3 solveInteract, AtomicInert,
4 InertSet, emptyInert, updInertSet, extractUnsolved, solveOne,
8 #include "HsVersions.h"
31 import Control.Monad ( when )
40 import qualified Data.Map as Map
43 import Control.Monad( zipWithM, unless )
44 import FastString ( sLit )
48 Note [InertSet invariants]
49 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
51 An InertSet is a bag of canonical constraints, with the following invariants:
53 1 No two constraints react with each other.
55 A tricky case is when there exists a given (solved) dictionary
56 constraint and a wanted identical constraint in the inert set, but do
57 not react because reaction would create loopy dictionary evidence for
58 the wanted. See note [Recursive dictionaries]
60 2 Given equalities form an idempotent substitution [none of the
61 given LHS's occur in any of the given RHS's or reactant parts]
63 3 Wanted equalities also form an idempotent substitution
64 4 The entire set of equalities is acyclic.
66 5 Wanted dictionaries are inert with the top-level axiom set
68 6 Equalities of the form tv1 ~ tv2 always have a touchable variable
69 on the left (if possible).
70 7 No wanted constraints tv1 ~ tv2 with tv1 touchable. Such constraints
71 will be marked as solved right before being pushed into the inert set.
72 See note [Touchables and givens].
74 Note that 6 and 7 are /not/ enforced by canonicalization but rather by
75 insertion in the inert list, ie by TcInteract.
77 During the process of solving, the inert set will contain some
78 previously given constraints, some wanted constraints, and some given
79 constraints which have arisen from solving wanted constraints. For
80 now we do not distinguish between given and solved constraints.
82 Note that we must switch wanted inert items to given when going under an
83 implication constraint (when in top-level inference mode).
85 Note [InertSet FlattenSkolemEqClass]
86 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
87 The inert_fsks field of the inert set contains an "inverse map" of all the
88 flatten skolem equalities in the inert set. For instance, if inert_cts looks
95 Then, the inert_fsks fields holds the following map:
96 fsk2 |-> { fsk1, fsk3 }
98 Along with the necessary coercions to convert fsk1 and fsk3 back to fsk2
99 and fsk4 back to fsk5. Hence, the invariants of the inert_fsks field are:
101 (a) All TcTyVars in the domain and range of inert_fsks are flatten skolems
102 (b) All TcTyVars in the domain of inert_fsk occur naked as rhs in some
103 equalities of inert_cts
104 (c) For every mapping fsk1 |-> { (fsk2,co), ... } it must be:
107 The role of the inert_fsks is to make it easy to maintain the equivalence
108 class of each flatten skolem, which is much needed to correctly do spontaneous
109 solving. See Note [Loopy Spontaneous Solving]
112 -- See Note [InertSet invariants]
114 = IS { inert_cts :: Bag.Bag CanonicalCt
115 , inert_fsks :: Map.Map TcTyVar [(TcTyVar,Coercion)] }
116 -- See Note [InertSet FlattenSkolemEqClass]
118 instance Outputable InertSet where
119 ppr is = vcat [ vcat (map ppr (Bag.bagToList $ inert_cts is))
120 , vcat (map (\(v,rest) -> ppr v <+> text "|->" <+> hsep (map (ppr.fst) rest))
121 (Map.toList $ inert_fsks is)
125 emptyInert :: InertSet
126 emptyInert = IS { inert_cts = Bag.emptyBag, inert_fsks = Map.empty }
128 updInertSet :: InertSet -> AtomicInert -> InertSet
129 -- Introduces an element in the inert set for the first time
130 updInertSet (IS { inert_cts = cts, inert_fsks = fsks })
131 item@(CTyEqCan { cc_id = cv
134 | Just tv2 <- tcGetTyVar_maybe xi,
135 FlatSkol {} <- tcTyVarDetails tv1,
136 FlatSkol {} <- tcTyVarDetails tv2
137 = let cts' = cts `Bag.snocBag` item
138 fsks' = Map.insertWith (++) tv2 [(tv1, mkCoVarCoercion cv)] fsks
139 -- See Note [InertSet FlattenSkolemEqClass]
140 in IS { inert_cts = cts', inert_fsks = fsks' }
141 updInertSet (IS { inert_cts = cts
142 , inert_fsks = fsks }) item
143 = let cts' = cts `Bag.snocBag` item
144 in IS { inert_cts = cts', inert_fsks = fsks }
146 foldlInertSetM :: (Monad m) => (a -> AtomicInert -> m a) -> a -> InertSet -> m a
147 foldlInertSetM k z (IS { inert_cts = cts })
148 = Bag.foldlBagM k z cts
150 extractUnsolved :: InertSet -> (InertSet, CanonicalCts)
151 extractUnsolved is@(IS {inert_cts = cts})
152 = (is { inert_cts = cts'}, unsolved)
153 where (unsolved, cts') = Bag.partitionBag isWantedCt cts
156 getFskEqClass :: InertSet -> TcTyVar -> [(TcTyVar,Coercion)]
157 -- Precondition: tv is a FlatSkol. See Note [InertSet FlattenSkolemEqClass]
158 getFskEqClass (IS { inert_cts = cts, inert_fsks = fsks }) tv
159 = case lkpTyEqCanByLhs of
160 Nothing -> fromMaybe [] (Map.lookup tv fsks)
162 case tcGetTyVar_maybe (cc_rhs ceq) of
163 Just tv_rhs | FlatSkol {} <- tcTyVarDetails tv_rhs
164 -> let ceq_co = mkSymCoercion $ mkCoVarCoercion (cc_id ceq)
165 mk_co (v,c) = (v, mkTransCoercion c ceq_co)
166 in (tv_rhs, ceq_co): map mk_co (fromMaybe [] $ Map.lookup tv fsks)
168 where lkpTyEqCanByLhs = Bag.foldlBag lkp Nothing cts
169 lkp :: Maybe CanonicalCt -> CanonicalCt -> Maybe CanonicalCt
170 lkp Nothing ct@(CTyEqCan {cc_tyvar = tv'}) | tv' == tv = Just ct
171 lkp other _ct = other
174 isWantedCt :: CanonicalCt -> Bool
175 isWantedCt ct = isWanted (cc_flavor ct)
178 data Inert = IS { class_inerts :: FiniteMap Class Atomics
179 ip_inerts :: FiniteMap Class Atomics
180 tyfun_inerts :: FiniteMap TyCon Atomics
181 tyvar_inerts :: FiniteMap TyVar Atomics
184 Later should we also separate out givens and wanteds?
189 Note [Touchables and givens]
190 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
191 Touchable variables will never show up in givens which are inputs to
192 the solver. However, touchables may show up in givens generated by the flattener.
207 which can be put in the inert set. Suppose we also have a wanted
211 We cannot rewrite the given G alpha ~g b using the wanted alpha ~w
212 Int. Instead, after reacting alpha ~w Int with the whole inert set,
213 we observe that we can solve it by unifying alpha with Int, so we mark
214 it as solved and put it back in the *work list*. [We also immediately unify
215 alpha := Int, without telling anyone, see trySpontaneousSolve function, to
216 avoid doing this in the end.]
218 Later, because it is solved (given, in effect), we can use it to rewrite
219 G alpha ~g b to G Int ~g b, which gets put back in the work list. Eventually,
220 we will dispatch the remaining wanted constraints using the top-level axioms.
222 Finally, note that after reacting a wanted equality with the entire inert set
223 we may end up with something like
227 which we should flip around to generate the solved constraint alpha ~s b.
229 %*********************************************************************
231 * Main Interaction Solver *
233 **********************************************************************
237 1. Canonicalise (unary)
238 2. Pairwise interaction (binary)
239 * Take one from work list
240 * Try all pair-wise interactions with each constraint in inert
241 3. Try to solve spontaneously for equalities involving touchables
242 4. Top-level interaction (binary wrt top-level)
243 Superclass decomposition belongs in (4), see note [Superclasses]
247 type AtomicInert = CanonicalCt -- constraint pulled from InertSet
248 type WorkItem = CanonicalCt -- constraint pulled from WorkList
250 type WorkList = CanonicalCts -- A mixture of Given, Wanted, and Solved
251 type SWorkList = WorkList -- A worklist of solved
254 listToWorkList :: [WorkItem] -> WorkList
255 listToWorkList = Bag.listToBag
257 unionWorkLists :: WorkList -> WorkList -> WorkList
258 unionWorkLists = Bag.unionBags
260 foldlWorkListM :: (Monad m) => (a -> WorkItem -> m a) -> a -> WorkList -> m a
261 foldlWorkListM = Bag.foldlBagM
263 isEmptyWorkList :: WorkList -> Bool
264 isEmptyWorkList = Bag.isEmptyBag
266 emptyWorkList :: WorkList
267 emptyWorkList = Bag.emptyBag
269 singletonWorkList :: CanonicalCt -> WorkList
270 singletonWorkList ct = singleCCan ct
273 = Stop -- Work item is consumed
274 | ContinueWith WorkItem -- Not consumed
276 instance Outputable StopOrContinue where
277 ppr Stop = ptext (sLit "Stop")
278 ppr (ContinueWith w) = ptext (sLit "ContinueWith") <+> ppr w
280 -- Results after interacting a WorkItem as far as possible with an InertSet
282 = SR { sr_inerts :: InertSet
283 -- The new InertSet to use (REPLACES the old InertSet)
284 , sr_new_work :: WorkList
285 -- Any new work items generated (should be ADDED to the old WorkList)
287 -- sr_stop = Just workitem => workitem is *not* in sr_inerts and
288 -- workitem is inert wrt to sr_inerts
289 , sr_stop :: StopOrContinue
292 instance Outputable StageResult where
293 ppr (SR { sr_inerts = inerts, sr_new_work = work, sr_stop = stop })
294 = ptext (sLit "SR") <+>
295 braces (sep [ ptext (sLit "inerts =") <+> ppr inerts <> comma
296 , ptext (sLit "new work =") <+> ppr work <> comma
297 , ptext (sLit "stop =") <+> ppr stop])
299 type SimplifierStage = WorkItem -> InertSet -> TcS StageResult
301 -- Combine a sequence of simplifier 'stages' to create a pipeline
302 runSolverPipeline :: [(String, SimplifierStage)]
303 -> InertSet -> WorkItem
304 -> TcS (InertSet, WorkList)
305 -- Precondition: non-empty list of stages
306 runSolverPipeline pipeline inerts workItem
307 = do { traceTcS "Start solver pipeline" $
308 vcat [ ptext (sLit "work item =") <+> ppr workItem
309 , ptext (sLit "inerts =") <+> ppr inerts]
311 ; let itr_in = SR { sr_inerts = inerts
312 , sr_new_work = emptyWorkList
313 , sr_stop = ContinueWith workItem }
314 ; itr_out <- run_pipeline pipeline itr_in
316 = case sr_stop itr_out of
317 Stop -> sr_inerts itr_out
318 ContinueWith item -> sr_inerts itr_out `updInertSet` item
319 ; return (new_inert, sr_new_work itr_out) }
321 run_pipeline :: [(String, SimplifierStage)]
322 -> StageResult -> TcS StageResult
323 run_pipeline [] itr = return itr
324 run_pipeline _ itr@(SR { sr_stop = Stop }) = return itr
326 run_pipeline ((name,stage):stages)
327 (SR { sr_new_work = accum_work
329 , sr_stop = ContinueWith work_item })
330 = do { itr <- stage work_item inerts
331 ; traceTcS ("Stage result (" ++ name ++ ")") (ppr itr)
