TcCanonical.hs 97.9 KB
Newer Older
1 2
{-# LANGUAGE CPP #-}

3
module TcCanonical(
4 5
     canonicalize,
     unifyDerived,
6
     makeSuperClasses, maybeSym,
7 8
     StopOrContinue(..), stopWith, continueWith,
     solveCallStack    -- For TcSimplify
9
  ) where
10 11 12

#include "HsVersions.h"

13 14
import GhcPrelude

15
import TcRnTypes
16
import TcUnify( swapOverTyVars, metaTyVarUpdateOK )
17
import TcType
18
import Type
19 20
import TcFlatten
import TcSMonad
21
import TcEvidence
22
import TcEvTerm
23 24
import Class
import TyCon
25
import TyCoRep   -- cleverly decomposes types, good for completeness checking
26
import Coercion
27 28
import CoreSyn
import Id( idType, mkTemplateLocals )
29 30
import FamInstEnv ( FamInstEnvs )
import FamInst ( tcTopNormaliseNewTypeTF_maybe )
31
import Var
32
import VarEnv( mkInScopeSet )
33
import VarSet( delVarSetList )
34
import Outputable
35
import DynFlags( DynFlags )
36
import NameSet
37
import RdrName
38
import HsTypes( HsIPName(..) )
39

40
import Pair
41
import Util
42
import Bag
43 44
import MonadUtils
import Control.Monad
45
import Data.Maybe ( isJust )
46
import Data.List  ( zip4, foldl' )
47
import BasicTypes
48

49 50
import Data.Bifunctor ( bimap )

Austin Seipp's avatar
Austin Seipp committed
51 52 53 54 55 56
{-
************************************************************************
*                                                                      *
*                      The Canonicaliser                               *
*                                                                      *
************************************************************************
57

58 59
Note [Canonicalization]
~~~~~~~~~~~~~~~~~~~~~~~
60

61
Canonicalization converts a simple constraint to a canonical form. It is
62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78
unary (i.e. treats individual constraints one at a time).

Constraints originating from user-written code come into being as
CNonCanonicals (except for CHoleCans, arising from holes). We know nothing
about these constraints. So, first:

     Classify CNonCanoncal constraints, depending on whether they
     are equalities, class predicates, or other.

Then proceed depending on the shape of the constraint. Generally speaking,
each constraint gets flattened and then decomposed into one of several forms
(see type Ct in TcRnTypes).

When an already-canonicalized constraint gets kicked out of the inert set,
it must be recanonicalized. But we know a bit about its shape from the
last time through, so we can skip the classification step.

Austin Seipp's avatar
Austin Seipp committed
79
-}
80

81 82 83
-- Top-level canonicalization
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

84
canonicalize :: Ct -> TcS (StopOrContinue Ct)
85 86
canonicalize (CNonCanonical { cc_ev = ev })
  = {-# SCC "canNC" #-}
87
    case classifyPredType pred of
88 89 90 91
      ClassPred cls tys     -> do traceTcS "canEvNC:cls" (ppr cls <+> ppr tys)
                                  canClassNC ev cls tys
      EqPred eq_rel ty1 ty2 -> do traceTcS "canEvNC:eq" (ppr ty1 $$ ppr ty2)
                                  canEqNC    ev eq_rel ty1 ty2
92
      IrredPred {}          -> do traceTcS "canEvNC:irred" (ppr pred)
93
                                  canIrred ev
94 95 96 97 98 99 100
      ForAllPred _ _ pred   -> do traceTcS "canEvNC:forall" (ppr pred)
                                  canForAll ev (isClassPred pred)
  where
    pred = ctEvPred ev

canonicalize (CQuantCan (QCI { qci_ev = ev, qci_pend_sc = pend_sc }))
  = canForAll ev pend_sc
101 102

canonicalize (CIrredCan { cc_ev = ev })
103 104 105 106 107 108 109 110 111
  | EqPred eq_rel ty1 ty2 <- classifyPredType (ctEvPred ev)
  = -- For insolubles (all of which are equalities, do /not/ flatten the arguments
    -- In Trac #14350 doing so led entire-unnecessary and ridiculously large
    -- type function expansion.  Instead, canEqNC just applies
    -- the substitution to the predicate, and may do decomposition;
    --    e.g. a ~ [a], where [G] a ~ [Int], can decompose
    canEqNC ev eq_rel ty1 ty2

  | otherwise
112
  = canIrred ev
113

114 115
canonicalize (CDictCan { cc_ev = ev, cc_class  = cls
                       , cc_tyargs = xis, cc_pend_sc = pend_sc })
116
  = {-# SCC "canClass" #-}
117 118
    canClass ev cls xis pend_sc

119
canonicalize (CTyEqCan { cc_ev = ev
120
                       , cc_tyvar  = tv
121 122
                       , cc_rhs    = xi
                       , cc_eq_rel = eq_rel })
123
  = {-# SCC "canEqLeafTyVarEq" #-}
124 125 126
    canEqNC ev eq_rel (mkTyVarTy tv) xi
      -- NB: Don't use canEqTyVar because that expects flattened types,
      -- and tv and xi may not be flat w.r.t. an updated inert set
127

128
canonicalize (CFunEqCan { cc_ev = ev
129 130
                        , cc_fun    = fn
                        , cc_tyargs = xis1
131
                        , cc_fsk    = fsk })
Simon Peyton Jones's avatar
Simon Peyton Jones committed
132
  = {-# SCC "canEqLeafFunEq" #-}
133
    canCFunEqCan ev fn xis1 fsk
134

135 136
canonicalize (CHoleCan { cc_ev = ev, cc_hole = hole })
  = canHole ev hole
137

Austin Seipp's avatar
Austin Seipp committed
138 139 140 141 142 143 144
{-
************************************************************************
*                                                                      *
*                      Class Canonicalization
*                                                                      *
************************************************************************
-}
145

146
canClassNC :: CtEvidence -> Class -> [Type] -> TcS (StopOrContinue Ct)
147
-- "NC" means "non-canonical"; that is, we have got here
Gabor Greif's avatar
Gabor Greif committed
148
-- from a NonCanonical constraint, not from a CDictCan
Simon Peyton Jones's avatar
Simon Peyton Jones committed
149
-- Precondition: EvVar is class evidence
150 151
canClassNC ev cls tys
  | isGiven ev  -- See Note [Eagerly expand given superclasses]
152
  = do { sc_cts <- mkStrictSuperClasses ev [] [] cls tys
153 154
       ; emitWork sc_cts
       ; canClass ev cls tys False }
155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174

  | isWanted ev
  , Just ip_name <- isCallStackPred cls tys
  , OccurrenceOf func <- ctLocOrigin loc
  -- If we're given a CallStack constraint that arose from a function
  -- call, we need to push the current call-site onto the stack instead
  -- of solving it directly from a given.
  -- See Note [Overview of implicit CallStacks] in TcEvidence
  -- and Note [Solving CallStack constraints] in TcSMonad
  = do { -- First we emit a new constraint that will capture the
         -- given CallStack.
       ; let new_loc = setCtLocOrigin loc (IPOccOrigin (HsIPName ip_name))
                            -- We change the origin to IPOccOrigin so
                            -- this rule does not fire again.
                            -- See Note [Overview of implicit CallStacks]

       ; new_ev <- newWantedEvVarNC new_loc pred

         -- Then we solve the wanted by pushing the call-site
         -- onto the newly emitted CallStack
175
       ; let ev_cs = EvCsPushCall func (ctLocSpan loc) (ctEvExpr new_ev)
176 177 178 179
       ; solveCallStack ev ev_cs

       ; canClass new_ev cls tys False }

180 181
  | otherwise
  = canClass ev cls tys (has_scs cls)
182

183 184
  where
    has_scs cls = not (null (classSCTheta cls))
185 186 187 188 189 190 191 192 193
    loc  = ctEvLoc ev
    pred = ctEvPred ev

solveCallStack :: CtEvidence -> EvCallStack -> TcS ()
-- Also called from TcSimplify when defaulting call stacks
solveCallStack ev ev_cs = do
  -- We're given ev_cs :: CallStack, but the evidence term should be a
  -- dictionary, so we have to coerce ev_cs to a dictionary for
  -- `IP ip CallStack`. See Note [Overview of implicit CallStacks]
194 195
  cs_tm <- evCallStack ev_cs
  let ev_tm = mkEvCast cs_tm (wrapIP (ctEvPred ev))
196
  setEvBindIfWanted ev ev_tm
197

198 199 200 201
canClass :: CtEvidence
         -> Class -> [Type]
         -> Bool            -- True <=> un-explored superclasses
         -> TcS (StopOrContinue Ct)
202
-- Precondition: EvVar is class evidence
203

204
canClass ev cls tys pend_sc
205 206
  =   -- all classes do *nominal* matching
    ASSERT2( ctEvRole ev == Nominal, ppr ev $$ ppr cls $$ ppr tys )
207 208 209
    do { (xis, cos, _kind_co) <- flattenArgsNom ev cls_tc tys
       ; MASSERT( isTcReflCo _kind_co )
       ; let co = mkTcTyConAppCo Nominal cls_tc cos
210
             xi = mkClassPred cls xis
211
             mk_ct new_ev = CDictCan { cc_ev = new_ev
212 213 214
                                     , cc_tyargs = xis
                                     , cc_class = cls
                                     , cc_pend_sc = pend_sc }
215
       ; mb <- rewriteEvidence ev xi co
216
       ; traceTcS "canClass" (vcat [ ppr ev
Simon Peyton Jones's avatar
Simon Peyton Jones committed
217
                                   , ppr xi, ppr mb ])
218
       ; return (fmap mk_ct mb) }
219 220
  where
    cls_tc = classTyCon cls
dimitris's avatar
dimitris committed
221

222 223 224 225
{- Note [The superclass story]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We need to add superclass constraints for two reasons:

226
* For givens [G], they give us a route to proof.  E.g.
227 228 229 230 231
    f :: Ord a => a -> Bool
    f x = x == x
  We get a Wanted (Eq a), which can only be solved from the superclass
  of the Given (Ord a).

232 233
* For wanteds [W], and deriveds [WD], [D], they may give useful
  functional dependencies.  E.g.
234 235
     class C a b | a -> b where ...
     class C a b => D a b where ...
236
  Now a [W] constraint (D Int beta) has (C Int beta) as a superclass
237
  and that might tell us about beta, via C's fundeps.  We can get this
238
  by generating a [D] (C Int beta) constraint.  It's derived because
239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263
  we don't actually have to cough up any evidence for it; it's only there
  to generate fundep equalities.

See Note [Why adding superclasses can help].

For these reasons we want to generate superclass constraints for both
Givens and Wanteds. But:

* (Minor) they are often not needed, so generating them aggressively
  is a waste of time.

* (Major) if we want recursive superclasses, there would be an infinite
  number of them.  Here is a real-life example (Trac #10318);

     class (Frac (Frac a) ~ Frac a,
            Fractional (Frac a),
            IntegralDomain (Frac a))
         => IntegralDomain a where
      type Frac a :: *

  Notice that IntegralDomain has an associated type Frac, and one
  of IntegralDomain's superclasses is another IntegralDomain constraint.

So here's the plan:

264 265
1. Eagerly generate superclasses for given (but not wanted)
   constraints; see Note [Eagerly expand given superclasses].
266 267
   This is done using mkStrictSuperClasses in canClassNC, when
   we take a non-canonical Given constraint and cannonicalise it.
268 269

   However stop if you encounter the same class twice.  That is,
270 271
   mkStrictSuperClasses expands eagerly, but has a conservative
   termination condition: see Note [Expanding superclasses] in TcType.
272

273 274 275 276
2. Solve the wanteds as usual, but do no further expansion of
   superclasses for canonical CDictCans in solveSimpleGivens or
   solveSimpleWanteds; Note [Danger of adding superclasses during solving]

277 278
   However, /do/ continue to eagerly expand superlasses for new /given/
   /non-canonical/ constraints (canClassNC does this).  As Trac #12175
279 280 281
   showed, a type-family application can expand to a class constraint,
   and we want to see its superclasses for just the same reason as
   Note [Eagerly expand given superclasses].
282

283 284 285
3. If we have any remaining unsolved wanteds
        (see Note [When superclasses help] in TcRnTypes)
   try harder: take both the Givens and Wanteds, and expand
286 287 288 289 290
   superclasses again.  See the calls to expandSuperClasses in
   TcSimplify.simpl_loop and solveWanteds.

   This may succeed in generating (a finite number of) extra Givens,
   and extra Deriveds. Both may help the proof.
291

292 293 294 295 296 297 298 299 300
3a An important wrinkle: only expand Givens from the current level.
   Two reasons:
      - We only want to expand it once, and that is best done at
        the level it is bound, rather than repeatedly at the leaves
        of the implication tree
      - We may be inside a type where we can't create term-level
        evidence anyway, so we can't superclass-expand, say,
        (a ~ b) to get (a ~# b).  This happened in Trac #15290.

301 302 303
4. Go round to (2) again.  This loop (2,3,4) is implemented
   in TcSimplify.simpl_loop.

304 305 306
The cc_pend_sc flag in a CDictCan records whether the superclasses of
this constraint have been expanded.  Specifically, in Step 3 we only
expand superclasses for constraints with cc_pend_sc set to true (i.e.
307 308
isPendingScDict holds).

309 310 311 312 313 314 315 316 317
Why do we do this?  Two reasons:

* To avoid repeated work, by repeatedly expanding the superclasses of
  same constraint,

* To terminate the above loop, at least in the -XNoRecursiveSuperClasses
  case.  If there are recursive superclasses we could, in principle,
  expand forever, always encountering new constraints.

318 319 320
When we take a CNonCanonical or CIrredCan, but end up classifying it
as a CDictCan, we set the cc_pend_sc flag to False.

321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338
Note [Superclass loops]
~~~~~~~~~~~~~~~~~~~~~~~
Suppose we have
  class C a => D a
  class D a => C a

