2 % (c) The University of Glasgow 2006
3 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
10 tcSimplifyInfer, tcSimplifyInferCheck,
11 tcSimplifyCheck, tcSimplifyRestricted,
12 tcSimplifyRuleLhs, tcSimplifyIPs,
13 tcSimplifySuperClasses,
14 tcSimplifyTop, tcSimplifyInteractive,
15 tcSimplifyBracket, tcSimplifyCheckPat,
17 tcSimplifyDeriv, tcSimplifyDefault,
25 #include "HsVersions.h"
27 import {-# SOURCE #-} TcUnify( unifyType )
31 import TcHsSyn ( hsLPatType )
39 import DsUtils -- Big-tuple functions
67 %************************************************************************
71 %************************************************************************
73 --------------------------------------
74 Notes on functional dependencies (a bug)
75 --------------------------------------
82 instance D a b => C a b -- Undecidable
83 -- (Not sure if it's crucial to this eg)
84 f :: C a b => a -> Bool
87 g :: C a b => a -> Bool
90 Here f typechecks, but g does not!! Reason: before doing improvement,
91 we reduce the (C a b1) constraint from the call of f to (D a b1).
93 Here is a more complicated example:
96 > class Foo a b | a->b
98 > class Bar a b | a->b
102 > instance Bar Obj Obj
104 > instance (Bar a b) => Foo a b
106 > foo:: (Foo a b) => a -> String
109 > runFoo:: (forall a b. (Foo a b) => a -> w) -> w
115 Could not deduce (Bar a b) from the context (Foo a b)
116 arising from use of `foo' at <interactive>:1
118 Add (Bar a b) to the expected type of an expression
119 In the first argument of `runFoo', namely `foo'
120 In the definition of `it': it = runFoo foo
122 Why all of the sudden does GHC need the constraint Bar a b? The
123 function foo didn't ask for that...
126 The trouble is that to type (runFoo foo), GHC has to solve the problem:
128 Given constraint Foo a b
129 Solve constraint Foo a b'
131 Notice that b and b' aren't the same. To solve this, just do
132 improvement and then they are the same. But GHC currently does
137 That is usually fine, but it isn't here, because it sees that Foo a b is
138 not the same as Foo a b', and so instead applies the instance decl for
139 instance Bar a b => Foo a b. And that's where the Bar constraint comes
142 The Right Thing is to improve whenever the constraint set changes at
143 all. Not hard in principle, but it'll take a bit of fiddling to do.
145 Note [Choosing which variables to quantify]
146 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
147 Suppose we are about to do a generalisation step. We have in our hand
150 T the type of the RHS
151 C the constraints from that RHS
153 The game is to figure out
155 Q the set of type variables over which to quantify
156 Ct the constraints we will *not* quantify over
157 Cq the constraints we will quantify over
159 So we're going to infer the type
163 and float the constraints Ct further outwards.
165 Here are the things that *must* be true:
167 (A) Q intersect fv(G) = EMPTY limits how big Q can be
168 (B) Q superset fv(Cq union T) \ oclose(fv(G),C) limits how small Q can be
170 (A) says we can't quantify over a variable that's free in the environment.
171 (B) says we must quantify over all the truly free variables in T, else
172 we won't get a sufficiently general type.
174 We do not *need* to quantify over any variable that is fixed by the
175 free vars of the environment G.
177 BETWEEN THESE TWO BOUNDS, ANY Q WILL DO!
179 Example: class H x y | x->y where ...
181 fv(G) = {a} C = {H a b, H c d}
184 (A) Q intersect {a} is empty
185 (B) Q superset {a,b,c,d} \ oclose({a}, C) = {a,b,c,d} \ {a,b} = {c,d}
187 So Q can be {c,d}, {b,c,d}
189 In particular, it's perfectly OK to quantify over more type variables
190 than strictly necessary; there is no need to quantify over 'b', since
191 it is determined by 'a' which is free in the envt, but it's perfectly
192 OK to do so. However we must not quantify over 'a' itself.
194 Other things being equal, however, we'd like to quantify over as few
195 variables as possible: smaller types, fewer type applications, more
196 constraints can get into Ct instead of Cq. Here's a good way to
199 Q = grow( fv(T), C ) \ oclose( fv(G), C )
201 That is, quantify over all variable that that MIGHT be fixed by the
202 call site (which influences T), but which aren't DEFINITELY fixed by
203 G. This choice definitely quantifies over enough type variables,
204 albeit perhaps too many.
206 Why grow( fv(T), C ) rather than fv(T)? Consider
208 class H x y | x->y where ...
213 If we used fv(T) = {c} we'd get the type
215 forall c. H c d => c -> b
217 And then if the fn was called at several different c's, each of
218 which fixed d differently, we'd get a unification error, because
219 d isn't quantified. Solution: quantify d. So we must quantify
220 everything that might be influenced by c.
222 Why not oclose( fv(T), C )? Because we might not be able to see
223 all the functional dependencies yet:
225 class H x y | x->y where ...
226 instance H x y => Eq (T x y) where ...
231 Now oclose(fv(T),C) = {c}, because the functional dependency isn't
232 apparent yet, and that's wrong. We must really quantify over d too.
234 There really isn't any point in quantifying over any more than
235 grow( fv(T), C ), because the call sites can't possibly influence
236 any other type variables.
240 -------------------------------------
242 -------------------------------------
244 It's very hard to be certain when a type is ambiguous. Consider
248 instance H x y => K (x,y)
250 Is this type ambiguous?
251 forall a b. (K (a,b), Eq b) => a -> a
253 Looks like it! But if we simplify (K (a,b)) we get (H a b) and
254 now we see that a fixes b. So we can't tell about ambiguity for sure
255 without doing a full simplification. And even that isn't possible if
256 the context has some free vars that may get unified. Urgle!
258 Here's another example: is this ambiguous?
259 forall a b. Eq (T b) => a -> a
260 Not if there's an insance decl (with no context)
261 instance Eq (T b) where ...
263 You may say of this example that we should use the instance decl right
264 away, but you can't always do that:
266 class J a b where ...
267 instance J Int b where ...
269 f :: forall a b. J a b => a -> a
271 (Notice: no functional dependency in J's class decl.)
272 Here f's type is perfectly fine, provided f is only called at Int.
273 It's premature to complain when meeting f's signature, or even
274 when inferring a type for f.
278 However, we don't *need* to report ambiguity right away. It'll always
279 show up at the call site.... and eventually at main, which needs special
280 treatment. Nevertheless, reporting ambiguity promptly is an excellent thing.
282 So here's the plan. We WARN about probable ambiguity if
284 fv(Cq) is not a subset of oclose(fv(T) union fv(G), C)
286 (all tested before quantification).
287 That is, all the type variables in Cq must be fixed by the the variables
288 in the environment, or by the variables in the type.
290 Notice that we union before calling oclose. Here's an example:
292 class J a b c | a b -> c
296 forall b c. (J a b c) => b -> b
298 Only if we union {a} from G with {b} from T before using oclose,
299 do we see that c is fixed.
301 It's a bit vague exactly which C we should use for this oclose call. If we
302 don't fix enough variables we might complain when we shouldn't (see
303 the above nasty example). Nothing will be perfect. That's why we can
304 only issue a warning.
307 Can we ever be *certain* about ambiguity? Yes: if there's a constraint
309 c in C such that fv(c) intersect (fv(G) union fv(T)) = EMPTY
311 then c is a "bubble"; there's no way it can ever improve, and it's
312 certainly ambiguous. UNLESS it is a constant (sigh). And what about
317 instance H x y => K (x,y)
319 Is this type ambiguous?
320 forall a b. (K (a,b), Eq b) => a -> a
322 Urk. The (Eq b) looks "definitely ambiguous" but it isn't. What we are after
323 is a "bubble" that's a set of constraints
325 Cq = Ca union Cq' st fv(Ca) intersect (fv(Cq') union fv(T) union fv(G)) = EMPTY
327 Hence another idea. To decide Q start with fv(T) and grow it
328 by transitive closure in Cq (no functional dependencies involved).
329 Now partition Cq using Q, leaving the definitely-ambiguous and probably-ok.
330 The definitely-ambiguous can then float out, and get smashed at top level
331 (which squashes out the constants, like Eq (T a) above)
334 --------------------------------------
335 Notes on principal types
336 --------------------------------------
341 f x = let g y = op (y::Int) in True
343 Here the principal type of f is (forall a. a->a)
344 but we'll produce the non-principal type
345 f :: forall a. C Int => a -> a
348 --------------------------------------
349 The need for forall's in constraints
350 --------------------------------------
352 [Exchange on Haskell Cafe 5/6 Dec 2000]
354 class C t where op :: t -> Bool
355 instance C [t] where op x = True
357 p y = (let f :: c -> Bool; f x = op (y >> return x) in f, y ++ [])
358 q y = (y ++ [], let f :: c -> Bool; f x = op (y >> return x) in f)
360 The definitions of p and q differ only in the order of the components in
361 the pair on their right-hand sides. And yet:
363 ghc and "Typing Haskell in Haskell" reject p, but accept q;
364 Hugs rejects q, but accepts p;
365 hbc rejects both p and q;
366 nhc98 ... (Malcolm, can you fill in the blank for us!).
368 The type signature for f forces context reduction to take place, and
369 the results of this depend on whether or not the type of y is known,
370 which in turn depends on which component of the pair the type checker
373 Solution: if y::m a, float out the constraints
374 Monad m, forall c. C (m c)
375 When m is later unified with [], we can solve both constraints.
378 --------------------------------------
379 Notes on implicit parameters
380 --------------------------------------
382 Note [Inheriting implicit parameters]
383 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
388 where f is *not* a top-level binding.
389 From the RHS of f we'll get the constraint (?y::Int).
390 There are two types we might infer for f:
394 (so we get ?y from the context of f's definition), or
396 f :: (?y::Int) => Int -> Int
398 At first you might think the first was better, becuase then
399 ?y behaves like a free variable of the definition, rather than
400 having to be passed at each call site. But of course, the WHOLE
401 IDEA is that ?y should be passed at each call site (that's what
402 dynamic binding means) so we'd better infer the second.
404 BOTTOM LINE: when *inferring types* you *must* quantify
405 over implicit parameters. See the predicate isFreeWhenInferring.
408 Note [Implicit parameters and ambiguity]
409 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
410 Only a *class* predicate can give rise to ambiguity
411 An *implicit parameter* cannot. For example:
412 foo :: (?x :: [a]) => Int
414 is fine. The call site will suppply a particular 'x'
416 Furthermore, the type variables fixed by an implicit parameter
417 propagate to the others. E.g.
418 foo :: (Show a, ?x::[a]) => Int
420 The type of foo looks ambiguous. But it isn't, because at a call site
422 let ?x = 5::Int in foo
423 and all is well. In effect, implicit parameters are, well, parameters,
424 so we can take their type variables into account as part of the
425 "tau-tvs" stuff. This is done in the function 'FunDeps.grow'.
428 Question 2: type signatures
429 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
430 BUT WATCH OUT: When you supply a type signature, we can't force you
431 to quantify over implicit parameters. For example:
435 This is perfectly reasonable. We do not want to insist on
437 (?x + 1) :: (?x::Int => Int)
439 That would be silly. Here, the definition site *is* the occurrence site,
440 so the above strictures don't apply. Hence the difference between
441 tcSimplifyCheck (which *does* allow implicit paramters to be inherited)
442 and tcSimplifyCheckBind (which does not).
444 What about when you supply a type signature for a binding?
445 Is it legal to give the following explicit, user type
446 signature to f, thus:
451 At first sight this seems reasonable, but it has the nasty property
452 that adding a type signature changes the dynamic semantics.
455 (let f x = (x::Int) + ?y
456 in (f 3, f 3 with ?y=5)) with ?y = 6
462 in (f 3, f 3 with ?y=5)) with ?y = 6
466 Indeed, simply inlining f (at the Haskell source level) would change the
469 Nevertheless, as Launchbury says (email Oct 01) we can't really give the
470 semantics for a Haskell program without knowing its typing, so if you
471 change the typing you may change the semantics.
473 To make things consistent in all cases where we are *checking* against
474 a supplied signature (as opposed to inferring a type), we adopt the
477 a signature does not need to quantify over implicit params.
479 [This represents a (rather marginal) change of policy since GHC 5.02,
480 which *required* an explicit signature to quantify over all implicit
481 params for the reasons mentioned above.]
483 But that raises a new question. Consider
485 Given (signature) ?x::Int
486 Wanted (inferred) ?x::Int, ?y::Bool
488 Clearly we want to discharge the ?x and float the ?y out. But
489 what is the criterion that distinguishes them? Clearly it isn't
490 what free type variables they have. The Right Thing seems to be
491 to float a constraint that
492 neither mentions any of the quantified type variables
493 nor any of the quantified implicit parameters
495 See the predicate isFreeWhenChecking.
498 Question 3: monomorphism
499 ~~~~~~~~~~~~~~~~~~~~~~~~
500 There's a nasty corner case when the monomorphism restriction bites:
504 The argument above suggests that we *must* generalise
505 over the ?y parameter, to get
506 z :: (?y::Int) => Int,
507 but the monomorphism restriction says that we *must not*, giving
509 Why does the momomorphism restriction say this? Because if you have
511 let z = x + ?y in z+z
513 you might not expect the addition to be done twice --- but it will if
514 we follow the argument of Question 2 and generalise over ?y.
517 Question 4: top level
518 ~~~~~~~~~~~~~~~~~~~~~
519 At the top level, monomorhism makes no sense at all.
522 main = let ?x = 5 in print foo
526 woggle :: (?x :: Int) => Int -> Int
529 We definitely don't want (foo :: Int) with a top-level implicit parameter
530 (?x::Int) becuase there is no way to bind it.
535 (A) Always generalise over implicit parameters
536 Bindings that fall under the monomorphism restriction can't
540 * Inlining remains valid
541 * No unexpected loss of sharing
542 * But simple bindings like
544 will be rejected, unless you add an explicit type signature
545 (to avoid the monomorphism restriction)
546 z :: (?y::Int) => Int
548 This seems unacceptable
550 (B) Monomorphism restriction "wins"
551 Bindings that fall under the monomorphism restriction can't
553 Always generalise over implicit parameters *except* for bindings
554 that fall under the monomorphism restriction
557 * Inlining isn't valid in general
558 * No unexpected loss of sharing
559 * Simple bindings like
561 accepted (get value of ?y from binding site)
563 (C) Always generalise over implicit parameters
564 Bindings that fall under the monomorphism restriction can't
565 be generalised, EXCEPT for implicit parameters
567 * Inlining remains valid
568 * Unexpected loss of sharing (from the extra generalisation)
569 * Simple bindings like
571 accepted (get value of ?y from occurrence sites)
576 None of these choices seems very satisfactory. But at least we should
577 decide which we want to do.
579 It's really not clear what is the Right Thing To Do. If you see
583 would you expect the value of ?y to be got from the *occurrence sites*
584 of 'z', or from the valuue of ?y at the *definition* of 'z'? In the
585 case of function definitions, the answer is clearly the former, but
586 less so in the case of non-fucntion definitions. On the other hand,
587 if we say that we get the value of ?y from the definition site of 'z',
588 then inlining 'z' might change the semantics of the program.
