::Integer->Integer: result was wrong for minBound::Int.

Correct off-by-one error in bounds checking when converting
NoRepIntegers into the real thing.  Not a bug, just a cleanup.

bump version to 4.06

Major improvements in interface processing, and minor supporting
improvements to CT-storage management.

* Make the iface parser return the complete interface as a single
  tree, which is processed later.  Added abs syntax tags
  I_INTERFACE .. I_VALUE to support this.

* Add tagged ("z") 2,3,4,5 tuples.  Because they are tagged, they can't
  be confused with lists, etc.  Selectors zfst, zsnd ... zsel45, zsel55
  check tags first.  Iface processing uses z-tuples wherever it can.

* Add unap as a safe "inverse" of ap; it checks tags.  So
  unap(TAG1, ap(TAG2,cell)) == cell but only if TAG1==TAG2, else
  assertion failure.

* In interface.c, clean up the startGHC*/endGHC* functions.
  processInterfaces() is the top-level driver; it makes 4
  passes over the supplied iface trees.

* Throw away iface symbols not mentioned in export lists.

* Use iface export lists to construct both the export and
  eval environments for a module.

* Don't use Texts to refer to things.  Instead use ConId and
  VarId.  Added ConId and VarId as synonyms for Cell in
  storage.h.

* Add findSimpleInstance in storage.c.

Unbreak finalization.

I was too enthusiastic about removing empty usage entries
from interface files.  This commit fixes my errors of yesterday.

Simon

allow certain keywords as scc labels.

eliminate recursive defn: HC <--> WithGhcHc

A bunch of small changes in the way that usage information
is generated to stuff into interface files.  I'm not certain
anything was really wrong before, but it's tidier now, and
there are slightly fewer dependencies generated.  Main
differences are in RnIfaces.getImportVersions

I also made the check for orphan rules a bit cleverer
(Rename.isOrphanDecl) so that we get fewer spurious orphan modules.

Simon

Remove bogus #! lines.

Obtained-from: Michael Weber <michael.weber@post.rwth-aachen.de>

cleanups & fixes

Obtained-from: Michael Weber <michael.weber@post.rwth-aachen.de>

clean sgmlverb.c

Obtained-from: Michael Weber <michael.weber@post.rwth-aachen.de>

recurse into libraries.

Obtained-from: Michael Weber <michael.weber@post.rwth-aachen.de>

clean PrelGHC.hi in its various flavours.

Obtained-from: Michael Weber <michael.weber@post.rwth-aachen.de>

misc ansification and -Wall cleanup

Add Marc Van Dongen's Integer improvements.  Specifically:

	- new primops: gcdInt#, gcdIntegerInt#, divExact#,
	  quotInteger#, remInteger#.

	- new definitions of quot and rem for Integer in
	  PrelNum (using the new quotInteger# and remInteger#
	  primops instead of quotRemInteger#).  Should be
	  slightly faster than before.  div & mod aren't
	  likewise optimised (yet).

	- specialisations of gcd for Int and Integer, and
	  lcm for Integer in PrelNum.

- add test cases for my URI library
- add George Russell's tests for the TimeExts library

oops, forgot to initialize bufStart in openStdFile.

Fix bogon in rule for parser/hschooks.c

New make variable: $(WithNofibHc) which defines the Haskell compiler
used to build nofib.

Define $(GHC_INPLACE) in fptools/mk/config.mk.in.  $(WithNofibHc) is
set to $(GHC_INPLACE) by default.

Fix $(MKDEPENDHS) in ghc/lib/std/Makefile.

remove cut-n-pasteo

Now that $(GHC) refers to an installed haskell compiler, use
$(GHC_INPLACE) to refer to the ghc in the build tree.

Remove long-outdated AnalFBWW

Now that $(GHC) means an installed copy of ghc, use $(GHC_INPLACE) to
refer to the ghc in the build tree.

Add drvrun008

Automatically detect installed copies of ghc, nhc and hbc and plug
them into the Makefile variables $(GHC), $(NHC) and $(HBC)
respectively.  $(HC) is now supposed to be a generic installed haskell
compiler, which defaults to $(GHC) and can be changed via ./configure
--with-hc=<blah>.

Reorganise and tidy up a few things in config.mk.in.

Derived instances should use *source* types not *representation*
types when doing their deriving stuff.  This bug prevented

	data F = F !Int deriving (Eq)

from working when -funbox-strict-fields was on


Simon

Package local configuration scripts allow to modularise the configuration (we
start by having one for ghc):
* Modularised configuration avoids unnecessary dependencies for individual
  packages.
* Furthermore, package local configuration (such as entering version
  information into rpm .spec files) is possible.

Of course, I had to get the name wrong - this has to be mangled by configure
(eg, to set the version number).

Also added dependency on gmp-devel (see discussion on glasgow-haskell-bugs).

OK, don't say you haven't been warned: Sven "Larry W." Panne waved his
Perl-wand and changed a / / into /\s+/. This fixes the problem with
multiple spaces between {-# OPTIONS ... -#}, but still fails miserably
when quotes are used, e.g. {-# OPTIONS -#include "My Own Header.h" #-}.

rpm build specifications in CVS are a first step to building packages
automatically from CVS (I test this with ghc, but support for the other
fptools components will follow)

Increase default c-t heap from 300000 to 350000 cells for nofib/real/anna.