GitLab

by passing the FamInstEnvs all the way down. This closes #7619.

Instead of first checking whether splitting is useful, and then firing
up the worker-wrapper-machinery, which will do the same checks again, we
now simply generate a worker and wrapper, and while doing so keep track
of whether what we did was in any way useful.

So now there is only one place left where we decide whether we want to
do w/w, and that place has access to more information, in particular the
actual types at hand.

Joachim and I are committing this onto a branch so that we can share it,
but we expect to do a bit more work before merging it onto head.

Nofib staus:
  - Most programs, no change
  - A few improve
  - A couple get worse (cacheprof, tak, rfib)
Investigating the "get worse" set is what's holding up putting this
on head.

The major issue is this.  Consider

    map (f g) ys

where f's demand signature looks like

   f :: <L,C1(C1(U))> -> <L,U> -> .

So 'f' is not saturated.  What demand do we place on g?
Answer
        C(C1(U))
That is, the inner C1 should stay, even though f is not saturated.

I found that this made a significant difference in the demand signatures
inferred in GHC.IO, which uses lots of higher-order exception handlers.

I also had to add used-once demand signatures for some of the
'catch' primops, so that we know their handlers are only called once.

We were re-using the super-magical "state token" type (which has
VoidRep and is zero bits wide) for situations in which we simply want
to lambda-abstract over a zero-bit argument. For example, join points:

   case (case x of { True -> e1; False -> e2 }) of
      Red  -> f1
      Blue -> True

==>

  let $j1 = \voidArg::Void# -> f1
  in
  case x of
    True -> case e1 of
              Red -> $j1 void
              Blue -> True
    False -> case e2 of
              Red -> $j1 void
              Blue -> True

This patch introduces

   * The new primitive type GHC.Prim.Void#, with PrimRep = VoidRep

   * A new global Id GHC.Prim.voidPrimId :: Void#.
     This has no binding because the code generator drops it,
     but is used as an argument (eg in the call of $j1)

   * A new local Id, MkId.voidArgId, which can be lambda-bound
     when you need to lambda-abstract over it.

and uses them throughout.

Now the State# thing is used only when we need state!

Fixes Trac #8037

A bit of a mess had accumulated, with unclear invariants.

* Remove splitNewTypeRepCo_maybe, in favour of topNormaliseNewType_maybe
  (which had the same signature but behaved subtly differently).

* Make topNormaliseNewType_maybe guaranteed to return a non-newtype
  if it says (Just ty).  This is what was causing the loop in #8467

* Apply similar tidying up to FamInstEnv.topNormaliseType

Roles are a solution to the GeneralizedNewtypeDeriving type-safety
problem.

Roles were first described in the "Generative type abstraction" paper,
by Stephanie Weirich, Dimitrios Vytiniotis, Simon PJ, and Steve Zdancewic.
The implementation is a little different than that paper. For a quick
primer, check out Note [Roles] in Coercion. Also see
http://ghc.haskell.org/trac/ghc/wiki/Roles
and
http://ghc.haskell.org/trac/ghc/wiki/RolesImplementation
For a more formal treatment, check out docs/core-spec/core-spec.pdf.

This fixes Trac #1496, #4846, #7148.

size_expr now ignores RealWorld lambdas, arguments, and applications.

Worker-wrapper previously removed all lambdas from a function, if they
were all unused. Removing *all* value lambdas is no longer
allowed. Instead (\_ -> E) will become (\_void -> E), where it used to
become E. The previous behavior can be recovered via the new
-ffun-to-thunk flag.

Nofib notables:

----------------------------------------------------------------
        Program               O2          O2 newly ignoring RealWorld
                                          and not turning function
                                          closures into thunks
----------------------------------------------------------------

 Allocations

  comp_lab_zift            333090392%           -5.0%
reverse-complem            155188304%           -3.2%

        rewrite             15380888%           +4.0%
         boyer2              3901064%           +7.5%

rewrite previously benefited from fortunate LoopBreaker choice that is
now disrupted.

A function in boyer2 goes from $wonewayunify1 size 700 to size 650,
thus gets inlined into rewritelemmas, thus exposing a parameter
scrutinisation, thus allowing SpecConstr, which unfortunately involves
reboxing.

Run Time

 fannkuch-redux                 7.89%          -15.9%

            hpg                 0.25%           +5.6%
           wang                 0.21%           +5.8%

/shrug

The main payload of this patch is to extend CPR so that it
detects when a function always returns a result constructed
with the *same* constructor, even if the constructor comes from
a sum type.  This doesn't matter very often, but it does improve
some things (results below).

Binary sizes increase a little bit, I think because there are more
wrappers.  This with -split-objs.  Without split-ojbs binary sizes
increased by 6% even for HelloWorld.hs.  It's hard to see exactly why,
but I think it was because System.Posix.Types.o got included in the
linked binary, whereas it didn't before.