Then, when we expand superclasses, we'll get back to the self-same
predicate, so we have reached a fixpoint in expansion and there is no
point in fruitlessly expanding further.  This case just falls out from
our strategy.  Consider
  f :: C a => a -> Bool
  f x = x==x
Then canClassNC gets the [G] d1: C a constraint, and eager emits superclasses
G] d2: D a, [G] d3: C a (psc).  (The "psc" means it has its sc_pend flag set.)
When processing d3 we find a match with d1 in the inert set, and we always
keep the inert item (d1) if possible: see Note [Replacement vs keeping] in
TcInteract.  So d3 dies a quick, happy death.

339 340 341
Note [Eagerly expand given superclasses]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In step (1) of Note [The superclass story], why do we eagerly expand
342 343 344 345
Given superclasses by one layer?  (By "one layer" we mean expand transitively
until you meet the same class again -- the conservative criterion embodied
in expandSuperClasses.  So a "layer" might be a whole stack of superclasses.)
We do this eagerly for Givens mainly because of some very obscure
346
cases like this:
347 348 349 350 351 352 353 354 355 356 357

   instance Bad a => Eq (T a)

   f :: (Ord (T a)) => blah
   f x = ....needs Eq (T a), Ord (T a)....

Here if we can't satisfy (Eq (T a)) from the givens we'll use the
instance declaration; but then we are stuck with (Bad a).  Sigh.
This is really a case of non-confluent proofs, but to stop our users
complaining we expand one layer in advance.

358 359 360 361 362 363 364 365 366 367 368 369 370
Note [Instance and Given overlap] in TcInteract.

We also want to do this if we have

   f :: F (T a) => blah

where
   type instance F (T a) = Ord (T a)

So we may need to do a little work on the givens to expose the
class that has the superclasses.  That's why the superclass
expansion for Givens happens in canClassNC.

371 372 373
Note [Why adding superclasses can help]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Examples of how adding superclasses can help:
374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389

    --- Example 1
        class C a b | a -> b
    Suppose we want to solve
         [G] C a b
         [W] C a beta
    Then adding [D] beta~b will let us solve it.

    -- Example 2 (similar but using a type-equality superclass)
        class (F a ~ b) => C a b
    And try to sllve:
         [G] C a b
         [W] C a beta
    Follow the superclass rules to add
         [G] F a ~ b
         [D] F a ~ beta
390
    Now we get [D] beta ~ b, and can solve that.
391

392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409
    -- Example (tcfail138)
      class L a b | a -> b
      class (G a, L a b) => C a b

      instance C a b' => G (Maybe a)
      instance C a b  => C (Maybe a) a
      instance L (Maybe a) a

    When solving the superclasses of the (C (Maybe a) a) instance, we get
      [G] C a b, and hance by superclasses, [G] G a, [G] L a b
      [W] G (Maybe a)
    Use the instance decl to get
      [W] C a beta
    Generate its derived superclass
      [D] L a beta.  Now using fundeps, combine with [G] L a b to get
      [D] beta ~ b
    which is what we want.