332 ; let itr' = itr { sr_new_work = sr_new_work itr
333 `unionWorkLists` accum_work }
334 ; run_pipeline stages itr' }
338 Inert: {c ~ d, F a ~ t, b ~ Int, a ~ ty} (all given)
339 Reagent: a ~ [b] (given)
341 React with (c~d) ==> IR (ContinueWith (a~[b])) True []
342 React with (F a ~ t) ==> IR (ContinueWith (a~[b])) False [F [b] ~ t]
343 React with (b ~ Int) ==> IR (ContinueWith (a~[Int]) True []
346 Inert: {c ~w d, F a ~g t, b ~w Int, a ~w ty}
349 React with (c ~w d) ==> IR (ContinueWith (a~[b])) True []
350 React with (F a ~g t) ==> IR (ContinueWith (a~[b])) True [] (can't rewrite given with wanted!)
354 Inert: {a ~ Int, F Int ~ b} (given)
355 Reagent: F a ~ b (wanted)
357 React with (a ~ Int) ==> IR (ContinueWith (F Int ~ b)) True []
358 React with (F Int ~ b) ==> IR Stop True [] -- after substituting we re-canonicalize and get nothing
361 -- Main interaction solver: we fully solve the worklist 'in one go',
362 -- returning an extended inert set.
364 -- See Note [Touchables and givens].
365 solveInteract :: InertSet -> WorkList -> TcS InertSet
366 solveInteract inert ws
367 = do { dyn_flags <- getDynFlags
368 ; solveInteractWithDepth (ctxtStkDepth dyn_flags,0,[]) inert ws
370 solveOne :: InertSet -> WorkItem -> TcS InertSet
371 solveOne inerts workItem
372 = do { dyn_flags <- getDynFlags
373 ; solveOneWithDepth (ctxtStkDepth dyn_flags,0,[]) inerts workItem
377 solveInteractWithDepth :: (Int, Int, [WorkItem])
378 -> InertSet -> WorkList -> TcS InertSet
379 solveInteractWithDepth ctxt@(max_depth,n,stack) inert ws
384 = solverDepthErrorTcS n stack
387 = do { traceTcS "solveInteractWithDepth" $
388 vcat [ text "Current depth =" <+> ppr n
389 , text "Max depth =" <+> ppr max_depth
391 ; foldlWorkListM (solveOneWithDepth ctxt) inert ws }
394 -- Fully interact the given work item with an inert set, and return a
395 -- new inert set which has assimilated the new information.
396 solveOneWithDepth :: (Int, Int, [WorkItem])
397 -> InertSet -> WorkItem -> TcS InertSet
398 solveOneWithDepth (max_depth, n, stack) inert work
399 = do { traceTcS0 (indent ++ "Solving {") (ppr work)
400 ; (new_inert, new_work) <- runSolverPipeline thePipeline inert work
402 ; traceTcS0 (indent ++ "Subgoals:") (ppr new_work)
404 -- Recursively solve the new work generated
405 -- from workItem, with a greater depth
406 ; res_inert <- solveInteractWithDepth (max_depth, n+1, work:stack)
409 ; traceTcS0 (indent ++ "Done }") (ppr work)
412 indent = replicate (2*n) ' '
414 thePipeline :: [(String,SimplifierStage)]
415 thePipeline = [ ("interact with inerts", interactWithInertsStage)
416 , ("spontaneous solve", spontaneousSolveStage)
417 , ("top-level reactions", topReactionsStage) ]
420 *********************************************************************************
422 The spontaneous-solve Stage
424 *********************************************************************************
427 spontaneousSolveStage :: SimplifierStage
428 spontaneousSolveStage workItem inerts
429 = do { mSolve <- trySpontaneousSolve workItem inerts
431 Nothing -> -- no spontaneous solution for him, keep going
432 return $ SR { sr_new_work = emptyWorkList
434 , sr_stop = ContinueWith workItem }
436 Just workList' -> -- He has been solved; workList' are all givens
437 return $ SR { sr_new_work = workList'
442 {-- This is all old code, but does not quite work now. The problem is that due to
443 Note [Loopy Spontaneous Solving] we may have unflattened a type, to be able to
444 perform a sneaky unification. This unflattening means that we may have to recanonicalize
445 a given (solved) equality, this is why the result of trySpontaneousSolve is now a list
446 of constraints (instead of an atomic solved constraint). We would have to react all of
447 them once again with the worklist but that is very tiresome. Instead we throw them back
450 | isWantedCt workItem
451 -- Original was wanted we have now made him given so
452 -- we have to ineract him with the inerts again because
453 -- of the change in his status. This may produce some work.
454 -> do { traceTcS "recursive interact with inerts {" $ vcat
455 [ text "work = " <+> ppr workItem'
456 , text "inerts = " <+> ppr inerts ]
457 ; itr_again <- interactWithInertsStage workItem' inerts
458 ; case sr_stop itr_again of
459 Stop -> pprPanic "BUG: Impossible to happen" $
460 vcat [ text "Original workitem:" <+> ppr workItem
461 , text "Spontaneously solved:" <+> ppr workItem'
462 , text "Solved was consumed, when reacting with inerts:"
463 , nest 2 (ppr inerts) ]
464 ContinueWith workItem'' -- Now *this* guy is inert wrt to inerts
465 -> do { traceTcS "end recursive interact }" $ ppr workItem''
466 ; return $ SR { sr_new_work = sr_new_work itr_again
467 , sr_inerts = sr_inerts itr_again
468 `extendInertSet` workItem''
472 -> return $ SR { sr_new_work = emptyWorkList
473 , sr_inerts = inerts `extendInertSet` workItem'
477 -- @trySpontaneousSolve wi@ solves equalities where one side is a
478 -- touchable unification variable. Returns:
479 -- * Nothing if we were not able to solve it
480 -- * Just wi' if we solved it, wi' (now a "given") should be put in the work list.
481 -- See Note [Touchables and givens]
482 -- Note, just passing the inerts through for the skolem equivalence classes
483 trySpontaneousSolve :: WorkItem -> InertSet -> TcS (Maybe SWorkList)
484 trySpontaneousSolve (CTyEqCan { cc_id = cv, cc_flavor = gw, cc_tyvar = tv1, cc_rhs = xi }) inerts
487 | Just tv2 <- tcGetTyVar_maybe xi
488 = do { tch1 <- isTouchableMetaTyVar tv1
489 ; tch2 <- isTouchableMetaTyVar tv2
490 ; case (tch1, tch2) of
491 (True, True) -> trySpontaneousEqTwoWay inerts cv gw tv1 tv2
492 (True, False) -> trySpontaneousEqOneWay inerts cv gw tv1 xi
493 (False, True) | tyVarKind tv1 `isSubKind` tyVarKind tv2
494 -> trySpontaneousEqOneWay inerts cv gw tv2 (mkTyVarTy tv1)
495 _ -> return Nothing }
497 = do { tch1 <- isTouchableMetaTyVar tv1
498 ; if tch1 then trySpontaneousEqOneWay inerts cv gw tv1 xi
499 else return Nothing }
502 -- trySpontaneousSolve (CFunEqCan ...) = ...
503 -- See Note [No touchables as FunEq RHS] in TcSMonad
504 trySpontaneousSolve _ _ = return Nothing
507 trySpontaneousEqOneWay :: InertSet -> CoVar -> CtFlavor -> TcTyVar -> Xi
508 -> TcS (Maybe SWorkList)
509 -- tv is a MetaTyVar, not untouchable
510 -- Precondition: kind(xi) is a sub-kind of kind(tv)
511 trySpontaneousEqOneWay inerts cv gw tv xi
512 | not (isSigTyVar tv) || isTyVarTy xi
513 = solveWithIdentity inerts cv gw tv xi
518 trySpontaneousEqTwoWay :: InertSet -> CoVar -> CtFlavor -> TcTyVar -> TcTyVar
519 -> TcS (Maybe SWorkList)
520 -- Both tyvars are *touchable* MetaTyvars
521 -- By the CTyEqCan invariant, k2 `isSubKind` k1
522 trySpontaneousEqTwoWay inerts cv gw tv1 tv2
524 , nicer_to_update_tv2 = solveWithIdentity inerts cv gw tv2 (mkTyVarTy tv1)
525 | otherwise = ASSERT( k2 `isSubKind` k1 )
526 solveWithIdentity inerts cv gw tv1 (mkTyVarTy tv2)
530 nicer_to_update_tv2 = isSigTyVar tv1 || isSystemName (Var.varName tv2)
533 Note [Loopy spontaneous solving]
534 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
535 Consider the original wanted:
536 wanted : Maybe (E alpha) ~ alpha
537 where E is a type family, such that E (T x) = x. After canonicalization,
538 as a result of flattening, we will get:
539 given : E alpha ~ fsk
540 wanted : alpha ~ Maybe fsk
541 where (fsk := E alpha, on the side). Now, if we spontaneously *solve*
542 (alpha := Maybe fsk) we are in trouble! Instead, we should refrain from solving
543 it and keep it as wanted. In inference mode we'll end up quantifying over
544 (alpha ~ Maybe (E alpha))
545 Hence, 'solveWithIdentity' performs a small occurs check before
546 actually solving. But this occurs check *must look through* flatten skolems.