        Program           Size    Allocs   Runtime   Elapsed  TotalMem
          fluid          +1.8%     -0.3%      0.01      0.01     +0.0%
            tak          +2.2%     -0.2%      0.02      0.02     +0.0%
           ansi          +1.7%     -0.3%      0.00      0.00     +0.0%
      cacheprof          +1.6%     -0.3%     +0.6%     +0.5%     +1.4%
        parstof          +1.4%     -4.4%      0.00      0.00     +0.0%
        reptile          +2.0%     +0.3%      0.02      0.02     +0.0%
----------------------------------------------------------------------
            Min          +1.1%     -4.4%     -4.7%     -4.7%    -15.0%
            Max          +2.3%     +0.3%     +8.3%     +9.4%    +50.0%
 Geometric Mean          +1.9%     -0.1%     +0.6%     +0.7%     +0.3%

Other things in this commit
~~~~~~~~~~~~~~~~~~~~~~~~~~~
* Got rid of the Lattice class in Demand

* Refactored the way that products and newtypes are
  decomposed (no change in functionality)

This patch is the result of Ilya Sergey's internship at MSR.  It
constitutes a thorough overhaul and simplification of the demand
analyser.  It makes a solid foundation on which we can now build.
Main changes are

* Instead of having one combined type for Demand, a Demand is
   now a pair (JointDmd) of
      - a StrDmd and
      - an AbsDmd.
   This allows strictness and absence to be though about quite
   orthogonally, and greatly reduces brain melt-down.

* Similarly in the DmdResult type, it's a pair of
     - a PureResult (indicating only divergence/non-divergence)
     - a CPRResult (which deals only with the CPR property

* In IdInfo, the
    strictnessInfo field contains a StrictSig, not a Maybe StrictSig
    demandInfo     field contains a Demand, not a Maybe Demand
  We don't need Nothing (to indicate no strictness/demand info)
  any more; topSig/topDmd will do.

* Remove "boxity" analysis entirely.  This was an attempt to
  avoid "reboxing", but it added complexity, is extremely
  ad-hoc, and makes very little difference in practice.

* Remove the "unboxing strategy" computation. This was an an
  attempt to ensure that a worker didn't get zillions of
  arguments by unboxing big tuples.  But in fact removing it
  DRAMATICALLY reduces allocation in an inner loop of the
  I/O library (where the threshold argument-count had been
  set just too low).  It's exceptional to have a zillion arguments
  and I don't think it's worth the complexity, especially since
  it turned out to have a serious performance hit.

* Remove quite a bit of ad-hoc cruft

* Move worthSplittingFun, worthSplittingThunk from WorkWrap to
  Demand. This allows JointDmd to be fully abstract, examined
  only inside Demand.

Everything else really follows from these changes.

All of this is really just refactoring, so we don't expect
big performance changes, but acutally the numbers look quite
good.  Here is a full nofib run with some highlights identified:

        Program           Size    Allocs   Runtime   Elapsed  TotalMem
--------------------------------------------------------------------------------
         expert          -2.6%    -15.5%      0.00      0.00     +0.0%
          fluid          -2.4%     -7.1%      0.01      0.01     +0.0%
             gg          -2.5%    -28.9%      0.02      0.02    -33.3%
      integrate          -2.6%     +3.2%     +2.6%     +2.6%     +0.0%
        mandel2          -2.6%     +4.2%      0.01      0.01     +0.0%
       nucleic2          -2.0%    -16.3%      0.11      0.11     +0.0%
           para          -2.6%    -20.0%    -11.8%    -11.7%     +0.0%
         parser          -2.5%    -17.9%      0.05      0.05     +0.0%
         prolog          -2.6%    -13.0%      0.00      0.00     +0.0%
         puzzle          -2.6%     +2.2%     +0.8%     +0.8%     +0.0%
        sorting          -2.6%    -35.9%      0.00      0.00     +0.0%
       treejoin          -2.6%    -52.2%     -9.8%     -9.9%     +0.0%
--------------------------------------------------------------------------------
            Min          -2.7%    -52.2%    -11.8%    -11.7%    -33.3%
            Max          -1.8%     +4.2%    +10.5%    +10.5%     +7.7%
 Geometric Mean          -2.5%     -2.8%     -0.4%     -0.5%     -0.4%

Things to note

* Binary sizes are smaller. I don't know why, but it's good.

* Allocation is sometiemes a *lot* smaller. I believe that all the big numbers
  (I checked treejoin, gg, sorting) arise from one place, namely a function
  GHC.IO.Encoding.UTF8.utf8_decode, which is strict in two Buffers both of
  which have several arugments.  Not w/w'ing both arguments (which is what
  we did before) has a big effect.  So the big win in actually somewhat
  accidental, gained by removing the "unboxing strategy" code.

* A couple of benchmarks allocate slightly more.  This turns out
  to be due to reboxing (integrate).  But the biggest increase is
  mandel2, and *that* turned out also to be a somewhat accidental
  loss of CSE, and pointed the way to doing better CSE: see Trac
  #7596.

* Runtimes are never very reliable, but seem to improve very slightly.

All in all, a good piece of work.  Thank you Ilya!

The situation was pretty dire.  The way in which data constructors
were handled, notably the mapping between their *source* argument types
and their *representation* argument types (after seq'ing and unpacking)
was scattered in three different places, and hard to keep in sync.

Now it is all in one place:

 * The dcRep field of a DataCon gives its representation,
   specified by a DataConRep

 * As well as having the wrapper, the DataConRep has a "boxer"
   of type DataConBoxer (defined in MkId for loopy reasons).
   The boxer used at a pattern match to reconstruct the source-level
   arguments from the rep-level bindings in the pattern match.

 * The unboxing in the wrapper and the boxing in the boxer are dual,
   and are now constructed together, by MkId.mkDataConRep. This is
   the key function of this change.

 * All the computeBoxingStrategy code in TcTyClsDcls disappears.

Much nicer.

There is a little bit of refactoring left to do; the strange
deepSplitProductType functions are now called only in WwLib, so
I moved them there, and I think they could be tidied up further.

A long-standing and egregious bug in the worker/wrapper code meant
that some functions with the CPR property weren't getting a CPR
w/w. And that had the effect of making a tail-recursive function not
tail recursive.  As well as increasing allocation.

Fixes Trac #5920, and #5997.

Nofib results (highlights):

        Program           Size    Allocs   Runtime   Elapsed  TotalMem
--------------------------------------------------------------------------------
         boyer2          -0.1%    -15.3%      0.01      0.01     +0.0%
        mandel2          -0.0%     -8.1%      0.01      0.01     +0.0%
           para          -0.1%    -11.8%     -7.9%     -7.8%     +0.0%
--------------------------------------------------------------------------------
            Min          -0.1%    -15.3%     -7.9%     -7.8%    -33.3%
            Max          +0.0%     +0.2%     +6.3%     +6.3%     +3.7%
 Geometric Mean          -0.0%     -0.4%     +0.1%     +0.1%     -0.5%

Looks like a clear win.  And I have not even recompiled the libraries, so
it'll probably be a bit better in the ed.

We only use it for "compiler" sources, i.e. not for libraries.