410 411
Note [Danger of adding superclasses during solving]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
412
Here's a serious, but now out-dated example, from Trac #4497:
Simon Peyton Jones's avatar
Simon Peyton Jones committed
413

414 415 416
   class Num (RealOf t) => Normed t
   type family RealOf x

Simon Peyton Jones's avatar
Simon Peyton Jones committed
417
Assume the generated wanted constraint is:
418 419 420
   [W] RealOf e ~ e
   [W] Normed e

Simon Peyton Jones's avatar
Simon Peyton Jones committed
421
If we were to be adding the superclasses during simplification we'd get:
422 423 424 425
   [W] RealOf e ~ e
   [W] Normed e
   [D] RealOf e ~ fuv
   [D] Num fuv
Simon Peyton Jones's avatar
Simon Peyton Jones committed
426
==>
427
   e := fuv, Num fuv, Normed fuv, RealOf fuv ~ fuv
Simon Peyton Jones's avatar
Simon Peyton Jones committed
428

429 430 431
While looks exactly like our original constraint. If we add the
superclass of (Normed fuv) again we'd loop.  By adding superclasses
definitely only once, during canonicalisation, this situation can't
432
happen.
433 434 435 436

Mind you, now that Wanteds cannot rewrite Derived, I think this particular
situation can't happen.
  -}
437

438 439 440 441
makeSuperClasses :: [Ct] -> TcS [Ct]
-- Returns strict superclasses, transitively, see Note [The superclasses story]
-- See Note [The superclass story]
-- The loop-breaking here follows Note [Expanding superclasses] in TcType
442 443 444 445 446 447 448 449 450 451
-- Specifically, for an incoming (C t) constraint, we return all of (C t)'s
--    superclasses, up to /and including/ the first repetition of C
--
-- Example:  class D a => C a
--           class C [a] => D a
-- makeSuperClasses (C x) will return (D x, C [x])
--
-- NB: the incoming constraints have had their cc_pend_sc flag already
--     flipped to False, by isPendingScDict, so we are /obliged/ to at
--     least produce the immediate superclasses
452 453 454
makeSuperClasses cts = concatMapM go cts
  where
    go (CDictCan { cc_ev = ev, cc_class = cls, cc_tyargs = tys })
455 456 457 458 459 460 461
      = mkStrictSuperClasses ev [] [] cls tys
    go (CQuantCan (QCI { qci_pred = pred, qci_ev = ev }))
      = ASSERT2( isClassPred pred, ppr pred )  -- The cts should all have
                                               -- class pred heads
        mkStrictSuperClasses ev tvs theta cls tys
      where
        (tvs, theta, cls, tys) = tcSplitDFunTy (ctEvPred ev)
462 463
    go ct = pprPanic "makeSuperClasses" (ppr ct)

464 465 466 467 468 469 470 471 472 473 474 475 476 477
mkStrictSuperClasses
    :: CtEvidence
    -> [TyVar] -> ThetaType  -- These two args are non-empty only when taking
                             -- superclasses of a /quantified/ constraint
    -> Class -> [Type] -> TcS [Ct]
-- Return constraints for the strict superclasses of
--   ev :: forall as. theta => cls tys
mkStrictSuperClasses ev tvs theta cls tys
  = mk_strict_superclasses (unitNameSet (className cls))
                           ev tvs theta cls tys

mk_strict_superclasses :: NameSet -> CtEvidence
                       -> [TyVar] -> ThetaType
                       -> Class -> [Type] -> TcS [Ct]
478 479 480
-- Always return the immediate superclasses of (cls tys);
-- and expand their superclasses, provided none of them are in rec_clss
-- nor are repeated
481
mk_strict_superclasses rec_clss ev tvs theta cls tys
482
  | CtGiven { ctev_evar = evar, ctev_loc = loc } <- ev
483 484 485 486 487
  = concatMapM (do_one_given evar (mk_given_loc loc)) $
    classSCSelIds cls
  where
    dict_ids  = mkTemplateLocals theta
    size      = sizeTypes tys
dimitris's avatar
dimitris committed
488

489 490 491 492 493 494 495 496 497 498 499
    do_one_given evar given_loc sel_id
      = do { let sc_pred = funResultTy (piResultTys (idType sel_id) tys)
                 given_ty = mkInfSigmaTy tvs theta sc_pred
           ; given_ev <- newGivenEvVar given_loc $
                         (given_ty, mk_sc_sel evar sel_id)
           ; mk_superclasses rec_clss given_ev tvs theta sc_pred }

    mk_sc_sel evar sel_id
      = EvExpr $ mkLams tvs $ mkLams dict_ids $
        Var sel_id `mkTyApps` tys `App`
        (evId evar `mkTyApps` mkTyVarTys tvs `mkVarApps` dict_ids)
500

501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517
    mk_given_loc loc
       | isCTupleClass cls
       = loc   -- For tuple predicates, just take them apart, without
               -- adding their (large) size into the chain.  When we
               -- get down to a base predicate, we'll include its size.
               -- Trac #10335

       | GivenOrigin skol_info <- ctLocOrigin loc
         -- See Note [Solving superclass constraints] in TcInstDcls
         -- for explantation of this transformation for givens
       = case skol_info of
            InstSkol -> loc { ctl_origin = GivenOrigin (InstSC size) }
            InstSC n -> loc { ctl_origin = GivenOrigin (InstSC (n `max` size)) }
            _        -> loc

       | otherwise  -- Probably doesn't happen, since this function
       = loc        -- is only used for Givens, but does no harm
518

519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575
mk_strict_superclasses rec_clss ev tvs theta cls tys
  | all noFreeVarsOfType tys
  = return [] -- Wanteds with no variables yield no deriveds.
              -- See Note [Improvement from Ground Wanteds]

  | otherwise -- Wanted/Derived case, just add Derived superclasses
              -- that can lead to improvement.
  = ASSERT2( null tvs && null theta, ppr tvs $$ ppr theta )
    concatMapM do_one_derived (immSuperClasses cls tys)
  where
    loc = ctEvLoc ev

    do_one_derived sc_pred
      = do { sc_ev <- newDerivedNC loc sc_pred
           ; mk_superclasses rec_clss sc_ev [] [] sc_pred }

mk_superclasses :: NameSet -> CtEvidence
                -> [TyVar] -> ThetaType -> PredType -> TcS [Ct]
-- Return this constraint, plus its superclasses, if any
mk_superclasses rec_clss ev tvs theta pred
  | ClassPred cls tys <- classifyPredType pred
  = mk_superclasses_of rec_clss ev tvs theta cls tys

  | otherwise   -- Superclass is not a class predicate
  = return [mkNonCanonical ev]

mk_superclasses_of :: NameSet -> CtEvidence
                   -> [TyVar] -> ThetaType -> Class -> [Type]
                   -> TcS [Ct]
-- Always return this class constraint,
-- and expand its superclasses
mk_superclasses_of rec_clss ev tvs theta cls tys
  | loop_found = do { traceTcS "mk_superclasses_of: loop" (ppr cls <+> ppr tys)
                    ; return [this_ct] }  -- cc_pend_sc of this_ct = True
  | otherwise  = do { traceTcS "mk_superclasses_of" (vcat [ ppr cls <+> ppr tys
                                                          , ppr (isCTupleClass cls)
                                                          , ppr rec_clss
                                                          ])
                    ; sc_cts <- mk_strict_superclasses rec_clss' ev tvs theta cls tys
                    ; return (this_ct : sc_cts) }
                                   -- cc_pend_sc of this_ct = False
  where
    cls_nm     = className cls
    loop_found = not (isCTupleClass cls) && cls_nm `elemNameSet` rec_clss
                 -- Tuples never contribute to recursion, and can be nested
    rec_clss'  = rec_clss `extendNameSet` cls_nm

    this_ct | null tvs, null theta
            = CDictCan { cc_ev = ev, cc_class = cls, cc_tyargs = tys
                       , cc_pend_sc = loop_found }
                 -- NB: If there is a loop, we cut off, so we have not
                 --     added the superclasses, hence cc_pend_sc = True
            | otherwise
            = CQuantCan (QCI { qci_tvs = tvs, qci_pred = mkClassPred cls tys
                             , qci_ev = ev
                             , qci_pend_sc = loop_found })

576

Austin Seipp's avatar
Austin Seipp committed
577 578 579 580 581 582 583
{-
************************************************************************
*                                                                      *
*                      Irreducibles canonicalization
*                                                                      *
************************************************************************
-}
584

