2 % (c) The University of Glasgow 2006
3 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
10 tcSimplifyInfer, tcSimplifyInferCheck,
11 tcSimplifyCheck, tcSimplifyRestricted,
12 tcSimplifyRuleLhs, tcSimplifyIPs,
13 tcSimplifySuperClasses,
14 tcSimplifyTop, tcSimplifyInteractive,
15 tcSimplifyBracket, tcSimplifyCheckPat,
17 tcSimplifyDeriv, tcSimplifyDefault,
25 #include "HsVersions.h"
27 import {-# SOURCE #-} TcUnify( unifyType )
31 import TcHsSyn ( hsLPatType )
39 import DsUtils -- Big-tuple functions
69 %************************************************************************
73 %************************************************************************
75 --------------------------------------
76 Notes on functional dependencies (a bug)
77 --------------------------------------
84 instance D a b => C a b -- Undecidable
85 -- (Not sure if it's crucial to this eg)
86 f :: C a b => a -> Bool
89 g :: C a b => a -> Bool
92 Here f typechecks, but g does not!! Reason: before doing improvement,
93 we reduce the (C a b1) constraint from the call of f to (D a b1).
95 Here is a more complicated example:
98 > class Foo a b | a->b
100 > class Bar a b | a->b
104 > instance Bar Obj Obj
106 > instance (Bar a b) => Foo a b
108 > foo:: (Foo a b) => a -> String
111 > runFoo:: (forall a b. (Foo a b) => a -> w) -> w
117 Could not deduce (Bar a b) from the context (Foo a b)
118 arising from use of `foo' at <interactive>:1
120 Add (Bar a b) to the expected type of an expression
121 In the first argument of `runFoo', namely `foo'
122 In the definition of `it': it = runFoo foo
124 Why all of the sudden does GHC need the constraint Bar a b? The
125 function foo didn't ask for that...
128 The trouble is that to type (runFoo foo), GHC has to solve the problem:
130 Given constraint Foo a b
131 Solve constraint Foo a b'
133 Notice that b and b' aren't the same. To solve this, just do
134 improvement and then they are the same. But GHC currently does
139 That is usually fine, but it isn't here, because it sees that Foo a b is
140 not the same as Foo a b', and so instead applies the instance decl for
141 instance Bar a b => Foo a b. And that's where the Bar constraint comes
144 The Right Thing is to improve whenever the constraint set changes at
145 all. Not hard in principle, but it'll take a bit of fiddling to do.
147 Note [Choosing which variables to quantify]
148 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
149 Suppose we are about to do a generalisation step. We have in our hand
152 T the type of the RHS
153 C the constraints from that RHS
155 The game is to figure out
157 Q the set of type variables over which to quantify
158 Ct the constraints we will *not* quantify over
159 Cq the constraints we will quantify over
161 So we're going to infer the type
165 and float the constraints Ct further outwards.
167 Here are the things that *must* be true:
169 (A) Q intersect fv(G) = EMPTY limits how big Q can be
170 (B) Q superset fv(Cq union T) \ oclose(fv(G),C) limits how small Q can be
172 (A) says we can't quantify over a variable that's free in the environment.
173 (B) says we must quantify over all the truly free variables in T, else
174 we won't get a sufficiently general type.
176 We do not *need* to quantify over any variable that is fixed by the
177 free vars of the environment G.
179 BETWEEN THESE TWO BOUNDS, ANY Q WILL DO!
181 Example: class H x y | x->y where ...
183 fv(G) = {a} C = {H a b, H c d}
186 (A) Q intersect {a} is empty
187 (B) Q superset {a,b,c,d} \ oclose({a}, C) = {a,b,c,d} \ {a,b} = {c,d}
189 So Q can be {c,d}, {b,c,d}
191 In particular, it's perfectly OK to quantify over more type variables
192 than strictly necessary; there is no need to quantify over 'b', since
193 it is determined by 'a' which is free in the envt, but it's perfectly
194 OK to do so. However we must not quantify over 'a' itself.
196 Other things being equal, however, we'd like to quantify over as few
197 variables as possible: smaller types, fewer type applications, more
198 constraints can get into Ct instead of Cq. Here's a good way to
201 Q = grow( fv(T), C ) \ oclose( fv(G), C )
203 That is, quantify over all variable that that MIGHT be fixed by the
204 call site (which influences T), but which aren't DEFINITELY fixed by
205 G. This choice definitely quantifies over enough type variables,
206 albeit perhaps too many.
208 Why grow( fv(T), C ) rather than fv(T)? Consider
210 class H x y | x->y where ...
215 If we used fv(T) = {c} we'd get the type
217 forall c. H c d => c -> b
219 And then if the fn was called at several different c's, each of
220 which fixed d differently, we'd get a unification error, because
221 d isn't quantified. Solution: quantify d. So we must quantify
222 everything that might be influenced by c.