Many modules have a -fno-warn-tabs kludge for now.

User visible changes
====================

Profilng
--------

Flags renamed (the old ones are still accepted for now):

  OLD            NEW
  ---------      ------------
  -auto-all      -fprof-auto
  -auto          -fprof-exported
  -caf-all       -fprof-cafs

New flags:

  -fprof-auto              Annotates all bindings (not just top-level
                           ones) with SCCs

  -fprof-top               Annotates just top-level bindings with SCCs

  -fprof-exported          Annotates just exported bindings with SCCs

  -fprof-no-count-entries  Do not maintain entry counts when profiling
                           (can make profiled code go faster; useful with
                           heap profiling where entry counts are not used)

Cost-centre stacks have a new semantics, which should in most cases
result in more useful and intuitive profiles.  If you find this not to
be the case, please let me know.  This is the area where I have been
experimenting most, and the current solution is probably not the
final version, however it does address all the outstanding bugs and
seems to be better than GHC 7.2.

Stack traces
------------

+RTS -xc now gives more information.  If the exception originates from
a CAF (as is common, because GHC tends to lift exceptions out to the
top-level), then the RTS walks up the stack and reports the stack in
the enclosing update frame(s).

Result: +RTS -xc is much more useful now - but you still have to
compile for profiling to get it.  I've played around a little with
adding 'head []' to GHC itself, and +RTS -xc does pinpoint the problem
quite accurately.

I plan to add more facilities for stack tracing (e.g. in GHCi) in the
future.

Coverage (HPC)
--------------

 * derived instances are now coloured yellow if they weren't used
 * likewise record field names
 * entry counts are more accurate (hpc --fun-entry-count)
 * tab width is now correct (markup was previously off in source with
   tabs)

Internal changes
================

In Core, the Note constructor has been replaced by

        Tick (Tickish b) (Expr b)

which is used to represent all the kinds of source annotation we
support: profiling SCCs, HPC ticks, and GHCi breakpoints.

Depending on the properties of the Tickish, different transformations
apply to Tick.  See CoreUtils.mkTick for details.

Tickets
=======

This commit closes the following tickets, test cases to follow:

  - Close #2552: not a bug, but the behaviour is now more intuitive
    (test is T2552)

  - Close #680 (test is T680)

  - Close #1531 (test is result001)

  - Close #949 (test is T949)

  - Close #2466: test case has bitrotted (doesn't compile against current
    version of vector-space package)

Basically as documented in http://hackage.haskell.org/trac/ghc/wiki/KindFact,
this patch adds a new kind Constraint such that:

  Show :: * -> Constraint
  (?x::Int) :: Constraint
  (Int ~ a) :: Constraint

And you can write *any* type with kind Constraint to the left of (=>):
even if that type is a type synonym, type variable, indexed type or so on.

The following (somewhat related) changes are also made:
 1. We now box equality evidence. This is required because we want
    to give (Int ~ a) the *lifted* kind Constraint
 2. For similar reasons, implicit parameters can now only be of
    a lifted kind. (?x::Int#) => ty is now ruled out
 3. Implicit parameter constraints are now allowed in superclasses
    and instance contexts (this just falls out as OK with the new
    constraint solver)

Internally the following major changes were made:
 1. There is now no PredTy in the Type data type. Instead
    GHC checks the kind of a type to figure out if it is a predicate
 2. There is now no AClass TyThing: we represent classes as TyThings
    just as a ATyCon (classes had TyCons anyway)
 3. What used to be (~) is now pretty-printed as (~#). The box
    constructor EqBox :: (a ~# b) -> (a ~ b)
 4. The type LCoercion is used internally in the constraint solver
    and type checker to represent coercions with free variables
    of type (a ~ b) rather than (a ~# b)

These turn out to be a useful special case of splitTyConApp_maybe.

A refactoring only; no change in behaviour

A coercion variable in a term must be wrapped in a coercion!
(Led to lint errors.)

See the paper "Practical aspects of evidence based compilation in System FC"

* Coercion becomes a data type, distinct from Type

* Coercions become value-level things, rather than type-level things,
  (although the value is zero bits wide, like the State token)
  A consequence is that a coerion abstraction increases the arity by 1
  (just like a dictionary abstraction)

* There is a new constructor in CoreExpr, namely Coercion, to inject
  coercions into terms

This is what was giving the "absent entered" messages
See Note [Absent errors] in WwLib.  We now return a 
suitable literal for absent values of unlifted type.

Previously we were simply passing arguments of unlifted
type to a wrapper, even if they were absent, which was
stupid.

See Note [Absent error Id] in WwLib.

This major patch implements the new OutsideIn constraint solving
algorithm in the typecheker, following our JFP paper "Modular type
inference with local assumptions".  

Done with major help from Dimitrios Vytiniotis and Brent Yorgey.

I finally got tired of the #ifdef OLD_STRICTNESS stuff.  I had been
keeping it around in the hope of doing old-to-new comparisions, but
have failed to do so for many years, so I don't think it's going to
happen.  This patch deletes the clutter.

This patch has been a long time in gestation and has, as a
result, accumulated some extra bits and bobs that are only
loosely related.  I separated the bits that are easy to split
off, but the rest comes as one big patch, I'm afraid.

Note that:
 * It comes together with a patch to the 'base' library
 * Interface file formats change slightly, so you need to
   recompile all libraries

The patch is mainly giant tidy-up, driven in part by the
particular stresses of the Data Parallel Haskell project. I don't
expect a big performance win for random programs.  Still, here are the
nofib results, relative to the state of affairs without the patch

        Program           Size    Allocs   Runtime   Elapsed
--------------------------------------------------------------------------------
            Min         -12.7%    -14.5%    -17.5%    -17.8%
            Max          +4.7%    +10.9%     +9.1%     +8.4%
 Geometric Mean          +0.9%     -0.1%     -5.6%     -7.3%

The +10.9% allocation outlier is rewrite, which happens to have a
very delicate optimisation opportunity involving an interaction
of CSE and inlining (see nofib/Simon-nofib-notes). The fact that
the 'before' case found the optimisation is somewhat accidental.
Runtimes seem to go down, but I never kno wwhether to really trust
this number.  Binary sizes wobble a bit, but nothing drastic.


The Main Ideas are as follows.

InlineRules
~~~~~~~~~~~
When you say 
      {-# INLINE f #-}
      f x = <rhs>
you intend that calls (f e) are replaced by <rhs>[e/x] So we
should capture (\x.<rhs>) in the Unfolding of 'f', and never meddle
with it.  Meanwhile, we can optimise <rhs> to our heart's content,
leaving the original unfolding intact in Unfolding of 'f'.