585
canIrred :: CtEvidence -> TcS (StopOrContinue Ct)
586
-- Precondition: ty not a tuple and no other evidence form
587
canIrred ev
588 589
  = do { let pred = ctEvPred ev
       ; traceTcS "can_pred" (text "IrredPred = " <+> ppr pred)
590 591
       ; (xi,co) <- flatten FM_FlattenAll ev pred -- co :: xi ~ pred
       ; rewriteEvidence ev xi co `andWhenContinue` \ new_ev ->
592 593
    do { -- Re-classify, in case flattening has improved its shape
       ; case classifyPredType (ctEvPred new_ev) of
594 595 596
           ClassPred cls tys     -> canClassNC new_ev cls tys
           EqPred eq_rel ty1 ty2 -> canEqNC new_ev eq_rel ty1 ty2
           _                     -> continueWith $
597
                                    mkIrredCt new_ev } }
598

599 600
canHole :: CtEvidence -> Hole -> TcS (StopOrContinue Ct)
canHole ev hole
601 602
  = do { let pred = ctEvPred ev
       ; (xi,co) <- flatten FM_SubstOnly ev pred -- co :: xi ~ pred
603
       ; rewriteEvidence ev xi co `andWhenContinue` \ new_ev ->
604 605
    do { updInertIrreds (`snocCts` (CHoleCan { cc_ev = new_ev
                                             , cc_hole = hole }))
606
       ; stopWith new_ev "Emit insoluble hole" } }
607

608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762

{- *********************************************************************
*                                                                      *
*                      Quantified predicates
*                                                                      *
********************************************************************* -}

{- Note [Quantified constraints]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The -XQuantifiedConstraints extension allows type-class contexts like this:

  data Rose f x = Rose x (f (Rose f x))

  instance (Eq a, forall b. Eq b => Eq (f b))
        => Eq (Rose f a)  where
    (Rose x1 rs1) == (Rose x2 rs2) = x1==x2 && rs1 == rs2

Note the (forall b. Eq b => Eq (f b)) in the instance contexts.
This quantified constraint is needed to solve the
 [W] (Eq (f (Rose f x)))
constraint which arises form the (==) definition.

The wiki page is
  https://ghc.haskell.org/trac/ghc/wiki/QuantifiedConstraints
which in turn contains a link to the GHC Proposal where the change
is specified, and a Haskell Symposium paper about it.

We implement two main extensions to the design in the paper:

 1. We allow a variable in the instance head, e.g.
      f :: forall m a. (forall b. m b) => D (m a)
    Notice the 'm' in the head of the quantified constraint, not
    a class.

 2. We suport superclasses to quantified constraints.
    For example (contrived):
      f :: (Ord b, forall b. Ord b => Ord (m b)) => m a -> m a -> Bool
      f x y = x==y
    Here we need (Eq (m a)); but the quantifed constraint deals only
    with Ord.  But we can make it work by using its superclass.

Here are the moving parts
  * Language extension {-# LANGUAGE QuantifiedConstraints #-}
    and add it to ghc-boot-th:GHC.LanguageExtensions.Type.Extension

  * A new form of evidence, EvDFun, that is used to discharge
    such wanted constraints

  * checkValidType gets some changes to accept forall-constraints
    only in the right places.

  * Type.PredTree gets a new constructor ForAllPred, and
    and classifyPredType analyses a PredType to decompose
    the new forall-constraints

  * TcSMonad.InertCans gets an extra field, inert_insts,
    which holds all the Given forall-constraints.  In effect,
    such Given constraints are like local instance decls.

  * When trying to solve a class constraint, via
    TcInteract.matchInstEnv, use the InstEnv from inert_insts
    so that we include the local Given forall-constraints
    in the lookup.  (See TcSMonad.getInstEnvs.)

  * TcCanonical.canForAll deals with solving a
    forall-constraint.  See
       Note [Solving a Wanted forall-constraint]

  * We augment the kick-out code to kick out an inert
    forall constraint if it can be rewritten by a new
    type equality; see TcSMonad.kick_out_rewritable

Note that a quantified constraint is never /inferred/
(by TcSimplify.simplifyInfer).  A function can only have a
quantified constraint in its type if it is given an explicit
type signature.

Note that we implement
-}

canForAll :: CtEvidence -> Bool -> TcS (StopOrContinue Ct)
-- We have a constraint (forall as. blah => C tys)
canForAll ev pend_sc
  = do { -- First rewrite it to apply the current substitution
         -- Do not bother with type-family reductions; we can't
         -- do them under a forall anyway (c.f. Flatten.flatten_one
         -- on a forall type)
         let pred = ctEvPred ev
       ; (xi,co) <- flatten FM_SubstOnly ev pred -- co :: xi ~ pred
       ; rewriteEvidence ev xi co `andWhenContinue` \ new_ev ->

    do { -- Now decompose into its pieces and solve it
         -- (It takes a lot less code to flatten before decomposing.)
       ; case classifyPredType (ctEvPred new_ev) of
           ForAllPred tv_bndrs theta pred
              -> solveForAll new_ev tv_bndrs theta pred pend_sc
           _  -> pprPanic "canForAll" (ppr new_ev)
    } }

solveForAll :: CtEvidence -> [TyVarBinder] -> TcThetaType -> PredType -> Bool
            -> TcS (StopOrContinue Ct)
solveForAll ev tv_bndrs theta pred pend_sc
  | CtWanted { ctev_dest = dest } <- ev
  = -- See Note [Solving a Wanted forall-constraint]
    do { let skol_info = QuantCtxtSkol
             empty_subst = mkEmptyTCvSubst $ mkInScopeSet $
                           tyCoVarsOfTypes (pred:theta) `delVarSetList` tvs
       ; (subst, skol_tvs) <- tcInstSkolTyVarsX empty_subst tvs
       ; given_ev_vars <- mapM newEvVar (substTheta subst theta)

       ; (w_id, ev_binds)
             <- checkConstraintsTcS skol_info skol_tvs given_ev_vars $
                do { wanted_ev <- newWantedEvVarNC loc $
                                  substTy subst pred
                   ; return ( ctEvEvId wanted_ev
                            , unitBag (mkNonCanonical wanted_ev)) }

      ; setWantedEvTerm dest $
        EvFun { et_tvs = skol_tvs, et_given = given_ev_vars
              , et_binds = ev_binds, et_body = w_id }

      ; stopWith ev "Wanted forall-constraint" }

  | isGiven ev   -- See Note [Solving a Given forall-constraint]
  = do { addInertForAll qci
       ; stopWith ev "Given forall-constraint" }

  | otherwise
  = stopWith ev "Derived forall-constraint"
  where
    loc = ctEvLoc ev
    tvs = binderVars tv_bndrs
    qci = QCI { qci_ev = ev, qci_tvs = tvs
              , qci_pred = pred, qci_pend_sc = pend_sc }

{- Note [Solving a Wanted forall-constraint]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Solving a wanted forall (quantified) constraint
  [W] df :: forall ab. (Eq a, Ord b) => C x a b
is delightfully easy.   Just build an implication constraint
    forall ab. (g1::Eq a, g2::Ord b) => [W] d :: C x a
and discharge df thus:
    df = /\ab. \g1 g2. let <binds> in d
where <binds> is filled in by solving the implication constraint.
All the machinery is to hand; there is little to do.

Note [Solving a Given forall-constraint]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
For a Given constraint
  [G] df :: forall ab. (Eq a, Ord b) => C x a b
we just add it to TcS's local InstEnv of known instances,
via addInertForall.  Then, if we look up (C x Int Bool), say,
we'll find a match in the InstEnv.


Austin Seipp's avatar
Austin Seipp committed
763 764 765 766 767
************************************************************************
*                                                                      *
*        Equalities
*                                                                      *
************************************************************************
768 769 770 771 772 773 774 775 776 777 778 779 780

Note [Canonicalising equalities]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In order to canonicalise an equality, we look at the structure of the
two types at hand, looking for similarities. A difficulty is that the
types may look dissimilar before flattening but similar after flattening.
However, we don't just want to jump in and flatten right away, because
this might be wasted effort. So, after looking for similarities and failing,
we flatten and then try again. Of course, we don't want to loop, so we
track whether or not we've already flattened.

It is conceivable to do a better job at tracking whether or not a type
is flattened, but this is left as future work. (Mar '15)
781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799


Note [FunTy and decomposing tycon applications]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

When can_eq_nc' attempts to decompose a tycon application we haven't yet zonked.
This means that we may very well have a FunTy containing a type of some unknown
kind. For instance, we may have,

    FunTy (a :: k) Int

Where k is a unification variable. tcRepSplitTyConApp_maybe panics in the event
that it sees such a type as it cannot determine the RuntimeReps which the (->)
is applied to. Consequently, it is vital that we instead use
tcRepSplitTyConApp_maybe', which simply returns Nothing in such a case.

