Compiler changes to:

	* remove PrimIO
	* change type of _ccall_ to IO.

(includes commits to basicTypes/Unique.lhs, deSugar/DsCCall.lhs, and
 prelude/PrelInfo.lhs, but the commit script messed up).

Compiler changes to:

	* remove PrimIO
	* change type of _ccall_ to IO.

add test for declarations of non-existent methods

Check for declarations of non-existant methods
	(bug: typecheck/should_fail/tcfail077.hs)

oops - xor# is dyadic, not monadic.

fix typo that crept in with last commit.

Replace use of map() with foreach, to make the driver perl4 compatible
again.

add xor# word operation

always print types with binders, and put them on separate lines.

Catch failures to output dependencies

New option --include-prelude that generates dependencies on imported prelude interfaces; renamed -W to -w

New option: -W, suppress duplicate interface file warnings

Quote -#include opts when passing them on (to mkdependHS); silently add ghc syslib with posix (it depends on it.)

Cosmetic fix to avoid cpp warnings under linux

Minor fix to suppress gcc warning

ErrorIO_innards: push realWorld# on B stack (cf. startStgWorld)

When needed , define __linux__ to silence cpp warnings

Added 'standard closures' to discussion of how Hugs enters an object

add pointer to Bradley C. Kuszmaul's CVS instructions.

add CVS guidelines file

changes for MachRegs.Addr -> MachRegs.Address

One square bracket too many

Renamed freeStdChannel to freeStdFile

Parallel Haskell changes(hwloidl)

Don't use ForeignObjs with Parallel Haskell; renamed finaliser for standard descriptors (now freeStdFile)

new prims made visible: parAtAbs#, parAtRel# and reallyUnsafePtrEquality#

Avoid exposing ForeignObjs when compiling Parallel Haskell

unionBy: remove duplicates in second argument (required for the following invariant to hold: if the first list argument contain no duplicates, then the result of applying unionBy doesn't either)

Improved glossary/terminology at start - added unpointed and unboxed.
Created a section at start to describe the source language.  At the moment,
 all it contains is a description of unboxed tuple constructors.
Replaced erroneous uses of "closure" with "heap object".  According to
 the glossary, closures are enterable - things like stack objects
 are not enterable so they can't be closures.
Clarified section 2.7 (heap and stack checks): why should we not move Hp
 during heap check?
Added comment that I don't believe in the notion of fixed headers.

fix for overloading-related space leak (typecheck/should_run/tcrun002)

fix reversed arguments to unpackNBytesST in getCurrentDirectory.

Comment on let-to-case added (simplNonRec)

Updated to reflect MachRegs.Addr to MachRegs.Address renaming

Updated to reflect MachRegs.Addr to MachRegs.Address renaming; various x86 bug fixes

Updated to reflect that MachRegs.Addr has been renamed to Address to avoid needless conflict with GlaExts

Updated to reflect MachRegs.Addr to MachRegs.Address renaming; x86: Hp and HpLim are located relative to BaseReg, not StorageMgrInfo

Updated to reflect MachRegs.Addr to MachRegs.Address renaming; x86 shift instructions need special treatment to get at %cl; added DEBUG msg if x86 MOV %r,%r instruction generated

Updated to reflect MachRegs.Addr to MachRegs.Address renaming; fixed incorrect handling of x86 shift instructions (second operand can be *both* immediate and register(ecx))

import MachRegs unconditionally

Added flag opt_UnfoldingKeenessFactor; new (internal) function: lookup_def_float