2 % (c) The University of Glasgow 2006
3 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
10 tcSimplifyInfer, tcSimplifyInferCheck,
11 tcSimplifyCheck, tcSimplifyRestricted,
12 tcSimplifyRuleLhs, tcSimplifyIPs,
13 tcSimplifySuperClasses,
14 tcSimplifyTop, tcSimplifyInteractive,
15 tcSimplifyBracket, tcSimplifyCheckPat,
17 tcSimplifyDeriv, tcSimplifyDefault,
18 bindInstsOfLocalFuns, bindIrreds,
21 #include "HsVersions.h"
23 import {-# SOURCE #-} TcUnify( unifyType )
59 %************************************************************************
63 %************************************************************************
65 --------------------------------------
66 Notes on functional dependencies (a bug)
67 --------------------------------------
74 instance D a b => C a b -- Undecidable
75 -- (Not sure if it's crucial to this eg)
76 f :: C a b => a -> Bool
79 g :: C a b => a -> Bool
82 Here f typechecks, but g does not!! Reason: before doing improvement,
83 we reduce the (C a b1) constraint from the call of f to (D a b1).
85 Here is a more complicated example:
87 | > class Foo a b | a->b
89 | > class Bar a b | a->b
93 | > instance Bar Obj Obj
95 | > instance (Bar a b) => Foo a b
97 | > foo:: (Foo a b) => a -> String
100 | > runFoo:: (forall a b. (Foo a b) => a -> w) -> w
106 | Could not deduce (Bar a b) from the context (Foo a b)
107 | arising from use of `foo' at <interactive>:1
109 | Add (Bar a b) to the expected type of an expression
110 | In the first argument of `runFoo', namely `foo'
111 | In the definition of `it': it = runFoo foo
113 | Why all of the sudden does GHC need the constraint Bar a b? The
114 | function foo didn't ask for that...
116 The trouble is that to type (runFoo foo), GHC has to solve the problem:
118 Given constraint Foo a b
119 Solve constraint Foo a b'
121 Notice that b and b' aren't the same. To solve this, just do
122 improvement and then they are the same. But GHC currently does
127 That is usually fine, but it isn't here, because it sees that Foo a b is
128 not the same as Foo a b', and so instead applies the instance decl for
129 instance Bar a b => Foo a b. And that's where the Bar constraint comes
132 The Right Thing is to improve whenever the constraint set changes at
133 all. Not hard in principle, but it'll take a bit of fiddling to do.
135 Note [Choosing which variables to quantify]
136 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
137 Suppose we are about to do a generalisation step. We have in our hand
140 T the type of the RHS
141 C the constraints from that RHS
143 The game is to figure out
145 Q the set of type variables over which to quantify
146 Ct the constraints we will *not* quantify over
147 Cq the constraints we will quantify over
149 So we're going to infer the type
153 and float the constraints Ct further outwards.
155 Here are the things that *must* be true:
157 (A) Q intersect fv(G) = EMPTY limits how big Q can be
158 (B) Q superset fv(Cq union T) \ oclose(fv(G),C) limits how small Q can be
160 (A) says we can't quantify over a variable that's free in the environment.
161 (B) says we must quantify over all the truly free variables in T, else
162 we won't get a sufficiently general type.
164 We do not *need* to quantify over any variable that is fixed by the
165 free vars of the environment G.
167 BETWEEN THESE TWO BOUNDS, ANY Q WILL DO!
169 Example: class H x y | x->y where ...
171 fv(G) = {a} C = {H a b, H c d}
174 (A) Q intersect {a} is empty
175 (B) Q superset {a,b,c,d} \ oclose({a}, C) = {a,b,c,d} \ {a,b} = {c,d}
177 So Q can be {c,d}, {b,c,d}
179 In particular, it's perfectly OK to quantify over more type variables
180 than strictly necessary; there is no need to quantify over 'b', since
181 it is determined by 'a' which is free in the envt, but it's perfectly
182 OK to do so. However we must not quantify over 'a' itself.
184 Other things being equal, however, we'd like to quantify over as few
185 variables as possible: smaller types, fewer type applications, more
186 constraints can get into Ct instead of Cq. Here's a good way to
189 Q = grow( fv(T), C ) \ oclose( fv(G), C )
191 That is, quantify over all variable that that MIGHT be fixed by the
192 call site (which influences T), but which aren't DEFINITELY fixed by
193 G. This choice definitely quantifies over enough type variables,
194 albeit perhaps too many.
196 Why grow( fv(T), C ) rather than fv(T)? Consider
198 class H x y | x->y where ...
203 If we used fv(T) = {c} we'd get the type
205 forall c. H c d => c -> b
207 And then if the fn was called at several different c's, each of
208 which fixed d differently, we'd get a unification error, because
209 d isn't quantified. Solution: quantify d. So we must quantify
210 everything that might be influenced by c.
212 Why not oclose( fv(T), C )? Because we might not be able to see
213 all the functional dependencies yet:
215 class H x y | x->y where ...
216 instance H x y => Eq (T x y) where ...
221 Now oclose(fv(T),C) = {c}, because the functional dependency isn't
222 apparent yet, and that's wrong. We must really quantify over d too.
224 There really isn't any point in quantifying over any more than
225 grow( fv(T), C ), because the call sites can't possibly influence
226 any other type variables.
230 -------------------------------------
232 -------------------------------------
234 It's very hard to be certain when a type is ambiguous. Consider
238 instance H x y => K (x,y)
240 Is this type ambiguous?
241 forall a b. (K (a,b), Eq b) => a -> a
243 Looks like it! But if we simplify (K (a,b)) we get (H a b) and
244 now we see that a fixes b. So we can't tell about ambiguity for sure
245 without doing a full simplification. And even that isn't possible if
246 the context has some free vars that may get unified. Urgle!
248 Here's another example: is this ambiguous?
249 forall a b. Eq (T b) => a -> a
250 Not if there's an insance decl (with no context)
251 instance Eq (T b) where ...
253 You may say of this example that we should use the instance decl right
254 away, but you can't always do that:
256 class J a b where ...
257 instance J Int b where ...
259 f :: forall a b. J a b => a -> a
261 (Notice: no functional dependency in J's class decl.)
262 Here f's type is perfectly fine, provided f is only called at Int.
263 It's premature to complain when meeting f's signature, or even
264 when inferring a type for f.
268 However, we don't *need* to report ambiguity right away. It'll always
269 show up at the call site.... and eventually at main, which needs special
270 treatment. Nevertheless, reporting ambiguity promptly is an excellent thing.
272 So here's the plan. We WARN about probable ambiguity if
274 fv(Cq) is not a subset of oclose(fv(T) union fv(G), C)
276 (all tested before quantification).
277 That is, all the type variables in Cq must be fixed by the the variables
278 in the environment, or by the variables in the type.
280 Notice that we union before calling oclose. Here's an example:
282 class J a b c | a b -> c
286 forall b c. (J a b c) => b -> b
288 Only if we union {a} from G with {b} from T before using oclose,
289 do we see that c is fixed.
291 It's a bit vague exactly which C we should use for this oclose call. If we
292 don't fix enough variables we might complain when we shouldn't (see
293 the above nasty example). Nothing will be perfect. That's why we can
294 only issue a warning.
297 Can we ever be *certain* about ambiguity? Yes: if there's a constraint
299 c in C such that fv(c) intersect (fv(G) union fv(T)) = EMPTY
301 then c is a "bubble"; there's no way it can ever improve, and it's
302 certainly ambiguous. UNLESS it is a constant (sigh). And what about
307 instance H x y => K (x,y)
309 Is this type ambiguous?
310 forall a b. (K (a,b), Eq b) => a -> a
312 Urk. The (Eq b) looks "definitely ambiguous" but it isn't. What we are after
313 is a "bubble" that's a set of constraints
315 Cq = Ca union Cq' st fv(Ca) intersect (fv(Cq') union fv(T) union fv(G)) = EMPTY
317 Hence another idea. To decide Q start with fv(T) and grow it
318 by transitive closure in Cq (no functional dependencies involved).
319 Now partition Cq using Q, leaving the definitely-ambiguous and probably-ok.
320 The definitely-ambiguous can then float out, and get smashed at top level
321 (which squashes out the constants, like Eq (T a) above)
324 --------------------------------------
325 Notes on principal types
326 --------------------------------------
331 f x = let g y = op (y::Int) in True
333 Here the principal type of f is (forall a. a->a)
334 but we'll produce the non-principal type
335 f :: forall a. C Int => a -> a
338 --------------------------------------
339 The need for forall's in constraints
340 --------------------------------------
342 [Exchange on Haskell Cafe 5/6 Dec 2000]
344 class C t where op :: t -> Bool
345 instance C [t] where op x = True
347 p y = (let f :: c -> Bool; f x = op (y >> return x) in f, y ++ [])
348 q y = (y ++ [], let f :: c -> Bool; f x = op (y >> return x) in f)
350 The definitions of p and q differ only in the order of the components in
351 the pair on their right-hand sides. And yet:
353 ghc and "Typing Haskell in Haskell" reject p, but accept q;
354 Hugs rejects q, but accepts p;
355 hbc rejects both p and q;
356 nhc98 ... (Malcolm, can you fill in the blank for us!).
358 The type signature for f forces context reduction to take place, and
359 the results of this depend on whether or not the type of y is known,
360 which in turn depends on which component of the pair the type checker
363 Solution: if y::m a, float out the constraints
364 Monad m, forall c. C (m c)
365 When m is later unified with [], we can solve both constraints.
368 --------------------------------------
369 Notes on implicit parameters
370 --------------------------------------
372 Note [Inheriting implicit parameters]
373 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
378 where f is *not* a top-level binding.
379 From the RHS of f we'll get the constraint (?y::Int).
380 There are two types we might infer for f:
384 (so we get ?y from the context of f's definition), or
386 f :: (?y::Int) => Int -> Int
388 At first you might think the first was better, becuase then
389 ?y behaves like a free variable of the definition, rather than
390 having to be passed at each call site. But of course, the WHOLE
391 IDEA is that ?y should be passed at each call site (that's what
392 dynamic binding means) so we'd better infer the second.
394 BOTTOM LINE: when *inferring types* you *must* quantify
395 over implicit parameters. See the predicate isFreeWhenInferring.
398 Note [Implicit parameters and ambiguity]
399 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
400 What type should we infer for this?
401 f x = (show ?y, x::Int)
402 Since we must quantify over the ?y, the most plausible type is
403 f :: (Show a, ?y::a) => Int -> (String, Int)
404 But notice that the type of the RHS is (String,Int), with no type
405 varibables mentioned at all! The type of f looks ambiguous. But
406 it isn't, because at a call site we might have
407 let ?y = 5::Int in f 7
408 and all is well. In effect, implicit parameters are, well, parameters,
409 so we can take their type variables into account as part of the
410 "tau-tvs" stuff. This is done in the function 'FunDeps.grow'.
413 Question 2: type signatures
414 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
415 BUT WATCH OUT: When you supply a type signature, we can't force you
416 to quantify over implicit parameters. For example:
420 This is perfectly reasonable. We do not want to insist on
422 (?x + 1) :: (?x::Int => Int)
424 That would be silly. Here, the definition site *is* the occurrence site,
425 so the above strictures don't apply. Hence the difference between
426 tcSimplifyCheck (which *does* allow implicit paramters to be inherited)
427 and tcSimplifyCheckBind (which does not).
429 What about when you supply a type signature for a binding?
430 Is it legal to give the following explicit, user type
431 signature to f, thus:
436 At first sight this seems reasonable, but it has the nasty property
437 that adding a type signature changes the dynamic semantics.
440 (let f x = (x::Int) + ?y
441 in (f 3, f 3 with ?y=5)) with ?y = 6
447 in (f 3, f 3 with ?y=5)) with ?y = 6
451 Indeed, simply inlining f (at the Haskell source level) would change the
454 Nevertheless, as Launchbury says (email Oct 01) we can't really give the
455 semantics for a Haskell program without knowing its typing, so if you
456 change the typing you may change the semantics.
458 To make things consistent in all cases where we are *checking* against
459 a supplied signature (as opposed to inferring a type), we adopt the
462 a signature does not need to quantify over implicit params.
464 [This represents a (rather marginal) change of policy since GHC 5.02,
465 which *required* an explicit signature to quantify over all implicit
466 params for the reasons mentioned above.]
468 But that raises a new question. Consider
470 Given (signature) ?x::Int
471 Wanted (inferred) ?x::Int, ?y::Bool
473 Clearly we want to discharge the ?x and float the ?y out. But
474 what is the criterion that distinguishes them? Clearly it isn't
475 what free type variables they have. The Right Thing seems to be
476 to float a constraint that
477 neither mentions any of the quantified type variables
478 nor any of the quantified implicit parameters
480 See the predicate isFreeWhenChecking.
483 Question 3: monomorphism
484 ~~~~~~~~~~~~~~~~~~~~~~~~
485 There's a nasty corner case when the monomorphism restriction bites:
489 The argument above suggests that we *must* generalise
490 over the ?y parameter, to get
491 z :: (?y::Int) => Int,
492 but the monomorphism restriction says that we *must not*, giving
494 Why does the momomorphism restriction say this? Because if you have
496 let z = x + ?y in z+z
498 you might not expect the addition to be done twice --- but it will if
499 we follow the argument of Question 2 and generalise over ?y.