When this happens can_eq_nc' will fail to decompose, zonk, and try again.
Zonking should fill the variable k, meaning that decomposition will succeed the
second time around.
Austin Seipp's avatar
Austin Seipp committed
800
-}
801

802 803
canEqNC :: CtEvidence -> EqRel -> Type -> Type -> TcS (StopOrContinue Ct)
canEqNC ev eq_rel ty1 ty2
804 805 806 807
  = do { result <- zonk_eq_types ty1 ty2
       ; case result of
           Left (Pair ty1' ty2') -> can_eq_nc False ev eq_rel ty1' ty1 ty2' ty2
           Right ty              -> canEqReflexive ev eq_rel ty }
808

809
can_eq_nc
810 811
   :: Bool            -- True => both types are flat
   -> CtEvidence
812
   -> EqRel
Austin Seipp's avatar
Austin Seipp committed
813 814
   -> Type -> Type    -- LHS, after and before type-synonym expansion, resp
   -> Type -> Type    -- RHS, after and before type-synonym expansion, resp
815
   -> TcS (StopOrContinue Ct)
816
can_eq_nc flat ev eq_rel ty1 ps_ty1 ty2 ps_ty2
Austin Seipp's avatar
Austin Seipp committed
817
  = do { traceTcS "can_eq_nc" $
818
         vcat [ ppr flat, ppr ev, ppr eq_rel, ppr ty1, ppr ps_ty1, ppr ty2, ppr ps_ty2 ]
819 820
       ; rdr_env <- getGlobalRdrEnvTcS
       ; fam_insts <- getFamInstEnvs
821
       ; can_eq_nc' flat rdr_env fam_insts ev eq_rel ty1 ps_ty1 ty2 ps_ty2 }
822 823

can_eq_nc'
824 825
   :: Bool           -- True => both input types are flattened
   -> GlobalRdrEnv   -- needed to see which newtypes are in scope
826 827 828 829 830 831
   -> FamInstEnvs    -- needed to unwrap data instances
   -> CtEvidence
   -> EqRel
   -> Type -> Type    -- LHS, after and before type-synonym expansion, resp
   -> Type -> Type    -- RHS, after and before type-synonym expansion, resp
   -> TcS (StopOrContinue Ct)
832 833

-- Expand synonyms first; see Note [Type synonyms and canonicalization]
834
can_eq_nc' flat _rdr_env _envs ev eq_rel ty1 ps_ty1 ty2 ps_ty2
Ben Gamari's avatar
Ben Gamari committed
835 836
  | Just ty1' <- tcView ty1 = can_eq_nc flat ev eq_rel ty1' ps_ty1 ty2  ps_ty2
  | Just ty2' <- tcView ty2 = can_eq_nc flat ev eq_rel ty1  ps_ty1 ty2' ps_ty2
837 838

-- need to check for reflexivity in the ReprEq case.
eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
839
-- See Note [Eager reflexivity check]
840 841 842
-- Check only when flat because the zonk_eq_types check in canEqNC takes
-- care of the non-flat case.
can_eq_nc' True _rdr_env _envs ev ReprEq ty1 _ ty2 _
eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
843
  | ty1 `tcEqType` ty2
844 845 846
  = canEqReflexive ev ReprEq ty1

-- When working with ReprEq, unwrap newtypes.
847 848 849 850
-- See Note [Unwrap newtypes first]
can_eq_nc' _flat rdr_env envs ev eq_rel ty1 ps_ty1 ty2 ps_ty2
  | ReprEq <- eq_rel
  , Just stuff1 <- tcTopNormaliseNewTypeTF_maybe envs rdr_env ty1
851
  = can_eq_newtype_nc ev NotSwapped ty1 stuff1 ty2 ps_ty2
852 853 854

  | ReprEq <- eq_rel
  , Just stuff2 <- tcTopNormaliseNewTypeTF_maybe envs rdr_env ty2
855
  = can_eq_newtype_nc ev IsSwapped  ty2 stuff2 ty1 ps_ty1
856

857 858 859 860 861 862
-- Then, get rid of casts
can_eq_nc' flat _rdr_env _envs ev eq_rel (CastTy ty1 co1) _ ty2 ps_ty2
  = canEqCast flat ev eq_rel NotSwapped ty1 co1 ty2 ps_ty2
can_eq_nc' flat _rdr_env _envs ev eq_rel ty1 ps_ty1 (CastTy ty2 co2) _
  = canEqCast flat ev eq_rel IsSwapped ty2 co2 ty1 ps_ty1

863 864 865 866 867 868 869
-- NB: pattern match on True: we want only flat types sent to canEqTyVar.
-- See also Note [No top-level newtypes on RHS of representational equalities]
can_eq_nc' True _rdr_env _envs ev eq_rel (TyVarTy tv1) ps_ty1 ty2 ps_ty2
  = canEqTyVar ev eq_rel NotSwapped tv1 ps_ty1 ty2 ps_ty2
can_eq_nc' True _rdr_env _envs ev eq_rel ty1 ps_ty1 (TyVarTy tv2) ps_ty2
  = canEqTyVar ev eq_rel IsSwapped tv2 ps_ty2 ty1 ps_ty1

870 871 872 873 874
----------------------
-- Otherwise try to decompose
----------------------

-- Literals
875
can_eq_nc' _flat _rdr_env _envs ev eq_rel ty1@(LitTy l1) _ (LitTy l2) _
876
 | l1 == l2
877
  = do { setEvBindIfWanted ev (evCoercion $ mkReflCo (eqRelRole eq_rel) ty1)
878
       ; stopWith ev "Equal LitTy" }
879

Simon Peyton Jones's avatar
Simon Peyton Jones committed
880 881
-- Try to decompose type constructor applications
-- Including FunTy (s -> t)
882
can_eq_nc' _flat _rdr_env _envs ev eq_rel ty1 _ ty2 _
883 884 885
    --- See Note [FunTy and decomposing type constructor applications].
  | Just (tc1, tys1) <- tcRepSplitTyConApp_maybe' ty1
  , Just (tc2, tys2) <- tcRepSplitTyConApp_maybe' ty2
886 887
  , not (isTypeFamilyTyCon tc1)
  , not (isTypeFamilyTyCon tc2)
888
  = canTyConApp ev eq_rel tc1 tys1 tc2 tys2
889

890
can_eq_nc' _flat _rdr_env _envs ev eq_rel
891
           s1@(ForAllTy {}) _ s2@(ForAllTy {}) _
892
  = can_eq_nc_forall ev eq_rel s1 s2
893

894
-- See Note [Canonicalising type applications] about why we require flat types
eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
895
can_eq_nc' True _rdr_env _envs ev eq_rel (AppTy t1 s1) _ ty2 _
896 897 898
  | NomEq <- eq_rel
  , Just (t2, s2) <- tcSplitAppTy_maybe ty2
  = can_eq_app ev t1 s1 t2 s2
eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
899
can_eq_nc' True _rdr_env _envs ev eq_rel ty1 _ (AppTy t2 s2) _
900 901 902
  | NomEq <- eq_rel
  , Just (t1, s1) <- tcSplitAppTy_maybe ty1
  = can_eq_app ev t1 s1 t2 s2
903

904
-- No similarity in type structure detected. Flatten and try again.
905 906 907
can_eq_nc' False rdr_env envs ev eq_rel _ ps_ty1 _ ps_ty2
  = do { (xi1, co1) <- flatten FM_FlattenAll ev ps_ty1
       ; (xi2, co2) <- flatten FM_FlattenAll ev ps_ty2
908 909
       ; new_ev <- rewriteEqEvidence ev NotSwapped xi1 xi2 co1 co2
       ; can_eq_nc' True rdr_env envs new_ev eq_rel xi1 xi1 xi2 xi2 }
910 911

-- We've flattened and the types don't match. Give up.
912
can_eq_nc' True _rdr_env _envs ev eq_rel _ ps_ty1 _ ps_ty2
913
  = do { traceTcS "can_eq_nc' catch-all case" (ppr ps_ty1 $$ ppr ps_ty2)
914 915 916 917 918 919 920 921 922 923 924 925 926 927
       ; case eq_rel of -- See Note [Unsolved equalities]
            ReprEq -> continueWith (mkIrredCt ev)
            NomEq  -> continueWith (mkInsolubleCt ev) }
          -- No need to call canEqFailure/canEqHardFailure because they
          -- flatten, and the types involved here are already flat

