2 % (c) The University of Glasgow 2006
3 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
10 -- The above warning supression flag is a temporary kludge.
11 -- While working on this module you are encouraged to remove it and fix
12 -- any warnings in the module. See
13 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
17 tcSimplifyInfer, tcSimplifyInferCheck,
18 tcSimplifyCheck, tcSimplifyRestricted,
19 tcSimplifyRuleLhs, tcSimplifyIPs,
20 tcSimplifySuperClasses,
21 tcSimplifyTop, tcSimplifyInteractive,
22 tcSimplifyBracket, tcSimplifyCheckPat,
24 tcSimplifyDeriv, tcSimplifyDefault,
30 #include "HsVersions.h"
32 import {-# SOURCE #-} TcUnify( unifyType )
72 %************************************************************************
76 %************************************************************************
78 --------------------------------------
79 Notes on functional dependencies (a bug)
80 --------------------------------------
87 instance D a b => C a b -- Undecidable
88 -- (Not sure if it's crucial to this eg)
89 f :: C a b => a -> Bool
92 g :: C a b => a -> Bool
95 Here f typechecks, but g does not!! Reason: before doing improvement,
96 we reduce the (C a b1) constraint from the call of f to (D a b1).
98 Here is a more complicated example:
100 | > class Foo a b | a->b
102 | > class Bar a b | a->b
106 | > instance Bar Obj Obj
108 | > instance (Bar a b) => Foo a b
110 | > foo:: (Foo a b) => a -> String
113 | > runFoo:: (forall a b. (Foo a b) => a -> w) -> w
119 | Could not deduce (Bar a b) from the context (Foo a b)
120 | arising from use of `foo' at <interactive>:1
122 | Add (Bar a b) to the expected type of an expression
123 | In the first argument of `runFoo', namely `foo'
124 | In the definition of `it': it = runFoo foo
126 | Why all of the sudden does GHC need the constraint Bar a b? The
127 | function foo didn't ask for that...
129 The trouble is that to type (runFoo foo), GHC has to solve the problem:
131 Given constraint Foo a b
132 Solve constraint Foo a b'
134 Notice that b and b' aren't the same. To solve this, just do
135 improvement and then they are the same. But GHC currently does
140 That is usually fine, but it isn't here, because it sees that Foo a b is
141 not the same as Foo a b', and so instead applies the instance decl for
142 instance Bar a b => Foo a b. And that's where the Bar constraint comes
145 The Right Thing is to improve whenever the constraint set changes at
146 all. Not hard in principle, but it'll take a bit of fiddling to do.
148 Note [Choosing which variables to quantify]
149 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
150 Suppose we are about to do a generalisation step. We have in our hand
153 T the type of the RHS
154 C the constraints from that RHS
156 The game is to figure out
158 Q the set of type variables over which to quantify
159 Ct the constraints we will *not* quantify over
160 Cq the constraints we will quantify over
162 So we're going to infer the type
166 and float the constraints Ct further outwards.
168 Here are the things that *must* be true:
170 (A) Q intersect fv(G) = EMPTY limits how big Q can be
171 (B) Q superset fv(Cq union T) \ oclose(fv(G),C) limits how small Q can be
173 (A) says we can't quantify over a variable that's free in the environment.
174 (B) says we must quantify over all the truly free variables in T, else
175 we won't get a sufficiently general type.
177 We do not *need* to quantify over any variable that is fixed by the
178 free vars of the environment G.
180 BETWEEN THESE TWO BOUNDS, ANY Q WILL DO!
182 Example: class H x y | x->y where ...
184 fv(G) = {a} C = {H a b, H c d}
187 (A) Q intersect {a} is empty
188 (B) Q superset {a,b,c,d} \ oclose({a}, C) = {a,b,c,d} \ {a,b} = {c,d}
190 So Q can be {c,d}, {b,c,d}
192 In particular, it's perfectly OK to quantify over more type variables
193 than strictly necessary; there is no need to quantify over 'b', since
194 it is determined by 'a' which is free in the envt, but it's perfectly
195 OK to do so. However we must not quantify over 'a' itself.
197 Other things being equal, however, we'd like to quantify over as few
198 variables as possible: smaller types, fewer type applications, more
199 constraints can get into Ct instead of Cq. Here's a good way to
202 Q = grow( fv(T), C ) \ oclose( fv(G), C )
204 That is, quantify over all variable that that MIGHT be fixed by the
205 call site (which influences T), but which aren't DEFINITELY fixed by
206 G. This choice definitely quantifies over enough type variables,
207 albeit perhaps too many.
209 Why grow( fv(T), C ) rather than fv(T)? Consider
211 class H x y | x->y where ...
216 If we used fv(T) = {c} we'd get the type
218 forall c. H c d => c -> b
220 And then if the fn was called at several different c's, each of
221 which fixed d differently, we'd get a unification error, because
222 d isn't quantified. Solution: quantify d. So we must quantify
223 everything that might be influenced by c.
225 Why not oclose( fv(T), C )? Because we might not be able to see
226 all the functional dependencies yet:
228 class H x y | x->y where ...
229 instance H x y => Eq (T x y) where ...
234 Now oclose(fv(T),C) = {c}, because the functional dependency isn't
235 apparent yet, and that's wrong. We must really quantify over d too.
237 There really isn't any point in quantifying over any more than
238 grow( fv(T), C ), because the call sites can't possibly influence
239 any other type variables.
243 -------------------------------------
245 -------------------------------------
247 It's very hard to be certain when a type is ambiguous. Consider
251 instance H x y => K (x,y)
253 Is this type ambiguous?
254 forall a b. (K (a,b), Eq b) => a -> a
256 Looks like it! But if we simplify (K (a,b)) we get (H a b) and
257 now we see that a fixes b. So we can't tell about ambiguity for sure
258 without doing a full simplification. And even that isn't possible if
259 the context has some free vars that may get unified. Urgle!
261 Here's another example: is this ambiguous?
262 forall a b. Eq (T b) => a -> a
263 Not if there's an insance decl (with no context)
264 instance Eq (T b) where ...
266 You may say of this example that we should use the instance decl right
267 away, but you can't always do that:
269 class J a b where ...
270 instance J Int b where ...
272 f :: forall a b. J a b => a -> a
274 (Notice: no functional dependency in J's class decl.)
275 Here f's type is perfectly fine, provided f is only called at Int.
276 It's premature to complain when meeting f's signature, or even
277 when inferring a type for f.
281 However, we don't *need* to report ambiguity right away. It'll always
282 show up at the call site.... and eventually at main, which needs special
283 treatment. Nevertheless, reporting ambiguity promptly is an excellent thing.
285 So here's the plan. We WARN about probable ambiguity if
287 fv(Cq) is not a subset of oclose(fv(T) union fv(G), C)
289 (all tested before quantification).
290 That is, all the type variables in Cq must be fixed by the the variables
291 in the environment, or by the variables in the type.
293 Notice that we union before calling oclose. Here's an example:
295 class J a b c | a b -> c
299 forall b c. (J a b c) => b -> b
301 Only if we union {a} from G with {b} from T before using oclose,
302 do we see that c is fixed.
304 It's a bit vague exactly which C we should use for this oclose call. If we
305 don't fix enough variables we might complain when we shouldn't (see
306 the above nasty example). Nothing will be perfect. That's why we can
307 only issue a warning.
310 Can we ever be *certain* about ambiguity? Yes: if there's a constraint
312 c in C such that fv(c) intersect (fv(G) union fv(T)) = EMPTY
314 then c is a "bubble"; there's no way it can ever improve, and it's
315 certainly ambiguous. UNLESS it is a constant (sigh). And what about
320 instance H x y => K (x,y)
322 Is this type ambiguous?
323 forall a b. (K (a,b), Eq b) => a -> a
325 Urk. The (Eq b) looks "definitely ambiguous" but it isn't. What we are after
326 is a "bubble" that's a set of constraints
328 Cq = Ca union Cq' st fv(Ca) intersect (fv(Cq') union fv(T) union fv(G)) = EMPTY
330 Hence another idea. To decide Q start with fv(T) and grow it
331 by transitive closure in Cq (no functional dependencies involved).
332 Now partition Cq using Q, leaving the definitely-ambiguous and probably-ok.
333 The definitely-ambiguous can then float out, and get smashed at top level
334 (which squashes out the constants, like Eq (T a) above)
337 --------------------------------------
338 Notes on principal types
339 --------------------------------------
344 f x = let g y = op (y::Int) in True
346 Here the principal type of f is (forall a. a->a)
347 but we'll produce the non-principal type
348 f :: forall a. C Int => a -> a
351 --------------------------------------
352 The need for forall's in constraints
353 --------------------------------------
355 [Exchange on Haskell Cafe 5/6 Dec 2000]
357 class C t where op :: t -> Bool
358 instance C [t] where op x = True
360 p y = (let f :: c -> Bool; f x = op (y >> return x) in f, y ++ [])
361 q y = (y ++ [], let f :: c -> Bool; f x = op (y >> return x) in f)
363 The definitions of p and q differ only in the order of the components in
364 the pair on their right-hand sides. And yet:
366 ghc and "Typing Haskell in Haskell" reject p, but accept q;
367 Hugs rejects q, but accepts p;
368 hbc rejects both p and q;
369 nhc98 ... (Malcolm, can you fill in the blank for us!).
371 The type signature for f forces context reduction to take place, and
372 the results of this depend on whether or not the type of y is known,
373 which in turn depends on which component of the pair the type checker
376 Solution: if y::m a, float out the constraints
377 Monad m, forall c. C (m c)
378 When m is later unified with [], we can solve both constraints.
381 --------------------------------------
382 Notes on implicit parameters
383 --------------------------------------
385 Note [Inheriting implicit parameters]
386 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
391 where f is *not* a top-level binding.
392 From the RHS of f we'll get the constraint (?y::Int).
393 There are two types we might infer for f:
397 (so we get ?y from the context of f's definition), or
399 f :: (?y::Int) => Int -> Int
401 At first you might think the first was better, becuase then
402 ?y behaves like a free variable of the definition, rather than
403 having to be passed at each call site. But of course, the WHOLE
404 IDEA is that ?y should be passed at each call site (that's what
405 dynamic binding means) so we'd better infer the second.
407 BOTTOM LINE: when *inferring types* you *must* quantify
408 over implicit parameters. See the predicate isFreeWhenInferring.
411 Note [Implicit parameters and ambiguity]
412 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
413 Only a *class* predicate can give rise to ambiguity
414 An *implicit parameter* cannot. For example:
415 foo :: (?x :: [a]) => Int
417 is fine. The call site will suppply a particular 'x'
419 Furthermore, the type variables fixed by an implicit parameter
420 propagate to the others. E.g.
421 foo :: (Show a, ?x::[a]) => Int
423 The type of foo looks ambiguous. But it isn't, because at a call site
425 let ?x = 5::Int in foo
426 and all is well. In effect, implicit parameters are, well, parameters,
427 so we can take their type variables into account as part of the
428 "tau-tvs" stuff. This is done in the function 'FunDeps.grow'.
431 Question 2: type signatures
432 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
433 BUT WATCH OUT: When you supply a type signature, we can't force you
434 to quantify over implicit parameters. For example:
438 This is perfectly reasonable. We do not want to insist on
440 (?x + 1) :: (?x::Int => Int)
442 That would be silly. Here, the definition site *is* the occurrence site,
443 so the above strictures don't apply. Hence the difference between
444 tcSimplifyCheck (which *does* allow implicit paramters to be inherited)
445 and tcSimplifyCheckBind (which does not).
447 What about when you supply a type signature for a binding?
448 Is it legal to give the following explicit, user type
449 signature to f, thus:
454 At first sight this seems reasonable, but it has the nasty property
455 that adding a type signature changes the dynamic semantics.
458 (let f x = (x::Int) + ?y
459 in (f 3, f 3 with ?y=5)) with ?y = 6
465 in (f 3, f 3 with ?y=5)) with ?y = 6
469 Indeed, simply inlining f (at the Haskell source level) would change the
472 Nevertheless, as Launchbury says (email Oct 01) we can't really give the
473 semantics for a Haskell program without knowing its typing, so if you
474 change the typing you may change the semantics.
476 To make things consistent in all cases where we are *checking* against
477 a supplied signature (as opposed to inferring a type), we adopt the
480 a signature does not need to quantify over implicit params.
482 [This represents a (rather marginal) change of policy since GHC 5.02,
483 which *required* an explicit signature to quantify over all implicit
484 params for the reasons mentioned above.]
486 But that raises a new question. Consider
488 Given (signature) ?x::Int
489 Wanted (inferred) ?x::Int, ?y::Bool
491 Clearly we want to discharge the ?x and float the ?y out. But
492 what is the criterion that distinguishes them? Clearly it isn't
493 what free type variables they have. The Right Thing seems to be
494 to float a constraint that
495 neither mentions any of the quantified type variables
496 nor any of the quantified implicit parameters
498 See the predicate isFreeWhenChecking.
501 Question 3: monomorphism
502 ~~~~~~~~~~~~~~~~~~~~~~~~
503 There's a nasty corner case when the monomorphism restriction bites:
507 The argument above suggests that we *must* generalise
508 over the ?y parameter, to get
509 z :: (?y::Int) => Int,
510 but the monomorphism restriction says that we *must not*, giving
512 Why does the momomorphism restriction say this? Because if you have
514 let z = x + ?y in z+z
516 you might not expect the addition to be done twice --- but it will if
517 we follow the argument of Question 2 and generalise over ?y.
520 Question 4: top level
521 ~~~~~~~~~~~~~~~~~~~~~
522 At the top level, monomorhism makes no sense at all.
525 main = let ?x = 5 in print foo
529 woggle :: (?x :: Int) => Int -> Int
532 We definitely don't want (foo :: Int) with a top-level implicit parameter
533 (?x::Int) becuase there is no way to bind it.
