GitLab

----------------------------
        Re-do kind inference (again)
	----------------------------

   [WARNING: interface file binary representation has
   (as usual) changed slightly; recompile your libraries!]

Inspired by the lambda-cube, for some time GHC has used
	type Kind = Type
That is, kinds were represented by the same data type as types.

But GHC also supports unboxed types and unboxed tuples, and these
complicate the kind system by requiring a sub-kind relationship.
Notably, an unboxed tuple is acceptable as the *result* of a
function but not as an *argument*.  So we have the following setup:

		 ?
		/ \
	       /   \
	      ??   (#)
	     /  \
            *   #

where	*    [LiftedTypeKind]   means a lifted type
	#    [UnliftedTypeKind] means an unlifted type
	(#)  [UbxTupleKind]     means unboxed tuple
	??   [ArgTypeKind]      is the lub of *,#
	?    [OpenTypeKind]	means any type at all

In particular:

  error :: forall a:?. String -> a
  (->)  :: ?? -> ? -> *
  (\(x::t) -> ...)	Here t::?? (i.e. not unboxed tuple)

All this has beome rather difficult to accommodate with Kind=Type, so this
commit splits the two.

  * Kind is a distinct type, defined in types/Kind.lhs

  * IfaceType.IfaceKind disappears: we just re-use Kind.Kind

  * TcUnify.unifyKind is a distinct unifier for kinds

  * TyCon no longer needs KindCon and SuperKindCon variants

  * TcUnify.zapExpectedType takes an expected Kind now, so that
    in TcPat.tcMonoPatBndr we can express that the bound variable
    must have an argTypeKind (??).

The big change is really that kind inference is much more systematic and
well behaved.  In particular, a kind variable can unify only with a
"simple kind", which is built from * and (->).  This deals neatly
with awkward questions about how we can combine sub-kinding with type
inference.

Lots of small consequential changes, especially to the kind-checking
plumbing in TcTyClsDecls.  (We played a bit fast and loose before, and
now we have to be more honest, in particular about how kind inference
works for type synonyms.  They can have kinds like (* -> #), so

This cures two long-standing SourceForge bugs

* 753777 (tcfail115.hs), which used erroneously to pass,
  but crashed in the code generator
      type T a = Int -> (# Int, Int #)
      f :: T a -> T a
      f t = \x -> case t x of r -> r

* 753780 (tc167.hs), which used erroneously to fail
      f :: (->) Int# Int#


Still, the result is not entirely satisfactory.  In particular

* The error message from tcfail115 is pretty obscure

* SourceForge bug 807249 (Instance match failure on openTypeKind)
  is not fixed.  Alas.

Import trimming

Added prototype for initAdjustor

Warning police + a little cleanup of #ifdefs relating to execPage. Added two
TODOs on the way (an adjustor spanning two pages + calling initAdjustor at RTS
startup). Note that execPage is still disabled (why?).

Add performMajorGC

Fix threaded RTS problem that is currently causing the conc007 test to
loop indefinitely.  Comment from the source regarding this change:

/* Major bogosity:
 *
 * In the threaded RTS, we can't set the virtual timer because the
 * thread which has the virtual timer might be sitting waiting for a
 * capability, and the virtual timer only ticks in CPU time.
 *
 * So, possible solutions:
 *
 * (1) tick in realtime.  Not very good, because this ticker is used for
 *     profiling, and this will give us unreliable time profiling
 *     results.  Furthermore, this requires picking a single OS thread
 *     to be the timekeeper, which is a bad idea because the thread in
 *     question might just be making a temporary call into Haskell land.
 *
 * (2) save/restore the virtual timer around excursions into STG land.
 *     Sounds great, but I tried it and the resolution of the virtual timer
 *     isn't good enough (on Linux) - most of our excursions fall
 *     within the timer's resolution and we never make any progress.
 *
 * (3) have a virtual timer in every OS thread.  Might be reasonable,
 *     because most of the time there is only ever one of these
 *     threads running, so it approximates a single virtual timer.
 *     But still quite bogus (and I got crashes when I tried this).
 *
 * For now, we're using (1), but this needs a better solution. --SDM
 */

Fix DEBUG & !RTS_SUPPORTS_THREADS build (bogus assertion failure)

Accept unboxed tuples in type signatures

---------------------------------------
  Precise locations for duplicate imports
  ---------------------------------------

As a displacment activity I added reporting for duplicate imports
And more precise locations for the imports themselves.  For example:

  module T where

  import Maybe (isJust,isJust)
  import Maybe (isJust)

gives

Foo.hs:
    Warning: `isJust' is imported more than once:
	       imported from Maybe at Foo.hs:4:16-21
	       imported from Maybe at Foo.hs:3:23-28
	       imported from Maybe at Foo.hs:3:16-21

Patch from Andres Loeh: Generate the *.raw-hs files with the inplace
GHC, otherwise one runs into a "fixed-point" problem where only
documentation known to the bootstrapping GHC could be built.