{- Note [Unsolved equalities]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
If we have an unsolved equality like
  (a b ~R# Int)
that is not necessarily insoluble!  Maybe 'a' will turn out to be a newtype.
So we want to make it a potentially-soluble Irred not an insoluble one.
Missing this point is what caused Trac #15431
-}
928

929 930 931 932 933 934 935 936 937 938 939 940 941 942
---------------------------------
can_eq_nc_forall :: CtEvidence -> EqRel
                 -> Type -> Type    -- LHS and RHS
                 -> TcS (StopOrContinue Ct)
-- (forall as. phi1) ~ (forall bs. phi2)
-- Check for length match of as, bs
-- Then build an implication constraint: forall as. phi1 ~ phi2[as/bs]
-- But remember also to unify the kinds of as and bs
--  (this is the 'go' loop), and actually substitute phi2[as |> cos / bs]
-- Remember also that we might have forall z (a:z). blah
--  so we must proceed one binder at a time (Trac #13879)

can_eq_nc_forall ev eq_rel s1 s2
 | CtWanted { ctev_loc = loc, ctev_dest = orig_dest } <- ev
943
 = do { let free_tvs       = tyCoVarsOfTypes [s1,s2]
944 945 946 947 948 949 950 951 952 953
            (bndrs1, phi1) = tcSplitForAllTyVarBndrs s1
            (bndrs2, phi2) = tcSplitForAllTyVarBndrs s2
      ; if not (equalLength bndrs1 bndrs2)
        then do { traceTcS "Forall failure" $
                     vcat [ ppr s1, ppr s2, ppr bndrs1, ppr bndrs2
                          , ppr (map binderArgFlag bndrs1)
                          , ppr (map binderArgFlag bndrs2) ]
                ; canEqHardFailure ev s1 s2 }
        else
   do { traceTcS "Creating implication for polytype equality" $ ppr ev
954
      ; let empty_subst1 = mkEmptyTCvSubst $ mkInScopeSet free_tvs
955 956 957 958 959 960 961
      ; (subst1, skol_tvs) <- tcInstSkolTyVarsX empty_subst1 $
                              binderVars bndrs1

      ; let skol_info = UnifyForAllSkol phi1
            phi1' = substTy subst1 phi1

            -- Unify the kinds, extend the substitution
962 963
            go :: [TcTyVar] -> TCvSubst -> [TyVarBinder]
               -> TcS (TcCoercion, Cts)
964 965
            go (skol_tv:skol_tvs) subst (bndr2:bndrs2)
              = do { let tv2 = binderVar bndr2
966 967
                   ; (kind_co, wanteds1) <- unify loc Nominal (tyVarKind skol_tv)
                                                  (substTy subst (tyVarKind tv2))
968 969
                   ; let subst' = extendTvSubst subst tv2
                                       (mkCastTy (mkTyVarTy skol_tv) kind_co)
970 971 972
                   ; (co, wanteds2) <- go skol_tvs subst' bndrs2
                   ; return ( mkTcForAllCo skol_tv kind_co co
                            , wanteds1 `unionBags` wanteds2 ) }
973 974 975 976

            -- Done: unify phi1 ~ phi2
            go [] subst bndrs2
              = ASSERT( null bndrs2 )
977
                unify loc (eqRelRole eq_rel) phi1' (substTy subst phi2)
978 979 980

            go _ _ _ = panic "cna_eq_nc_forall"  -- case (s:ss) []

981
            empty_subst2 = mkEmptyTCvSubst (getTCvInScope subst1)
982

983
      ; all_co <- checkTvConstraintsTcS skol_info skol_tvs $
984 985
                  go skol_tvs empty_subst2 bndrs2

986 987 988 989 990 991 992 993
      ; setWantedEq orig_dest all_co
      ; stopWith ev "Deferred polytype equality" } }

 | otherwise
 = do { traceTcS "Omitting decomposition of given polytype equality" $
        pprEq s1 s2    -- See Note [Do not decompose given polytype equalities]
      ; stopWith ev "Discard given polytype equality" }

994 995 996 997 998 999 1000 1001 1002 1003 1004
 where
    unify :: CtLoc -> Role -> TcType -> TcType -> TcS (TcCoercion, Cts)
    -- This version returns the wanted constraint rather
    -- than putting it in the work list
    unify loc role ty1 ty2
      | ty1 `tcEqType` ty2
      = return (mkTcReflCo role ty1, emptyBag)
      | otherwise
      = do { (wanted, co) <- newWantedEq loc role ty1 ty2
           ; return (co, unitBag (mkNonCanonical wanted)) }

1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028
---------------------------------
-- | Compare types for equality, while zonking as necessary. Gives up
-- as soon as it finds that two types are not equal.
-- This is quite handy when some unification has made two
-- types in an inert wanted to be equal. We can discover the equality without
-- flattening, which is sometimes very expensive (in the case of type functions).
-- In particular, this function makes a ~20% improvement in test case
-- perf/compiler/T5030.
--
-- Returns either the (partially zonked) types in the case of
-- inequality, or the one type in the case of equality. canEqReflexive is
-- a good next step in the 'Right' case. Returning 'Left' is always safe.
--
-- NB: This does *not* look through type synonyms. In fact, it treats type
-- synonyms as rigid constructors. In the future, it might be convenient
-- to look at only those arguments of type synonyms that actually appear
-- in the synonym RHS. But we're not there yet.
zonk_eq_types :: TcType -> TcType -> TcS (Either (Pair TcType) TcType)
zonk_eq_types = go
  where
    go (TyVarTy tv1) (TyVarTy tv2) = tyvar_tyvar tv1 tv2
    go (TyVarTy tv1) ty2           = tyvar NotSwapped tv1 ty2
    go ty1 (TyVarTy tv2)           = tyvar IsSwapped  tv2 ty1

1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048
    -- We handle FunTys explicitly here despite the fact that they could also be
    -- treated as an application. Why? Well, for one it's cheaper to just look
    -- at two types (the argument and result types) than four (the argument,
    -- result, and their RuntimeReps). Also, we haven't completely zonked yet,
    -- so we may run into an unzonked type variable while trying to compute the
    -- RuntimeReps of the argument and result types. This can be observed in
    -- testcase tc269.
    go ty1 ty2
      | Just (arg1, res1) <- split1
      , Just (arg2, res2) <- split2
      = do { res_a <- go arg1 arg2
           ; res_b <- go res1 res2
           ; return $ combine_rev mkFunTy res_b res_a
           }
      | isJust split1 || isJust split2
      = bale_out ty1 ty2
      where
        split1 = tcSplitFunTy_maybe ty1
        split2 = tcSplitFunTy_maybe ty2

1049 1050 1051
    go ty1 ty2
      | Just (tc1, tys1) <- tcRepSplitTyConApp_maybe ty1
      , Just (tc2, tys2) <- tcRepSplitTyConApp_maybe ty2
1052 1053 1054 1055 1056 1057 1058 1059 1060
      = if tc1 == tc2 && tys1 `equalLength` tys2
          -- Crucial to check for equal-length args, because
          -- we cannot assume that the two args to 'go' have
          -- the same kind.  E.g go (Proxy *      (Maybe Int))
          --                        (Proxy (*->*) Maybe)
          -- We'll call (go (Maybe Int) Maybe)
          -- See Trac #13083
        then tycon tc1 tys1 tys2
        else bale_out ty1 ty2
1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072

    go ty1 ty2
      | Just (ty1a, ty1b) <- tcRepSplitAppTy_maybe ty1
      , Just (ty2a, ty2b) <- tcRepSplitAppTy_maybe ty2
      = do { res_a <- go ty1a ty2a
           ; res_b <- go ty1b ty2b
           ; return $ combine_rev mkAppTy res_b res_a }

    go ty1@(LitTy lit1) (LitTy lit2)
      | lit1 == lit2
      = return (Right ty1)