590 Choice (C) really says "the monomorphism restriction doesn't apply
591 to implicit parameters". Which is fine, but remember that every
592 innocent binding 'x = ...' that mentions an implicit parameter in
593 the RHS becomes a *function* of that parameter, called at each
594 use of 'x'. Now, the chances are that there are no intervening 'with'
595 clauses that bind ?y, so a decent compiler should common up all
596 those function calls. So I think I strongly favour (C). Indeed,
597 one could make a similar argument for abolishing the monomorphism
598 restriction altogether.
600 BOTTOM LINE: we choose (B) at present. See tcSimplifyRestricted
604 %************************************************************************
606 \subsection{tcSimplifyInfer}
608 %************************************************************************
610 tcSimplify is called when we *inferring* a type. Here's the overall game plan:
612 1. Compute Q = grow( fvs(T), C )
614 2. Partition C based on Q into Ct and Cq. Notice that ambiguous
615 predicates will end up in Ct; we deal with them at the top level
617 3. Try improvement, using functional dependencies
619 4. If Step 3 did any unification, repeat from step 1
620 (Unification can change the result of 'grow'.)
622 Note: we don't reduce dictionaries in step 2. For example, if we have
623 Eq (a,b), we don't simplify to (Eq a, Eq b). So Q won't be different
624 after step 2. However note that we may therefore quantify over more
625 type variables than we absolutely have to.
627 For the guts, we need a loop, that alternates context reduction and
628 improvement with unification. E.g. Suppose we have
630 class C x y | x->y where ...
632 and tcSimplify is called with:
634 Then improvement unifies a with b, giving
637 If we need to unify anything, we rattle round the whole thing all over
644 -> TcTyVarSet -- fv(T); type vars
646 -> TcM ([TcTyVar], -- Tyvars to quantify (zonked and quantified)
647 [Inst], -- Dict Ids that must be bound here (zonked)
648 TcDictBinds) -- Bindings
649 -- Any free (escaping) Insts are tossed into the environment
654 tcSimplifyInfer doc tau_tvs wanted
655 = do { tau_tvs1 <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
656 ; wanted' <- mapM zonkInst wanted -- Zonk before deciding quantified tyvars
657 ; gbl_tvs <- tcGetGlobalTyVars
658 ; let preds1 = fdPredsOfInsts wanted'
659 gbl_tvs1 = oclose preds1 gbl_tvs
660 qtvs = growInstsTyVars wanted' tau_tvs1 `minusVarSet` gbl_tvs1
661 -- See Note [Choosing which variables to quantify]
663 -- To maximise sharing, remove from consideration any
664 -- constraints that don't mention qtvs at all
665 ; let (free, bound) = partition (isFreeWhenInferring qtvs) wanted'
668 -- To make types simple, reduce as much as possible
669 ; traceTc (text "infer" <+> (ppr preds1 $$ ppr (growInstsTyVars wanted' tau_tvs1) $$ ppr gbl_tvs $$
670 ppr gbl_tvs1 $$ ppr free $$ ppr bound))
671 ; (irreds1, binds1) <- tryHardCheckLoop doc bound
673 -- Note [Inference and implication constraints]
674 ; let want_dict d = tyVarsOfInst d `intersectsVarSet` qtvs
675 ; (irreds2, binds2) <- approximateImplications doc want_dict irreds1
677 -- Now work out all over again which type variables to quantify,
678 -- exactly in the same way as before, but starting from irreds2. Why?
679 -- a) By now improvment may have taken place, and we must *not*
680 -- quantify over any variable free in the environment
681 -- tc137 (function h inside g) is an example
683 -- b) Do not quantify over constraints that *now* do not
684 -- mention quantified type variables, because they are
685 -- simply ambiguous (or might be bound further out). Example:
686 -- f :: Eq b => a -> (a, b)
688 -- From the RHS of g we get the MethodInst f77 :: alpha -> (alpha, beta)
689 -- We decide to quantify over 'alpha' alone, but free1 does not include f77
690 -- because f77 mentions 'alpha'. Then reducing leaves only the (ambiguous)
691 -- constraint (Eq beta), which we dump back into the free set
692 -- See test tcfail181
694 -- c) irreds may contain type variables not previously mentioned,
695 -- e.g. instance D a x => Foo [a]
697 -- Then after simplifying we'll get (D a x), and x is fresh
698 -- We must quantify over x else it'll be totally unbound
699 ; tau_tvs2 <- zonkTcTyVarsAndFV (varSetElems tau_tvs1)
700 ; gbl_tvs2 <- zonkTcTyVarsAndFV (varSetElems gbl_tvs1)
701 -- Note that we start from gbl_tvs1
702 -- We use tcGetGlobalTyVars, then oclose wrt preds2, because
703 -- we've already put some of the original preds1 into frees
704 -- E.g. wanteds = C a b (where a->b)
707 -- Then b is fixed by gbl_tvs, so (C a b) will be in free, and
708 -- irreds2 will be empty. But we don't want to generalise over b!
709 ; let preds2 = fdPredsOfInsts irreds2 -- irreds2 is zonked
710 qtvs = growInstsTyVars irreds2 tau_tvs2 `minusVarSet` oclose preds2 gbl_tvs2
711 ---------------------------------------------------
712 -- BUG WARNING: there's a nasty bug lurking here
713 -- fdPredsOfInsts may return preds that mention variables quantified in
714 -- one of the implication constraints in irreds2; and that is clearly wrong:
715 -- we might quantify over too many variables through accidental capture
716 ---------------------------------------------------
717 ; let (free, irreds3) = partition (isFreeWhenInferring qtvs) irreds2
720 -- Turn the quantified meta-type variables into real type variables
721 ; qtvs2 <- zonkQuantifiedTyVars (varSetElems qtvs)
723 -- We can't abstract over any remaining unsolved
724 -- implications so instead just float them outwards. Ugh.
725 ; let (q_dicts0, implics) = partition isAbstractableInst irreds3
726 ; loc <- getInstLoc (ImplicOrigin doc)
727 ; implic_bind <- bindIrreds loc qtvs2 q_dicts0 implics
729 -- Prepare equality instances for quantification
730 ; let (q_eqs0,q_dicts) = partition isEqInst q_dicts0
731 ; q_eqs <- mapM finalizeEqInst q_eqs0
733 ; return (qtvs2, q_eqs ++ q_dicts, binds1 `unionBags` binds2 `unionBags` implic_bind) }
734 -- NB: when we are done, we might have some bindings, but
735 -- the final qtvs might be empty. See Note [NO TYVARS] below.
737 approximateImplications :: SDoc -> (Inst -> Bool) -> [Inst] -> TcM ([Inst], TcDictBinds)
738 -- Note [Inference and implication constraints]
739 -- Given a bunch of Dict and ImplicInsts, try to approximate the implications by
740 -- - fetching any dicts inside them that are free
741 -- - using those dicts as cruder constraints, to solve the implications
742 -- - returning the extra ones too
744 approximateImplications doc want_dict irreds
746 = return (irreds, emptyBag)
748 = do { extra_dicts' <- mapM cloneDict extra_dicts
749 ; tryHardCheckLoop doc (extra_dicts' ++ irreds) }
750 -- By adding extra_dicts', we make them
751 -- available to solve the implication constraints
753 extra_dicts = get_dicts (filter isImplicInst irreds)
755 get_dicts :: [Inst] -> [Inst] -- Returns only Dicts
756 -- Find the wanted constraints in implication constraints that satisfy
757 -- want_dict, and are not bound by forall's in the constraint itself
758 get_dicts ds = concatMap get_dict ds
760 get_dict d@(Dict {}) | want_dict d = [d]
762 get_dict (ImplicInst {tci_tyvars = tvs, tci_wanted = wanteds})
763 = [ d | let tv_set = mkVarSet tvs
764 , d <- get_dicts wanteds
765 , not (tyVarsOfInst d `intersectsVarSet` tv_set)]
766 get_dict i@(EqInst {}) | want_dict i = [i]
768 get_dict other = pprPanic "approximateImplications" (ppr other)
771 Note [Inference and implication constraints]
772 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
773 Suppose we have a wanted implication constraint (perhaps arising from
774 a nested pattern match) like
776 and we are now trying to quantify over 'a' when inferring the type for
777 a function. In principle it's possible that there might be an instance
778 instance (C a, E a) => D [a]
779 so the context (E a) would suffice. The Right Thing is to abstract over
780 the implication constraint, but we don't do that (a) because it'll be
781 surprising to programmers and (b) because we don't have the machinery to deal
782 with 'given' implications.
784 So our best approximation is to make (D [a]) part of the inferred
785 context, so we can use that to discharge the implication. Hence
786 the strange function get_dicts in approximateImplications.
788 The common cases are more clear-cut, when we have things like
790 Here, abstracting over (C b) is not an approximation at all -- but see
791 Note [Freeness and implications].
793 See Trac #1430 and test tc228.
797 -----------------------------------------------------------
798 -- tcSimplifyInferCheck is used when we know the constraints we are to simplify
799 -- against, but we don't know the type variables over which we are going to quantify.
800 -- This happens when we have a type signature for a mutually recursive group
803 -> TcTyVarSet -- fv(T)
806 -> TcM ([TyVar], -- Fully zonked, and quantified
807 TcDictBinds) -- Bindings
809 tcSimplifyInferCheck loc tau_tvs givens wanteds
810 = do { traceTc (text "tcSimplifyInferCheck <-" <+> ppr wanteds)
811 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
813 -- Figure out which type variables to quantify over
814 -- You might think it should just be the signature tyvars,
815 -- but in bizarre cases you can get extra ones
816 -- f :: forall a. Num a => a -> a
817 -- f x = fst (g (x, head [])) + 1
819 -- Here we infer g :: forall a b. a -> b -> (b,a)
820 -- We don't want g to be monomorphic in b just because
821 -- f isn't quantified over b.
822 ; let all_tvs = varSetElems (tau_tvs `unionVarSet` tyVarsOfInsts givens)
823 ; all_tvs <- zonkTcTyVarsAndFV all_tvs
824 ; gbl_tvs <- tcGetGlobalTyVars
825 ; let qtvs = varSetElems (all_tvs `minusVarSet` gbl_tvs)
826 -- We could close gbl_tvs, but its not necessary for
827 -- soundness, and it'll only affect which tyvars, not which
828 -- dictionaries, we quantify over
830 ; qtvs' <- zonkQuantifiedTyVars qtvs
832 -- Now we are back to normal (c.f. tcSimplCheck)
833 ; implic_bind <- bindIrreds loc qtvs' givens irreds
835 ; traceTc (text "tcSimplifyInferCheck ->" <+> ppr (implic_bind))
836 ; return (qtvs', binds `unionBags` implic_bind) }
839 Note [Squashing methods]
840 ~~~~~~~~~~~~~~~~~~~~~~~~~
841 Be careful if you want to float methods more:
842 truncate :: forall a. RealFrac a => forall b. Integral b => a -> b
843 From an application (truncate f i) we get
846 If we have also have a second occurrence of truncate, we get
849 When simplifying with i,f free, we might still notice that
850 t1=t3; but alas, the binding for t2 (which mentions t1)
851 may continue to float out!
856 class Y a b | a -> b where
859 instance Y [[a]] a where
862 k :: X a -> X a -> X a
864 g :: Num a => [X a] -> [X a]
867 h ys = ys ++ map (k (y [[0]])) xs
869 The excitement comes when simplifying the bindings for h. Initially
870 try to simplify {y @ [[t1]] t2, 0 @ t1}, with initial qtvs = {t2}.
871 From this we get t1~t2, but also various bindings. We can't forget
872 the bindings (because of [LOOP]), but in fact t1 is what g is
875 The net effect of [NO TYVARS]
878 isFreeWhenInferring :: TyVarSet -> Inst -> Bool
879 isFreeWhenInferring qtvs inst
880 = isFreeWrtTyVars qtvs inst -- Constrains no quantified vars
881 && isInheritableInst inst -- and no implicit parameter involved
882 -- see Note [Inheriting implicit parameters]
884 {- No longer used (with implication constraints)
885 isFreeWhenChecking :: TyVarSet -- Quantified tyvars
886 -> NameSet -- Quantified implicit parameters
888 isFreeWhenChecking qtvs ips inst
889 = isFreeWrtTyVars qtvs inst
890 && isFreeWrtIPs ips inst
893 isFreeWrtTyVars :: VarSet -> Inst -> Bool
894 isFreeWrtTyVars qtvs inst = tyVarsOfInst inst `disjointVarSet` qtvs
895 isFreeWrtIPs :: NameSet -> Inst -> Bool
896 isFreeWrtIPs ips inst = not (any (`elemNameSet` ips) (ipNamesOfInst inst))
900 %************************************************************************
902 \subsection{tcSimplifyCheck}
904 %************************************************************************
906 @tcSimplifyCheck@ is used when we know exactly the set of variables
907 we are going to quantify over. For example, a class or instance declaration.
910 -----------------------------------------------------------
911 -- tcSimplifyCheck is used when checking expression type signatures,
912 -- class decls, instance decls etc.
913 tcSimplifyCheck :: InstLoc
914 -> [TcTyVar] -- Quantify over these
917 -> TcM TcDictBinds -- Bindings
918 tcSimplifyCheck loc qtvs givens wanteds
919 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
920 do { traceTc (text "tcSimplifyCheck")
921 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
922 ; implic_bind <- bindIrreds loc qtvs givens irreds
923 ; return (binds `unionBags` implic_bind) }
925 -----------------------------------------------------------
926 -- tcSimplifyCheckPat is used for existential pattern match
927 tcSimplifyCheckPat :: InstLoc
928 -> [TcTyVar] -- Quantify over these
931 -> TcM TcDictBinds -- Bindings
932 tcSimplifyCheckPat loc qtvs givens wanteds
933 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
934 do { traceTc (text "tcSimplifyCheckPat")
935 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
936 ; implic_bind <- bindIrredsR loc qtvs givens irreds
937 ; return (binds `unionBags` implic_bind) }
939 -----------------------------------------------------------
940 bindIrreds :: InstLoc -> [TcTyVar]
943 bindIrreds loc qtvs givens irreds
944 = bindIrredsR loc qtvs givens irreds
946 bindIrredsR :: InstLoc -> [TcTyVar] -> [Inst] -> [Inst] -> TcM TcDictBinds
947 -- Make a binding that binds 'irreds', by generating an implication
948 -- constraint for them, *and* throwing the constraint into the LIE
949 bindIrredsR loc qtvs givens irreds
953 = do { let givens' = filter isAbstractableInst givens
954 -- The givens can (redundantly) include methods
955 -- We want to retain both EqInsts and Dicts
956 -- There should be no implicadtion constraints
957 -- See Note [Pruning the givens in an implication constraint]
959 -- If there are no 'givens', then it's safe to
960 -- partition the 'wanteds' by their qtvs, thereby trimming irreds
961 -- See Note [Freeness and implications]
962 ; irreds' <- if null givens'
964 { let qtv_set = mkVarSet qtvs
965 (frees, real_irreds) = partition (isFreeWrtTyVars qtv_set) irreds
967 ; return real_irreds }
970 ; (implics, bind) <- makeImplicationBind loc qtvs givens' irreds'
971 -- This call does the real work
972 -- If irreds' is empty, it does something sensible
977 makeImplicationBind :: InstLoc -> [TcTyVar]
979 -> TcM ([Inst], TcDictBinds)
980 -- Make a binding that binds 'irreds', by generating an implication
981 -- constraint for them.
