GitLab

It controls the number of simplifier phases run during optimisation. These are
numbered from n to 1 (by default, n=2). Phase 0 is always run regardless of
this flag. The flag is ignored with -O0 since (practically) no optimisation is
performed in that case.

The float-out transformation must handle the case where a coercion
variable is free, which in turn mentions type variables in its kind.
Just like a term variable really.

I did a bit of refactoring at the same time.

Test is tc241

MERGE to stable branch

This patch (which is part of the fix for Trac #2018) makes coercion variables
be handled more uniformly.  Generally, they are treated like dictionaries
in the type checker, not like type variables, but in a couple of places we
were treating them like type variables.  Also when zonking we should use
zonkDictBndr not zonkIdBndr.

Somehow we didn't have a separate flag for impredicativity; now we do.

Furthermore, Trac #2019 showed up a missing test for monotypes in data
constructor return types.  And I realised that we were even allowing
things like
	Num (forall a. a) => ...
which we definitely should not.  

This patch insists on monotypes in several places where we were (wrongly)
too liberal before.

Could be merged to 6.8 but no big deal.

This allows the instance of UserOfLocalRegs to be within Haskell98, and IMHO
 makes the code a little cleaner generally.
This is one small (though tedious) step towards making GHC's code more
 portable...

was instance Outputable CmmGraph
type CmmGraph = LGraph Middle Last
now instance (ctx) => Outputable (LGraph m l),
in module with LGraph where it belongs
This also let us reduce the context of DebugNodes to Haskell98,
leaving that class's only extension being multi-parameter.
(also Outputable (LGraph M Last) with M = ExtendWithSpills Middle
was another redundant instance that was then removed)

UserOfLocalRegs (ZLast Last) isn't Haskell98, but it was just as
good an instance to be UserOfLocalRegs a => UserOfLocalRegs (ZLast a)

See Trac #2004, and Note [Flags and equational constraints] in TcPat.

To enable this, set UseLibFFI=YES in mk/build.mk.  

The main advantage here is that this reduces the porting effort for
new platforms: libffi works on more architectures than our current
adjustor code, and it is probably more heavily tested.  We could
potentially replace our existing code, but since it is probably faster
than libffi (just a guess, I'll measure later) and is already working,
it doesn't seem worthwhile.

Right now, you must have libffi installed on your system.  I used the
one supplied by Debian/Ubuntu.

This flag does nothing, and should have been removed ages ago. (GHC
no longer does update-in-place.)

MERGE to 6.8 branch

cmmMiniInline counts the uses of local variables, so it can easily
eliminate assigments to unused locals.  This almost never gets
triggered, as we don't generate any dead assignments, but it will be
needed by a forthcoming cleanup in CgUtils.emitSwitch.

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)