502 Question 4: top level
503 ~~~~~~~~~~~~~~~~~~~~~
504 At the top level, monomorhism makes no sense at all.
507 main = let ?x = 5 in print foo
511 woggle :: (?x :: Int) => Int -> Int
514 We definitely don't want (foo :: Int) with a top-level implicit parameter
515 (?x::Int) becuase there is no way to bind it.
520 (A) Always generalise over implicit parameters
521 Bindings that fall under the monomorphism restriction can't
525 * Inlining remains valid
526 * No unexpected loss of sharing
527 * But simple bindings like
529 will be rejected, unless you add an explicit type signature
530 (to avoid the monomorphism restriction)
531 z :: (?y::Int) => Int
533 This seems unacceptable
535 (B) Monomorphism restriction "wins"
536 Bindings that fall under the monomorphism restriction can't
538 Always generalise over implicit parameters *except* for bindings
539 that fall under the monomorphism restriction
542 * Inlining isn't valid in general
543 * No unexpected loss of sharing
544 * Simple bindings like
546 accepted (get value of ?y from binding site)
548 (C) Always generalise over implicit parameters
549 Bindings that fall under the monomorphism restriction can't
550 be generalised, EXCEPT for implicit parameters
552 * Inlining remains valid
553 * Unexpected loss of sharing (from the extra generalisation)
554 * Simple bindings like
556 accepted (get value of ?y from occurrence sites)
561 None of these choices seems very satisfactory. But at least we should
562 decide which we want to do.
564 It's really not clear what is the Right Thing To Do. If you see
568 would you expect the value of ?y to be got from the *occurrence sites*
569 of 'z', or from the valuue of ?y at the *definition* of 'z'? In the
570 case of function definitions, the answer is clearly the former, but
571 less so in the case of non-fucntion definitions. On the other hand,
572 if we say that we get the value of ?y from the definition site of 'z',
573 then inlining 'z' might change the semantics of the program.
575 Choice (C) really says "the monomorphism restriction doesn't apply
576 to implicit parameters". Which is fine, but remember that every
577 innocent binding 'x = ...' that mentions an implicit parameter in
578 the RHS becomes a *function* of that parameter, called at each
579 use of 'x'. Now, the chances are that there are no intervening 'with'
580 clauses that bind ?y, so a decent compiler should common up all
581 those function calls. So I think I strongly favour (C). Indeed,
582 one could make a similar argument for abolishing the monomorphism
583 restriction altogether.
585 BOTTOM LINE: we choose (B) at present. See tcSimplifyRestricted
589 %************************************************************************
591 \subsection{tcSimplifyInfer}
593 %************************************************************************
595 tcSimplify is called when we *inferring* a type. Here's the overall game plan:
597 1. Compute Q = grow( fvs(T), C )
599 2. Partition C based on Q into Ct and Cq. Notice that ambiguous
600 predicates will end up in Ct; we deal with them at the top level
602 3. Try improvement, using functional dependencies
604 4. If Step 3 did any unification, repeat from step 1
605 (Unification can change the result of 'grow'.)
607 Note: we don't reduce dictionaries in step 2. For example, if we have
608 Eq (a,b), we don't simplify to (Eq a, Eq b). So Q won't be different
609 after step 2. However note that we may therefore quantify over more
610 type variables than we absolutely have to.
612 For the guts, we need a loop, that alternates context reduction and
613 improvement with unification. E.g. Suppose we have
615 class C x y | x->y where ...
617 and tcSimplify is called with:
619 Then improvement unifies a with b, giving
622 If we need to unify anything, we rattle round the whole thing all over
629 -> TcTyVarSet -- fv(T); type vars
631 -> TcM ([TcTyVar], -- Tyvars to quantify (zonked and quantified)
632 [Inst], -- Dict Ids that must be bound here (zonked)
633 TcDictBinds) -- Bindings
634 -- Any free (escaping) Insts are tossed into the environment
639 tcSimplifyInfer doc tau_tvs wanted
640 = do { tau_tvs' <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
641 ; wanted' <- mappM zonkInst wanted -- Zonk before deciding quantified tyvars
642 ; gbl_tvs <- tcGetGlobalTyVars
643 ; let preds = fdPredsOfInsts wanted'
644 qtvs = grow preds tau_tvs' `minusVarSet` oclose preds gbl_tvs
645 -- See Note [Choosing which variables to quantify]
647 -- To maximise sharing, remove from consideration any
648 -- constraints that don't mention qtvs at all
649 ; let (free1, bound) = partition (isFreeWhenInferring qtvs) wanted'
652 -- To make types simple, reduce as much as possible
653 ; traceTc (text "infer" <+> (ppr preds $$ ppr (grow preds tau_tvs') $$ ppr gbl_tvs $$
654 ppr (oclose preds gbl_tvs) $$ ppr free1 $$ ppr bound))
655 ; (irreds1, binds1) <- tryHardCheckLoop doc bound
657 -- Note [Inference and implication constraints]
658 ; let want_dict d = tyVarsOfInst d `intersectsVarSet` qtvs
659 ; (irreds2, binds2) <- approximateImplications doc want_dict irreds1
661 -- By now improvment may have taken place, and we must *not*
662 -- quantify over any variable free in the environment
663 -- tc137 (function h inside g) is an example
664 ; gbl_tvs <- tcGetGlobalTyVars
665 ; qtvs1 <- zonkTcTyVarsAndFV (varSetElems qtvs)
666 ; qtvs2 <- zonkQuantifiedTyVars (varSetElems (qtvs1 `minusVarSet` gbl_tvs))
668 -- Do not quantify over constraints that *now* do not
669 -- mention quantified type variables, because they are
670 -- simply ambiguous (or might be bound further out). Example:
671 -- f :: Eq b => a -> (a, b)
673 -- From the RHS of g we get the MethodInst f77 :: alpha -> (alpha, beta)
674 -- We decide to quantify over 'alpha' alone, but free1 does not include f77
675 -- because f77 mentions 'alpha'. Then reducing leaves only the (ambiguous)
676 -- constraint (Eq beta), which we dump back into the free set
677 -- See test tcfail181
678 ; let (free3, irreds3) = partition (isFreeWhenInferring (mkVarSet qtvs2)) irreds2
681 -- We can't abstract over any remaining unsolved
682 -- implications so instead just float them outwards. Ugh.
683 ; let (q_dicts, implics) = partition isDict irreds3
684 ; loc <- getInstLoc (ImplicOrigin doc)
685 ; implic_bind <- bindIrreds loc qtvs2 q_dicts implics
687 ; return (qtvs2, q_dicts, binds1 `unionBags` binds2 `unionBags` implic_bind) }
688 -- NB: when we are done, we might have some bindings, but
689 -- the final qtvs might be empty. See Note [NO TYVARS] below.
691 approximateImplications :: SDoc -> (Inst -> Bool) -> [Inst] -> TcM ([Inst], TcDictBinds)
692 -- Note [Inference and implication constraints]
693 -- Given a bunch of Dict and ImplicInsts, try to approximate the implications by
694 -- - fetching any dicts inside them that are free
695 -- - using those dicts as cruder constraints, to solve the implications
696 -- - returning the extra ones too
698 approximateImplications doc want_dict irreds
700 = return (irreds, emptyBag)
702 = do { extra_dicts' <- mapM cloneDict extra_dicts
703 ; tryHardCheckLoop doc (extra_dicts' ++ irreds) }
704 -- By adding extra_dicts', we make them
705 -- available to solve the implication constraints
707 extra_dicts = get_dicts (filter isImplicInst irreds)
709 get_dicts :: [Inst] -> [Inst] -- Returns only Dicts
710 -- Find the wanted constraints in implication constraints that satisfy
711 -- want_dict, and are not bound by forall's in the constraint itself
712 get_dicts ds = concatMap get_dict ds
714 get_dict d@(Dict {}) | want_dict d = [d]
716 get_dict (ImplicInst {tci_tyvars = tvs, tci_wanted = wanteds})
717 = [ d | let tv_set = mkVarSet tvs
718 , d <- get_dicts wanteds
719 , not (tyVarsOfInst d `intersectsVarSet` tv_set)]
720 get_dict other = pprPanic "approximateImplications" (ppr other)
723 Note [Inference and implication constraints]
724 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
725 Suppose we have a wanted implication constraint (perhaps arising from
726 a nested pattern match) like
728 and we are now trying to quantify over 'a' when inferring the type for
729 a function. In principle it's possible that there might be an instance
730 instance (C a, E a) => D [a]
731 so the context (E a) would suffice. The Right Thing is to abstract over
732 the implication constraint, but we don't do that (a) because it'll be
733 surprising to programmers and (b) because we don't have the machinery to deal
734 with 'given' implications.
736 So our best approximation is to make (D [a]) part of the inferred
737 context, so we can use that to discharge the implication. Hence
738 the strange function getImplicWanteds.
740 The common cases are more clear-cut, when we have things like
742 Here, abstracting over (C b) is not an approximation at all -- but see
743 Note [Freeness and implications].
745 See Trac #1430 and test tc228.
749 -----------------------------------------------------------
750 -- tcSimplifyInferCheck is used when we know the constraints we are to simplify
751 -- against, but we don't know the type variables over which we are going to quantify.
752 -- This happens when we have a type signature for a mutually recursive group
755 -> TcTyVarSet -- fv(T)
758 -> TcM ([TyVar], -- Fully zonked, and quantified
759 TcDictBinds) -- Bindings
761 tcSimplifyInferCheck loc tau_tvs givens wanteds
762 = do { (irreds, binds) <- gentleCheckLoop loc givens wanteds
764 -- Figure out which type variables to quantify over
765 -- You might think it should just be the signature tyvars,
766 -- but in bizarre cases you can get extra ones
767 -- f :: forall a. Num a => a -> a
768 -- f x = fst (g (x, head [])) + 1
770 -- Here we infer g :: forall a b. a -> b -> (b,a)
771 -- We don't want g to be monomorphic in b just because
772 -- f isn't quantified over b.
773 ; let all_tvs = varSetElems (tau_tvs `unionVarSet` tyVarsOfInsts givens)
774 ; all_tvs <- zonkTcTyVarsAndFV all_tvs
775 ; gbl_tvs <- tcGetGlobalTyVars
776 ; let qtvs = varSetElems (all_tvs `minusVarSet` gbl_tvs)
777 -- We could close gbl_tvs, but its not necessary for
778 -- soundness, and it'll only affect which tyvars, not which
779 -- dictionaries, we quantify over
781 ; qtvs' <- zonkQuantifiedTyVars qtvs
783 -- Now we are back to normal (c.f. tcSimplCheck)
784 ; implic_bind <- bindIrreds loc qtvs' givens irreds
786 ; return (qtvs', binds `unionBags` implic_bind) }
789 Note [Squashing methods]
790 ~~~~~~~~~~~~~~~~~~~~~~~~~
791 Be careful if you want to float methods more:
792 truncate :: forall a. RealFrac a => forall b. Integral b => a -> b
793 From an application (truncate f i) we get
796 If we have also have a second occurrence of truncate, we get
799 When simplifying with i,f free, we might still notice that
800 t1=t3; but alas, the binding for t2 (which mentions t1)
801 may continue to float out!
806 class Y a b | a -> b where
809 instance Y [[a]] a where
812 k :: X a -> X a -> X a
814 g :: Num a => [X a] -> [X a]
817 h ys = ys ++ map (k (y [[0]])) xs
819 The excitement comes when simplifying the bindings for h. Initially
820 try to simplify {y @ [[t1]] t2, 0 @ t1}, with initial qtvs = {t2}.
821 From this we get t1:=:t2, but also various bindings. We can't forget
822 the bindings (because of [LOOP]), but in fact t1 is what g is
825 The net effect of [NO TYVARS]
828 isFreeWhenInferring :: TyVarSet -> Inst -> Bool
829 isFreeWhenInferring qtvs inst
830 = isFreeWrtTyVars qtvs inst -- Constrains no quantified vars
831 && isInheritableInst inst -- and no implicit parameter involved
832 -- see Note [Inheriting implicit parameters]
834 {- No longer used (with implication constraints)
835 isFreeWhenChecking :: TyVarSet -- Quantified tyvars
836 -> NameSet -- Quantified implicit parameters
838 isFreeWhenChecking qtvs ips inst
839 = isFreeWrtTyVars qtvs inst
840 && isFreeWrtIPs ips inst
843 isFreeWrtTyVars qtvs inst = tyVarsOfInst inst `disjointVarSet` qtvs
844 isFreeWrtIPs ips inst = not (any (`elemNameSet` ips) (ipNamesOfInst inst))
848 %************************************************************************
850 \subsection{tcSimplifyCheck}
852 %************************************************************************
854 @tcSimplifyCheck@ is used when we know exactly the set of variables
855 we are going to quantify over. For example, a class or instance declaration.