983 -- The binding looks like
984 -- (ir1, .., irn) = f qtvs givens
985 -- where f is (evidence for) the new implication constraint
986 -- f :: forall qtvs. givens => (ir1, .., irn)
987 -- qtvs includes coercion variables
989 -- This binding must line up the 'rhs' in reduceImplication
990 makeImplicationBind loc all_tvs
991 givens -- Guaranteed all Dicts or EqInsts
993 | null irreds -- If there are no irreds, we are done
994 = return ([], emptyBag)
995 | otherwise -- Otherwise we must generate a binding
996 = do { uniq <- newUnique
997 ; span <- getSrcSpanM
998 ; let (eq_givens, dict_givens) = partition isEqInst givens
1000 -- extract equality binders
1001 eq_cotvs = map eqInstType eq_givens
1003 -- make the implication constraint instance
1004 name = mkInternalName uniq (mkVarOcc "ic") span
1005 implic_inst = ImplicInst { tci_name = name,
1006 tci_tyvars = all_tvs,
1007 tci_given = eq_givens ++ dict_givens,
1008 -- same order as binders
1009 tci_wanted = irreds,
1012 -- create binders for the irreducible dictionaries
1013 dict_irreds = filter (not . isEqInst) irreds
1014 dict_irred_ids = map instToId dict_irreds
1015 lpat = mkBigLHsPatTup (map (L span . VarPat) dict_irred_ids)
1017 -- create the binding
1018 rhs = L span (mkHsWrap co (HsVar (instToId implic_inst)))
1019 co = mkWpApps (map instToId dict_givens)
1020 <.> mkWpTyApps eq_cotvs
1021 <.> mkWpTyApps (mkTyVarTys all_tvs)
1022 bind | [dict_irred_id] <- dict_irred_ids
1023 = VarBind dict_irred_id rhs
1025 = PatBind { pat_lhs = lpat
1026 , pat_rhs = unguardedGRHSs rhs
1027 , pat_rhs_ty = hsLPatType lpat
1028 , bind_fvs = placeHolderNames
1031 ; traceTc $ text "makeImplicationBind" <+> ppr implic_inst
1032 ; return ([implic_inst], unitBag (L span bind))
1035 -----------------------------------------------------------
1036 tryHardCheckLoop :: SDoc
1038 -> TcM ([Inst], TcDictBinds)
1040 tryHardCheckLoop doc wanteds
1041 = do { (irreds,binds) <- checkLoop (mkInferRedEnv doc try_me) wanteds
1042 ; return (irreds,binds)
1046 -- Here's the try-hard bit
1048 -----------------------------------------------------------
1049 gentleCheckLoop :: InstLoc
1052 -> TcM ([Inst], TcDictBinds)
1054 gentleCheckLoop inst_loc givens wanteds
1055 = do { (irreds,binds) <- checkLoop env wanteds
1056 ; return (irreds,binds)
1059 env = mkRedEnv (pprInstLoc inst_loc) try_me givens
1061 try_me inst | isMethodOrLit inst = ReduceMe
1063 -- When checking against a given signature
1064 -- we MUST be very gentle: Note [Check gently]
1066 gentleInferLoop :: SDoc -> [Inst]
1067 -> TcM ([Inst], TcDictBinds)
1068 gentleInferLoop doc wanteds
1069 = do { (irreds, binds) <- checkLoop env wanteds
1070 ; return (irreds, binds) }
1072 env = mkInferRedEnv doc try_me
1073 try_me inst | isMethodOrLit inst = ReduceMe
1078 ~~~~~~~~~~~~~~~~~~~~
1079 We have to very careful about not simplifying too vigorously
1084 f :: Show b => T b -> b
1085 f (MkT x) = show [x]
1087 Inside the pattern match, which binds (a:*, x:a), we know that
1089 Hence we have a dictionary for Show [a] available; and indeed we
1090 need it. We are going to build an implication contraint
1091 forall a. (b~[a]) => Show [a]
1092 Later, we will solve this constraint using the knowledge (Show b)
1094 But we MUST NOT reduce (Show [a]) to (Show a), else the whole
1095 thing becomes insoluble. So we simplify gently (get rid of literals
1096 and methods only, plus common up equal things), deferring the real
1097 work until top level, when we solve the implication constraint
1098 with tryHardCheckLooop.
1102 -----------------------------------------------------------
1105 -> TcM ([Inst], TcDictBinds)
1106 -- Precondition: givens are completely rigid
1107 -- Postcondition: returned Insts are zonked
1109 checkLoop env wanteds
1111 where go env wanteds
1112 = do { -- We do need to zonk the givens; cf Note [Zonking RedEnv]
1113 ; env' <- zonkRedEnv env
1114 ; wanteds' <- zonkInsts wanteds
1116 ; (improved, tybinds, binds, irreds)
1117 <- reduceContext env' wanteds'
1118 ; execTcTyVarBinds tybinds
1120 ; if null irreds || not improved then
1121 return (irreds, binds)
1124 -- If improvement did some unification, we go round again.
1125 -- We start again with irreds, not wanteds
1126 -- Using an instance decl might have introduced a fresh type
1127 -- variable which might have been unified, so we'd get an
1128 -- infinite loop if we started again with wanteds!
1130 { (irreds1, binds1) <- go env' irreds
1131 ; return (irreds1, binds `unionBags` binds1) } }
1134 Note [Zonking RedEnv]
1135 ~~~~~~~~~~~~~~~~~~~~~
1136 It might appear as if the givens in RedEnv are always rigid, but that is not
1137 necessarily the case for programs involving higher-rank types that have class
1138 contexts constraining the higher-rank variables. An example from tc237 in the
1141 class Modular s a | s -> a
1143 wim :: forall a w. Integral a
1144 => a -> (forall s. Modular s a => M s w) -> w
1145 wim i k = error "urk"
1147 test5 :: (Modular s a, Integral a) => M s a
1150 test4 = wim 4 test4'
1152 Notice how the variable 'a' of (Modular s a) in the rank-2 type of wim is
1153 quantified further outside. When type checking test4, we have to check
1154 whether the signature of test5 is an instance of
1156 (forall s. Modular s a => M s w)
1158 Consequently, we will get (Modular s t_a), where t_a is a TauTv into the
1161 Given the FD of Modular in this example, class improvement will instantiate
1162 t_a to 'a', where 'a' is the skolem from test5's signatures (due to the
1163 Modular s a predicate in that signature). If we don't zonk (Modular s t_a) in
1164 the givens, we will get into a loop as improveOne uses the unification engine
1165 Unify.tcUnifyTys, which doesn't know about mutable type variables.
1170 class If b t e r | b t e -> r
1173 class Lte a b c | a b -> c where lte :: a -> b -> c
1175 instance (Lte a b l,If l b a c) => Max a b c
1177 Wanted: Max Z (S x) y
1179 Then we'll reduce using the Max instance to:
1180 (Lte Z (S x) l, If l (S x) Z y)
1181 and improve by binding l->T, after which we can do some reduction
1182 on both the Lte and If constraints. What we *can't* do is start again
1183 with (Max Z (S x) y)!
1187 %************************************************************************
1189 tcSimplifySuperClasses
1191 %************************************************************************
1193 Note [SUPERCLASS-LOOP 1]
1194 ~~~~~~~~~~~~~~~~~~~~~~~~
1195 We have to be very, very careful when generating superclasses, lest we
1196 accidentally build a loop. Here's an example:
1200 class S a => C a where { opc :: a -> a }
1201 class S b => D b where { opd :: b -> b }
1203 instance C Int where
1206 instance D Int where
1209 From (instance C Int) we get the constraint set {ds1:S Int, dd:D Int}
1210 Simplifying, we may well get:
1211 $dfCInt = :C ds1 (opd dd)
1214 Notice that we spot that we can extract ds1 from dd.
1216 Alas! Alack! We can do the same for (instance D Int):
1218 $dfDInt = :D ds2 (opc dc)
1222 And now we've defined the superclass in terms of itself.
1223 Two more nasty cases are in
1228 - Satisfy the superclass context *all by itself*
1229 (tcSimplifySuperClasses)
1230 - And do so completely; i.e. no left-over constraints
1231 to mix with the constraints arising from method declarations
1234 Note [Recursive instances and superclases]
1235 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1236 Consider this code, which arises in the context of "Scrap Your
1237 Boilerplate with Class".
1241 instance Sat (ctx Char) => Data ctx Char
1242 instance (Sat (ctx [a]), Data ctx a) => Data ctx [a]
1244 class Data Maybe a => Foo a
1246 instance Foo t => Sat (Maybe t)
1248 instance Data Maybe a => Foo a
1249 instance Foo a => Foo [a]
1252 In the instance for Foo [a], when generating evidence for the superclasses
1253 (ie in tcSimplifySuperClasses) we need a superclass (Data Maybe [a]).
1254 Using the instance for Data, we therefore need
1255 (Sat (Maybe [a], Data Maybe a)
1256 But we are given (Foo a), and hence its superclass (Data Maybe a).
1257 So that leaves (Sat (Maybe [a])). Using the instance for Sat means
1258 we need (Foo [a]). And that is the very dictionary we are bulding
1259 an instance for! So we must put that in the "givens". So in this
1261 Given: Foo a, Foo [a]
1262 Watend: Data Maybe [a]
1264 BUT we must *not not not* put the *superclasses* of (Foo [a]) in
1265 the givens, which is what 'addGiven' would normally do. Why? Because
1266 (Data Maybe [a]) is the superclass, so we'd "satisfy" the wanted
1267 by selecting a superclass from Foo [a], which simply makes a loop.
1269 On the other hand we *must* put the superclasses of (Foo a) in
1270 the givens, as you can see from the derivation described above.
1272 Conclusion: in the very special case of tcSimplifySuperClasses
1273 we have one 'given' (namely the "this" dictionary) whose superclasses
1274 must not be added to 'givens' by addGiven.
1276 There is a complication though. Suppose there are equalities
1277 instance (Eq a, a~b) => Num (a,b)
1278 Then we normalise the 'givens' wrt the equalities, so the original
1279 given "this" dictionary is cast to one of a different type. So it's a
1280 bit trickier than before to identify the "special" dictionary whose
1281 superclasses must not be added. See test
1282 indexed-types/should_run/EqInInstance
1284 We need a persistent property of the dictionary to record this
1285 special-ness. Current I'm using the InstLocOrigin (a bit of a hack,
1286 but cool), which is maintained by dictionary normalisation.
1287 Specifically, the InstLocOrigin is
1289 then the no-superclass thing kicks in. WATCH OUT if you fiddle
1293 tcSimplifySuperClasses
1295 -> Inst -- The dict whose superclasses
1296 -- are being figured out
1300 tcSimplifySuperClasses loc this givens sc_wanteds
1301 = do { traceTc (text "tcSimplifySuperClasses")
1303 -- Note [Recursive instances and superclases]
1304 ; no_sc_loc <- getInstLoc NoScOrigin
1305 ; let no_sc_this = setInstLoc this no_sc_loc
1307 ; let env = RedEnv { red_doc = pprInstLoc loc,
1308 red_try_me = try_me,
1309 red_givens = no_sc_this : givens,
1311 red_improve = False } -- No unification vars
1314 ; (irreds,binds1) <- checkLoop env sc_wanteds
1315 ; let (tidy_env, tidy_irreds) = tidyInsts irreds
1316 ; reportNoInstances tidy_env (Just (loc, givens)) [] tidy_irreds
1319 try_me _ = ReduceMe -- Try hard, so we completely solve the superclass
1320 -- constraints right here. See Note [SUPERCLASS-LOOP 1]
1324 %************************************************************************
1326 \subsection{tcSimplifyRestricted}
1328 %************************************************************************
1330 tcSimplifyRestricted infers which type variables to quantify for a
1331 group of restricted bindings. This isn't trivial.
1334 We want to quantify over a to get id :: forall a. a->a
1337 We do not want to quantify over a, because there's an Eq a
1338 constraint, so we get eq :: a->a->Bool (notice no forall)
1341 RHS has type 'tau', whose free tyvars are tau_tvs
1342 RHS has constraints 'wanteds'
1345 Quantify over (tau_tvs \ ftvs(wanteds))
1346 This is bad. The constraints may contain (Monad (ST s))
1347 where we have instance Monad (ST s) where...
1348 so there's no need to be monomorphic in s!
1350 Also the constraint might be a method constraint,
1351 whose type mentions a perfectly innocent tyvar:
1352 op :: Num a => a -> b -> a
1353 Here, b is unconstrained. A good example would be
1355 We want to infer the polymorphic type
1356 foo :: forall b. b -> b
1359 Plan B (cunning, used for a long time up to and including GHC 6.2)
1360 Step 1: Simplify the constraints as much as possible (to deal
1361 with Plan A's problem). Then set
1362 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1364 Step 2: Now simplify again, treating the constraint as 'free' if
1365 it does not mention qtvs, and trying to reduce it otherwise.
1366 The reasons for this is to maximise sharing.
1368 This fails for a very subtle reason. Suppose that in the Step 2
1369 a constraint (Foo (Succ Zero) (Succ Zero) b) gets thrown upstairs as 'free'.
1370 In the Step 1 this constraint might have been simplified, perhaps to
1371 (Foo Zero Zero b), AND THEN THAT MIGHT BE IMPROVED, to bind 'b' to 'T'.
1372 This won't happen in Step 2... but that in turn might prevent some other
1373 constraint (Baz [a] b) being simplified (e.g. via instance Baz [a] T where {..})
1374 and that in turn breaks the invariant that no constraints are quantified over.
1376 Test typecheck/should_compile/tc177 (which failed in GHC 6.2) demonstrates
1381 Step 1: Simplify the constraints as much as possible (to deal
1382 with Plan A's problem). Then set
1383 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1384 Return the bindings from Step 1.
1387 A note about Plan C (arising from "bug" reported by George Russel March 2004)
1390 instance (HasBinary ty IO) => HasCodedValue ty
1392 foo :: HasCodedValue a => String -> IO a
1394 doDecodeIO :: HasCodedValue a => () -> () -> IO a
1395 doDecodeIO codedValue view
1396 = let { act = foo "foo" } in act
1398 You might think this should work becuase the call to foo gives rise to a constraint
1399 (HasCodedValue t), which can be satisfied by the type sig for doDecodeIO. But the
1400 restricted binding act = ... calls tcSimplifyRestricted, and PlanC simplifies the
1401 constraint using the (rather bogus) instance declaration, and now we are stuffed.
1403 I claim this is not really a bug -- but it bit Sergey as well as George. So here's
1407 Plan D (a variant of plan B)
1408 Step 1: Simplify the constraints as much as possible (to deal
1409 with Plan A's problem), BUT DO NO IMPROVEMENT. Then set
1410 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1412 Step 2: Now simplify again, treating the constraint as 'free' if
1413 it does not mention qtvs, and trying to reduce it otherwise.
1415 The point here is that it's generally OK to have too few qtvs; that is,
1416 to make the thing more monomorphic than it could be. We don't want to
1417 do that in the common cases, but in wierd cases it's ok: the programmer
1418 can always add a signature.
