Add some quick-start instructions.

update a little.

Move the RTS document into the ghc tree where it belongs.

Only require happy if compiling ghc, and ask for version 1.6

add gcdIntegerzh.

Oops, back out most of last revision.  Other changes crept in by mistake.

fix a couple of typos in the rules docs.

oops, 'g' and 'h' aren't hex digits.

nullary recpat reg. test

syntactically invalid

Exhaustiveness checks for patterns containing a nullary recpat
panic'ed. Fixed.

Fix -monly-3-regs problem.

typePrimRep should work on the repType, avoiding foralls etc.

Removed rather odd quote-removal sed patterns; surely not needed... (they were causing CPP trouble)

Fixed bug/limitation in command-line handling.

Recent CPP-related wibblement meant that _WIN32 wasn't defined when
the pre-proc was run (cf. -undef usage.) Avoid the use of _WIN32

Another cosmetic change - avoid desugar warning

showPrimRep - dead code removal. No need to carry over to 4.04

test for bit shifts by 32.

Document changes to shift operations.

Shift operations where the shift count is greater than the width of
the type being shifted are now defined as returning zero (or -1 in the
case of a negative integer being shifted right with iShiftRA).

The hscpp saga continues, this time mysterious 2's are appearing in my
files.

Improve error msg in tcPolyExpr; Simon pls merge into release

- Use explicit file modes when installing binaries & scripts (not essential,
  but someone complained about it).

- Use $(INSTALL_SCRIPT) for scripts

Sigh, another go at getting #line directives right.

Allow unboxed tuple components to have polymorphic types.

Sigh, didn't work in the  cpp, !literate case.

oops, debugging code committed by mistake.

Fix filenames in {-# LINE #-} directives when using both unlit and
hscpp (I think).

fix typo in hppa1.1 support.

remove spurious ^M's

add unofficial/undocumented function:

    mkWeakIORef :: IORef a -> IO () -> IO (Weak (IORef a))

for weak pointers to IORefs.  This avoids some of the problems with
inlining disturbing the semantics of mkWeak by making the weak key
point to the primitive MutVar instead of the box.

Allow mkWeak# to take an unlifted type as the key argument.  The
intention is that boxed-but-unlifted types can be used as keys, not
unlifted types in general but there's no way to enforce this at the
moment.

+RTS -s<blah>  ==>  +RTS -S<blah>

for -Rghc-timing.  The former now gives only summary statistics, as
advertised.

If calling a 'foreign export'ed Haskell function resulted in an error,
localise the error by supplying the module name as well as the name
of the function.

Improve the UI a little for modules contain 'foreign export' declarations -
no longer demand that module name == (basename filename)

Something is very screwy here.  This file does not appear to have been
removed, but it doesn't have any tags beyond 2.04.

Back out yesterday's change to Parser.y, and throw out illegal do
expressions in the renamer instead.  It turned out to be hard to get
the optional semicolons right in the grammar at the same time as
checking that the last statement is an expression.

Add a bit more info about hi-boot files

Do a more correct job of explicit for-alls in types