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,
23 #include "HsVersions.h"
25 import {-# SOURCE #-} TcUnify( unifyType )
29 import TcHsSyn ( hsLPatType )
37 import DsUtils -- Big-tuple functions
66 %************************************************************************
70 %************************************************************************
72 --------------------------------------
73 Notes on functional dependencies (a bug)
74 --------------------------------------
81 instance D a b => C a b -- Undecidable
82 -- (Not sure if it's crucial to this eg)
83 f :: C a b => a -> Bool
86 g :: C a b => a -> Bool
89 Here f typechecks, but g does not!! Reason: before doing improvement,
90 we reduce the (C a b1) constraint from the call of f to (D a b1).
92 Here is a more complicated example:
94 | > class Foo a b | a->b
96 | > class Bar a b | a->b
100 | > instance Bar Obj Obj
102 | > instance (Bar a b) => Foo a b
104 | > foo:: (Foo a b) => a -> String
107 | > runFoo:: (forall a b. (Foo a b) => a -> w) -> w
113 | Could not deduce (Bar a b) from the context (Foo a b)
114 | arising from use of `foo' at <interactive>:1
116 | Add (Bar a b) to the expected type of an expression
117 | In the first argument of `runFoo', namely `foo'
118 | In the definition of `it': it = runFoo foo
120 | Why all of the sudden does GHC need the constraint Bar a b? The
121 | function foo didn't ask for that...
123 The trouble is that to type (runFoo foo), GHC has to solve the problem:
125 Given constraint Foo a b
126 Solve constraint Foo a b'
128 Notice that b and b' aren't the same. To solve this, just do
129 improvement and then they are the same. But GHC currently does
134 That is usually fine, but it isn't here, because it sees that Foo a b is
135 not the same as Foo a b', and so instead applies the instance decl for
136 instance Bar a b => Foo a b. And that's where the Bar constraint comes
139 The Right Thing is to improve whenever the constraint set changes at
140 all. Not hard in principle, but it'll take a bit of fiddling to do.
142 Note [Choosing which variables to quantify]
143 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
144 Suppose we are about to do a generalisation step. We have in our hand
147 T the type of the RHS
148 C the constraints from that RHS
150 The game is to figure out
152 Q the set of type variables over which to quantify
153 Ct the constraints we will *not* quantify over
154 Cq the constraints we will quantify over
156 So we're going to infer the type
160 and float the constraints Ct further outwards.
162 Here are the things that *must* be true:
164 (A) Q intersect fv(G) = EMPTY limits how big Q can be
165 (B) Q superset fv(Cq union T) \ oclose(fv(G),C) limits how small Q can be
167 (A) says we can't quantify over a variable that's free in the environment.
168 (B) says we must quantify over all the truly free variables in T, else
169 we won't get a sufficiently general type.
171 We do not *need* to quantify over any variable that is fixed by the
172 free vars of the environment G.
174 BETWEEN THESE TWO BOUNDS, ANY Q WILL DO!
176 Example: class H x y | x->y where ...
178 fv(G) = {a} C = {H a b, H c d}
181 (A) Q intersect {a} is empty
182 (B) Q superset {a,b,c,d} \ oclose({a}, C) = {a,b,c,d} \ {a,b} = {c,d}
184 So Q can be {c,d}, {b,c,d}
186 In particular, it's perfectly OK to quantify over more type variables
187 than strictly necessary; there is no need to quantify over 'b', since
188 it is determined by 'a' which is free in the envt, but it's perfectly
189 OK to do so. However we must not quantify over 'a' itself.
191 Other things being equal, however, we'd like to quantify over as few
192 variables as possible: smaller types, fewer type applications, more
193 constraints can get into Ct instead of Cq. Here's a good way to
196 Q = grow( fv(T), C ) \ oclose( fv(G), C )
198 That is, quantify over all variable that that MIGHT be fixed by the
199 call site (which influences T), but which aren't DEFINITELY fixed by
200 G. This choice definitely quantifies over enough type variables,
201 albeit perhaps too many.
203 Why grow( fv(T), C ) rather than fv(T)? Consider
205 class H x y | x->y where ...
210 If we used fv(T) = {c} we'd get the type
212 forall c. H c d => c -> b
214 And then if the fn was called at several different c's, each of
215 which fixed d differently, we'd get a unification error, because
216 d isn't quantified. Solution: quantify d. So we must quantify
217 everything that might be influenced by c.
219 Why not oclose( fv(T), C )? Because we might not be able to see
220 all the functional dependencies yet:
222 class H x y | x->y where ...
223 instance H x y => Eq (T x y) where ...
228 Now oclose(fv(T),C) = {c}, because the functional dependency isn't
229 apparent yet, and that's wrong. We must really quantify over d too.
231 There really isn't any point in quantifying over any more than
232 grow( fv(T), C ), because the call sites can't possibly influence
233 any other type variables.
237 -------------------------------------
239 -------------------------------------
241 It's very hard to be certain when a type is ambiguous. Consider
245 instance H x y => K (x,y)
247 Is this type ambiguous?
248 forall a b. (K (a,b), Eq b) => a -> a
250 Looks like it! But if we simplify (K (a,b)) we get (H a b) and
251 now we see that a fixes b. So we can't tell about ambiguity for sure
252 without doing a full simplification. And even that isn't possible if
253 the context has some free vars that may get unified. Urgle!
255 Here's another example: is this ambiguous?
256 forall a b. Eq (T b) => a -> a
257 Not if there's an insance decl (with no context)
258 instance Eq (T b) where ...
260 You may say of this example that we should use the instance decl right
261 away, but you can't always do that:
263 class J a b where ...
264 instance J Int b where ...
266 f :: forall a b. J a b => a -> a
268 (Notice: no functional dependency in J's class decl.)
269 Here f's type is perfectly fine, provided f is only called at Int.
270 It's premature to complain when meeting f's signature, or even
271 when inferring a type for f.
275 However, we don't *need* to report ambiguity right away. It'll always
276 show up at the call site.... and eventually at main, which needs special
277 treatment. Nevertheless, reporting ambiguity promptly is an excellent thing.
279 So here's the plan. We WARN about probable ambiguity if
281 fv(Cq) is not a subset of oclose(fv(T) union fv(G), C)
283 (all tested before quantification).
284 That is, all the type variables in Cq must be fixed by the the variables
285 in the environment, or by the variables in the type.
287 Notice that we union before calling oclose. Here's an example:
289 class J a b c | a b -> c
293 forall b c. (J a b c) => b -> b
295 Only if we union {a} from G with {b} from T before using oclose,
296 do we see that c is fixed.
298 It's a bit vague exactly which C we should use for this oclose call. If we
299 don't fix enough variables we might complain when we shouldn't (see
300 the above nasty example). Nothing will be perfect. That's why we can
301 only issue a warning.
304 Can we ever be *certain* about ambiguity? Yes: if there's a constraint
306 c in C such that fv(c) intersect (fv(G) union fv(T)) = EMPTY
308 then c is a "bubble"; there's no way it can ever improve, and it's
309 certainly ambiguous. UNLESS it is a constant (sigh). And what about
314 instance H x y => K (x,y)
316 Is this type ambiguous?
317 forall a b. (K (a,b), Eq b) => a -> a
319 Urk. The (Eq b) looks "definitely ambiguous" but it isn't. What we are after
320 is a "bubble" that's a set of constraints
322 Cq = Ca union Cq' st fv(Ca) intersect (fv(Cq') union fv(T) union fv(G)) = EMPTY
324 Hence another idea. To decide Q start with fv(T) and grow it
325 by transitive closure in Cq (no functional dependencies involved).
326 Now partition Cq using Q, leaving the definitely-ambiguous and probably-ok.
327 The definitely-ambiguous can then float out, and get smashed at top level
328 (which squashes out the constants, like Eq (T a) above)
331 --------------------------------------
332 Notes on principal types
333 --------------------------------------
338 f x = let g y = op (y::Int) in True
340 Here the principal type of f is (forall a. a->a)
341 but we'll produce the non-principal type
342 f :: forall a. C Int => a -> a
345 --------------------------------------
346 The need for forall's in constraints
347 --------------------------------------
349 [Exchange on Haskell Cafe 5/6 Dec 2000]
351 class C t where op :: t -> Bool
352 instance C [t] where op x = True
354 p y = (let f :: c -> Bool; f x = op (y >> return x) in f, y ++ [])
355 q y = (y ++ [], let f :: c -> Bool; f x = op (y >> return x) in f)
357 The definitions of p and q differ only in the order of the components in
358 the pair on their right-hand sides. And yet:
360 ghc and "Typing Haskell in Haskell" reject p, but accept q;
361 Hugs rejects q, but accepts p;
362 hbc rejects both p and q;
363 nhc98 ... (Malcolm, can you fill in the blank for us!).
365 The type signature for f forces context reduction to take place, and
366 the results of this depend on whether or not the type of y is known,
367 which in turn depends on which component of the pair the type checker
370 Solution: if y::m a, float out the constraints
371 Monad m, forall c. C (m c)
372 When m is later unified with [], we can solve both constraints.
375 --------------------------------------
376 Notes on implicit parameters
377 --------------------------------------
379 Note [Inheriting implicit parameters]
380 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
385 where f is *not* a top-level binding.
386 From the RHS of f we'll get the constraint (?y::Int).
387 There are two types we might infer for f:
391 (so we get ?y from the context of f's definition), or
393 f :: (?y::Int) => Int -> Int
395 At first you might think the first was better, becuase then
396 ?y behaves like a free variable of the definition, rather than
397 having to be passed at each call site. But of course, the WHOLE
398 IDEA is that ?y should be passed at each call site (that's what
399 dynamic binding means) so we'd better infer the second.
401 BOTTOM LINE: when *inferring types* you *must* quantify
402 over implicit parameters. See the predicate isFreeWhenInferring.
405 Note [Implicit parameters and ambiguity]
406 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
407 Only a *class* predicate can give rise to ambiguity
408 An *implicit parameter* cannot. For example:
409 foo :: (?x :: [a]) => Int
411 is fine. The call site will suppply a particular 'x'
413 Furthermore, the type variables fixed by an implicit parameter
414 propagate to the others. E.g.
415 foo :: (Show a, ?x::[a]) => Int
417 The type of foo looks ambiguous. But it isn't, because at a call site
419 let ?x = 5::Int in foo
420 and all is well. In effect, implicit parameters are, well, parameters,
421 so we can take their type variables into account as part of the
422 "tau-tvs" stuff. This is done in the function 'FunDeps.grow'.
425 Question 2: type signatures
426 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
427 BUT WATCH OUT: When you supply a type signature, we can't force you
428 to quantify over implicit parameters. For example:
432 This is perfectly reasonable. We do not want to insist on
434 (?x + 1) :: (?x::Int => Int)
436 That would be silly. Here, the definition site *is* the occurrence site,
437 so the above strictures don't apply. Hence the difference between
438 tcSimplifyCheck (which *does* allow implicit paramters to be inherited)
439 and tcSimplifyCheckBind (which does not).
441 What about when you supply a type signature for a binding?
442 Is it legal to give the following explicit, user type
443 signature to f, thus:
448 At first sight this seems reasonable, but it has the nasty property
449 that adding a type signature changes the dynamic semantics.
452 (let f x = (x::Int) + ?y
453 in (f 3, f 3 with ?y=5)) with ?y = 6
459 in (f 3, f 3 with ?y=5)) with ?y = 6
463 Indeed, simply inlining f (at the Haskell source level) would change the
466 Nevertheless, as Launchbury says (email Oct 01) we can't really give the
467 semantics for a Haskell program without knowing its typing, so if you
468 change the typing you may change the semantics.
470 To make things consistent in all cases where we are *checking* against
471 a supplied signature (as opposed to inferring a type), we adopt the
474 a signature does not need to quantify over implicit params.
476 [This represents a (rather marginal) change of policy since GHC 5.02,
477 which *required* an explicit signature to quantify over all implicit
478 params for the reasons mentioned above.]
480 But that raises a new question. Consider
482 Given (signature) ?x::Int
483 Wanted (inferred) ?x::Int, ?y::Bool
485 Clearly we want to discharge the ?x and float the ?y out. But
486 what is the criterion that distinguishes them? Clearly it isn't
487 what free type variables they have. The Right Thing seems to be
488 to float a constraint that
489 neither mentions any of the quantified type variables
490 nor any of the quantified implicit parameters
492 See the predicate isFreeWhenChecking.
495 Question 3: monomorphism
496 ~~~~~~~~~~~~~~~~~~~~~~~~
497 There's a nasty corner case when the monomorphism restriction bites:
501 The argument above suggests that we *must* generalise
502 over the ?y parameter, to get
503 z :: (?y::Int) => Int,
504 but the monomorphism restriction says that we *must not*, giving
506 Why does the momomorphism restriction say this? Because if you have
508 let z = x + ?y in z+z
510 you might not expect the addition to be done twice --- but it will if
511 we follow the argument of Question 2 and generalise over ?y.
514 Question 4: top level
515 ~~~~~~~~~~~~~~~~~~~~~
516 At the top level, monomorhism makes no sense at all.
519 main = let ?x = 5 in print foo
523 woggle :: (?x :: Int) => Int -> Int
526 We definitely don't want (foo :: Int) with a top-level implicit parameter
527 (?x::Int) becuase there is no way to bind it.
532 (A) Always generalise over implicit parameters
533 Bindings that fall under the monomorphism restriction can't
537 * Inlining remains valid
538 * No unexpected loss of sharing
539 * But simple bindings like
541 will be rejected, unless you add an explicit type signature
542 (to avoid the monomorphism restriction)
543 z :: (?y::Int) => Int
545 This seems unacceptable
547 (B) Monomorphism restriction "wins"
548 Bindings that fall under the monomorphism restriction can't
550 Always generalise over implicit parameters *except* for bindings
551 that fall under the monomorphism restriction
554 * Inlining isn't valid in general
555 * No unexpected loss of sharing
556 * Simple bindings like
558 accepted (get value of ?y from binding site)
560 (C) Always generalise over implicit parameters
561 Bindings that fall under the monomorphism restriction can't
562 be generalised, EXCEPT for implicit parameters
564 * Inlining remains valid
565 * Unexpected loss of sharing (from the extra generalisation)
566 * Simple bindings like
568 accepted (get value of ?y from occurrence sites)
573 None of these choices seems very satisfactory. But at least we should
574 decide which we want to do.
