80-columnify

Don't create a new infotable for every constructor application.
Amazingly, that's what the codegen.c used to do.  It didn't generate
vast numbers of redundant infotables until recently, when I changed
translate.c to generate saturated constructor applications in line.
Prior to that, there was only ever one application of each constructor,
so the old scheme was ok.

Also, fix the assembler so that info table ptrs are commoned up.
Eg, for [1,2,3], store only one copy of the address of the (:) itbl,
rather than 3, in the BCO.

Fixed a FunDep leak in tcSimplifyToDicts (they weren't being filtered out),
and fixed bug in instance improvement (matching wasn't being done correctly
for polymorphic instances).

Remove compile-time address space allocated to tuples, and remove all
references to TUPMIN.  I think I have fixed the address mapping for
when TREX is defined, but you never know.

Remove Hugs' special treatment of tuples, and instead have them as
just another Tycon.  This is to make interworking with GHC simpler.

Put tuple entries in the Tycon table.  Modify isTycon, isTuple, tupleOf,
mkTuple and whatIs so that client code doesn't see any difference.
Add allocTupleTycon to manufacture tuple Tycon entries as startup.

Only detect (# and #) as special tokens when parsing interface files.

Add missing stdout file

Some minor tidying-up that should remove an occurrence
of an empty Let Rec that confused CoreLint.dumpLoc.

Simon

Fix a major bug in exporting unfoldings involving existentials.

Change core printing so that we put an '@' before type
variables in case patterns.  This only affects existentials.

	case x of
	  C @ a x y -> ...

Here 'a' is an existentially quantified type variable, and the
'@' signifies this.  We continue to omit kinds and type on case-bound
variables; the type checker can fill them in.

The reason for this change is that type variables and term variables
live in a different name space, so we need to know which name space
is  involved when binding one.  How this ever worked I will never know.


While I was at it, I also arranged that -ddump-rn prints out whatever
it has even if it finds errors.  Adding -dppr-debug prints even the
unfoldings on imported things.

Simon

Change printDump so that it prints in user style by default.
This means that (eg) -ddump-simpl output is much more readable...
but you may get confused by variables that look the same but aren't.

To recover the previous behaviour use -dppr-debug

This change only affects compiler hackers; let me know if it
has any good or bad effects.

Simon

Install an interface file for PrelGHC, too.

fixIO versionitis

Replaced $(H_FILES) $(H_CONFIG) with $(wildcard *.h), it seemed to be
a cut-n-paste bug from another Makefile where they are defined.

Extend getTyVarsToGen to take the closure of the set of tyvars
with respect to functional dependencies.  Really simple programs
using functional dependencies work now.  Also fixed a small glitch
where trivial (empty) FunDeps were being tossed into the context willy nilly.

Remove some hooks into the Prelude which Hugs no longer needs to know about.

More mods to interface file parsing:
* Strictness annotations in data decls
* Allow qualified names in a couple more places

Add Andy Gill's Monad library into the build.

Move fixST and fixIO out of the prelude.

Corrected time stamp calculation for .hp files.

Make the combined/standalone switch, +c/-c, work at all :-)

Add initial support for loading GHC Prelude (doesn't work yet):

* Command line flag, +c/-c to start up in combined or standalone mode.
  In combined mode, looks for GHC's prelude in ghc/interpreter/GhcPrel
  (pro tem).

* Parse unboxed tuple types and usage annotations in interface files.

Directory wibbles

Fergot the booties...

Basic functional dependency implementation.  Most notable change to be
aware of is that the datatype `Inst' now has a new constructor `FunDep'
used to track functional dependencies.  The FunDep predicates are filtered
out in tcSimplify* so that they don't escape.

Change the convention for cost-centre labels to be <name>_cc and
cost-centre stacks to be <name>_ccs.  This makes cost-centre labels
more consistent with our other naming conventions, and fixes some
problems caused by cost-centre labels being misinterpreted by the
mangler.

This fixes one cause of profiled programs crashing; if you're seeing
this symptom then this patch may help.

"compilation IS required" ===>  ""

Relegate the "ignoring heap-size setting option..." messages to -v.

Relegate the 'Out of date: <blah>' messages to -dshow-rn-trace, they
don't mean a great deal to the average punter.

The infamous "NOTE: simplifier still going..." message is now
protected by #ifdef DEBUG.  It doesn't mean much to the average user,
and we tend to use DEBUG when hacking anyway.

Only print out the module-version info messages when the -v flag is
on.

Something-similar-requested-ages-ago-by: Sven Panne

Fix up code to print out the phase & signal number if one of the
compilation phases receives a signal.  The previous commit
accidentally sneaked in a non-working version of this stuff.

add missing #include

disallow case expressions with no alternatives.

Add some tests for the new async exception behaviour.

initialize the blocked_exceptions field of a TSO properly.

Oops, forgot to scavenge the new fields in a TSO.

Common up no less than 3 copies of the TSO scavenging code in the
process, thus fixing a bug in one of the copies.

Fix the case where we're killing a thread that is blocked on
delivering an exception.

Support for blocking & unblocking asynchronous exceptions.

  - new primops:

	blockAsyncExceptions#, unblockAsyncExceptions# :: IO a -> IO a

  - raiseInThread will block if the target thread is currently
    blocking async exceptions.

  - async exceptions are currently implicitly blocked inside
    an exception handler.  This decision might be reversed when
    we have more experience with this stuff.

  - Move exception-related stuff in the RTS into its own file,
    Exception.{h,hc}.

Remove empty sections for arch's supported in StgCRun.c.

Alpha fixes.