548 However, it may be the case that the flatten skolem in hand is equal to some other
549 flatten skolem whith *does not* mention our unification variable. Here's a typical example:
554 After canonicalization:
559 After some reactions:
564 At this point, we will try to spontaneously solve (alpha ~ f2) which remains as yet unsolved.
565 We will look inside f2, which immediately mentions (F alpha), so it's not good to unify! However
566 by looking at the equivalence class of the flatten skolems, we can see that it is fine to
567 unify (alpha ~ f1) which solves our goals!
569 A similar problem happens because of other spontaneous solving. Suppose we have the
570 following wanteds, arriving in this exact order:
571 (first) w: beta ~ alpha
572 (second) w: alpha ~ fsk
573 (third) g: F beta ~ fsk
574 Then, we first spontaneously solve the first constraint, making (beta := alpha), and having
575 (beta ~ alpha) as given. *Then* we encounter the second wanted (alpha ~ fsk). "fsk" does not
576 obviously mention alpha, so naively we can also spontaneously solve (alpha := fsk). But
577 that is wrong since fsk mentions beta, which has already secretly been unified to alpha!
579 To avoid this problem, the same occurs check must unveil rewritings that can happen because
580 of spontaneously having solved other constraints.
583 Note [Avoid double unifications]
584 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
585 The spontaneous solver has to return a given which mentions the unified unification
586 variable *on the left* of the equality. Here is what happens if not:
587 Original wanted: (a ~ alpha), (alpha ~ Int)
588 We spontaneously solve the first wanted, without changing the order!
589 given : a ~ alpha [having unified alpha := a]
590 Now the second wanted comes along, but he cannot rewrite the given, so we simply continue.
591 At the end we spontaneously solve that guy, *reunifying* [alpha := Int]
593 We avoid this problem by orienting the given so that the unification
594 variable is on the left. [Note that alternatively we could attempt to
595 enforce this at canonicalization]
597 See also Note [No touchables as FunEq RHS] in TcSMonad; avoiding
598 double unifications is the main reason we disallow touchable
599 unification variables as RHS of type family equations: F xis ~ alpha.
603 solveWithIdentity :: InertSet
604 -> CoVar -> CtFlavor -> TcTyVar -> Xi
605 -> TcS (Maybe SWorkList)
606 -- Solve with the identity coercion
607 -- Precondition: kind(xi) is a sub-kind of kind(tv)
608 -- Precondition: CtFlavor is Wanted or Derived
609 -- See [New Wanted Superclass Work] to see why solveWithIdentity
610 -- must work for Derived as well as Wanted
611 solveWithIdentity inerts cv gw tv xi
612 = do { tybnds <- getTcSTyBindsMap
613 ; case occurCheck tybnds inerts tv xi of
614 Nothing -> return Nothing
615 Just (xi_unflat,coi) -> solve_with xi_unflat coi }
617 solve_with xi_unflat coi -- coi : xi_unflat ~ xi
618 = do { traceTcS "Sneaky unification:" $
619 vcat [text "Coercion variable: " <+> ppr gw,
620 text "Coercion: " <+> pprEq (mkTyVarTy tv) xi,
621 text "Left Kind is : " <+> ppr (typeKind (mkTyVarTy tv)),
622 text "Right Kind is : " <+> ppr (typeKind xi)
624 ; setWantedTyBind tv xi_unflat -- Set tv := xi_unflat
625 ; cv_given <- newGivOrDerCoVar (mkTyVarTy tv) xi_unflat xi_unflat
626 ; let flav = mkGivenFlavor gw UnkSkol
627 ; (cts, co) <- case coi of
628 ACo co -> do { can_eqs <- canEq flav cv_given (mkTyVarTy tv) xi_unflat
629 ; return (can_eqs, co) }
631 (singleCCan (CTyEqCan { cc_id = cv_given
632 , cc_flavor = mkGivenFlavor gw UnkSkol
633 , cc_tyvar = tv, cc_rhs = xi }
634 -- xi, *not* xi_unflat because
635 -- xi_unflat may require flattening!
638 Wanted {} -> setWantedCoBind cv co
639 Derived {} -> setDerivedCoBind cv co
640 _ -> pprPanic "Can't spontaneously solve *given*" empty
641 -- See Note [Avoid double unifications]
642 ; return (Just cts) }
644 occurCheck :: VarEnv (TcTyVar, TcType) -> InertSet
645 -> TcTyVar -> TcType -> Maybe (TcType,CoercionI)
646 -- Traverse @ty@ to make sure that @tv@ does not appear under some flatten skolem.
647 -- If it appears under some flatten skolem look in that flatten skolem equivalence class
648 -- (see Note [InertSet FlattenSkolemEqClass], [Loopy Spontaneous Solving]) to see if you
649 -- can find a different flatten skolem to use, that is, one that does not mention @tv@.
651 -- Postcondition: Just (ty', coi) = occurCheck binds inerts tv ty
653 -- NB: The returned type ty' may not be flat!
655 occurCheck ty_binds inerts the_tv the_ty
656 = ok emptyVarSet the_ty
658 -- If (fsk `elem` bad) then tv occurs in any rendering
659 -- of the type under the expansion of fsk
660 ok bad this_ty@(TyConApp tc tys)
661 | Just tys_cois <- allMaybes (map (ok bad) tys)
662 , (tys',cois') <- unzip tys_cois
663 = Just (TyConApp tc tys', mkTyConAppCoI tc cois')
664 | isSynTyCon tc, Just ty_expanded <- tcView this_ty
665 = ok bad ty_expanded -- See Note [Type synonyms and the occur check] in TcUnify
667 | Just (sty',coi) <- ok_pred bad sty
668 = Just (PredTy sty', coi)
669 ok bad (FunTy arg res)
670 | Just (arg', coiarg) <- ok bad arg, Just (res', coires) <- ok bad res
671 = Just (FunTy arg' res', mkFunTyCoI coiarg coires)
672 ok bad (AppTy fun arg)
673 | Just (fun', coifun) <- ok bad fun, Just (arg', coiarg) <- ok bad arg
674 = Just (AppTy fun' arg', mkAppTyCoI coifun coiarg)
675 ok bad (ForAllTy tv1 ty1)
676 -- WARNING: What if it is a (t1 ~ t2) => t3? It's not handled properly at the moment.
677 | Just (ty1', coi) <- ok bad ty1
678 = Just (ForAllTy tv1 ty1', mkForAllTyCoI tv1 coi)
681 ok bad this_ty@(TyVarTy tv)
682 | tv == the_tv = Nothing -- Occurs check error
683 | not (isTcTyVar tv) = Just (this_ty, IdCo this_ty) -- Bound var
684 | FlatSkol zty <- tcTyVarDetails tv = ok_fsk bad tv zty
685 | Just (_,ty) <- lookupVarEnv ty_binds tv = ok bad ty
686 | otherwise = Just (this_ty, IdCo this_ty)
688 -- Check if there exists a ty bind already, as a result of sneaky unification.
690 ok _bad _ty = Nothing
693 ok_pred bad (ClassP cn tys)
694 | Just tys_cois <- allMaybes $ map (ok bad) tys
695 = let (tys', cois') = unzip tys_cois
696 in Just (ClassP cn tys', mkClassPPredCoI cn cois')
697 ok_pred bad (IParam nm ty)
698 | Just (ty',co') <- ok bad ty
699 = Just (IParam nm ty', mkIParamPredCoI nm co')
700 ok_pred bad (EqPred ty1 ty2)
701 | Just (ty1',coi1) <- ok bad ty1, Just (ty2',coi2) <- ok bad ty2
702 = Just (EqPred ty1' ty2', mkEqPredCoI coi1 coi2)
703 ok_pred _ _ = Nothing
707 | fsk `elemVarSet` bad
708 -- We are already trying to find a rendering of fsk,
709 -- and to do that it seems we need a rendering, so fail
712 = firstJusts (ok new_bad zty : map (go_under_fsk new_bad) fsk_equivs)
714 fsk_equivs = getFskEqClass inerts fsk
715 new_bad = bad `extendVarSetList` (fsk : map fst fsk_equivs)
718 go_under_fsk bad_tvs (fsk,co)
719 | FlatSkol zty <- tcTyVarDetails fsk
720 = case ok bad_tvs zty of
722 Just (ty,coi') -> Just (ty, mkTransCoI coi' (ACo co))
723 | otherwise = pprPanic "go_down_equiv" (ppr fsk)
727 *********************************************************************************
729 The interact-with-inert Stage
731 *********************************************************************************
734 -- Interaction result of WorkItem <~> AtomicInert
736 = IR { ir_stop :: StopOrContinue
738 -- => Reagent (work item) consumed.
739 -- ContinueWith new_reagent
740 -- => Reagent transformed but keep gathering interactions.
741 -- The transformed item remains inert with respect
742 -- to any previously encountered inerts.
744 , ir_inert_action :: InertAction
745 -- Whether the inert item should remain in the InertSet.
747 , ir_new_work :: WorkList
748 -- new work items to add to the WorkList
751 -- What to do with the inert reactant.