576 It's really not clear what is the Right Thing To Do. If you see
580 would you expect the value of ?y to be got from the *occurrence sites*
581 of 'z', or from the valuue of ?y at the *definition* of 'z'? In the
582 case of function definitions, the answer is clearly the former, but
583 less so in the case of non-fucntion definitions. On the other hand,
584 if we say that we get the value of ?y from the definition site of 'z',
585 then inlining 'z' might change the semantics of the program.
587 Choice (C) really says "the monomorphism restriction doesn't apply
588 to implicit parameters". Which is fine, but remember that every
589 innocent binding 'x = ...' that mentions an implicit parameter in
590 the RHS becomes a *function* of that parameter, called at each
591 use of 'x'. Now, the chances are that there are no intervening 'with'
592 clauses that bind ?y, so a decent compiler should common up all
593 those function calls. So I think I strongly favour (C). Indeed,
594 one could make a similar argument for abolishing the monomorphism
595 restriction altogether.
597 BOTTOM LINE: we choose (B) at present. See tcSimplifyRestricted
601 %************************************************************************
603 \subsection{tcSimplifyInfer}
605 %************************************************************************
607 tcSimplify is called when we *inferring* a type. Here's the overall game plan:
609 1. Compute Q = grow( fvs(T), C )
611 2. Partition C based on Q into Ct and Cq. Notice that ambiguous
612 predicates will end up in Ct; we deal with them at the top level
614 3. Try improvement, using functional dependencies
616 4. If Step 3 did any unification, repeat from step 1
617 (Unification can change the result of 'grow'.)
619 Note: we don't reduce dictionaries in step 2. For example, if we have
620 Eq (a,b), we don't simplify to (Eq a, Eq b). So Q won't be different
621 after step 2. However note that we may therefore quantify over more
622 type variables than we absolutely have to.
624 For the guts, we need a loop, that alternates context reduction and
625 improvement with unification. E.g. Suppose we have
627 class C x y | x->y where ...
629 and tcSimplify is called with:
631 Then improvement unifies a with b, giving
634 If we need to unify anything, we rattle round the whole thing all over
641 -> TcTyVarSet -- fv(T); type vars
643 -> TcM ([TcTyVar], -- Tyvars to quantify (zonked and quantified)
644 [Inst], -- Dict Ids that must be bound here (zonked)
645 TcDictBinds) -- Bindings
646 -- Any free (escaping) Insts are tossed into the environment
651 tcSimplifyInfer doc tau_tvs wanted
652 = do { tau_tvs1 <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
653 ; wanted' <- mapM zonkInst wanted -- Zonk before deciding quantified tyvars
654 ; gbl_tvs <- tcGetGlobalTyVars
655 ; let preds1 = fdPredsOfInsts wanted'
656 gbl_tvs1 = oclose preds1 gbl_tvs
657 qtvs = grow preds1 tau_tvs1 `minusVarSet` gbl_tvs1
658 -- See Note [Choosing which variables to quantify]
660 -- To maximise sharing, remove from consideration any
661 -- constraints that don't mention qtvs at all
662 ; let (free, bound) = partition (isFreeWhenInferring qtvs) wanted'
665 -- To make types simple, reduce as much as possible
666 ; traceTc (text "infer" <+> (ppr preds1 $$ ppr (grow preds1 tau_tvs1) $$ ppr gbl_tvs $$
667 ppr gbl_tvs1 $$ ppr free $$ ppr bound))
668 ; (irreds1, binds1) <- tryHardCheckLoop doc bound
670 -- Note [Inference and implication constraints]
671 ; let want_dict d = tyVarsOfInst d `intersectsVarSet` qtvs
672 ; (irreds2, binds2) <- approximateImplications doc want_dict irreds1
674 -- Now work out all over again which type variables to quantify,
675 -- exactly in the same way as before, but starting from irreds2. Why?
676 -- a) By now improvment may have taken place, and we must *not*
677 -- quantify over any variable free in the environment
678 -- tc137 (function h inside g) is an example
680 -- b) Do not quantify over constraints that *now* do not
681 -- mention quantified type variables, because they are
682 -- simply ambiguous (or might be bound further out). Example:
683 -- f :: Eq b => a -> (a, b)
685 -- From the RHS of g we get the MethodInst f77 :: alpha -> (alpha, beta)
686 -- We decide to quantify over 'alpha' alone, but free1 does not include f77
687 -- because f77 mentions 'alpha'. Then reducing leaves only the (ambiguous)
688 -- constraint (Eq beta), which we dump back into the free set
689 -- See test tcfail181
691 -- c) irreds may contain type variables not previously mentioned,
692 -- e.g. instance D a x => Foo [a]
694 -- Then after simplifying we'll get (D a x), and x is fresh
695 -- We must quantify over x else it'll be totally unbound
696 ; tau_tvs2 <- zonkTcTyVarsAndFV (varSetElems tau_tvs1)
697 ; gbl_tvs2 <- zonkTcTyVarsAndFV (varSetElems gbl_tvs1)
698 -- Note that we start from gbl_tvs1
699 -- We use tcGetGlobalTyVars, then oclose wrt preds2, because
700 -- we've already put some of the original preds1 into frees
701 -- E.g. wanteds = C a b (where a->b)
704 -- Then b is fixed by gbl_tvs, so (C a b) will be in free, and
705 -- irreds2 will be empty. But we don't want to generalise over b!
706 ; let preds2 = fdPredsOfInsts irreds2 -- irreds2 is zonked
707 qtvs = grow preds2 tau_tvs2 `minusVarSet` oclose preds2 gbl_tvs2
708 ; let (free, irreds3) = partition (isFreeWhenInferring qtvs) irreds2
711 -- Turn the quantified meta-type variables into real type variables
712 ; qtvs2 <- zonkQuantifiedTyVars (varSetElems qtvs)
714 -- We can't abstract over any remaining unsolved
715 -- implications so instead just float them outwards. Ugh.
716 ; let (q_dicts0, implics) = partition isAbstractableInst irreds3
717 ; loc <- getInstLoc (ImplicOrigin doc)
718 ; implic_bind <- bindIrreds loc qtvs2 q_dicts0 implics
720 -- Prepare equality instances for quantification
721 ; let (q_eqs0,q_dicts) = partition isEqInst q_dicts0
722 ; q_eqs <- mapM finalizeEqInst q_eqs0
724 ; return (qtvs2, q_eqs ++ q_dicts, binds1 `unionBags` binds2 `unionBags` implic_bind) }
725 -- NB: when we are done, we might have some bindings, but
726 -- the final qtvs might be empty. See Note [NO TYVARS] below.
728 approximateImplications :: SDoc -> (Inst -> Bool) -> [Inst] -> TcM ([Inst], TcDictBinds)
729 -- Note [Inference and implication constraints]
730 -- Given a bunch of Dict and ImplicInsts, try to approximate the implications by
731 -- - fetching any dicts inside them that are free
732 -- - using those dicts as cruder constraints, to solve the implications
733 -- - returning the extra ones too
735 approximateImplications doc want_dict irreds
737 = return (irreds, emptyBag)
739 = do { extra_dicts' <- mapM cloneDict extra_dicts
740 ; tryHardCheckLoop doc (extra_dicts' ++ irreds) }
741 -- By adding extra_dicts', we make them
742 -- available to solve the implication constraints
744 extra_dicts = get_dicts (filter isImplicInst irreds)
746 get_dicts :: [Inst] -> [Inst] -- Returns only Dicts
747 -- Find the wanted constraints in implication constraints that satisfy
748 -- want_dict, and are not bound by forall's in the constraint itself
749 get_dicts ds = concatMap get_dict ds
751 get_dict d@(Dict {}) | want_dict d = [d]
753 get_dict (ImplicInst {tci_tyvars = tvs, tci_wanted = wanteds})
754 = [ d | let tv_set = mkVarSet tvs
755 , d <- get_dicts wanteds
756 , not (tyVarsOfInst d `intersectsVarSet` tv_set)]
757 get_dict i@(EqInst {}) | want_dict i = [i]
759 get_dict other = pprPanic "approximateImplications" (ppr other)
762 Note [Inference and implication constraints]
763 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
764 Suppose we have a wanted implication constraint (perhaps arising from
765 a nested pattern match) like
767 and we are now trying to quantify over 'a' when inferring the type for
768 a function. In principle it's possible that there might be an instance
769 instance (C a, E a) => D [a]
770 so the context (E a) would suffice. The Right Thing is to abstract over
771 the implication constraint, but we don't do that (a) because it'll be
772 surprising to programmers and (b) because we don't have the machinery to deal
773 with 'given' implications.
775 So our best approximation is to make (D [a]) part of the inferred
776 context, so we can use that to discharge the implication. Hence
777 the strange function get_dicts in approximateImplications.
779 The common cases are more clear-cut, when we have things like
781 Here, abstracting over (C b) is not an approximation at all -- but see
782 Note [Freeness and implications].
784 See Trac #1430 and test tc228.
788 -----------------------------------------------------------
789 -- tcSimplifyInferCheck is used when we know the constraints we are to simplify
790 -- against, but we don't know the type variables over which we are going to quantify.
791 -- This happens when we have a type signature for a mutually recursive group
794 -> TcTyVarSet -- fv(T)
797 -> TcM ([TyVar], -- Fully zonked, and quantified
798 TcDictBinds) -- Bindings
800 tcSimplifyInferCheck loc tau_tvs givens wanteds
801 = do { traceTc (text "tcSimplifyInferCheck <-" <+> ppr wanteds)
802 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
804 -- Figure out which type variables to quantify over
805 -- You might think it should just be the signature tyvars,
806 -- but in bizarre cases you can get extra ones
807 -- f :: forall a. Num a => a -> a
808 -- f x = fst (g (x, head [])) + 1
810 -- Here we infer g :: forall a b. a -> b -> (b,a)
811 -- We don't want g to be monomorphic in b just because
812 -- f isn't quantified over b.
813 ; let all_tvs = varSetElems (tau_tvs `unionVarSet` tyVarsOfInsts givens)
814 ; all_tvs <- zonkTcTyVarsAndFV all_tvs
815 ; gbl_tvs <- tcGetGlobalTyVars
816 ; let qtvs = varSetElems (all_tvs `minusVarSet` gbl_tvs)
817 -- We could close gbl_tvs, but its not necessary for
818 -- soundness, and it'll only affect which tyvars, not which
819 -- dictionaries, we quantify over
821 ; qtvs' <- zonkQuantifiedTyVars qtvs
823 -- Now we are back to normal (c.f. tcSimplCheck)
824 ; implic_bind <- bindIrreds loc qtvs' givens irreds
826 ; traceTc (text "tcSimplifyInferCheck ->" <+> ppr (implic_bind))
827 ; return (qtvs', binds `unionBags` implic_bind) }
830 Note [Squashing methods]
831 ~~~~~~~~~~~~~~~~~~~~~~~~~
832 Be careful if you want to float methods more:
833 truncate :: forall a. RealFrac a => forall b. Integral b => a -> b
834 From an application (truncate f i) we get
837 If we have also have a second occurrence of truncate, we get
840 When simplifying with i,f free, we might still notice that
841 t1=t3; but alas, the binding for t2 (which mentions t1)
842 may continue to float out!
847 class Y a b | a -> b where
850 instance Y [[a]] a where
853 k :: X a -> X a -> X a
855 g :: Num a => [X a] -> [X a]
858 h ys = ys ++ map (k (y [[0]])) xs
860 The excitement comes when simplifying the bindings for h. Initially
861 try to simplify {y @ [[t1]] t2, 0 @ t1}, with initial qtvs = {t2}.
862 From this we get t1:=:t2, but also various bindings. We can't forget
863 the bindings (because of [LOOP]), but in fact t1 is what g is
866 The net effect of [NO TYVARS]
869 isFreeWhenInferring :: TyVarSet -> Inst -> Bool
870 isFreeWhenInferring qtvs inst
871 = isFreeWrtTyVars qtvs inst -- Constrains no quantified vars
872 && isInheritableInst inst -- and no implicit parameter involved
873 -- see Note [Inheriting implicit parameters]
875 {- No longer used (with implication constraints)
876 isFreeWhenChecking :: TyVarSet -- Quantified tyvars
877 -> NameSet -- Quantified implicit parameters
879 isFreeWhenChecking qtvs ips inst
880 = isFreeWrtTyVars qtvs inst
881 && isFreeWrtIPs ips inst
884 isFreeWrtTyVars :: VarSet -> Inst -> Bool
885 isFreeWrtTyVars qtvs inst = tyVarsOfInst inst `disjointVarSet` qtvs
886 isFreeWrtIPs :: NameSet -> Inst -> Bool
887 isFreeWrtIPs ips inst = not (any (`elemNameSet` ips) (ipNamesOfInst inst))
891 %************************************************************************
893 \subsection{tcSimplifyCheck}
895 %************************************************************************
897 @tcSimplifyCheck@ is used when we know exactly the set of variables
898 we are going to quantify over. For example, a class or instance declaration.
901 -----------------------------------------------------------
902 -- tcSimplifyCheck is used when checking expression type signatures,
903 -- class decls, instance decls etc.