1420 Too few qtvs => too many wanteds, which is what happens if you do less
1425 tcSimplifyRestricted -- Used for restricted binding groups
1426 -- i.e. ones subject to the monomorphism restriction
1429 -> [Name] -- Things bound in this group
1430 -> TcTyVarSet -- Free in the type of the RHSs
1431 -> [Inst] -- Free in the RHSs
1432 -> TcM ([TyVar], -- Tyvars to quantify (zonked and quantified)
1433 TcDictBinds) -- Bindings
1434 -- tcSimpifyRestricted returns no constraints to
1435 -- quantify over; by definition there are none.
1436 -- They are all thrown back in the LIE
1438 tcSimplifyRestricted doc top_lvl bndrs tau_tvs wanteds
1439 -- Zonk everything in sight
1440 = do { traceTc (text "tcSimplifyRestricted")
1441 ; wanteds_z <- zonkInsts wanteds
1443 -- 'ReduceMe': Reduce as far as we can. Don't stop at
1444 -- dicts; the idea is to get rid of as many type
1445 -- variables as possible, and we don't want to stop
1446 -- at (say) Monad (ST s), because that reduces
1447 -- immediately, with no constraint on s.
1449 -- BUT do no improvement! See Plan D above
1450 -- HOWEVER, some unification may take place, if we instantiate
1451 -- a method Inst with an equality constraint
1452 ; let env = mkNoImproveRedEnv doc (\_ -> ReduceMe)
1453 ; (_imp, _tybinds, _binds, constrained_dicts)
1454 <- reduceContext env wanteds_z
1456 -- Next, figure out the tyvars we will quantify over
1457 ; tau_tvs' <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
1458 ; gbl_tvs' <- tcGetGlobalTyVars
1459 ; constrained_dicts' <- zonkInsts constrained_dicts
1461 ; let qtvs1 = tau_tvs' `minusVarSet` oclose (fdPredsOfInsts constrained_dicts) gbl_tvs'
1462 -- As in tcSimplifyInfer
1464 -- Do not quantify over constrained type variables:
1465 -- this is the monomorphism restriction
1466 constrained_tvs' = tyVarsOfInsts constrained_dicts'
1467 qtvs = qtvs1 `minusVarSet` constrained_tvs'
1468 pp_bndrs = pprWithCommas (quotes . ppr) bndrs
1471 ; warn_mono <- doptM Opt_WarnMonomorphism
1472 ; warnTc (warn_mono && (constrained_tvs' `intersectsVarSet` qtvs1))
1473 (vcat[ ptext (sLit "the Monomorphism Restriction applies to the binding")
1474 <> plural bndrs <+> ptext (sLit "for") <+> pp_bndrs,
1475 ptext (sLit "Consider giving a type signature for") <+> pp_bndrs])
1477 ; traceTc (text "tcSimplifyRestricted" <+> vcat [
1478 pprInsts wanteds, pprInsts constrained_dicts',
1480 ppr constrained_tvs', ppr tau_tvs', ppr qtvs ])
1482 -- The first step may have squashed more methods than
1483 -- necessary, so try again, this time more gently, knowing the exact
1484 -- set of type variables to quantify over.
1486 -- We quantify only over constraints that are captured by qtvs;
1487 -- these will just be a subset of non-dicts. This in contrast
1488 -- to normal inference (using isFreeWhenInferring) in which we quantify over
1489 -- all *non-inheritable* constraints too. This implements choice
1490 -- (B) under "implicit parameter and monomorphism" above.
1492 -- Remember that we may need to do *some* simplification, to
1493 -- (for example) squash {Monad (ST s)} into {}. It's not enough
1494 -- just to float all constraints
1496 -- At top level, we *do* squash methods becuase we want to
1497 -- expose implicit parameters to the test that follows
1498 ; let is_nested_group = isNotTopLevel top_lvl
1499 try_me inst | isFreeWrtTyVars qtvs inst,
1500 (is_nested_group || isDict inst) = Stop
1501 | otherwise = ReduceMe
1502 env = mkNoImproveRedEnv doc try_me
1503 ; (_imp, tybinds, binds, irreds) <- reduceContext env wanteds_z
1504 ; execTcTyVarBinds tybinds
1506 -- See "Notes on implicit parameters, Question 4: top level"
1507 ; ASSERT( all (isFreeWrtTyVars qtvs) irreds ) -- None should be captured
1508 if is_nested_group then
1510 else do { let (bad_ips, non_ips) = partition isIPDict irreds
1511 ; addTopIPErrs bndrs bad_ips
1512 ; extendLIEs non_ips }
1514 ; qtvs' <- zonkQuantifiedTyVars (varSetElems qtvs)
1515 ; return (qtvs', binds) }
1519 %************************************************************************
1523 %************************************************************************
1525 On the LHS of transformation rules we only simplify methods and constants,
1526 getting dictionaries. We want to keep all of them unsimplified, to serve
1527 as the available stuff for the RHS of the rule.
1529 Example. Consider the following left-hand side of a rule
1531 f (x == y) (y > z) = ...
1533 If we typecheck this expression we get constraints
1535 d1 :: Ord a, d2 :: Eq a
1537 We do NOT want to "simplify" to the LHS
1539 forall x::a, y::a, z::a, d1::Ord a.
1540 f ((==) (eqFromOrd d1) x y) ((>) d1 y z) = ...
1544 forall x::a, y::a, z::a, d1::Ord a, d2::Eq a.
1545 f ((==) d2 x y) ((>) d1 y z) = ...
1547 Here is another example:
1549 fromIntegral :: (Integral a, Num b) => a -> b
1550 {-# RULES "foo" fromIntegral = id :: Int -> Int #-}
1552 In the rule, a=b=Int, and Num Int is a superclass of Integral Int. But
1553 we *dont* want to get
1555 forall dIntegralInt.
1556 fromIntegral Int Int dIntegralInt (scsel dIntegralInt) = id Int
1558 because the scsel will mess up RULE matching. Instead we want
1560 forall dIntegralInt, dNumInt.
1561 fromIntegral Int Int dIntegralInt dNumInt = id Int
1565 g (x == y) (y == z) = ..
1567 where the two dictionaries are *identical*, we do NOT WANT
1569 forall x::a, y::a, z::a, d1::Eq a
1570 f ((==) d1 x y) ((>) d1 y z) = ...
1572 because that will only match if the dict args are (visibly) equal.
1573 Instead we want to quantify over the dictionaries separately.
1575 In short, tcSimplifyRuleLhs must *only* squash LitInst and MethInts, leaving
1576 all dicts unchanged, with absolutely no sharing. It's simpler to do this
1577 from scratch, rather than further parameterise simpleReduceLoop etc.
1578 Simpler, maybe, but alas not simple (see Trac #2494)
1580 * Type errors may give rise to an (unsatisfiable) equality constraint
1582 * Applications of a higher-rank function on the LHS may give
1583 rise to an implication constraint, esp if there are unsatisfiable
1584 equality constraints inside.
1587 tcSimplifyRuleLhs :: [Inst] -> TcM ([Inst], TcDictBinds)
1588 tcSimplifyRuleLhs wanteds
1589 = do { wanteds' <- zonkInsts wanteds
1590 ; (irreds, binds) <- go [] emptyBag wanteds'
1591 ; let (dicts, bad_irreds) = partition isDict irreds
1592 ; traceTc (text "tcSimplifyrulelhs" <+> pprInsts bad_irreds)
1593 ; addNoInstanceErrs (nub bad_irreds)
1594 -- The nub removes duplicates, which has
1595 -- not happened otherwise (see notes above)
1596 ; return (dicts, binds) }
1598 go :: [Inst] -> TcDictBinds -> [Inst] -> TcM ([Inst], TcDictBinds)
1600 = return (irreds, binds)
1601 go irreds binds (w:ws)
1603 = go (w:irreds) binds ws
1604 | isImplicInst w -- Have a go at reducing the implication
1605 = do { (binds1, irreds1) <- reduceImplication red_env w
1606 ; let (bad_irreds, ok_irreds) = partition isImplicInst irreds1
1607 ; go (bad_irreds ++ irreds)
1608 (binds `unionBags` binds1)
1611 = do { w' <- zonkInst w -- So that (3::Int) does not generate a call
1612 -- to fromInteger; this looks fragile to me
1613 ; lookup_result <- lookupSimpleInst w'
1614 ; case lookup_result of
1615 NoInstance -> go (w:irreds) binds ws
1616 GenInst ws' rhs -> go irreds binds' (ws' ++ ws)
1618 binds' = addInstToDictBind binds w rhs
1621 -- Sigh: we need to reduce inside implications
1622 red_env = mkInferRedEnv doc try_me
1623 doc = ptext (sLit "Implication constraint in RULE lhs")
1624 try_me inst | isMethodOrLit inst = ReduceMe
1625 | otherwise = Stop -- Be gentle
1628 tcSimplifyBracket is used when simplifying the constraints arising from
1629 a Template Haskell bracket [| ... |]. We want to check that there aren't
1630 any constraints that can't be satisfied (e.g. Show Foo, where Foo has no
1631 Show instance), but we aren't otherwise interested in the results.
1632 Nor do we care about ambiguous dictionaries etc. We will type check
1633 this bracket again at its usage site.
1636 tcSimplifyBracket :: [Inst] -> TcM ()
1637 tcSimplifyBracket wanteds
1638 = do { _ <- tryHardCheckLoop doc wanteds
1641 doc = text "tcSimplifyBracket"
1645 %************************************************************************
1647 \subsection{Filtering at a dynamic binding}
1649 %************************************************************************
1654 we must discharge all the ?x constraints from B. We also do an improvement
1655 step; if we have ?x::t1 and ?x::t2 we must unify t1, t2.
1657 Actually, the constraints from B might improve the types in ?x. For example
1659 f :: (?x::Int) => Char -> Char
1662 then the constraint (?x::Int) arising from the call to f will
1663 force the binding for ?x to be of type Int.
1666 tcSimplifyIPs :: [Inst] -- The implicit parameters bound here
1669 -- We need a loop so that we do improvement, and then
1670 -- (next time round) generate a binding to connect the two
1672 -- Here the two ?x's have different types, and improvement
1673 -- makes them the same.
1675 tcSimplifyIPs given_ips wanteds
1676 = do { wanteds' <- zonkInsts wanteds
1677 ; given_ips' <- zonkInsts given_ips
1678 -- Unusually for checking, we *must* zonk the given_ips
1680 ; let env = mkRedEnv doc try_me given_ips'
1681 ; (improved, tybinds, binds, irreds) <- reduceContext env wanteds'
1682 ; execTcTyVarBinds tybinds
1684 ; if null irreds || not improved then
1685 ASSERT( all is_free irreds )
1686 do { extendLIEs irreds
1689 -- If improvement did some unification, we go round again.
1690 -- We start again with irreds, not wanteds
1691 -- Using an instance decl might have introduced a fresh type
1692 -- variable which might have been unified, so we'd get an
1693 -- infinite loop if we started again with wanteds!
1695 { binds1 <- tcSimplifyIPs given_ips' irreds
1696 ; return $ binds `unionBags` binds1
1699 doc = text "tcSimplifyIPs" <+> ppr given_ips
1700 ip_set = mkNameSet (ipNamesOfInsts given_ips)
1701 is_free inst = isFreeWrtIPs ip_set inst
1703 -- Simplify any methods that mention the implicit parameter
1704 try_me inst | is_free inst = Stop
1705 | otherwise = ReduceMe
1709 %************************************************************************
1711 \subsection[binds-for-local-funs]{@bindInstsOfLocalFuns@}
1713 %************************************************************************
1715 When doing a binding group, we may have @Insts@ of local functions.
1716 For example, we might have...
1718 let f x = x + 1 -- orig local function (overloaded)
1719 f.1 = f Int -- two instances of f
1724 The point is: we must drop the bindings for @f.1@ and @f.2@ here,
1725 where @f@ is in scope; those @Insts@ must certainly not be passed
1726 upwards towards the top-level. If the @Insts@ were binding-ified up
1727 there, they would have unresolvable references to @f@.
1729 We pass in an @init_lie@ of @Insts@ and a list of locally-bound @Ids@.
1730 For each method @Inst@ in the @init_lie@ that mentions one of the
1731 @Ids@, we create a binding. We return the remaining @Insts@ (in an
1732 @LIE@), as well as the @HsBinds@ generated.
1735 bindInstsOfLocalFuns :: [Inst] -> [TcId] -> TcM TcDictBinds
1736 -- Simlifies only MethodInsts, and generate only bindings of form
1738 -- We're careful not to even generate bindings of the form
1740 -- You'd think that'd be fine, but it interacts with what is
1741 -- arguably a bug in Match.tidyEqnInfo (see notes there)
1743 bindInstsOfLocalFuns wanteds local_ids
1744 | null overloaded_ids = do
1747 return emptyLHsBinds
1750 = do { (irreds, binds) <- gentleInferLoop doc for_me
1751 ; extendLIEs not_for_me
1755 doc = text "bindInsts" <+> ppr local_ids
1756 overloaded_ids = filter is_overloaded local_ids
1757 is_overloaded id = isOverloadedTy (idType id)
1758 (for_me, not_for_me) = partition (isMethodFor overloaded_set) wanteds
1760 overloaded_set = mkVarSet overloaded_ids -- There can occasionally be a lot of them
1761 -- so it's worth building a set, so that
1762 -- lookup (in isMethodFor) is faster
1766 %************************************************************************
1768 \subsection{Data types for the reduction mechanism}
1770 %************************************************************************
1772 The main control over context reduction is here
1776 = RedEnv { red_doc :: SDoc -- The context
1777 , red_try_me :: Inst -> WhatToDo
1778 , red_improve :: Bool -- True <=> do improvement
1779 , red_givens :: [Inst] -- All guaranteed rigid
1780 -- Always dicts & equalities
1781 -- but see Note [Rigidity]
1783 , red_stack :: (Int, [Inst]) -- Recursion stack (for err msg)
1784 -- See Note [RedStack]
1788 -- The red_givens are rigid so far as cmpInst is concerned.
1789 -- There is one case where they are not totally rigid, namely in tcSimplifyIPs
1790 -- let ?x = e in ...
1791 -- Here, the given is (?x::a), where 'a' is not necy a rigid type
1792 -- But that doesn't affect the comparison, which is based only on mame.
1795 -- The red_stack pair (n,insts) pair is just used for error reporting.
1796 -- 'n' is always the depth of the stack.
1797 -- The 'insts' is the stack of Insts being reduced: to produce X
1798 -- I had to produce Y, to produce Y I had to produce Z, and so on.
1801 mkRedEnv :: SDoc -> (Inst -> WhatToDo) -> [Inst] -> RedEnv
1802 mkRedEnv doc try_me givens
1803 = RedEnv { red_doc = doc, red_try_me = try_me,
1804 red_givens = givens,
1806 red_improve = True }
1808 mkInferRedEnv :: SDoc -> (Inst -> WhatToDo) -> RedEnv
1810 mkInferRedEnv doc try_me
1811 = RedEnv { red_doc = doc, red_try_me = try_me,
1814 red_improve = True }
1816 mkNoImproveRedEnv :: SDoc -> (Inst -> WhatToDo) -> RedEnv
1817 -- Do not do improvement; no givens
1818 mkNoImproveRedEnv doc try_me
1819 = RedEnv { red_doc = doc, red_try_me = try_me,
1822 red_improve = True }
1825 = ReduceMe -- Try to reduce this
1826 -- If there's no instance, add the inst to the
1827 -- irreductible ones, but don't produce an error
1828 -- message of any kind.