752 data InertAction = KeepInert | DropInert
755 mkIRContinue :: Monad m => WorkItem -> InertAction -> WorkList -> m InteractResult
756 mkIRContinue wi keep newWork = return $ IR (ContinueWith wi) keep newWork
758 mkIRStop :: Monad m => InertAction -> WorkList -> m InteractResult
759 mkIRStop keep newWork = return $ IR Stop keep newWork
761 dischargeWorkItem :: Monad m => m InteractResult
762 dischargeWorkItem = mkIRStop KeepInert emptyCCan
764 noInteraction :: Monad m => WorkItem -> m InteractResult
765 noInteraction workItem = mkIRContinue workItem KeepInert emptyCCan
767 data WhichComesFromInert = LeftComesFromInert | RightComesFromInert
769 ---------------------------------------------------
770 -- Interact a single WorkItem with an InertSet as far as possible, i.e. until we get a Stop
771 -- result from an individual interaction (i.e. when the WorkItem is consumed), or until we've
772 -- interacted the WorkItem with the entire InertSet.
774 -- Postcondition: the new InertSet in the resulting StageResult is subset
775 -- of the input InertSet.
777 interactWithInertsStage :: SimplifierStage
778 interactWithInertsStage workItem inert
779 = foldlInertSetM interactNext initITR inert
781 initITR = SR { sr_inerts = emptyInert
782 , sr_new_work = emptyCCan
783 , sr_stop = ContinueWith workItem }
786 interactNext :: StageResult -> AtomicInert -> TcS StageResult
787 interactNext it inert
788 | ContinueWith workItem <- sr_stop it
789 = do { ir <- interactWithInert inert workItem
790 ; let inerts = sr_inerts it
791 ; return $ SR { sr_inerts = if ir_inert_action ir == KeepInert
792 then inerts `updInertSet` inert
794 , sr_new_work = sr_new_work it `unionWorkLists` ir_new_work ir
795 , sr_stop = ir_stop ir } }
796 | otherwise = return $ itrAddInert inert it
799 itrAddInert :: AtomicInert -> StageResult -> StageResult
800 itrAddInert inert itr = itr { sr_inerts = (sr_inerts itr) `updInertSet` inert }
802 -- Do a single interaction of two constraints.
803 interactWithInert :: AtomicInert -> WorkItem -> TcS InteractResult
804 interactWithInert inert workitem
805 = do { ctxt <- getTcSContext
806 ; let is_allowed = allowedInteraction (simplEqsOnly ctxt) inert workitem
807 inert_ev = cc_id inert
808 work_ev = cc_id workitem
810 -- Never interact a wanted and a derived where the derived's evidence
811 -- mentions the wanted evidence in an unguarded way.
812 -- See Note [Superclasses and recursive dictionaries]
813 -- and Note [New Wanted Superclass Work]
814 -- We don't have to do this for givens, as we fully know the evidence for them.
816 case (cc_flavor inert, cc_flavor workitem) of
817 (Wanted loc, Derived _) -> isGoodRecEv work_ev (WantedEvVar inert_ev loc)
818 (Derived _, Wanted loc) -> isGoodRecEv inert_ev (WantedEvVar work_ev loc)
821 ; if is_allowed && rec_ev_ok then
822 doInteractWithInert inert workitem
824 noInteraction workitem
827 allowedInteraction :: Bool -> AtomicInert -> WorkItem -> Bool
828 -- Allowed interactions
829 allowedInteraction eqs_only (CDictCan {}) (CDictCan {}) = not eqs_only
830 allowedInteraction eqs_only (CIPCan {}) (CIPCan {}) = not eqs_only
831 allowedInteraction _ _ _ = True
833 --------------------------------------------
834 doInteractWithInert :: CanonicalCt -> CanonicalCt -> TcS InteractResult
835 -- Identical class constraints.
838 (CDictCan { cc_id = d1, cc_flavor = fl1, cc_class = cls1, cc_tyargs = tys1 })
839 workItem@(CDictCan { cc_id = d2, cc_flavor = fl2, cc_class = cls2, cc_tyargs = tys2 })
840 | cls1 == cls2 && (and $ zipWith tcEqType tys1 tys2)
841 = solveOneFromTheOther (d1,fl1) workItem
843 | cls1 == cls2 && (not (isGiven fl1 && isGiven fl2))
844 = -- See Note [When improvement happens]
845 do { let work_item_pred_loc = (ClassP cls2 tys2, ppr d2)
846 inert_pred_loc = (ClassP cls1 tys1, ppr d1)
847 loc = combineCtLoc fl1 fl2
848 eqn_pred_locs = improveFromAnother work_item_pred_loc inert_pred_loc
849 ; wevvars <- mkWantedFunDepEqns loc eqn_pred_locs
850 -- See Note [Generating extra equalities]
851 ; workList <- canWanteds wevvars
852 ; mkIRContinue workItem KeepInert workList -- Keep the inert there so we avoid
853 -- re-introducing the fundep equalities
854 -- See Note [FunDep Reactions]
857 -- Class constraint and given equality: use the equality to rewrite
858 -- the class constraint.
859 doInteractWithInert (CTyEqCan { cc_id = cv, cc_flavor = ifl, cc_tyvar = tv, cc_rhs = xi })
860 (CDictCan { cc_id = dv, cc_flavor = wfl, cc_class = cl, cc_tyargs = xis })
861 | ifl `canRewrite` wfl
862 , tv `elemVarSet` tyVarsOfTypes xis
863 -- substitute for tv in xis. Note that the resulting class
864 -- constraint is still canonical, since substituting xi-types in
865 -- xi-types generates xi-types. However, it may no longer be
866 -- inert with respect to the inert set items we've already seen.
867 -- For example, consider the inert set
872 -- and the work item D a (w). D a does not interact with D Int.
873 -- Next, it does interact with a ~g Int, getting rewritten to D
874 -- Int (w). But now we must go back through the rest of the inert
875 -- set again, to find that it can now be discharged by the given D
877 = do { rewritten_dict <- rewriteDict (cv,tv,xi) (dv,wfl,cl,xis)
878 ; mkIRStop KeepInert (singleCCan rewritten_dict) }
880 doInteractWithInert (CDictCan { cc_id = dv, cc_flavor = ifl, cc_class = cl, cc_tyargs = xis })
881 workItem@(CTyEqCan { cc_id = cv, cc_flavor = wfl, cc_tyvar = tv, cc_rhs = xi })
882 | wfl `canRewrite` ifl
883 , tv `elemVarSet` tyVarsOfTypes xis
884 = do { rewritten_dict <- rewriteDict (cv,tv,xi) (dv,ifl,cl,xis)
885 ; mkIRContinue workItem DropInert (singleCCan rewritten_dict) }
887 -- Class constraint and given equality: use the equality to rewrite
888 -- the class constraint.
889 doInteractWithInert (CTyEqCan { cc_id = cv, cc_flavor = ifl, cc_tyvar = tv, cc_rhs = xi })
890 (CIPCan { cc_id = ipid, cc_flavor = wfl, cc_ip_nm = nm, cc_ip_ty = ty })
891 | ifl `canRewrite` wfl
892 , tv `elemVarSet` tyVarsOfType ty
893 = do { rewritten_ip <- rewriteIP (cv,tv,xi) (ipid,wfl,nm,ty)
894 ; mkIRStop KeepInert (singleCCan rewritten_ip) }
896 doInteractWithInert (CIPCan { cc_id = ipid, cc_flavor = ifl, cc_ip_nm = nm, cc_ip_ty = ty })
897 workItem@(CTyEqCan { cc_id = cv, cc_flavor = wfl, cc_tyvar = tv, cc_rhs = xi })
898 | wfl `canRewrite` ifl
899 , tv `elemVarSet` tyVarsOfType ty
900 = do { rewritten_ip <- rewriteIP (cv,tv,xi) (ipid,ifl,nm,ty)
901 ; mkIRContinue workItem DropInert (singleCCan rewritten_ip) }
903 -- Two implicit parameter constraints. If the names are the same,
904 -- but their types are not, we generate a wanted type equality
905 -- that equates the type (this is "improvement").
906 -- However, we don't actually need the coercion evidence,
907 -- so we just generate a fresh coercion variable that isn't used anywhere.
908 doInteractWithInert (CIPCan { cc_id = id1, cc_flavor = ifl, cc_ip_nm = nm1, cc_ip_ty = ty1 })
909 workItem@(CIPCan { cc_flavor = wfl, cc_ip_nm = nm2, cc_ip_ty = ty2 })
910 | nm1 == nm2 && isGiven wfl && isGiven ifl
911 = -- See Note [Overriding implicit parameters]
912 -- Dump the inert item, override totally with the new one
913 -- Do not require type equality
914 mkIRContinue workItem DropInert emptyCCan
916 | nm1 == nm2 && ty1 `tcEqType` ty2
917 = solveOneFromTheOther (id1,ifl) workItem
920 = -- See Note [When improvement happens]
921 do { co_var <- newWantedCoVar ty1 ty2
922 ; let flav = Wanted (combineCtLoc ifl wfl)
923 ; mkCanonical flav co_var >>= mkIRContinue workItem KeepInert }
926 -- Inert: equality, work item: function equality
928 -- Never rewrite a given with a wanted equality, and a type function
929 -- equality can never rewrite an equality. Note also that if we have
930 -- F x1 ~ x2 and a ~ x3, and a occurs in x2, we don't rewrite it. We
931 -- can wait until F x1 ~ x2 matches another F x1 ~ x4, and only then
932 -- we will ``expose'' x2 and x4 to rewriting.
934 -- Otherwise, we can try rewriting the type function equality with the equality.