904 tcSimplifyCheck :: InstLoc
905 -> [TcTyVar] -- Quantify over these
908 -> TcM TcDictBinds -- Bindings
909 tcSimplifyCheck loc qtvs givens wanteds
910 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
911 do { traceTc (text "tcSimplifyCheck")
912 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
913 ; implic_bind <- bindIrreds loc qtvs givens irreds
914 ; return (binds `unionBags` implic_bind) }
916 -----------------------------------------------------------
917 -- tcSimplifyCheckPat is used for existential pattern match
918 tcSimplifyCheckPat :: InstLoc
919 -> [TcTyVar] -- Quantify over these
922 -> TcM TcDictBinds -- Bindings
923 tcSimplifyCheckPat loc qtvs givens wanteds
924 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
925 do { traceTc (text "tcSimplifyCheckPat")
926 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
927 ; implic_bind <- bindIrredsR loc qtvs givens irreds
928 ; return (binds `unionBags` implic_bind) }
930 -----------------------------------------------------------
931 bindIrreds :: InstLoc -> [TcTyVar]
934 bindIrreds loc qtvs givens irreds
935 = bindIrredsR loc qtvs givens irreds
937 bindIrredsR :: InstLoc -> [TcTyVar] -> [Inst] -> [Inst] -> TcM TcDictBinds
938 -- Make a binding that binds 'irreds', by generating an implication
939 -- constraint for them, *and* throwing the constraint into the LIE
940 bindIrredsR loc qtvs givens irreds
944 = do { let givens' = filter isAbstractableInst givens
945 -- The givens can (redundantly) include methods
946 -- We want to retain both EqInsts and Dicts
947 -- There should be no implicadtion constraints
948 -- See Note [Pruning the givens in an implication constraint]
950 -- If there are no 'givens', then it's safe to
951 -- partition the 'wanteds' by their qtvs, thereby trimming irreds
952 -- See Note [Freeness and implications]
953 ; irreds' <- if null givens'
955 { let qtv_set = mkVarSet qtvs
956 (frees, real_irreds) = partition (isFreeWrtTyVars qtv_set) irreds
958 ; return real_irreds }
961 ; (implics, bind) <- makeImplicationBind loc qtvs givens' irreds'
962 -- This call does the real work
963 -- If irreds' is empty, it does something sensible
968 makeImplicationBind :: InstLoc -> [TcTyVar]
970 -> TcM ([Inst], TcDictBinds)
971 -- Make a binding that binds 'irreds', by generating an implication
972 -- constraint for them, *and* throwing the constraint into the LIE
973 -- The binding looks like
974 -- (ir1, .., irn) = f qtvs givens
975 -- where f is (evidence for) the new implication constraint
976 -- f :: forall qtvs. {reft} givens => (ir1, .., irn)
977 -- qtvs includes coercion variables
979 -- This binding must line up the 'rhs' in reduceImplication
980 makeImplicationBind loc all_tvs
981 givens -- Guaranteed all Dicts
984 | null irreds -- If there are no irreds, we are done
985 = return ([], emptyBag)
986 | otherwise -- Otherwise we must generate a binding
987 = do { uniq <- newUnique
988 ; span <- getSrcSpanM
989 ; let (eq_givens, dict_givens) = partition isEqInst givens
990 eq_tyvar_cos = mkTyVarTys (varSetElems $ tyVarsOfTypes $ map eqInstType eq_givens)
991 -- Urgh! See line 2187 or thereabouts. I believe that all these
992 -- 'givens' must be a simple CoVar. This MUST be cleaned up.
994 ; let name = mkInternalName uniq (mkVarOcc "ic") span
995 implic_inst = ImplicInst { tci_name = name,
996 tci_tyvars = all_tvs,
997 tci_given = (eq_givens ++ dict_givens),
998 tci_wanted = irreds, tci_loc = loc }
999 ; let -- only create binder for dict_irreds
1000 (_, dict_irreds) = partition isEqInst irreds
1001 dict_irred_ids = map instToId dict_irreds
1002 lpat = mkBigLHsPatTup (map (L span . VarPat) dict_irred_ids)
1003 rhs = L span (mkHsWrap co (HsVar (instToId implic_inst)))
1004 co = mkWpApps (map instToId dict_givens)
1005 <.> mkWpTyApps eq_tyvar_cos
1006 <.> mkWpTyApps (mkTyVarTys all_tvs)
1007 bind | [dict_irred_id] <- dict_irred_ids = VarBind dict_irred_id rhs
1008 | otherwise = PatBind { pat_lhs = lpat,
1009 pat_rhs = unguardedGRHSs rhs,
1010 pat_rhs_ty = hsLPatType lpat,
1011 bind_fvs = placeHolderNames }
1012 ; traceTc $ text "makeImplicationBind" <+> ppr implic_inst
1013 ; return ([implic_inst], unitBag (L span bind))
1016 -----------------------------------------------------------
1017 tryHardCheckLoop :: SDoc
1019 -> TcM ([Inst], TcDictBinds)
1021 tryHardCheckLoop doc wanteds
1022 = do { (irreds,binds) <- checkLoop (mkRedEnv doc try_me []) wanteds
1023 ; return (irreds,binds)
1026 try_me _ = ReduceMe AddSCs
1027 -- Here's the try-hard bit
1029 -----------------------------------------------------------
1030 gentleCheckLoop :: InstLoc
1033 -> TcM ([Inst], TcDictBinds)
1035 gentleCheckLoop inst_loc givens wanteds
1036 = do { (irreds,binds) <- checkLoop env wanteds
1037 ; return (irreds,binds)
1040 env = mkRedEnv (pprInstLoc inst_loc) try_me givens
1042 try_me inst | isMethodOrLit inst = ReduceMe AddSCs
1044 -- When checking against a given signature
1045 -- we MUST be very gentle: Note [Check gently]
1047 gentleInferLoop :: SDoc -> [Inst]
1048 -> TcM ([Inst], TcDictBinds)
1049 gentleInferLoop doc wanteds
1050 = do { (irreds, binds) <- checkLoop env wanteds
1051 ; return (irreds, binds) }
1053 env = mkRedEnv doc try_me []
1054 try_me inst | isMethodOrLit inst = ReduceMe AddSCs
1059 ~~~~~~~~~~~~~~~~~~~~
1060 We have to very careful about not simplifying too vigorously
1065 f :: Show b => T b -> b
1066 f (MkT x) = show [x]
1068 Inside the pattern match, which binds (a:*, x:a), we know that
1070 Hence we have a dictionary for Show [a] available; and indeed we
1071 need it. We are going to build an implication contraint
1072 forall a. (b~[a]) => Show [a]
1073 Later, we will solve this constraint using the knowledge (Show b)
1075 But we MUST NOT reduce (Show [a]) to (Show a), else the whole
1076 thing becomes insoluble. So we simplify gently (get rid of literals
1077 and methods only, plus common up equal things), deferring the real
1078 work until top level, when we solve the implication constraint
1079 with tryHardCheckLooop.
1083 -----------------------------------------------------------
1086 -> TcM ([Inst], TcDictBinds)
1087 -- Precondition: givens are completely rigid
1088 -- Postcondition: returned Insts are zonked
1090 checkLoop env wanteds
1091 = go env wanteds (return ())
1092 where go env wanteds elim_skolems
1093 = do { -- We do need to zonk the givens; cf Note [Zonking RedEnv]
1094 ; env' <- zonkRedEnv env
1095 ; wanteds' <- zonkInsts wanteds
1097 ; (improved, binds, irreds, elim_more_skolems)
1098 <- reduceContext env' wanteds'
1099 ; let elim_skolems' = elim_skolems >> elim_more_skolems
1101 ; if not improved then
1102 elim_skolems' >> return (irreds, binds)
1105 -- If improvement did some unification, we go round again.
1106 -- We start again with irreds, not wanteds
1107 -- Using an instance decl might have introduced a fresh type
1108 -- variable which might have been unified, so we'd get an
1109 -- infinite loop if we started again with wanteds!
1111 { (irreds1, binds1) <- go env' irreds elim_skolems'
1112 ; return (irreds1, binds `unionBags` binds1) } }
1115 Note [Zonking RedEnv]
1116 ~~~~~~~~~~~~~~~~~~~~~
1117 It might appear as if the givens in RedEnv are always rigid, but that is not
1118 necessarily the case for programs involving higher-rank types that have class
1119 contexts constraining the higher-rank variables. An example from tc237 in the
1122 class Modular s a | s -> a
1124 wim :: forall a w. Integral a
1125 => a -> (forall s. Modular s a => M s w) -> w
1126 wim i k = error "urk"
1128 test5 :: (Modular s a, Integral a) => M s a
1131 test4 = wim 4 test4'
1133 Notice how the variable 'a' of (Modular s a) in the rank-2 type of wim is
1134 quantified further outside. When type checking test4, we have to check
1135 whether the signature of test5 is an instance of
1137 (forall s. Modular s a => M s w)
1139 Consequently, we will get (Modular s t_a), where t_a is a TauTv into the
1142 Given the FD of Modular in this example, class improvement will instantiate
1143 t_a to 'a', where 'a' is the skolem from test5's signatures (due to the
1144 Modular s a predicate in that signature). If we don't zonk (Modular s t_a) in
1145 the givens, we will get into a loop as improveOne uses the unification engine
1146 Unify.tcUnifyTys, which doesn't know about mutable type variables.
1151 class If b t e r | b t e -> r
1154 class Lte a b c | a b -> c where lte :: a -> b -> c
1156 instance (Lte a b l,If l b a c) => Max a b c
1158 Wanted: Max Z (S x) y
1160 Then we'll reduce using the Max instance to:
1161 (Lte Z (S x) l, If l (S x) Z y)
1162 and improve by binding l->T, after which we can do some reduction
1163 on both the Lte and If constraints. What we *can't* do is start again
1164 with (Max Z (S x) y)!
1168 %************************************************************************
1170 tcSimplifySuperClasses
1172 %************************************************************************
1174 Note [SUPERCLASS-LOOP 1]
1175 ~~~~~~~~~~~~~~~~~~~~~~~~
1176 We have to be very, very careful when generating superclasses, lest we
1177 accidentally build a loop. Here's an example:
1181 class S a => C a where { opc :: a -> a }
1182 class S b => D b where { opd :: b -> b }
1184 instance C Int where
1187 instance D Int where
1190 From (instance C Int) we get the constraint set {ds1:S Int, dd:D Int}
1191 Simplifying, we may well get:
1192 $dfCInt = :C ds1 (opd dd)
1195 Notice that we spot that we can extract ds1 from dd.
1197 Alas! Alack! We can do the same for (instance D Int):
1199 $dfDInt = :D ds2 (opc dc)
1203 And now we've defined the superclass in terms of itself.
1205 Solution: never generate a superclass selectors at all when
1206 satisfying the superclass context of an instance declaration.
1208 Two more nasty cases are in
1213 tcSimplifySuperClasses
1218 tcSimplifySuperClasses loc givens sc_wanteds
1219 = do { traceTc (text "tcSimplifySuperClasses")
1220 ; (irreds,binds1) <- checkLoop env sc_wanteds
1221 ; let (tidy_env, tidy_irreds) = tidyInsts irreds
1222 ; reportNoInstances tidy_env (Just (loc, givens)) tidy_irreds
1225 env = mkRedEnv (pprInstLoc loc) try_me givens
1226 try_me _ = ReduceMe NoSCs
1227 -- Like tryHardCheckLoop, but with NoSCs
1231 %************************************************************************
1233 \subsection{tcSimplifyRestricted}
1235 %************************************************************************
1237 tcSimplifyRestricted infers which type variables to quantify for a
1238 group of restricted bindings. This isn't trivial.
1241 We want to quantify over a to get id :: forall a. a->a
1244 We do not want to quantify over a, because there's an Eq a
1245 constraint, so we get eq :: a->a->Bool (notice no forall)
1248 RHS has type 'tau', whose free tyvars are tau_tvs
1249 RHS has constraints 'wanteds'
1252 Quantify over (tau_tvs \ ftvs(wanteds))
1253 This is bad. The constraints may contain (Monad (ST s))
1254 where we have instance Monad (ST s) where...
1255 so there's no need to be monomorphic in s!
1257 Also the constraint might be a method constraint,
1258 whose type mentions a perfectly innocent tyvar:
1259 op :: Num a => a -> b -> a
1260 Here, b is unconstrained. A good example would be
1262 We want to infer the polymorphic type
1263 foo :: forall b. b -> b
1266 Plan B (cunning, used for a long time up to and including GHC 6.2)
1267 Step 1: Simplify the constraints as much as possible (to deal
1268 with Plan A's problem). Then set
1269 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1271 Step 2: Now simplify again, treating the constraint as 'free' if
1272 it does not mention qtvs, and trying to reduce it otherwise.
1273 The reasons for this is to maximise sharing.
1275 This fails for a very subtle reason. Suppose that in the Step 2
1276 a constraint (Foo (Succ Zero) (Succ Zero) b) gets thrown upstairs as 'free'.
1277 In the Step 1 this constraint might have been simplified, perhaps to
1278 (Foo Zero Zero b), AND THEN THAT MIGHT BE IMPROVED, to bind 'b' to 'T'.
1279 This won't happen in Step 2... but that in turn might prevent some other
1280 constraint (Baz [a] b) being simplified (e.g. via instance Baz [a] T where {..})
1281 and that in turn breaks the invariant that no constraints are quantified over.
1283 Test typecheck/should_compile/tc177 (which failed in GHC 6.2) demonstrates
1288 Step 1: Simplify the constraints as much as possible (to deal
1289 with Plan A's problem). Then set
1290 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1291 Return the bindings from Step 1.
1294 A note about Plan C (arising from "bug" reported by George Russel March 2004)
1297 instance (HasBinary ty IO) => HasCodedValue ty
1299 foo :: HasCodedValue a => String -> IO a
1301 doDecodeIO :: HasCodedValue a => () -> () -> IO a
1302 doDecodeIO codedValue view
1303 = let { act = foo "foo" } in act
1305 You might think this should work becuase the call to foo gives rise to a constraint
1306 (HasCodedValue t), which can be satisfied by the type sig for doDecodeIO. But the
1307 restricted binding act = ... calls tcSimplifyRestricted, and PlanC simplifies the
1308 constraint using the (rather bogus) instance declaration, and now we are stuffed.
1310 I claim this is not really a bug -- but it bit Sergey as well as George. So here's
1314 Plan D (a variant of plan B)
1315 Step 1: Simplify the constraints as much as possible (to deal
1316 with Plan A's problem), BUT DO NO IMPROVEMENT. Then set
1317 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1319 Step 2: Now simplify again, treating the constraint as 'free' if
1320 it does not mention qtvs, and trying to reduce it otherwise.
1322 The point here is that it's generally OK to have too few qtvs; that is,
1323 to make the thing more monomorphic than it could be. We don't want to
1324 do that in the common cases, but in wierd cases it's ok: the programmer
1325 can always add a signature.
1327 Too few qtvs => too many wanteds, which is what happens if you do less
1332 tcSimplifyRestricted -- Used for restricted binding groups
1333 -- i.e. ones subject to the monomorphism restriction
1336 -> [Name] -- Things bound in this group
1337 -> TcTyVarSet -- Free in the type of the RHSs
1338 -> [Inst] -- Free in the RHSs
1339 -> TcM ([TyVar], -- Tyvars to quantify (zonked and quantified)
1340 TcDictBinds) -- Bindings
1341 -- tcSimpifyRestricted returns no constraints to
1342 -- quantify over; by definition there are none.