858 -----------------------------------------------------------
859 -- tcSimplifyCheck is used when checking expression type signatures,
860 -- class decls, instance decls etc.
861 tcSimplifyCheck :: InstLoc
862 -> [TcTyVar] -- Quantify over these
865 -> TcM TcDictBinds -- Bindings
866 tcSimplifyCheck loc qtvs givens wanteds
867 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
868 do { (irreds, binds) <- gentleCheckLoop loc givens wanteds
869 ; implic_bind <- bindIrreds loc qtvs givens irreds
870 ; return (binds `unionBags` implic_bind) }
872 -----------------------------------------------------------
873 -- tcSimplifyCheckPat is used for existential pattern match
874 tcSimplifyCheckPat :: InstLoc
875 -> [CoVar] -> Refinement
876 -> [TcTyVar] -- Quantify over these
879 -> TcM TcDictBinds -- Bindings
880 tcSimplifyCheckPat loc co_vars reft qtvs givens wanteds
881 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
882 do { (irreds, binds) <- gentleCheckLoop loc givens wanteds
883 ; implic_bind <- bindIrredsR loc qtvs co_vars reft
885 ; return (binds `unionBags` implic_bind) }
887 -----------------------------------------------------------
888 bindIrreds :: InstLoc -> [TcTyVar]
891 bindIrreds loc qtvs givens irreds
892 = bindIrredsR loc qtvs [] emptyRefinement givens irreds
894 bindIrredsR :: InstLoc -> [TcTyVar] -> [CoVar]
895 -> Refinement -> [Inst] -> [Inst]
897 -- Make a binding that binds 'irreds', by generating an implication
898 -- constraint for them, *and* throwing the constraint into the LIE
899 bindIrredsR loc qtvs co_vars reft givens irreds
903 = do { let givens' = filter isDict givens
904 -- The givens can include methods
905 -- See Note [Pruning the givens in an implication constraint]
907 -- If there are no 'givens' *and* the refinement is empty
908 -- (the refinement is like more givens), then it's safe to
909 -- partition the 'wanteds' by their qtvs, thereby trimming irreds
910 -- See Note [Freeness and implications]
911 ; irreds' <- if null givens' && isEmptyRefinement reft
913 { let qtv_set = mkVarSet qtvs
914 (frees, real_irreds) = partition (isFreeWrtTyVars qtv_set) irreds
916 ; return real_irreds }
919 ; let all_tvs = qtvs ++ co_vars -- Abstract over all these
920 ; (implics, bind) <- makeImplicationBind loc all_tvs reft givens' irreds'
921 -- This call does the real work
922 -- If irreds' is empty, it does something sensible
927 makeImplicationBind :: InstLoc -> [TcTyVar] -> Refinement
929 -> TcM ([Inst], TcDictBinds)
930 -- Make a binding that binds 'irreds', by generating an implication
931 -- constraint for them, *and* throwing the constraint into the LIE
932 -- The binding looks like
933 -- (ir1, .., irn) = f qtvs givens
934 -- where f is (evidence for) the new implication constraint
935 -- f :: forall qtvs. {reft} givens => (ir1, .., irn)
936 -- qtvs includes coercion variables
938 -- This binding must line up the 'rhs' in reduceImplication
939 makeImplicationBind loc all_tvs reft
940 givens -- Guaranteed all Dicts
942 | null irreds -- If there are no irreds, we are done
943 = return ([], emptyBag)
944 | otherwise -- Otherwise we must generate a binding
945 = do { uniq <- newUnique
946 ; span <- getSrcSpanM
947 ; let name = mkInternalName uniq (mkVarOcc "ic") span
948 implic_inst = ImplicInst { tci_name = name, tci_reft = reft,
949 tci_tyvars = all_tvs,
951 tci_wanted = irreds, tci_loc = loc }
953 ; let n_irreds = length irreds
954 irred_ids = map instToId irreds
955 tup_ty = mkTupleTy Boxed n_irreds (map idType irred_ids)
956 pat = TuplePat (map nlVarPat irred_ids) Boxed tup_ty
957 rhs = L span (mkHsWrap co (HsVar (instToId implic_inst)))
958 co = mkWpApps (map instToId givens) <.> mkWpTyApps (mkTyVarTys all_tvs)
959 bind | n_irreds==1 = VarBind (head irred_ids) rhs
960 | otherwise = PatBind { pat_lhs = L span pat,
961 pat_rhs = unguardedGRHSs rhs,
963 bind_fvs = placeHolderNames }
964 ; -- pprTrace "Make implic inst" (ppr implic_inst) $
965 return ([implic_inst], unitBag (L span bind)) }
967 -----------------------------------------------------------
968 tryHardCheckLoop :: SDoc
970 -> TcM ([Inst], TcDictBinds)
972 tryHardCheckLoop doc wanteds
973 = checkLoop (mkRedEnv doc try_me []) wanteds
975 try_me inst = ReduceMe AddSCs
976 -- Here's the try-hard bit
978 -----------------------------------------------------------
979 gentleCheckLoop :: InstLoc
982 -> TcM ([Inst], TcDictBinds)
984 gentleCheckLoop inst_loc givens wanteds
985 = checkLoop env wanteds
987 env = mkRedEnv (pprInstLoc inst_loc) try_me givens
989 try_me inst | isMethodOrLit inst = ReduceMe AddSCs
991 -- When checking against a given signature
992 -- we MUST be very gentle: Note [Check gently]
997 We have to very careful about not simplifying too vigorously
1002 f :: Show b => T b -> b
1003 f (MkT x) = show [x]
1005 Inside the pattern match, which binds (a:*, x:a), we know that
1007 Hence we have a dictionary for Show [a] available; and indeed we
1008 need it. We are going to build an implication contraint
1009 forall a. (b~[a]) => Show [a]
1010 Later, we will solve this constraint using the knowledg e(Show b)
1012 But we MUST NOT reduce (Show [a]) to (Show a), else the whole
1013 thing becomes insoluble. So we simplify gently (get rid of literals
1014 and methods only, plus common up equal things), deferring the real
1015 work until top level, when we solve the implication constraint
1016 with tryHardCheckLooop.
1020 -----------------------------------------------------------
1023 -> TcM ([Inst], TcDictBinds)
1024 -- Precondition: givens are completely rigid
1025 -- Postcondition: returned Insts are zonked
1027 checkLoop env wanteds
1028 = do { -- Givens are skolems, so no need to zonk them
1029 wanteds' <- mappM zonkInst wanteds
1031 ; (improved, binds, irreds) <- reduceContext env wanteds'
1033 ; if not improved then
1034 return (irreds, binds)
1037 -- If improvement did some unification, we go round again.
1038 -- We start again with irreds, not wanteds
1039 -- Using an instance decl might have introduced a fresh type variable
1040 -- which might have been unified, so we'd get an infinite loop
1041 -- if we started again with wanteds! See Note [LOOP]
1042 { (irreds1, binds1) <- checkLoop env irreds
1043 ; return (irreds1, binds `unionBags` binds1) } }
1048 class If b t e r | b t e -> r
1051 class Lte a b c | a b -> c where lte :: a -> b -> c
1053 instance (Lte a b l,If l b a c) => Max a b c
1055 Wanted: Max Z (S x) y
1057 Then we'll reduce using the Max instance to:
1058 (Lte Z (S x) l, If l (S x) Z y)
1059 and improve by binding l->T, after which we can do some reduction
1060 on both the Lte and If constraints. What we *can't* do is start again
1061 with (Max Z (S x) y)!
1065 %************************************************************************
1067 tcSimplifySuperClasses
1069 %************************************************************************
1071 Note [SUPERCLASS-LOOP 1]
1072 ~~~~~~~~~~~~~~~~~~~~~~~~
1073 We have to be very, very careful when generating superclasses, lest we
1074 accidentally build a loop. Here's an example:
1078 class S a => C a where { opc :: a -> a }
1079 class S b => D b where { opd :: b -> b }
1081 instance C Int where
1084 instance D Int where
1087 From (instance C Int) we get the constraint set {ds1:S Int, dd:D Int}
1088 Simplifying, we may well get:
1089 $dfCInt = :C ds1 (opd dd)
1092 Notice that we spot that we can extract ds1 from dd.
1094 Alas! Alack! We can do the same for (instance D Int):
1096 $dfDInt = :D ds2 (opc dc)
1100 And now we've defined the superclass in terms of itself.
1102 Solution: never generate a superclass selectors at all when
1103 satisfying the superclass context of an instance declaration.
1105 Two more nasty cases are in
1110 tcSimplifySuperClasses
1115 tcSimplifySuperClasses loc givens sc_wanteds
1116 = do { (irreds, binds1) <- checkLoop env sc_wanteds
1117 ; let (tidy_env, tidy_irreds) = tidyInsts irreds
1118 ; reportNoInstances tidy_env (Just (loc, givens)) tidy_irreds
1121 env = mkRedEnv (pprInstLoc loc) try_me givens
1122 try_me inst = ReduceMe NoSCs
1123 -- Like tryHardCheckLoop, but with NoSCs
1127 %************************************************************************
1129 \subsection{tcSimplifyRestricted}
1131 %************************************************************************
1133 tcSimplifyRestricted infers which type variables to quantify for a
1134 group of restricted bindings. This isn't trivial.
1137 We want to quantify over a to get id :: forall a. a->a
1140 We do not want to quantify over a, because there's an Eq a
1141 constraint, so we get eq :: a->a->Bool (notice no forall)
1144 RHS has type 'tau', whose free tyvars are tau_tvs
1145 RHS has constraints 'wanteds'
1148 Quantify over (tau_tvs \ ftvs(wanteds))
1149 This is bad. The constraints may contain (Monad (ST s))
1150 where we have instance Monad (ST s) where...
1151 so there's no need to be monomorphic in s!
1153 Also the constraint might be a method constraint,
1154 whose type mentions a perfectly innocent tyvar:
1155 op :: Num a => a -> b -> a
1156 Here, b is unconstrained. A good example would be
1158 We want to infer the polymorphic type
1159 foo :: forall b. b -> b
1162 Plan B (cunning, used for a long time up to and including GHC 6.2)
1163 Step 1: Simplify the constraints as much as possible (to deal
1164 with Plan A's problem). Then set
1165 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1167 Step 2: Now simplify again, treating the constraint as 'free' if
1168 it does not mention qtvs, and trying to reduce it otherwise.
1169 The reasons for this is to maximise sharing.
1171 This fails for a very subtle reason. Suppose that in the Step 2
1172 a constraint (Foo (Succ Zero) (Succ Zero) b) gets thrown upstairs as 'free'.
1173 In the Step 1 this constraint might have been simplified, perhaps to
1174 (Foo Zero Zero b), AND THEN THAT MIGHT BE IMPROVED, to bind 'b' to 'T'.
1175 This won't happen in Step 2... but that in turn might prevent some other
1176 constraint (Baz [a] b) being simplified (e.g. via instance Baz [a] T where {..})
1177 and that in turn breaks the invariant that no constraints are quantified over.
1179 Test typecheck/should_compile/tc177 (which failed in GHC 6.2) demonstrates
1184 Step 1: Simplify the constraints as much as possible (to deal
1185 with Plan A's problem). Then set
1186 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1187 Return the bindings from Step 1.
1190 A note about Plan C (arising from "bug" reported by George Russel March 2004)
1193 instance (HasBinary ty IO) => HasCodedValue ty
1195 foo :: HasCodedValue a => String -> IO a
1197 doDecodeIO :: HasCodedValue a => () -> () -> IO a
1198 doDecodeIO codedValue view
1199 = let { act = foo "foo" } in act
1201 You might think this should work becuase the call to foo gives rise to a constraint
1202 (HasCodedValue t), which can be satisfied by the type sig for doDecodeIO. But the
1203 restricted binding act = ... calls tcSimplifyRestricted, and PlanC simplifies the
1204 constraint using the (rather bogus) instance declaration, and now we are stuffed.
1206 I claim this is not really a bug -- but it bit Sergey as well as George. So here's
1210 Plan D (a variant of plan B)
1211 Step 1: Simplify the constraints as much as possible (to deal
1212 with Plan A's problem), BUT DO NO IMPROVEMENT. Then set
1213 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1215 Step 2: Now simplify again, treating the constraint as 'free' if
1216 it does not mention qtvs, and trying to reduce it otherwise.
1218 The point here is that it's generally OK to have too few qtvs; that is,
1219 to make the thing more monomorphic than it could be. We don't want to
1220 do that in the common cases, but in wierd cases it's ok: the programmer
1221 can always add a signature.