935 doInteractWithInert (CTyEqCan { cc_id = cv1, cc_flavor = ifl, cc_tyvar = tv, cc_rhs = xi1 })
936 (CFunEqCan { cc_id = cv2, cc_flavor = wfl, cc_fun = tc
937 , cc_tyargs = args, cc_rhs = xi2 })
938 | ifl `canRewrite` wfl
939 , tv `elemVarSet` tyVarsOfTypes args
940 = do { rewritten_funeq <- rewriteFunEq (cv1,tv,xi1) (cv2,wfl,tc,args,xi2)
941 ; mkIRStop KeepInert (singleCCan rewritten_funeq) }
943 -- Inert: function equality, work item: equality
945 doInteractWithInert (CFunEqCan {cc_id = cv1, cc_flavor = ifl, cc_fun = tc
946 , cc_tyargs = args, cc_rhs = xi1 })
947 workItem@(CTyEqCan { cc_id = cv2, cc_flavor = wfl, cc_tyvar = tv, cc_rhs = xi2 })
948 | wfl `canRewrite` ifl
949 , tv `elemVarSet` tyVarsOfTypes args
950 = do { rewritten_funeq <- rewriteFunEq (cv2,tv,xi2) (cv1,ifl,tc,args,xi1)
951 ; mkIRContinue workItem DropInert (singleCCan rewritten_funeq) }
953 doInteractWithInert (CFunEqCan { cc_id = cv1, cc_flavor = fl1, cc_fun = tc1
954 , cc_tyargs = args1, cc_rhs = xi1 })
955 workItem@(CFunEqCan { cc_id = cv2, cc_flavor = fl2, cc_fun = tc2
956 , cc_tyargs = args2, cc_rhs = xi2 })
957 | fl1 `canRewrite` fl2 && lhss_match
958 = do { cans <- rewriteEqLHS LeftComesFromInert (mkCoVarCoercion cv1,xi1) (cv2,fl2,xi2)
959 ; mkIRStop KeepInert cans }
960 | fl2 `canRewrite` fl1 && lhss_match
961 = do { cans <- rewriteEqLHS RightComesFromInert (mkCoVarCoercion cv2,xi2) (cv1,fl1,xi1)
962 ; mkIRContinue workItem DropInert cans }
964 lhss_match = tc1 == tc2 && and (zipWith tcEqType args1 args2)
967 inert@(CTyEqCan { cc_id = cv1, cc_flavor = fl1, cc_tyvar = tv1, cc_rhs = xi1 })
968 workItem@(CTyEqCan { cc_id = cv2, cc_flavor = fl2, cc_tyvar = tv2, cc_rhs = xi2 })
969 -- Check for matching LHS
970 | fl1 `canRewrite` fl2 && tv1 == tv2
971 = do { cans <- rewriteEqLHS LeftComesFromInert (mkCoVarCoercion cv1,xi1) (cv2,fl2,xi2)
972 ; mkIRStop KeepInert cans }
974 | fl2 `canRewrite` fl1 && tv1 == tv2
975 = do { cans <- rewriteEqLHS RightComesFromInert (mkCoVarCoercion cv2,xi2) (cv1,fl1,xi1)
976 ; mkIRContinue workItem DropInert cans }
978 -- Check for rewriting RHS
979 | fl1 `canRewrite` fl2 && tv1 `elemVarSet` tyVarsOfType xi2
980 = do { rewritten_eq <- rewriteEqRHS (cv1,tv1,xi1) (cv2,fl2,tv2,xi2)
981 ; mkIRStop KeepInert rewritten_eq }
982 | fl2 `canRewrite` fl1 && tv2 `elemVarSet` tyVarsOfType xi1
983 = do { rewritten_eq <- rewriteEqRHS (cv2,tv2,xi2) (cv1,fl1,tv1,xi1)
984 ; mkIRContinue workItem DropInert rewritten_eq }
986 -- Finally, if workitem is a Flatten Equivalence Class constraint and the
987 -- inert is a wanted constraint, even when the workitem cannot rewrite the
988 -- inert, drop the inert out because you may have to reconsider solving the
989 -- inert *using* the equivalence class you created. See note [Loopy Spontaneous Solving]
990 -- and [InertSet FlattenSkolemEqClass]
992 | not $ isGiven fl1, -- The inert is wanted or derived
993 isMetaTyVar tv1, -- and has a unification variable lhs
994 FlatSkol {} <- tcTyVarDetails tv2, -- And workitem is a flatten skolem equality
995 Just tv2' <- tcGetTyVar_maybe xi2, FlatSkol {} <- tcTyVarDetails tv2'
996 = mkIRContinue workItem DropInert (singletonWorkList inert)
999 -- Fall-through case for all other situations
1000 doInteractWithInert _ workItem = noInteraction workItem
1002 -------------------------
1003 -- Equational Rewriting
1004 rewriteDict :: (CoVar, TcTyVar, Xi) -> (DictId, CtFlavor, Class, [Xi]) -> TcS CanonicalCt
1005 rewriteDict (cv,tv,xi) (dv,gw,cl,xis)
1006 = do { let cos = substTysWith [tv] [mkCoVarCoercion cv] xis -- xis[tv] ~ xis[xi]
1007 args = substTysWith [tv] [xi] xis
1009 dict_co = mkTyConCoercion con cos
1010 ; dv' <- newDictVar cl args
1012 Wanted {} -> setDictBind dv (EvCast dv' (mkSymCoercion dict_co))
1013 _given_or_derived -> setDictBind dv' (EvCast dv dict_co)
1014 ; return (CDictCan { cc_id = dv'
1017 , cc_tyargs = args }) }
1019 rewriteIP :: (CoVar,TcTyVar,Xi) -> (EvVar,CtFlavor, IPName Name, TcType) -> TcS CanonicalCt
1020 rewriteIP (cv,tv,xi) (ipid,gw,nm,ty)
1021 = do { let ip_co = substTyWith [tv] [mkCoVarCoercion cv] ty -- ty[tv] ~ t[xi]
1022 ty' = substTyWith [tv] [xi] ty
1023 ; ipid' <- newIPVar nm ty'
1025 Wanted {} -> setIPBind ipid (EvCast ipid' (mkSymCoercion ip_co))
1026 _given_or_derived -> setIPBind ipid' (EvCast ipid ip_co)
1027 ; return (CIPCan { cc_id = ipid'
1030 , cc_ip_ty = ty' }) }
1032 rewriteFunEq :: (CoVar,TcTyVar,Xi) -> (CoVar,CtFlavor,TyCon, [Xi], Xi) -> TcS CanonicalCt
1033 rewriteFunEq (cv1,tv,xi1) (cv2,gw, tc,args,xi2)
1034 = do { let arg_cos = substTysWith [tv] [mkCoVarCoercion cv1] args
1035 args' = substTysWith [tv] [xi1] args
1036 fun_co = mkTyConCoercion tc arg_cos
1037 ; cv2' <- case gw of
1038 Wanted {} -> do { cv2' <- newWantedCoVar (mkTyConApp tc args') xi2
1039 ; setWantedCoBind cv2 $
1040 mkTransCoercion fun_co (mkCoVarCoercion cv2')
1042 _giv_or_der -> newGivOrDerCoVar (mkTyConApp tc args') xi2 $
1043 mkTransCoercion (mkSymCoercion fun_co) (mkCoVarCoercion cv2)
1044 ; return (CFunEqCan { cc_id = cv2'
1051 rewriteEqRHS :: (CoVar,TcTyVar,Xi) -> (CoVar,CtFlavor,TcTyVar,Xi) -> TcS CanonicalCts
1052 -- Use the first equality to rewrite the second, flavors already checked.
1053 -- E.g. c1 : tv1 ~ xi1 c2 : tv2 ~ xi2
1054 -- rewrites c2 to give
1055 -- c2' : tv2 ~ xi2[xi1/tv1]
1056 -- We must do an occurs check to sure the new constraint is canonical
1057 -- So we might return an empty bag
1058 rewriteEqRHS (cv1,tv1,xi1) (cv2,gw,tv2,xi2)
1059 | Just tv2' <- tcGetTyVar_maybe xi2'
1060 , tv2 == tv2' -- In this case xi2[xi1/tv1] = tv2, so we have tv2~tv2
1061 = do { when (isWanted gw) (setWantedCoBind cv2 (mkSymCoercion co2'))
1062 ; return emptyCCan }
1067 -> do { cv2' <- newWantedCoVar (mkTyVarTy tv2) xi2'
1068 ; setWantedCoBind cv2 $
1069 mkCoVarCoercion cv2' `mkTransCoercion` mkSymCoercion co2'
1072 -> newGivOrDerCoVar (mkTyVarTy tv2) xi2' $
1073 mkCoVarCoercion cv2 `mkTransCoercion` co2'
1075 ; xi2'' <- canOccursCheck gw tv2 xi2' -- we know xi2' is *not* tv2
1076 ; return (singleCCan $ CTyEqCan { cc_id = cv2'
1082 xi2' = substTyWith [tv1] [xi1] xi2
1083 co2' = substTyWith [tv1] [mkCoVarCoercion cv1] xi2 -- xi2 ~ xi2[xi1/tv1]
1086 rewriteEqLHS :: WhichComesFromInert -> (Coercion,Xi) -> (CoVar,CtFlavor,Xi) -> TcS CanonicalCts
1087 -- Used to ineratct two equalities of the following form:
1088 -- First Equality: co1: (XXX ~ xi1)
1089 -- Second Equality: cv2: (XXX ~ xi2)
1090 -- Where the cv1 `canRewrite` cv2 equality
1091 -- We have an option of creating new work (xi1 ~ xi2) OR (xi2 ~ xi1). This
1092 -- depends on whether the left or the right equality comes from the inert set.