1343 -- They are all thrown back in the LIE
1345 tcSimplifyRestricted doc top_lvl bndrs tau_tvs wanteds
1346 -- Zonk everything in sight
1347 = do { traceTc (text "tcSimplifyRestricted")
1348 ; wanteds' <- zonkInsts wanteds
1350 -- 'ReduceMe': Reduce as far as we can. Don't stop at
1351 -- dicts; the idea is to get rid of as many type
1352 -- variables as possible, and we don't want to stop
1353 -- at (say) Monad (ST s), because that reduces
1354 -- immediately, with no constraint on s.
1356 -- BUT do no improvement! See Plan D above
1357 -- HOWEVER, some unification may take place, if we instantiate
1358 -- a method Inst with an equality constraint
1359 ; let env = mkNoImproveRedEnv doc (\_ -> ReduceMe AddSCs)
1360 ; (_imp, _binds, constrained_dicts, elim_skolems)
1361 <- reduceContext env wanteds'
1364 -- Next, figure out the tyvars we will quantify over
1365 ; tau_tvs' <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
1366 ; gbl_tvs' <- tcGetGlobalTyVars
1367 ; constrained_dicts' <- zonkInsts constrained_dicts
1369 ; let qtvs1 = tau_tvs' `minusVarSet` oclose (fdPredsOfInsts constrained_dicts) gbl_tvs'
1370 -- As in tcSimplifyInfer
1372 -- Do not quantify over constrained type variables:
1373 -- this is the monomorphism restriction
1374 constrained_tvs' = tyVarsOfInsts constrained_dicts'
1375 qtvs = qtvs1 `minusVarSet` constrained_tvs'
1376 pp_bndrs = pprWithCommas (quotes . ppr) bndrs
1379 ; warn_mono <- doptM Opt_WarnMonomorphism
1380 ; warnTc (warn_mono && (constrained_tvs' `intersectsVarSet` qtvs1))
1381 (vcat[ ptext (sLit "the Monomorphism Restriction applies to the binding")
1382 <> plural bndrs <+> ptext (sLit "for") <+> pp_bndrs,
1383 ptext (sLit "Consider giving a type signature for") <+> pp_bndrs])
1385 ; traceTc (text "tcSimplifyRestricted" <+> vcat [
1386 pprInsts wanteds, pprInsts constrained_dicts',
1388 ppr constrained_tvs', ppr tau_tvs', ppr qtvs ])
1390 -- The first step may have squashed more methods than
1391 -- necessary, so try again, this time more gently, knowing the exact
1392 -- set of type variables to quantify over.
1394 -- We quantify only over constraints that are captured by qtvs;
1395 -- these will just be a subset of non-dicts. This in contrast
1396 -- to normal inference (using isFreeWhenInferring) in which we quantify over
1397 -- all *non-inheritable* constraints too. This implements choice
1398 -- (B) under "implicit parameter and monomorphism" above.
1400 -- Remember that we may need to do *some* simplification, to
1401 -- (for example) squash {Monad (ST s)} into {}. It's not enough
1402 -- just to float all constraints
1404 -- At top level, we *do* squash methods becuase we want to
1405 -- expose implicit parameters to the test that follows
1406 ; let is_nested_group = isNotTopLevel top_lvl
1407 try_me inst | isFreeWrtTyVars qtvs inst,
1408 (is_nested_group || isDict inst) = Stop
1409 | otherwise = ReduceMe AddSCs
1410 env = mkNoImproveRedEnv doc try_me
1411 ; (_imp, binds, irreds, elim_skolems) <- reduceContext env wanteds'
1414 -- See "Notes on implicit parameters, Question 4: top level"
1415 ; ASSERT( all (isFreeWrtTyVars qtvs) irreds ) -- None should be captured
1416 if is_nested_group then
1418 else do { let (bad_ips, non_ips) = partition isIPDict irreds
1419 ; addTopIPErrs bndrs bad_ips
1420 ; extendLIEs non_ips }
1422 ; qtvs' <- zonkQuantifiedTyVars (varSetElems qtvs)
1423 ; return (qtvs', binds) }
1427 %************************************************************************
1431 %************************************************************************
1433 On the LHS of transformation rules we only simplify methods and constants,
1434 getting dictionaries. We want to keep all of them unsimplified, to serve
1435 as the available stuff for the RHS of the rule.
1437 Example. Consider the following left-hand side of a rule
1439 f (x == y) (y > z) = ...
1441 If we typecheck this expression we get constraints
1443 d1 :: Ord a, d2 :: Eq a
1445 We do NOT want to "simplify" to the LHS
1447 forall x::a, y::a, z::a, d1::Ord a.
1448 f ((==) (eqFromOrd d1) x y) ((>) d1 y z) = ...
1452 forall x::a, y::a, z::a, d1::Ord a, d2::Eq a.
1453 f ((==) d2 x y) ((>) d1 y z) = ...
1455 Here is another example:
1457 fromIntegral :: (Integral a, Num b) => a -> b
1458 {-# RULES "foo" fromIntegral = id :: Int -> Int #-}
1460 In the rule, a=b=Int, and Num Int is a superclass of Integral Int. But
1461 we *dont* want to get
1463 forall dIntegralInt.
1464 fromIntegral Int Int dIntegralInt (scsel dIntegralInt) = id Int
1466 because the scsel will mess up RULE matching. Instead we want
1468 forall dIntegralInt, dNumInt.
1469 fromIntegral Int Int dIntegralInt dNumInt = id Int
1473 g (x == y) (y == z) = ..
1475 where the two dictionaries are *identical*, we do NOT WANT
1477 forall x::a, y::a, z::a, d1::Eq a
1478 f ((==) d1 x y) ((>) d1 y z) = ...
1480 because that will only match if the dict args are (visibly) equal.
1481 Instead we want to quantify over the dictionaries separately.
1483 In short, tcSimplifyRuleLhs must *only* squash LitInst and MethInts, leaving
1484 all dicts unchanged, with absolutely no sharing. It's simpler to do this
1485 from scratch, rather than further parameterise simpleReduceLoop etc
1488 tcSimplifyRuleLhs :: [Inst] -> TcM ([Inst], TcDictBinds)
1489 tcSimplifyRuleLhs wanteds
1490 = go [] emptyBag wanteds
1493 = return (dicts, binds)
1494 go dicts binds (w:ws)
1496 = go (w:dicts) binds ws
1498 = do { w' <- zonkInst w -- So that (3::Int) does not generate a call
1499 -- to fromInteger; this looks fragile to me
1500 ; lookup_result <- lookupSimpleInst w'
1501 ; case lookup_result of
1503 go dicts (addInstToDictBind binds w rhs) (ws' ++ ws)
1504 NoInstance -> pprPanic "tcSimplifyRuleLhs" (ppr w)
1508 tcSimplifyBracket is used when simplifying the constraints arising from
1509 a Template Haskell bracket [| ... |]. We want to check that there aren't
1510 any constraints that can't be satisfied (e.g. Show Foo, where Foo has no
1511 Show instance), but we aren't otherwise interested in the results.
1512 Nor do we care about ambiguous dictionaries etc. We will type check
1513 this bracket again at its usage site.
1516 tcSimplifyBracket :: [Inst] -> TcM ()
1517 tcSimplifyBracket wanteds
1518 = do { tryHardCheckLoop doc wanteds
1521 doc = text "tcSimplifyBracket"
1525 %************************************************************************
1527 \subsection{Filtering at a dynamic binding}
1529 %************************************************************************
1534 we must discharge all the ?x constraints from B. We also do an improvement
1535 step; if we have ?x::t1 and ?x::t2 we must unify t1, t2.
1537 Actually, the constraints from B might improve the types in ?x. For example
1539 f :: (?x::Int) => Char -> Char
1542 then the constraint (?x::Int) arising from the call to f will
1543 force the binding for ?x to be of type Int.
1546 tcSimplifyIPs :: [Inst] -- The implicit parameters bound here
1549 -- We need a loop so that we do improvement, and then
1550 -- (next time round) generate a binding to connect the two
1552 -- Here the two ?x's have different types, and improvement
1553 -- makes them the same.
1555 tcSimplifyIPs given_ips wanteds
1556 = do { wanteds' <- zonkInsts wanteds
1557 ; given_ips' <- zonkInsts given_ips
1558 -- Unusually for checking, we *must* zonk the given_ips
1560 ; let env = mkRedEnv doc try_me given_ips'
1561 ; (improved, binds, irreds, elim_skolems) <- reduceContext env wanteds'
1564 ; if not improved then
1565 ASSERT( all is_free irreds )
1566 do { extendLIEs irreds
1569 tcSimplifyIPs given_ips wanteds }
1571 doc = text "tcSimplifyIPs" <+> ppr given_ips
1572 ip_set = mkNameSet (ipNamesOfInsts given_ips)
1573 is_free inst = isFreeWrtIPs ip_set inst
1575 -- Simplify any methods that mention the implicit parameter
1576 try_me inst | is_free inst = Stop
1577 | otherwise = ReduceMe NoSCs
1581 %************************************************************************
1583 \subsection[binds-for-local-funs]{@bindInstsOfLocalFuns@}
1585 %************************************************************************
1587 When doing a binding group, we may have @Insts@ of local functions.
1588 For example, we might have...
1590 let f x = x + 1 -- orig local function (overloaded)
1591 f.1 = f Int -- two instances of f
1596 The point is: we must drop the bindings for @f.1@ and @f.2@ here,
1597 where @f@ is in scope; those @Insts@ must certainly not be passed
1598 upwards towards the top-level. If the @Insts@ were binding-ified up
1599 there, they would have unresolvable references to @f@.
1601 We pass in an @init_lie@ of @Insts@ and a list of locally-bound @Ids@.
1602 For each method @Inst@ in the @init_lie@ that mentions one of the
1603 @Ids@, we create a binding. We return the remaining @Insts@ (in an
1604 @LIE@), as well as the @HsBinds@ generated.
1607 bindInstsOfLocalFuns :: [Inst] -> [TcId] -> TcM TcDictBinds
1608 -- Simlifies only MethodInsts, and generate only bindings of form
1610 -- We're careful not to even generate bindings of the form
1612 -- You'd think that'd be fine, but it interacts with what is
1613 -- arguably a bug in Match.tidyEqnInfo (see notes there)
1615 bindInstsOfLocalFuns wanteds local_ids
1616 | null overloaded_ids = do
1619 return emptyLHsBinds
1622 = do { (irreds, binds) <- gentleInferLoop doc for_me
1623 ; extendLIEs not_for_me
1627 doc = text "bindInsts" <+> ppr local_ids
1628 overloaded_ids = filter is_overloaded local_ids
1629 is_overloaded id = isOverloadedTy (idType id)
1630 (for_me, not_for_me) = partition (isMethodFor overloaded_set) wanteds
1632 overloaded_set = mkVarSet overloaded_ids -- There can occasionally be a lot of them
1633 -- so it's worth building a set, so that
1634 -- lookup (in isMethodFor) is faster
1638 %************************************************************************
1640 \subsection{Data types for the reduction mechanism}
1642 %************************************************************************
1644 The main control over context reduction is here
1648 = RedEnv { red_doc :: SDoc -- The context
1649 , red_try_me :: Inst -> WhatToDo
1650 , red_improve :: Bool -- True <=> do improvement
1651 , red_givens :: [Inst] -- All guaranteed rigid
1653 -- but see Note [Rigidity]
1654 , red_stack :: (Int, [Inst]) -- Recursion stack (for err msg)
1655 -- See Note [RedStack]
1659 -- The red_givens are rigid so far as cmpInst is concerned.
1660 -- There is one case where they are not totally rigid, namely in tcSimplifyIPs
1661 -- let ?x = e in ...
1662 -- Here, the given is (?x::a), where 'a' is not necy a rigid type
1663 -- But that doesn't affect the comparison, which is based only on mame.
1666 -- The red_stack pair (n,insts) pair is just used for error reporting.
1667 -- 'n' is always the depth of the stack.
1668 -- The 'insts' is the stack of Insts being reduced: to produce X
1669 -- I had to produce Y, to produce Y I had to produce Z, and so on.
1672 mkRedEnv :: SDoc -> (Inst -> WhatToDo) -> [Inst] -> RedEnv
1673 mkRedEnv doc try_me givens
1674 = RedEnv { red_doc = doc, red_try_me = try_me,
1675 red_givens = givens,
1677 red_improve = True }
1679 mkNoImproveRedEnv :: SDoc -> (Inst -> WhatToDo) -> RedEnv
1680 -- Do not do improvement; no givens
1681 mkNoImproveRedEnv doc try_me
1682 = RedEnv { red_doc = doc, red_try_me = try_me,
1685 red_improve = True }
1688 = ReduceMe WantSCs -- Try to reduce this
1689 -- If there's no instance, add the inst to the
1690 -- irreductible ones, but don't produce an error
1691 -- message of any kind.
1692 -- It might be quite legitimate such as (Eq a)!
1694 | Stop -- Return as irreducible unless it can
1695 -- be reduced to a constant in one step
1696 -- Do not add superclasses; see
1698 data WantSCs = NoSCs | AddSCs -- Tells whether we should add the superclasses
1699 -- of a predicate when adding it to the avails
1700 -- The reason for this flag is entirely the super-class loop problem
1701 -- Note [SUPER-CLASS LOOP 1]
1703 zonkRedEnv :: RedEnv -> TcM RedEnv
1705 = do { givens' <- mapM zonkInst (red_givens env)
1706 ; return $ env {red_givens = givens'}
1711 %************************************************************************
1713 \subsection[reduce]{@reduce@}
1715 %************************************************************************
1717 Note [Ancestor Equalities]
1718 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1719 During context reduction, we add to the wanted equalities also those
1720 equalities that (transitively) occur in superclass contexts of wanted
1721 class constraints. Consider the following code
1723 class a ~ Int => C a
1726 If (C a) is wanted, we want to add (a ~ Int), which will be discharged by
1727 substituting Int for a. Hence, we ultimately want (C Int), which we
1728 discharge with the explicit instance.