So the representation of an Unfolding has changed quite a bit
(see CoreSyn).  An INLINE pragma gives rise to an InlineRule 
unfolding.  

Moreover, it's only used when 'f' is applied to the
specified number of arguments; that is, the number of argument on 
the LHS of the '=' sign in the original source definition. 
For example, (.) is now defined in the libraries like this
   {-# INLINE (.) #-}
   (.) f g = \x -> f (g x)
so that it'll inline when applied to two arguments. If 'x' appeared
on the left, thus
   (.) f g x = f (g x)
it'd only inline when applied to three arguments.  This slightly-experimental
change was requested by Roman, but it seems to make sense.

Other associated changes

* Moving the deck chairs in DsBinds, which processes the INLINE pragmas

* In the old system an INLINE pragma made the RHS look like
   (Note InlineMe <rhs>)
  The Note switched off optimisation in <rhs>.  But it was quite
  fragile in corner cases. The new system is more robust, I believe.
  In any case, the InlineMe note has disappeared 

* The workerInfo of an Id has also been combined into its Unfolding,
  so it's no longer a separate field of the IdInfo.

* Many changes in CoreUnfold, esp in callSiteInline, which is the critical
  function that decides which function to inline.  Lots of comments added!

* exprIsConApp_maybe has moved to CoreUnfold, since it's so strongly
  associated with "does this expression unfold to a constructor application".
  It can now do some limited beta reduction too, which Roman found 
  was an important.

Instance declarations
~~~~~~~~~~~~~~~~~~~~~
It's always been tricky to get the dfuns generated from instance
declarations to work out well.  This is particularly important in 
the Data Parallel Haskell project, and I'm now on my fourth attempt,
more or less.

There is a detailed description in TcInstDcls, particularly in
Note [How instance declarations are translated].   Roughly speaking
we now generate a top-level helper function for every method definition
in an instance declaration, so that the dfun takes a particularly
stylised form:
  dfun a d1 d2 = MkD (op1 a d1 d2) (op2 a d1 d2) ...etc...

In fact, it's *so* stylised that we never need to unfold a dfun.
Instead ClassOps have a special rewrite rule that allows us to
short-cut dictionary selection.  Suppose dfun :: Ord a -> Ord [a]
                                            d :: Ord a
Then   
    compare (dfun a d)  -->   compare_list a d 
in one rewrite, without first inlining the 'compare' selector
and the body of the dfun.

To support this
a) ClassOps have a BuiltInRule (see MkId.dictSelRule)
b) DFuns have a special form of unfolding (CoreSyn.DFunUnfolding)
   which is exploited in CoreUnfold.exprIsConApp_maybe

Implmenting all this required a root-and-branch rework of TcInstDcls
and bits of TcClassDcl.


Default methods
~~~~~~~~~~~~~~~
If you give an INLINE pragma to a default method, it should be just
as if you'd written out that code in each instance declaration, including
the INLINE pragma.  I think that it now *is* so.  As a result, library
code can be simpler; less duplication.