538 (A) Always generalise over implicit parameters
539 Bindings that fall under the monomorphism restriction can't
543 * Inlining remains valid
544 * No unexpected loss of sharing
545 * But simple bindings like
547 will be rejected, unless you add an explicit type signature
548 (to avoid the monomorphism restriction)
549 z :: (?y::Int) => Int
551 This seems unacceptable
553 (B) Monomorphism restriction "wins"
554 Bindings that fall under the monomorphism restriction can't
556 Always generalise over implicit parameters *except* for bindings
557 that fall under the monomorphism restriction
560 * Inlining isn't valid in general
561 * No unexpected loss of sharing
562 * Simple bindings like
564 accepted (get value of ?y from binding site)
566 (C) Always generalise over implicit parameters
567 Bindings that fall under the monomorphism restriction can't
568 be generalised, EXCEPT for implicit parameters
570 * Inlining remains valid
571 * Unexpected loss of sharing (from the extra generalisation)
572 * Simple bindings like
574 accepted (get value of ?y from occurrence sites)
579 None of these choices seems very satisfactory. But at least we should
580 decide which we want to do.
582 It's really not clear what is the Right Thing To Do. If you see
586 would you expect the value of ?y to be got from the *occurrence sites*
587 of 'z', or from the valuue of ?y at the *definition* of 'z'? In the
588 case of function definitions, the answer is clearly the former, but
589 less so in the case of non-fucntion definitions. On the other hand,
590 if we say that we get the value of ?y from the definition site of 'z',
591 then inlining 'z' might change the semantics of the program.
593 Choice (C) really says "the monomorphism restriction doesn't apply
594 to implicit parameters". Which is fine, but remember that every
595 innocent binding 'x = ...' that mentions an implicit parameter in
596 the RHS becomes a *function* of that parameter, called at each
597 use of 'x'. Now, the chances are that there are no intervening 'with'
598 clauses that bind ?y, so a decent compiler should common up all
599 those function calls. So I think I strongly favour (C). Indeed,
600 one could make a similar argument for abolishing the monomorphism
601 restriction altogether.
603 BOTTOM LINE: we choose (B) at present. See tcSimplifyRestricted
607 %************************************************************************
609 \subsection{tcSimplifyInfer}
611 %************************************************************************
613 tcSimplify is called when we *inferring* a type. Here's the overall game plan:
615 1. Compute Q = grow( fvs(T), C )
617 2. Partition C based on Q into Ct and Cq. Notice that ambiguous
618 predicates will end up in Ct; we deal with them at the top level
620 3. Try improvement, using functional dependencies
622 4. If Step 3 did any unification, repeat from step 1
623 (Unification can change the result of 'grow'.)
625 Note: we don't reduce dictionaries in step 2. For example, if we have
626 Eq (a,b), we don't simplify to (Eq a, Eq b). So Q won't be different
627 after step 2. However note that we may therefore quantify over more
628 type variables than we absolutely have to.
630 For the guts, we need a loop, that alternates context reduction and
631 improvement with unification. E.g. Suppose we have
633 class C x y | x->y where ...
635 and tcSimplify is called with:
637 Then improvement unifies a with b, giving
640 If we need to unify anything, we rattle round the whole thing all over
647 -> TcTyVarSet -- fv(T); type vars
649 -> TcM ([TcTyVar], -- Tyvars to quantify (zonked and quantified)
650 [Inst], -- Dict Ids that must be bound here (zonked)
651 TcDictBinds) -- Bindings
652 -- Any free (escaping) Insts are tossed into the environment
657 tcSimplifyInfer doc tau_tvs wanted
658 = do { tau_tvs1 <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
659 ; wanted' <- mapM zonkInst wanted -- Zonk before deciding quantified tyvars
660 ; gbl_tvs <- tcGetGlobalTyVars
661 ; let preds1 = fdPredsOfInsts wanted'
662 gbl_tvs1 = oclose preds1 gbl_tvs
663 qtvs = grow preds1 tau_tvs1 `minusVarSet` gbl_tvs1
664 -- See Note [Choosing which variables to quantify]
666 -- To maximise sharing, remove from consideration any
667 -- constraints that don't mention qtvs at all
668 ; let (free, bound) = partition (isFreeWhenInferring qtvs) wanted'
671 -- To make types simple, reduce as much as possible
672 ; traceTc (text "infer" <+> (ppr preds1 $$ ppr (grow preds1 tau_tvs1) $$ ppr gbl_tvs $$
673 ppr gbl_tvs1 $$ ppr free $$ ppr bound))
674 ; (irreds1, binds1) <- tryHardCheckLoop doc bound
676 -- Note [Inference and implication constraints]
677 ; let want_dict d = tyVarsOfInst d `intersectsVarSet` qtvs
678 ; (irreds2, binds2) <- approximateImplications doc want_dict irreds1
680 -- Now work out all over again which type variables to quantify,
681 -- exactly in the same way as before, but starting from irreds2. Why?
682 -- a) By now improvment may have taken place, and we must *not*
683 -- quantify over any variable free in the environment
684 -- tc137 (function h inside g) is an example
686 -- b) Do not quantify over constraints that *now* do not
687 -- mention quantified type variables, because they are
688 -- simply ambiguous (or might be bound further out). Example:
689 -- f :: Eq b => a -> (a, b)
691 -- From the RHS of g we get the MethodInst f77 :: alpha -> (alpha, beta)
692 -- We decide to quantify over 'alpha' alone, but free1 does not include f77
693 -- because f77 mentions 'alpha'. Then reducing leaves only the (ambiguous)
694 -- constraint (Eq beta), which we dump back into the free set
695 -- See test tcfail181
697 -- c) irreds may contain type variables not previously mentioned,
698 -- e.g. instance D a x => Foo [a]
700 -- Then after simplifying we'll get (D a x), and x is fresh
701 -- We must quantify over x else it'll be totally unbound
702 ; tau_tvs2 <- zonkTcTyVarsAndFV (varSetElems tau_tvs1)
703 ; gbl_tvs2 <- zonkTcTyVarsAndFV (varSetElems gbl_tvs1)
704 -- Note that we start from gbl_tvs1
705 -- We use tcGetGlobalTyVars, then oclose wrt preds2, because
706 -- we've already put some of the original preds1 into frees
707 -- E.g. wanteds = C a b (where a->b)
710 -- Then b is fixed by gbl_tvs, so (C a b) will be in free, and
711 -- irreds2 will be empty. But we don't want to generalise over b!
712 ; let preds2 = fdPredsOfInsts irreds2 -- irreds2 is zonked
713 qtvs = grow preds2 tau_tvs2 `minusVarSet` oclose preds2 gbl_tvs2
714 ; let (free, irreds3) = partition (isFreeWhenInferring qtvs) irreds2
717 -- Turn the quantified meta-type variables into real type variables
718 ; qtvs2 <- zonkQuantifiedTyVars (varSetElems qtvs)
720 -- We can't abstract over any remaining unsolved
721 -- implications so instead just float them outwards. Ugh.
722 ; let (q_dicts0, implics) = partition isAbstractableInst irreds3
723 ; loc <- getInstLoc (ImplicOrigin doc)
724 ; implic_bind <- bindIrreds loc qtvs2 q_dicts0 implics
726 -- Prepare equality instances for quantification
727 ; let (q_eqs0,q_dicts) = partition isEqInst q_dicts0
728 ; q_eqs <- mapM finalizeEqInst q_eqs0
730 ; return (qtvs2, q_eqs ++ q_dicts, binds1 `unionBags` binds2 `unionBags` implic_bind) }
731 -- NB: when we are done, we might have some bindings, but
732 -- the final qtvs might be empty. See Note [NO TYVARS] below.
734 approximateImplications :: SDoc -> (Inst -> Bool) -> [Inst] -> TcM ([Inst], TcDictBinds)
735 -- Note [Inference and implication constraints]
736 -- Given a bunch of Dict and ImplicInsts, try to approximate the implications by
737 -- - fetching any dicts inside them that are free
738 -- - using those dicts as cruder constraints, to solve the implications
739 -- - returning the extra ones too
741 approximateImplications doc want_dict irreds
743 = return (irreds, emptyBag)
745 = do { extra_dicts' <- mapM cloneDict extra_dicts
746 ; tryHardCheckLoop doc (extra_dicts' ++ irreds) }
747 -- By adding extra_dicts', we make them
748 -- available to solve the implication constraints
750 extra_dicts = get_dicts (filter isImplicInst irreds)
752 get_dicts :: [Inst] -> [Inst] -- Returns only Dicts
753 -- Find the wanted constraints in implication constraints that satisfy
754 -- want_dict, and are not bound by forall's in the constraint itself
755 get_dicts ds = concatMap get_dict ds
757 get_dict d@(Dict {}) | want_dict d = [d]
759 get_dict (ImplicInst {tci_tyvars = tvs, tci_wanted = wanteds})
760 = [ d | let tv_set = mkVarSet tvs
761 , d <- get_dicts wanteds
762 , not (tyVarsOfInst d `intersectsVarSet` tv_set)]
763 get_dict i@(EqInst {}) | want_dict i = [i]
765 get_dict other = pprPanic "approximateImplications" (ppr other)
768 Note [Inference and implication constraints]
769 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
770 Suppose we have a wanted implication constraint (perhaps arising from
771 a nested pattern match) like
773 and we are now trying to quantify over 'a' when inferring the type for
774 a function. In principle it's possible that there might be an instance
775 instance (C a, E a) => D [a]
776 so the context (E a) would suffice. The Right Thing is to abstract over
777 the implication constraint, but we don't do that (a) because it'll be
778 surprising to programmers and (b) because we don't have the machinery to deal
779 with 'given' implications.
781 So our best approximation is to make (D [a]) part of the inferred
782 context, so we can use that to discharge the implication. Hence
783 the strange function get_dicts in approximateImplications.
785 The common cases are more clear-cut, when we have things like
787 Here, abstracting over (C b) is not an approximation at all -- but see
788 Note [Freeness and implications].
790 See Trac #1430 and test tc228.
794 -----------------------------------------------------------
795 -- tcSimplifyInferCheck is used when we know the constraints we are to simplify
796 -- against, but we don't know the type variables over which we are going to quantify.
797 -- This happens when we have a type signature for a mutually recursive group
800 -> TcTyVarSet -- fv(T)
803 -> TcM ([TyVar], -- Fully zonked, and quantified
804 TcDictBinds) -- Bindings
806 tcSimplifyInferCheck loc tau_tvs givens wanteds
807 = do { traceTc (text "tcSimplifyInferCheck <-" <+> ppr wanteds)
808 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
810 -- Figure out which type variables to quantify over
811 -- You might think it should just be the signature tyvars,
812 -- but in bizarre cases you can get extra ones
813 -- f :: forall a. Num a => a -> a
814 -- f x = fst (g (x, head [])) + 1
816 -- Here we infer g :: forall a b. a -> b -> (b,a)
817 -- We don't want g to be monomorphic in b just because
818 -- f isn't quantified over b.
819 ; let all_tvs = varSetElems (tau_tvs `unionVarSet` tyVarsOfInsts givens)
820 ; all_tvs <- zonkTcTyVarsAndFV all_tvs
821 ; gbl_tvs <- tcGetGlobalTyVars
822 ; let qtvs = varSetElems (all_tvs `minusVarSet` gbl_tvs)
823 -- We could close gbl_tvs, but its not necessary for
824 -- soundness, and it'll only affect which tyvars, not which
825 -- dictionaries, we quantify over
827 ; qtvs' <- zonkQuantifiedTyVars qtvs
829 -- Now we are back to normal (c.f. tcSimplCheck)
830 ; implic_bind <- bindIrreds loc qtvs' givens irreds
832 ; traceTc (text "tcSimplifyInferCheck ->" <+> ppr (implic_bind))
833 ; return (qtvs', binds `unionBags` implic_bind) }
836 Note [Squashing methods]
837 ~~~~~~~~~~~~~~~~~~~~~~~~~
838 Be careful if you want to float methods more:
839 truncate :: forall a. RealFrac a => forall b. Integral b => a -> b
840 From an application (truncate f i) we get
843 If we have also have a second occurrence of truncate, we get
846 When simplifying with i,f free, we might still notice that
847 t1=t3; but alas, the binding for t2 (which mentions t1)
848 may continue to float out!
853 class Y a b | a -> b where
856 instance Y [[a]] a where
859 k :: X a -> X a -> X a
861 g :: Num a => [X a] -> [X a]
864 h ys = ys ++ map (k (y [[0]])) xs
866 The excitement comes when simplifying the bindings for h. Initially
867 try to simplify {y @ [[t1]] t2, 0 @ t1}, with initial qtvs = {t2}.
868 From this we get t1:=:t2, but also various bindings. We can't forget
869 the bindings (because of [LOOP]), but in fact t1 is what g is
872 The net effect of [NO TYVARS]
875 isFreeWhenInferring :: TyVarSet -> Inst -> Bool
876 isFreeWhenInferring qtvs inst
877 = isFreeWrtTyVars qtvs inst -- Constrains no quantified vars
878 && isInheritableInst inst -- and no implicit parameter involved
879 -- see Note [Inheriting implicit parameters]
881 {- No longer used (with implication constraints)
882 isFreeWhenChecking :: TyVarSet -- Quantified tyvars
883 -> NameSet -- Quantified implicit parameters
885 isFreeWhenChecking qtvs ips inst
886 = isFreeWrtTyVars qtvs inst
887 && isFreeWrtIPs ips inst
890 isFreeWrtTyVars qtvs inst = tyVarsOfInst inst `disjointVarSet` qtvs
891 isFreeWrtIPs ips inst = not (any (`elemNameSet` ips) (ipNamesOfInst inst))
895 %************************************************************************
897 \subsection{tcSimplifyCheck}
899 %************************************************************************
901 @tcSimplifyCheck@ is used when we know exactly the set of variables
902 we are going to quantify over. For example, a class or instance declaration.
905 -----------------------------------------------------------
906 -- tcSimplifyCheck is used when checking expression type signatures,
907 -- class decls, instance decls etc.
