GitLab

Fixes Trac #1966. 

Suspensions in the Term datatype used for RTTI
always get assigned a Type, so there is no reason
to juggle around with a (Maybe Type) anymore. 

    
    This comes as result of the short discussion linked below.
    
    http://www.haskell.org/pipermail/cvs-ghc/2007-December/040049.html

Teach GHCi to show a "perhaps you forgot to use :trace?" when
it finds that the user is trying to retrieve an empty :history

- The information of whether given or wanted class dictionaries where
  normalised by rewriting wasn't always correctly propagated in TcTyFuns,
  which lead to malformed dictionary bindings.
- Also fixes a bug in TcPat.tcConPat where GADT equalities where emitted in
  the wrong position in case bindings (which led to CoreLint failures).

* This patch implements the use of equality constraints instead of GADT
  refinements that we have been discussing for a while.
* It just changes TcPat.tcConPat.  It doesn't have any of the simplification
  and dead code removal that is possible due to this change.
* At the moment, this patch breaks a fair number of GADT regression tests.

We also need to do the GHC.Prim hack when registering inplace or the
tests that use it fail.

    x && y
is not the same as
    if x; then y; fi
as the latter doesn't fail when x fails

This fixes building on NixOS. I'm not sure why it worked everywhere else,
but not on NixOS, before.

The old ones caused lots of
    unknown scattered relocation type 4
errors. Patch from Chris Kuklewicz.

Also fixes it with Solaris's sh, spotted by Christian Maeder

This fixes the 1959 test, but will do more recompilation than is
strictly necessary (but only when -O is on).  Still, more
recompilation is better than segfaults, link errors or other random
breakage.

We were building a mapping from ModuleName to [Occ] from the usage
list, using the usg_mod field as the key.  Unfortunately, due to a
very poor naming decision, usg_mod is actually the module version, not
the ModuleName.  usg_name is the ModuleName.  Since Version is also an
instance of Uniquable, there was no type error: all that happened was
lookups in the map never succeeded.  I shall rename the fields of
Usage in a separate patch.

This doesn't completely fix #1959, but it gets part of the way there.

I have to take partial blame as the person who wrote this fragment of
code in late 2006 (patch "Interface file optimisation and removal of
nameParent").

Prelude Data.IORef> :p l
l = (_t4::Maybe Integer) : (_t5::[Maybe Integer])
Prelude Data.IORef> p <- newIORef l
Prelude Data.IORef> :p p
p = GHC.IOBase.IORef (GHC.STRef.STRef {((_t6::Maybe Integer) :
                                        (_t7::[Maybe Integer]))})
Prelude Data.IORef> :sp p
p = GHC.IOBase.IORef (GHC.STRef.STRef {(_ : _)})


I used braces to denote the contents of a reference.
Perhaps there is a more appropriate notation?

Eta reduction was wrongly transforming
	f = \x. f x
to
	f = f

Solution: don't trust f's arity information; instead look at its
unfolding.  See Note [Eta reduction conditions]

Almost all the new lines are comments!

	
(No need to merge to 6.8, but no harm if a subsequent patch needs it.)

The proximate cause for this patch is to improve the inlining for INLINE
things that are not functions; this came up in the NDP project.  See
Note [Lone variables] in CoreUnfold.

This caused some refactoring that actually made things simpler.  In 
particular, more of the inlining logic has moved from SimplUtils to 
CoreUnfold, where it belongs.

using the new block-inheriting forkIO (#1048)