The CONLIKE pragma
~~~~~~~~~~~~~~~~~~
In the DPH project, Roman found cases where he had

   p n k = let x = replicate n k
           in ...(f x)...(g x)....

   {-# RULE f (replicate x) = f_rep x #-}

Normally the RULE would not fire, because doing so involves 
(in effect) duplicating the redex (replicate n k).  A new
experimental modifier to the INLINE pragma, {-# INLINE CONLIKE
replicate #-}, allows you to tell GHC to be prepared to duplicate
a call of this function if it allows a RULE to fire.

See Note [CONLIKE pragma] in BasicTypes


Join points
~~~~~~~~~~~
See Note [Case binders and join points] in Simplify


Other refactoring
~~~~~~~~~~~~~~~~~
* I moved endPass from CoreLint to CoreMonad, with associated jigglings

* Better pretty-printing of Core

* The top-level RULES (ones that are not rules for locally-defined things)
  are now substituted on every simplifier iteration.  I'm not sure how
  we got away without doing this before.  This entails a bit more plumbing
  in SimplCore.

* The necessary stuff to serialise and deserialise the new
  info across interface files.

* Something about bottoming floats in SetLevels
      Note [Bottoming floats]

* substUnfolding has moved from SimplEnv to CoreSubs, where it belongs


--------------------------------------------------------------------------------
        Program           Size    Allocs   Runtime   Elapsed
--------------------------------------------------------------------------------
           anna          +2.4%     -0.5%      0.16      0.17
           ansi          +2.6%     -0.1%      0.00      0.00
           atom          -3.8%     -0.0%     -1.0%     -2.5%
         awards          +3.0%     +0.7%      0.00      0.00
         banner          +3.3%     -0.0%      0.00      0.00
     bernouilli          +2.7%     +0.0%     -4.6%     -6.9%
          boyer          +2.6%     +0.0%      0.06      0.07
         boyer2          +4.4%     +0.2%      0.01      0.01
           bspt          +3.2%     +9.6%      0.02      0.02
      cacheprof          +1.4%     -1.0%    -12.2%    -13.6%
       calendar          +2.7%     -1.7%      0.00      0.00
       cichelli          +3.7%     -0.0%      0.13      0.14
        circsim          +3.3%     +0.0%     -2.3%     -9.9%
       clausify          +2.7%     +0.0%      0.05      0.06
  comp_lab_zift          +2.6%     -0.3%     -7.2%     -7.9%
       compress          +3.3%     +0.0%     -8.5%     -9.6%
      compress2          +3.6%     +0.0%    -15.1%    -17.8%
    constraints          +2.7%     -0.6%    -10.0%    -10.7%
   cryptarithm1          +4.5%     +0.0%     -4.7%     -5.7%
   cryptarithm2          +4.3%    -14.5%      0.02      0.02
            cse          +4.4%     -0.0%      0.00      0.00
          eliza          +2.8%     -0.1%      0.00      0.00
          event          +2.6%     -0.0%     -4.9%     -4.4%
         exp3_8          +2.8%     +0.0%     -4.5%     -9.5%
         expert          +2.7%     +0.3%      0.00      0.00
            fem          -2.0%     +0.6%      0.04      0.04
            fft          -6.0%     +1.8%      0.05      0.06
           fft2          -4.8%     +2.7%      0.13      0.14
       fibheaps          +2.6%     -0.6%      0.05      0.05
           fish          +4.1%     +0.0%      0.03      0.04
          fluid          -2.1%     -0.2%      0.01      0.01
         fulsom          -4.8%     +9.2%     +9.1%     +8.4%
         gamteb          -7.1%     -1.3%      0.10      0.11
            gcd          +2.7%     +0.0%      0.05      0.05
    gen_regexps          +3.9%     -0.0%      0.00      0.00
         genfft          +2.7%     -0.1%      0.05      0.06
             gg          -2.7%     -0.1%      0.02      0.02
           grep          +3.2%     -0.0%      0.00      0.00
         hidden          -0.5%     +0.0%    -11.9%    -13.3%
            hpg          -3.0%     -1.8%     +0.0%     -2.4%
            ida          +2.6%     -1.2%      0.17     -9.0%
          infer          +1.7%     -0.8%      0.08      0.09
        integer          +2.5%     -0.0%     -2.6%     -2.2%
      integrate          -5.0%     +0.0%     -1.3%     -2.9%
        knights          +4.3%     -1.5%      0.01      0.01
           lcss          +2.5%     -0.1%     -7.5%     -9.4%
           life          +4.2%     +0.0%     -3.1%     -3.3%
           lift          +2.4%     -3.2%      0.00      0.00
      listcompr          +4.0%     -1.6%      0.16      0.17
       listcopy          +4.0%     -1.4%      0.17      0.18
       maillist          +4.1%     +0.1%      0.09      0.14
         mandel          +2.9%     +0.0%      0.11      0.12
        mandel2          +4.7%     +0.0%      0.01      0.01
        minimax          +3.8%     -0.0%      0.00      0.00
        mkhprog          +3.2%     -4.2%      0.00      0.00
     multiplier          +2.5%     -0.4%     +0.7%     -1.3%
       nucleic2          -9.3%     +0.0%      0.10      0.10
           para          +2.9%     +0.1%     -0.7%     -1.2%
      paraffins         -10.4%     +0.0%      0.20     -1.9%
         parser          +3.1%     -0.0%      0.05      0.05
        parstof          +1.9%     -0.0%      0.00      0.01
            pic          -2.8%     -0.8%      0.01      0.02
          power          +2.1%     +0.1%     -8.5%     -9.0%
         pretty         -12.7%     +0.1%      0.00      0.00
         primes          +2.8%     +0.0%      0.11      0.11
      primetest          +2.5%     -0.0%     -2.1%     -3.1%
         prolog          +3.2%     -7.2%      0.00      0.00
         puzzle          +4.1%     +0.0%     -3.5%     -8.0%
         queens          +2.8%     +0.0%      0.03      0.03