908 tcSimplifyCheck :: InstLoc
909 -> [TcTyVar] -- Quantify over these
912 -> TcM TcDictBinds -- Bindings
913 tcSimplifyCheck loc qtvs givens wanteds
914 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
915 do { traceTc (text "tcSimplifyCheck")
916 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
917 ; implic_bind <- bindIrreds loc qtvs givens irreds
918 ; return (binds `unionBags` implic_bind) }
920 -----------------------------------------------------------
921 -- tcSimplifyCheckPat is used for existential pattern match
922 tcSimplifyCheckPat :: InstLoc
923 -> [TcTyVar] -- Quantify over these
926 -> TcM TcDictBinds -- Bindings
927 tcSimplifyCheckPat loc qtvs givens wanteds
928 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
929 do { traceTc (text "tcSimplifyCheckPat")
930 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
931 ; implic_bind <- bindIrredsR loc qtvs givens irreds
932 ; return (binds `unionBags` implic_bind) }
934 -----------------------------------------------------------
935 bindIrreds :: InstLoc -> [TcTyVar]
938 bindIrreds loc qtvs givens irreds
939 = bindIrredsR loc qtvs givens irreds
941 bindIrredsR :: InstLoc -> [TcTyVar] -> [Inst] -> [Inst] -> TcM TcDictBinds
942 -- Make a binding that binds 'irreds', by generating an implication
943 -- constraint for them, *and* throwing the constraint into the LIE
944 bindIrredsR loc qtvs givens irreds
948 = do { let givens' = filter isAbstractableInst givens
949 -- The givens can (redundantly) include methods
950 -- We want to retain both EqInsts and Dicts
951 -- There should be no implicadtion constraints
952 -- See Note [Pruning the givens in an implication constraint]
954 -- If there are no 'givens', then it's safe to
955 -- partition the 'wanteds' by their qtvs, thereby trimming irreds
956 -- See Note [Freeness and implications]
957 ; irreds' <- if null givens'
959 { let qtv_set = mkVarSet qtvs
960 (frees, real_irreds) = partition (isFreeWrtTyVars qtv_set) irreds
962 ; return real_irreds }
965 ; (implics, bind) <- makeImplicationBind loc qtvs givens' irreds'
966 -- This call does the real work
967 -- If irreds' is empty, it does something sensible
972 makeImplicationBind :: InstLoc -> [TcTyVar]
974 -> TcM ([Inst], TcDictBinds)
975 -- Make a binding that binds 'irreds', by generating an implication
976 -- constraint for them, *and* throwing the constraint into the LIE
977 -- The binding looks like
978 -- (ir1, .., irn) = f qtvs givens
979 -- where f is (evidence for) the new implication constraint
980 -- f :: forall qtvs. {reft} givens => (ir1, .., irn)
981 -- qtvs includes coercion variables
983 -- This binding must line up the 'rhs' in reduceImplication
984 makeImplicationBind loc all_tvs
985 givens -- Guaranteed all Dicts
988 | null irreds -- If there are no irreds, we are done
989 = return ([], emptyBag)
990 | otherwise -- Otherwise we must generate a binding
991 = do { uniq <- newUnique
992 ; span <- getSrcSpanM
993 ; let (eq_givens, dict_givens) = partition isEqInst givens
994 eq_tyvar_cos = mkTyVarTys (varSetElems $ tyVarsOfTypes $ map eqInstType eq_givens)
995 -- Urgh! See line 2187 or thereabouts. I believe that all these
996 -- 'givens' must be a simple CoVar. This MUST be cleaned up.
998 ; let name = mkInternalName uniq (mkVarOcc "ic") span
999 implic_inst = ImplicInst { tci_name = name,
1000 tci_tyvars = all_tvs,
1001 tci_given = (eq_givens ++ dict_givens),
1002 tci_wanted = irreds, tci_loc = loc }
1003 ; let -- only create binder for dict_irreds
1004 (eq_irreds, dict_irreds) = partition isEqInst irreds
1005 n_dict_irreds = length dict_irreds
1006 dict_irred_ids = map instToId dict_irreds
1007 tup_ty = mkTupleTy Boxed n_dict_irreds (map idType dict_irred_ids)
1008 pat = TuplePat (map nlVarPat dict_irred_ids) Boxed tup_ty
1009 rhs = L span (mkHsWrap co (HsVar (instToId implic_inst)))
1010 co = mkWpApps (map instToId dict_givens)
1011 <.> mkWpTyApps eq_tyvar_cos
1012 <.> mkWpTyApps (mkTyVarTys all_tvs)
1013 bind | [dict_irred_id] <- dict_irred_ids = VarBind dict_irred_id rhs
1014 | otherwise = PatBind { pat_lhs = L span pat,
1015 pat_rhs = unguardedGRHSs rhs,
1016 pat_rhs_ty = tup_ty,
1017 bind_fvs = placeHolderNames }
1018 ; traceTc $ text "makeImplicationBind" <+> ppr implic_inst
1019 ; return ([implic_inst], unitBag (L span bind))
1022 -----------------------------------------------------------
1023 tryHardCheckLoop :: SDoc
1025 -> TcM ([Inst], TcDictBinds)
1027 tryHardCheckLoop doc wanteds
1028 = do { (irreds,binds) <- checkLoop (mkRedEnv doc try_me []) wanteds
1029 ; return (irreds,binds)
1032 try_me inst = ReduceMe AddSCs
1033 -- Here's the try-hard bit
1035 -----------------------------------------------------------
1036 gentleCheckLoop :: InstLoc
1039 -> TcM ([Inst], TcDictBinds)
1041 gentleCheckLoop inst_loc givens wanteds
1042 = do { (irreds,binds) <- checkLoop env wanteds
1043 ; return (irreds,binds)
1046 env = mkRedEnv (pprInstLoc inst_loc) try_me givens
1048 try_me inst | isMethodOrLit inst = ReduceMe AddSCs
1050 -- When checking against a given signature
1051 -- we MUST be very gentle: Note [Check gently]
1053 gentleInferLoop :: SDoc -> [Inst]
1054 -> TcM ([Inst], TcDictBinds)
1055 gentleInferLoop doc wanteds
1056 = do { (irreds, binds) <- checkLoop env wanteds
1057 ; return (irreds, binds) }
1059 env = mkRedEnv doc try_me []
1060 try_me inst | isMethodOrLit inst = ReduceMe AddSCs
1065 ~~~~~~~~~~~~~~~~~~~~
1066 We have to very careful about not simplifying too vigorously
1071 f :: Show b => T b -> b
1072 f (MkT x) = show [x]
1074 Inside the pattern match, which binds (a:*, x:a), we know that
1076 Hence we have a dictionary for Show [a] available; and indeed we
1077 need it. We are going to build an implication contraint
1078 forall a. (b~[a]) => Show [a]
1079 Later, we will solve this constraint using the knowledge (Show b)
1081 But we MUST NOT reduce (Show [a]) to (Show a), else the whole
1082 thing becomes insoluble. So we simplify gently (get rid of literals
1083 and methods only, plus common up equal things), deferring the real
1084 work until top level, when we solve the implication constraint
1085 with tryHardCheckLooop.
1089 -----------------------------------------------------------
1092 -> TcM ([Inst], TcDictBinds)
1093 -- Precondition: givens are completely rigid
1094 -- Postcondition: returned Insts are zonked
1096 checkLoop env wanteds
1097 = go env wanteds (return ())
1098 where go env wanteds elim_skolems
1099 = do { -- We do need to zonk the givens; cf Note [Zonking RedEnv]
1100 ; env' <- zonkRedEnv env
1101 ; wanteds' <- zonkInsts wanteds
1103 ; (improved, binds, irreds, elim_more_skolems)
1104 <- reduceContext env' wanteds'
1105 ; let elim_skolems' = elim_skolems >> elim_more_skolems
1107 ; if not improved then
1108 elim_skolems' >> return (irreds, binds)
1111 -- If improvement did some unification, we go round again.
1112 -- We start again with irreds, not wanteds
1113 -- Using an instance decl might have introduced a fresh type
1114 -- variable which might have been unified, so we'd get an
1115 -- infinite loop if we started again with wanteds!
1117 { (irreds1, binds1) <- go env' irreds elim_skolems'
1118 ; return (irreds1, binds `unionBags` binds1) } }
1121 Note [Zonking RedEnv]
1122 ~~~~~~~~~~~~~~~~~~~~~
1123 It might appear as if the givens in RedEnv are always rigid, but that is not
1124 necessarily the case for programs involving higher-rank types that have class
1125 contexts constraining the higher-rank variables. An example from tc237 in the
1128 class Modular s a | s -> a
1130 wim :: forall a w. Integral a
1131 => a -> (forall s. Modular s a => M s w) -> w
1132 wim i k = error "urk"
1134 test5 :: (Modular s a, Integral a) => M s a
1137 test4 = wim 4 test4'
1139 Notice how the variable 'a' of (Modular s a) in the rank-2 type of wim is
1140 quantified further outside. When type checking test4, we have to check
1141 whether the signature of test5 is an instance of
1143 (forall s. Modular s a => M s w)
1145 Consequently, we will get (Modular s t_a), where t_a is a TauTv into the
1148 Given the FD of Modular in this example, class improvement will instantiate
1149 t_a to 'a', where 'a' is the skolem from test5's signatures (due to the
1150 Modular s a predicate in that signature). If we don't zonk (Modular s t_a) in
1151 the givens, we will get into a loop as improveOne uses the unification engine
1152 TcGadt.tcUnifyTys, which doesn't know about mutable type variables.
1157 class If b t e r | b t e -> r
1160 class Lte a b c | a b -> c where lte :: a -> b -> c
1162 instance (Lte a b l,If l b a c) => Max a b c
1164 Wanted: Max Z (S x) y
1166 Then we'll reduce using the Max instance to:
1167 (Lte Z (S x) l, If l (S x) Z y)
1168 and improve by binding l->T, after which we can do some reduction
1169 on both the Lte and If constraints. What we *can't* do is start again
1170 with (Max Z (S x) y)!
1174 %************************************************************************
1176 tcSimplifySuperClasses
1178 %************************************************************************
1180 Note [SUPERCLASS-LOOP 1]
1181 ~~~~~~~~~~~~~~~~~~~~~~~~
1182 We have to be very, very careful when generating superclasses, lest we
1183 accidentally build a loop. Here's an example:
1187 class S a => C a where { opc :: a -> a }
1188 class S b => D b where { opd :: b -> b }
1190 instance C Int where
1193 instance D Int where
1196 From (instance C Int) we get the constraint set {ds1:S Int, dd:D Int}
1197 Simplifying, we may well get:
1198 $dfCInt = :C ds1 (opd dd)
1201 Notice that we spot that we can extract ds1 from dd.
1203 Alas! Alack! We can do the same for (instance D Int):
1205 $dfDInt = :D ds2 (opc dc)
1209 And now we've defined the superclass in terms of itself.
1211 Solution: never generate a superclass selectors at all when
1212 satisfying the superclass context of an instance declaration.
1214 Two more nasty cases are in
1219 tcSimplifySuperClasses
1224 tcSimplifySuperClasses loc givens sc_wanteds
1225 = do { traceTc (text "tcSimplifySuperClasses")
1226 ; (irreds,binds1) <- checkLoop env sc_wanteds
1227 ; let (tidy_env, tidy_irreds) = tidyInsts irreds
1228 ; reportNoInstances tidy_env (Just (loc, givens)) tidy_irreds
1231 env = mkRedEnv (pprInstLoc loc) try_me givens
1232 try_me inst = ReduceMe NoSCs
1233 -- Like tryHardCheckLoop, but with NoSCs
1237 %************************************************************************
1239 \subsection{tcSimplifyRestricted}
1241 %************************************************************************
1243 tcSimplifyRestricted infers which type variables to quantify for a
1244 group of restricted bindings. This isn't trivial.
1247 We want to quantify over a to get id :: forall a. a->a
1250 We do not want to quantify over a, because there's an Eq a
1251 constraint, so we get eq :: a->a->Bool (notice no forall)
1254 RHS has type 'tau', whose free tyvars are tau_tvs
1255 RHS has constraints 'wanteds'
1258 Quantify over (tau_tvs \ ftvs(wanteds))
1259 This is bad. The constraints may contain (Monad (ST s))
1260 where we have instance Monad (ST s) where...
1261 so there's no need to be monomorphic in s!
1263 Also the constraint might be a method constraint,
1264 whose type mentions a perfectly innocent tyvar:
1265 op :: Num a => a -> b -> a
1266 Here, b is unconstrained. A good example would be
1268 We want to infer the polymorphic type
1269 foo :: forall b. b -> b
1272 Plan B (cunning, used for a long time up to and including GHC 6.2)
1273 Step 1: Simplify the constraints as much as possible (to deal
1274 with Plan A's problem). Then set
1275 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1277 Step 2: Now simplify again, treating the constraint as 'free' if
1278 it does not mention qtvs, and trying to reduce it otherwise.
1279 The reasons for this is to maximise sharing.
1281 This fails for a very subtle reason. Suppose that in the Step 2
1282 a constraint (Foo (Succ Zero) (Succ Zero) b) gets thrown upstairs as 'free'.
1283 In the Step 1 this constraint might have been simplified, perhaps to
1284 (Foo Zero Zero b), AND THEN THAT MIGHT BE IMPROVED, to bind 'b' to 'T'.
1285 This won't happen in Step 2... but that in turn might prevent some other
1286 constraint (Baz [a] b) being simplified (e.g. via instance Baz [a] T where {..})
1287 and that in turn breaks the invariant that no constraints are quantified over.
1289 Test typecheck/should_compile/tc177 (which failed in GHC 6.2) demonstrates
1294 Step 1: Simplify the constraints as much as possible (to deal
1295 with Plan A's problem). Then set
1296 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1297 Return the bindings from Step 1.
1300 A note about Plan C (arising from "bug" reported by George Russel March 2004)
1303 instance (HasBinary ty IO) => HasCodedValue ty
1305 foo :: HasCodedValue a => String -> IO a
1307 doDecodeIO :: HasCodedValue a => () -> () -> IO a
1308 doDecodeIO codedValue view
1309 = let { act = foo "foo" } in act
1311 You might think this should work becuase the call to foo gives rise to a constraint
1312 (HasCodedValue t), which can be satisfied by the type sig for doDecodeIO. But the
1313 restricted binding act = ... calls tcSimplifyRestricted, and PlanC simplifies the
1314 constraint using the (rather bogus) instance declaration, and now we are stuffed.