1731 reduceContext :: RedEnv
1733 -> TcM (ImprovementDone,
1734 TcDictBinds, -- Dictionary bindings
1735 [Inst], -- Irreducible
1736 TcM ()) -- Undo skolems from SkolemOccurs
1738 reduceContext env wanteds
1739 = do { traceTc (text "reduceContext" <+> (vcat [
1740 text "----------------------",
1742 text "given" <+> ppr (red_givens env),
1743 text "wanted" <+> ppr wanteds,
1744 text "----------------------"
1748 ; let givens = red_givens env
1749 (given_eqs0, given_dicts0) = partition isEqInst givens
1750 (wanted_eqs0, wanted_non_eqs) = partition isEqInst wanteds
1751 (wanted_implics0, wanted_dicts) = partition isImplicInst wanted_non_eqs
1753 -- We want to add as wanted equalities those that (transitively)
1754 -- occur in superclass contexts of wanted class constraints.
1755 -- See Note [Ancestor Equalities]
1756 ; ancestor_eqs <- ancestorEqualities wanted_dicts
1757 ; let wanted_eqs = wanted_eqs0 ++ ancestor_eqs
1758 ; traceTc $ text "reduceContext: ancestor eqs" <+> ppr ancestor_eqs
1760 -- 1. Normalise the *given* *equality* constraints
1761 ; (given_eqs, eliminate_skolems) <- normaliseGivenEqs given_eqs0
1763 -- 2. Normalise the *given* *dictionary* constraints
1764 -- wrt. the toplevel and given equations
1765 ; (given_dicts, given_binds) <- normaliseGivenDicts given_eqs
1768 -- 5. Build the Avail mapping from "given_dicts"
1769 ; (init_state, _) <- getLIE $ do
1770 { init_state <- foldlM addGiven emptyAvails given_dicts
1774 -- *** ToDo: what to do with the "extra_givens"? For the
1775 -- moment I'm simply discarding them, which is probably wrong
1777 -- 6. Solve the *wanted* *dictionary* constraints (not implications)
1778 -- This may expose some further equational constraints...
1779 ; (avails, extra_eqs) <- getLIE (reduceList env wanted_dicts init_state)
1780 ; (dict_binds, bound_dicts, dict_irreds)
1781 <- extractResults avails wanted_dicts
1782 ; traceTc $ text "reduceContext extractresults" <+> vcat
1783 [ppr avails, ppr wanted_dicts, ppr dict_binds]
1785 -- Solve the wanted *implications*. In doing so, we can provide
1786 -- as "given" all the dicts that were originally given,
1787 -- *or* for which we now have bindings,
1788 -- *or* which are now irreds
1789 ; let implic_env = env { red_givens = givens ++ bound_dicts
1791 ; (implic_binds_s, implic_irreds_s)
1792 <- mapAndUnzipM (reduceImplication implic_env) wanted_implics0
1793 ; let implic_binds = unionManyBags implic_binds_s
1794 implic_irreds = concat implic_irreds_s
1796 -- Normalise the wanted equality constraints
1797 ; eq_irreds <- normaliseWantedEqs given_eqs (wanted_eqs ++ extra_eqs)
1799 -- Normalise the wanted dictionaries
1800 ; let irreds = dict_irreds ++ implic_irreds
1801 eqs = eq_irreds ++ given_eqs
1802 ; (norm_irreds, normalise_binds) <- normaliseWantedDicts eqs irreds
1804 -- Figure out whether we should go round again. We do so in either
1806 -- (1) If any of the mutable tyvars in givens or irreds has been
1807 -- filled in by improvement, there is merit in going around
1808 -- again, because we may make further progress.
1809 -- (2) If we managed to normalise any dicts, there is merit in going
1810 -- around gain, because reduceList may be able to get further.
1812 -- ToDo: We may have exposed new
1813 -- equality constraints and should probably go round again
1814 -- then as well. But currently we are dropping them on the
1817 ; let all_irreds = norm_irreds ++ eq_irreds
1818 ; improvedMetaTy <- anyM isFilledMetaTyVar $ varSetElems $
1819 tyVarsOfInsts (givens ++ all_irreds)
1820 ; let improvedDicts = not $ isEmptyBag normalise_binds
1821 improved = improvedMetaTy || improvedDicts
1823 -- The old plan (fragile)
1824 -- improveed = availsImproved avails
1825 -- || (not $ isEmptyBag normalise_binds1)
1826 -- || (not $ isEmptyBag normalise_binds2)
1827 -- || (any isEqInst irreds)
1829 ; traceTc (text "reduceContext end" <+> (vcat [
1830 text "----------------------",
1832 text "given" <+> ppr givens,
1833 text "given_eqs" <+> ppr given_eqs,
1834 text "wanted" <+> ppr wanteds,
1835 text "wanted_dicts" <+> ppr wanted_dicts,
1837 text "avails" <+> pprAvails avails,
1838 text "improved =" <+> ppr improved,
1839 text "(all) irreds = " <+> ppr all_irreds,
1840 text "dict-binds = " <+> ppr dict_binds,
1841 text "implic-binds = " <+> ppr implic_binds,
1842 text "----------------------"
1846 given_binds `unionBags` normalise_binds
1847 `unionBags` dict_binds
1848 `unionBags` implic_binds,
1853 tcImproveOne :: Avails -> Inst -> TcM ImprovementDone
1854 tcImproveOne avails inst
1855 | not (isDict inst) = return False
1857 = do { inst_envs <- tcGetInstEnvs
1858 ; let eqns = improveOne (classInstances inst_envs)
1859 (dictPred inst, pprInstArising inst)
1860 [ (dictPred p, pprInstArising p)
1861 | p <- availsInsts avails, isDict p ]
1862 -- Avails has all the superclasses etc (good)
1863 -- It also has all the intermediates of the deduction (good)
1864 -- It does not have duplicates (good)
1865 -- NB that (?x::t1) and (?x::t2) will be held separately in avails
1866 -- so that improve will see them separate
1867 ; traceTc (text "improveOne" <+> ppr inst)
1870 unifyEqns :: [(Equation,(PredType,SDoc),(PredType,SDoc))]
1871 -> TcM ImprovementDone
1872 unifyEqns [] = return False
1874 = do { traceTc (ptext (sLit "Improve:") <+> vcat (map pprEquationDoc eqns))
1878 unify ((qtvs, pairs), what1, what2)
1879 = addErrCtxtM (mkEqnMsg what1 what2) $ do
1880 (_, _, tenv) <- tcInstTyVars (varSetElems qtvs)
1881 mapM_ (unif_pr tenv) pairs
1882 unif_pr tenv (ty1,ty2) = unifyType (substTy tenv ty1) (substTy tenv ty2)
1884 pprEquationDoc :: (Equation, (PredType, SDoc), (PredType, SDoc)) -> SDoc
1885 pprEquationDoc (eqn, (p1, _), (p2, _)) = vcat [pprEquation eqn, nest 2 (ppr p1), nest 2 (ppr p2)]
1887 mkEqnMsg :: (TcPredType, SDoc) -> (TcPredType, SDoc) -> TidyEnv
1888 -> TcM (TidyEnv, SDoc)
1889 mkEqnMsg (pred1,from1) (pred2,from2) tidy_env
1890 = do { pred1' <- zonkTcPredType pred1; pred2' <- zonkTcPredType pred2
1891 ; let { pred1'' = tidyPred tidy_env pred1'; pred2'' = tidyPred tidy_env pred2' }
1892 ; let msg = vcat [ptext (sLit "When using functional dependencies to combine"),
1893 nest 2 (sep [ppr pred1'' <> comma, nest 2 from1]),
1894 nest 2 (sep [ppr pred2'' <> comma, nest 2 from2])]
1895 ; return (tidy_env, msg) }
1898 The main context-reduction function is @reduce@. Here's its game plan.
1901 reduceList :: RedEnv -> [Inst] -> Avails -> TcM Avails
1902 reduceList env@(RedEnv {red_stack = (n,stk)}) wanteds state
1903 = do { traceTc (text "reduceList " <+> (ppr wanteds $$ ppr state))
1905 ; when (debugIsOn && (n > 8)) $ do
1906 debugDumpTcRn (hang (ptext (sLit "Interesting! Context reduction stack depth") <+> int n)
1907 2 (ifPprDebug (nest 2 (pprStack stk))))
1908 ; if n >= ctxtStkDepth dopts then
1909 failWithTc (reduceDepthErr n stk)
1913 go [] state = return state
1914 go (w:ws) state = do { state' <- reduce (env {red_stack = (n+1, w:stk)}) w state
1917 -- Base case: we're done!
1918 reduce :: RedEnv -> Inst -> Avails -> TcM Avails
1919 reduce env wanted avails
1920 -- It's the same as an existing inst, or a superclass thereof
1921 | Just _ <- findAvail avails wanted
1922 = do { traceTc (text "reduce: found " <+> ppr wanted)
1927 = do { traceTc (text "reduce" <+> ppr wanted $$ ppr avails)
1928 ; case red_try_me env wanted of {
1929 Stop -> try_simple (addIrred NoSCs);
1930 -- See Note [No superclasses for Stop]
1932 ReduceMe want_scs -> do -- It should be reduced
1933 { (avails, lookup_result) <- reduceInst env avails wanted
1934 ; case lookup_result of
1935 NoInstance -> addIrred want_scs avails wanted
1936 -- Add it and its superclasses
1938 GenInst [] rhs -> addWanted want_scs avails wanted rhs []
1940 GenInst wanteds' rhs
1941 -> do { avails1 <- addIrred NoSCs avails wanted
1942 ; avails2 <- reduceList env wanteds' avails1
1943 ; addWanted want_scs avails2 wanted rhs wanteds' } }
1944 -- Temporarily do addIrred *before* the reduceList,
1945 -- which has the effect of adding the thing we are trying
1946 -- to prove to the database before trying to prove the things it
1947 -- needs. See note [RECURSIVE DICTIONARIES]
1948 -- NB: we must not do an addWanted before, because that adds the
1949 -- superclasses too, and that can lead to a spurious loop; see
1950 -- the examples in [SUPERCLASS-LOOP]
1951 -- So we do an addIrred before, and then overwrite it afterwards with addWanted
1954 -- First, see if the inst can be reduced to a constant in one step
1955 -- Works well for literals (1::Int) and constant dictionaries (d::Num Int)
1956 -- Don't bother for implication constraints, which take real work
1957 try_simple do_this_otherwise
1958 = do { res <- lookupSimpleInst wanted
1960 GenInst [] rhs -> addWanted AddSCs avails wanted rhs []
1961 _ -> do_this_otherwise avails wanted }
1965 Note [SUPERCLASS-LOOP 2]
1966 ~~~~~~~~~~~~~~~~~~~~~~~~
1967 But the above isn't enough. Suppose we are *given* d1:Ord a,
1968 and want to deduce (d2:C [a]) where
1970 class Ord a => C a where
1971 instance Ord [a] => C [a] where ...
1973 Then we'll use the instance decl to deduce C [a] from Ord [a], and then add the
1974 superclasses of C [a] to avails. But we must not overwrite the binding
1975 for Ord [a] (which is obtained from Ord a) with a superclass selection or we'll just
1978 Here's another variant, immortalised in tcrun020
1979 class Monad m => C1 m
1980 class C1 m => C2 m x
1981 instance C2 Maybe Bool
1982 For the instance decl we need to build (C1 Maybe), and it's no good if
1983 we run around and add (C2 Maybe Bool) and its superclasses to the avails
1984 before we search for C1 Maybe.
1986 Here's another example
1987 class Eq b => Foo a b
1988 instance Eq a => Foo [a] a
1992 we'll first deduce that it holds (via the instance decl). We must not
1993 then overwrite the Eq t constraint with a superclass selection!
1995 At first I had a gross hack, whereby I simply did not add superclass constraints
1996 in addWanted, though I did for addGiven and addIrred. This was sub-optimal,
1997 becuase it lost legitimate superclass sharing, and it still didn't do the job:
1998 I found a very obscure program (now tcrun021) in which improvement meant the
1999 simplifier got two bites a the cherry... so something seemed to be an Stop
2000 first time, but reducible next time.
2002 Now we implement the Right Solution, which is to check for loops directly
2003 when adding superclasses. It's a bit like the occurs check in unification.
2006 Note [RECURSIVE DICTIONARIES]
2007 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2009 data D r = ZeroD | SuccD (r (D r));
2011 instance (Eq (r (D r))) => Eq (D r) where
2012 ZeroD == ZeroD = True
2013 (SuccD a) == (SuccD b) = a == b
2016 equalDC :: D [] -> D [] -> Bool;
2019 We need to prove (Eq (D [])). Here's how we go:
2023 by instance decl, holds if
2027 by instance decl of Eq, holds if
2029 where d2 = dfEqList d3
2032 But now we can "tie the knot" to give
2038 and it'll even run! The trick is to put the thing we are trying to prove
2039 (in this case Eq (D []) into the database before trying to prove its
2040 contributing clauses.
2043 %************************************************************************
2045 Reducing a single constraint
2047 %************************************************************************
2050 ---------------------------------------------
2051 reduceInst :: RedEnv -> Avails -> Inst -> TcM (Avails, LookupInstResult)
2052 reduceInst _ avails other_inst
2053 = do { result <- lookupSimpleInst other_inst
2054 ; return (avails, result) }
2057 Note [Equational Constraints in Implication Constraints]
2058 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2060 An implication constraint is of the form
2062 where Given and Wanted may contain both equational and dictionary
2063 constraints. The delay and reduction of these two kinds of constraints
2066 -) In the generated code, wanted Dictionary constraints are wrapped up in an
2067 implication constraint that is created at the code site where the wanted
2068 dictionaries can be reduced via a let-binding. This let-bound implication
2069 constraint is deconstructed at the use-site of the wanted dictionaries.
2071 -) While the reduction of equational constraints is also delayed, the delay
2072 is not manifest in the generated code. The required evidence is generated
2073 in the code directly at the use-site. There is no let-binding and deconstruction
2074 necessary. The main disadvantage is that we cannot exploit sharing as the
2075 same evidence may be generated at multiple use-sites. However, this disadvantage
2076 is limited because it only concerns coercions which are erased.