        reptile          +2.2%     -2.2%      0.02      0.02
        rewrite          +3.1%    +10.9%      0.03      0.03
           rfib          -5.2%     +0.2%      0.03      0.03
            rsa          +2.6%     +0.0%      0.05      0.06
            scc          +4.6%     +0.4%      0.00      0.00
          sched          +2.7%     +0.1%      0.03      0.03
            scs          -2.6%     -0.9%     -9.6%    -11.6%
         simple          -4.0%     +0.4%    -14.6%    -14.9%
          solid          -5.6%     -0.6%     -9.3%    -14.3%
        sorting          +3.8%     +0.0%      0.00      0.00
         sphere          -3.6%     +8.5%      0.15      0.16
         symalg          -1.3%     +0.2%      0.03      0.03
            tak          +2.7%     +0.0%      0.02      0.02
      transform          +2.0%     -2.9%     -8.0%     -8.8%
       treejoin          +3.1%     +0.0%    -17.5%    -17.8%
      typecheck          +2.9%     -0.3%     -4.6%     -6.6%
        veritas          +3.9%     -0.3%      0.00      0.00
           wang          -6.2%     +0.0%      0.18     -9.8%
      wave4main         -10.3%     +2.6%     -2.1%     -2.3%
   wheel-sieve1          +2.7%     -0.0%     +0.3%     -0.6%
   wheel-sieve2          +2.7%     +0.0%     -3.7%     -7.5%
           x2n1          -4.1%     +0.1%      0.03      0.04
--------------------------------------------------------------------------------
            Min         -12.7%    -14.5%    -17.5%    -17.8%
            Max          +4.7%    +10.9%     +9.1%     +8.4%
 Geometric Mean          +0.9%     -0.1%     -5.6%     -7.3%

This biggish patch addresses Trac #2670.  The main effect is to make
record selectors into ordinary functions, whose unfoldings appear in
interface files, in contrast to their previous existence as magic
"implicit Ids".  This means that the usual machinery of optimisation,
analysis, and inlining applies to them, which was failing before when
the selector was somewhat complicated.  (Which it can be when
strictness annotations, unboxing annotations, and GADTs are involved.)

The change involves the following points

* Changes in Var.lhs to the representation of Var.  Now a LocalId can
  have an IdDetails as well as a GlobalId.  In particular, the
  information that an Id is a record selector is kept in the
  IdDetails.  While compiling the current module, the record selector
  *must* be a LocalId, so that it participates properly in compilation
  (free variables etc).

  This led me to change the (hidden) representation of Var, so that there
  is now only one constructor for Id, not two.

* The IdDetails is persisted into interface files, so that an
  importing module can see which Ids are records selectors.

* In TcTyClDecls, we generate the record-selector bindings in renamed,
  but not typechecked form.  In this way, we can get the typechecker
  to add all the types and so on, which is jolly helpful especially
  when GADTs or type families are involved.  Just like derived
  instance declarations.

  This is the big new chunk of 180 lines of code (much of which is
  commentary).  A call to the same function, mkAuxBinds, is needed in
  TcInstDcls for associated types.

* The typechecker therefore has to pin the correct IdDetails on to 
  the record selector, when it typechecks it.  There was a neat way
  to do this, by adding a new sort of signature to HsBinds.Sig, namely
  IdSig.  This contains an Id (with the correct Name, Type, and IdDetails);
  the type checker uses it as the binder for the final binding.  This
  worked out rather easily.

* Record selectors are no longer "implicit ids", which entails changes to
     IfaceSyn.ifaceDeclSubBndrs
     HscTypes.implicitTyThings
     TidyPgm.getImplicitBinds
  (These three functions must agree.)

* MkId.mkRecordSelectorId is deleted entirely, some 300+ lines (incl
  comments) of very error prone code.  Happy days.

* A TyCon no longer contains the list of record selectors: 
  algTcSelIds is gone

The renamer is unaffected, including the way that import and export of
record selectors is handled.

Other small things

* IfaceSyn.ifaceDeclSubBndrs had a fragile test for whether a data
  constructor had a wrapper.  I've replaced that with an explicit flag
  in the interface file. More robust I hope.

* I renamed isIdVar to isId, which touched a few otherwise-unrelated files.

rolling back:

Fri Dec  5 16:54:00 GMT 2008  simonpj@microsoft.com
  * Completely new treatment of INLINE pragmas (big patch)
  
  This is a major patch, which changes the way INLINE pragmas work.
  Although lots of files are touched, the net is only +21 lines of
  code -- and I bet that most of those are comments!
  
  HEADS UP: interface file format has changed, so you'll need to
  recompile everything.
  
  There is not much effect on overall performance for nofib, 
  probably because those programs don't make heavy use of INLINE pragmas.
  
          Program           Size    Allocs   Runtime   Elapsed
              Min         -11.3%     -6.9%     -9.2%     -8.2%
              Max          -0.1%     +4.6%     +7.5%     +8.9%
   Geometric Mean          -2.2%     -0.2%     -1.0%     -0.8%
  
  (The +4.6% for on allocs is cichelli; see other patch relating to
  -fpass-case-bndr-to-join-points.)
  
  The old INLINE system
  ~~~~~~~~~~~~~~~~~~~~~
  The old system worked like this. A function with an INLINE pragam
  got a right-hand side which looked like
       f = __inline_me__ (\xy. e)
  The __inline_me__ part was an InlineNote, and was treated specially
  in various ways.  Notably, the simplifier didn't inline inside an
  __inline_me__ note.  
  