1316 I claim this is not really a bug -- but it bit Sergey as well as George. So here's
1320 Plan D (a variant of plan B)
1321 Step 1: Simplify the constraints as much as possible (to deal
1322 with Plan A's problem), BUT DO NO IMPROVEMENT. Then set
1323 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1325 Step 2: Now simplify again, treating the constraint as 'free' if
1326 it does not mention qtvs, and trying to reduce it otherwise.
1328 The point here is that it's generally OK to have too few qtvs; that is,
1329 to make the thing more monomorphic than it could be. We don't want to
1330 do that in the common cases, but in wierd cases it's ok: the programmer
1331 can always add a signature.
1333 Too few qtvs => too many wanteds, which is what happens if you do less
1338 tcSimplifyRestricted -- Used for restricted binding groups
1339 -- i.e. ones subject to the monomorphism restriction
1342 -> [Name] -- Things bound in this group
1343 -> TcTyVarSet -- Free in the type of the RHSs
1344 -> [Inst] -- Free in the RHSs
1345 -> TcM ([TyVar], -- Tyvars to quantify (zonked and quantified)
1346 TcDictBinds) -- Bindings
1347 -- tcSimpifyRestricted returns no constraints to
1348 -- quantify over; by definition there are none.
1349 -- They are all thrown back in the LIE
1351 tcSimplifyRestricted doc top_lvl bndrs tau_tvs wanteds
1352 -- Zonk everything in sight
1353 = do { traceTc (text "tcSimplifyRestricted")
1354 ; wanteds' <- zonkInsts wanteds
1356 -- 'ReduceMe': Reduce as far as we can. Don't stop at
1357 -- dicts; the idea is to get rid of as many type
1358 -- variables as possible, and we don't want to stop
1359 -- at (say) Monad (ST s), because that reduces
1360 -- immediately, with no constraint on s.
1362 -- BUT do no improvement! See Plan D above
1363 -- HOWEVER, some unification may take place, if we instantiate
1364 -- a method Inst with an equality constraint
1365 ; let env = mkNoImproveRedEnv doc (\i -> ReduceMe AddSCs)
1366 ; (_imp, _binds, constrained_dicts, elim_skolems)
1367 <- reduceContext env wanteds'
1370 -- Next, figure out the tyvars we will quantify over
1371 ; tau_tvs' <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
1372 ; gbl_tvs' <- tcGetGlobalTyVars
1373 ; constrained_dicts' <- zonkInsts constrained_dicts
1375 ; let qtvs1 = tau_tvs' `minusVarSet` oclose (fdPredsOfInsts constrained_dicts) gbl_tvs'
1376 -- As in tcSimplifyInfer
1378 -- Do not quantify over constrained type variables:
1379 -- this is the monomorphism restriction
1380 constrained_tvs' = tyVarsOfInsts constrained_dicts'
1381 qtvs = qtvs1 `minusVarSet` constrained_tvs'
1382 pp_bndrs = pprWithCommas (quotes . ppr) bndrs
1385 ; warn_mono <- doptM Opt_WarnMonomorphism
1386 ; warnTc (warn_mono && (constrained_tvs' `intersectsVarSet` qtvs1))
1387 (vcat[ ptext SLIT("the Monomorphism Restriction applies to the binding")
1388 <> plural bndrs <+> ptext SLIT("for") <+> pp_bndrs,
1389 ptext SLIT("Consider giving a type signature for") <+> pp_bndrs])
1391 ; traceTc (text "tcSimplifyRestricted" <+> vcat [
1392 pprInsts wanteds, pprInsts constrained_dicts',
1394 ppr constrained_tvs', ppr tau_tvs', ppr qtvs ])
1396 -- The first step may have squashed more methods than
1397 -- necessary, so try again, this time more gently, knowing the exact
1398 -- set of type variables to quantify over.
1400 -- We quantify only over constraints that are captured by qtvs;
1401 -- these will just be a subset of non-dicts. This in contrast
1402 -- to normal inference (using isFreeWhenInferring) in which we quantify over
1403 -- all *non-inheritable* constraints too. This implements choice
1404 -- (B) under "implicit parameter and monomorphism" above.
1406 -- Remember that we may need to do *some* simplification, to
1407 -- (for example) squash {Monad (ST s)} into {}. It's not enough
1408 -- just to float all constraints
1410 -- At top level, we *do* squash methods becuase we want to
1411 -- expose implicit parameters to the test that follows
1412 ; let is_nested_group = isNotTopLevel top_lvl
1413 try_me inst | isFreeWrtTyVars qtvs inst,
1414 (is_nested_group || isDict inst) = Stop
1415 | otherwise = ReduceMe AddSCs
1416 env = mkNoImproveRedEnv doc try_me
1417 ; (_imp, binds, irreds, elim_skolems) <- reduceContext env wanteds'
1420 -- See "Notes on implicit parameters, Question 4: top level"
1421 ; ASSERT( all (isFreeWrtTyVars qtvs) irreds ) -- None should be captured
1422 if is_nested_group then
1424 else do { let (bad_ips, non_ips) = partition isIPDict irreds
1425 ; addTopIPErrs bndrs bad_ips
1426 ; extendLIEs non_ips }
1428 ; qtvs' <- zonkQuantifiedTyVars (varSetElems qtvs)
1429 ; return (qtvs', binds) }
1433 %************************************************************************
1437 %************************************************************************
1439 On the LHS of transformation rules we only simplify methods and constants,
1440 getting dictionaries. We want to keep all of them unsimplified, to serve
1441 as the available stuff for the RHS of the rule.
1443 Example. Consider the following left-hand side of a rule
1445 f (x == y) (y > z) = ...
1447 If we typecheck this expression we get constraints
1449 d1 :: Ord a, d2 :: Eq a
1451 We do NOT want to "simplify" to the LHS
1453 forall x::a, y::a, z::a, d1::Ord a.
1454 f ((==) (eqFromOrd d1) x y) ((>) d1 y z) = ...
1458 forall x::a, y::a, z::a, d1::Ord a, d2::Eq a.
1459 f ((==) d2 x y) ((>) d1 y z) = ...
1461 Here is another example:
1463 fromIntegral :: (Integral a, Num b) => a -> b
1464 {-# RULES "foo" fromIntegral = id :: Int -> Int #-}
1466 In the rule, a=b=Int, and Num Int is a superclass of Integral Int. But
1467 we *dont* want to get
1469 forall dIntegralInt.
1470 fromIntegral Int Int dIntegralInt (scsel dIntegralInt) = id Int
1472 because the scsel will mess up RULE matching. Instead we want
1474 forall dIntegralInt, dNumInt.
1475 fromIntegral Int Int dIntegralInt dNumInt = id Int
1479 g (x == y) (y == z) = ..
1481 where the two dictionaries are *identical*, we do NOT WANT
1483 forall x::a, y::a, z::a, d1::Eq a
1484 f ((==) d1 x y) ((>) d1 y z) = ...
1486 because that will only match if the dict args are (visibly) equal.
1487 Instead we want to quantify over the dictionaries separately.
1489 In short, tcSimplifyRuleLhs must *only* squash LitInst and MethInts, leaving
1490 all dicts unchanged, with absolutely no sharing. It's simpler to do this
1491 from scratch, rather than further parameterise simpleReduceLoop etc
1494 tcSimplifyRuleLhs :: [Inst] -> TcM ([Inst], TcDictBinds)
1495 tcSimplifyRuleLhs wanteds
1496 = go [] emptyBag wanteds
1499 = return (dicts, binds)
1500 go dicts binds (w:ws)
1502 = go (w:dicts) binds ws
1504 = do { w' <- zonkInst w -- So that (3::Int) does not generate a call
1505 -- to fromInteger; this looks fragile to me
1506 ; lookup_result <- lookupSimpleInst w'
1507 ; case lookup_result of
1509 go dicts (addInstToDictBind binds w rhs) (ws' ++ ws)
1510 NoInstance -> pprPanic "tcSimplifyRuleLhs" (ppr w)
1514 tcSimplifyBracket is used when simplifying the constraints arising from
1515 a Template Haskell bracket [| ... |]. We want to check that there aren't
1516 any constraints that can't be satisfied (e.g. Show Foo, where Foo has no
1517 Show instance), but we aren't otherwise interested in the results.
1518 Nor do we care about ambiguous dictionaries etc. We will type check
1519 this bracket again at its usage site.
1522 tcSimplifyBracket :: [Inst] -> TcM ()
1523 tcSimplifyBracket wanteds
1524 = do { tryHardCheckLoop doc wanteds
1527 doc = text "tcSimplifyBracket"
1531 %************************************************************************
1533 \subsection{Filtering at a dynamic binding}
1535 %************************************************************************
1540 we must discharge all the ?x constraints from B. We also do an improvement
1541 step; if we have ?x::t1 and ?x::t2 we must unify t1, t2.
1543 Actually, the constraints from B might improve the types in ?x. For example
1545 f :: (?x::Int) => Char -> Char
1548 then the constraint (?x::Int) arising from the call to f will
1549 force the binding for ?x to be of type Int.
1552 tcSimplifyIPs :: [Inst] -- The implicit parameters bound here
1555 -- We need a loop so that we do improvement, and then
1556 -- (next time round) generate a binding to connect the two
1558 -- Here the two ?x's have different types, and improvement
1559 -- makes them the same.
1561 tcSimplifyIPs given_ips wanteds
1562 = do { wanteds' <- zonkInsts wanteds
1563 ; given_ips' <- zonkInsts given_ips
1564 -- Unusually for checking, we *must* zonk the given_ips
1566 ; let env = mkRedEnv doc try_me given_ips'
1567 ; (improved, binds, irreds, elim_skolems) <- reduceContext env wanteds'
1570 ; if not improved then
1571 ASSERT( all is_free irreds )
1572 do { extendLIEs irreds
1575 tcSimplifyIPs given_ips wanteds }
1577 doc = text "tcSimplifyIPs" <+> ppr given_ips
1578 ip_set = mkNameSet (ipNamesOfInsts given_ips)
1579 is_free inst = isFreeWrtIPs ip_set inst
1581 -- Simplify any methods that mention the implicit parameter
1582 try_me inst | is_free inst = Stop
1583 | otherwise = ReduceMe NoSCs
1587 %************************************************************************
1589 \subsection[binds-for-local-funs]{@bindInstsOfLocalFuns@}
1591 %************************************************************************
1593 When doing a binding group, we may have @Insts@ of local functions.
1594 For example, we might have...
1596 let f x = x + 1 -- orig local function (overloaded)
1597 f.1 = f Int -- two instances of f
1602 The point is: we must drop the bindings for @f.1@ and @f.2@ here,
1603 where @f@ is in scope; those @Insts@ must certainly not be passed
1604 upwards towards the top-level. If the @Insts@ were binding-ified up
1605 there, they would have unresolvable references to @f@.
1607 We pass in an @init_lie@ of @Insts@ and a list of locally-bound @Ids@.
1608 For each method @Inst@ in the @init_lie@ that mentions one of the
1609 @Ids@, we create a binding. We return the remaining @Insts@ (in an
1610 @LIE@), as well as the @HsBinds@ generated.
1613 bindInstsOfLocalFuns :: [Inst] -> [TcId] -> TcM TcDictBinds
1614 -- Simlifies only MethodInsts, and generate only bindings of form
1616 -- We're careful not to even generate bindings of the form
1618 -- You'd think that'd be fine, but it interacts with what is
1619 -- arguably a bug in Match.tidyEqnInfo (see notes there)
1621 bindInstsOfLocalFuns wanteds local_ids
1622 | null overloaded_ids = do
1625 return emptyLHsBinds
1628 = do { (irreds, binds) <- gentleInferLoop doc for_me
1629 ; extendLIEs not_for_me
1633 doc = text "bindInsts" <+> ppr local_ids
1634 overloaded_ids = filter is_overloaded local_ids
1635 is_overloaded id = isOverloadedTy (idType id)
1636 (for_me, not_for_me) = partition (isMethodFor overloaded_set) wanteds
1638 overloaded_set = mkVarSet overloaded_ids -- There can occasionally be a lot of them
1639 -- so it's worth building a set, so that
1640 -- lookup (in isMethodFor) is faster
1644 %************************************************************************
1646 \subsection{Data types for the reduction mechanism}
1648 %************************************************************************
1650 The main control over context reduction is here
1654 = RedEnv { red_doc :: SDoc -- The context
1655 , red_try_me :: Inst -> WhatToDo
1656 , red_improve :: Bool -- True <=> do improvement
1657 , red_givens :: [Inst] -- All guaranteed rigid
1659 -- but see Note [Rigidity]
1660 , red_stack :: (Int, [Inst]) -- Recursion stack (for err msg)
1661 -- See Note [RedStack]
1665 -- The red_givens are rigid so far as cmpInst is concerned.
1666 -- There is one case where they are not totally rigid, namely in tcSimplifyIPs
1667 -- let ?x = e in ...
1668 -- Here, the given is (?x::a), where 'a' is not necy a rigid type
1669 -- But that doesn't affect the comparison, which is based only on mame.
1672 -- The red_stack pair (n,insts) pair is just used for error reporting.
1673 -- 'n' is always the depth of the stack.
1674 -- The 'insts' is the stack of Insts being reduced: to produce X
1675 -- I had to produce Y, to produce Y I had to produce Z, and so on.
1678 mkRedEnv :: SDoc -> (Inst -> WhatToDo) -> [Inst] -> RedEnv
1679 mkRedEnv doc try_me givens
1680 = RedEnv { red_doc = doc, red_try_me = try_me,
1681 red_givens = givens,
1683 red_improve = True }
1685 mkNoImproveRedEnv :: SDoc -> (Inst -> WhatToDo) -> RedEnv
1686 -- Do not do improvement; no givens
1687 mkNoImproveRedEnv doc try_me
1688 = RedEnv { red_doc = doc, red_try_me = try_me,
1691 red_improve = True }
1694 = ReduceMe WantSCs -- Try to reduce this
1695 -- If there's no instance, add the inst to the
1696 -- irreductible ones, but don't produce an error
1697 -- message of any kind.
1698 -- It might be quite legitimate such as (Eq a)!