224 Why not oclose( fv(T), C )? Because we might not be able to see
225 all the functional dependencies yet:
227 class H x y | x->y where ...
228 instance H x y => Eq (T x y) where ...
233 Now oclose(fv(T),C) = {c}, because the functional dependency isn't
234 apparent yet, and that's wrong. We must really quantify over d too.
236 There really isn't any point in quantifying over any more than
237 grow( fv(T), C ), because the call sites can't possibly influence
238 any other type variables.
242 -------------------------------------
244 -------------------------------------
246 It's very hard to be certain when a type is ambiguous. Consider
250 instance H x y => K (x,y)
252 Is this type ambiguous?
253 forall a b. (K (a,b), Eq b) => a -> a
255 Looks like it! But if we simplify (K (a,b)) we get (H a b) and
256 now we see that a fixes b. So we can't tell about ambiguity for sure
257 without doing a full simplification. And even that isn't possible if
258 the context has some free vars that may get unified. Urgle!
260 Here's another example: is this ambiguous?
261 forall a b. Eq (T b) => a -> a
262 Not if there's an insance decl (with no context)
263 instance Eq (T b) where ...
265 You may say of this example that we should use the instance decl right
266 away, but you can't always do that:
268 class J a b where ...
269 instance J Int b where ...
271 f :: forall a b. J a b => a -> a
273 (Notice: no functional dependency in J's class decl.)
274 Here f's type is perfectly fine, provided f is only called at Int.
275 It's premature to complain when meeting f's signature, or even
276 when inferring a type for f.
280 However, we don't *need* to report ambiguity right away. It'll always
281 show up at the call site.... and eventually at main, which needs special
282 treatment. Nevertheless, reporting ambiguity promptly is an excellent thing.
284 So here's the plan. We WARN about probable ambiguity if
286 fv(Cq) is not a subset of oclose(fv(T) union fv(G), C)
288 (all tested before quantification).
289 That is, all the type variables in Cq must be fixed by the the variables
290 in the environment, or by the variables in the type.
292 Notice that we union before calling oclose. Here's an example:
294 class J a b c | a b -> c
298 forall b c. (J a b c) => b -> b
300 Only if we union {a} from G with {b} from T before using oclose,
301 do we see that c is fixed.
303 It's a bit vague exactly which C we should use for this oclose call. If we
304 don't fix enough variables we might complain when we shouldn't (see
305 the above nasty example). Nothing will be perfect. That's why we can
306 only issue a warning.
309 Can we ever be *certain* about ambiguity? Yes: if there's a constraint
311 c in C such that fv(c) intersect (fv(G) union fv(T)) = EMPTY
313 then c is a "bubble"; there's no way it can ever improve, and it's
314 certainly ambiguous. UNLESS it is a constant (sigh). And what about
319 instance H x y => K (x,y)
321 Is this type ambiguous?
322 forall a b. (K (a,b), Eq b) => a -> a
324 Urk. The (Eq b) looks "definitely ambiguous" but it isn't. What we are after
325 is a "bubble" that's a set of constraints
327 Cq = Ca union Cq' st fv(Ca) intersect (fv(Cq') union fv(T) union fv(G)) = EMPTY
329 Hence another idea. To decide Q start with fv(T) and grow it
330 by transitive closure in Cq (no functional dependencies involved).
331 Now partition Cq using Q, leaving the definitely-ambiguous and probably-ok.
332 The definitely-ambiguous can then float out, and get smashed at top level
333 (which squashes out the constants, like Eq (T a) above)
336 --------------------------------------
337 Notes on principal types
338 --------------------------------------
343 f x = let g y = op (y::Int) in True
345 Here the principal type of f is (forall a. a->a)
346 but we'll produce the non-principal type
347 f :: forall a. C Int => a -> a
350 --------------------------------------
351 The need for forall's in constraints
352 --------------------------------------
354 [Exchange on Haskell Cafe 5/6 Dec 2000]
356 class C t where op :: t -> Bool
357 instance C [t] where op x = True
359 p y = (let f :: c -> Bool; f x = op (y >> return x) in f, y ++ [])
360 q y = (y ++ [], let f :: c -> Bool; f x = op (y >> return x) in f)
362 The definitions of p and q differ only in the order of the components in
363 the pair on their right-hand sides. And yet:
365 ghc and "Typing Haskell in Haskell" reject p, but accept q;
366 Hugs rejects q, but accepts p;
367 hbc rejects both p and q;
368 nhc98 ... (Malcolm, can you fill in the blank for us!).