  As a result, the code for f itself was pretty crappy. That matters
  if you say (map f xs), because then you execute the code for f,
  rather than inlining a copy at the call site.
  
  The new story: InlineRules
  ~~~~~~~~~~~~~~~~~~~~~~~~~~
  The new system removes the InlineMe Note altogether.  Instead there
  is a new constructor InlineRule in CoreSyn.Unfolding.  This is a 
  bit like a RULE, in that it remembers the template to be inlined inside
  the InlineRule.  No simplification or inlining is done on an InlineRule,
  just like RULEs.  
  
  An Id can have an InlineRule *or* a CoreUnfolding (since these are two
  constructors from Unfolding). The simplifier treats them differently:
  
    - An InlineRule is has the substitution applied (like RULES) but 
      is otherwise left undisturbed.
  
    - A CoreUnfolding is updated with the new RHS of the definition,
      on each iteration of the simplifier.
  
  An InlineRule fires regardless of size, but *only* when the function
  is applied to enough arguments.  The "arity" of the rule is specified
  (by the programmer) as the number of args on the LHS of the "=".  So
  it makes a difference whether you say
    	{-# INLINE f #-}
  	f x = \y -> e     or     f x y = e
  This is one of the big new features that InlineRule gives us, and it
  is one that Roman really wanted.
  
  In contrast, a CoreUnfolding can fire when it is applied to fewer
  args than than the function has lambdas, provided the result is small
  enough.
  
  
  Consequential stuff
  ~~~~~~~~~~~~~~~~~~~
  * A 'wrapper' no longer has a WrapperInfo in the IdInfo.  Instead,
    the InlineRule has a field identifying wrappers.
  
  * Of course, IfaceSyn and interface serialisation changes appropriately.
  
  * Making implication constraints inline nicely was a bit fiddly. In
    the end I added a var_inline field to HsBInd.VarBind, which is why
    this patch affects the type checker slightly
  
  * I made some changes to the way in which eta expansion happens in
    CorePrep, mainly to ensure that *arguments* that become let-bound
    are also eta-expanded.  I'm still not too happy with the clarity
    and robustness fo the result.
  
  * We now complain if the programmer gives an INLINE pragma for
    a recursive function (prevsiously we just ignored it).  Reason for
    change: we don't want an InlineRule on a LoopBreaker, because then
    we'd have to check for loop-breaker-hood at occurrence sites (which
    isn't currenlty done).  Some tests need changing as a result.
  
  This patch has been in my tree for quite a while, so there are
  probably some other minor changes.
  

    M ./compiler/basicTypes/Id.lhs -11
    M ./compiler/basicTypes/IdInfo.lhs -82
    M ./compiler/basicTypes/MkId.lhs -2 +2
    M ./compiler/coreSyn/CoreFVs.lhs -2 +25
    M ./compiler/coreSyn/CoreLint.lhs -5 +1
    M ./compiler/coreSyn/CorePrep.lhs -59 +53
    M ./compiler/coreSyn/CoreSubst.lhs -22 +31
    M ./compiler/coreSyn/CoreSyn.lhs -66 +92
    M ./compiler/coreSyn/CoreUnfold.lhs -112 +112
    M ./compiler/coreSyn/CoreUtils.lhs -185 +184
    M ./compiler/coreSyn/MkExternalCore.lhs -1
    M ./compiler/coreSyn/PprCore.lhs -4 +40
    M ./compiler/deSugar/DsBinds.lhs -70 +118
    M ./compiler/deSugar/DsForeign.lhs -2 +4
    M ./compiler/deSugar/DsMeta.hs -4 +3
    M ./compiler/hsSyn/HsBinds.lhs -3 +3
    M ./compiler/hsSyn/HsUtils.lhs -2 +7
    M ./compiler/iface/BinIface.hs -11 +25
    M ./compiler/iface/IfaceSyn.lhs -13 +21
    M ./compiler/iface/MkIface.lhs -24 +19
    M ./compiler/iface/TcIface.lhs -29 +23
    M ./compiler/main/TidyPgm.lhs -55 +49
    M ./compiler/parser/ParserCore.y -5 +6
    M ./compiler/simplCore/CSE.lhs -2 +1
    M ./compiler/simplCore/FloatIn.lhs -6 +1
    M ./compiler/simplCore/FloatOut.lhs -23
    M ./compiler/simplCore/OccurAnal.lhs -36 +5
    M ./compiler/simplCore/SetLevels.lhs -59 +54
    M ./compiler/simplCore/SimplCore.lhs -48 +52
    M ./compiler/simplCore/SimplEnv.lhs -26 +22
    M ./compiler/simplCore/SimplUtils.lhs -28 +4
    M ./compiler/simplCore/Simplify.lhs -91 +109
    M ./compiler/specialise/Specialise.lhs -15 +18
    M ./compiler/stranal/WorkWrap.lhs -14 +11
    M ./compiler/stranal/WwLib.lhs -2 +2
    M ./compiler/typecheck/Inst.lhs -1 +3
    M ./compiler/typecheck/TcBinds.lhs -17 +27
    M ./compiler/typecheck/TcClassDcl.lhs -1 +2
    M ./compiler/typecheck/TcExpr.lhs -4 +6
    M ./compiler/typecheck/TcForeign.lhs -1 +1
    M ./compiler/typecheck/TcGenDeriv.lhs -14 +13
    M ./compiler/typecheck/TcHsSyn.lhs -3 +2
    M ./compiler/typecheck/TcInstDcls.lhs -5 +4
    M ./compiler/typecheck/TcRnDriver.lhs -2 +11
    M ./compiler/typecheck/TcSimplify.lhs -10 +17
    M ./compiler/vectorise/VectType.hs +7

Mon Dec  8 12:43:10 GMT 2008  simonpj@microsoft.com
  * White space only

    M ./compiler/simplCore/Simplify.lhs -2

Mon Dec  8 12:48:40 GMT 2008  simonpj@microsoft.com
  * Move simpleOptExpr from CoreUnfold to CoreSubst