1094 -- prefer to create (xi2 ~ xi1) if the first comes from the inert
1095 -- prefer to create (xi1 ~ xi2) if the second comes from the inert
1096 rewriteEqLHS which (co1,xi1) (cv2,gw,xi2)
1097 = do { cv2' <- case (isWanted gw, which) of
1098 (True,LeftComesFromInert) ->
1099 do { cv2' <- newWantedCoVar xi2 xi1
1100 ; setWantedCoBind cv2 $
1101 co1 `mkTransCoercion` mkSymCoercion (mkCoVarCoercion cv2')
1103 (True,RightComesFromInert) ->
1104 do { cv2' <- newWantedCoVar xi1 xi2
1105 ; setWantedCoBind cv2 $
1106 co1 `mkTransCoercion` mkCoVarCoercion cv2'
1108 (False,LeftComesFromInert) ->
1109 newGivOrDerCoVar xi2 xi1 $
1110 mkSymCoercion (mkCoVarCoercion cv2) `mkTransCoercion` co1
1111 (False,RightComesFromInert) ->
1112 newGivOrDerCoVar xi1 xi2 $
1113 mkSymCoercion co1 `mkTransCoercion` mkCoVarCoercion cv2
1114 ; mkCanonical gw cv2' }
1118 solveOneFromTheOther :: (EvVar, CtFlavor) -> CanonicalCt -> TcS InteractResult
1119 -- First argument inert, second argument workitem. They both represent
1120 -- wanted/given/derived evidence for the *same* predicate so we try here to
1121 -- discharge one directly from the other.
1123 -- Precondition: value evidence only (implicit parameters, classes)
1125 solveOneFromTheOther (iid,ifl) workItem
1126 -- Both derived needs a special case. You might think that we do not need
1127 -- two evidence terms for the same claim. But, since the evidence is partial,
1128 -- either evidence may do in some cases; see TcSMonad.isGoodRecEv.
1129 -- See also Example 3 in Note [Superclasses and recursive dictionaries]
1130 | isDerived ifl && isDerived wfl
1131 = noInteraction workItem
1133 | ifl `canRewrite` wfl
1134 = do { unless (isGiven wfl) $ setEvBind wid (EvId iid)
1135 -- Overwrite the binding, if one exists
1136 -- For Givens, which are lambda-bound, nothing to overwrite,
1137 ; dischargeWorkItem }
1139 | otherwise -- wfl `canRewrite` ifl
1140 = do { unless (isGiven ifl) $ setEvBind iid (EvId wid)
1141 ; mkIRContinue workItem DropInert emptyCCan }
1144 wfl = cc_flavor workItem
1145 wid = cc_id workItem
1148 Note [Superclasses and recursive dictionaries]
1149 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1150 Overlaps with Note [SUPERCLASS-LOOP 1]
1151 Note [SUPERCLASS-LOOP 2]
1152 Note [Recursive instances and superclases]
1153 ToDo: check overlap and delete redundant stuff
1155 Right before adding a given into the inert set, we must
1156 produce some more work, that will bring the superclasses
1157 of the given into scope. The superclass constraints go into
1160 When we simplify a wanted constraint, if we first see a matching
1161 instance, we may produce new wanted work. To (1) avoid doing this work
1162 twice in the future and (2) to handle recursive dictionaries we may ``cache''
1163 this item as solved (in effect, given) into our inert set and with that add
1164 its superclass constraints (as given) in our worklist.
1166 But now we have added partially solved constraints to the worklist which may
1167 interact with other wanteds. Consider the example:
1171 class Eq b => Foo a b --- 0-th selector
1172 instance Eq a => Foo [a] a --- fooDFun
1174 and wanted (Foo [t] t). We are first going to see that the instance matches
1175 and create an inert set that includes the solved (Foo [t] t) and its
1177 d1 :_g Foo [t] t d1 := EvDFunApp fooDFun d3
1178 d2 :_g Eq t d2 := EvSuperClass d1 0
1179 Our work list is going to contain a new *wanted* goal
1181 It is wrong to react the wanted (Eq t) with the given (Eq t) because that would
1182 construct loopy evidence. Hence the check isGoodRecEv in doInteractWithInert.
1184 OK, so we have ruled out bad behaviour, but how do we ge recursive dictionaries,
1189 data D r = ZeroD | SuccD (r (D r));
1191 instance (Eq (r (D r))) => Eq (D r) where
1192 ZeroD == ZeroD = True
1193 (SuccD a) == (SuccD b) = a == b
1196 equalDC :: D [] -> D [] -> Bool;
1199 We need to prove (Eq (D [])). Here's how we go:
1203 by instance decl, holds if
1207 *BUT* we have an inert set which gives us (no superclasses):
1209 By the instance declaration of Eq we can show the 'd2' goal if
1211 where d2 = dfEqList d3
1213 Now, however this wanted can interact with our inert d1 to set:
1215 and solve the goal. Why was this interaction OK? Because, if we chase the
1216 evidence of d1 ~~> dfEqD d2 ~~-> dfEqList d3, so by setting d3 := d1 we
1218 d3 := dfEqD2 (dfEqList d3)
1219 which is FINE because the use of d3 is protected by the instance function
1222 So, our strategy is to try to put solved wanted dictionaries into the
1223 inert set along with their superclasses (when this is meaningful,
1224 i.e. when new wanted goals are generated) but solve a wanted dictionary
1225 from a given only in the case where the evidence variable of the
1226 wanted is mentioned in the evidence of the given (recursively through
1227 the evidence binds) in a protected way: more instance function applications
1228 than superclass selectors.
1230 Here are some more examples from GHC's previous type checker
1234 This code arises in the context of "Scrap Your Boilerplate with Class"
1238 instance Sat (ctx Char) => Data ctx Char -- dfunData1
1239 instance (Sat (ctx [a]), Data ctx a) => Data ctx [a] -- dfunData2
1241 class Data Maybe a => Foo a
1243 instance Foo t => Sat (Maybe t) -- dfunSat
1245 instance Data Maybe a => Foo a -- dfunFoo1
1246 instance Foo a => Foo [a] -- dfunFoo2
1247 instance Foo [Char] -- dfunFoo3
1249 Consider generating the superclasses of the instance declaration
1250 instance Foo a => Foo [a]
1252 So our problem is this
1254 d1 :_w Data Maybe [t]
1256 We may add the given in the inert set, along with its superclasses
1257 [assuming we don't fail because there is a matching instance, see
1258 tryTopReact, given case ]
1262 d01 :_g Data Maybe t -- d2 := EvDictSuperClass d0 0
1263 d1 :_w Data Maybe [t]
1264 Then d2 can readily enter the inert, and we also do solving of the wanted
1267 d1 :_s Data Maybe [t] d1 := dfunData2 d2 d3
1269 d2 :_w Sat (Maybe [t])
1271 d01 :_g Data Maybe t
1272 Now, we may simplify d2 more:
1275 d1 :_s Data Maybe [t] d1 := dfunData2 d2 d3
1276 d1 :_g Data Maybe [t]
1277 d2 :_g Sat (Maybe [t]) d2 := dfunSat d4
1281 d01 :_g Data Maybe t
1283 Now, we can just solve d3.
1286 d1 :_s Data Maybe [t] d1 := dfunData2 d2 d3
1287 d2 :_g Sat (Maybe [t]) d2 := dfunSat d4
1290 d01 :_g Data Maybe t
1291 And now we can simplify d4 again, but since it has superclasses we *add* them to the worklist:
1294 d1 :_s Data Maybe [t] d1 := dfunData2 d2 d3
1295 d2 :_g Sat (Maybe [t]) d2 := dfunSat d4
1296 d4 :_g Foo [t] d4 := dfunFoo2 d5
1299 d6 :_g Data Maybe [t] d6 := EvDictSuperClass d4 0
1300 d01 :_g Data Maybe t
1301 Now, d5 can be solved! (and its superclass enter scope)
1304 d1 :_s Data Maybe [t] d1 := dfunData2 d2 d3
1305 d2 :_g Sat (Maybe [t]) d2 := dfunSat d4
1306 d4 :_g Foo [t] d4 := dfunFoo2 d5
1307 d5 :_g Foo t d5 := dfunFoo1 d7
1310 d6 :_g Data Maybe [t]
1311 d8 :_g Data Maybe t d8 := EvDictSuperClass d5 0
1312 d01 :_g Data Maybe t
1315 [1] Suppose we pick d8 and we react him with d01. Which of the two givens should
1316 we keep? Well, we *MUST NOT* drop d01 because d8 contains recursive evidence
1317 that must not be used (look at case interactInert where both inert and workitem
1318 are givens). So we have several options:
1319 - Drop the workitem always (this will drop d8)
1320 This feels very unsafe -- what if the work item was the "good" one
1321 that should be used later to solve another wanted?
1322 - Don't drop anyone: the inert set may contain multiple givens!
1323 [This is currently implemented]
1325 The "don't drop anyone" seems the most safe thing to do, so now we come to problem 2:
1326 [2] We have added both d6 and d01 in the inert set, and we are interacting our wanted
1327 d7. Now the [isRecDictEv] function in the ineration solver
1328 [case inert-given workitem-wanted] will prevent us from interacting d7 := d8
1329 precisely because chasing the evidence of d8 leads us to an unguarded use of d7.
1331 So, no interaction happens there. Then we meet d01 and there is no recursion
1332 problem there [isRectDictEv] gives us the OK to interact and we do solve d7 := d01!
1334 Note [SUPERCLASS-LOOP 1]
1335 ~~~~~~~~~~~~~~~~~~~~~~~~
1336 We have to be very, very careful when generating superclasses, lest we
1337 accidentally build a loop. Here's an example:
1341 class S a => C a where { opc :: a -> a }
1342 class S b => D b where { opd :: b -> b }
1344 instance C Int where
1347 instance D Int where
1350 From (instance C Int) we get the constraint set {ds1:S Int, dd:D Int}
1351 Simplifying, we may well get:
1352 $dfCInt = :C ds1 (opd dd)
1355 Notice that we spot that we can extract ds1 from dd.
1357 Alas! Alack! We can do the same for (instance D Int):
1359 $dfDInt = :D ds2 (opc dc)
1363 And now we've defined the superclass in terms of itself.
1364 Two more nasty cases are in
1369 - Satisfy the superclass context *all by itself*
1370 (tcSimplifySuperClasses)
1371 - And do so completely; i.e. no left-over constraints
1372 to mix with the constraints arising from method declarations
1375 Note [SUPERCLASS-LOOP 2]
1376 ~~~~~~~~~~~~~~~~~~~~~~~~
1377 We need to be careful when adding "the constaint we are trying to prove".
1378 Suppose we are *given* d1:Ord a, and want to deduce (d2:C [a]) where
1380 class Ord a => C a where
1381 instance Ord [a] => C [a] where ...