370 The type signature for f forces context reduction to take place, and
371 the results of this depend on whether or not the type of y is known,
372 which in turn depends on which component of the pair the type checker
375 Solution: if y::m a, float out the constraints
376 Monad m, forall c. C (m c)
377 When m is later unified with [], we can solve both constraints.
380 --------------------------------------
381 Notes on implicit parameters
382 --------------------------------------
384 Note [Inheriting implicit parameters]
385 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
390 where f is *not* a top-level binding.
391 From the RHS of f we'll get the constraint (?y::Int).
392 There are two types we might infer for f:
396 (so we get ?y from the context of f's definition), or
398 f :: (?y::Int) => Int -> Int
400 At first you might think the first was better, becuase then
401 ?y behaves like a free variable of the definition, rather than
402 having to be passed at each call site. But of course, the WHOLE
403 IDEA is that ?y should be passed at each call site (that's what
404 dynamic binding means) so we'd better infer the second.
406 BOTTOM LINE: when *inferring types* you *must* quantify
407 over implicit parameters. See the predicate isFreeWhenInferring.
410 Note [Implicit parameters and ambiguity]
411 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
412 Only a *class* predicate can give rise to ambiguity
413 An *implicit parameter* cannot. For example:
414 foo :: (?x :: [a]) => Int
416 is fine. The call site will suppply a particular 'x'
418 Furthermore, the type variables fixed by an implicit parameter
419 propagate to the others. E.g.
420 foo :: (Show a, ?x::[a]) => Int
422 The type of foo looks ambiguous. But it isn't, because at a call site
424 let ?x = 5::Int in foo
425 and all is well. In effect, implicit parameters are, well, parameters,
426 so we can take their type variables into account as part of the
427 "tau-tvs" stuff. This is done in the function 'FunDeps.grow'.
430 Question 2: type signatures
431 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
432 BUT WATCH OUT: When you supply a type signature, we can't force you
433 to quantify over implicit parameters. For example:
437 This is perfectly reasonable. We do not want to insist on
439 (?x + 1) :: (?x::Int => Int)
441 That would be silly. Here, the definition site *is* the occurrence site,
442 so the above strictures don't apply. Hence the difference between
443 tcSimplifyCheck (which *does* allow implicit paramters to be inherited)
444 and tcSimplifyCheckBind (which does not).
446 What about when you supply a type signature for a binding?
447 Is it legal to give the following explicit, user type
448 signature to f, thus:
453 At first sight this seems reasonable, but it has the nasty property
454 that adding a type signature changes the dynamic semantics.
457 (let f x = (x::Int) + ?y
458 in (f 3, f 3 with ?y=5)) with ?y = 6
464 in (f 3, f 3 with ?y=5)) with ?y = 6
468 Indeed, simply inlining f (at the Haskell source level) would change the
471 Nevertheless, as Launchbury says (email Oct 01) we can't really give the
472 semantics for a Haskell program without knowing its typing, so if you
473 change the typing you may change the semantics.
475 To make things consistent in all cases where we are *checking* against
476 a supplied signature (as opposed to inferring a type), we adopt the
479 a signature does not need to quantify over implicit params.
481 [This represents a (rather marginal) change of policy since GHC 5.02,
482 which *required* an explicit signature to quantify over all implicit
483 params for the reasons mentioned above.]
485 But that raises a new question. Consider
487 Given (signature) ?x::Int
488 Wanted (inferred) ?x::Int, ?y::Bool
490 Clearly we want to discharge the ?x and float the ?y out. But
491 what is the criterion that distinguishes them? Clearly it isn't
492 what free type variables they have. The Right Thing seems to be
493 to float a constraint that
494 neither mentions any of the quantified type variables
495 nor any of the quantified implicit parameters
497 See the predicate isFreeWhenChecking.
500 Question 3: monomorphism
501 ~~~~~~~~~~~~~~~~~~~~~~~~
502 There's a nasty corner case when the monomorphism restriction bites:
506 The argument above suggests that we *must* generalise
507 over the ?y parameter, to get
508 z :: (?y::Int) => Int,
509 but the monomorphism restriction says that we *must not*, giving
511 Why does the momomorphism restriction say this? Because if you have
513 let z = x + ?y in z+z
515 you might not expect the addition to be done twice --- but it will if
516 we follow the argument of Question 2 and generalise over ?y.
519 Question 4: top level
520 ~~~~~~~~~~~~~~~~~~~~~
521 At the top level, monomorhism makes no sense at all.
524 main = let ?x = 5 in print foo
528 woggle :: (?x :: Int) => Int -> Int
531 We definitely don't want (foo :: Int) with a top-level implicit parameter
532 (?x::Int) becuase there is no way to bind it.
