GitLab

Add dynlibdir target to config.mk.in, setting it to @libdir@.
Invoke installPackage with dynlibdir at libraries/Makefile
Make installPackage.hs hand dynlibdir to Cabal.

Rename interfacedir to haddockdir
Change empty(Copy|Register)Flags to default(Copy|Register)Flags
Wrap content of RegisterFlags with toFlag (the Flag type is actually just Maybe)

Added API support for compiling Haskell to simplified Core, and for
compiling Core to machine code. The latter, especially, should be
considered experimental and has only been given cursory testing. Also
fixed warnings in DriverPipeline. Merry Christmas.

In a few places we want to "look through" newtypes to get to the
representation type.  But we need to be careful that  we don't fall 
into an ininite loop with e.g.
	newtype T = MkT T

The old mechansim for doing this was to have a field nt_rep, inside 
a newtype TyCon, that gave the "ultimate representation" of the type.
But that failed for Trac 1495, which looked like this:
   newtype Fix a = Fix (a (Fix a))
   data I a = I a
Then, expanding the type (Fix I) went on for ever.

The right thing to do seems to be to check for loops when epxanding
the *type*, rather than in the *tycon*.  This patch does that, 
	- Removes nt_rep from TyCon
	- Make Type.repType check for loops
See Note [Expanding newtypes] in Type.lhs.

At the same time I also fixed a bug for Roman, where newtypes were not
being expanded properly in FamInstEnv.topNormaliseType.  This function
and Type.repType share a common structure.


	Ian, see if this merges easily to the branch
	If not, I don't think it's essential to fix 6.8

We were crashing when we saw
	case x of DEFAULT -> rhs
where x had a type-family type.  This patch fixes it.

MERGE to the 6.8 branch.

	MERGE to 6.8 branch

In computeChangedOccs we look up the old version of a Name.
But a WiredIn Name doesn't have an old version, because WiredIn things
don't appear in interface files at all.

Result: ghc-6.9: panic! (the 'impossible' happened)
  (GHC version 6.9 for x86_64-unknown-linux):
	lookupVers1 base:GHC.Prim chr#{v}

This fixes the problem.  The patch should merge easily onto the branch.

The ru_fn field was wrong when we moved RULES from one Id to another.
The fix is simple enough.

However, looking at this makes me realise that the worker/wrapper stuff
for recursive newtypes isn't very clever: we generate demand info but
then don't properly exploit it.  

This patch fixes the crash though.

  This patch implements generalised list comprehensions, as described in 
  the paper "Comprehensive comprehensions" (Peyton Jones & Wadler, Haskell
  Workshop 2007).  If you don't use the new comprehensions, nothing
  should change.
  
  The syntax is not exactly as in the paper; see the user manual entry 
  for details.
  
  You need an accompanying patch to the base library for this stuff 
  to work.
  
  The patch is the work of Max Bolingbroke [batterseapower@hotmail.com], 
  with some advice from Simon PJ.
  
  The related GHC Wiki page is 
    http://hackage.haskell.org/trac/ghc/wiki/SQLLikeComprehensions 

- Ensure the stage1 compiler uses ghc's own GMP library on Mac OS
- Need to rebuild installPackage and ifBuildable with stage1 compiler as they
  go into bindists

Patch from Christian Maeder

The library doc index thought that the docs were in $module.html, rather
than $package/$module.html.

We have to explicity check for empty arrays in each alternative as recursive
algorithms wouldn't terminate otherwise.

This is a temporary hack which allows us to vectorise literals.

This patch replaces -fspec-threshold by -fspec-constr-threshold and
-fliberate-case-threshold. The thresholds can be disabled by
-fno-spec-constr-threshold and -fno-liberate-case-threshold.

As well as fixing the immediate problem (Trac #1972) this patch does
a signficant simplification and refactoring of pattern renaming.

Fewer functions, fewer parameters passed....it's all good.  But it
took much longer than I expected to figure out.

The most significant change is that the NameMaker type does *binding*
as well as *making* and, in the matchNameMaker case, checks for unused
bindings as well.  This is much tider.

(No need to merge to the 6.8 branch, but no harm either.)

The number of iterations during the first run of phase 0 was erroneously
hardcoded to 3. It should be *at least* 3 (see comments in code) but can be
more.