    M ./compiler/coreSyn/CoreSubst.lhs -1 +87
    M ./compiler/coreSyn/CoreUnfold.lhs -72 +1

Mon Dec  8 17:30:18 GMT 2008  simonpj@microsoft.com
  * Use CoreSubst.simpleOptExpr in place of the ad-hoc simpleSubst (reduces code too)

    M ./compiler/deSugar/DsBinds.lhs -50 +16

Tue Dec  9 17:03:02 GMT 2008  simonpj@microsoft.com
  * Fix Trac #2861: bogus eta expansion
  
  Urghlhl!  I "tided up" the treatment of the "state hack" in CoreUtils, but
  missed an unexpected interaction with the way that a bottoming function
  simply swallows excess arguments.  There's a long
       Note [State hack and bottoming functions]
  to explain (which accounts for most of the new lines of code).
  

    M ./compiler/coreSyn/CoreUtils.lhs -16 +53

Mon Dec 15 10:02:21 GMT 2008  Simon Marlow <marlowsd@gmail.com>
  * Revert CorePrep part of "Completely new treatment of INLINE pragmas..."
  
  The original patch said:
  
  * I made some changes to the way in which eta expansion happens in
    CorePrep, mainly to ensure that *arguments* that become let-bound
    are also eta-expanded.  I'm still not too happy with the clarity
    and robustness fo the result.
    
  Unfortunately this change apparently broke some invariants that were
  relied on elsewhere, and in particular lead to panics when compiling
  with profiling on.
  
  Will re-investigate in the new year.

    M ./compiler/coreSyn/CorePrep.lhs -53 +58
    M ./configure.ac -1 +1

Mon Dec 15 12:28:51 GMT 2008  Simon Marlow <marlowsd@gmail.com>
  * revert accidental change to configure.ac

    M ./configure.ac -1 +1

This is a major patch, which changes the way INLINE pragmas work.
Although lots of files are touched, the net is only +21 lines of
code -- and I bet that most of those are comments!

HEADS UP: interface file format has changed, so you'll need to
recompile everything.

There is not much effect on overall performance for nofib, 
probably because those programs don't make heavy use of INLINE pragmas.

        Program           Size    Allocs   Runtime   Elapsed
            Min         -11.3%     -6.9%     -9.2%     -8.2%
            Max          -0.1%     +4.6%     +7.5%     +8.9%
 Geometric Mean          -2.2%     -0.2%     -1.0%     -0.8%

(The +4.6% for on allocs is cichelli; see other patch relating to
-fpass-case-bndr-to-join-points.)

The old INLINE system
~~~~~~~~~~~~~~~~~~~~~
The old system worked like this. A function with an INLINE pragam
got a right-hand side which looked like
     f = __inline_me__ (\xy. e)
The __inline_me__ part was an InlineNote, and was treated specially
in various ways.  Notably, the simplifier didn't inline inside an
__inline_me__ note.  

As a result, the code for f itself was pretty crappy. That matters
if you say (map f xs), because then you execute the code for f,
rather than inlining a copy at the call site.

The new story: InlineRules
~~~~~~~~~~~~~~~~~~~~~~~~~~
The new system removes the InlineMe Note altogether.  Instead there
is a new constructor InlineRule in CoreSyn.Unfolding.  This is a 
bit like a RULE, in that it remembers the template to be inlined inside
the InlineRule.  No simplification or inlining is done on an InlineRule,
just like RULEs.  

An Id can have an InlineRule *or* a CoreUnfolding (since these are two
constructors from Unfolding). The simplifier treats them differently:

  - An InlineRule is has the substitution applied (like RULES) but 
    is otherwise left undisturbed.

  - A CoreUnfolding is updated with the new RHS of the definition,
    on each iteration of the simplifier.

An InlineRule fires regardless of size, but *only* when the function
is applied to enough arguments.  The "arity" of the rule is specified
(by the programmer) as the number of args on the LHS of the "=".  So
it makes a difference whether you say
  	{-# INLINE f #-}
	f x = \y -> e     or     f x y = e
This is one of the big new features that InlineRule gives us, and it
is one that Roman really wanted.

In contrast, a CoreUnfolding can fire when it is applied to fewer
args than than the function has lambdas, provided the result is small
enough.


Consequential stuff
~~~~~~~~~~~~~~~~~~~
* A 'wrapper' no longer has a WrapperInfo in the IdInfo.  Instead,
  the InlineRule has a field identifying wrappers.

* Of course, IfaceSyn and interface serialisation changes appropriately.

* Making implication constraints inline nicely was a bit fiddly. In
  the end I added a var_inline field to HsBInd.VarBind, which is why
  this patch affects the type checker slightly

* I made some changes to the way in which eta expansion happens in
  CorePrep, mainly to ensure that *arguments* that become let-bound
  are also eta-expanded.  I'm still not too happy with the clarity
  and robustness fo the result.

* We now complain if the programmer gives an INLINE pragma for
  a recursive function (prevsiously we just ignored it).  Reason for
  change: we don't want an InlineRule on a LoopBreaker, because then
  we'd have to check for loop-breaker-hood at occurrence sites (which
  isn't currenlty done).  Some tests need changing as a result.

This patch has been in my tree for quite a while, so there are
probably some other minor changes.

See Note [Freshen type variables] in WwLib.  We need to clone type
variables when building a worker/wrapper split, else we simply get
bogus code, admittedly in rather obscure situations.  I can't quite
remember what program showed this up, unfortunately, but there 
definitely *was* one!  (You get a Lint error.)