1829 -- It might be quite legitimate such as (Eq a)!
1831 | Stop -- Return as irreducible unless it can
1832 -- be reduced to a constant in one step
1833 -- Do not add superclasses; see
1835 data WantSCs = NoSCs | AddSCs -- Tells whether we should add the superclasses
1836 -- of a predicate when adding it to the avails
1837 -- The reason for this flag is entirely the super-class loop problem
1838 -- Note [SUPER-CLASS LOOP 1]
1840 zonkRedEnv :: RedEnv -> TcM RedEnv
1842 = do { givens' <- mapM zonkInst (red_givens env)
1843 ; return $ env {red_givens = givens'}
1848 %************************************************************************
1850 \subsection[reduce]{@reduce@}
1852 %************************************************************************
1854 Note [Ancestor Equalities]
1855 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1856 During context reduction, we add to the wanted equalities also those
1857 equalities that (transitively) occur in superclass contexts of wanted
1858 class constraints. Consider the following code
1860 class a ~ Int => C a
1863 If (C a) is wanted, we want to add (a ~ Int), which will be discharged by
1864 substituting Int for a. Hence, we ultimately want (C Int), which we
1865 discharge with the explicit instance.
1868 reduceContext :: RedEnv
1870 -> TcM (ImprovementDone,
1871 TcTyVarBinds, -- Type variable bindings
1872 TcDictBinds, -- Dictionary bindings
1873 [Inst]) -- Irreducible
1875 reduceContext env wanteds0
1876 = do { traceTc (text "reduceContext" <+> (vcat [
1877 text "----------------------",
1879 text "given" <+> ppr (red_givens env),
1880 text "wanted" <+> ppr wanteds0,
1881 text "----------------------"
1884 -- We want to add as wanted equalities those that (transitively)
1885 -- occur in superclass contexts of wanted class constraints.
1886 -- See Note [Ancestor Equalities]
1887 ; ancestor_eqs <- ancestorEqualities wanteds0
1888 ; traceTc $ text "reduceContext: ancestor eqs" <+> ppr ancestor_eqs
1890 -- Normalise and solve all equality constraints as far as possible
1891 -- and normalise all dictionary constraints wrt to the reduced
1892 -- equalities. The returned wanted constraints include the
1893 -- irreducible wanted equalities.
1894 ; let wanteds = wanteds0 ++ ancestor_eqs
1895 givens = red_givens env
1899 normalise_binds) <- tcReduceEqs givens wanteds
1900 ; traceTc $ text "reduceContext: tcReduceEqs result" <+> vcat
1901 [ppr givens', ppr wanteds', ppr tybinds,
1902 ppr normalise_binds]
1904 -- Build the Avail mapping from "given_dicts"
1905 ; (init_state, _) <- getLIE $ do
1906 { init_state <- foldlM addGiven emptyAvails givens'
1910 -- Solve the *wanted* *dictionary* constraints (not implications)
1911 -- This may expose some further equational constraints in the course
1912 -- of improvement due to functional dependencies if any of the
1913 -- involved unifications gets deferred.
1914 ; let (wanted_implics, wanted_dicts) = partition isImplicInst wanteds'
1915 ; (avails, extra_eqs) <- getLIE (reduceList env wanted_dicts init_state)
1916 -- The getLIE is reqd because reduceList does improvement
1917 -- (via extendAvails) which may in turn do unification
1920 dict_irreds) <- extractResults avails wanted_dicts
1921 ; traceTc $ text "reduceContext: extractResults" <+> vcat
1922 [ppr avails, ppr wanted_dicts, ppr dict_binds]
1924 -- Solve the wanted *implications*. In doing so, we can provide
1925 -- as "given" all the dicts that were originally given,
1926 -- *or* for which we now have bindings,
1927 -- *or* which are now irreds
1928 -- NB: Equality irreds need to be converted, as the recursive
1929 -- invocation of the solver will still treat them as wanteds
1931 ; let implic_env = env { red_givens
1932 = givens ++ bound_dicts ++
1933 map wantedToLocalEqInst dict_irreds }
1934 ; (implic_binds_s, implic_irreds_s)
1935 <- mapAndUnzipM (reduceImplication implic_env) wanted_implics
1936 ; let implic_binds = unionManyBags implic_binds_s
1937 implic_irreds = concat implic_irreds_s
1939 -- Collect all irreducible instances, and determine whether we should
1940 -- go round again. We do so in either of two cases:
1941 -- (1) If dictionary reduction or equality solving led to
1942 -- improvement (i.e., bindings for type variables).
1943 -- (2) If we reduced dictionaries (i.e., got dictionary bindings),
1944 -- they may have exposed further opportunities to normalise
1945 -- family applications. See Note [Dictionary Improvement]
1947 -- NB: We do *not* go around for new extra_eqs. Morally, we should,
1948 -- but we can't without risking non-termination (see #2688). By
1949 -- not going around, we miss some legal programs mixing FDs and
1950 -- TFs, but we never claimed to support such programs in the
1951 -- current implementation anyway.
1953 ; let all_irreds = dict_irreds ++ implic_irreds ++ extra_eqs
1954 avails_improved = availsImproved avails
1955 eq_improved = anyBag (not . isCoVarBind) tybinds
1956 improvedFlexible = avails_improved || eq_improved
1957 reduced_dicts = not (isEmptyBag dict_binds)
1958 improved = improvedFlexible || reduced_dicts
1960 improvedHint = (if avails_improved then " [AVAILS]" else "") ++
1961 (if eq_improved then " [EQ]" else "")
1963 ; traceTc (text "reduceContext end" <+> (vcat [
1964 text "----------------------",
1966 text "given" <+> ppr givens,
1967 text "wanted" <+> ppr wanteds0,
1969 text "tybinds" <+> ppr tybinds,
1970 text "avails" <+> pprAvails avails,
1971 text "improved =" <+> ppr improved <+> text improvedHint,
1972 text "(all) irreds = " <+> ppr all_irreds,
1973 text "dict-binds = " <+> ppr dict_binds,
1974 text "implic-binds = " <+> ppr implic_binds,
1975 text "----------------------"
1980 normalise_binds `unionBags` dict_binds
1981 `unionBags` implic_binds,
1985 isCoVarBind (TcTyVarBind tv _) = isCoVar tv
1987 tcImproveOne :: Avails -> Inst -> TcM ImprovementDone
1988 tcImproveOne avails inst
1989 | not (isDict inst) = return False
1991 = do { inst_envs <- tcGetInstEnvs
1992 ; let eqns = improveOne (classInstances inst_envs)
1993 (dictPred inst, pprInstArising inst)
1994 [ (dictPred p, pprInstArising p)
1995 | p <- availsInsts avails, isDict p ]
1996 -- Avails has all the superclasses etc (good)
1997 -- It also has all the intermediates of the deduction (good)
1998 -- It does not have duplicates (good)
1999 -- NB that (?x::t1) and (?x::t2) will be held separately in
2000 -- avails so that improve will see them separate
2001 ; traceTc (text "improveOne" <+> ppr inst)
2004 unifyEqns :: [(Equation, (PredType, SDoc), (PredType, SDoc))]
2005 -> TcM ImprovementDone
2006 unifyEqns [] = return False
2008 = do { traceTc (ptext (sLit "Improve:") <+> vcat (map pprEquationDoc eqns))
2009 ; improved <- mapM unify eqns
2010 ; return $ or improved
2013 unify ((qtvs, pairs), what1, what2)
2014 = addErrCtxtM (mkEqnMsg what1 what2) $
2015 do { let freeTyVars = unionVarSets (map tvs_pr pairs)
2017 ; (_, _, tenv) <- tcInstTyVars (varSetElems qtvs)
2018 ; mapM_ (unif_pr tenv) pairs
2019 ; anyM isFilledMetaTyVar $ varSetElems freeTyVars
2022 unif_pr tenv (ty1, ty2) = unifyType (substTy tenv ty1) (substTy tenv ty2)
2024 tvs_pr (ty1, ty2) = tyVarsOfType ty1 `unionVarSet` tyVarsOfType ty2
2026 pprEquationDoc :: (Equation, (PredType, SDoc), (PredType, SDoc)) -> SDoc
2027 pprEquationDoc (eqn, (p1, _), (p2, _))
2028 = vcat [pprEquation eqn, nest 2 (ppr p1), nest 2 (ppr p2)]
2030 mkEqnMsg :: (TcPredType, SDoc) -> (TcPredType, SDoc) -> TidyEnv
2031 -> TcM (TidyEnv, SDoc)
2032 mkEqnMsg (pred1,from1) (pred2,from2) tidy_env
2033 = do { pred1' <- zonkTcPredType pred1
2034 ; pred2' <- zonkTcPredType pred2
2035 ; let { pred1'' = tidyPred tidy_env pred1'
2036 ; pred2'' = tidyPred tidy_env pred2' }
2037 ; let msg = vcat [ptext (sLit "When using functional dependencies to combine"),
2038 nest 2 (sep [ppr pred1'' <> comma, nest 2 from1]),
2039 nest 2 (sep [ppr pred2'' <> comma, nest 2 from2])]
2040 ; return (tidy_env, msg) }
2043 Note [Dictionary Improvement]
2044 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2045 In reduceContext, we first reduce equalities and then class constraints.
2046 However, the letter may expose further opportunities for the former. Hence,
2047 we need to go around again if dictionary reduction produced any dictionary
2048 bindings. The following example demonstrated the point:
2050 data EX _x _y (p :: * -> *)
2055 class Base (Def p) => Prop p where
2059 instance Prop () where
2062 instance (Base (Def (p ANY))) => Base (EX _x _y p)
2063 instance (Prop (p ANY)) => Prop (EX _x _y p) where
2064 type Def (EX _x _y p) = EX _x _y p
2067 instance Prop (FOO x) where
2068 type Def (FOO x) = ()
2071 instance Prop BAR where
2072 type Def BAR = EX () () FOO
2074 During checking the last instance declaration, we need to check the superclass
2075 cosntraint Base (Def BAR), which family normalisation reduced to
2076 Base (EX () () FOO). Chasing the instance for Base (EX _x _y p), gives us
2077 Base (Def (FOO ANY)), which again requires family normalisation of Def to
2078 Base () before we can finish.
2081 The main context-reduction function is @reduce@. Here's its game plan.
2084 reduceList :: RedEnv -> [Inst] -> Avails -> TcM Avails
2085 reduceList env@(RedEnv {red_stack = (n,stk)}) wanteds state
2086 = do { traceTc (text "reduceList " <+> (ppr wanteds $$ ppr state))
2088 ; when (debugIsOn && (n > 8)) $ do
2089 debugDumpTcRn (hang (ptext (sLit "Interesting! Context reduction stack depth") <+> int n)
2090 2 (ifPprDebug (nest 2 (pprStack stk))))
2091 ; if n >= ctxtStkDepth dopts then
2092 failWithTc (reduceDepthErr n stk)
2096 go [] state = return state
2097 go (w:ws) state = do { state' <- reduce (env {red_stack = (n+1, w:stk)}) w state
2100 -- Base case: we're done!
2101 reduce :: RedEnv -> Inst -> Avails -> TcM Avails
2102 reduce env wanted avails
2104 -- We don't reduce equalities here (and they must not end up as irreds
2109 -- It's the same as an existing inst, or a superclass thereof
2110 | Just _ <- findAvail avails wanted
2111 = do { traceTc (text "reduce: found " <+> ppr wanted)
2116 = do { traceTc (text "reduce" <+> ppr wanted $$ ppr avails)
2117 ; case red_try_me env wanted of {
2118 Stop -> try_simple (addIrred NoSCs);
2119 -- See Note [No superclasses for Stop]
2121 ReduceMe -> do -- It should be reduced
2122 { (avails, lookup_result) <- reduceInst env avails wanted
2123 ; case lookup_result of
2124 NoInstance -> addIrred AddSCs avails wanted
2125 -- Add it and its superclasses
2127 GenInst [] rhs -> addWanted AddSCs avails wanted rhs []
2129 GenInst wanteds' rhs
2130 -> do { avails1 <- addIrred NoSCs avails wanted
2131 ; avails2 <- reduceList env wanteds' avails1
2132 ; addWanted AddSCs avails2 wanted rhs wanteds' } }
2133 -- Temporarily do addIrred *before* the reduceList,
2134 -- which has the effect of adding the thing we are trying
2135 -- to prove to the database before trying to prove the things it
2136 -- needs. See note [RECURSIVE DICTIONARIES]
2137 -- NB: we must not do an addWanted before, because that adds the
2138 -- superclasses too, and that can lead to a spurious loop; see
2139 -- the examples in [SUPERCLASS-LOOP]
2140 -- So we do an addIrred before, and then overwrite it afterwards with addWanted
2143 -- First, see if the inst can be reduced to a constant in one step
2144 -- Works well for literals (1::Int) and constant dictionaries (d::Num Int)
2145 -- Don't bother for implication constraints, which take real work
2146 try_simple do_this_otherwise
2147 = do { res <- lookupSimpleInst wanted
2149 GenInst [] rhs -> addWanted AddSCs avails wanted rhs []
2150 _ -> do_this_otherwise avails wanted }
2154 Note [RECURSIVE DICTIONARIES]
2155 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2157 data D r = ZeroD | SuccD (r (D r));
2159 instance (Eq (r (D r))) => Eq (D r) where
2160 ZeroD == ZeroD = True
2161 (SuccD a) == (SuccD b) = a == b
2164 equalDC :: D [] -> D [] -> Bool;
2167 We need to prove (Eq (D [])). Here's how we go:
2171 by instance decl, holds if
2175 by instance decl of Eq, holds if
2177 where d2 = dfEqList d3
2180 But now we can "tie the knot" to give
2186 and it'll even run! The trick is to put the thing we are trying to prove
2187 (in this case Eq (D []) into the database before trying to prove its
2188 contributing clauses.
2190 Note [SUPERCLASS-LOOP 2]
2191 ~~~~~~~~~~~~~~~~~~~~~~~~
2192 We need to be careful when adding "the constaint we are trying to prove".
2193 Suppose we are *given* d1:Ord a, and want to deduce (d2:C [a]) where
2195 class Ord a => C a where
2196 instance Ord [a] => C [a] where ...
2198 Then we'll use the instance decl to deduce C [a] from Ord [a], and then add the
2199 superclasses of C [a] to avails. But we must not overwrite the binding
2200 for Ord [a] (which is obtained from Ord a) with a superclass selection or we'll just
2203 Here's another variant, immortalised in tcrun020
2204 class Monad m => C1 m
2205 class C1 m => C2 m x
2206 instance C2 Maybe Bool
2207 For the instance decl we need to build (C1 Maybe), and it's no good if
2208 we run around and add (C2 Maybe Bool) and its superclasses to the avails
2209 before we search for C1 Maybe.
2211 Here's another example
2212 class Eq b => Foo a b
2213 instance Eq a => Foo [a] a
2217 we'll first deduce that it holds (via the instance decl). We must not
2218 then overwrite the Eq t constraint with a superclass selection!
2220 At first I had a gross hack, whereby I simply did not add superclass constraints
2221 in addWanted, though I did for addGiven and addIrred. This was sub-optimal,
2222 becuase it lost legitimate superclass sharing, and it still didn't do the job:
2223 I found a very obscure program (now tcrun021) in which improvement meant the
2224 simplifier got two bites a the cherry... so something seemed to be an Stop
2225 first time, but reducible next time.