1383 Then we'll use the instance decl to deduce C [a] from Ord [a], and then add the
1384 superclasses of C [a] to avails. But we must not overwrite the binding
1385 for Ord [a] (which is obtained from Ord a) with a superclass selection or we'll just
1388 Here's another variant, immortalised in tcrun020
1389 class Monad m => C1 m
1390 class C1 m => C2 m x
1391 instance C2 Maybe Bool
1392 For the instance decl we need to build (C1 Maybe), and it's no good if
1393 we run around and add (C2 Maybe Bool) and its superclasses to the avails
1394 before we search for C1 Maybe.
1396 Here's another example
1397 class Eq b => Foo a b
1398 instance Eq a => Foo [a] a
1402 we'll first deduce that it holds (via the instance decl). We must not
1403 then overwrite the Eq t constraint with a superclass selection!
1405 At first I had a gross hack, whereby I simply did not add superclass constraints
1406 in addWanted, though I did for addGiven and addIrred. This was sub-optimal,
1407 becuase it lost legitimate superclass sharing, and it still didn't do the job:
1408 I found a very obscure program (now tcrun021) in which improvement meant the
1409 simplifier got two bites a the cherry... so something seemed to be an Stop
1410 first time, but reducible next time.
1412 Now we implement the Right Solution, which is to check for loops directly
1413 when adding superclasses. It's a bit like the occurs check in unification.
1415 Note [Recursive instances and superclases]
1416 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1417 Consider this code, which arises in the context of "Scrap Your
1418 Boilerplate with Class".
1422 instance Sat (ctx Char) => Data ctx Char
1423 instance (Sat (ctx [a]), Data ctx a) => Data ctx [a]
1425 class Data Maybe a => Foo a
1427 instance Foo t => Sat (Maybe t)
1429 instance Data Maybe a => Foo a
1430 instance Foo a => Foo [a]
1433 In the instance for Foo [a], when generating evidence for the superclasses
1434 (ie in tcSimplifySuperClasses) we need a superclass (Data Maybe [a]).
1435 Using the instance for Data, we therefore need
1436 (Sat (Maybe [a], Data Maybe a)
1437 But we are given (Foo a), and hence its superclass (Data Maybe a).
1438 So that leaves (Sat (Maybe [a])). Using the instance for Sat means
1439 we need (Foo [a]). And that is the very dictionary we are bulding
1440 an instance for! So we must put that in the "givens". So in this
1442 Given: Foo a, Foo [a]
1443 Wanted: Data Maybe [a]
1445 BUT we must *not not not* put the *superclasses* of (Foo [a]) in
1446 the givens, which is what 'addGiven' would normally do. Why? Because
1447 (Data Maybe [a]) is the superclass, so we'd "satisfy" the wanted
1448 by selecting a superclass from Foo [a], which simply makes a loop.
1450 On the other hand we *must* put the superclasses of (Foo a) in
1451 the givens, as you can see from the derivation described above.
1453 Conclusion: in the very special case of tcSimplifySuperClasses
1454 we have one 'given' (namely the "this" dictionary) whose superclasses
1455 must not be added to 'givens' by addGiven.
1457 There is a complication though. Suppose there are equalities
1458 instance (Eq a, a~b) => Num (a,b)
1459 Then we normalise the 'givens' wrt the equalities, so the original
1460 given "this" dictionary is cast to one of a different type. So it's a
1461 bit trickier than before to identify the "special" dictionary whose
1462 superclasses must not be added. See test
1463 indexed-types/should_run/EqInInstance
1465 We need a persistent property of the dictionary to record this
1466 special-ness. Current I'm using the InstLocOrigin (a bit of a hack,
1467 but cool), which is maintained by dictionary normalisation.
1468 Specifically, the InstLocOrigin is
1470 then the no-superclass thing kicks in. WATCH OUT if you fiddle
1473 Note [MATCHING-SYNONYMS]
1474 ~~~~~~~~~~~~~~~~~~~~~~~~
1475 When trying to match a dictionary (D tau) to a top-level instance, or a
1476 type family equation (F taus_1 ~ tau_2) to a top-level family instance,
1477 we do *not* need to expand type synonyms because the matcher will do that for us.
1480 Note [RHS-FAMILY-SYNONYMS]
1481 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1482 The RHS of a family instance is represented as yet another constructor which is
1483 like a type synonym for the real RHS the programmer declared. Eg:
1484 type instance F (a,a) = [a]
1486 :R32 a = [a] -- internal type synonym introduced
1487 F (a,a) ~ :R32 a -- instance
1489 When we react a family instance with a type family equation in the work list
1490 we keep the synonym-using RHS without expansion.
1493 *********************************************************************************
1495 The top-reaction Stage
1497 *********************************************************************************
1500 -- If a work item has any form of interaction with top-level we get this
1501 data TopInteractResult
1502 = NoTopInt -- No top-level interaction
1504 { tir_new_work :: WorkList -- Sub-goals or new work (could be given,
1505 -- for superclasses)
1506 , tir_new_inert :: StopOrContinue -- The input work item, ready to become *inert* now:
1507 } -- NB: in ``given'' (solved) form if the
1508 -- original was wanted or given and instance match
1509 -- was found, but may also be in wanted form if we
1510 -- only reacted with functional dependencies
1511 -- arising from top-level instances.
1513 topReactionsStage :: SimplifierStage
1514 topReactionsStage workItem inerts
1515 = do { tir <- tryTopReact workItem
1518 return $ SR { sr_inerts = inerts
1519 , sr_new_work = emptyWorkList
1520 , sr_stop = ContinueWith workItem }
1521 SomeTopInt tir_new_work tir_new_inert ->
1522 return $ SR { sr_inerts = inerts
1523 , sr_new_work = tir_new_work
1524 , sr_stop = tir_new_inert
1528 tryTopReact :: WorkItem -> TcS TopInteractResult
1529 tryTopReact workitem
1530 = do { -- A flag controls the amount of interaction allowed
1531 -- See Note [Simplifying RULE lhs constraints]
1532 ctxt <- getTcSContext
1533 ; if allowedTopReaction (simplEqsOnly ctxt) workitem
1534 then do { traceTcS "tryTopReact / calling doTopReact" (ppr workitem)
1535 ; doTopReact workitem }
1536 else return NoTopInt
1539 allowedTopReaction :: Bool -> WorkItem -> Bool
1540 allowedTopReaction eqs_only (CDictCan {}) = not eqs_only
1541 allowedTopReaction _ _ = True
1544 doTopReact :: WorkItem -> TcS TopInteractResult
1545 -- The work item does not react with the inert set,
1546 -- so try interaction with top-level instances
1547 doTopReact workItem@(CDictCan { cc_id = dv, cc_flavor = Wanted loc
1548 , cc_class = cls, cc_tyargs = xis })
1549 = do { -- See Note [MATCHING-SYNONYMS]
1550 ; lkp_inst_res <- matchClassInst cls xis loc
1551 ; case lkp_inst_res of
1552 NoInstance -> do { traceTcS "doTopReact/ no class instance for" (ppr dv)
1554 GenInst wtvs ev_term -> -- Solved
1555 -- No need to do fundeps stuff here; the instance
1556 -- matches already so we won't get any more info
1557 -- from functional dependencies
1558 do { traceTcS "doTopReact/ found class instance for" (ppr dv)
1559 ; setDictBind dv ev_term
1560 ; workList <- canWanteds wtvs
1562 -- Solved in one step and no new wanted work produced.
1563 -- i.e we directly matched a top-level instance
1564 -- No point in caching this in 'inert', nor in adding superclasses
1565 then return $ SomeTopInt { tir_new_work = emptyCCan
1566 , tir_new_inert = Stop }
1568 -- Solved and new wanted work produced, you may cache the
1569 -- (tentatively solved) dictionary as Derived and its superclasses
1570 else do { let solved = makeSolved workItem
1571 ; sc_work <- newSCWorkFromFlavored dv (Derived loc) cls xis
1572 ; return $ SomeTopInt
1573 { tir_new_work = workList `unionWorkLists` sc_work
1574 , tir_new_inert = ContinueWith solved } }
1578 -- Try for a fundep reaction beween the wanted item
1579 -- and a top-level instance declaration
1581 = do { instEnvs <- getInstEnvs
1582 ; let eqn_pred_locs = improveFromInstEnv (classInstances instEnvs)
1583 (ClassP cls xis, ppr dv)
1584 ; wevvars <- mkWantedFunDepEqns loc eqn_pred_locs
1585 -- NB: fundeps generate some wanted equalities, but
1586 -- we don't use their evidence for anything
1587 ; fd_work <- canWanteds wevvars
1588 ; sc_work <- newSCWorkFromFlavored dv (Derived loc) cls xis
1589 ; return $ SomeTopInt { tir_new_work = fd_work `unionWorkLists` sc_work
1590 , tir_new_inert = ContinueWith workItem }
1591 -- NB: workItem is inert, but it isn't solved
1592 -- keep it as inert, although it's not solved because we
1593 -- have now reacted all its top-level fundep-induced equalities!
