Previously MVars were always on the mutable list of the old
generation, which meant every MVar was visited during every minor GC.
With lots of MVars hanging around, this gets expensive.  We addressed
this problem for MUT_VARs (aka IORefs) a while ago, the solution is to
use a traditional GC write-barrier when the object is modified.  This
patch does the same thing for MVars.

TVars are still done the old way, they could probably benefit from the
same treatment too.

This fixes the unreg build, and in general building the RTS code
via-C. I'm not sure at what stage this was broken, but I think it
was working accidentally before.

GHC spots that an 'otherwise' guard is true, and uses that knowledge to 
avoid reporting spurious missing-pattern or overlaps with -Wall.

The HPC ticks were disguising the 'otherwise', which led to this failure.
Now we check.  The key change is defining DsGRHSs.isTrueLHsExpr.

Test is ds062

There are a number of situations in which you aren't allowed to use
a quoted Name in a TH program, such as
	\x -> 'x
But we weren't checking for that!  Now we are.

Merge to stable branch.

Test is TH_qname.

Fix Trac #1678; be more careful about catching and reporting exceptions in spliced TH monadic computations

Many of the new lines are comments to explain the slightly-convoluted
in which exceptions get propagated out of the Q monad.

This fixes Trac 1679; test is TH_runIO (as well as the exising TH_fail).

Please merge

The extra safe points introduced for breakpoints were previously
compiled as normal updatable thunks, but they are guaranteed
single-entry, so we can use non-updatable thunks here.  This restores
the tail-call property where it was lost in some cases (although stack
squeezing probably often recovered it), and should improve
performance.

Previously inline candidates were given higher preference as
non-loop-breakers than constructor applications, but the reason for
this was that making a wrapper into a loop-breaker is to be avoided at
all costs.  This patch refines the algorithm slightly so that wrappers
are explicitly avoided by giving them a much higher score, and other
inline candidates are given lower scores than constructor
applications.

This makes almost zero difference to a complete nofib run, so it
amounts to just a tidyup.

The combination of "type refinement" for GADTs and the new equality
constraints has made TcSimplify rather complicated.  And wrong:
it generated bogus code for cholewo-eval.

This patch is still far from entirely satisfactory.  There are
too many places where invariants are unclear, and the code is 
still a bit of a mess.  But I believe it's better, and it passes
the regression tests!  So I think it's good enough for the 6.8 release.

Please merge.

The main changes are:

- get rid of extractLocalResults (which was always suspicious)

- instead, treat the 'refinement' along with 'givens', by 
  adding a field to RedEnv, red_reft which travels with red_givens

- I also reworked extractResults a bit, which looked wrong to me
  This entailed changing the Given constructor in Avail to take
  an Inst rather than a TcId

The Cast case of the rule-matcher was simply wrong.

This patch fixes it; see Trac #1746.

I also fixed the rule generation in SpecConstr to generate a wild-card
for the cast expression, which we don't want to match on.  This makes
the rule more widely applicable; it wasn't the cause of the bug.

This required moving PackageId from PackageConfig to Module

This fixes a problem with spurious recompilations: each time a module
was recompiled, the order of the children would change, causing extra
recompilation.

MERGE TO STABLE