2227 Now we implement the Right Solution, which is to check for loops directly
2228 when adding superclasses. It's a bit like the occurs check in unification.
2232 %************************************************************************
2234 Reducing a single constraint
2236 %************************************************************************
2239 ---------------------------------------------
2240 reduceInst :: RedEnv -> Avails -> Inst -> TcM (Avails, LookupInstResult)
2241 reduceInst _ avails other_inst
2242 = do { result <- lookupSimpleInst other_inst
2243 ; return (avails, result) }
2246 Note [Equational Constraints in Implication Constraints]
2247 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2249 An implication constraint is of the form
2251 where Given and Wanted may contain both equational and dictionary
2252 constraints. The delay and reduction of these two kinds of constraints
2255 -) In the generated code, wanted Dictionary constraints are wrapped up in an
2256 implication constraint that is created at the code site where the wanted
2257 dictionaries can be reduced via a let-binding. This let-bound implication
2258 constraint is deconstructed at the use-site of the wanted dictionaries.
2260 -) While the reduction of equational constraints is also delayed, the delay
2261 is not manifest in the generated code. The required evidence is generated
2262 in the code directly at the use-site. There is no let-binding and deconstruction
2263 necessary. The main disadvantage is that we cannot exploit sharing as the
2264 same evidence may be generated at multiple use-sites. However, this disadvantage
2265 is limited because it only concerns coercions which are erased.
2267 The different treatment is motivated by the different in representation. Dictionary
2268 constraints require manifest runtime dictionaries, while equations require coercions
2272 ---------------------------------------------
2273 reduceImplication :: RedEnv
2275 -> TcM (TcDictBinds, [Inst])
2278 Suppose we are simplifying the constraint
2279 forall bs. extras => wanted
2280 in the context of an overall simplification problem with givens 'givens'.
2283 * The 'givens' need not mention any of the quantified type variables
2284 e.g. forall {}. Eq a => Eq [a]
2285 forall {}. C Int => D (Tree Int)
2287 This happens when you have something like
2289 T1 :: Eq a => a -> T a
2292 f x = ...(case x of { T1 v -> v==v })...
2295 -- ToDo: should we instantiate tvs? I think it's not necessary
2297 -- Note on coercion variables:
2299 -- The extra given coercion variables are bound at two different
2302 -- -) in the creation context of the implication constraint
2303 -- the solved equational constraints use these binders
2305 -- -) at the solving site of the implication constraint
2306 -- the solved dictionaries use these binders;
2307 -- these binders are generated by reduceImplication
2309 -- Note [Binders for equalities]
2310 -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2311 -- To reuse the binders of local/given equalities in the binders of
2312 -- implication constraints, it is crucial that these given equalities
2313 -- always have the form
2315 -- where cotv is a simple coercion type variable (and not a more
2316 -- complex coercion term). We require that the extra_givens always
2317 -- have this form and exploit the special form when generating binders.
2318 reduceImplication env
2319 orig_implic@(ImplicInst { tci_name = name, tci_loc = inst_loc,
2321 tci_given = extra_givens, tci_wanted = wanteds
2323 = do { -- Solve the sub-problem
2324 ; let try_me _ = ReduceMe -- Note [Freeness and implications]
2325 env' = env { red_givens = extra_givens ++ red_givens env
2326 , red_doc = sep [ptext (sLit "reduceImplication for")
2328 nest 2 (parens $ ptext (sLit "within")
2330 , red_try_me = try_me }
2332 ; traceTc (text "reduceImplication" <+> vcat
2333 [ ppr (red_givens env), ppr extra_givens,
2335 ; (irreds, binds) <- checkLoop env' wanteds
2337 ; traceTc (text "reduceImplication result" <+> vcat
2338 [ppr irreds, ppr binds])
2340 ; -- extract superclass binds
2341 -- (sc_binds,_) <- extractResults avails []
2342 -- ; traceTc (text "reduceImplication sc_binds" <+> vcat
2343 -- [ppr sc_binds, ppr avails])
2346 -- SLPJ Sept 07: what if improvement happened inside the checkLoop?
2347 -- Then we must iterate the outer loop too!
2349 ; didntSolveWantedEqs <- allM wantedEqInstIsUnsolved wanteds
2350 -- we solve wanted eqs by side effect!
2352 -- Progress is no longer measered by the number of bindings
2353 -- If there are any irreds, but no bindings and no solved
2354 -- equalities, we back off and do nothing
2355 ; let backOff = isEmptyLHsBinds binds && -- no new bindings
2356 (not $ null irreds) && -- but still some irreds
2357 didntSolveWantedEqs -- no instantiated cotv
2359 ; if backOff then -- No progress
2360 return (emptyBag, [orig_implic])
2362 { (simpler_implic_insts, bind)
2363 <- makeImplicationBind inst_loc tvs extra_givens irreds
2364 -- This binding is useless if the recursive simplification
2365 -- made no progress; but currently we don't try to optimise that
2366 -- case. After all, we only try hard to reduce at top level, or
2367 -- when inferring types.
2369 ; let -- extract Id binders for dicts and CoTyVar binders for eqs;
2370 -- see Note [Binders for equalities]
2371 (extra_eq_givens, extra_dict_givens) = partition isEqInst
2373 eq_cotvs = map instToVar extra_eq_givens
2374 dict_ids = map instToId extra_dict_givens
2376 -- Note [Always inline implication constraints]
2377 wrap_inline | null dict_ids = idHsWrapper
2378 | otherwise = WpInline
2381 <.> mkWpTyLams eq_cotvs
2382 <.> mkWpLams dict_ids
2383 <.> WpLet (binds `unionBags` bind)
2384 rhs = mkLHsWrap co payload
2385 loc = instLocSpan inst_loc
2386 -- wanted equalities are solved by updating their
2387 -- cotv; we don't generate bindings for them
2388 dict_bndrs = map (L loc . HsVar . instToId)
2389 . filter (not . isEqInst)
2391 payload = mkBigLHsTup dict_bndrs
2394 ; traceTc (vcat [text "reduceImplication" <+> ppr name,
2395 ppr simpler_implic_insts,
2396 text "->" <+> ppr rhs])
2397 ; return (unitBag (L loc (VarBind (instToId orig_implic) rhs)),
2398 simpler_implic_insts)
2401 reduceImplication _ i = pprPanic "reduceImplication" (ppr i)
2404 Note [Always inline implication constraints]
2405 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2406 Suppose an implication constraint floats out of an INLINE function.
2407 Then although the implication has a single call site, it won't be
2408 inlined. And that is bad because it means that even if there is really
2409 *no* overloading (type signatures specify the exact types) there will
2410 still be dictionary passing in the resulting code. To avert this,
2411 we mark the implication constraints themselves as INLINE, at least when
2412 there is no loss of sharing as a result.
2414 Note [Freeness and implications]
2415 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2416 It's hard to say when an implication constraint can be floated out. Consider
2417 forall {} Eq a => Foo [a]
2418 The (Foo [a]) doesn't mention any of the quantified variables, but it
2419 still might be partially satisfied by the (Eq a).
2421 There is a useful special case when it *is* easy to partition the
2422 constraints, namely when there are no 'givens'. Consider
2423 forall {a}. () => Bar b
2424 There are no 'givens', and so there is no reason to capture (Bar b).
2425 We can let it float out. But if there is even one constraint we
2426 must be much more careful:
2427 forall {a}. C a b => Bar (m b)
2428 because (C a b) might have a superclass (D b), from which we might
2429 deduce (Bar [b]) when m later gets instantiated to []. Ha!
2431 Here is an even more exotic example
2433 Now consider the constraint
2434 forall b. D Int b => C Int
2435 We can satisfy the (C Int) from the superclass of D, so we don't want
2436 to float the (C Int) out, even though it mentions no type variable in
2439 One more example: the constraint
2441 instance (C a, E c) => E (a,c)
2443 constraint: forall b. D Int b => E (Int,c)
2445 You might think that the (D Int b) can't possibly contribute
2446 to solving (E (Int,c)), since the latter mentions 'c'. But
2447 in fact it can, because solving the (E (Int,c)) constraint needs
2450 and the (C Int) can be satisfied from the superclass of (D Int b).
2451 So we must still not float (E (Int,c)) out.
2453 To think about: special cases for unary type classes?
2455 Note [Pruning the givens in an implication constraint]
2456 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2457 Suppose we are about to form the implication constraint
2458 forall tvs. Eq a => Ord b
2459 The (Eq a) cannot contribute to the (Ord b), because it has no access to
2460 the type variable 'b'. So we could filter out the (Eq a) from the givens.
2461 But BE CAREFUL of the examples above in [Freeness and implications].
2463 Doing so would be a bit tidier, but all the implication constraints get
2464 simplified away by the optimiser, so it's no great win. So I don't take
2465 advantage of that at the moment.
2467 If you do, BE CAREFUL of wobbly type variables.
2470 %************************************************************************
2472 Avails and AvailHow: the pool of evidence
2474 %************************************************************************
2478 data Avails = Avails !ImprovementDone !AvailEnv
2480 type ImprovementDone = Bool -- True <=> some unification has happened
2481 -- so some Irreds might now be reducible
2482 -- keys that are now
2484 type AvailEnv = FiniteMap Inst AvailHow
2486 = IsIrred -- Used for irreducible dictionaries,
2487 -- which are going to be lambda bound
2489 | Given Inst -- Used for dictionaries for which we have a binding
2490 -- e.g. those "given" in a signature
2492 | Rhs -- Used when there is a RHS
2493 (LHsExpr TcId) -- The RHS
2494 [Inst] -- Insts free in the RHS; we need these too
2496 instance Outputable Avails where
2499 pprAvails :: Avails -> SDoc
2500 pprAvails (Avails imp avails)
2501 = vcat [ ptext (sLit "Avails") <> (if imp then ptext (sLit "[improved]") else empty)
2503 vcat [ sep [ppr inst, nest 2 (equals <+> ppr avail)]
2504 | (inst,avail) <- fmToList avails ]]
2506 instance Outputable AvailHow where
2509 -------------------------
2510 pprAvail :: AvailHow -> SDoc
2511 pprAvail IsIrred = text "Irred"
2512 pprAvail (Given x) = text "Given" <+> ppr x
2513 pprAvail (Rhs rhs bs) = sep [text "Rhs" <+> ppr bs,
2516 -------------------------
2517 extendAvailEnv :: AvailEnv -> Inst -> AvailHow -> AvailEnv
2518 extendAvailEnv env inst avail = addToFM env inst avail
2520 findAvailEnv :: AvailEnv -> Inst -> Maybe AvailHow
2521 findAvailEnv env wanted = lookupFM env wanted
2522 -- NB 1: the Ord instance of Inst compares by the class/type info
2523 -- *not* by unique. So
2524 -- d1::C Int == d2::C Int
2526 emptyAvails :: Avails
2527 emptyAvails = Avails False emptyFM
2529 findAvail :: Avails -> Inst -> Maybe AvailHow
2530 findAvail (Avails _ avails) wanted = findAvailEnv avails wanted
2532 elemAvails :: Inst -> Avails -> Bool
2533 elemAvails wanted (Avails _ avails) = wanted `elemFM` avails
2535 extendAvails :: Avails -> Inst -> AvailHow -> TcM Avails
2537 extendAvails avails@(Avails imp env) inst avail
2538 = do { imp1 <- tcImproveOne avails inst -- Do any improvement
2539 ; return (Avails (imp || imp1) (extendAvailEnv env inst avail)) }
2541 availsInsts :: Avails -> [Inst]
2542 availsInsts (Avails _ avails) = keysFM avails
2544 availsImproved :: Avails -> ImprovementDone
2545 availsImproved (Avails imp _) = imp
2548 Extracting the bindings from a bunch of Avails.
2549 The bindings do *not* come back sorted in dependency order.
2550 We assume that they'll be wrapped in a big Rec, so that the
2551 dependency analyser can sort them out later
2554 type DoneEnv = FiniteMap Inst [Id]
2555 -- Tracks which things we have evidence for
2557 extractResults :: Avails
2559 -> TcM (TcDictBinds, -- Bindings
2560 [Inst], -- The insts bound by the bindings
2561 [Inst]) -- Irreducible ones
2562 -- Note [Reducing implication constraints]
2564 extractResults (Avails _ avails) wanteds
2565 = go emptyBag [] [] emptyFM wanteds
2567 go :: TcDictBinds -- Bindings for dicts
2568 -> [Inst] -- Bound by the bindings
2570 -> DoneEnv -- Has an entry for each inst in the above three sets
2572 -> TcM (TcDictBinds, [Inst], [Inst])
2573 go binds bound_dicts irreds _ []
2574 = return (binds, bound_dicts, irreds)
2576 go binds bound_dicts irreds done (w:ws)
2578 = go binds bound_dicts (w:irreds) done' ws
2580 | Just done_ids@(done_id : rest_done_ids) <- lookupFM done w
2581 = if w_id `elem` done_ids then
2582 go binds bound_dicts irreds done ws
2584 go (add_bind (nlHsVar done_id)) bound_dicts irreds
2585 (addToFM done w (done_id : w_id : rest_done_ids)) ws
2587 | otherwise -- Not yet done
2588 = case findAvailEnv avails w of
2589 Nothing -> pprTrace "Urk: extractResults" (ppr w) $
2590 go binds bound_dicts irreds done ws
2592 Just IsIrred -> go binds bound_dicts (w:irreds) done' ws
2594 Just (Rhs rhs ws') -> go (add_bind rhs) (w:bound_dicts) irreds done' (ws' ++ ws)
2596 Just (Given g) -> go binds' bound_dicts irreds (addToFM done w [g_id]) ws
2599 binds' | w_id == g_id = binds
2600 | otherwise = add_bind (nlHsVar g_id)
2603 done' = addToFM done w [w_id]
2604 add_bind rhs = addInstToDictBind binds w rhs
2608 Note [No superclasses for Stop]
2609 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2610 When we decide not to reduce an Inst -- the 'WhatToDo' --- we still
2611 add it to avails, so that any other equal Insts will be commoned up
2612 right here. However, we do *not* add superclasses. If we have
2615 but a is not bound here, then we *don't* want to derive dn from df
2616 here lest we lose sharing.
