This lets the threaded RTS use SIGVTALRM rather than SIGALRM for its
interval timer signal, so the threaded and non-threaded RTS are
compatible.  It unfortunately doesn't completely fix #850/#1156, for
that we really have to use a restartable sleep instead of usleep().

Also I cleaned up the timer API a little: instead of returning an
error value that ultimately gets ignored, we now report errors from
system calls and exit.

We had a pattern that matched (base+(index<<N)), but not
(base+(index<<N)+M).

We weren't catching some literals that actually fit in 32 bits before,
because e.g. -1 was sometimes represented as 0xffffffffffffffff in CmmInt.

This patch was originally by dinko.tenev@gmail.com, but I re-recorded
it in order to add a better log message.

The effect of this patch is to add entries for primitive types in the
documentation: Int#, Char#, etc. and to document the built-in
identifiers (seq, lazy, inline, unsafeCoerce#).

There were two bugs:
 * we were generating the symbol name for the CAF
   cost centre from the OccName, which isn't unique enough
   in the case of system-generated non-external names
 * :Main.main caused problems, because we were assuming that
   every top-level CAF was from the current module.

This is necessary because GHC's recompilation checker doesn't check
across package boundaries.

We recently discovered that they aren't a win any more, and just cost
code size.

A RULE is supposed to match even if there is an intervening let:
	RULE f (x:xs) = ....
	target	   f (let x = thing in x:xs)

It's surprisingly tricky to get this right; in effect we are doing
let-floating on the fly. I managed to get it wrong before, or at least
be over-conservative.  And in "fixing" that I got it wrong again in a
different way, which made it far too conservative. In particular, it
failed to match	f (let x = y+y in let z=x+y in z:xs)
because the binder x was cloned and looked "locally-bound". See the
ever growing comments with the Let rule for details.

That patch reverts to the previous story, which is still a bit too 
conservative, but not so egregiously so.  Fixes Romans's problem.

From heatsink, in trac #1150.

This is a simplification & minor optimisation for GHCi

This patch adds data constructor names into their info tables. 
This is useful in the ghci debugger. It replaces the old scheme which
was based on tracking data con names in the linker. 

Previously one had to edit configure.ac to build a release version. This is
evil, so one can now specify this on the configure invocation line like:

   RELEASE=YES ./configure

Note that non-released versions are still the default.

My previous patch was creating problems with hs-boot files. 

This patch performs several optimizations with the goal of minimizing the cost of building the arguments to breakpointJump:
  - Group them all in a single tuple, to minimize closure creation in heap
  - Wrap this with the GHC.Base.lazy combinator, to induce max laziness
  - Remove as many literal strings as possible 
    * injecting a module-local CAF to store the module name and use that 
    * eliminating the package string (not needed).

It now outputs "<function>" instead of showing them as thunks

The breakpointJump functions never show up in the code that the typechecker sees, as they are inserted by the desugarer later.

Fixes Trac #1154.   Please merge.
Tests are tc225, and tcfail176.

Conal Eliott (Trac #1145) exposed a nasty flaw in the way in which
orphan instances are computed, when there are functional dependencies
in the class.  It took me some time to figure out what was going on,
and led to more refactoring.

Briefly:

* Elaborate comments about orphan-hood and versioning added to IfaceSyn
* The is_orph field vanishes from InstEnv.Instance
* Similarly ru_orph vanishes from CoreSyn.CoreRule
* Orphan-hood is computed in MkIface.instanceToIfaceInst, and
	MkIface.coreRuleToIfaceRule

Elsewhere just tidying up.

I forgot to to this when refactoring some months ago!

This one is a hangover from something I did a month or two ago, but
didn't get quite right.  tcSimplifyDefault should not check for no-instances;
instead the checkValidInstance in TcDeriv does so.

Conal's DeepArrow needs this fix.  Test is drv015.

This patch fixes the typechecking of the default declaration itself,
when overloaded strings are involved.  It also documents the behaviour
in the user manual.

nofib/spectral/power should work again now!