GitLab

This patch does a fairly major clean-up of the code that implements 'deriving.

* The big changes are in TcDeriv, which is dramatically cleaned up.
  In particular, there is a clear split into
	a) inference of instance contexts for deriving clauses
	b) generation of the derived code, given a context 
  Step (a) is skipped for standalone instance decls, which 
  have an explicitly provided context.

* The handling of "taggery", which is cooperative between TcDeriv and
  TcGenDeriv, is cleaned up a lot

* I have added documentation for standalone deriving (which was 
  previously wrong).

* The Haskell report is vague on exactly when a deriving clause should
  succeed.  Prodded by Conal I have loosened the rules slightly, thereyb
  making drv015 work again, and documented the rules in the user manual.

I believe this patch validates ok (once I've update the test suite)
and can go into the 6.8 branch.

trivColorable was soaking up total 31% time, 41% alloc when
compiling SHA1.lhs with -O2 -fregs-graph on x86.

Refactoring to use unboxed accumulators and walk directly
over the UniqFM holding the set of conflicts reduces this 
to 17% time, 6% alloc.

The type parameter to a C-- procedure now represents a control-flow
graph, not a single instruction.  The newtype ListGraph preserves the 
current representation while enabling other representations and a
sensible way of prettyprinting.  Except for a few changes in the
prettyprinter the new compiler binary should be bit-for-bit identical
to the old.

In fact hs-boot files had nothing to do with it: the problem was that
GHCi would forget the breakpoint information for a module that had
been reloaded but not recompiled.  It's amazing that we never noticed
this before.

The ModBreaks were in the ModDetails, which was the wrong place.  When
we avoid recompiling a module, ModDetails is regenerated from ModIface
by typecheckIface, and at that point it has no idea what the ModBreaks
should be, so typecheckIface made it empty.  The right place for the
ModBreaks to go is with the Linkable, which is retained when
compilation is avoided.  So now I've placed the ModBreaks in with the
CompiledByteCode, which also makes it clear that only byte-code
modules have breakpoints.

This fixes break022/break023

Two problems here: find needs to dereference symbolic links (-L
option, I really hope that's portable), and we need to notice when
aclocal.m4 is updated.  

Somehow I think this was easier when it just always ran
autoreconf... what was wrong with that?

Avoid coalescing nodes in the register conflict graph if the
new node will not be trivially colorable. Also remove the
front end aggressive coalescing pass.
  
For typical Haskell code the graph coloring allocator now does
about as well as the linear allocator.
  
For code with a large amount of register pressure it does much
better, but takes longer.
  
For SHA1.lhs from darcs on x86
   
          spills    reloads    reg-reg-moves
          inserted   inserted  left in code   compile-time
  linear    1068      1311        229            7.69(s)
  graph      387       902        340           16.12(s)

On OS X 10.4 and newer, we have to use dlsym because the old NS* interface has
been deprecated. The patch checks for HAVE_DLFCN_H instead of switching on
the OS version.

There is one additional quirk: although OS X prefixes global symbols with an
underscore, dlsym expects its argument NOT to have a leading underscore. As a
hack, we simply strip it off in lookupSymbol. Something a bit more elaborate
might be cleaner.

Previously failed to account for equational
class context for wanted dictionary contraints, e.g. wanted C a
in 

	class a ~ Int => C a
	instance C Int

should give rise to wanted a ~ Int and consequently discharge a ~ Int by
unifying a with Int and then discharge C Int with the instance.

All ancestor equalities are taken into account.

We then set datarootdir to something else later on so that things still
work when configure doesn't set it.

Formerly, the ghc-pkg was called to get the HTML dirs for other packages, but
of course doing this at *build* time is totally wrong. Now we use a relative
path, just like before. This is probably not perfect, but much better than
before.

As a sidenote: Cabal calls the relevant flag "html-location", ghc-pkg calls the
field "haddock-html", and Haddock itself uses it as part of "read-interface".
Too much creativity is sometimes a bad thing...

This applies to EnterCriticalSection and LeaveCriticalSection in the RTS

- This fixes the first part of #1651; ie, the panic in ghci.

Older GHCs can't parse OPTIONS_GHC.
This also changes the URL referenced for the -w options from
WorkingConventions#Warnings to CodingStyle#Warnings for the compiler
modules.

Patch from Audrey Tang.

Conservative and iterative coalescing come next.