2078 The different treatment is motivated by the different in representation. Dictionary
2079 constraints require manifest runtime dictionaries, while equations require coercions
2083 ---------------------------------------------
2084 reduceImplication :: RedEnv
2086 -> TcM (TcDictBinds, [Inst])
2089 Suppose we are simplifying the constraint
2090 forall bs. extras => wanted
2091 in the context of an overall simplification problem with givens 'givens'.
2094 * The 'givens' need not mention any of the quantified type variables
2095 e.g. forall {}. Eq a => Eq [a]
2096 forall {}. C Int => D (Tree Int)
2098 This happens when you have something like
2100 T1 :: Eq a => a -> T a
2103 f x = ...(case x of { T1 v -> v==v })...
2106 -- ToDo: should we instantiate tvs? I think it's not necessary
2108 -- Note on coercion variables:
2110 -- The extra given coercion variables are bound at two different sites:
2111 -- -) in the creation context of the implication constraint
2112 -- the solved equational constraints use these binders
2114 -- -) at the solving site of the implication constraint
2115 -- the solved dictionaries use these binders
2116 -- these binders are generated by reduceImplication
2118 reduceImplication env
2119 orig_implic@(ImplicInst { tci_name = name, tci_loc = inst_loc,
2121 tci_given = extra_givens, tci_wanted = wanteds })
2122 = do { -- Solve the sub-problem
2123 ; let try_me _ = ReduceMe AddSCs -- Note [Freeness and implications]
2124 env' = env { red_givens = extra_givens ++ red_givens env
2125 , red_doc = sep [ptext (sLit "reduceImplication for")
2127 nest 2 (parens $ ptext (sLit "within")
2129 , red_try_me = try_me }
2131 ; traceTc (text "reduceImplication" <+> vcat
2132 [ ppr (red_givens env), ppr extra_givens,
2134 ; (irreds, binds) <- checkLoop env' wanteds
2135 ; let (extra_eq_givens, extra_dict_givens) = partition isEqInst extra_givens
2136 -- SLPJ Sept 07: I think this is bogus; currently
2137 -- there are no Eqinsts in extra_givens
2138 dict_ids = map instToId extra_dict_givens
2140 -- Note [Reducing implication constraints]
2141 -- Tom -- update note, put somewhere!
2143 ; traceTc (text "reduceImplication result" <+> vcat
2144 [ppr irreds, ppr binds])
2146 ; -- extract superclass binds
2147 -- (sc_binds,_) <- extractResults avails []
2148 -- ; traceTc (text "reduceImplication sc_binds" <+> vcat
2149 -- [ppr sc_binds, ppr avails])
2152 -- SLPJ Sept 07: what if improvement happened inside the checkLoop?
2153 -- Then we must iterate the outer loop too!
2155 ; traceTc (text "reduceImplication condition" <+> ppr ((isEmptyLHsBinds binds) || (null irreds)))
2157 -- Progress is no longer measered by the number of bindings
2158 ; if (isEmptyLHsBinds binds) && (not $ null irreds) then -- No progress
2159 -- If there are any irreds, we back off and do nothing
2160 return (emptyBag, [orig_implic])
2162 { (simpler_implic_insts, bind)
2163 <- makeImplicationBind inst_loc tvs extra_givens irreds
2164 -- This binding is useless if the recursive simplification
2165 -- made no progress; but currently we don't try to optimise that
2166 -- case. After all, we only try hard to reduce at top level, or
2167 -- when inferring types.
2169 ; let dict_wanteds = filter (not . isEqInst) wanteds
2170 -- TOMDO: given equational constraints bug!
2171 -- we need a different evidence for given
2172 -- equations depending on whether we solve
2173 -- dictionary constraints or equational constraints
2175 eq_tyvars = varSetElems $ tyVarsOfTypes $ map eqInstType extra_eq_givens
2176 -- SLPJ Sept07: this looks Utterly Wrong to me, but I think
2177 -- that current extra_givens has no EqInsts, so
2178 -- it makes no difference
2179 co = wrap_inline -- Note [Always inline implication constraints]
2181 <.> mkWpLams eq_tyvars
2182 <.> mkWpLams dict_ids
2183 <.> WpLet (binds `unionBags` bind)
2184 wrap_inline | null dict_ids = idHsWrapper
2185 | otherwise = WpInline
2186 rhs = mkLHsWrap co payload
2187 loc = instLocSpan inst_loc
2188 payload = mkBigLHsTup (map (L loc . HsVar . instToId) dict_wanteds)
2191 ; traceTc (vcat [text "reduceImplication" <+> ppr name,
2192 ppr simpler_implic_insts,
2193 text "->" <+> ppr rhs])
2194 ; return (unitBag (L loc (VarBind (instToId orig_implic) rhs)),
2195 simpler_implic_insts)
2198 reduceImplication _ i = pprPanic "reduceImplication" (ppr i)
2201 Note [Always inline implication constraints]
2202 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2203 Suppose an implication constraint floats out of an INLINE function.
2204 Then although the implication has a single call site, it won't be
2205 inlined. And that is bad because it means that even if there is really
2206 *no* overloading (type signatures specify the exact types) there will
2207 still be dictionary passing in the resulting code. To avert this,
2208 we mark the implication constraints themselves as INLINE, at least when
2209 there is no loss of sharing as a result.
2211 Note [Freeness and implications]
2212 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2213 It's hard to say when an implication constraint can be floated out. Consider
2214 forall {} Eq a => Foo [a]
2215 The (Foo [a]) doesn't mention any of the quantified variables, but it
2216 still might be partially satisfied by the (Eq a).
2218 There is a useful special case when it *is* easy to partition the
2219 constraints, namely when there are no 'givens'. Consider
2220 forall {a}. () => Bar b
2221 There are no 'givens', and so there is no reason to capture (Bar b).
2222 We can let it float out. But if there is even one constraint we
2223 must be much more careful:
2224 forall {a}. C a b => Bar (m b)
2225 because (C a b) might have a superclass (D b), from which we might
2226 deduce (Bar [b]) when m later gets instantiated to []. Ha!
2228 Here is an even more exotic example
2230 Now consider the constraint
2231 forall b. D Int b => C Int
2232 We can satisfy the (C Int) from the superclass of D, so we don't want
2233 to float the (C Int) out, even though it mentions no type variable in
2236 One more example: the constraint
2238 instance (C a, E c) => E (a,c)
2240 constraint: forall b. D Int b => E (Int,c)
2242 You might think that the (D Int b) can't possibly contribute
2243 to solving (E (Int,c)), since the latter mentions 'c'. But
2244 in fact it can, because solving the (E (Int,c)) constraint needs
2247 and the (C Int) can be satisfied from the superclass of (D Int b).
2248 So we must still not float (E (Int,c)) out.
2250 To think about: special cases for unary type classes?
2252 Note [Pruning the givens in an implication constraint]
2253 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2254 Suppose we are about to form the implication constraint
2255 forall tvs. Eq a => Ord b
2256 The (Eq a) cannot contribute to the (Ord b), because it has no access to
2257 the type variable 'b'. So we could filter out the (Eq a) from the givens.
2258 But BE CAREFUL of the examples above in [Freeness and implications].
2260 Doing so would be a bit tidier, but all the implication constraints get
2261 simplified away by the optimiser, so it's no great win. So I don't take
2262 advantage of that at the moment.
2264 If you do, BE CAREFUL of wobbly type variables.
2267 %************************************************************************
2269 Avails and AvailHow: the pool of evidence
2271 %************************************************************************
2275 data Avails = Avails !ImprovementDone !AvailEnv
2277 type ImprovementDone = Bool -- True <=> some unification has happened
2278 -- so some Irreds might now be reducible
2279 -- keys that are now
2281 type AvailEnv = FiniteMap Inst AvailHow
2283 = IsIrred -- Used for irreducible dictionaries,
2284 -- which are going to be lambda bound
2286 | Given Inst -- Used for dictionaries for which we have a binding
2287 -- e.g. those "given" in a signature
2289 | Rhs -- Used when there is a RHS
2290 (LHsExpr TcId) -- The RHS
2291 [Inst] -- Insts free in the RHS; we need these too
2293 instance Outputable Avails where
2296 pprAvails :: Avails -> SDoc
2297 pprAvails (Avails imp avails)
2298 = vcat [ ptext (sLit "Avails") <> (if imp then ptext (sLit "[improved]") else empty)
2300 vcat [ sep [ppr inst, nest 2 (equals <+> ppr avail)]
2301 | (inst,avail) <- fmToList avails ]]
2303 instance Outputable AvailHow where
2306 -------------------------
2307 pprAvail :: AvailHow -> SDoc
2308 pprAvail IsIrred = text "Irred"
2309 pprAvail (Given x) = text "Given" <+> ppr x
2310 pprAvail (Rhs rhs bs) = sep [text "Rhs" <+> ppr bs,
2313 -------------------------
2314 extendAvailEnv :: AvailEnv -> Inst -> AvailHow -> AvailEnv
2315 extendAvailEnv env inst avail = addToFM env inst avail
2317 findAvailEnv :: AvailEnv -> Inst -> Maybe AvailHow
2318 findAvailEnv env wanted = lookupFM env wanted
2319 -- NB 1: the Ord instance of Inst compares by the class/type info
2320 -- *not* by unique. So
2321 -- d1::C Int == d2::C Int
2323 emptyAvails :: Avails
2324 emptyAvails = Avails False emptyFM
2326 findAvail :: Avails -> Inst -> Maybe AvailHow
2327 findAvail (Avails _ avails) wanted = findAvailEnv avails wanted
2329 elemAvails :: Inst -> Avails -> Bool
2330 elemAvails wanted (Avails _ avails) = wanted `elemFM` avails
2332 extendAvails :: Avails -> Inst -> AvailHow -> TcM Avails
2334 extendAvails avails@(Avails imp env) inst avail
2335 = do { imp1 <- tcImproveOne avails inst -- Do any improvement
2336 ; return (Avails (imp || imp1) (extendAvailEnv env inst avail)) }
2338 availsInsts :: Avails -> [Inst]
2339 availsInsts (Avails _ avails) = keysFM avails
2341 _availsImproved :: Avails -> ImprovementDone
2342 _availsImproved (Avails imp _) = imp
2345 Extracting the bindings from a bunch of Avails.
2346 The bindings do *not* come back sorted in dependency order.
2347 We assume that they'll be wrapped in a big Rec, so that the
2348 dependency analyser can sort them out later
2351 type DoneEnv = FiniteMap Inst [Id]
2352 -- Tracks which things we have evidence for
2354 extractResults :: Avails
2356 -> TcM (TcDictBinds, -- Bindings
2357 [Inst], -- The insts bound by the bindings
2358 [Inst]) -- Irreducible ones
2359 -- Note [Reducing implication constraints]
2361 extractResults (Avails _ avails) wanteds
2362 = go emptyBag [] [] emptyFM wanteds
2364 go :: TcDictBinds -- Bindings for dicts
2365 -> [Inst] -- Bound by the bindings
2367 -> DoneEnv -- Has an entry for each inst in the above three sets
2369 -> TcM (TcDictBinds, [Inst], [Inst])
2370 go binds bound_dicts irreds _ []
2371 = return (binds, bound_dicts, irreds)
2373 go binds bound_dicts irreds done (w:ws)
2374 | Just done_ids@(done_id : rest_done_ids) <- lookupFM done w
2375 = if w_id `elem` done_ids then
2376 go binds bound_dicts irreds done ws
2378 go (add_bind (nlHsVar done_id)) bound_dicts irreds
2379 (addToFM done w (done_id : w_id : rest_done_ids)) ws
2381 | otherwise -- Not yet done
2382 = case findAvailEnv avails w of
2383 Nothing -> pprTrace "Urk: extractResults" (ppr w) $
2384 go binds bound_dicts irreds done ws
2386 Just IsIrred -> go binds bound_dicts (w:irreds) done' ws
2388 Just (Rhs rhs ws') -> go (add_bind rhs) (w:bound_dicts) irreds done' (ws' ++ ws)
2390 Just (Given g) -> go binds' bound_dicts irreds (addToFM done w [g_id]) ws
2393 binds' | w_id == g_id = binds
2394 | otherwise = add_bind (nlHsVar g_id)
2397 done' = addToFM done w [w_id]
2398 add_bind rhs = addInstToDictBind binds w rhs
2402 Note [No superclasses for Stop]
2403 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2404 When we decide not to reduce an Inst -- the 'WhatToDo' --- we still
2405 add it to avails, so that any other equal Insts will be commoned up
2406 right here. However, we do *not* add superclasses. If we have
2409 but a is not bound here, then we *don't* want to derive dn from df
2410 here lest we lose sharing.
2413 addWanted :: WantSCs -> Avails -> Inst -> LHsExpr TcId -> [Inst] -> TcM Avails
2414 addWanted want_scs avails wanted rhs_expr wanteds
2415 = addAvailAndSCs want_scs avails wanted avail
2417 avail = Rhs rhs_expr wanteds
2419 addGiven :: Avails -> Inst -> TcM Avails
2420 addGiven avails given = addAvailAndSCs AddSCs avails given (Given given)
2421 -- Always add superclasses for 'givens'
2423 -- No ASSERT( not (given `elemAvails` avails) ) because in an instance
2424 -- decl for Ord t we can add both Ord t and Eq t as 'givens',
2425 -- so the assert isn't true
2429 addIrred :: WantSCs -> Avails -> Inst -> TcM Avails
2430 addIrred want_scs avails irred = ASSERT2( not (irred `elemAvails` avails), ppr irred $$ ppr avails )
2431 addAvailAndSCs want_scs avails irred IsIrred
2433 addAvailAndSCs :: WantSCs -> Avails -> Inst -> AvailHow -> TcM Avails
2434 addAvailAndSCs want_scs avails inst avail
2435 | not (isClassDict inst) = extendAvails avails inst avail
2436 | NoSCs <- want_scs = extendAvails avails inst avail
2437 | otherwise = do { traceTc (text "addAvailAndSCs" <+> vcat [ppr inst, ppr deps])
2438 ; avails' <- extendAvails avails inst avail
2439 ; addSCs is_loop avails' inst }
2441 is_loop pred = any (`tcEqType` mkPredTy pred) dep_tys
2442 -- Note: this compares by *type*, not by Unique
2443 deps = findAllDeps (unitVarSet (instToVar inst)) avail
2444 dep_tys = map idType (varSetElems deps)
2446 findAllDeps :: IdSet -> AvailHow -> IdSet
2447 -- Find all the Insts that this one depends on
2448 -- See Note [SUPERCLASS-LOOP 2]
2449 -- Watch out, though. Since the avails may contain loops
2450 -- (see Note [RECURSIVE DICTIONARIES]), so we need to track the ones we've seen so far
2451 findAllDeps so_far (Rhs _ kids) = foldl find_all so_far kids
2452 findAllDeps so_far _ = so_far
2454 find_all :: IdSet -> Inst -> IdSet
2456 | isEqInst kid = so_far
2457 | kid_id `elemVarSet` so_far = so_far
2458 | Just avail <- findAvail avails kid = findAllDeps so_far' avail
2459 | otherwise = so_far'
2461 so_far' = extendVarSet so_far kid_id -- Add the new kid to so_far
2462 kid_id = instToId kid
2464 addSCs :: (TcPredType -> Bool) -> Avails -> Inst -> TcM Avails
2465 -- Add all the superclasses of the Inst to Avails
2466 -- The first param says "don't do this because the original thing
2467 -- depends on this one, so you'd build a loop"
2468 -- Invariant: the Inst is already in Avails.