1700 | Stop -- Return as irreducible unless it can
1701 -- be reduced to a constant in one step
1702 -- Do not add superclasses; see
1704 data WantSCs = NoSCs | AddSCs -- Tells whether we should add the superclasses
1705 -- of a predicate when adding it to the avails
1706 -- The reason for this flag is entirely the super-class loop problem
1707 -- Note [SUPER-CLASS LOOP 1]
1709 zonkRedEnv :: RedEnv -> TcM RedEnv
1711 = do { givens' <- mapM zonkInst (red_givens env)
1712 ; return $ env {red_givens = givens'}
1717 %************************************************************************
1719 \subsection[reduce]{@reduce@}
1721 %************************************************************************
1723 Note [Ancestor Equalities]
1724 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1725 During context reduction, we add to the wanted equalities also those
1726 equalities that (transitively) occur in superclass contexts of wanted
1727 class constraints. Consider the following code
1729 class a ~ Int => C a
1732 If (C a) is wanted, we want to add (a ~ Int), which will be discharged by
1733 substituting Int for a. Hence, we ultimately want (C Int), which we
1734 discharge with the explicit instance.
1737 reduceContext :: RedEnv
1739 -> TcM (ImprovementDone,
1740 TcDictBinds, -- Dictionary bindings
1741 [Inst], -- Irreducible
1742 TcM ()) -- Undo skolems from SkolemOccurs
1744 reduceContext env wanteds
1745 = do { traceTc (text "reduceContext" <+> (vcat [
1746 text "----------------------",
1748 text "given" <+> ppr (red_givens env),
1749 text "wanted" <+> ppr wanteds,
1750 text "----------------------"
1754 ; let givens = red_givens env
1755 (given_eqs0, given_dicts0) = partition isEqInst givens
1756 (wanted_eqs0, wanted_non_eqs) = partition isEqInst wanteds
1757 (wanted_implics0, wanted_dicts) = partition isImplicInst wanted_non_eqs
1759 -- We want to add as wanted equalities those that (transitively)
1760 -- occur in superclass contexts of wanted class constraints.
1761 -- See Note [Ancestor Equalities]
1762 ; ancestor_eqs <- ancestorEqualities wanted_dicts
1763 ; let wanted_eqs = wanted_eqs0 ++ ancestor_eqs
1764 ; traceTc $ text "reduceContext: ancestor eqs" <+> ppr ancestor_eqs
1766 -- 1. Normalise the *given* *equality* constraints
1767 ; (given_eqs, eliminate_skolems) <- normaliseGivenEqs given_eqs0
1769 -- 2. Normalise the *given* *dictionary* constraints
1770 -- wrt. the toplevel and given equations
1771 ; (given_dicts, given_binds) <- normaliseGivenDicts given_eqs
1774 -- 5. Build the Avail mapping from "given_dicts"
1775 ; (init_state, extra_givens) <- getLIE $ do
1776 { init_state <- foldlM addGiven emptyAvails given_dicts
1780 -- *** ToDo: what to do with the "extra_givens"? For the
1781 -- moment I'm simply discarding them, which is probably wrong
1783 -- 6. Solve the *wanted* *dictionary* constraints (not implications)
1784 -- This may expose some further equational constraints...
1785 ; (avails, extra_eqs) <- getLIE (reduceList env wanted_dicts init_state)
1786 ; (dict_binds, bound_dicts, dict_irreds)
1787 <- extractResults avails wanted_dicts
1788 ; traceTc $ text "reduceContext extractresults" <+> vcat
1789 [ppr avails, ppr wanted_dicts, ppr dict_binds]
1791 -- Solve the wanted *implications*. In doing so, we can provide
1792 -- as "given" all the dicts that were originally given,
1793 -- *or* for which we now have bindings,
1794 -- *or* which are now irreds
1795 ; let implic_env = env { red_givens = givens ++ bound_dicts
1797 ; (implic_binds_s, implic_irreds_s)
1798 <- mapAndUnzipM (reduceImplication implic_env) wanted_implics0
1799 ; let implic_binds = unionManyBags implic_binds_s
1800 implic_irreds = concat implic_irreds_s
1802 -- Normalise the wanted equality constraints
1803 ; eq_irreds <- normaliseWantedEqs given_eqs (wanted_eqs ++ extra_eqs)
1805 -- Normalise the wanted dictionaries
1806 ; let irreds = dict_irreds ++ implic_irreds
1807 eqs = eq_irreds ++ given_eqs
1808 ; (norm_irreds, normalise_binds) <- normaliseWantedDicts eqs irreds
1810 -- Figure out whether we should go round again. We do so in either
1812 -- (1) If any of the mutable tyvars in givens or irreds has been
1813 -- filled in by improvement, there is merit in going around
1814 -- again, because we may make further progress.
1815 -- (2) If we managed to normalise any dicts, there is merit in going
1816 -- around gain, because reduceList may be able to get further.
1818 -- ToDo: We may have exposed new
1819 -- equality constraints and should probably go round again
1820 -- then as well. But currently we are dropping them on the
1823 ; let all_irreds = norm_irreds ++ eq_irreds
1824 ; improvedMetaTy <- anyM isFilledMetaTyVar $ varSetElems $
1825 tyVarsOfInsts (givens ++ all_irreds)
1826 ; let improvedDicts = not $ isEmptyBag normalise_binds
1827 improved = improvedMetaTy || improvedDicts
1829 -- The old plan (fragile)
1830 -- improveed = availsImproved avails
1831 -- || (not $ isEmptyBag normalise_binds1)
1832 -- || (not $ isEmptyBag normalise_binds2)
1833 -- || (any isEqInst irreds)
1835 ; traceTc (text "reduceContext end" <+> (vcat [
1836 text "----------------------",
1838 text "given" <+> ppr givens,
1839 text "given_eqs" <+> ppr given_eqs,
1840 text "wanted" <+> ppr wanteds,
1841 text "wanted_dicts" <+> ppr wanted_dicts,
1843 text "avails" <+> pprAvails avails,
1844 text "improved =" <+> ppr improved,
1845 text "(all) irreds = " <+> ppr all_irreds,
1846 text "dict-binds = " <+> ppr dict_binds,
1847 text "implic-binds = " <+> ppr implic_binds,
1848 text "----------------------"
1852 given_binds `unionBags` normalise_binds
1853 `unionBags` dict_binds
1854 `unionBags` implic_binds,
1859 tcImproveOne :: Avails -> Inst -> TcM ImprovementDone
1860 tcImproveOne avails inst
1861 | not (isDict inst) = return False
1863 = do { inst_envs <- tcGetInstEnvs
1864 ; let eqns = improveOne (classInstances inst_envs)
1865 (dictPred inst, pprInstArising inst)
1866 [ (dictPred p, pprInstArising p)
1867 | p <- availsInsts avails, isDict p ]
1868 -- Avails has all the superclasses etc (good)
1869 -- It also has all the intermediates of the deduction (good)
1870 -- It does not have duplicates (good)
1871 -- NB that (?x::t1) and (?x::t2) will be held separately in avails
1872 -- so that improve will see them separate
1873 ; traceTc (text "improveOne" <+> ppr inst)
1876 unifyEqns :: [(Equation,(PredType,SDoc),(PredType,SDoc))]
1877 -> TcM ImprovementDone
1878 unifyEqns [] = return False
1880 = do { traceTc (ptext SLIT("Improve:") <+> vcat (map pprEquationDoc eqns))
1884 unify ((qtvs, pairs), what1, what2)
1885 = addErrCtxtM (mkEqnMsg what1 what2) $ do
1886 (_, _, tenv) <- tcInstTyVars (varSetElems qtvs)
1887 mapM_ (unif_pr tenv) pairs
1888 unif_pr tenv (ty1,ty2) = unifyType (substTy tenv ty1) (substTy tenv ty2)
1890 pprEquationDoc (eqn, (p1,w1), (p2,w2)) = vcat [pprEquation eqn, nest 2 (ppr p1), nest 2 (ppr p2)]
1892 mkEqnMsg (pred1,from1) (pred2,from2) tidy_env
1893 = do { pred1' <- zonkTcPredType pred1; pred2' <- zonkTcPredType pred2
1894 ; let { pred1'' = tidyPred tidy_env pred1'; pred2'' = tidyPred tidy_env pred2' }
1895 ; let msg = vcat [ptext SLIT("When using functional dependencies to combine"),
1896 nest 2 (sep [ppr pred1'' <> comma, nest 2 from1]),
1897 nest 2 (sep [ppr pred2'' <> comma, nest 2 from2])]
1898 ; return (tidy_env, msg) }
1901 The main context-reduction function is @reduce@. Here's its game plan.
1904 reduceList :: RedEnv -> [Inst] -> Avails -> TcM Avails
1905 reduceList env@(RedEnv {red_stack = (n,stk)}) wanteds state
1906 = do { traceTc (text "reduceList " <+> (ppr wanteds $$ ppr state))
1910 dumpTcRn (hang (ptext SLIT("Interesting! Context reduction stack depth") <+> int n)
1911 2 (ifPprDebug (nest 2 (pprStack stk))))
1914 ; if n >= ctxtStkDepth dopts then
1915 failWithTc (reduceDepthErr n stk)
1919 go [] state = return state
1920 go (w:ws) state = do { state' <- reduce (env {red_stack = (n+1, w:stk)}) w state
1923 -- Base case: we're done!
1924 reduce env wanted avails
1925 -- It's the same as an existing inst, or a superclass thereof
1926 | Just avail <- findAvail avails wanted
1927 = do { traceTc (text "reduce: found " <+> ppr wanted)
1932 = do { traceTc (text "reduce" <+> ppr wanted $$ ppr avails)
1933 ; case red_try_me env wanted of {
1934 Stop -> try_simple (addIrred NoSCs);
1935 -- See Note [No superclasses for Stop]
1937 ReduceMe want_scs -> do -- It should be reduced
1938 { (avails, lookup_result) <- reduceInst env avails wanted
1939 ; case lookup_result of
1940 NoInstance -> addIrred want_scs avails wanted
1941 -- Add it and its superclasses
1943 GenInst [] rhs -> addWanted want_scs avails wanted rhs []
1945 GenInst wanteds' rhs
1946 -> do { avails1 <- addIrred NoSCs avails wanted
1947 ; avails2 <- reduceList env wanteds' avails1
1948 ; addWanted want_scs avails2 wanted rhs wanteds' } }
1949 -- Temporarily do addIrred *before* the reduceList,
1950 -- which has the effect of adding the thing we are trying
1951 -- to prove to the database before trying to prove the things it
1952 -- needs. See note [RECURSIVE DICTIONARIES]
1953 -- NB: we must not do an addWanted before, because that adds the
1954 -- superclasses too, and that can lead to a spurious loop; see
1955 -- the examples in [SUPERCLASS-LOOP]
1956 -- So we do an addIrred before, and then overwrite it afterwards with addWanted
1959 -- First, see if the inst can be reduced to a constant in one step
1960 -- Works well for literals (1::Int) and constant dictionaries (d::Num Int)
1961 -- Don't bother for implication constraints, which take real work
1962 try_simple do_this_otherwise
1963 = do { res <- lookupSimpleInst wanted
1965 GenInst [] rhs -> addWanted AddSCs avails wanted rhs []
1966 other -> do_this_otherwise avails wanted }
1970 Note [SUPERCLASS-LOOP 2]
1971 ~~~~~~~~~~~~~~~~~~~~~~~~
1972 But the above isn't enough. Suppose we are *given* d1:Ord a,
1973 and want to deduce (d2:C [a]) where
1975 class Ord a => C a where
1976 instance Ord [a] => C [a] where ...
1978 Then we'll use the instance decl to deduce C [a] from Ord [a], and then add the
1979 superclasses of C [a] to avails. But we must not overwrite the binding
1980 for Ord [a] (which is obtained from Ord a) with a superclass selection or we'll just
1983 Here's another variant, immortalised in tcrun020
1984 class Monad m => C1 m
1985 class C1 m => C2 m x
1986 instance C2 Maybe Bool
1987 For the instance decl we need to build (C1 Maybe), and it's no good if
1988 we run around and add (C2 Maybe Bool) and its superclasses to the avails
1989 before we search for C1 Maybe.
1991 Here's another example
1992 class Eq b => Foo a b
1993 instance Eq a => Foo [a] a
1997 we'll first deduce that it holds (via the instance decl). We must not
1998 then overwrite the Eq t constraint with a superclass selection!
2000 At first I had a gross hack, whereby I simply did not add superclass constraints
2001 in addWanted, though I did for addGiven and addIrred. This was sub-optimal,
2002 becuase it lost legitimate superclass sharing, and it still didn't do the job:
2003 I found a very obscure program (now tcrun021) in which improvement meant the
2004 simplifier got two bites a the cherry... so something seemed to be an Stop
2005 first time, but reducible next time.
2007 Now we implement the Right Solution, which is to check for loops directly
2008 when adding superclasses. It's a bit like the occurs check in unification.
2011 Note [RECURSIVE DICTIONARIES]
2012 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2014 data D r = ZeroD | SuccD (r (D r));
2016 instance (Eq (r (D r))) => Eq (D r) where
2017 ZeroD == ZeroD = True
2018 (SuccD a) == (SuccD b) = a == b
2021 equalDC :: D [] -> D [] -> Bool;
2024 We need to prove (Eq (D [])). Here's how we go:
2028 by instance decl, holds if
2032 by instance decl of Eq, holds if
2034 where d2 = dfEqList d3
2037 But now we can "tie the knot" to give
2043 and it'll even run! The trick is to put the thing we are trying to prove
2044 (in this case Eq (D []) into the database before trying to prove its
2045 contributing clauses.
2048 %************************************************************************
2050 Reducing a single constraint
2052 %************************************************************************
2055 ---------------------------------------------
2056 reduceInst :: RedEnv -> Avails -> Inst -> TcM (Avails, LookupInstResult)
2057 reduceInst env avails other_inst
2058 = do { result <- lookupSimpleInst other_inst
2059 ; return (avails, result) }
2062 Note [Equational Constraints in Implication Constraints]
2063 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2065 An implication constraint is of the form
2067 where Given and Wanted may contain both equational and dictionary
2068 constraints. The delay and reduction of these two kinds of constraints
2071 -) In the generated code, wanted Dictionary constraints are wrapped up in an
2072 implication constraint that is created at the code site where the wanted
2073 dictionaries can be reduced via a let-binding. This let-bound implication
2074 constraint is deconstructed at the use-site of the wanted dictionaries.