1073 1074 1075 1076
    go ty1 ty2 = bale_out ty1 ty2
      -- We don't handle more complex forms here

    bale_out ty1 ty2 = return $ Left (Pair ty1 ty2)
1077 1078 1079

    tyvar :: SwapFlag -> TcTyVar -> TcType
          -> TcS (Either (Pair TcType) TcType)
1080
      -- Try to do as little as possible, as anything we do here is redundant
1081 1082 1083 1084 1085 1086 1087 1088
      -- with flattening. In particular, no need to zonk kinds. That's why
      -- we don't use the already-defined zonking functions
    tyvar swapped tv ty
      = case tcTyVarDetails tv of
          MetaTv { mtv_ref = ref }
            -> do { cts <- readTcRef ref
                  ; case cts of
                      Flexi        -> give_up
Simon Peyton Jones's avatar
Simon Peyton Jones committed
1089 1090
                      Indirect ty' -> do { trace_indirect tv ty'
                                         ; unSwap swapped go ty' ty } }
1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
          _ -> give_up
      where
        give_up = return $ Left $ unSwap swapped Pair (mkTyVarTy tv) ty

    tyvar_tyvar tv1 tv2
      | tv1 == tv2 = return (Right (mkTyVarTy tv1))
      | otherwise  = do { (ty1', progress1) <- quick_zonk tv1
                        ; (ty2', progress2) <- quick_zonk tv2
                        ; if progress1 || progress2
                          then go ty1' ty2'
                          else return $ Left (Pair (TyVarTy tv1) (TyVarTy tv2)) }

Simon Peyton Jones's avatar
Simon Peyton Jones committed
1103 1104 1105 1106
    trace_indirect tv ty
       = traceTcS "Following filled tyvar (zonk_eq_types)"
                  (ppr tv <+> equals <+> ppr ty)

1107 1108 1109 1110 1111
    quick_zonk tv = case tcTyVarDetails tv of
      MetaTv { mtv_ref = ref }
        -> do { cts <- readTcRef ref
              ; case cts of
                  Flexi        -> return (TyVarTy tv, False)
Simon Peyton Jones's avatar
Simon Peyton Jones committed
1112 1113
                  Indirect ty' -> do { trace_indirect tv ty'
                                     ; return (ty', True) } }
1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140
      _ -> return (TyVarTy tv, False)

      -- This happens for type families, too. But recall that failure
      -- here just means to try harder, so it's OK if the type function
      -- isn't injective.
    tycon :: TyCon -> [TcType] -> [TcType]
          -> TcS (Either (Pair TcType) TcType)
    tycon tc tys1 tys2
      = do { results <- zipWithM go tys1 tys2
           ; return $ case combine_results results of
               Left tys  -> Left (mkTyConApp tc <$> tys)
               Right tys -> Right (mkTyConApp tc tys) }

    combine_results :: [Either (Pair TcType) TcType]
                    -> Either (Pair [TcType]) [TcType]
    combine_results = bimap (fmap reverse) reverse .
                      foldl' (combine_rev (:)) (Right [])

      -- combine (in reverse) a new result onto an already-combined result
    combine_rev :: (a -> b -> c)
                -> Either (Pair b) b
                -> Either (Pair a) a
                -> Either (Pair c) c
    combine_rev f (Left list) (Left elt) = Left (f <$> elt     <*> list)
    combine_rev f (Left list) (Right ty) = Left (f <$> pure ty <*> list)
    combine_rev f (Right tys) (Left elt) = Left (f <$> elt     <*> pure tys)
    combine_rev f (Right tys) (Right ty) = Right (f ty tys)
1141

1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161
{- See Note [Unwrap newtypes first]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider
  newtype N m a = MkN (m a)
Then N will get a conservative, Nominal role for its second paramter 'a',
because it appears as an argument to the unknown 'm'. Now consider
  [W] N Maybe a  ~R#  N Maybe b

If we decompose, we'll get
  [W] a ~N# b

But if instead we unwrap we'll get
  [W] Maybe a ~R# Maybe b
which in turn gives us
  [W] a ~R# b
which is easier to satisfy.

Bottom line: unwrap newtypes before decomposing them!
c.f. Trac #9123 comment:52,53 for a compelling example.

1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179
Note [Newtypes can blow the stack]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Suppose we have

  newtype X = MkX (Int -> X)
  newtype Y = MkY (Int -> Y)

and now wish to prove

  [W] X ~R Y

This Wanted will loop, expanding out the newtypes ever deeper looking
for a solid match or a solid discrepancy. Indeed, there is something
appropriate to this looping, because X and Y *do* have the same representation,
in the limit -- they're both (Fix ((->) Int)). However, no finitely-sized
coercion will ever witness it. This loop won't actually cause GHC to hang,
though, because we check our depth when unwrapping newtypes.

1180 1181 1182 1183 1184 1185 1186 1187 1188
Note [Eager reflexivity check]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Suppose we have

  newtype X = MkX (Int -> X)

and

  [W] X ~R X
1189

1190 1191 1192 1193 1194
Naively, we would start unwrapping X and end up in a loop. Instead,
we do this eager reflexivity check. This is necessary only for representational
equality because the flattener technology deals with the similar case
(recursive type families) for nominal equality.

1195 1196
Note that this check does not catch all cases, but it will catch the cases
we're most worried about, types like X above that are actually inhabited.
1197

eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
1198
Here's another place where this reflexivity check is key:
1199 1200 1201
Consider trying to prove (f a) ~R (f a). The AppTys in there can't
be decomposed, because representational equality isn't congruent with respect
to AppTy. So, when canonicalising the equality above, we get stuck and
1202
would normally produce a CIrredCan. However, we really do want to
1203 1204 1205 1206 1207 1208 1209 1210 1211
be able to solve (f a) ~R (f a). So, in the representational case only,
we do a reflexivity check.

(This would be sound in the nominal case, but unnecessary, and I [Richard
E.] am worried that it would slow down the common case.)
-}

------------------------
-- | We're able to unwrap a newtype. Update the bits accordingly.
1212
can_eq_newtype_nc :: CtEvidence           -- ^ :: ty1 ~ ty2
1213
                  -> SwapFlag
1214 1215
                  -> TcType                                    -- ^ ty1
                  -> ((Bag GlobalRdrElt, TcCoercion), TcType)  -- ^ :: ty1 ~ ty1'
1216 1217 1218
                  -> TcType               -- ^ ty2
                  -> TcType               -- ^ ty2, with type synonyms
                  -> TcS (StopOrContinue Ct)
1219
can_eq_newtype_nc ev swapped ty1 ((gres, co), ty1') ty2 ps_ty2
1220
  = do { traceTcS "can_eq_newtype_nc" $
1221
         vcat [ ppr ev, ppr swapped, ppr co, ppr gres, ppr ty1', ppr ty2 ]
1222 1223

         -- check for blowing our stack:
1224 1225
         -- See Note [Newtypes can blow the stack]
       ; checkReductionDepth (ctEvLoc ev) ty1
1226
       ; addUsedGREs (bagToList gres)
1227 1228 1229
           -- we have actually used the newtype constructor here, so
           -- make sure we don't warn about importing it!

1230 1231 1232
       ; new_ev <- rewriteEqEvidence ev swapped ty1' ps_ty2
                                     (mkTcSymCo co) (mkTcReflCo Representational ps_ty2)
       ; can_eq_nc False new_ev ReprEq ty1' ty1' ty2 ps_ty2 }
1233

1234
---------
eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
1235
-- ^ Decompose a type application.
1236
-- All input types must be flat. See Note [Canonicalising type applications]
1237 1238
-- Nominal equality only!
can_eq_app :: CtEvidence       -- :: s1 t1 ~N s2 t2
1239 1240 1241
           -> Xi -> Xi         -- s1 t1
           -> Xi -> Xi         -- s2 t2
           -> TcS (StopOrContinue Ct)
eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
1242 1243

-- AppTys only decompose for nominal equality, so this case just leads
eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
1244
-- to an irreducible constraint; see typecheck/should_compile/T10494
Simon Peyton Jones's avatar
Simon Peyton Jones committed
1245
-- See Note [Decomposing equality], note {4}
1246
can_eq_app ev s1 t1 s2 t2
1247
  | CtDerived { ctev_loc = loc } <- ev
1248 1249 1250 1251
  = do { unifyDeriveds loc [Nominal, Nominal] [s1, t1] [s2, t2]
       ; stopWith ev "Decomposed [D] AppTy" }
  | CtWanted { ctev_dest = dest, ctev_loc = loc } <- ev
  = do { co_s <- unifyWanted loc Nominal s1 s2
1252 1253 1254 1255
       ; let arg_loc
               | isNextArgVisible s1 = loc
               | otherwise           = updateCtLocOrigin loc toInvisibleOrigin
       ; co_t <- unifyWanted arg_loc Nominal t1 t2
1256 1257 1258
       ; let co = mkAppCo co_s co_t
       ; setWantedEq dest co
       ; stopWith ev "Decomposed [W] AppTy" }
1259 1260 1261 1262 1263 1264 1265 1266 1267