2470 addSCs is_loop avails dict
2471 = ASSERT( isDict dict )
2472 do { sc_dicts <- newDictBndrs (instLoc dict) sc_theta'
2473 ; foldlM add_sc avails (zipEqual "add_scs" sc_dicts sc_sels) }
2475 (clas, tys) = getDictClassTys dict
2476 (tyvars, sc_theta, sc_sels, _) = classBigSig clas
2477 sc_theta' = filter (not . isEqPred) $
2478 substTheta (zipTopTvSubst tyvars tys) sc_theta
2480 add_sc avails (sc_dict, sc_sel)
2481 | is_loop (dictPred sc_dict) = return avails -- See Note [SUPERCLASS-LOOP 2]
2482 | is_given sc_dict = return avails
2483 | otherwise = do { avails' <- extendAvails avails sc_dict (Rhs sc_sel_rhs [dict])
2484 ; addSCs is_loop avails' sc_dict }
2486 sc_sel_rhs = L (instSpan dict) (HsWrap co_fn (HsVar sc_sel))
2487 co_fn = WpApp (instToVar dict) <.> mkWpTyApps tys
2489 is_given :: Inst -> Bool
2490 is_given sc_dict = case findAvail avails sc_dict of
2491 Just (Given _) -> True -- Given is cheaper than superclass selection
2494 -- From the a set of insts obtain all equalities that (transitively) occur in
2495 -- superclass contexts of class constraints (aka the ancestor equalities).
2497 ancestorEqualities :: [Inst] -> TcM [Inst]
2499 = mapM mkWantedEqInst -- turn only equality predicates..
2500 . filter isEqPred -- ..into wanted equality insts
2502 . addAEsToBag emptyBag -- collect the superclass constraints..
2503 . map dictPred -- ..of all predicates in a bag
2504 . filter isClassDict
2506 addAEsToBag :: Bag PredType -> [PredType] -> Bag PredType
2507 addAEsToBag bag [] = bag
2508 addAEsToBag bag (pred:preds)
2509 | pred `elemBag` bag = addAEsToBag bag preds
2510 | isEqPred pred = addAEsToBag bagWithPred preds
2511 | isClassPred pred = addAEsToBag bagWithPred predsWithSCs
2512 | otherwise = addAEsToBag bag preds
2514 bagWithPred = bag `snocBag` pred
2515 predsWithSCs = preds ++ substTheta (zipTopTvSubst tyvars tys) sc_theta
2517 (tyvars, sc_theta, _, _) = classBigSig clas
2518 (clas, tys) = getClassPredTys pred
2522 %************************************************************************
2524 \section{tcSimplifyTop: defaulting}
2526 %************************************************************************
2529 @tcSimplifyTop@ is called once per module to simplify all the constant
2530 and ambiguous Insts.
2532 We need to be careful of one case. Suppose we have
2534 instance Num a => Num (Foo a b) where ...
2536 and @tcSimplifyTop@ is given a constraint (Num (Foo x y)). Then it'll simplify
2537 to (Num x), and default x to Int. But what about y??
2539 It's OK: the final zonking stage should zap y to (), which is fine.
2543 tcSimplifyTop, tcSimplifyInteractive :: [Inst] -> TcM TcDictBinds
2544 tcSimplifyTop wanteds
2545 = tc_simplify_top doc False wanteds
2547 doc = text "tcSimplifyTop"
2549 tcSimplifyInteractive wanteds
2550 = tc_simplify_top doc True wanteds
2552 doc = text "tcSimplifyInteractive"
2554 -- The TcLclEnv should be valid here, solely to improve
2555 -- error message generation for the monomorphism restriction
2556 tc_simplify_top :: SDoc -> Bool -> [Inst] -> TcM (Bag (LHsBind TcId))
2557 tc_simplify_top doc interactive wanteds
2558 = do { dflags <- getDOpts
2559 ; wanteds <- zonkInsts wanteds
2560 ; mapM_ zonkTopTyVar (varSetElems (tyVarsOfInsts wanteds))
2562 ; traceTc (text "tc_simplify_top 0: " <+> ppr wanteds)
2563 ; (irreds1, binds1) <- tryHardCheckLoop doc1 wanteds
2564 -- ; (irreds1, binds1) <- gentleInferLoop doc1 wanteds
2565 ; traceTc (text "tc_simplify_top 1: " <+> ppr irreds1)
2566 ; (irreds2, binds2) <- approximateImplications doc2 (\_ -> True) irreds1
2567 ; traceTc (text "tc_simplify_top 2: " <+> ppr irreds2)
2569 -- Use the defaulting rules to do extra unification
2570 -- NB: irreds2 are already zonked
2571 ; (irreds3, binds3) <- disambiguate doc3 interactive dflags irreds2
2573 -- Deal with implicit parameters
2574 ; let (bad_ips, non_ips) = partition isIPDict irreds3
2575 (ambigs, others) = partition isTyVarDict non_ips
2577 ; topIPErrs bad_ips -- Can arise from f :: Int -> Int
2579 ; addNoInstanceErrs others
2580 ; addTopAmbigErrs ambigs
2582 ; return (binds1 `unionBags` binds2 `unionBags` binds3) }
2584 doc1 = doc <+> ptext (sLit "(first round)")
2585 doc2 = doc <+> ptext (sLit "(approximate)")
2586 doc3 = doc <+> ptext (sLit "(disambiguate)")
2589 If a dictionary constrains a type variable which is
2590 * not mentioned in the environment
2591 * and not mentioned in the type of the expression
2592 then it is ambiguous. No further information will arise to instantiate
2593 the type variable; nor will it be generalised and turned into an extra
2594 parameter to a function.
2596 It is an error for this to occur, except that Haskell provided for
2597 certain rules to be applied in the special case of numeric types.
2599 * at least one of its classes is a numeric class, and
2600 * all of its classes are numeric or standard
2601 then the type variable can be defaulted to the first type in the
2602 default-type list which is an instance of all the offending classes.
2604 So here is the function which does the work. It takes the ambiguous
2605 dictionaries and either resolves them (producing bindings) or
2606 complains. It works by splitting the dictionary list by type
2607 variable, and using @disambigOne@ to do the real business.
2609 @disambigOne@ assumes that its arguments dictionaries constrain all
2610 the same type variable.
2612 ADR Comment 20/6/94: I've changed the @CReturnable@ case to default to
2613 @()@ instead of @Int@. I reckon this is the Right Thing to do since
2614 the most common use of defaulting is code like:
2616 _ccall_ foo `seqPrimIO` bar
2618 Since we're not using the result of @foo@, the result if (presumably)
2622 disambiguate :: SDoc -> Bool -> DynFlags -> [Inst] -> TcM ([Inst], TcDictBinds)
2623 -- Just does unification to fix the default types
2624 -- The Insts are assumed to be pre-zonked
2625 disambiguate doc interactive dflags insts
2627 = return (insts, emptyBag)
2629 | null defaultable_groups
2630 = do { traceTc (text "disambigutate, no defaultable groups" <+> vcat [ppr unaries, ppr insts, ppr bad_tvs, ppr defaultable_groups])
2631 ; return (insts, emptyBag) }
2634 = do { -- Figure out what default types to use
2635 default_tys <- getDefaultTys extended_defaulting ovl_strings
2637 ; traceTc (text "disambiguate1" <+> vcat [ppr insts, ppr unaries, ppr bad_tvs, ppr defaultable_groups])
2638 ; mapM_ (disambigGroup default_tys) defaultable_groups
2640 -- disambigGroup does unification, hence try again
2641 ; tryHardCheckLoop doc insts }
2644 extended_defaulting = interactive || dopt Opt_ExtendedDefaultRules dflags
2645 ovl_strings = dopt Opt_OverloadedStrings dflags
2647 unaries :: [(Inst, Class, TcTyVar)] -- (C tv) constraints
2648 bad_tvs :: TcTyVarSet -- Tyvars mentioned by *other* constraints
2649 (unaries, bad_tvs_s) = partitionWith find_unary insts
2650 bad_tvs = unionVarSets bad_tvs_s
2652 -- Finds unary type-class constraints
2653 find_unary d@(Dict {tci_pred = ClassP cls [ty]})
2654 | Just tv <- tcGetTyVar_maybe ty = Left (d,cls,tv)
2655 find_unary inst = Right (tyVarsOfInst inst)
2657 -- Group by type variable
2658 defaultable_groups :: [[(Inst,Class,TcTyVar)]]
2659 defaultable_groups = filter defaultable_group (equivClasses cmp_tv unaries)
2660 cmp_tv (_,_,tv1) (_,_,tv2) = tv1 `compare` tv2
2662 defaultable_group :: [(Inst,Class,TcTyVar)] -> Bool
2663 defaultable_group ds@((_,_,tv):_)
2664 = isTyConableTyVar tv -- Note [Avoiding spurious errors]
2665 && not (tv `elemVarSet` bad_tvs)
2666 && defaultable_classes [c | (_,c,_) <- ds]
2667 defaultable_group [] = panic "defaultable_group"
2669 defaultable_classes clss
2670 | extended_defaulting = any isInteractiveClass clss
2671 | otherwise = all is_std_class clss && (any is_num_class clss)
2673 -- In interactive mode, or with -fextended-default-rules,
2674 -- we default Show a to Show () to avoid graututious errors on "show []"
2675 isInteractiveClass cls
2676 = is_num_class cls || (classKey cls `elem` [showClassKey, eqClassKey, ordClassKey])
2678 is_num_class cls = isNumericClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2679 -- is_num_class adds IsString to the standard numeric classes,
2680 -- when -foverloaded-strings is enabled
2682 is_std_class cls = isStandardClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2683 -- Similarly is_std_class
2685 -----------------------
2686 disambigGroup :: [Type] -- The default types
2687 -> [(Inst,Class,TcTyVar)] -- All standard classes of form (C a)
2688 -> TcM () -- Just does unification, to fix the default types
2690 disambigGroup default_tys dicts
2691 = try_default default_tys
2693 (_,_,tyvar) = ASSERT(not (null dicts)) head dicts -- Should be non-empty
2694 classes = [c | (_,c,_) <- dicts]
2696 try_default [] = return ()
2697 try_default (default_ty : default_tys)
2698 = tryTcLIE_ (try_default default_tys) $
2699 do { tcSimplifyDefault [mkClassPred clas [default_ty] | clas <- classes]
2700 -- This may fail; then the tryTcLIE_ kicks in
2701 -- Failure here is caused by there being no type in the
2702 -- default list which can satisfy all the ambiguous classes.
2703 -- For example, if Real a is reqd, but the only type in the
2704 -- default list is Int.
2706 -- After this we can't fail
2707 ; warnDefault dicts default_ty
2708 ; unifyType default_ty (mkTyVarTy tyvar)
2709 ; return () -- TOMDO: do something with the coercion
2713 -----------------------
2714 getDefaultTys :: Bool -> Bool -> TcM [Type]
2715 getDefaultTys extended_deflts ovl_strings
2716 = do { mb_defaults <- getDeclaredDefaultTys
2717 ; case mb_defaults of {
2718 Just tys -> return tys ; -- User-supplied defaults
2721 -- No use-supplied default
2722 -- Use [Integer, Double], plus modifications
2723 { integer_ty <- tcMetaTy integerTyConName
2724 ; checkWiredInTyCon doubleTyCon
2725 ; string_ty <- tcMetaTy stringTyConName
2726 ; return (opt_deflt extended_deflts unitTy
2727 -- Note [Default unitTy]
2729 [integer_ty,doubleTy]
2731 opt_deflt ovl_strings string_ty) } } }
2733 opt_deflt True ty = [ty]
2734 opt_deflt False _ = []
2737 Note [Default unitTy]
2738 ~~~~~~~~~~~~~~~~~~~~~
2739 In interative mode (or with -fextended-default-rules) we add () as the first type we
2740 try when defaulting. This has very little real impact, except in the following case.
2742 Text.Printf.printf "hello"
2743 This has type (forall a. IO a); it prints "hello", and returns 'undefined'. We don't
2744 want the GHCi repl loop to try to print that 'undefined'. The neatest thing is to
2745 default the 'a' to (), rather than to Integer (which is what would otherwise happen;
2746 and then GHCi doesn't attempt to print the (). So in interactive mode, we add
2747 () to the list of defaulting types. See Trac #1200.
2749 Note [Avoiding spurious errors]
2750 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2751 When doing the unification for defaulting, we check for skolem
2752 type variables, and simply don't default them. For example:
2753 f = (*) -- Monomorphic
2754 g :: Num a => a -> a
2756 Here, we get a complaint when checking the type signature for g,
2757 that g isn't polymorphic enough; but then we get another one when
2758 dealing with the (Num a) context arising from f's definition;
2759 we try to unify a with Int (to default it), but find that it's
2760 already been unified with the rigid variable from g's type sig
2763 %************************************************************************
2765 \subsection[simple]{@Simple@ versions}
2767 %************************************************************************
2769 Much simpler versions when there are no bindings to make!
2771 @tcSimplifyThetas@ simplifies class-type constraints formed by
2772 @deriving@ declarations and when specialising instances. We are
2773 only interested in the simplified bunch of class/type constraints.