2076 -) While the reduction of equational constraints is also delayed, the delay
2077 is not manifest in the generated code. The required evidence is generated
2078 in the code directly at the use-site. There is no let-binding and deconstruction
2079 necessary. The main disadvantage is that we cannot exploit sharing as the
2080 same evidence may be generated at multiple use-sites. However, this disadvantage
2081 is limited because it only concerns coercions which are erased.
2083 The different treatment is motivated by the different in representation. Dictionary
2084 constraints require manifest runtime dictionaries, while equations require coercions
2088 ---------------------------------------------
2089 reduceImplication :: RedEnv
2091 -> TcM (TcDictBinds, [Inst])
2094 Suppose we are simplifying the constraint
2095 forall bs. extras => wanted
2096 in the context of an overall simplification problem with givens 'givens'.
2099 * The 'givens' need not mention any of the quantified type variables
2100 e.g. forall {}. Eq a => Eq [a]
2101 forall {}. C Int => D (Tree Int)
2103 This happens when you have something like
2105 T1 :: Eq a => a -> T a
2108 f x = ...(case x of { T1 v -> v==v })...
2111 -- ToDo: should we instantiate tvs? I think it's not necessary
2113 -- Note on coercion variables:
2115 -- The extra given coercion variables are bound at two different sites:
2116 -- -) in the creation context of the implication constraint
2117 -- the solved equational constraints use these binders
2119 -- -) at the solving site of the implication constraint
2120 -- the solved dictionaries use these binders
2121 -- these binders are generated by reduceImplication
2123 reduceImplication env
2124 orig_implic@(ImplicInst { tci_name = name, tci_loc = inst_loc,
2126 tci_given = extra_givens, tci_wanted = wanteds })
2127 = do { -- Solve the sub-problem
2128 ; let try_me inst = ReduceMe AddSCs -- Note [Freeness and implications]
2129 env' = env { red_givens = extra_givens ++ red_givens env
2130 , red_doc = sep [ptext SLIT("reduceImplication for")
2132 nest 2 (parens $ ptext SLIT("within")
2134 , red_try_me = try_me }
2136 ; traceTc (text "reduceImplication" <+> vcat
2137 [ ppr (red_givens env), ppr extra_givens,
2139 ; (irreds, binds) <- checkLoop env' wanteds
2140 ; let (extra_eq_givens, extra_dict_givens) = partition isEqInst extra_givens
2141 -- SLPJ Sept 07: I think this is bogus; currently
2142 -- there are no Eqinsts in extra_givens
2143 dict_ids = map instToId extra_dict_givens
2145 -- Note [Reducing implication constraints]
2146 -- Tom -- update note, put somewhere!
2148 ; traceTc (text "reduceImplication result" <+> vcat
2149 [ppr irreds, ppr binds])
2151 ; -- extract superclass binds
2152 -- (sc_binds,_) <- extractResults avails []
2153 -- ; traceTc (text "reduceImplication sc_binds" <+> vcat
2154 -- [ppr sc_binds, ppr avails])
2157 -- SLPJ Sept 07: what if improvement happened inside the checkLoop?
2158 -- Then we must iterate the outer loop too!
2160 ; traceTc (text "reduceImplication condition" <+> ppr ((isEmptyLHsBinds binds) || (null irreds)))
2162 -- Progress is no longer measered by the number of bindings
2163 ; if (isEmptyLHsBinds binds) && (not $ null irreds) then -- No progress
2164 -- If there are any irreds, we back off and do nothing
2165 return (emptyBag, [orig_implic])
2167 { (simpler_implic_insts, bind)
2168 <- makeImplicationBind inst_loc tvs extra_givens irreds
2169 -- This binding is useless if the recursive simplification
2170 -- made no progress; but currently we don't try to optimise that
2171 -- case. After all, we only try hard to reduce at top level, or
2172 -- when inferring types.
2174 ; let dict_wanteds = filter (not . isEqInst) wanteds
2175 -- TOMDO: given equational constraints bug!
2176 -- we need a different evidence for given
2177 -- equations depending on whether we solve
2178 -- dictionary constraints or equational constraints
2180 eq_tyvars = varSetElems $ tyVarsOfTypes $ map eqInstType extra_eq_givens
2181 -- SLPJ Sept07: this looks Utterly Wrong to me, but I think
2182 -- that current extra_givens has no EqInsts, so
2183 -- it makes no difference
2184 co = wrap_inline -- Note [Always inline implication constraints]
2186 <.> mkWpLams eq_tyvars
2187 <.> mkWpLams dict_ids
2188 <.> WpLet (binds `unionBags` bind)
2189 wrap_inline | null dict_ids = idHsWrapper
2190 | otherwise = WpInline
2191 rhs = mkHsWrap co payload
2192 loc = instLocSpan inst_loc
2193 payload | [dict_wanted] <- dict_wanteds = HsVar (instToId dict_wanted)
2194 | otherwise = ExplicitTuple (map (L loc . HsVar . instToId) dict_wanteds) Boxed
2197 ; traceTc (vcat [text "reduceImplication" <+> ppr name,
2198 ppr simpler_implic_insts,
2199 text "->" <+> ppr rhs])
2200 ; return (unitBag (L loc (VarBind (instToId orig_implic) (L loc rhs))),
2201 simpler_implic_insts)
2206 Note [Always inline implication constraints]
2207 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2208 Suppose an implication constraint floats out of an INLINE function.
2209 Then although the implication has a single call site, it won't be
2210 inlined. And that is bad because it means that even if there is really
2211 *no* overloading (type signatures specify the exact types) there will
2212 still be dictionary passing in the resulting code. To avert this,
2213 we mark the implication constraints themselves as INLINE, at least when
2214 there is no loss of sharing as a result.
2216 Note [Freeness and implications]
2217 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2218 It's hard to say when an implication constraint can be floated out. Consider
2219 forall {} Eq a => Foo [a]
2220 The (Foo [a]) doesn't mention any of the quantified variables, but it
2221 still might be partially satisfied by the (Eq a).
2223 There is a useful special case when it *is* easy to partition the
2224 constraints, namely when there are no 'givens'. Consider
2225 forall {a}. () => Bar b
2226 There are no 'givens', and so there is no reason to capture (Bar b).
2227 We can let it float out. But if there is even one constraint we
2228 must be much more careful:
2229 forall {a}. C a b => Bar (m b)
2230 because (C a b) might have a superclass (D b), from which we might
2231 deduce (Bar [b]) when m later gets instantiated to []. Ha!
2233 Here is an even more exotic example
2235 Now consider the constraint
2236 forall b. D Int b => C Int
2237 We can satisfy the (C Int) from the superclass of D, so we don't want
2238 to float the (C Int) out, even though it mentions no type variable in
2241 Note [Pruning the givens in an implication constraint]
2242 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2243 Suppose we are about to form the implication constraint
2244 forall tvs. Eq a => Ord b
2245 The (Eq a) cannot contribute to the (Ord b), because it has no access to
2246 the type variable 'b'. So we could filter out the (Eq a) from the givens.
2248 Doing so would be a bit tidier, but all the implication constraints get
2249 simplified away by the optimiser, so it's no great win. So I don't take
2250 advantage of that at the moment.
2252 If you do, BE CAREFUL of wobbly type variables.
2255 %************************************************************************
2257 Avails and AvailHow: the pool of evidence
2259 %************************************************************************
2263 data Avails = Avails !ImprovementDone !AvailEnv
2265 type ImprovementDone = Bool -- True <=> some unification has happened
2266 -- so some Irreds might now be reducible
2267 -- keys that are now
2269 type AvailEnv = FiniteMap Inst AvailHow
2271 = IsIrred -- Used for irreducible dictionaries,
2272 -- which are going to be lambda bound
2274 | Given Inst -- Used for dictionaries for which we have a binding
2275 -- e.g. those "given" in a signature
2277 | Rhs -- Used when there is a RHS
2278 (LHsExpr TcId) -- The RHS
2279 [Inst] -- Insts free in the RHS; we need these too
2281 instance Outputable Avails where
2284 pprAvails (Avails imp avails)
2285 = vcat [ ptext SLIT("Avails") <> (if imp then ptext SLIT("[improved]") else empty)
2287 vcat [ sep [ppr inst, nest 2 (equals <+> ppr avail)]
2288 | (inst,avail) <- fmToList avails ]]
2290 instance Outputable AvailHow where
2293 -------------------------
2294 pprAvail :: AvailHow -> SDoc
2295 pprAvail IsIrred = text "Irred"
2296 pprAvail (Given x) = text "Given" <+> ppr x
2297 pprAvail (Rhs rhs bs) = sep [text "Rhs" <+> ppr bs,
2300 -------------------------
2301 extendAvailEnv :: AvailEnv -> Inst -> AvailHow -> AvailEnv
2302 extendAvailEnv env inst avail = addToFM env inst avail
2304 findAvailEnv :: AvailEnv -> Inst -> Maybe AvailHow
2305 findAvailEnv env wanted = lookupFM env wanted
2306 -- NB 1: the Ord instance of Inst compares by the class/type info
2307 -- *not* by unique. So
2308 -- d1::C Int == d2::C Int
2310 emptyAvails :: Avails
2311 emptyAvails = Avails False emptyFM
2313 findAvail :: Avails -> Inst -> Maybe AvailHow
2314 findAvail (Avails _ avails) wanted = findAvailEnv avails wanted
2316 elemAvails :: Inst -> Avails -> Bool
2317 elemAvails wanted (Avails _ avails) = wanted `elemFM` avails
2319 extendAvails :: Avails -> Inst -> AvailHow -> TcM Avails
2321 extendAvails avails@(Avails imp env) inst avail
2322 = do { imp1 <- tcImproveOne avails inst -- Do any improvement
2323 ; return (Avails (imp || imp1) (extendAvailEnv env inst avail)) }
2325 availsInsts :: Avails -> [Inst]
2326 availsInsts (Avails _ avails) = keysFM avails
2328 availsImproved (Avails imp _) = imp
2330 updateImprovement :: Avails -> Avails -> Avails
2331 -- (updateImprovement a1 a2) sets a1's improvement flag from a2
2332 updateImprovement (Avails _ avails1) (Avails imp2 _) = Avails imp2 avails1
2335 Extracting the bindings from a bunch of Avails.
2336 The bindings do *not* come back sorted in dependency order.
2337 We assume that they'll be wrapped in a big Rec, so that the
2338 dependency analyser can sort them out later
2341 type DoneEnv = FiniteMap Inst [Id]
2342 -- Tracks which things we have evidence for
2344 extractResults :: Avails
2346 -> TcM (TcDictBinds, -- Bindings
2347 [Inst], -- The insts bound by the bindings
2348 [Inst]) -- Irreducible ones
2349 -- Note [Reducing implication constraints]
2351 extractResults (Avails _ avails) wanteds
2352 = go emptyBag [] [] emptyFM wanteds
2354 go :: TcDictBinds -- Bindings for dicts
2355 -> [Inst] -- Bound by the bindings
2357 -> DoneEnv -- Has an entry for each inst in the above three sets
2359 -> TcM (TcDictBinds, [Inst], [Inst])
2360 go binds bound_dicts irreds done []
2361 = return (binds, bound_dicts, irreds)
2363 go binds bound_dicts irreds done (w:ws)
2364 | Just done_ids@(done_id : rest_done_ids) <- lookupFM done w
2365 = if w_id `elem` done_ids then
2366 go binds bound_dicts irreds done ws
2368 go (add_bind (nlHsVar done_id)) bound_dicts irreds
2369 (addToFM done w (done_id : w_id : rest_done_ids)) ws
2371 | otherwise -- Not yet done
2372 = case findAvailEnv avails w of
2373 Nothing -> pprTrace "Urk: extractResults" (ppr w) $
2374 go binds bound_dicts irreds done ws
2376 Just IsIrred -> go binds bound_dicts (w:irreds) done' ws
2378 Just (Rhs rhs ws') -> go (add_bind rhs) (w:bound_dicts) irreds done' (ws' ++ ws)
2380 Just (Given g) -> go binds' bound_dicts irreds (addToFM done w [g_id]) ws
2383 binds' | w_id == g_id = binds
2384 | otherwise = add_bind (nlHsVar g_id)
2387 done' = addToFM done w [w_id]
2388 add_bind rhs = addInstToDictBind binds w rhs
2392 Note [No superclasses for Stop]
2393 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2394 When we decide not to reduce an Inst -- the 'WhatToDo' --- we still
2395 add it to avails, so that any other equal Insts will be commoned up
2396 right here. However, we do *not* add superclasses. If we have
2399 but a is not bound here, then we *don't* want to derive dn from df
2400 here lest we lose sharing.