1223 Too few qtvs => too many wanteds, which is what happens if you do less
1228 tcSimplifyRestricted -- Used for restricted binding groups
1229 -- i.e. ones subject to the monomorphism restriction
1232 -> [Name] -- Things bound in this group
1233 -> TcTyVarSet -- Free in the type of the RHSs
1234 -> [Inst] -- Free in the RHSs
1235 -> TcM ([TyVar], -- Tyvars to quantify (zonked and quantified)
1236 TcDictBinds) -- Bindings
1237 -- tcSimpifyRestricted returns no constraints to
1238 -- quantify over; by definition there are none.
1239 -- They are all thrown back in the LIE
1241 tcSimplifyRestricted doc top_lvl bndrs tau_tvs wanteds
1242 -- Zonk everything in sight
1243 = do { wanteds' <- mappM zonkInst wanteds
1245 -- 'ReduceMe': Reduce as far as we can. Don't stop at
1246 -- dicts; the idea is to get rid of as many type
1247 -- variables as possible, and we don't want to stop
1248 -- at (say) Monad (ST s), because that reduces
1249 -- immediately, with no constraint on s.
1251 -- BUT do no improvement! See Plan D above
1252 -- HOWEVER, some unification may take place, if we instantiate
1253 -- a method Inst with an equality constraint
1254 ; let env = mkNoImproveRedEnv doc (\i -> ReduceMe AddSCs)
1255 ; (_imp, _binds, constrained_dicts) <- reduceContext env wanteds'
1257 -- Next, figure out the tyvars we will quantify over
1258 ; tau_tvs' <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
1259 ; gbl_tvs' <- tcGetGlobalTyVars
1260 ; constrained_dicts' <- mappM zonkInst constrained_dicts
1262 ; let qtvs1 = tau_tvs' `minusVarSet` oclose (fdPredsOfInsts constrained_dicts) gbl_tvs'
1263 -- As in tcSimplifyInfer
1265 -- Do not quantify over constrained type variables:
1266 -- this is the monomorphism restriction
1267 constrained_tvs' = tyVarsOfInsts constrained_dicts'
1268 qtvs = qtvs1 `minusVarSet` constrained_tvs'
1269 pp_bndrs = pprWithCommas (quotes . ppr) bndrs
1272 ; warn_mono <- doptM Opt_WarnMonomorphism
1273 ; warnTc (warn_mono && (constrained_tvs' `intersectsVarSet` qtvs1))
1274 (vcat[ ptext SLIT("the Monomorphism Restriction applies to the binding")
1275 <> plural bndrs <+> ptext SLIT("for") <+> pp_bndrs,
1276 ptext SLIT("Consider giving a type signature for") <+> pp_bndrs])
1278 ; traceTc (text "tcSimplifyRestricted" <+> vcat [
1279 pprInsts wanteds, pprInsts constrained_dicts',
1281 ppr constrained_tvs', ppr tau_tvs', ppr qtvs ])
1283 -- The first step may have squashed more methods than
1284 -- necessary, so try again, this time more gently, knowing the exact
1285 -- set of type variables to quantify over.
1287 -- We quantify only over constraints that are captured by qtvs;
1288 -- these will just be a subset of non-dicts. This in contrast
1289 -- to normal inference (using isFreeWhenInferring) in which we quantify over
1290 -- all *non-inheritable* constraints too. This implements choice
1291 -- (B) under "implicit parameter and monomorphism" above.
1293 -- Remember that we may need to do *some* simplification, to
1294 -- (for example) squash {Monad (ST s)} into {}. It's not enough
1295 -- just to float all constraints
1297 -- At top level, we *do* squash methods becuase we want to
1298 -- expose implicit parameters to the test that follows
1299 ; let is_nested_group = isNotTopLevel top_lvl
1300 try_me inst | isFreeWrtTyVars qtvs inst,
1301 (is_nested_group || isDict inst) = Stop
1302 | otherwise = ReduceMe AddSCs
1303 env = mkNoImproveRedEnv doc try_me
1304 ; (_imp, binds, irreds) <- reduceContext env wanteds'
1306 -- See "Notes on implicit parameters, Question 4: top level"
1307 ; ASSERT( all (isFreeWrtTyVars qtvs) irreds ) -- None should be captured
1308 if is_nested_group then
1310 else do { let (bad_ips, non_ips) = partition isIPDict irreds
1311 ; addTopIPErrs bndrs bad_ips
1312 ; extendLIEs non_ips }
1314 ; qtvs' <- zonkQuantifiedTyVars (varSetElems qtvs)
1315 ; return (qtvs', binds) }
1319 %************************************************************************
1323 %************************************************************************
1325 On the LHS of transformation rules we only simplify methods and constants,
1326 getting dictionaries. We want to keep all of them unsimplified, to serve
1327 as the available stuff for the RHS of the rule.
1329 Example. Consider the following left-hand side of a rule
1331 f (x == y) (y > z) = ...
1333 If we typecheck this expression we get constraints
1335 d1 :: Ord a, d2 :: Eq a
1337 We do NOT want to "simplify" to the LHS
1339 forall x::a, y::a, z::a, d1::Ord a.
1340 f ((==) (eqFromOrd d1) x y) ((>) d1 y z) = ...
1344 forall x::a, y::a, z::a, d1::Ord a, d2::Eq a.
1345 f ((==) d2 x y) ((>) d1 y z) = ...
1347 Here is another example:
1349 fromIntegral :: (Integral a, Num b) => a -> b
1350 {-# RULES "foo" fromIntegral = id :: Int -> Int #-}
1352 In the rule, a=b=Int, and Num Int is a superclass of Integral Int. But
1353 we *dont* want to get
1355 forall dIntegralInt.
1356 fromIntegral Int Int dIntegralInt (scsel dIntegralInt) = id Int
1358 because the scsel will mess up RULE matching. Instead we want
1360 forall dIntegralInt, dNumInt.
1361 fromIntegral Int Int dIntegralInt dNumInt = id Int
1365 g (x == y) (y == z) = ..
1367 where the two dictionaries are *identical*, we do NOT WANT
1369 forall x::a, y::a, z::a, d1::Eq a
1370 f ((==) d1 x y) ((>) d1 y z) = ...
1372 because that will only match if the dict args are (visibly) equal.
1373 Instead we want to quantify over the dictionaries separately.
1375 In short, tcSimplifyRuleLhs must *only* squash LitInst and MethInts, leaving
1376 all dicts unchanged, with absolutely no sharing. It's simpler to do this
1377 from scratch, rather than further parameterise simpleReduceLoop etc
1380 tcSimplifyRuleLhs :: [Inst] -> TcM ([Inst], TcDictBinds)
1381 tcSimplifyRuleLhs wanteds
1382 = go [] emptyBag wanteds
1385 = return (dicts, binds)
1386 go dicts binds (w:ws)
1388 = go (w:dicts) binds ws
1390 = do { w' <- zonkInst w -- So that (3::Int) does not generate a call
1391 -- to fromInteger; this looks fragile to me
1392 ; lookup_result <- lookupSimpleInst w'
1393 ; case lookup_result of
1394 GenInst ws' rhs -> go dicts (addBind binds (instToId w) rhs) (ws' ++ ws)
1395 NoInstance -> pprPanic "tcSimplifyRuleLhs" (ppr w)
1399 tcSimplifyBracket is used when simplifying the constraints arising from
1400 a Template Haskell bracket [| ... |]. We want to check that there aren't
1401 any constraints that can't be satisfied (e.g. Show Foo, where Foo has no
1402 Show instance), but we aren't otherwise interested in the results.
1403 Nor do we care about ambiguous dictionaries etc. We will type check
1404 this bracket again at its usage site.
1407 tcSimplifyBracket :: [Inst] -> TcM ()
1408 tcSimplifyBracket wanteds
1409 = do { tryHardCheckLoop doc wanteds
1412 doc = text "tcSimplifyBracket"
1416 %************************************************************************
1418 \subsection{Filtering at a dynamic binding}
1420 %************************************************************************
1425 we must discharge all the ?x constraints from B. We also do an improvement
1426 step; if we have ?x::t1 and ?x::t2 we must unify t1, t2.
1428 Actually, the constraints from B might improve the types in ?x. For example
1430 f :: (?x::Int) => Char -> Char
1433 then the constraint (?x::Int) arising from the call to f will
1434 force the binding for ?x to be of type Int.
1437 tcSimplifyIPs :: [Inst] -- The implicit parameters bound here
1440 -- We need a loop so that we do improvement, and then
1441 -- (next time round) generate a binding to connect the two
1443 -- Here the two ?x's have different types, and improvement
1444 -- makes them the same.
1446 tcSimplifyIPs given_ips wanteds
1447 = do { wanteds' <- mappM zonkInst wanteds
1448 ; given_ips' <- mappM zonkInst given_ips
1449 -- Unusually for checking, we *must* zonk the given_ips
1451 ; let env = mkRedEnv doc try_me given_ips'
1452 ; (improved, binds, irreds) <- reduceContext env wanteds'
1454 ; if not improved then
1455 ASSERT( all is_free irreds )
1456 do { extendLIEs irreds
1459 tcSimplifyIPs given_ips wanteds }
1461 doc = text "tcSimplifyIPs" <+> ppr given_ips
1462 ip_set = mkNameSet (ipNamesOfInsts given_ips)
1463 is_free inst = isFreeWrtIPs ip_set inst
1465 -- Simplify any methods that mention the implicit parameter
1466 try_me inst | is_free inst = Stop
1467 | otherwise = ReduceMe NoSCs
1471 %************************************************************************
1473 \subsection[binds-for-local-funs]{@bindInstsOfLocalFuns@}
1475 %************************************************************************
1477 When doing a binding group, we may have @Insts@ of local functions.
1478 For example, we might have...
1480 let f x = x + 1 -- orig local function (overloaded)
1481 f.1 = f Int -- two instances of f
1486 The point is: we must drop the bindings for @f.1@ and @f.2@ here,
1487 where @f@ is in scope; those @Insts@ must certainly not be passed
1488 upwards towards the top-level. If the @Insts@ were binding-ified up
1489 there, they would have unresolvable references to @f@.
1491 We pass in an @init_lie@ of @Insts@ and a list of locally-bound @Ids@.
1492 For each method @Inst@ in the @init_lie@ that mentions one of the
1493 @Ids@, we create a binding. We return the remaining @Insts@ (in an
1494 @LIE@), as well as the @HsBinds@ generated.
1497 bindInstsOfLocalFuns :: [Inst] -> [TcId] -> TcM TcDictBinds
1498 -- Simlifies only MethodInsts, and generate only bindings of form
1500 -- We're careful not to even generate bindings of the form
1502 -- You'd think that'd be fine, but it interacts with what is
1503 -- arguably a bug in Match.tidyEqnInfo (see notes there)
1505 bindInstsOfLocalFuns wanteds local_ids
1506 | null overloaded_ids
1508 = extendLIEs wanteds `thenM_`
1509 returnM emptyLHsBinds
1512 = do { (irreds, binds) <- checkLoop env for_me
1513 ; extendLIEs not_for_me
1517 env = mkRedEnv doc try_me []
1518 doc = text "bindInsts" <+> ppr local_ids
1519 overloaded_ids = filter is_overloaded local_ids
1520 is_overloaded id = isOverloadedTy (idType id)
1521 (for_me, not_for_me) = partition (isMethodFor overloaded_set) wanteds
1523 overloaded_set = mkVarSet overloaded_ids -- There can occasionally be a lot of them
1524 -- so it's worth building a set, so that
1525 -- lookup (in isMethodFor) is faster
1526 try_me inst | isMethod inst = ReduceMe NoSCs
1531 %************************************************************************
1533 \subsection{Data types for the reduction mechanism}
1535 %************************************************************************
1537 The main control over context reduction is here
1541 = RedEnv { red_doc :: SDoc -- The context
1542 , red_try_me :: Inst -> WhatToDo
1543 , red_improve :: Bool -- True <=> do improvement
1544 , red_givens :: [Inst] -- All guaranteed rigid
1546 -- but see Note [Rigidity]
1547 , red_stack :: (Int, [Inst]) -- Recursion stack (for err msg)
1548 -- See Note [RedStack]
1552 -- The red_givens are rigid so far as cmpInst is concerned.
1553 -- There is one case where they are not totally rigid, namely in tcSimplifyIPs
1554 -- let ?x = e in ...
1555 -- Here, the given is (?x::a), where 'a' is not necy a rigid type
1556 -- But that doesn't affect the comparison, which is based only on mame.
1559 -- The red_stack pair (n,insts) pair is just used for error reporting.
1560 -- 'n' is always the depth of the stack.
1561 -- The 'insts' is the stack of Insts being reduced: to produce X
1562 -- I had to produce Y, to produce Y I had to produce Z, and so on.