1595 -- See Note [FunDep Reactions]
1598 -- Otherwise, we have a given or derived
1599 doTopReact workItem@(CDictCan { cc_id = dv, cc_flavor = fl
1600 , cc_class = cls, cc_tyargs = xis })
1601 = do { sc_work <- newSCWorkFromFlavored dv fl cls xis
1602 ; return $ SomeTopInt sc_work (ContinueWith workItem) }
1603 -- See Note [Given constraint that matches an instance declaration]
1606 doTopReact (CFunEqCan { cc_id = cv, cc_flavor = fl
1607 , cc_fun = tc, cc_tyargs = args, cc_rhs = xi })
1608 = ASSERT (isSynFamilyTyCon tc) -- No associated data families have reached that far
1609 do { match_res <- matchFam tc args -- See Note [MATCHING-SYNONYMS]
1613 MatchInstSingle (rep_tc, rep_tys)
1614 -> do { let Just coe_tc = tyConFamilyCoercion_maybe rep_tc
1615 Just rhs_ty = tcView (mkTyConApp rep_tc rep_tys)
1616 -- Eagerly expand away the type synonym on the
1617 -- RHS of a type function, so that it never
1618 -- appears in an error message
1619 -- See Note [Type synonym families] in TyCon
1620 coe = mkTyConApp coe_tc rep_tys
1622 Wanted {} -> do { cv' <- newWantedCoVar rhs_ty xi
1623 ; setWantedCoBind cv $
1624 coe `mkTransCoercion`
1627 _ -> newGivOrDerCoVar xi rhs_ty $
1628 mkSymCoercion (mkCoVarCoercion cv) `mkTransCoercion` coe
1630 ; workList <- mkCanonical fl cv'
1631 ; return $ SomeTopInt workList Stop }
1633 -> panicTcS $ text "TcSMonad.matchFam returned multiple instances!"
1637 -- Any other work item does not react with any top-level equations
1638 doTopReact _workItem = return NoTopInt
1641 Note [FunDep and implicit parameter reactions]
1642 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1643 Currently, our story of interacting two dictionaries (or a dictionary
1644 and top-level instances) for functional dependencies, and implicit
1645 paramters, is that we simply produce new wanted equalities. So for example
1647 class D a b | a -> b where ...
1653 We generate the extra work item
1655 where 'cv' is currently unused. However, this new item reacts with d2,
1656 discharging it in favour of a new constraint d2' thus:
1658 d2 := d2' |> D Int cv
1659 Now d2' can be discharged from d1
1661 We could be more aggressive and try to *immediately* solve the dictionary
1662 using those extra equalities. With the same inert set and work item we
1663 might dischard d2 directly:
1666 d2 := d1 |> D Int cv
1668 But in general it's a bit painful to figure out the necessary coercion,
1669 so we just take the first approach.
1671 It's exactly the same with implicit parameters, except that the
1672 "aggressive" approach would be much easier to implement.
1674 Note [When improvement happens]
1675 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1676 We fire an improvement rule when
1678 * Two constraints match (modulo the fundep)
1679 e.g. C t1 t2, C t1 t3 where C a b | a->b
1680 The two match because the first arg is identical
1682 * At least one is not Given. If they are both given, we don't fire
1683 the reaction because we have no way of constructing evidence for a
1684 new equality nor does it seem right to create a new wanted goal
1685 (because the goal will most likely contain untouchables, which
1686 can't be solved anyway)!
1688 Note that we *do* fire the improvement if one is Given and one is Derived.
1689 The latter can be a superclass of a wanted goal. Example (tcfail138)
1690 class L a b | a -> b
1691 class (G a, L a b) => C a b
1693 instance C a b' => G (Maybe a)
1694 instance C a b => C (Maybe a) a
1695 instance L (Maybe a) a
1697 When solving the superclasses of the (C (Maybe a) a) instance, we get
1698 Given: C a b ... and hance by superclasses, (G a, L a b)
1700 Use the instance decl to get
1702 The (C a b') is inert, so we generate its Derived superclasses (L a b'),
1703 and now we need improvement between that derived superclass an the Given (L a b)
1705 Note [Overriding implicit parameters]
1706 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1708 f :: (?x::a) -> Bool -> a
1710 g v = let ?x::Int = 3
1711 in (f v, let ?x::Bool = True in f v)
1713 This should probably be well typed, with
1714 g :: Bool -> (Int, Bool)
1716 So the inner binding for ?x::Bool *overrides* the outer one.
1717 Hence a work-item Given overrides an inert-item Given.
1719 Note [Given constraint that matches an instance declaration]
1720 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1721 What should we do when we discover that one (or more) top-level
1722 instances match a given (or solved) class constraint? We have
1725 1. Reject the program. The reason is that there may not be a unique
1726 best strategy for the solver. Example, from the OutsideIn(X) paper:
1727 instance P x => Q [x]
1728 instance (x ~ y) => R [x] y
1730 wob :: forall a b. (Q [b], R b a) => a -> Int
1732 g :: forall a. Q [a] => [a] -> Int
1735 will generate the impliation constraint:
1736 Q [a] => (Q [beta], R beta [a])
1737 If we react (Q [beta]) with its top-level axiom, we end up with a
1738 (P beta), which we have no way of discharging. On the other hand,
1739 if we react R beta [a] with the top-level we get (beta ~ a), which
1740 is solvable and can help us rewrite (Q [beta]) to (Q [a]) which is
1741 now solvable by the given Q [a].
1743 However, this option is restrictive, for instance [Example 3] from
1744 Note [Recursive dictionaries] will fail to work.
1746 2. Ignore the problem, hoping that the situations where there exist indeed
1747 such multiple strategies are rare: Indeed the cause of the previous
1748 problem is that (R [x] y) yields the new work (x ~ y) which can be
1749 *spontaneously* solved, not using the givens.
1751 We are choosing option 2 below but we might consider having a flag as well.
1754 Note [New Wanted Superclass Work]
1755 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1756 Even in the case of wanted constraints, we add all of its superclasses as
1757 new given work. There are several reasons for this:
1758 a) to minimise error messages;
1759 eg suppose we have wanted (Eq a, Ord a)
1760 then we report only (Ord a) unsoluble
1762 b) to make the smallest number of constraints when *inferring* a type
1763 (same Eq/Ord example)
1765 c) for recursive dictionaries we *must* add the superclasses
1766 so that we can use them when solving a sub-problem
1768 d) To allow FD-like improvement for type families. Assume that
1770 class C a b | a -> b
1771 and we have to solve the implication constraint:
1773 Then, FD improvement can help us to produce a new wanted (beta ~ b)
1775 We want to have the same effect with the type family encoding of
1776 functional dependencies. Namely, consider:
1777 class (F a ~ b) => C a b
1778 Now suppose that we have:
1781 By interacting the given we will get given (F a ~ b) which is not
1782 enough by itself to make us discharge (C a beta). However, we
1783 may create a new derived equality from the super-class of the
1784 wanted constraint (C a beta), namely derived (F a ~ beta).
1785 Now we may interact this with given (F a ~ b) to get:
1787 But 'beta' is a touchable unification variable, and hence OK to
1788 unify it with 'b', replacing the derived evidence with the identity.
1790 This requires trySpontaneousSolve to solve *derived*
1791 equalities that have a touchable in their RHS, *in addition*
1792 to solving wanted equalities.
1794 Here is another example where this is useful.
1798 class (F a ~ b) => C a b
1799 And we are given the wanteds:
1803 We surely do *not* want to quantify over (b ~ c), since if someone provides
1804 dictionaries for (C a b) and (C a c), these dictionaries can provide a proof
1805 of (b ~ c), hence no extra evidence is necessary. Here is what will happen:
1807 Step 1: We will get new *given* superclass work,
1808 provisionally to our solving of w1 and w2
1810 g1: F a ~ b, g2 : F a ~ c,
1811 w1 : C a b, w2 : C a c, w3 : b ~ c
1813 The evidence for g1 and g2 is a superclass evidence term:
1815 g1 := sc w1, g2 := sc w2
1817 Step 2: The givens will solve the wanted w3, so that
1818 w3 := sym (sc w1) ; sc w2
1820 Step 3: Now, one may naively assume that then w2 can be solve from w1
1821 after rewriting with the (now solved equality) (b ~ c).
1823 But this rewriting is ruled out by the isGoodRectDict!
1825 Conclusion, we will (correctly) end up with the unsolved goals
1828 NB: The desugarer needs be more clever to deal with equalities
1829 that participate in recursive dictionary bindings.
1832 newSCWorkFromFlavored :: EvVar -> CtFlavor -> Class -> [Xi]
1834 newSCWorkFromFlavored ev flavor cls xis
1835 | Given loc <- flavor -- The NoScSkol says "don't add superclasses"
1836 , NoScSkol <- ctLocOrigin loc -- Very important!
1837 = return emptyWorkList
1840 = do { let (tyvars, sc_theta, _, _) = classBigSig cls
1841 sc_theta1 = substTheta (zipTopTvSubst tyvars xis) sc_theta
1842 -- Add *all* its superclasses (equalities or not) as new given work
1843 -- See Note [New Wanted Superclass Work]
1844 ; sc_vars <- zipWithM inst_one sc_theta1 [0..]
1845 ; mkCanonicals flavor sc_vars }
1847 inst_one pred n = newGivOrDerEvVar pred (EvSuperClass ev n)
1849 data LookupInstResult
1851 | GenInst [WantedEvVar] EvTerm
1853 matchClassInst :: Class -> [Type] -> WantedLoc -> TcS LookupInstResult
1854 matchClassInst clas tys loc
1855 = do { let pred = mkClassPred clas tys
1856 ; mb_result <- matchClass clas tys
1858 MatchInstNo -> return NoInstance
1859 MatchInstMany -> return NoInstance -- defer any reactions of a multitude until
1860 -- we learn more about the reagent
1861 MatchInstSingle (dfun_id, mb_inst_tys) ->
1862 do { checkWellStagedDFun pred dfun_id loc
1864 -- It's possible that not all the tyvars are in
1865 -- the substitution, tenv. For example:
1866 -- instance C X a => D X where ...
1867 -- (presumably there's a functional dependency in class C)
1868 -- Hence mb_inst_tys :: Either TyVar TcType
1870 ; tys <- instDFunTypes mb_inst_tys
1871 ; let (theta, _) = tcSplitPhiTy (applyTys (idType dfun_id) tys)
1872 ; if null theta then
1873 return (GenInst [] (EvDFunApp dfun_id tys []))
1875 { ev_vars <- instDFunConstraints theta
1876 ; let wevs = [WantedEvVar w loc | w <- ev_vars]
1877 ; return $ GenInst wevs (EvDFunApp dfun_id tys ev_vars) }