Forgot to include in previous commit.

New primop: integer2Word#. Only used in fromInteger methods on Word*

Equip all locally bound names with new uniques

Still a lingering lost-Eval-dict bug; but nearly there!

scc printing: Only output duplicated flag in scc label when debugging

import decls:Allow the use of 'as' without 'qualified'

Simplifier fixed - I think!

Include srcloc in panic msg

"Testing GHC" section: Don't suggest compiling a version of 'nfib'
for 'Complex Float', since this is not an Ord type.

Added IOExts.openFileEx + IOExts.IOModeEx

More simplifier modifications; may not even compile; will fix first thing tomorrow

jtos: leave Integer-land ASAP

newtype fix

- remove breakage introduced by ADR
 - add ALIGNMENT_LONG

we don't use GNUCPP anywhere.

Use $(CPP), not $(RAWCPP).  This involves saving the cpp output to a
temporary file.  This change should fix 'make depend' on RedHat 5.

set CPP to @CPP@, not @RAWCPP@.

add all missing SIZEOF_* and ALIGNMENT_* defines.

- move SIZEOF checks together (previous commit added SIZEOF checks to
   a different part of the file).

 - check alignment of longs:  unsigned ints are no good, because we need to
   check whether f.e. a float can be put on the stack, which is
   StgWord aligned, and currently an StgWord is basically a long.

add
	RTS_PREFIX = $(FPTOOLS_TOP)/common-rts/

(Harmless unless compiling for the new RTS).

added .hc-->.s and .S-->.o rules.

New specialiser again; I think the simpifier is OK

New specialiser; warning: simplifier *may* be broken

New specialiser

indentation fix.

Exercise defaulting of _ccall_ results (bot not args) to ()

Default ambiguous _ccall_ results to (), not the arguments

Don't split object files when booting from .hc

Misc fixes to setup for booting via .hc files

Don't run mkdependHS when booting via .hc files

addAvailEnv: restrict previous change to not redo
duplicate checks.

Added alignment tests

addAvailEnv: special handling of explicit exports of class methods.

Add some size tests.

startPerformIO: Only save C stack pointer when
compiling the conc RTS on x86s.

back out last commit until we've decided how to do these properly.

unifyKindErr: drop the "Compiler bug:" prefix

Added more feature tests

Generate a little bit less code for record selectors.

Fix for the minBound case, obtained from GMP 2.