1565 mkRedEnv :: SDoc -> (Inst -> WhatToDo) -> [Inst] -> RedEnv
1566 mkRedEnv doc try_me givens
1567 = RedEnv { red_doc = doc, red_try_me = try_me,
1568 red_givens = givens, red_stack = (0,[]),
1569 red_improve = True }
1571 mkNoImproveRedEnv :: SDoc -> (Inst -> WhatToDo) -> RedEnv
1572 -- Do not do improvement; no givens
1573 mkNoImproveRedEnv doc try_me
1574 = RedEnv { red_doc = doc, red_try_me = try_me,
1575 red_givens = [], red_stack = (0,[]),
1576 red_improve = True }
1579 = ReduceMe WantSCs -- Try to reduce this
1580 -- If there's no instance, add the inst to the
1581 -- irreductible ones, but don't produce an error
1582 -- message of any kind.
1583 -- It might be quite legitimate such as (Eq a)!
1585 | Stop -- Return as irreducible unless it can
1586 -- be reduced to a constant in one step
1587 -- Do not add superclasses; see
1589 data WantSCs = NoSCs | AddSCs -- Tells whether we should add the superclasses
1590 -- of a predicate when adding it to the avails
1591 -- The reason for this flag is entirely the super-class loop problem
1592 -- Note [SUPER-CLASS LOOP 1]
1595 %************************************************************************
1597 \subsection[reduce]{@reduce@}
1599 %************************************************************************
1603 reduceContext :: RedEnv
1605 -> TcM (ImprovementDone,
1606 TcDictBinds, -- Dictionary bindings
1607 [Inst]) -- Irreducible
1609 reduceContext env wanteds
1610 = do { traceTc (text "reduceContext" <+> (vcat [
1611 text "----------------------",
1613 text "given" <+> ppr (red_givens env),
1614 text "wanted" <+> ppr wanteds,
1615 text "----------------------"
1618 -- Build the Avail mapping from "givens"
1619 ; init_state <- foldlM addGiven emptyAvails (red_givens env)
1622 -- Process non-implication constraints first, so that they are
1623 -- available to help solving the implication constraints
1624 -- ToDo: seems a bit inefficient and ad-hoc
1625 ; let (implics, rest) = partition isImplicInst wanteds
1626 ; avails <- reduceList env (rest ++ implics) init_state
1628 ; let improved = availsImproved avails
1629 ; (binds, irreds) <- extractResults avails wanteds
1631 ; traceTc (text "reduceContext end" <+> (vcat [
1632 text "----------------------",
1634 text "given" <+> ppr (red_givens env),
1635 text "wanted" <+> ppr wanteds,
1637 text "avails" <+> pprAvails avails,
1638 text "improved =" <+> ppr improved,
1639 text "irreds = " <+> ppr irreds,
1640 text "----------------------"
1643 ; return (improved, binds, irreds) }
1645 tcImproveOne :: Avails -> Inst -> TcM ImprovementDone
1646 tcImproveOne avails inst
1647 | not (isDict inst) = return False
1649 = do { inst_envs <- tcGetInstEnvs
1650 ; let eqns = improveOne (classInstances inst_envs)
1651 (dictPred inst, pprInstArising inst)
1652 [ (dictPred p, pprInstArising p)
1653 | p <- availsInsts avails, isDict p ]
1654 -- Avails has all the superclasses etc (good)
1655 -- It also has all the intermediates of the deduction (good)
1656 -- It does not have duplicates (good)
1657 -- NB that (?x::t1) and (?x::t2) will be held separately in avails
1658 -- so that improve will see them separate
1659 ; traceTc (text "improveOne" <+> ppr inst)
1662 unifyEqns :: [(Equation,(PredType,SDoc),(PredType,SDoc))]
1663 -> TcM ImprovementDone
1664 unifyEqns [] = return False
1666 = do { traceTc (ptext SLIT("Improve:") <+> vcat (map pprEquationDoc eqns))
1670 unify ((qtvs, pairs), what1, what2)
1671 = addErrCtxtM (mkEqnMsg what1 what2) $
1672 tcInstTyVars (varSetElems qtvs) `thenM` \ (_, _, tenv) ->
1673 mapM_ (unif_pr tenv) pairs
1674 unif_pr tenv (ty1,ty2) = unifyType (substTy tenv ty1) (substTy tenv ty2)
1676 pprEquationDoc (eqn, (p1,w1), (p2,w2)) = vcat [pprEquation eqn, nest 2 (ppr p1), nest 2 (ppr p2)]
1678 mkEqnMsg (pred1,from1) (pred2,from2) tidy_env
1679 = do { pred1' <- zonkTcPredType pred1; pred2' <- zonkTcPredType pred2
1680 ; let { pred1'' = tidyPred tidy_env pred1'; pred2'' = tidyPred tidy_env pred2' }
1681 ; let msg = vcat [ptext SLIT("When using functional dependencies to combine"),
1682 nest 2 (sep [ppr pred1'' <> comma, nest 2 from1]),
1683 nest 2 (sep [ppr pred2'' <> comma, nest 2 from2])]
1684 ; return (tidy_env, msg) }
1687 The main context-reduction function is @reduce@. Here's its game plan.
1690 reduceList :: RedEnv -> [Inst] -> Avails -> TcM Avails
1691 reduceList env@(RedEnv {red_stack = (n,stk)}) wanteds state
1692 = do { dopts <- getDOpts
1695 dumpTcRn (hang (ptext SLIT("Interesting! Context reduction stack depth") <+> int n)
1696 2 (ifPprDebug (nest 2 (pprStack stk))))
1699 ; if n >= ctxtStkDepth dopts then
1700 failWithTc (reduceDepthErr n stk)
1704 go [] state = return state
1705 go (w:ws) state = do { state' <- reduce (env {red_stack = (n+1, w:stk)}) w state
1708 -- Base case: we're done!
1709 reduce env wanted avails
1710 -- It's the same as an existing inst, or a superclass thereof
1711 | Just avail <- findAvail avails wanted
1715 = case red_try_me env wanted of {
1716 ; Stop -> try_simple (addIrred NoSCs) -- See Note [No superclasses for Stop]
1718 ; ReduceMe want_scs -> -- It should be reduced
1719 reduceInst env avails wanted `thenM` \ (avails, lookup_result) ->
1720 case lookup_result of
1721 NoInstance -> -- No such instance!
1722 -- Add it and its superclasses
1723 addIrred want_scs avails wanted
1725 GenInst [] rhs -> addWanted want_scs avails wanted rhs []
1727 GenInst wanteds' rhs -> do { avails1 <- addIrred NoSCs avails wanted
1728 ; avails2 <- reduceList env wanteds' avails1
1729 ; addWanted want_scs avails2 wanted rhs wanteds' }
1730 -- Temporarily do addIrred *before* the reduceList,
1731 -- which has the effect of adding the thing we are trying
1732 -- to prove to the database before trying to prove the things it
1733 -- needs. See note [RECURSIVE DICTIONARIES]
1734 -- NB: we must not do an addWanted before, because that adds the
1735 -- superclasses too, and thaat can lead to a spurious loop; see
1736 -- the examples in [SUPERCLASS-LOOP]
1737 -- So we do an addIrred before, and then overwrite it afterwards with addWanted
1741 -- First, see if the inst can be reduced to a constant in one step
1742 -- Works well for literals (1::Int) and constant dictionaries (d::Num Int)
1743 -- Don't bother for implication constraints, which take real work
1744 try_simple do_this_otherwise
1745 = do { res <- lookupSimpleInst wanted
1747 GenInst [] rhs -> addWanted AddSCs avails wanted rhs []
1748 other -> do_this_otherwise avails wanted }
1752 Note [SUPERCLASS-LOOP 2]
1753 ~~~~~~~~~~~~~~~~~~~~~~~~
1754 But the above isn't enough. Suppose we are *given* d1:Ord a,
1755 and want to deduce (d2:C [a]) where
1757 class Ord a => C a where
1758 instance Ord [a] => C [a] where ...
1760 Then we'll use the instance decl to deduce C [a] from Ord [a], and then add the
1761 superclasses of C [a] to avails. But we must not overwrite the binding
1762 for Ord [a] (which is obtained from Ord a) with a superclass selection or we'll just
1765 Here's another variant, immortalised in tcrun020
1766 class Monad m => C1 m
1767 class C1 m => C2 m x
1768 instance C2 Maybe Bool
1769 For the instance decl we need to build (C1 Maybe), and it's no good if
1770 we run around and add (C2 Maybe Bool) and its superclasses to the avails
1771 before we search for C1 Maybe.
1773 Here's another example
1774 class Eq b => Foo a b
1775 instance Eq a => Foo [a] a
1779 we'll first deduce that it holds (via the instance decl). We must not
1780 then overwrite the Eq t constraint with a superclass selection!
1782 At first I had a gross hack, whereby I simply did not add superclass constraints
1783 in addWanted, though I did for addGiven and addIrred. This was sub-optimal,
1784 becuase it lost legitimate superclass sharing, and it still didn't do the job:
1785 I found a very obscure program (now tcrun021) in which improvement meant the
1786 simplifier got two bites a the cherry... so something seemed to be an Stop
1787 first time, but reducible next time.
1789 Now we implement the Right Solution, which is to check for loops directly
1790 when adding superclasses. It's a bit like the occurs check in unification.
1793 Note [RECURSIVE DICTIONARIES]
1794 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1796 data D r = ZeroD | SuccD (r (D r));
1798 instance (Eq (r (D r))) => Eq (D r) where
1799 ZeroD == ZeroD = True
1800 (SuccD a) == (SuccD b) = a == b
1803 equalDC :: D [] -> D [] -> Bool;
1806 We need to prove (Eq (D [])). Here's how we go:
1810 by instance decl, holds if
1814 by instance decl of Eq, holds if
1816 where d2 = dfEqList d3
1819 But now we can "tie the knot" to give
1825 and it'll even run! The trick is to put the thing we are trying to prove
1826 (in this case Eq (D []) into the database before trying to prove its
1827 contributing clauses.
1830 %************************************************************************
1832 Reducing a single constraint
1834 %************************************************************************
1837 ---------------------------------------------
1838 reduceInst :: RedEnv -> Avails -> Inst -> TcM (Avails, LookupInstResult)
1839 reduceInst env avails (ImplicInst { tci_tyvars = tvs, tci_reft = reft, tci_loc = loc,
1840 tci_given = extra_givens, tci_wanted = wanteds })
1841 = reduceImplication env avails reft tvs extra_givens wanteds loc
1843 reduceInst env avails other_inst
1844 = do { result <- lookupSimpleInst other_inst
1845 ; return (avails, result) }
1849 ---------------------------------------------
1850 reduceImplication :: RedEnv
1852 -> Refinement -- May refine the givens; often empty
1853 -> [TcTyVar] -- Quantified type variables; all skolems
1854 -> [Inst] -- Extra givens; all rigid
1857 -> TcM (Avails, LookupInstResult)
1860 Suppose we are simplifying the constraint
1861 forall bs. extras => wanted
1862 in the context of an overall simplification problem with givens 'givens',
1863 and refinment 'reft'.
1866 * The refinement is often empty
1868 * The 'extra givens' need not mention any of the quantified type variables
1869 e.g. forall {}. Eq a => Eq [a]
1870 forall {}. C Int => D (Tree Int)
1872 This happens when you have something like
1874 T1 :: Eq a => a -> T a
1877 f x = ...(case x of { T1 v -> v==v })...
1880 -- ToDo: should we instantiate tvs? I think it's not necessary
1882 -- ToDo: what about improvement? There may be some improvement
1883 -- exposed as a result of the simplifications done by reduceList
1884 -- which are discarded if we back off.
1885 -- This is almost certainly Wrong, but we'll fix it when dealing
1886 -- better with equality constraints
1887 reduceImplication env orig_avails reft tvs extra_givens wanteds inst_loc
1888 = do { -- Add refined givens, and the extra givens
1889 (refined_red_givens, avails)
1890 <- if isEmptyRefinement reft then return (red_givens env, orig_avails)
1891 else foldlM (addRefinedGiven reft) ([], orig_avails) (red_givens env)
1892 ; avails <- foldlM addGiven avails extra_givens
1894 -- Solve the sub-problem
1895 ; let try_me inst = ReduceMe AddSCs -- Note [Freeness and implications]
1896 env' = env { red_givens = refined_red_givens ++ extra_givens
1897 , red_try_me = try_me }
1899 ; traceTc (text "reduceImplication" <+> vcat
1901 ppr (red_givens env), ppr extra_givens,
1902 ppr reft, ppr wanteds, ppr avails ])
1903 ; avails <- reduceList env' wanteds avails
1905 -- Extract the binding
1906 ; (binds, irreds) <- extractResults avails wanteds
1908 ; traceTc (text "reduceImplication result" <+> vcat
1909 [ ppr irreds, ppr binds])
1911 -- We always discard the extra avails we've generated;
1912 -- but we remember if we have done any (global) improvement
1913 ; let ret_avails = updateImprovement orig_avails avails
1915 ; if isEmptyLHsBinds binds then -- No progress
1916 return (ret_avails, NoInstance)
1918 { (implic_insts, bind) <- makeImplicationBind inst_loc tvs reft extra_givens irreds
1920 ; let dict_ids = map instToId extra_givens
1921 co = mkWpTyLams tvs <.> mkWpLams dict_ids <.> WpLet (binds `unionBags` bind)
1922 rhs = mkHsWrap co payload
1923 loc = instLocSpan inst_loc
1924 payload | [wanted] <- wanteds = HsVar (instToId wanted)
1925 | otherwise = ExplicitTuple (map (L loc . HsVar . instToId) wanteds) Boxed
1927 -- If there are any irreds, we back off and return NoInstance
1928 ; return (ret_avails, GenInst implic_insts (L loc rhs))
1932 Note [Freeness and implications]
1933 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1934 It's hard to say when an implication constraint can be floated out. Consider
1935 forall {} Eq a => Foo [a]
1936 The (Foo [a]) doesn't mention any of the quantified variables, but it
1937 still might be partially satisfied by the (Eq a).