2403 addWanted :: WantSCs -> Avails -> Inst -> LHsExpr TcId -> [Inst] -> TcM Avails
2404 addWanted want_scs avails wanted rhs_expr wanteds
2405 = addAvailAndSCs want_scs avails wanted avail
2407 avail = Rhs rhs_expr wanteds
2409 addGiven :: Avails -> Inst -> TcM Avails
2410 addGiven avails given = addAvailAndSCs AddSCs avails given (Given given)
2411 -- Always add superclasses for 'givens'
2413 -- No ASSERT( not (given `elemAvails` avails) ) because in an instance
2414 -- decl for Ord t we can add both Ord t and Eq t as 'givens',
2415 -- so the assert isn't true
2419 addIrred :: WantSCs -> Avails -> Inst -> TcM Avails
2420 addIrred want_scs avails irred = ASSERT2( not (irred `elemAvails` avails), ppr irred $$ ppr avails )
2421 addAvailAndSCs want_scs avails irred IsIrred
2423 addAvailAndSCs :: WantSCs -> Avails -> Inst -> AvailHow -> TcM Avails
2424 addAvailAndSCs want_scs avails inst avail
2425 | not (isClassDict inst) = extendAvails avails inst avail
2426 | NoSCs <- want_scs = extendAvails avails inst avail
2427 | otherwise = do { traceTc (text "addAvailAndSCs" <+> vcat [ppr inst, ppr deps])
2428 ; avails' <- extendAvails avails inst avail
2429 ; addSCs is_loop avails' inst }
2431 is_loop pred = any (`tcEqType` mkPredTy pred) dep_tys
2432 -- Note: this compares by *type*, not by Unique
2433 deps = findAllDeps (unitVarSet (instToVar inst)) avail
2434 dep_tys = map idType (varSetElems deps)
2436 findAllDeps :: IdSet -> AvailHow -> IdSet
2437 -- Find all the Insts that this one depends on
2438 -- See Note [SUPERCLASS-LOOP 2]
2439 -- Watch out, though. Since the avails may contain loops
2440 -- (see Note [RECURSIVE DICTIONARIES]), so we need to track the ones we've seen so far
2441 findAllDeps so_far (Rhs _ kids) = foldl find_all so_far kids
2442 findAllDeps so_far other = so_far
2444 find_all :: IdSet -> Inst -> IdSet
2446 | isEqInst kid = so_far
2447 | kid_id `elemVarSet` so_far = so_far
2448 | Just avail <- findAvail avails kid = findAllDeps so_far' avail
2449 | otherwise = so_far'
2451 so_far' = extendVarSet so_far kid_id -- Add the new kid to so_far
2452 kid_id = instToId kid
2454 addSCs :: (TcPredType -> Bool) -> Avails -> Inst -> TcM Avails
2455 -- Add all the superclasses of the Inst to Avails
2456 -- The first param says "don't do this because the original thing
2457 -- depends on this one, so you'd build a loop"
2458 -- Invariant: the Inst is already in Avails.
2460 addSCs is_loop avails dict
2461 = ASSERT( isDict dict )
2462 do { sc_dicts <- newDictBndrs (instLoc dict) sc_theta'
2463 ; foldlM add_sc avails (zipEqual "add_scs" sc_dicts sc_sels) }
2465 (clas, tys) = getDictClassTys dict
2466 (tyvars, sc_theta, sc_sels, _) = classBigSig clas
2467 sc_theta' = filter (not . isEqPred) $
2468 substTheta (zipTopTvSubst tyvars tys) sc_theta
2470 add_sc avails (sc_dict, sc_sel)
2471 | is_loop (dictPred sc_dict) = return avails -- See Note [SUPERCLASS-LOOP 2]
2472 | is_given sc_dict = return avails
2473 | otherwise = do { avails' <- extendAvails avails sc_dict (Rhs sc_sel_rhs [dict])
2474 ; addSCs is_loop avails' sc_dict }
2476 sc_sel_rhs = L (instSpan dict) (HsWrap co_fn (HsVar sc_sel))
2477 co_fn = WpApp (instToVar dict) <.> mkWpTyApps tys
2479 is_given :: Inst -> Bool
2480 is_given sc_dict = case findAvail avails sc_dict of
2481 Just (Given _) -> True -- Given is cheaper than superclass selection
2484 -- From the a set of insts obtain all equalities that (transitively) occur in
2485 -- superclass contexts of class constraints (aka the ancestor equalities).
2487 ancestorEqualities :: [Inst] -> TcM [Inst]
2489 = mapM mkWantedEqInst -- turn only equality predicates..
2490 . filter isEqPred -- ..into wanted equality insts
2492 . addAEsToBag emptyBag -- collect the superclass constraints..
2493 . map dictPred -- ..of all predicates in a bag
2494 . filter isClassDict
2496 addAEsToBag :: Bag PredType -> [PredType] -> Bag PredType
2497 addAEsToBag bag [] = bag
2498 addAEsToBag bag (pred:preds)
2499 | pred `elemBag` bag = addAEsToBag bag preds
2500 | isEqPred pred = addAEsToBag bagWithPred preds
2501 | isClassPred pred = addAEsToBag bagWithPred predsWithSCs
2502 | otherwise = addAEsToBag bag preds
2504 bagWithPred = bag `snocBag` pred
2505 predsWithSCs = preds ++ substTheta (zipTopTvSubst tyvars tys) sc_theta
2507 (tyvars, sc_theta, _, _) = classBigSig clas
2508 (clas, tys) = getClassPredTys pred
2512 %************************************************************************
2514 \section{tcSimplifyTop: defaulting}
2516 %************************************************************************
2519 @tcSimplifyTop@ is called once per module to simplify all the constant
2520 and ambiguous Insts.
2522 We need to be careful of one case. Suppose we have
2524 instance Num a => Num (Foo a b) where ...
2526 and @tcSimplifyTop@ is given a constraint (Num (Foo x y)). Then it'll simplify
2527 to (Num x), and default x to Int. But what about y??
2529 It's OK: the final zonking stage should zap y to (), which is fine.
2533 tcSimplifyTop, tcSimplifyInteractive :: [Inst] -> TcM TcDictBinds
2534 tcSimplifyTop wanteds
2535 = tc_simplify_top doc False wanteds
2537 doc = text "tcSimplifyTop"
2539 tcSimplifyInteractive wanteds
2540 = tc_simplify_top doc True wanteds
2542 doc = text "tcSimplifyInteractive"
2544 -- The TcLclEnv should be valid here, solely to improve
2545 -- error message generation for the monomorphism restriction
2546 tc_simplify_top doc interactive wanteds
2547 = do { dflags <- getDOpts
2548 ; wanteds <- zonkInsts wanteds
2549 ; mapM_ zonkTopTyVar (varSetElems (tyVarsOfInsts wanteds))
2551 ; traceTc (text "tc_simplify_top 0: " <+> ppr wanteds)
2552 ; (irreds1, binds1) <- tryHardCheckLoop doc1 wanteds
2553 -- ; (irreds1, binds1) <- gentleInferLoop doc1 wanteds
2554 ; traceTc (text "tc_simplify_top 1: " <+> ppr irreds1)
2555 ; (irreds2, binds2) <- approximateImplications doc2 (\d -> True) irreds1
2556 ; traceTc (text "tc_simplify_top 2: " <+> ppr irreds2)
2558 -- Use the defaulting rules to do extra unification
2559 -- NB: irreds2 are already zonked
2560 ; (irreds3, binds3) <- disambiguate doc3 interactive dflags irreds2
2562 -- Deal with implicit parameters
2563 ; let (bad_ips, non_ips) = partition isIPDict irreds3
2564 (ambigs, others) = partition isTyVarDict non_ips
2566 ; topIPErrs bad_ips -- Can arise from f :: Int -> Int
2568 ; addNoInstanceErrs others
2569 ; addTopAmbigErrs ambigs
2571 ; return (binds1 `unionBags` binds2 `unionBags` binds3) }
2573 doc1 = doc <+> ptext SLIT("(first round)")
2574 doc2 = doc <+> ptext SLIT("(approximate)")
2575 doc3 = doc <+> ptext SLIT("(disambiguate)")
2578 If a dictionary constrains a type variable which is
2579 * not mentioned in the environment
2580 * and not mentioned in the type of the expression
2581 then it is ambiguous. No further information will arise to instantiate
2582 the type variable; nor will it be generalised and turned into an extra
2583 parameter to a function.
2585 It is an error for this to occur, except that Haskell provided for
2586 certain rules to be applied in the special case of numeric types.
2588 * at least one of its classes is a numeric class, and
2589 * all of its classes are numeric or standard
2590 then the type variable can be defaulted to the first type in the
2591 default-type list which is an instance of all the offending classes.
2593 So here is the function which does the work. It takes the ambiguous
2594 dictionaries and either resolves them (producing bindings) or
2595 complains. It works by splitting the dictionary list by type
2596 variable, and using @disambigOne@ to do the real business.
2598 @disambigOne@ assumes that its arguments dictionaries constrain all
2599 the same type variable.
2601 ADR Comment 20/6/94: I've changed the @CReturnable@ case to default to
2602 @()@ instead of @Int@. I reckon this is the Right Thing to do since
2603 the most common use of defaulting is code like:
2605 _ccall_ foo `seqPrimIO` bar
2607 Since we're not using the result of @foo@, the result if (presumably)
2611 disambiguate :: SDoc -> Bool -> DynFlags -> [Inst] -> TcM ([Inst], TcDictBinds)
2612 -- Just does unification to fix the default types
2613 -- The Insts are assumed to be pre-zonked
2614 disambiguate doc interactive dflags insts
2616 = return (insts, emptyBag)
2618 | null defaultable_groups
2619 = do { traceTc (text "disambigutate, no defaultable groups" <+> vcat [ppr unaries, ppr insts, ppr bad_tvs, ppr defaultable_groups])
2620 ; return (insts, emptyBag) }
2623 = do { -- Figure out what default types to use
2624 default_tys <- getDefaultTys extended_defaulting ovl_strings
2626 ; traceTc (text "disambiguate1" <+> vcat [ppr insts, ppr unaries, ppr bad_tvs, ppr defaultable_groups])
2627 ; mapM_ (disambigGroup default_tys) defaultable_groups
2629 -- disambigGroup does unification, hence try again
2630 ; tryHardCheckLoop doc insts }
2633 extended_defaulting = interactive || dopt Opt_ExtendedDefaultRules dflags
2634 ovl_strings = dopt Opt_OverloadedStrings dflags
2636 unaries :: [(Inst, Class, TcTyVar)] -- (C tv) constraints
2637 bad_tvs :: TcTyVarSet -- Tyvars mentioned by *other* constraints
2638 (unaries, bad_tvs_s) = partitionWith find_unary insts
2639 bad_tvs = unionVarSets bad_tvs_s
2641 -- Finds unary type-class constraints
2642 find_unary d@(Dict {tci_pred = ClassP cls [ty]})
2643 | Just tv <- tcGetTyVar_maybe ty = Left (d,cls,tv)
2644 find_unary inst = Right (tyVarsOfInst inst)
2646 -- Group by type variable
2647 defaultable_groups :: [[(Inst,Class,TcTyVar)]]
2648 defaultable_groups = filter defaultable_group (equivClasses cmp_tv unaries)
2649 cmp_tv (_,_,tv1) (_,_,tv2) = tv1 `compare` tv2
2651 defaultable_group :: [(Inst,Class,TcTyVar)] -> Bool
2652 defaultable_group ds@((_,_,tv):_)
2653 = isTyConableTyVar tv -- Note [Avoiding spurious errors]
2654 && not (tv `elemVarSet` bad_tvs)
2655 && defaultable_classes [c | (_,c,_) <- ds]
2656 defaultable_group [] = panic "defaultable_group"
2658 defaultable_classes clss
2659 | extended_defaulting = any isInteractiveClass clss
2660 | otherwise = all is_std_class clss && (any is_num_class clss)
2662 -- In interactive mode, or with -fextended-default-rules,
2663 -- we default Show a to Show () to avoid graututious errors on "show []"
2664 isInteractiveClass cls
2665 = is_num_class cls || (classKey cls `elem` [showClassKey, eqClassKey, ordClassKey])
2667 is_num_class cls = isNumericClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2668 -- is_num_class adds IsString to the standard numeric classes,
2669 -- when -foverloaded-strings is enabled
2671 is_std_class cls = isStandardClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2672 -- Similarly is_std_class
2674 -----------------------
2675 disambigGroup :: [Type] -- The default types
2676 -> [(Inst,Class,TcTyVar)] -- All standard classes of form (C a)
2677 -> TcM () -- Just does unification, to fix the default types
2679 disambigGroup default_tys dicts
2680 = try_default default_tys
2682 (_,_,tyvar) = ASSERT(not (null dicts)) head dicts -- Should be non-empty
2683 classes = [c | (_,c,_) <- dicts]
2685 try_default [] = return ()
2686 try_default (default_ty : default_tys)
2687 = tryTcLIE_ (try_default default_tys) $
2688 do { tcSimplifyDefault [mkClassPred clas [default_ty] | clas <- classes]
2689 -- This may fail; then the tryTcLIE_ kicks in
2690 -- Failure here is caused by there being no type in the
2691 -- default list which can satisfy all the ambiguous classes.
2692 -- For example, if Real a is reqd, but the only type in the
2693 -- default list is Int.
2695 -- After this we can't fail
2696 ; warnDefault dicts default_ty
2697 ; unifyType default_ty (mkTyVarTy tyvar)
2698 ; return () -- TOMDO: do something with the coercion
2702 -----------------------
2703 getDefaultTys :: Bool -> Bool -> TcM [Type]
2704 getDefaultTys extended_deflts ovl_strings
2705 = do { mb_defaults <- getDeclaredDefaultTys
2706 ; case mb_defaults of {
2707 Just tys -> return tys ; -- User-supplied defaults
2710 -- No use-supplied default
2711 -- Use [Integer, Double], plus modifications
2712 { integer_ty <- tcMetaTy integerTyConName
2713 ; checkWiredInTyCon doubleTyCon
2714 ; string_ty <- tcMetaTy stringTyConName
2715 ; return (opt_deflt extended_deflts unitTy
2716 -- Note [Default unitTy]
2718 [integer_ty,doubleTy]
2720 opt_deflt ovl_strings string_ty) } } }
2722 opt_deflt True ty = [ty]
2723 opt_deflt False ty = []
2726 Note [Default unitTy]
2727 ~~~~~~~~~~~~~~~~~~~~~
2728 In interative mode (or with -fextended-default-rules) we add () as the first type we
2729 try when defaulting. This has very little real impact, except in the following case.
2731 Text.Printf.printf "hello"
2732 This has type (forall a. IO a); it prints "hello", and returns 'undefined'. We don't
2733 want the GHCi repl loop to try to print that 'undefined'. The neatest thing is to
2734 default the 'a' to (), rather than to Integer (which is what would otherwise happen;
2735 and then GHCi doesn't attempt to print the (). So in interactive mode, we add
2736 () to the list of defaulting types. See Trac #1200.
2738 Note [Avoiding spurious errors]
2739 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2740 When doing the unification for defaulting, we check for skolem
2741 type variables, and simply don't default them. For example:
2742 f = (*) -- Monomorphic
2743 g :: Num a => a -> a
2745 Here, we get a complaint when checking the type signature for g,
2746 that g isn't polymorphic enough; but then we get another one when
2747 dealing with the (Num a) context arising from f's definition;
2748 we try to unify a with Int (to default it), but find that it's
2749 already been unified with the rigid variable from g's type sig
2752 %************************************************************************
2754 \subsection[simple]{@Simple@ versions}
2756 %************************************************************************
2758 Much simpler versions when there are no bindings to make!
2760 @tcSimplifyThetas@ simplifies class-type constraints formed by
2761 @deriving@ declarations and when specialising instances. We are
2762 only interested in the simplified bunch of class/type constraints.