2775 It simplifies to constraints of the form (C a b c) where
2776 a,b,c are type variables. This is required for the context of
2777 instance declarations.
2780 tcSimplifyDeriv :: InstOrigin
2782 -> ThetaType -- Wanted
2783 -> TcM ThetaType -- Needed
2784 -- Given instance (wanted) => C inst_ty
2785 -- Simplify 'wanted' as much as possible
2787 tcSimplifyDeriv orig tyvars theta
2788 = do { (tvs, _, tenv) <- tcInstTyVars tyvars
2789 -- The main loop may do unification, and that may crash if
2790 -- it doesn't see a TcTyVar, so we have to instantiate. Sigh
2791 -- ToDo: what if two of them do get unified?
2792 ; wanteds <- newDictBndrsO orig (substTheta tenv theta)
2793 ; (irreds, _) <- tryHardCheckLoop doc wanteds
2795 ; let (tv_dicts, others) = partition ok irreds
2796 ; addNoInstanceErrs others
2797 -- See Note [Exotic derived instance contexts] in TcMType
2799 ; let rev_env = zipTopTvSubst tvs (mkTyVarTys tyvars)
2800 simpl_theta = substTheta rev_env (map dictPred tv_dicts)
2801 -- This reverse-mapping is a pain, but the result
2802 -- should mention the original TyVars not TcTyVars
2804 ; return simpl_theta }
2806 doc = ptext (sLit "deriving classes for a data type")
2808 ok dict | isDict dict = validDerivPred (dictPred dict)
2813 @tcSimplifyDefault@ just checks class-type constraints, essentially;
2814 used with \tr{default} declarations. We are only interested in
2815 whether it worked or not.
2818 tcSimplifyDefault :: ThetaType -- Wanted; has no type variables in it
2821 tcSimplifyDefault theta = do
2822 wanteds <- newDictBndrsO DefaultOrigin theta
2823 (irreds, _) <- tryHardCheckLoop doc wanteds
2824 addNoInstanceErrs irreds
2828 traceTc (ptext (sLit "tcSimplifyDefault failing")) >> failM
2830 doc = ptext (sLit "default declaration")
2834 %************************************************************************
2836 \section{Errors and contexts}
2838 %************************************************************************
2840 ToDo: for these error messages, should we note the location as coming
2841 from the insts, or just whatever seems to be around in the monad just
2845 groupErrs :: ([Inst] -> TcM ()) -- Deal with one group
2846 -> [Inst] -- The offending Insts
2848 -- Group together insts with the same origin
2849 -- We want to report them together in error messages
2853 groupErrs report_err (inst:insts)
2854 = do { do_one (inst:friends)
2855 ; groupErrs report_err others }
2857 -- (It may seem a bit crude to compare the error messages,
2858 -- but it makes sure that we combine just what the user sees,
2859 -- and it avoids need equality on InstLocs.)
2860 (friends, others) = partition is_friend insts
2861 loc_msg = showSDoc (pprInstLoc (instLoc inst))
2862 is_friend friend = showSDoc (pprInstLoc (instLoc friend)) == loc_msg
2863 do_one insts = addInstCtxt (instLoc (head insts)) (report_err insts)
2864 -- Add location and context information derived from the Insts
2866 -- Add the "arising from..." part to a message about bunch of dicts
2867 addInstLoc :: [Inst] -> Message -> Message
2868 addInstLoc insts msg = msg $$ nest 2 (pprInstArising (head insts))
2870 addTopIPErrs :: [Name] -> [Inst] -> TcM ()
2873 addTopIPErrs bndrs ips
2874 = do { dflags <- getDOpts
2875 ; addErrTcM (tidy_env, mk_msg dflags tidy_ips) }
2877 (tidy_env, tidy_ips) = tidyInsts ips
2879 = vcat [sep [ptext (sLit "Implicit parameters escape from"),
2880 nest 2 (ptext (sLit "the monomorphic top-level binding")
2881 <> plural bndrs <+> ptext (sLit "of")
2882 <+> pprBinders bndrs <> colon)],
2883 nest 2 (vcat (map ppr_ip ips)),
2884 monomorphism_fix dflags]
2885 ppr_ip ip = pprPred (dictPred ip) <+> pprInstArising ip
2887 topIPErrs :: [Inst] -> TcM ()
2889 = groupErrs report tidy_dicts
2891 (tidy_env, tidy_dicts) = tidyInsts dicts
2892 report dicts = addErrTcM (tidy_env, mk_msg dicts)
2893 mk_msg dicts = addInstLoc dicts (ptext (sLit "Unbound implicit parameter") <>
2894 plural tidy_dicts <+> pprDictsTheta tidy_dicts)
2896 addNoInstanceErrs :: [Inst] -- Wanted (can include implications)
2898 addNoInstanceErrs insts
2899 = do { let (tidy_env, tidy_insts) = tidyInsts insts
2900 ; reportNoInstances tidy_env Nothing tidy_insts }
2904 -> Maybe (InstLoc, [Inst]) -- Context
2905 -- Nothing => top level
2906 -- Just (d,g) => d describes the construct
2908 -> [Inst] -- What is wanted (can include implications)
2911 reportNoInstances tidy_env mb_what insts
2912 = groupErrs (report_no_instances tidy_env mb_what) insts
2914 report_no_instances :: TidyEnv -> Maybe (InstLoc, [Inst]) -> [Inst] -> TcM ()
2915 report_no_instances tidy_env mb_what insts
2916 = do { inst_envs <- tcGetInstEnvs
2917 ; let (implics, insts1) = partition isImplicInst insts
2918 (insts2, overlaps) = partitionWith (check_overlap inst_envs) insts1
2919 (eqInsts, insts3) = partition isEqInst insts2
2920 ; traceTc (text "reportNoInstances" <+> vcat
2921 [ppr insts, ppr implics, ppr insts1, ppr insts2])
2922 ; mapM_ complain_implic implics
2923 ; mapM_ (\doc -> addErrTcM (tidy_env, doc)) overlaps
2924 ; groupErrs complain_no_inst insts3
2925 ; mapM_ (addErrTcM . mk_eq_err) eqInsts
2928 complain_no_inst insts = addErrTcM (tidy_env, mk_no_inst_err insts)
2930 complain_implic inst -- Recurse!
2931 = reportNoInstances tidy_env
2932 (Just (tci_loc inst, tci_given inst))
2935 check_overlap :: (InstEnv,InstEnv) -> Inst -> Either Inst SDoc
2936 -- Right msg => overlap message
2937 -- Left inst => no instance
2938 check_overlap inst_envs wanted
2939 | not (isClassDict wanted) = Left wanted
2941 = case lookupInstEnv inst_envs clas tys of
2942 ([], _) -> Left wanted -- No match
2943 -- The case of exactly one match and no unifiers means a
2944 -- successful lookup. That can't happen here, because dicts
2945 -- only end up here if they didn't match in Inst.lookupInst
2947 | debugIsOn -> pprPanic "reportNoInstance" (ppr wanted)
2948 res -> Right (mk_overlap_msg wanted res)
2950 (clas,tys) = getDictClassTys wanted
2952 mk_overlap_msg dict (matches, unifiers)
2953 = ASSERT( not (null matches) )
2954 vcat [ addInstLoc [dict] ((ptext (sLit "Overlapping instances for")
2955 <+> pprPred (dictPred dict))),
2956 sep [ptext (sLit "Matching instances") <> colon,
2957 nest 2 (vcat [pprInstances ispecs, pprInstances unifiers])],
2958 if not (isSingleton matches)
2959 then -- Two or more matches
2961 else -- One match, plus some unifiers
2962 ASSERT( not (null unifiers) )
2963 parens (vcat [ptext (sLit "The choice depends on the instantiation of") <+>
2964 quotes (pprWithCommas ppr (varSetElems (tyVarsOfInst dict))),
2965 ptext (sLit "To pick the first instance above, use -fallow-incoherent-instances"),
2966 ptext (sLit "when compiling the other instance declarations")])]
2968 ispecs = [ispec | (ispec, _) <- matches]
2970 mk_eq_err :: Inst -> (TidyEnv, SDoc)
2971 mk_eq_err inst = misMatchMsg tidy_env (eqInstTys inst)
2973 mk_no_inst_err insts
2974 | null insts = empty
2976 | Just (loc, givens) <- mb_what, -- Nested (type signatures, instance decls)
2977 not (isEmptyVarSet (tyVarsOfInsts insts))
2978 = vcat [ addInstLoc insts $
2979 sep [ ptext (sLit "Could not deduce") <+> pprDictsTheta insts
2980 , nest 2 $ ptext (sLit "from the context") <+> pprDictsTheta givens]
2981 , show_fixes (fix1 loc : fixes2) ]
2983 | otherwise -- Top level
2984 = vcat [ addInstLoc insts $
2985 ptext (sLit "No instance") <> plural insts
2986 <+> ptext (sLit "for") <+> pprDictsTheta insts
2987 , show_fixes fixes2 ]
2990 fix1 loc = sep [ ptext (sLit "add") <+> pprDictsTheta insts
2991 <+> ptext (sLit "to the context of"),
2992 nest 2 (ppr (instLocOrigin loc)) ]
2993 -- I'm not sure it helps to add the location
2994 -- nest 2 (ptext (sLit "at") <+> ppr (instLocSpan loc)) ]
2996 fixes2 | null instance_dicts = []
2997 | otherwise = [sep [ptext (sLit "add an instance declaration for"),
2998 pprDictsTheta instance_dicts]]
2999 instance_dicts = [d | d <- insts, isClassDict d, not (isTyVarDict d)]
3000 -- Insts for which it is worth suggesting an adding an instance declaration
3001 -- Exclude implicit parameters, and tyvar dicts
3003 show_fixes :: [SDoc] -> SDoc
3004 show_fixes [] = empty
3005 show_fixes (f:fs) = sep [ptext (sLit "Possible fix:"),
3006 nest 2 (vcat (f : map (ptext (sLit "or") <+>) fs))]
3008 addTopAmbigErrs :: [Inst] -> TcRn ()
3009 addTopAmbigErrs dicts
3010 -- Divide into groups that share a common set of ambiguous tyvars
3011 = ifErrsM (return ()) $ -- Only report ambiguity if no other errors happened
3012 -- See Note [Avoiding spurious errors]
3013 mapM_ report (equivClasses cmp [(d, tvs_of d) | d <- tidy_dicts])
3015 (tidy_env, tidy_dicts) = tidyInsts dicts
3017 tvs_of :: Inst -> [TcTyVar]
3018 tvs_of d = varSetElems (tyVarsOfInst d)
3019 cmp (_,tvs1) (_,tvs2) = tvs1 `compare` tvs2
3021 report :: [(Inst,[TcTyVar])] -> TcM ()
3022 report pairs@((inst,tvs) : _) = do -- The pairs share a common set of ambiguous tyvars
3023 (tidy_env, mono_msg) <- mkMonomorphismMsg tidy_env tvs
3024 setSrcSpan (instSpan inst) $
3025 -- the location of the first one will do for the err message
3026 addErrTcM (tidy_env, msg $$ mono_msg)
3028 dicts = map fst pairs
3029 msg = sep [text "Ambiguous type variable" <> plural tvs <+>
3030 pprQuotedList tvs <+> in_msg,
3031 nest 2 (pprDictsInFull dicts)]
3032 in_msg = text "in the constraint" <> plural dicts <> colon
3033 report [] = panic "addTopAmbigErrs"
3036 mkMonomorphismMsg :: TidyEnv -> [TcTyVar] -> TcM (TidyEnv, Message)
3037 -- There's an error with these Insts; if they have free type variables
3038 -- it's probably caused by the monomorphism restriction.
3039 -- Try to identify the offending variable
3040 -- ASSUMPTION: the Insts are fully zonked
3041 mkMonomorphismMsg tidy_env inst_tvs
3042 = do { dflags <- getDOpts
3043 ; (tidy_env, docs) <- findGlobals (mkVarSet inst_tvs) tidy_env
3044 ; return (tidy_env, mk_msg dflags docs) }
3046 mk_msg _ _ | any isRuntimeUnk inst_tvs
3047 = vcat [ptext (sLit "Cannot resolve unknown runtime types:") <+>
3048 (pprWithCommas ppr inst_tvs),
3049 ptext (sLit "Use :print or :force to determine these types")]
3050 mk_msg _ [] = ptext (sLit "Probable fix: add a type signature that fixes these type variable(s)")
3051 -- This happens in things like
3052 -- f x = show (read "foo")
3053 -- where monomorphism doesn't play any role
3055 = vcat [ptext (sLit "Possible cause: the monomorphism restriction applied to the following:"),
3057 monomorphism_fix dflags]
3059 monomorphism_fix :: DynFlags -> SDoc
3060 monomorphism_fix dflags
3061 = ptext (sLit "Probable fix:") <+> vcat
3062 [ptext (sLit "give these definition(s) an explicit type signature"),
3063 if dopt Opt_MonomorphismRestriction dflags
3064 then ptext (sLit "or use -fno-monomorphism-restriction")
3065 else empty] -- Only suggest adding "-fno-monomorphism-restriction"
3066 -- if it is not already set!
3068 warnDefault :: [(Inst, Class, Var)] -> Type -> TcM ()
3069 warnDefault ups default_ty = do
3070 warn_flag <- doptM Opt_WarnTypeDefaults
3071 addInstCtxt (instLoc (head (dicts))) (warnTc warn_flag warn_msg)
3073 dicts = [d | (d,_,_) <- ups]
3076 (_, tidy_dicts) = tidyInsts dicts
3077 warn_msg = vcat [ptext (sLit "Defaulting the following constraint(s) to type") <+>
3078 quotes (ppr default_ty),
3079 pprDictsInFull tidy_dicts]
3081 reduceDepthErr :: Int -> [Inst] -> SDoc
3082 reduceDepthErr n stack
3083 = vcat [ptext (sLit "Context reduction stack overflow; size =") <+> int n,
3084 ptext (sLit "Use -fcontext-stack=N to increase stack size to N"),
3085 nest 4 (pprStack stack)]
3087 pprStack :: [Inst] -> SDoc
3088 pprStack stack = vcat (map pprInstInFull stack)