1939 There is a useful special case when it *is* easy to partition the
1940 constraints, namely when there are no 'givens'. Consider
1941 forall {a}. () => Bar b
1942 There are no 'givens', and so there is no reason to capture (Bar b).
1943 We can let it float out. But if there is even one constraint we
1944 must be much more careful:
1945 forall {a}. C a b => Bar (m b)
1946 because (C a b) might have a superclass (D b), from which we might
1947 deduce (Bar [b]) when m later gets instantiated to []. Ha!
1949 Here is an even more exotic example
1951 Now consider the constraint
1952 forall b. D Int b => C Int
1953 We can satisfy the (C Int) from the superclass of D, so we don't want
1954 to float the (C Int) out, even though it mentions no type variable in
1957 Note [Pruning the givens in an implication constraint]
1958 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1959 Suppose we are about to form the implication constraint
1960 forall tvs. Eq a => Ord b
1961 The (Eq a) cannot contribute to the (Ord b), because it has no access to
1962 the type variable 'b'. So we could filter out the (Eq a) from the givens.
1964 Doing so would be a bit tidier, but all the implication constraints get
1965 simplified away by the optimiser, so it's no great win. So I don't take
1966 advantage of that at the moment.
1968 If you do, BE CAREFUL of wobbly type variables.
1971 %************************************************************************
1973 Avails and AvailHow: the pool of evidence
1975 %************************************************************************
1979 data Avails = Avails !ImprovementDone !AvailEnv
1981 type ImprovementDone = Bool -- True <=> some unification has happened
1982 -- so some Irreds might now be reducible
1983 -- keys that are now
1985 type AvailEnv = FiniteMap Inst AvailHow
1987 = IsIrred TcId -- Used for irreducible dictionaries,
1988 -- which are going to be lambda bound
1990 | Given TcId -- Used for dictionaries for which we have a binding
1991 -- e.g. those "given" in a signature
1993 | Rhs -- Used when there is a RHS
1994 (LHsExpr TcId) -- The RHS
1995 [Inst] -- Insts free in the RHS; we need these too
1997 instance Outputable Avails where
2000 pprAvails (Avails imp avails)
2001 = vcat [ ptext SLIT("Avails") <> (if imp then ptext SLIT("[improved]") else empty)
2002 , nest 2 (vcat [sep [ppr inst, nest 2 (equals <+> ppr avail)]
2003 | (inst,avail) <- fmToList avails ])]
2005 instance Outputable AvailHow where
2008 -------------------------
2009 pprAvail :: AvailHow -> SDoc
2010 pprAvail (IsIrred x) = text "Irred" <+> ppr x
2011 pprAvail (Given x) = text "Given" <+> ppr x
2012 pprAvail (Rhs rhs bs) = text "Rhs" <+> ppr rhs <+> braces (ppr bs)
2014 -------------------------
2015 extendAvailEnv :: AvailEnv -> Inst -> AvailHow -> AvailEnv
2016 extendAvailEnv env inst avail = addToFM env inst avail
2018 findAvailEnv :: AvailEnv -> Inst -> Maybe AvailHow
2019 findAvailEnv env wanted = lookupFM env wanted
2020 -- NB 1: the Ord instance of Inst compares by the class/type info
2021 -- *not* by unique. So
2022 -- d1::C Int == d2::C Int
2024 emptyAvails :: Avails
2025 emptyAvails = Avails False emptyFM
2027 findAvail :: Avails -> Inst -> Maybe AvailHow
2028 findAvail (Avails _ avails) wanted = findAvailEnv avails wanted
2030 elemAvails :: Inst -> Avails -> Bool
2031 elemAvails wanted (Avails _ avails) = wanted `elemFM` avails
2033 extendAvails :: Avails -> Inst -> AvailHow -> TcM Avails
2035 extendAvails avails@(Avails imp env) inst avail
2036 = do { imp1 <- tcImproveOne avails inst -- Do any improvement
2037 ; return (Avails (imp || imp1) (extendAvailEnv env inst avail)) }
2039 availsInsts :: Avails -> [Inst]
2040 availsInsts (Avails _ avails) = keysFM avails
2042 availsImproved (Avails imp _) = imp
2044 updateImprovement :: Avails -> Avails -> Avails
2045 -- (updateImprovement a1 a2) sets a1's improvement flag from a2
2046 updateImprovement (Avails _ avails1) (Avails imp2 _) = Avails imp2 avails1
2049 Extracting the bindings from a bunch of Avails.
2050 The bindings do *not* come back sorted in dependency order.
2051 We assume that they'll be wrapped in a big Rec, so that the
2052 dependency analyser can sort them out later
2055 extractResults :: Avails
2057 -> TcM ( TcDictBinds, -- Bindings
2058 [Inst]) -- Irreducible ones
2060 extractResults (Avails _ avails) wanteds
2061 = go avails emptyBag [] wanteds
2063 go :: AvailEnv -> TcDictBinds -> [Inst] -> [Inst]
2064 -> TcM (TcDictBinds, [Inst])
2065 go avails binds irreds []
2066 = returnM (binds, irreds)
2068 go avails binds irreds (w:ws)
2069 = case findAvailEnv avails w of
2070 Nothing -> pprTrace "Urk: extractResults" (ppr w) $
2071 go avails binds irreds ws
2074 | id == w_id -> go avails binds irreds ws
2075 | otherwise -> go avails (addBind binds w_id (nlHsVar id)) irreds ws
2076 -- The sought Id can be one of the givens, via a superclass chain
2077 -- and then we definitely don't want to generate an x=x binding!
2080 | id == w_id -> go (add_given avails w) binds (w:irreds) ws
2081 | otherwise -> go avails (addBind binds w_id (nlHsVar id)) irreds ws
2082 -- The add_given handles the case where we want (Ord a, Eq a), and we
2083 -- don't want to emit *two* Irreds for Ord a, one via the superclass chain
2084 -- This showed up in a dupliated Ord constraint in the error message for
2087 Just (Rhs rhs ws') -> go (add_given avails w) new_binds irreds (ws' ++ ws)
2089 new_binds = addBind binds w_id rhs
2093 add_given avails w = extendAvailEnv avails w (Given (instToId w))
2094 -- Don't add the same binding twice
2096 addBind binds id rhs = binds `unionBags` unitBag (L (getSrcSpan id) (VarBind id rhs))
2100 Note [No superclasses for Stop]
2101 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2102 When we decide not to reduce an Inst -- the 'WhatToDo' --- we still
2103 add it to avails, so that any other equal Insts will be commoned up
2104 right here. However, we do *not* add superclasses. If we have
2107 but a is not bound here, then we *don't* want to derive dn from df
2108 here lest we lose sharing.
2111 addWanted :: WantSCs -> Avails -> Inst -> LHsExpr TcId -> [Inst] -> TcM Avails
2112 addWanted want_scs avails wanted rhs_expr wanteds
2113 = addAvailAndSCs want_scs avails wanted avail
2115 avail = Rhs rhs_expr wanteds
2117 addGiven :: Avails -> Inst -> TcM Avails
2118 addGiven avails given = addAvailAndSCs AddSCs avails given (Given (instToId given))
2119 -- Always add superclasses for 'givens'
2121 -- No ASSERT( not (given `elemAvails` avails) ) because in an instance
2122 -- decl for Ord t we can add both Ord t and Eq t as 'givens',
2123 -- so the assert isn't true
2125 addRefinedGiven :: Refinement -> ([Inst], Avails) -> Inst -> TcM ([Inst], Avails)
2126 addRefinedGiven reft (refined_givens, avails) given
2127 | isDict given -- We sometimes have 'given' methods, but they
2128 -- are always optional, so we can drop them
2129 , let pred = dictPred given
2130 , isRefineablePred pred -- See Note [ImplicInst rigidity]
2131 , Just (co, pred) <- refinePred reft pred
2132 = do { new_given <- newDictBndr (instLoc given) pred
2133 ; let rhs = L (instSpan given) $
2134 HsWrap (WpCo co) (HsVar (instToId given))
2135 ; avails <- addAvailAndSCs AddSCs avails new_given (Rhs rhs [given])
2136 ; return (new_given:refined_givens, avails) }
2137 -- ToDo: the superclasses of the original given all exist in Avails
2138 -- so we could really just cast them, but it's more awkward to do,
2139 -- and hopefully the optimiser will spot the duplicated work
2141 = return (refined_givens, avails)
2144 Note [ImplicInst rigidity]
2145 ~~~~~~~~~~~~~~~~~~~~~~~~~~
2147 C :: forall ab. (Eq a, Ord b) => b -> T a
2149 ...(case x of C v -> <body>)...
2151 From the case (where x::T ty) we'll get an implication constraint
2152 forall b. (Eq ty, Ord b) => <body-constraints>
2153 Now suppose <body-constraints> itself has an implication constraint
2155 forall c. <reft> => <payload>
2156 Then, we can certainly apply the refinement <reft> to the Ord b, becuase it is
2157 existential, but we probably should not apply it to the (Eq ty) because it may
2158 be wobbly. Hence the isRigidInst
2160 @Insts@ are ordered by their class/type info, rather than by their
2161 unique. This allows the context-reduction mechanism to use standard finite
2162 maps to do their stuff. It's horrible that this code is here, rather
2163 than with the Avails handling stuff in TcSimplify
2166 addIrred :: WantSCs -> Avails -> Inst -> TcM Avails
2167 addIrred want_scs avails irred = ASSERT2( not (irred `elemAvails` avails), ppr irred $$ ppr avails )
2168 addAvailAndSCs want_scs avails irred (IsIrred (instToId irred))
2170 addAvailAndSCs :: WantSCs -> Avails -> Inst -> AvailHow -> TcM Avails
2171 addAvailAndSCs want_scs avails inst avail
2172 | not (isClassDict inst) = extendAvails avails inst avail
2173 | NoSCs <- want_scs = extendAvails avails inst avail
2174 | otherwise = do { traceTc (text "addAvailAndSCs" <+> vcat [ppr inst, ppr deps])
2175 ; avails' <- extendAvails avails inst avail
2176 ; addSCs is_loop avails' inst }
2178 is_loop pred = any (`tcEqType` mkPredTy pred) dep_tys
2179 -- Note: this compares by *type*, not by Unique
2180 deps = findAllDeps (unitVarSet (instToId inst)) avail
2181 dep_tys = map idType (varSetElems deps)
2183 findAllDeps :: IdSet -> AvailHow -> IdSet
2184 -- Find all the Insts that this one depends on
2185 -- See Note [SUPERCLASS-LOOP 2]
2186 -- Watch out, though. Since the avails may contain loops
2187 -- (see Note [RECURSIVE DICTIONARIES]), so we need to track the ones we've seen so far
2188 findAllDeps so_far (Rhs _ kids) = foldl find_all so_far kids
2189 findAllDeps so_far other = so_far
2191 find_all :: IdSet -> Inst -> IdSet
2193 | kid_id `elemVarSet` so_far = so_far
2194 | Just avail <- findAvail avails kid = findAllDeps so_far' avail
2195 | otherwise = so_far'
2197 so_far' = extendVarSet so_far kid_id -- Add the new kid to so_far
2198 kid_id = instToId kid
2200 addSCs :: (TcPredType -> Bool) -> Avails -> Inst -> TcM Avails
2201 -- Add all the superclasses of the Inst to Avails
2202 -- The first param says "dont do this because the original thing
2203 -- depends on this one, so you'd build a loop"
2204 -- Invariant: the Inst is already in Avails.