2619 addWanted :: WantSCs -> Avails -> Inst -> LHsExpr TcId -> [Inst] -> TcM Avails
2620 addWanted want_scs avails wanted rhs_expr wanteds
2621 = addAvailAndSCs want_scs avails wanted avail
2623 avail = Rhs rhs_expr wanteds
2625 addGiven :: Avails -> Inst -> TcM Avails
2626 addGiven avails given
2627 = addAvailAndSCs want_scs avails given (Given given)
2629 want_scs = case instLocOrigin (instLoc given) of
2632 -- Conditionally add superclasses for 'given'
2633 -- See Note [Recursive instances and superclases]
2635 -- No ASSERT( not (given `elemAvails` avails) ) because in an
2636 -- instance decl for Ord t we can add both Ord t and Eq t as
2637 -- 'givens', so the assert isn't true
2641 addIrred :: WantSCs -> Avails -> Inst -> TcM Avails
2642 addIrred want_scs avails irred = ASSERT2( not (irred `elemAvails` avails), ppr irred $$ ppr avails )
2643 addAvailAndSCs want_scs avails irred IsIrred
2645 addAvailAndSCs :: WantSCs -> Avails -> Inst -> AvailHow -> TcM Avails
2646 addAvailAndSCs want_scs avails inst avail
2647 | not (isClassDict inst) = extendAvails avails inst avail
2648 | NoSCs <- want_scs = extendAvails avails inst avail
2649 | otherwise = do { traceTc (text "addAvailAndSCs" <+> vcat [ppr inst, ppr deps])
2650 ; avails' <- extendAvails avails inst avail
2651 ; addSCs is_loop avails' inst }
2653 is_loop pred = any (`tcEqType` mkPredTy pred) dep_tys
2654 -- Note: this compares by *type*, not by Unique
2655 deps = findAllDeps (unitVarSet (instToVar inst)) avail
2656 dep_tys = map idType (varSetElems deps)
2658 findAllDeps :: IdSet -> AvailHow -> IdSet
2659 -- Find all the Insts that this one depends on
2660 -- See Note [SUPERCLASS-LOOP 2]
2661 -- Watch out, though. Since the avails may contain loops
2662 -- (see Note [RECURSIVE DICTIONARIES]), so we need to track the ones we've seen so far
2663 findAllDeps so_far (Rhs _ kids) = foldl find_all so_far kids
2664 findAllDeps so_far _ = so_far
2666 find_all :: IdSet -> Inst -> IdSet
2668 | isEqInst kid = so_far
2669 | kid_id `elemVarSet` so_far = so_far
2670 | Just avail <- findAvail avails kid = findAllDeps so_far' avail
2671 | otherwise = so_far'
2673 so_far' = extendVarSet so_far kid_id -- Add the new kid to so_far
2674 kid_id = instToId kid
2676 addSCs :: (TcPredType -> Bool) -> Avails -> Inst -> TcM Avails
2677 -- Add all the superclasses of the Inst to Avails
2678 -- The first param says "don't do this because the original thing
2679 -- depends on this one, so you'd build a loop"
2680 -- Invariant: the Inst is already in Avails.
2682 addSCs is_loop avails dict
2683 = ASSERT( isDict dict )
2684 do { sc_dicts <- newDictBndrs (instLoc dict) sc_theta'
2685 ; foldlM add_sc avails (zipEqual "add_scs" sc_dicts sc_sels) }
2687 (clas, tys) = getDictClassTys dict
2688 (tyvars, sc_theta, sc_sels, _) = classBigSig clas
2689 sc_theta' = filter (not . isEqPred) $
2690 substTheta (zipTopTvSubst tyvars tys) sc_theta
2692 add_sc avails (sc_dict, sc_sel)
2693 | is_loop (dictPred sc_dict) = return avails -- See Note [SUPERCLASS-LOOP 2]
2694 | is_given sc_dict = return avails
2695 | otherwise = do { avails' <- extendAvails avails sc_dict (Rhs sc_sel_rhs [dict])
2696 ; addSCs is_loop avails' sc_dict }
2698 sc_sel_rhs = L (instSpan dict) (HsWrap co_fn (HsVar sc_sel))
2699 co_fn = WpApp (instToVar dict) <.> mkWpTyApps tys
2701 is_given :: Inst -> Bool
2702 is_given sc_dict = case findAvail avails sc_dict of
2703 Just (Given _) -> True -- Given is cheaper than superclass selection
2706 -- From the a set of insts obtain all equalities that (transitively) occur in
2707 -- superclass contexts of class constraints (aka the ancestor equalities).
2709 ancestorEqualities :: [Inst] -> TcM [Inst]
2711 = mapM mkWantedEqInst -- turn only equality predicates..
2712 . filter isEqPred -- ..into wanted equality insts
2714 . addAEsToBag emptyBag -- collect the superclass constraints..
2715 . map dictPred -- ..of all predicates in a bag
2716 . filter isClassDict
2718 addAEsToBag :: Bag PredType -> [PredType] -> Bag PredType
2719 addAEsToBag bag [] = bag
2720 addAEsToBag bag (pred:preds)
2721 | pred `elemBag` bag = addAEsToBag bag preds
2722 | isEqPred pred = addAEsToBag bagWithPred preds
2723 | isClassPred pred = addAEsToBag bagWithPred predsWithSCs
2724 | otherwise = addAEsToBag bag preds
2726 bagWithPred = bag `snocBag` pred
2727 predsWithSCs = preds ++ substTheta (zipTopTvSubst tyvars tys) sc_theta
2729 (tyvars, sc_theta, _, _) = classBigSig clas
2730 (clas, tys) = getClassPredTys pred
2734 %************************************************************************
2736 \section{tcSimplifyTop: defaulting}
2738 %************************************************************************
2741 @tcSimplifyTop@ is called once per module to simplify all the constant
2742 and ambiguous Insts.
2744 We need to be careful of one case. Suppose we have
2746 instance Num a => Num (Foo a b) where ...
2748 and @tcSimplifyTop@ is given a constraint (Num (Foo x y)). Then it'll simplify
2749 to (Num x), and default x to Int. But what about y??
2751 It's OK: the final zonking stage should zap y to (), which is fine.
2755 tcSimplifyTop, tcSimplifyInteractive :: [Inst] -> TcM TcDictBinds
2756 tcSimplifyTop wanteds
2757 = tc_simplify_top doc False wanteds
2759 doc = text "tcSimplifyTop"
2761 tcSimplifyInteractive wanteds
2762 = tc_simplify_top doc True wanteds
2764 doc = text "tcSimplifyInteractive"
2766 -- The TcLclEnv should be valid here, solely to improve
2767 -- error message generation for the monomorphism restriction
2768 tc_simplify_top :: SDoc -> Bool -> [Inst] -> TcM (Bag (LHsBind TcId))
2769 tc_simplify_top doc interactive wanteds
2770 = do { dflags <- getDOpts
2771 ; wanteds <- zonkInsts wanteds
2772 ; mapM_ zonkTopTyVar (varSetElems (tyVarsOfInsts wanteds))
2774 ; traceTc (text "tc_simplify_top 0: " <+> ppr wanteds)
2775 ; (irreds1, binds1) <- tryHardCheckLoop doc1 wanteds
2776 -- ; (irreds1, binds1) <- gentleInferLoop doc1 wanteds
2777 ; traceTc (text "tc_simplify_top 1: " <+> ppr irreds1)
2778 ; (irreds2, binds2) <- approximateImplications doc2 (\_ -> True) irreds1
2779 ; traceTc (text "tc_simplify_top 2: " <+> ppr irreds2)
2781 -- Use the defaulting rules to do extra unification
2782 -- NB: irreds2 are already zonked
2783 ; (irreds3, binds3) <- disambiguate doc3 interactive dflags irreds2
2785 -- Deal with implicit parameters
2786 ; let (bad_ips, non_ips) = partition isIPDict irreds3
2787 (ambigs, others) = partition isTyVarDict non_ips
2789 ; topIPErrs bad_ips -- Can arise from f :: Int -> Int
2791 ; addNoInstanceErrs others
2792 ; addTopAmbigErrs ambigs
2794 ; return (binds1 `unionBags` binds2 `unionBags` binds3) }
2796 doc1 = doc <+> ptext (sLit "(first round)")
2797 doc2 = doc <+> ptext (sLit "(approximate)")
2798 doc3 = doc <+> ptext (sLit "(disambiguate)")
2801 If a dictionary constrains a type variable which is
2802 * not mentioned in the environment
2803 * and not mentioned in the type of the expression
2804 then it is ambiguous. No further information will arise to instantiate
2805 the type variable; nor will it be generalised and turned into an extra
2806 parameter to a function.
2808 It is an error for this to occur, except that Haskell provided for
2809 certain rules to be applied in the special case of numeric types.
2811 * at least one of its classes is a numeric class, and
2812 * all of its classes are numeric or standard
2813 then the type variable can be defaulted to the first type in the
2814 default-type list which is an instance of all the offending classes.
2816 So here is the function which does the work. It takes the ambiguous
2817 dictionaries and either resolves them (producing bindings) or
2818 complains. It works by splitting the dictionary list by type
2819 variable, and using @disambigOne@ to do the real business.
2821 @disambigOne@ assumes that its arguments dictionaries constrain all
2822 the same type variable.
2824 ADR Comment 20/6/94: I've changed the @CReturnable@ case to default to
2825 @()@ instead of @Int@. I reckon this is the Right Thing to do since
2826 the most common use of defaulting is code like:
2828 _ccall_ foo `seqPrimIO` bar
2830 Since we're not using the result of @foo@, the result if (presumably)
2834 disambiguate :: SDoc -> Bool -> DynFlags -> [Inst] -> TcM ([Inst], TcDictBinds)
2835 -- Just does unification to fix the default types
2836 -- The Insts are assumed to be pre-zonked
2837 disambiguate doc interactive dflags insts
2839 = return (insts, emptyBag)
2841 | null defaultable_groups
2842 = do { traceTc (text "disambigutate, no defaultable groups" <+> vcat [ppr unaries, ppr insts, ppr bad_tvs, ppr defaultable_groups])
2843 ; return (insts, emptyBag) }
2846 = do { -- Figure out what default types to use
2847 default_tys <- getDefaultTys extended_defaulting ovl_strings
2849 ; traceTc (text "disambiguate1" <+> vcat [ppr insts, ppr unaries, ppr bad_tvs, ppr defaultable_groups])
2850 ; mapM_ (disambigGroup default_tys) defaultable_groups
2852 -- disambigGroup does unification, hence try again
2853 ; tryHardCheckLoop doc insts }
2856 extended_defaulting = interactive || dopt Opt_ExtendedDefaultRules dflags
2857 -- See also Trac #1974
2858 ovl_strings = dopt Opt_OverloadedStrings dflags
2860 unaries :: [(Inst, Class, TcTyVar)] -- (C tv) constraints
2861 bad_tvs :: TcTyVarSet -- Tyvars mentioned by *other* constraints
2862 (unaries, bad_tvs_s) = partitionWith find_unary insts
2863 bad_tvs = unionVarSets bad_tvs_s
2865 -- Finds unary type-class constraints
2866 find_unary d@(Dict {tci_pred = ClassP cls [ty]})
2867 | Just tv <- tcGetTyVar_maybe ty = Left (d,cls,tv)
2868 find_unary inst = Right (tyVarsOfInst inst)
2870 -- Group by type variable
2871 defaultable_groups :: [[(Inst,Class,TcTyVar)]]
2872 defaultable_groups = filter defaultable_group (equivClasses cmp_tv unaries)
2873 cmp_tv (_,_,tv1) (_,_,tv2) = tv1 `compare` tv2
2875 defaultable_group :: [(Inst,Class,TcTyVar)] -> Bool
2876 defaultable_group ds@((_,_,tv):_)
2877 = isTyConableTyVar tv -- Note [Avoiding spurious errors]
2878 && not (tv `elemVarSet` bad_tvs)
2879 && defaultable_classes [c | (_,c,_) <- ds]
2880 defaultable_group [] = panic "defaultable_group"
2882 defaultable_classes clss
2883 | extended_defaulting = any isInteractiveClass clss
2884 | otherwise = all is_std_class clss && (any is_num_class clss)
2886 -- In interactive mode, or with -XExtendedDefaultRules,
2887 -- we default Show a to Show () to avoid graututious errors on "show []"
2888 isInteractiveClass cls
2889 = is_num_class cls || (classKey cls `elem` [showClassKey, eqClassKey, ordClassKey])
2891 is_num_class cls = isNumericClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2892 -- is_num_class adds IsString to the standard numeric classes,
2893 -- when -foverloaded-strings is enabled
2895 is_std_class cls = isStandardClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2896 -- Similarly is_std_class
2898 -----------------------
2899 disambigGroup :: [Type] -- The default types
2900 -> [(Inst,Class,TcTyVar)] -- All standard classes of form (C a)
2901 -> TcM () -- Just does unification, to fix the default types
2903 disambigGroup default_tys dicts
2904 = do { mb_chosen_ty <- try_default default_tys
2905 ; case mb_chosen_ty of
2906 Nothing -> return ()
2907 Just chosen_ty -> do { _ <- unifyType chosen_ty (mkTyVarTy tyvar)
2908 ; warnDefault dicts chosen_ty } }
2910 (_,_,tyvar) = ASSERT(not (null dicts)) head dicts -- Should be non-empty
2911 classes = [c | (_,c,_) <- dicts]
2913 try_default [] = return Nothing
2914 try_default (default_ty : default_tys)
2915 = tryTcLIE_ (try_default default_tys) $
2916 do { tcSimplifyDefault [mkClassPred clas [default_ty] | clas <- classes]
2917 -- This may fail; then the tryTcLIE_ kicks in
2918 -- Failure here is caused by there being no type in the
2919 -- default list which can satisfy all the ambiguous classes.
2920 -- For example, if Real a is reqd, but the only type in the
2921 -- default list is Int.
2923 ; return (Just default_ty) -- TOMDO: do something with the coercion
2927 -----------------------
2928 getDefaultTys :: Bool -> Bool -> TcM [Type]
2929 getDefaultTys extended_deflts ovl_strings
2930 = do { mb_defaults <- getDeclaredDefaultTys
2931 ; case mb_defaults of {
2932 Just tys -> return tys ; -- User-supplied defaults
2935 -- No use-supplied default
2936 -- Use [Integer, Double], plus modifications
2937 { integer_ty <- tcMetaTy integerTyConName
2938 ; checkWiredInTyCon doubleTyCon
2939 ; string_ty <- tcMetaTy stringTyConName
2940 ; return (opt_deflt extended_deflts unitTy
2941 -- Note [Default unitTy]
2943 [integer_ty,doubleTy]
2945 opt_deflt ovl_strings string_ty) } } }
2947 opt_deflt True ty = [ty]
2948 opt_deflt False _ = []
2951 Note [Default unitTy]
2952 ~~~~~~~~~~~~~~~~~~~~~
2953 In interative mode (or with -XExtendedDefaultRules) we add () as the first type we
2954 try when defaulting. This has very little real impact, except in the following case.
2956 Text.Printf.printf "hello"
2957 This has type (forall a. IO a); it prints "hello", and returns 'undefined'. We don't
2958 want the GHCi repl loop to try to print that 'undefined'. The neatest thing is to
2959 default the 'a' to (), rather than to Integer (which is what would otherwise happen;
2960 and then GHCi doesn't attempt to print the (). So in interactive mode, we add
2961 () to the list of defaulting types. See Trac #1200.
2963 Note [Avoiding spurious errors]
2964 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2965 When doing the unification for defaulting, we check for skolem
2966 type variables, and simply don't default them. For example:
2967 f = (*) -- Monomorphic
2968 g :: Num a => a -> a
2970 Here, we get a complaint when checking the type signature for g,
2971 that g isn't polymorphic enough; but then we get another one when
2972 dealing with the (Num a) context arising from f's definition;
2973 we try to unify a with Int (to default it), but find that it's
2974 already been unified with the rigid variable from g's type sig
2977 %************************************************************************
2979 \subsection[simple]{@Simple@ versions}
2981 %************************************************************************
2983 Much simpler versions when there are no bindings to make!
2985 @tcSimplifyThetas@ simplifies class-type constraints formed by
2986 @deriving@ declarations and when specialising instances. We are
2987 only interested in the simplified bunch of class/type constraints.