2764 It simplifies to constraints of the form (C a b c) where
2765 a,b,c are type variables. This is required for the context of
2766 instance declarations.
2769 tcSimplifyDeriv :: InstOrigin
2771 -> ThetaType -- Wanted
2772 -> TcM ThetaType -- Needed
2773 -- Given instance (wanted) => C inst_ty
2774 -- Simplify 'wanted' as much as possible
2776 tcSimplifyDeriv orig tyvars theta
2777 = do { (tvs, _, tenv) <- tcInstTyVars tyvars
2778 -- The main loop may do unification, and that may crash if
2779 -- it doesn't see a TcTyVar, so we have to instantiate. Sigh
2780 -- ToDo: what if two of them do get unified?
2781 ; wanteds <- newDictBndrsO orig (substTheta tenv theta)
2782 ; (irreds, _) <- tryHardCheckLoop doc wanteds
2784 ; let (tv_dicts, others) = partition ok irreds
2785 ; addNoInstanceErrs others
2786 -- See Note [Exotic derived instance contexts] in TcMType
2788 ; let rev_env = zipTopTvSubst tvs (mkTyVarTys tyvars)
2789 simpl_theta = substTheta rev_env (map dictPred tv_dicts)
2790 -- This reverse-mapping is a pain, but the result
2791 -- should mention the original TyVars not TcTyVars
2793 ; return simpl_theta }
2795 doc = ptext SLIT("deriving classes for a data type")
2797 ok dict | isDict dict = validDerivPred (dictPred dict)
2802 @tcSimplifyDefault@ just checks class-type constraints, essentially;
2803 used with \tr{default} declarations. We are only interested in
2804 whether it worked or not.
2807 tcSimplifyDefault :: ThetaType -- Wanted; has no type variables in it
2810 tcSimplifyDefault theta = do
2811 wanteds <- newDictBndrsO DefaultOrigin theta
2812 (irreds, _) <- tryHardCheckLoop doc wanteds
2813 addNoInstanceErrs irreds
2817 traceTc (ptext SLIT("tcSimplifyDefault failing")) >> failM
2819 doc = ptext SLIT("default declaration")
2823 %************************************************************************
2825 \section{Errors and contexts}
2827 %************************************************************************
2829 ToDo: for these error messages, should we note the location as coming
2830 from the insts, or just whatever seems to be around in the monad just
2834 groupErrs :: ([Inst] -> TcM ()) -- Deal with one group
2835 -> [Inst] -- The offending Insts
2837 -- Group together insts with the same origin
2838 -- We want to report them together in error messages
2840 groupErrs report_err []
2842 groupErrs report_err (inst:insts)
2843 = do { do_one (inst:friends)
2844 ; groupErrs report_err others }
2846 -- (It may seem a bit crude to compare the error messages,
2847 -- but it makes sure that we combine just what the user sees,
2848 -- and it avoids need equality on InstLocs.)
2849 (friends, others) = partition is_friend insts
2850 loc_msg = showSDoc (pprInstLoc (instLoc inst))
2851 is_friend friend = showSDoc (pprInstLoc (instLoc friend)) == loc_msg
2852 do_one insts = addInstCtxt (instLoc (head insts)) (report_err insts)
2853 -- Add location and context information derived from the Insts
2855 -- Add the "arising from..." part to a message about bunch of dicts
2856 addInstLoc :: [Inst] -> Message -> Message
2857 addInstLoc insts msg = msg $$ nest 2 (pprInstArising (head insts))
2859 addTopIPErrs :: [Name] -> [Inst] -> TcM ()
2860 addTopIPErrs bndrs []
2862 addTopIPErrs bndrs ips
2863 = do { dflags <- getDOpts
2864 ; addErrTcM (tidy_env, mk_msg dflags tidy_ips) }
2866 (tidy_env, tidy_ips) = tidyInsts ips
2868 = vcat [sep [ptext SLIT("Implicit parameters escape from"),
2869 nest 2 (ptext SLIT("the monomorphic top-level binding")
2870 <> plural bndrs <+> ptext SLIT("of")
2871 <+> pprBinders bndrs <> colon)],
2872 nest 2 (vcat (map ppr_ip ips)),
2873 monomorphism_fix dflags]
2874 ppr_ip ip = pprPred (dictPred ip) <+> pprInstArising ip
2876 topIPErrs :: [Inst] -> TcM ()
2878 = groupErrs report tidy_dicts
2880 (tidy_env, tidy_dicts) = tidyInsts dicts
2881 report dicts = addErrTcM (tidy_env, mk_msg dicts)
2882 mk_msg dicts = addInstLoc dicts (ptext SLIT("Unbound implicit parameter") <>
2883 plural tidy_dicts <+> pprDictsTheta tidy_dicts)
2885 addNoInstanceErrs :: [Inst] -- Wanted (can include implications)
2887 addNoInstanceErrs insts
2888 = do { let (tidy_env, tidy_insts) = tidyInsts insts
2889 ; reportNoInstances tidy_env Nothing tidy_insts }
2893 -> Maybe (InstLoc, [Inst]) -- Context
2894 -- Nothing => top level
2895 -- Just (d,g) => d describes the construct
2897 -> [Inst] -- What is wanted (can include implications)
2900 reportNoInstances tidy_env mb_what insts
2901 = groupErrs (report_no_instances tidy_env mb_what) insts
2903 report_no_instances tidy_env mb_what insts
2904 = do { inst_envs <- tcGetInstEnvs
2905 ; let (implics, insts1) = partition isImplicInst insts
2906 (insts2, overlaps) = partitionWith (check_overlap inst_envs) insts1
2907 (eqInsts, insts3) = partition isEqInst insts2
2908 ; traceTc (text "reportNoInstances" <+> vcat
2909 [ppr insts, ppr implics, ppr insts1, ppr insts2])
2910 ; mapM_ complain_implic implics
2911 ; mapM_ (\doc -> addErrTcM (tidy_env, doc)) overlaps
2912 ; groupErrs complain_no_inst insts3
2913 ; mapM_ (addErrTcM . mk_eq_err) eqInsts
2916 complain_no_inst insts = addErrTcM (tidy_env, mk_no_inst_err insts)
2918 complain_implic inst -- Recurse!
2919 = reportNoInstances tidy_env
2920 (Just (tci_loc inst, tci_given inst))
2923 check_overlap :: (InstEnv,InstEnv) -> Inst -> Either Inst SDoc
2924 -- Right msg => overlap message
2925 -- Left inst => no instance
2926 check_overlap inst_envs wanted
2927 | not (isClassDict wanted) = Left wanted
2929 = case lookupInstEnv inst_envs clas tys of
2930 -- The case of exactly one match and no unifiers means a
2931 -- successful lookup. That can't happen here, because dicts
2932 -- only end up here if they didn't match in Inst.lookupInst
2934 ([m],[]) -> pprPanic "reportNoInstance" (ppr wanted)
2936 ([], _) -> Left wanted -- No match
2937 res -> Right (mk_overlap_msg wanted res)
2939 (clas,tys) = getDictClassTys wanted
2941 mk_overlap_msg dict (matches, unifiers)
2942 = ASSERT( not (null matches) )
2943 vcat [ addInstLoc [dict] ((ptext SLIT("Overlapping instances for")
2944 <+> pprPred (dictPred dict))),
2945 sep [ptext SLIT("Matching instances") <> colon,
2946 nest 2 (vcat [pprInstances ispecs, pprInstances unifiers])],
2947 if not (isSingleton matches)
2948 then -- Two or more matches
2950 else -- One match, plus some unifiers
2951 ASSERT( not (null unifiers) )
2952 parens (vcat [ptext SLIT("The choice depends on the instantiation of") <+>
2953 quotes (pprWithCommas ppr (varSetElems (tyVarsOfInst dict))),
2954 ptext SLIT("To pick the first instance above, use -fallow-incoherent-instances"),
2955 ptext SLIT("when compiling the other instance declarations")])]
2957 ispecs = [ispec | (ispec, _) <- matches]
2959 mk_eq_err :: Inst -> (TidyEnv, SDoc)
2960 mk_eq_err inst = misMatchMsg tidy_env (eqInstTys inst)
2962 mk_no_inst_err insts
2963 | null insts = empty
2965 | Just (loc, givens) <- mb_what, -- Nested (type signatures, instance decls)
2966 not (isEmptyVarSet (tyVarsOfInsts insts))
2967 = vcat [ addInstLoc insts $
2968 sep [ ptext SLIT("Could not deduce") <+> pprDictsTheta insts
2969 , nest 2 $ ptext SLIT("from the context") <+> pprDictsTheta givens]
2970 , show_fixes (fix1 loc : fixes2) ]
2972 | otherwise -- Top level
2973 = vcat [ addInstLoc insts $
2974 ptext SLIT("No instance") <> plural insts
2975 <+> ptext SLIT("for") <+> pprDictsTheta insts
2976 , show_fixes fixes2 ]
2979 fix1 loc = sep [ ptext SLIT("add") <+> pprDictsTheta insts
2980 <+> ptext SLIT("to the context of"),
2981 nest 2 (ppr (instLocOrigin loc)) ]
2982 -- I'm not sure it helps to add the location
2983 -- nest 2 (ptext SLIT("at") <+> ppr (instLocSpan loc)) ]
2985 fixes2 | null instance_dicts = []
2986 | otherwise = [sep [ptext SLIT("add an instance declaration for"),
2987 pprDictsTheta instance_dicts]]
2988 instance_dicts = [d | d <- insts, isClassDict d, not (isTyVarDict d)]
2989 -- Insts for which it is worth suggesting an adding an instance declaration
2990 -- Exclude implicit parameters, and tyvar dicts
2992 show_fixes :: [SDoc] -> SDoc
2993 show_fixes [] = empty
2994 show_fixes (f:fs) = sep [ptext SLIT("Possible fix:"),
2995 nest 2 (vcat (f : map (ptext SLIT("or") <+>) fs))]
2997 addTopAmbigErrs dicts
2998 -- Divide into groups that share a common set of ambiguous tyvars
2999 = ifErrsM (return ()) $ -- Only report ambiguity if no other errors happened
3000 -- See Note [Avoiding spurious errors]
3001 mapM_ report (equivClasses cmp [(d, tvs_of d) | d <- tidy_dicts])
3003 (tidy_env, tidy_dicts) = tidyInsts dicts
3005 tvs_of :: Inst -> [TcTyVar]
3006 tvs_of d = varSetElems (tyVarsOfInst d)
3007 cmp (_,tvs1) (_,tvs2) = tvs1 `compare` tvs2
3009 report :: [(Inst,[TcTyVar])] -> TcM ()
3010 report pairs@((inst,tvs) : _) = do -- The pairs share a common set of ambiguous tyvars
3011 (tidy_env, mono_msg) <- mkMonomorphismMsg tidy_env tvs
3012 setSrcSpan (instSpan inst) $
3013 -- the location of the first one will do for the err message
3014 addErrTcM (tidy_env, msg $$ mono_msg)
3016 dicts = map fst pairs
3017 msg = sep [text "Ambiguous type variable" <> plural tvs <+>
3018 pprQuotedList tvs <+> in_msg,
3019 nest 2 (pprDictsInFull dicts)]
3020 in_msg = text "in the constraint" <> plural dicts <> colon
3021 report [] = panic "addTopAmbigErrs"
3024 mkMonomorphismMsg :: TidyEnv -> [TcTyVar] -> TcM (TidyEnv, Message)
3025 -- There's an error with these Insts; if they have free type variables
3026 -- it's probably caused by the monomorphism restriction.
3027 -- Try to identify the offending variable
3028 -- ASSUMPTION: the Insts are fully zonked
3029 mkMonomorphismMsg tidy_env inst_tvs
3030 = do { dflags <- getDOpts
3031 ; (tidy_env, docs) <- findGlobals (mkVarSet inst_tvs) tidy_env
3032 ; return (tidy_env, mk_msg dflags docs) }
3034 mk_msg _ _ | any isRuntimeUnk inst_tvs
3035 = vcat [ptext SLIT("Cannot resolve unknown runtime types:") <+>
3036 (pprWithCommas ppr inst_tvs),
3037 ptext SLIT("Use :print or :force to determine these types")]
3038 mk_msg _ [] = ptext SLIT("Probable fix: add a type signature that fixes these type variable(s)")
3039 -- This happens in things like
3040 -- f x = show (read "foo")
3041 -- where monomorphism doesn't play any role
3043 = vcat [ptext SLIT("Possible cause: the monomorphism restriction applied to the following:"),
3045 monomorphism_fix dflags]
3047 monomorphism_fix :: DynFlags -> SDoc
3048 monomorphism_fix dflags
3049 = ptext SLIT("Probable fix:") <+> vcat
3050 [ptext SLIT("give these definition(s) an explicit type signature"),
3051 if dopt Opt_MonomorphismRestriction dflags
3052 then ptext SLIT("or use -fno-monomorphism-restriction")
3053 else empty] -- Only suggest adding "-fno-monomorphism-restriction"
3054 -- if it is not already set!
3056 warnDefault ups default_ty = do
3057 warn_flag <- doptM Opt_WarnTypeDefaults
3058 addInstCtxt (instLoc (head (dicts))) (warnTc warn_flag warn_msg)
3060 dicts = [d | (d,_,_) <- ups]
3063 (_, tidy_dicts) = tidyInsts dicts
3064 warn_msg = vcat [ptext SLIT("Defaulting the following constraint(s) to type") <+>
3065 quotes (ppr default_ty),
3066 pprDictsInFull tidy_dicts]
3068 reduceDepthErr n stack
3069 = vcat [ptext SLIT("Context reduction stack overflow; size =") <+> int n,
3070 ptext SLIT("Use -fcontext-stack=N to increase stack size to N"),
3071 nest 4 (pprStack stack)]
3073 pprStack stack = vcat (map pprInstInFull stack)