2206 addSCs is_loop avails dict
2207 = ASSERT( isDict dict )
2208 do { sc_dicts <- newDictBndrs (instLoc dict) sc_theta'
2209 ; foldlM add_sc avails (zipEqual "add_scs" sc_dicts sc_sels) }
2211 (clas, tys) = getDictClassTys dict
2212 (tyvars, sc_theta, sc_sels, _) = classBigSig clas
2213 sc_theta' = substTheta (zipTopTvSubst tyvars tys) sc_theta
2215 add_sc avails (sc_dict, sc_sel)
2216 | is_loop (dictPred sc_dict) = return avails -- See Note [SUPERCLASS-LOOP 2]
2217 | is_given sc_dict = return avails
2218 | otherwise = do { avails' <- extendAvails avails sc_dict (Rhs sc_sel_rhs [dict])
2219 ; addSCs is_loop avails' sc_dict }
2221 sc_sel_rhs = L (instSpan dict) (HsWrap co_fn (HsVar sc_sel))
2222 co_fn = WpApp (instToId dict) <.> mkWpTyApps tys
2224 is_given :: Inst -> Bool
2225 is_given sc_dict = case findAvail avails sc_dict of
2226 Just (Given _) -> True -- Given is cheaper than superclass selection
2230 %************************************************************************
2232 \section{tcSimplifyTop: defaulting}
2234 %************************************************************************
2237 @tcSimplifyTop@ is called once per module to simplify all the constant
2238 and ambiguous Insts.
2240 We need to be careful of one case. Suppose we have
2242 instance Num a => Num (Foo a b) where ...
2244 and @tcSimplifyTop@ is given a constraint (Num (Foo x y)). Then it'll simplify
2245 to (Num x), and default x to Int. But what about y??
2247 It's OK: the final zonking stage should zap y to (), which is fine.
2251 tcSimplifyTop, tcSimplifyInteractive :: [Inst] -> TcM TcDictBinds
2252 tcSimplifyTop wanteds
2253 = tc_simplify_top doc False wanteds
2255 doc = text "tcSimplifyTop"
2257 tcSimplifyInteractive wanteds
2258 = tc_simplify_top doc True wanteds
2260 doc = text "tcSimplifyInteractive"
2262 -- The TcLclEnv should be valid here, solely to improve
2263 -- error message generation for the monomorphism restriction
2264 tc_simplify_top doc interactive wanteds
2265 = do { dflags <- getDOpts
2266 ; wanteds <- mapM zonkInst wanteds
2267 ; mapM_ zonkTopTyVar (varSetElems (tyVarsOfInsts wanteds))
2269 ; (irreds1, binds1) <- tryHardCheckLoop doc1 wanteds
2270 ; (irreds2, binds2) <- approximateImplications doc2 (\d -> True) irreds1
2272 -- Use the defaulting rules to do extra unification
2273 -- NB: irreds2 are already zonked
2274 ; (irreds3, binds3) <- disambiguate doc3 interactive dflags irreds2
2276 -- Deal with implicit parameters
2277 ; let (bad_ips, non_ips) = partition isIPDict irreds3
2278 (ambigs, others) = partition isTyVarDict non_ips
2280 ; topIPErrs bad_ips -- Can arise from f :: Int -> Int
2282 ; addNoInstanceErrs others
2283 ; addTopAmbigErrs ambigs
2285 ; return (binds1 `unionBags` binds2 `unionBags` binds3) }
2287 doc1 = doc <+> ptext SLIT("(first round)")
2288 doc2 = doc <+> ptext SLIT("(approximate)")
2289 doc3 = doc <+> ptext SLIT("(disambiguate)")
2292 If a dictionary constrains a type variable which is
2293 * not mentioned in the environment
2294 * and not mentioned in the type of the expression
2295 then it is ambiguous. No further information will arise to instantiate
2296 the type variable; nor will it be generalised and turned into an extra
2297 parameter to a function.
2299 It is an error for this to occur, except that Haskell provided for
2300 certain rules to be applied in the special case of numeric types.
2302 * at least one of its classes is a numeric class, and
2303 * all of its classes are numeric or standard
2304 then the type variable can be defaulted to the first type in the
2305 default-type list which is an instance of all the offending classes.
2307 So here is the function which does the work. It takes the ambiguous
2308 dictionaries and either resolves them (producing bindings) or
2309 complains. It works by splitting the dictionary list by type
2310 variable, and using @disambigOne@ to do the real business.
2312 @disambigOne@ assumes that its arguments dictionaries constrain all
2313 the same type variable.
2315 ADR Comment 20/6/94: I've changed the @CReturnable@ case to default to
2316 @()@ instead of @Int@. I reckon this is the Right Thing to do since
2317 the most common use of defaulting is code like:
2319 _ccall_ foo `seqPrimIO` bar
2321 Since we're not using the result of @foo@, the result if (presumably)
2325 disambiguate :: SDoc -> Bool -> DynFlags -> [Inst] -> TcM ([Inst], TcDictBinds)
2326 -- Just does unification to fix the default types
2327 -- The Insts are assumed to be pre-zonked
2328 disambiguate doc interactive dflags insts
2330 = return (insts, emptyBag)
2332 | null defaultable_groups
2333 = do { traceTc (text "disambigutate" <+> vcat [ppr unaries, ppr bad_tvs, ppr defaultable_groups])
2334 ; return (insts, emptyBag) }
2337 = do { -- Figure out what default types to use
2338 default_tys <- getDefaultTys extended_defaulting ovl_strings
2340 ; traceTc (text "disambigutate" <+> vcat [ppr unaries, ppr bad_tvs, ppr defaultable_groups])
2341 ; mapM_ (disambigGroup default_tys) defaultable_groups
2343 -- disambigGroup does unification, hence try again
2344 ; tryHardCheckLoop doc insts }
2347 extended_defaulting = interactive || dopt Opt_ExtendedDefaultRules dflags
2348 ovl_strings = dopt Opt_OverloadedStrings dflags
2350 unaries :: [(Inst, Class, TcTyVar)] -- (C tv) constraints
2351 bad_tvs :: TcTyVarSet -- Tyvars mentioned by *other* constraints
2352 (unaries, bad_tvs_s) = partitionWith find_unary insts
2353 bad_tvs = unionVarSets bad_tvs_s
2355 -- Finds unary type-class constraints
2356 find_unary d@(Dict {tci_pred = ClassP cls [ty]})
2357 | Just tv <- tcGetTyVar_maybe ty = Left (d,cls,tv)
2358 find_unary inst = Right (tyVarsOfInst inst)
2360 -- Group by type variable
2361 defaultable_groups :: [[(Inst,Class,TcTyVar)]]
2362 defaultable_groups = filter defaultable_group (equivClasses cmp_tv unaries)
2363 cmp_tv (_,_,tv1) (_,_,tv2) = tv1 `compare` tv2
2365 defaultable_group :: [(Inst,Class,TcTyVar)] -> Bool
2366 defaultable_group ds@((_,_,tv):_)
2367 = isTyConableTyVar tv -- Note [Avoiding spurious errors]
2368 && not (tv `elemVarSet` bad_tvs)
2369 && defaultable_classes [c | (_,c,_) <- ds]
2370 defaultable_group [] = panic "defaultable_group"
2372 defaultable_classes clss
2373 | extended_defaulting = any isInteractiveClass clss
2374 | otherwise = all is_std_class clss && (any is_num_class clss)
2376 -- In interactive mode, or with -fextended-default-rules,
2377 -- we default Show a to Show () to avoid graututious errors on "show []"
2378 isInteractiveClass cls
2379 = is_num_class cls || (classKey cls `elem` [showClassKey, eqClassKey, ordClassKey])
2381 is_num_class cls = isNumericClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2382 -- is_num_class adds IsString to the standard numeric classes,
2383 -- when -foverloaded-strings is enabled
2385 is_std_class cls = isStandardClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2386 -- Similarly is_std_class
2388 -----------------------
2389 disambigGroup :: [Type] -- The default types
2390 -> [(Inst,Class,TcTyVar)] -- All standard classes of form (C a)
2391 -> TcM () -- Just does unification, to fix the default types
2393 disambigGroup default_tys dicts
2394 = try_default default_tys
2396 (_,_,tyvar) = head dicts -- Should be non-empty
2397 classes = [c | (_,c,_) <- dicts]
2399 try_default [] = return ()
2400 try_default (default_ty : default_tys)
2401 = tryTcLIE_ (try_default default_tys) $
2402 do { tcSimplifyDefault [mkClassPred clas [default_ty] | clas <- classes]
2403 -- This may fail; then the tryTcLIE_ kicks in
2404 -- Failure here is caused by there being no type in the
2405 -- default list which can satisfy all the ambiguous classes.
2406 -- For example, if Real a is reqd, but the only type in the
2407 -- default list is Int.
2409 -- After this we can't fail
2410 ; warnDefault dicts default_ty
2411 ; unifyType default_ty (mkTyVarTy tyvar) }
2414 -----------------------
2415 getDefaultTys :: Bool -> Bool -> TcM [Type]
2416 getDefaultTys extended_deflts ovl_strings
2417 = do { mb_defaults <- getDeclaredDefaultTys
2418 ; case mb_defaults of {
2419 Just tys -> return tys ; -- User-supplied defaults
2422 -- No use-supplied default
2423 -- Use [Integer, Double], plus modifications
2424 { integer_ty <- tcMetaTy integerTyConName
2425 ; checkWiredInTyCon doubleTyCon
2426 ; string_ty <- tcMetaTy stringTyConName
2427 ; return (opt_deflt extended_deflts unitTy
2428 -- Note [Default unitTy]
2430 [integer_ty,doubleTy]
2432 opt_deflt ovl_strings string_ty) } } }
2434 opt_deflt True ty = [ty]
2435 opt_deflt False ty = []
2438 Note [Default unitTy]
2439 ~~~~~~~~~~~~~~~~~~~~~
2440 In interative mode (or with -fextended-default-rules) we add () as the first type we
2441 try when defaulting. This has very little real impact, except in the following case.
2443 Text.Printf.printf "hello"
2444 This has type (forall a. IO a); it prints "hello", and returns 'undefined'. We don't
2445 want the GHCi repl loop to try to print that 'undefined'. The neatest thing is to
2446 default the 'a' to (), rather than to Integer (which is what would otherwise happen;
2447 and then GHCi doesn't attempt to print the (). So in interactive mode, we add
2448 () to the list of defaulting types. See Trac #1200.
2450 Note [Avoiding spurious errors]
2451 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2452 When doing the unification for defaulting, we check for skolem
2453 type variables, and simply don't default them. For example:
2454 f = (*) -- Monomorphic
2455 g :: Num a => a -> a
2457 Here, we get a complaint when checking the type signature for g,
2458 that g isn't polymorphic enough; but then we get another one when
2459 dealing with the (Num a) context arising from f's definition;
2460 we try to unify a with Int (to default it), but find that it's
2461 already been unified with the rigid variable from g's type sig
2464 %************************************************************************
2466 \subsection[simple]{@Simple@ versions}
2468 %************************************************************************
2470 Much simpler versions when there are no bindings to make!
2472 @tcSimplifyThetas@ simplifies class-type constraints formed by
2473 @deriving@ declarations and when specialising instances. We are
2474 only interested in the simplified bunch of class/type constraints.
2476 It simplifies to constraints of the form (C a b c) where
2477 a,b,c are type variables. This is required for the context of
2478 instance declarations.
2481 tcSimplifyDeriv :: InstOrigin
2483 -> ThetaType -- Wanted
2484 -> TcM ThetaType -- Needed
2485 -- Given instance (wanted) => C inst_ty
2486 -- Simplify 'wanted' as much as possible
2487 -- The inst_ty is needed only for the termination check
2489 tcSimplifyDeriv orig tyvars theta
2490 = do { (tvs, _, tenv) <- tcInstTyVars tyvars
2491 -- The main loop may do unification, and that may crash if
2492 -- it doesn't see a TcTyVar, so we have to instantiate. Sigh
2493 -- ToDo: what if two of them do get unified?
2494 ; wanteds <- newDictBndrsO orig (substTheta tenv theta)
2495 ; (irreds, _) <- tryHardCheckLoop doc wanteds
2497 ; let (tv_dicts, others) = partition isTyVarDict irreds
2498 ; addNoInstanceErrs others
2500 ; let rev_env = zipTopTvSubst tvs (mkTyVarTys tyvars)
2501 simpl_theta = substTheta rev_env (map dictPred tv_dicts)
2502 -- This reverse-mapping is a pain, but the result
2503 -- should mention the original TyVars not TcTyVars
2505 ; return simpl_theta }
2507 doc = ptext SLIT("deriving classes for a data type")
2510 Note [Exotic derived instance contexts]
2511 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2513 data T a b c = MkT (Foo a b c) deriving( Eq )
2514 instance (C Int a, Eq b, Eq c) => Eq (Foo a b c)
2516 Notice that this instance (just) satisfies the Paterson termination
2517 conditions. Then we *could* derive an instance decl like this:
2519 instance (C Int a, Eq b, Eq c) => Eq (T a b c)
2521 even though there is no instance for (C Int a), because there just
2522 *might* be an instance for, say, (C Int Bool) at a site where we
2523 need the equality instance for T's.
2525 However, this seems pretty exotic, and it's quite tricky to allow
2526 this, and yet give sensible error messages in the (much more common)
2527 case where we really want that instance decl for C.
2529 So for now we simply require that the derived instance context
2530 should have only type-variable constraints.
2532 Here is another example:
2533 data Fix f = In (f (Fix f)) deriving( Eq )
2534 Here, if we are prepared to allow -fallow-undecidable-instances we
2535 could derive the instance
2536 instance Eq (f (Fix f)) => Eq (Fix f)
2537 but this is so delicate that I don't think it should happen inside
2538 'deriving'. If you want this, write it yourself!