    -- If there is a ForAll/(->) mismatch, the use of the Left coercion
    -- below is ill-typed, potentially leading to a panic in splitTyConApp
    -- Test case: typecheck/should_run/Typeable1
    -- We could also include this mismatch check above (for W and D), but it's slow
    -- and we'll get a better error message not doing it
  | s1k `mismatches` s2k
  = canEqHardFailure ev (s1 `mkAppTy` t1) (s2 `mkAppTy` t2)

1268 1269
  | CtGiven { ctev_evar = evar, ctev_loc = loc } <- ev
  = do { let co   = mkTcCoVarCo evar
1270 1271
             co_s = mkTcLRCo CLeft  co
             co_t = mkTcLRCo CRight co
1272
       ; evar_s <- newGivenEvVar loc ( mkTcEqPredLikeEv ev s1 s2
1273
                                     , evCoercion co_s )
1274
       ; evar_t <- newGivenEvVar loc ( mkTcEqPredLikeEv ev t1 t2
1275
                                     , evCoercion co_t )
1276 1277
       ; emitWorkNC [evar_t]
       ; canEqNC evar_s NomEq s1 s2 }
1278 1279 1280 1281 1282 1283 1284 1285

  where
    s1k = typeKind s1
    s2k = typeKind s2

    k1 `mismatches` k2
      =  isForAllTy k1 && not (isForAllTy k2)
      || not (isForAllTy k1) && isForAllTy k2
1286

1287 1288
-----------------------
-- | Break apart an equality over a casted type
Simon Peyton Jones's avatar
Simon Peyton Jones committed
1289
-- looking like   (ty1 |> co1) ~ ty2   (modulo a swap-flag)
1290 1291 1292 1293
canEqCast :: Bool         -- are both types flat?
          -> CtEvidence
          -> EqRel
          -> SwapFlag
Simon Peyton Jones's avatar
Simon Peyton Jones committed
1294 1295
          -> TcType -> Coercion   -- LHS (res. RHS), ty1 |> co1
          -> TcType -> TcType     -- RHS (res. LHS), ty2 both normal and pretty
1296 1297 1298 1299 1300
          -> TcS (StopOrContinue Ct)
canEqCast flat ev eq_rel swapped ty1 co1 ty2 ps_ty2
  = do { traceTcS "Decomposing cast" (vcat [ ppr ev
                                           , ppr ty1 <+> text "|>" <+> ppr co1
                                           , ppr ps_ty2 ])
1301
       ; new_ev <- rewriteEqEvidence ev swapped ty1 ps_ty2
Ningning Xie's avatar
Ningning Xie committed
1302
                                     (mkTcGReflRightCo role ty1 co1)
1303 1304
                                     (mkTcReflCo role ps_ty2)
       ; can_eq_nc flat new_ev eq_rel ty1 ty1 ty2 ps_ty2 }
1305 1306 1307
  where
    role = eqRelRole eq_rel

1308
------------------------
1309 1310 1311 1312
canTyConApp :: CtEvidence -> EqRel
            -> TyCon -> [TcType]
            -> TyCon -> [TcType]
            -> TcS (StopOrContinue Ct)
1313
-- See Note [Decomposing TyConApps]
1314
canTyConApp ev eq_rel tc1 tys1 tc2 tys2
1315
  | tc1 == tc2
1316
  , tys1 `equalLength` tys2
1317
  = do { inerts <- getTcSInerts
1318 1319
       ; if can_decompose inerts
         then do { traceTcS "canTyConApp"
1320 1321 1322 1323
                       (ppr ev $$ ppr eq_rel $$ ppr tc1 $$ ppr tys1 $$ ppr tys2)
                 ; canDecomposableTyConAppOK ev eq_rel tc1 tys1 tys2
                 ; stopWith ev "Decomposed TyConApp" }
         else canEqFailure ev eq_rel ty1 ty2 }
1324

1325 1326
  -- See Note [Skolem abstract data] (at tyConSkolem)
  | tyConSkolem tc1 || tyConSkolem tc2
1327
  = do { traceTcS "canTyConApp: skolem abstract" (ppr tc1 $$ ppr tc2)
1328
       ; continueWith (mkIrredCt ev) }
1329

1330 1331
  -- Fail straight away for better error messages
  -- See Note [Use canEqFailure in canDecomposableTyConApp]
1332 1333
  | eq_rel == ReprEq && not (isGenerativeTyCon tc1 Representational &&
                             isGenerativeTyCon tc2 Representational)
1334 1335
  = canEqFailure ev eq_rel ty1 ty2
  | otherwise
1336
  = canEqHardFailure ev ty1 ty2
1337 1338 1339 1340
  where
    ty1 = mkTyConApp tc1 tys1
    ty2 = mkTyConApp tc2 tys2

1341 1342 1343
    loc  = ctEvLoc ev
    pred = ctEvPred ev

1344 1345 1346 1347 1348
     -- See Note [Decomposing equality]
    can_decompose inerts
      =  isInjectiveTyCon tc1 (eqRelRole eq_rel)
      || (ctEvFlavour ev /= Given && isEmptyBag (matchableGivens loc pred inerts))

1349 1350 1351 1352 1353
{-
Note [Use canEqFailure in canDecomposableTyConApp]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We must use canEqFailure, not canEqHardFailure here, because there is
the possibility of success if working with a representational equality.
eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
1354
Here is one case:
1355 1356 1357 1358 1359

  type family TF a where TF Char = Bool
  data family DF a
  newtype instance DF Bool = MkDF Int

eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
1360
Suppose we are canonicalising (Int ~R DF (TF a)), where we don't yet
1361 1362
know `a`. This is *not* a hard failure, because we might soon learn
that `a` is, in fact, Char, and then the equality succeeds.
1363

eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
1364 1365
Here is another case:

eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
1366
  [G] Age ~R Int
eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
1367 1368 1369 1370

where Age's constructor is not in scope. We don't want to report
an "inaccessible code" error in the context of this Given!

eir@cis.upenn.edu's avatar
eir@cis.upenn.edu committed
1371 1372 1373 1374 1375 1376 1377 1378 1379 1380
For example, see typecheck/should_compile/T10493, repeated here:

  import Data.Ord (Down)  -- no constructor

  foo :: Coercible (Down Int) Int => Down Int -> Int
  foo = coerce

That should compile, but only because we use canEqFailure and not
canEqHardFailure.

1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419
Note [Decomposing equality]
~~~~~~~~~~~~~~~~~~~~~~~~~~~
If we have a constraint (of any flavour and role) that looks like
T tys1 ~ T tys2, what can we conclude about tys1 and tys2? The answer,
of course, is "it depends". This Note spells it all out.

In this Note, "decomposition" refers to taking the constraint
  [fl] (T tys1 ~X T tys2)
(for some flavour fl and some role X) and replacing it with
  [fls'] (tys1 ~Xs' tys2)
where that notation indicates a list of new constraints, where the
new constraints may have different flavours and different roles.

The key property to consider is injectivity. When decomposing a Given the
decomposition is sound if and only if T is injective in all of its type
arguments. When decomposing a Wanted, the decomposition is sound (assuming the
correct roles in the produced equality constraints), but it may be a guess --
that is, an unforced decision by the constraint solver. Decomposing Wanteds
over injective TyCons does not entail guessing. But sometimes we want to
decompose a Wanted even when the TyCon involved is not injective! (See below.)

So, in broad strokes, we want this rule:

(*) Decompose a constraint (T tys1 ~X T tys2) if and only if T is injective
at role X.

Pursuing the details requires exploring three axes:
* Flavour: Given vs. Derived vs. Wanted
* Role: Nominal vs. Representational
* TyCon species: datatype vs. newtype vs. data family vs. type family vs. type variable

(So a type variable isn't a TyCon, but it's convenient to put the AppTy case
in the same table.)

Right away, we can say that Derived behaves just as Wanted for the purposes
of decomposition. The difference between Derived and Wanted is the handling of
evidence. Since decomposition in these cases isn't a matter of soundness but of
guessing, we want the same behavior regardless of evidence.

Simon Peyton Jones's avatar
Simon Peyton Jones committed
1420 1421 1422 1423 1424 1425
Here is a table (discussion following) detailing where decomposition of
   (T s1 ... sn) ~r (T t1 .. tn)
is allowed.  The first four lines (Data types ... type family) refer
to TyConApps with various TyCons T; the last line is for AppTy, where
there is presumably a type variable at the head, so it's actually
   (s s1 ... sn) ~r (t t1 .. tn)