(fixes various ffi test failures on x86.  This change was supposed to
be part of 9f61598c, but somehow it
got lost).

Previously we were completely ignoring these, due to the difficulties
of traversing the pinned blocks (the objects are not necessarily
end-to-end, we can't tell how large the gaps are).  Now just count the
whole block as a big ARR_WORDS, so at least we're accounting for the
memory and it has the right type.

a link to the Commentary where the current story is described.

When floating a case outwards we must be careful to clone
the binders, since their scope is widening.

Plus lots of tidying up.

It was inconsistent before, now it's right

trustworthy module in the same package.

In arrow commands, the function position in an application (HsApp)
is a command, not an expression.

as it maintains 16-byte alignment of the stack pointer (see #5250)

head-normal-form, and similarly for spark#.

  seq#   :: a -> State# s -> (# State# s, a #)
  spark# :: a -> State# s -> (# State# s, a #)

seq# is a version of seq that can be used in a State#-passing
context.  We will use it to implement Control.Exception.evaluate and
thus fix #5129.  Also we have plans to use it to fix #5262.

spark# is to seq# as par is to pseq.  That is, it creates a spark in a
State#-passing context.  We will use spark# and seq# to implement rpar
and rseq respectively in an improved implementation of the Eval monad.

HPC expects the end column of a span to be one less than the
convention used by SrcSpan, and we had lost the "-1", causing various
HPC tests to fail.

calling convention.

Patch based on one by Karel Gardas.

wrong platform by mistake.

The OS X ABI requires the C stack pointer to be 16-byte aligned at a
function call.  As far as I know this is not a requirement on other
x86 ABIs, but it seems that gcc is now generating SSE2 code that
assumes stack alignment (-mincoming-stack-boundary defaults to 4), so
we have to respect 16-byte alignment.

It looks like where x86 assembly is using '@' character,
ARM assembly requires '%' character. This makes a problem in the patch
814edf44 'Force re-linking if
the options have changed (#4451)' which makes linking assembly
file uncompilable on ARM. This patch fixes this.

There was an extra % in the strftime string used in the fingerprint
script to name the output files. It worked fine in python 2.6 on mac
os, but was producing bad file names in python 2.7 on linux.

There are two things in this patch. First, a new feature.
Given     (case x of I# y -> ...)
where 'x' is known to be evaluated, the float-out pass
will float the case outwards towards x's binding.  Of
course this doesn't happen if 'x' is evaluated because
of an enclosing case (becuase then the inner case would
be eliminated) but it *does* happen when x is bound by
a constructor with a strict field.  This happens in DPH.
Trac #4081.

The second change is a significant refactoring of the
way the let-floater works.  Now SetLevels makes a decision
about whether the let (or case) will move, and records
that decision in the FloatSpec flag.  This change makes
the whole caboodle much easier to think about.