2540 NB: if you want to lift this condition, make sure you still meet the
2541 termination conditions! If not, the deriving mechanism generates
2542 larger and larger constraints. Example:
2544 data Seq a = Cons a (Seq (Succ a)) | Nil deriving Show
2546 Note the lack of a Show instance for Succ. First we'll generate
2547 instance (Show (Succ a), Show a) => Show (Seq a)
2549 instance (Show (Succ (Succ a)), Show (Succ a), Show a) => Show (Seq a)
2550 and so on. Instead we want to complain of no instance for (Show (Succ a)).
2553 @tcSimplifyDefault@ just checks class-type constraints, essentially;
2554 used with \tr{default} declarations. We are only interested in
2555 whether it worked or not.
2558 tcSimplifyDefault :: ThetaType -- Wanted; has no type variables in it
2561 tcSimplifyDefault theta
2562 = newDictBndrsO DefaultOrigin theta `thenM` \ wanteds ->
2563 tryHardCheckLoop doc wanteds `thenM` \ (irreds, _) ->
2564 addNoInstanceErrs irreds `thenM_`
2570 doc = ptext SLIT("default declaration")
2574 %************************************************************************
2576 \section{Errors and contexts}
2578 %************************************************************************
2580 ToDo: for these error messages, should we note the location as coming
2581 from the insts, or just whatever seems to be around in the monad just
2585 groupErrs :: ([Inst] -> TcM ()) -- Deal with one group
2586 -> [Inst] -- The offending Insts
2588 -- Group together insts with the same origin
2589 -- We want to report them together in error messages
2591 groupErrs report_err []
2593 groupErrs report_err (inst:insts)
2594 = do_one (inst:friends) `thenM_`
2595 groupErrs report_err others
2598 -- (It may seem a bit crude to compare the error messages,
2599 -- but it makes sure that we combine just what the user sees,
2600 -- and it avoids need equality on InstLocs.)
2601 (friends, others) = partition is_friend insts
2602 loc_msg = showSDoc (pprInstLoc (instLoc inst))
2603 is_friend friend = showSDoc (pprInstLoc (instLoc friend)) == loc_msg
2604 do_one insts = addInstCtxt (instLoc (head insts)) (report_err insts)
2605 -- Add location and context information derived from the Insts
2607 -- Add the "arising from..." part to a message about bunch of dicts
2608 addInstLoc :: [Inst] -> Message -> Message
2609 addInstLoc insts msg = msg $$ nest 2 (pprInstArising (head insts))
2611 addTopIPErrs :: [Name] -> [Inst] -> TcM ()
2612 addTopIPErrs bndrs []
2614 addTopIPErrs bndrs ips
2615 = do { dflags <- getDOpts
2616 ; addErrTcM (tidy_env, mk_msg dflags tidy_ips) }
2618 (tidy_env, tidy_ips) = tidyInsts ips
2620 = vcat [sep [ptext SLIT("Implicit parameters escape from"),
2621 nest 2 (ptext SLIT("the monomorphic top-level binding")
2622 <> plural bndrs <+> ptext SLIT("of")
2623 <+> pprBinders bndrs <> colon)],
2624 nest 2 (vcat (map ppr_ip ips)),
2625 monomorphism_fix dflags]
2626 ppr_ip ip = pprPred (dictPred ip) <+> pprInstArising ip
2628 topIPErrs :: [Inst] -> TcM ()
2630 = groupErrs report tidy_dicts
2632 (tidy_env, tidy_dicts) = tidyInsts dicts
2633 report dicts = addErrTcM (tidy_env, mk_msg dicts)
2634 mk_msg dicts = addInstLoc dicts (ptext SLIT("Unbound implicit parameter") <>
2635 plural tidy_dicts <+> pprDictsTheta tidy_dicts)
2637 addNoInstanceErrs :: [Inst] -- Wanted (can include implications)
2639 addNoInstanceErrs insts
2640 = do { let (tidy_env, tidy_insts) = tidyInsts insts
2641 ; reportNoInstances tidy_env Nothing tidy_insts }
2645 -> Maybe (InstLoc, [Inst]) -- Context
2646 -- Nothing => top level
2647 -- Just (d,g) => d describes the construct
2649 -> [Inst] -- What is wanted (can include implications)
2652 reportNoInstances tidy_env mb_what insts
2653 = groupErrs (report_no_instances tidy_env mb_what) insts
2655 report_no_instances tidy_env mb_what insts
2656 = do { inst_envs <- tcGetInstEnvs
2657 ; let (implics, insts1) = partition isImplicInst insts
2658 (insts2, overlaps) = partitionWith (check_overlap inst_envs) insts1
2659 ; traceTc (text "reportNoInstnces" <+> vcat
2660 [ppr implics, ppr insts1, ppr insts2])
2661 ; mapM_ complain_implic implics
2662 ; mapM_ (\doc -> addErrTcM (tidy_env, doc)) overlaps
2663 ; groupErrs complain_no_inst insts2 }
2665 complain_no_inst insts = addErrTcM (tidy_env, mk_no_inst_err insts)
2667 complain_implic inst -- Recurse!
2668 = reportNoInstances tidy_env
2669 (Just (tci_loc inst, tci_given inst))
2672 check_overlap :: (InstEnv,InstEnv) -> Inst -> Either Inst SDoc
2673 -- Right msg => overlap message
2674 -- Left inst => no instance
2675 check_overlap inst_envs wanted
2676 | not (isClassDict wanted) = Left wanted
2678 = case lookupInstEnv inst_envs clas tys of
2679 -- The case of exactly one match and no unifiers means
2680 -- a successful lookup. That can't happen here, becuase
2681 -- dicts only end up here if they didn't match in Inst.lookupInst
2683 ([m],[]) -> pprPanic "reportNoInstance" (ppr wanted)
2685 ([], _) -> Left wanted -- No match
2686 res -> Right (mk_overlap_msg wanted res)
2688 (clas,tys) = getDictClassTys wanted
2690 mk_overlap_msg dict (matches, unifiers)
2691 = ASSERT( not (null matches) )
2692 vcat [ addInstLoc [dict] ((ptext SLIT("Overlapping instances for")
2693 <+> pprPred (dictPred dict))),
2694 sep [ptext SLIT("Matching instances") <> colon,
2695 nest 2 (vcat [pprInstances ispecs, pprInstances unifiers])],
2696 if not (isSingleton matches)
2697 then -- Two or more matches
2699 else -- One match, plus some unifiers
2700 ASSERT( not (null unifiers) )
2701 parens (vcat [ptext SLIT("The choice depends on the instantiation of") <+>
2702 quotes (pprWithCommas ppr (varSetElems (tyVarsOfInst dict))),
2703 ptext SLIT("To pick the first instance above, use -fallow-incoherent-instances"),
2704 ptext SLIT("when compiling the other instances")])]
2706 ispecs = [ispec | (ispec, _) <- matches]
2708 mk_no_inst_err insts
2709 | null insts = empty
2711 | Just (loc, givens) <- mb_what, -- Nested (type signatures, instance decls)
2712 not (isEmptyVarSet (tyVarsOfInsts insts))
2713 = vcat [ addInstLoc insts $
2714 sep [ ptext SLIT("Could not deduce") <+> pprDictsTheta insts
2715 , nest 2 $ ptext SLIT("from the context") <+> pprDictsTheta givens]
2716 , show_fixes (fix1 loc : fixes2) ]
2718 | otherwise -- Top level
2719 = vcat [ addInstLoc insts $
2720 ptext SLIT("No instance") <> plural insts
2721 <+> ptext SLIT("for") <+> pprDictsTheta insts
2722 , show_fixes fixes2 ]
2725 fix1 loc = sep [ ptext SLIT("add") <+> pprDictsTheta insts
2726 <+> ptext SLIT("to the context of"),
2727 nest 2 (ppr (instLocOrigin loc)) ]
2728 -- I'm not sure it helps to add the location
2729 -- nest 2 (ptext SLIT("at") <+> ppr (instLocSpan loc)) ]
2731 fixes2 | null instance_dicts = []
2732 | otherwise = [sep [ptext SLIT("add an instance declaration for"),
2733 pprDictsTheta instance_dicts]]
2734 instance_dicts = [d | d <- insts, isClassDict d, not (isTyVarDict d)]
2735 -- Insts for which it is worth suggesting an adding an instance declaration
2736 -- Exclude implicit parameters, and tyvar dicts
2738 show_fixes :: [SDoc] -> SDoc
2739 show_fixes [] = empty
2740 show_fixes (f:fs) = sep [ptext SLIT("Possible fix:"),
2741 nest 2 (vcat (f : map (ptext SLIT("or") <+>) fs))]
2743 addTopAmbigErrs dicts
2744 -- Divide into groups that share a common set of ambiguous tyvars
2745 = ifErrsM (return ()) $ -- Only report ambiguity if no other errors happened
2746 -- See Note [Avoiding spurious errors]
2747 mapM_ report (equivClasses cmp [(d, tvs_of d) | d <- tidy_dicts])
2749 (tidy_env, tidy_dicts) = tidyInsts dicts
2751 tvs_of :: Inst -> [TcTyVar]
2752 tvs_of d = varSetElems (tyVarsOfInst d)
2753 cmp (_,tvs1) (_,tvs2) = tvs1 `compare` tvs2
2755 report :: [(Inst,[TcTyVar])] -> TcM ()
2756 report pairs@((inst,tvs) : _) -- The pairs share a common set of ambiguous tyvars
2757 = mkMonomorphismMsg tidy_env tvs `thenM` \ (tidy_env, mono_msg) ->
2758 setSrcSpan (instSpan inst) $
2759 -- the location of the first one will do for the err message
2760 addErrTcM (tidy_env, msg $$ mono_msg)
2762 dicts = map fst pairs
2763 msg = sep [text "Ambiguous type variable" <> plural tvs <+>
2764 pprQuotedList tvs <+> in_msg,
2765 nest 2 (pprDictsInFull dicts)]
2766 in_msg = text "in the constraint" <> plural dicts <> colon
2767 report [] = panic "addTopAmbigErrs"
2770 mkMonomorphismMsg :: TidyEnv -> [TcTyVar] -> TcM (TidyEnv, Message)
2771 -- There's an error with these Insts; if they have free type variables
2772 -- it's probably caused by the monomorphism restriction.
2773 -- Try to identify the offending variable
2774 -- ASSUMPTION: the Insts are fully zonked
2775 mkMonomorphismMsg tidy_env inst_tvs
2776 = do { dflags <- getDOpts
2777 ; (tidy_env, docs) <- findGlobals (mkVarSet inst_tvs) tidy_env
2778 ; return (tidy_env, mk_msg dflags docs) }
2780 mk_msg _ _ | any isRuntimeUnk inst_tvs
2781 = vcat [ptext SLIT("Cannot resolve unknown runtime types:") <+>
2782 (pprWithCommas ppr inst_tvs),
2783 ptext SLIT("Use :print or :force to determine these types")]
2784 mk_msg _ [] = ptext SLIT("Probable fix: add a type signature that fixes these type variable(s)")
2785 -- This happens in things like
2786 -- f x = show (read "foo")
2787 -- where monomorphism doesn't play any role
2789 = vcat [ptext SLIT("Possible cause: the monomorphism restriction applied to the following:"),
2791 monomorphism_fix dflags]
2793 isRuntimeUnk :: TcTyVar -> Bool
2794 isRuntimeUnk x | SkolemTv RuntimeUnkSkol <- tcTyVarDetails x = True
2797 monomorphism_fix :: DynFlags -> SDoc
2798 monomorphism_fix dflags
2799 = ptext SLIT("Probable fix:") <+> vcat
2800 [ptext SLIT("give these definition(s) an explicit type signature"),
2801 if dopt Opt_MonomorphismRestriction dflags
2802 then ptext SLIT("or use -fno-monomorphism-restriction")
2803 else empty] -- Only suggest adding "-fno-monomorphism-restriction"
2804 -- if it is not already set!
2806 warnDefault ups default_ty
2807 = doptM Opt_WarnTypeDefaults `thenM` \ warn_flag ->
2808 addInstCtxt (instLoc (head (dicts))) (warnTc warn_flag warn_msg)
2810 dicts = [d | (d,_,_) <- ups]
2813 (_, tidy_dicts) = tidyInsts dicts
2814 warn_msg = vcat [ptext SLIT("Defaulting the following constraint(s) to type") <+>
2815 quotes (ppr default_ty),
2816 pprDictsInFull tidy_dicts]
2818 reduceDepthErr n stack
2819 = vcat [ptext SLIT("Context reduction stack overflow; size =") <+> int n,
2820 ptext SLIT("Use -fcontext-stack=N to increase stack size to N"),
2821 nest 4 (pprStack stack)]
2823 pprStack stack = vcat (map pprInstInFull stack)