2989 It simplifies to constraints of the form (C a b c) where
2990 a,b,c are type variables. This is required for the context of
2991 instance declarations.
2994 tcSimplifyDeriv :: InstOrigin
2996 -> ThetaType -- Wanted
2997 -> TcM ThetaType -- Needed
2998 -- Given instance (wanted) => C inst_ty
2999 -- Simplify 'wanted' as much as possible
3001 tcSimplifyDeriv orig tyvars theta
3002 = do { (tvs, _, tenv) <- tcInstTyVars tyvars
3003 -- The main loop may do unification, and that may crash if
3004 -- it doesn't see a TcTyVar, so we have to instantiate. Sigh
3005 -- ToDo: what if two of them do get unified?
3006 ; wanteds <- newDictBndrsO orig (substTheta tenv theta)
3007 ; (irreds, _) <- tryHardCheckLoop doc wanteds
3009 ; let (tv_dicts, others) = partition ok irreds
3010 (tidy_env, tidy_insts) = tidyInsts others
3011 ; reportNoInstances tidy_env Nothing [alt_fix] tidy_insts
3012 -- See Note [Exotic derived instance contexts] in TcMType
3014 ; let rev_env = zipTopTvSubst tvs (mkTyVarTys tyvars)
3015 simpl_theta = substTheta rev_env (map dictPred tv_dicts)
3016 -- This reverse-mapping is a pain, but the result
3017 -- should mention the original TyVars not TcTyVars
3019 ; return simpl_theta }
3021 doc = ptext (sLit "deriving classes for a data type")
3023 ok dict | isDict dict = validDerivPred (dictPred dict)
3025 alt_fix = vcat [ptext (sLit "use a standalone 'deriving instance' declaration instead,"),
3026 ptext (sLit "so you can specify the instance context yourself")]
3030 @tcSimplifyDefault@ just checks class-type constraints, essentially;
3031 used with \tr{default} declarations. We are only interested in
3032 whether it worked or not.
3035 tcSimplifyDefault :: ThetaType -- Wanted; has no type variables in it
3038 tcSimplifyDefault theta = do
3039 wanteds <- newDictBndrsO DefaultOrigin theta
3040 (irreds, _) <- tryHardCheckLoop doc wanteds
3041 addNoInstanceErrs irreds
3045 traceTc (ptext (sLit "tcSimplifyDefault failing")) >> failM
3047 doc = ptext (sLit "default declaration")
3050 @tcSimplifyStagedExpr@ performs a simplification but does so at a new
3051 stage. This is used when typechecking annotations and splices.
3055 tcSimplifyStagedExpr :: ThStage -> TcM a -> TcM (a, TcDictBinds)
3056 -- Type check an expression that runs at a top level stage as if
3057 -- it were going to be spliced and then simplify it
3058 tcSimplifyStagedExpr stage tc_action
3059 = setStage stage $ do {
3060 -- Typecheck the expression
3061 (thing', lie) <- getLIE tc_action
3063 -- Solve the constraints
3064 ; const_binds <- tcSimplifyTop lie
3066 ; return (thing', const_binds) }
3071 %************************************************************************
3073 \section{Errors and contexts}
3075 %************************************************************************
3077 ToDo: for these error messages, should we note the location as coming
3078 from the insts, or just whatever seems to be around in the monad just
3082 groupErrs :: ([Inst] -> TcM ()) -- Deal with one group
3083 -> [Inst] -- The offending Insts
3085 -- Group together insts with the same origin
3086 -- We want to report them together in error messages
3090 groupErrs report_err (inst:insts)
3091 = do { do_one (inst:friends)
3092 ; groupErrs report_err others }
3094 -- (It may seem a bit crude to compare the error messages,
3095 -- but it makes sure that we combine just what the user sees,
3096 -- and it avoids need equality on InstLocs.)
3097 (friends, others) = partition is_friend insts
3098 loc_msg = showSDoc (pprInstLoc (instLoc inst))
3099 is_friend friend = showSDoc (pprInstLoc (instLoc friend)) == loc_msg
3100 do_one insts = addInstCtxt (instLoc (head insts)) (report_err insts)
3101 -- Add location and context information derived from the Insts
3103 -- Add the "arising from..." part to a message about bunch of dicts
3104 addInstLoc :: [Inst] -> Message -> Message
3105 addInstLoc insts msg = msg $$ nest 2 (pprInstArising (head insts))
3107 addTopIPErrs :: [Name] -> [Inst] -> TcM ()
3110 addTopIPErrs bndrs ips
3111 = do { dflags <- getDOpts
3112 ; addErrTcM (tidy_env, mk_msg dflags tidy_ips) }
3114 (tidy_env, tidy_ips) = tidyInsts ips
3116 = vcat [sep [ptext (sLit "Implicit parameters escape from"),
3117 nest 2 (ptext (sLit "the monomorphic top-level binding")
3118 <> plural bndrs <+> ptext (sLit "of")
3119 <+> pprBinders bndrs <> colon)],
3120 nest 2 (vcat (map ppr_ip ips)),
3121 monomorphism_fix dflags]
3122 ppr_ip ip = pprPred (dictPred ip) <+> pprInstArising ip
3124 topIPErrs :: [Inst] -> TcM ()
3126 = groupErrs report tidy_dicts
3128 (tidy_env, tidy_dicts) = tidyInsts dicts
3129 report dicts = addErrTcM (tidy_env, mk_msg dicts)
3130 mk_msg dicts = addInstLoc dicts (ptext (sLit "Unbound implicit parameter") <>
3131 plural tidy_dicts <+> pprDictsTheta tidy_dicts)
3133 addNoInstanceErrs :: [Inst] -- Wanted (can include implications)
3135 addNoInstanceErrs insts
3136 = do { let (tidy_env, tidy_insts) = tidyInsts insts
3137 ; reportNoInstances tidy_env Nothing [] tidy_insts }
3141 -> Maybe (InstLoc, [Inst]) -- Context
3142 -- Nothing => top level
3143 -- Just (d,g) => d describes the construct
3145 -> [SDoc] -- Alternative fix for no-such-instance
3146 -> [Inst] -- What is wanted (can include implications)
3149 reportNoInstances tidy_env mb_what alt_fix insts
3150 = groupErrs (report_no_instances tidy_env mb_what alt_fix) insts
3152 report_no_instances :: TidyEnv -> Maybe (InstLoc, [Inst]) -> [SDoc] -> [Inst] -> TcM ()
3153 report_no_instances tidy_env mb_what alt_fixes insts
3154 = do { inst_envs <- tcGetInstEnvs
3155 ; let (implics, insts1) = partition isImplicInst insts
3156 (insts2, overlaps) = partitionWith (check_overlap inst_envs) insts1
3157 (eqInsts, insts3) = partition isEqInst insts2
3158 ; traceTc (text "reportNoInstances" <+> vcat
3159 [ppr insts, ppr implics, ppr insts1, ppr insts2])
3160 ; mapM_ complain_implic implics
3161 ; mapM_ (\doc -> addErrTcM (tidy_env, doc)) overlaps
3162 ; groupErrs complain_no_inst insts3
3163 ; mapM_ (addErrTcM . mk_eq_err) eqInsts
3166 complain_no_inst insts = addErrTcM (tidy_env, mk_no_inst_err insts)
3168 complain_implic inst -- Recurse!
3169 = reportNoInstances tidy_env
3170 (Just (tci_loc inst, tci_given inst))
3171 alt_fixes (tci_wanted inst)
3173 check_overlap :: (InstEnv,InstEnv) -> Inst -> Either Inst SDoc
3174 -- Right msg => overlap message
3175 -- Left inst => no instance
3176 check_overlap inst_envs wanted
3177 | not (isClassDict wanted) = Left wanted
3179 = case lookupInstEnv inst_envs clas tys of
3180 ([], _) -> Left wanted -- No match
3181 -- The case of exactly one match and no unifiers means a
3182 -- successful lookup. That can't happen here, because dicts
3183 -- only end up here if they didn't match in Inst.lookupInst
3185 | debugIsOn -> pprPanic "reportNoInstance" (ppr wanted)
3186 res -> Right (mk_overlap_msg wanted res)
3188 (clas,tys) = getDictClassTys wanted
3190 mk_overlap_msg dict (matches, unifiers)
3191 = ASSERT( not (null matches) )
3192 vcat [ addInstLoc [dict] ((ptext (sLit "Overlapping instances for")
3193 <+> pprPred (dictPred dict))),
3194 sep [ptext (sLit "Matching instances") <> colon,
3195 nest 2 (vcat [pprInstances ispecs, pprInstances unifiers])],
3196 if not (isSingleton matches)
3197 then -- Two or more matches
3199 else -- One match, plus some unifiers
3200 ASSERT( not (null unifiers) )
3201 parens (vcat [ptext (sLit "The choice depends on the instantiation of") <+>
3202 quotes (pprWithCommas ppr (varSetElems (tyVarsOfInst dict))),
3203 ptext (sLit "To pick the first instance above, use -XIncoherentInstances"),
3204 ptext (sLit "when compiling the other instance declarations")])]
3206 ispecs = [ispec | (ispec, _) <- matches]
3208 mk_eq_err :: Inst -> (TidyEnv, SDoc)
3209 mk_eq_err inst = misMatchMsg tidy_env (eqInstTys inst)
3211 mk_no_inst_err insts
3212 | null insts = empty
3214 | Just (loc, givens) <- mb_what, -- Nested (type signatures, instance decls)
3215 not (isEmptyVarSet (tyVarsOfInsts insts))
3216 = vcat [ addInstLoc insts $
3217 sep [ ptext (sLit "Could not deduce") <+> pprDictsTheta insts
3218 , nest 2 $ ptext (sLit "from the context") <+> pprDictsTheta givens]
3219 , show_fixes (fix1 loc : fixes2 ++ alt_fixes) ]
3221 | otherwise -- Top level
3222 = vcat [ addInstLoc insts $
3223 ptext (sLit "No instance") <> plural insts
3224 <+> ptext (sLit "for") <+> pprDictsTheta insts
3225 , show_fixes (fixes2 ++ alt_fixes) ]
3228 fix1 loc = sep [ ptext (sLit "add") <+> pprDictsTheta insts
3229 <+> ptext (sLit "to the context of"),
3230 nest 2 (ppr (instLocOrigin loc)) ]
3231 -- I'm not sure it helps to add the location
3232 -- nest 2 (ptext (sLit "at") <+> ppr (instLocSpan loc)) ]
3234 fixes2 | null instance_dicts = []
3235 | otherwise = [sep [ptext (sLit "add an instance declaration for"),
3236 pprDictsTheta instance_dicts]]
3237 instance_dicts = [d | d <- insts, isClassDict d, not (isTyVarDict d)]
3238 -- Insts for which it is worth suggesting an adding an instance declaration
3239 -- Exclude implicit parameters, and tyvar dicts
3241 show_fixes :: [SDoc] -> SDoc
3242 show_fixes [] = empty
3243 show_fixes (f:fs) = sep [ptext (sLit "Possible fix:"),
3244 nest 2 (vcat (f : map (ptext (sLit "or") <+>) fs))]
3246 addTopAmbigErrs :: [Inst] -> TcRn ()
3247 addTopAmbigErrs dicts
3248 -- Divide into groups that share a common set of ambiguous tyvars
3249 = ifErrsM (return ()) $ -- Only report ambiguity if no other errors happened
3250 -- See Note [Avoiding spurious errors]
3251 mapM_ report (equivClasses cmp [(d, tvs_of d) | d <- tidy_dicts])
3253 (tidy_env, tidy_dicts) = tidyInsts dicts
3255 tvs_of :: Inst -> [TcTyVar]
3256 tvs_of d = varSetElems (tyVarsOfInst d)
3257 cmp (_,tvs1) (_,tvs2) = tvs1 `compare` tvs2
3259 report :: [(Inst,[TcTyVar])] -> TcM ()
3260 report pairs@((inst,tvs) : _) = do -- The pairs share a common set of ambiguous tyvars
3261 (tidy_env, mono_msg) <- mkMonomorphismMsg tidy_env tvs
3262 setSrcSpan (instSpan inst) $
3263 -- the location of the first one will do for the err message
3264 addErrTcM (tidy_env, msg $$ mono_msg)
3266 dicts = map fst pairs
3267 msg = sep [text "Ambiguous type variable" <> plural tvs <+>
3268 pprQuotedList tvs <+> in_msg,
3269 nest 2 (pprDictsInFull dicts)]
3270 in_msg = text "in the constraint" <> plural dicts <> colon
3271 report [] = panic "addTopAmbigErrs"
3274 mkMonomorphismMsg :: TidyEnv -> [TcTyVar] -> TcM (TidyEnv, Message)
3275 -- There's an error with these Insts; if they have free type variables
3276 -- it's probably caused by the monomorphism restriction.
3277 -- Try to identify the offending variable
3278 -- ASSUMPTION: the Insts are fully zonked
3279 mkMonomorphismMsg tidy_env inst_tvs
3280 = do { dflags <- getDOpts
3281 ; (tidy_env, docs) <- findGlobals (mkVarSet inst_tvs) tidy_env
3282 ; return (tidy_env, mk_msg dflags docs) }
3284 mk_msg _ _ | any isRuntimeUnk inst_tvs
3285 = vcat [ptext (sLit "Cannot resolve unknown runtime types:") <+>
3286 (pprWithCommas ppr inst_tvs),
3287 ptext (sLit "Use :print or :force to determine these types")]
3288 mk_msg _ [] = ptext (sLit "Probable fix: add a type signature that fixes these type variable(s)")
3289 -- This happens in things like
3290 -- f x = show (read "foo")
3291 -- where monomorphism doesn't play any role
3293 = vcat [ptext (sLit "Possible cause: the monomorphism restriction applied to the following:"),
3295 monomorphism_fix dflags]
3297 monomorphism_fix :: DynFlags -> SDoc
3298 monomorphism_fix dflags
3299 = ptext (sLit "Probable fix:") <+> vcat
3300 [ptext (sLit "give these definition(s) an explicit type signature"),
3301 if dopt Opt_MonomorphismRestriction dflags
3302 then ptext (sLit "or use -XNoMonomorphismRestriction")
3303 else empty] -- Only suggest adding "-XNoMonomorphismRestriction"
3304 -- if it is not already set!
3306 warnDefault :: [(Inst, Class, Var)] -> Type -> TcM ()
3307 warnDefault ups default_ty = do
3308 warn_flag <- doptM Opt_WarnTypeDefaults
3309 addInstCtxt (instLoc (head (dicts))) (warnTc warn_flag warn_msg)
3311 dicts = [d | (d,_,_) <- ups]
3314 (_, tidy_dicts) = tidyInsts dicts
3315 warn_msg = vcat [ptext (sLit "Defaulting the following constraint(s) to type") <+>
3316 quotes (ppr default_ty),
3317 pprDictsInFull tidy_dicts]
3319 reduceDepthErr :: Int -> [Inst] -> SDoc
3320 reduceDepthErr n stack
3321 = vcat [ptext (sLit "Context reduction stack overflow; size =") <+> int n,
3322 ptext (sLit "Use -fcontext-stack=N to increase stack size to N"),
3323 nest 4 (pprStack stack)]
3325 pprStack :: [Inst] -> SDoc
3326 pprStack stack = vcat (map pprInstInFull stack)