Synched with latest changes in Hugs' aclocal.m4 (wglGetProcAddress
fiddling)

Optimisation in awaitEvent(): reset the abandonRequestWait() signal if
it isn't needed (compensates for not using PulseEvent() in
abandonRequestWait(), see AsyncIO.c for details).

Code cleanup in awaitEvent(): don't test the various exit conditions twice.

Fix some threaded RTS bugs:

 - in awaitRequests(), it was possible to return immediately even
   though wait was set to rtsTrue.  This lead to a busy wait loop,
   so I've disabled that case (see code for details).

 - using PulseEvent in abandonRequestWait() wasn't the right thing,
   because it exposed a race condition.  Again, see the comment in
   the code for details.

Add missing 'continue' after 'passCapability'

- make --disable-foo work for each of the --enable options
- tidyup: use AC_HELP_STRING everywhere

no_glu is spelled no_GLU, and similarly for no_GLUT

Undo part of previous merge: the HEAD doesn't export saved_termios (it
is accessed by functions in RtsStartup instead).

merge rev. 1.106.2.3

merge rev. 1.128.2.11

Fix up following changes to the Capability API.  (strange, I'm *sure*
I tested this stuff...)

---------------------------------
  Gate in a few more instance decls
  ---------------------------------

When we have
	instance T a where...

we need to suck in the instance decl rather more aggressively than we
were.  (A case I hadn't thought of because it only happens with
undecideable instances.)

-----------------------------------------------------
  Fix a subtle loop in the context-reduction machinery
  ----------------------------------------------------

This bug was provoked by a recent change: when trying to prove
a constraint C, TcSimplify.reduce now adds C to the database before
trying to prove C, thus building recursive dictionaries.

Two bugs
a) If we add C's superclasses (which we were) we can now build a
   bogusly-recursive dictionary (see Note [SUPERCLASS-LOOP]).
   Solution: in reduce, add C only (via addIrred NoSCs) and then
   later use addWanted to add its definition plus SCs.

b) Since we can have recursive definitions, the superclass-loop
   handling machinery (findAllDeps) must carry its visited-set
   with it (which it was not doing before)


The main file is TcSimplify; but I modified a bunch of others to
take advantage of new function extendVarSetList

To get wglGetProcAddress on Windows, we have to link with
opengl32.dll, too, even when we are using Cygwin with X11.

--------------------
	Towards type splices
	--------------------

Starts the move to supporting type splices, by making
HsExpr.HsSplice a separate type of its own, and adding
HsSpliceTy constructor to HsType.

Trim exports

Typo in comment

Fix newtype deriving for Enum

Remove unused parameter to groupAvails

Dont report bindings beginning with _ as unused

Rule selection test was exactly backwards

Clarify newtype deriving documentation

Clarify warn-unused-bindings documention

comments only

Clean up Capability API
~~~~~~~~~~~~~~~~~~~~~~~

- yieldToReturningWorker() is now yieldCapability(), and performs all
  kinds of yielding (both to returning workers and passing to other
  OS threads).  yieldCapabiltiy() does *not* re-acquire a capability.

- waitForWorkCapabilty() is now waitForCapability().

- releaseCapbility() also releases the capability when passing to
  another OS thread.  It is the only way to release a capability (apart
  from yieldCapability(), which calls releaseCapability() internally).

- passCapability() and passCapabilityToWorker() now do not release the
  capability.  They just set a flag to indicate where the capabiliy
  should go when it it next released.


Other cleanups:

  - Removed all the SMP stuff from Schedule.c.  It had extensive bitrot,
    and was just obfuscating the code.  If it is ever needed again,
    it can be resurrected from CVS.

  - Removed some other dead code in Schedule.c, in an attempt to make
    this file more manageable.

ANSIfy

Add assertion.

Debugging output wibble

Fix a deadlock: an OS thread returning from a C call could enter
grabReturnCapability, grabbing the capability that was in the process
of being passed to another thread via passCapability.  This leads to a
deadlock shortly afterward, because the passCapability flag is still
set, so a normal worker won't pick up the capability when it is
released.

Fix (not sure if this is the best fix, though): don't grab the
capability in grabReturnCapability() if passCapabilty is set.

Fix locking bug in awaitEvent(): in one code path it could return with
sched_lock unlocked.

Fix bogosity in implementation of ACQUIRE_LOCK/RELEASE_LOCK on Win32.
These functions were essentially doing nothing, due to a missing
dereference on the argument.

I've rewritten them as inlines (to catch type errors) and added some
checking of the return values, which should help catch errors like
this in the future.