537 (A) Always generalise over implicit parameters
538 Bindings that fall under the monomorphism restriction can't
542 * Inlining remains valid
543 * No unexpected loss of sharing
544 * But simple bindings like
546 will be rejected, unless you add an explicit type signature
547 (to avoid the monomorphism restriction)
548 z :: (?y::Int) => Int
550 This seems unacceptable
552 (B) Monomorphism restriction "wins"
553 Bindings that fall under the monomorphism restriction can't
555 Always generalise over implicit parameters *except* for bindings
556 that fall under the monomorphism restriction
559 * Inlining isn't valid in general
560 * No unexpected loss of sharing
561 * Simple bindings like
563 accepted (get value of ?y from binding site)
565 (C) Always generalise over implicit parameters
566 Bindings that fall under the monomorphism restriction can't
567 be generalised, EXCEPT for implicit parameters
569 * Inlining remains valid
570 * Unexpected loss of sharing (from the extra generalisation)
571 * Simple bindings like
573 accepted (get value of ?y from occurrence sites)
578 None of these choices seems very satisfactory. But at least we should
579 decide which we want to do.
581 It's really not clear what is the Right Thing To Do. If you see
585 would you expect the value of ?y to be got from the *occurrence sites*
586 of 'z', or from the valuue of ?y at the *definition* of 'z'? In the
587 case of function definitions, the answer is clearly the former, but
588 less so in the case of non-fucntion definitions. On the other hand,
589 if we say that we get the value of ?y from the definition site of 'z',
590 then inlining 'z' might change the semantics of the program.
592 Choice (C) really says "the monomorphism restriction doesn't apply
593 to implicit parameters". Which is fine, but remember that every
594 innocent binding 'x = ...' that mentions an implicit parameter in
595 the RHS becomes a *function* of that parameter, called at each
596 use of 'x'. Now, the chances are that there are no intervening 'with'
597 clauses that bind ?y, so a decent compiler should common up all
598 those function calls. So I think I strongly favour (C). Indeed,
599 one could make a similar argument for abolishing the monomorphism
600 restriction altogether.
602 BOTTOM LINE: we choose (B) at present. See tcSimplifyRestricted
606 %************************************************************************
608 \subsection{tcSimplifyInfer}
610 %************************************************************************
612 tcSimplify is called when we *inferring* a type. Here's the overall game plan:
614 1. Compute Q = grow( fvs(T), C )
616 2. Partition C based on Q into Ct and Cq. Notice that ambiguous
617 predicates will end up in Ct; we deal with them at the top level
619 3. Try improvement, using functional dependencies
621 4. If Step 3 did any unification, repeat from step 1
622 (Unification can change the result of 'grow'.)
624 Note: we don't reduce dictionaries in step 2. For example, if we have
625 Eq (a,b), we don't simplify to (Eq a, Eq b). So Q won't be different
626 after step 2. However note that we may therefore quantify over more
627 type variables than we absolutely have to.
629 For the guts, we need a loop, that alternates context reduction and
630 improvement with unification. E.g. Suppose we have
632 class C x y | x->y where ...
634 and tcSimplify is called with:
636 Then improvement unifies a with b, giving
639 If we need to unify anything, we rattle round the whole thing all over
646 -> TcTyVarSet -- fv(T); type vars
648 -> TcM ([TcTyVar], -- Tyvars to quantify (zonked and quantified)
649 [Inst], -- Dict Ids that must be bound here (zonked)
650 TcDictBinds) -- Bindings
651 -- Any free (escaping) Insts are tossed into the environment
656 tcSimplifyInfer doc tau_tvs wanted
657 = do { tau_tvs1 <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
658 ; wanted' <- mapM zonkInst wanted -- Zonk before deciding quantified tyvars
659 ; gbl_tvs <- tcGetGlobalTyVars
660 ; let preds1 = fdPredsOfInsts wanted'
661 gbl_tvs1 = oclose preds1 gbl_tvs
662 qtvs = growInstsTyVars wanted' tau_tvs1 `minusVarSet` gbl_tvs1
663 -- See Note [Choosing which variables to quantify]
665 -- To maximise sharing, remove from consideration any
666 -- constraints that don't mention qtvs at all
667 ; let (free, bound) = partition (isFreeWhenInferring qtvs) wanted'
670 -- To make types simple, reduce as much as possible
671 ; traceTc (text "infer" <+> (ppr preds1 $$ ppr (growInstsTyVars wanted' tau_tvs1) $$ ppr gbl_tvs $$
672 ppr gbl_tvs1 $$ ppr free $$ ppr bound))
673 ; (irreds1, binds1) <- tryHardCheckLoop doc bound
675 -- Note [Inference and implication constraints]
676 ; let want_dict d = tyVarsOfInst d `intersectsVarSet` qtvs
677 ; (irreds2, binds2) <- approximateImplications doc want_dict irreds1
679 -- Now work out all over again which type variables to quantify,
680 -- exactly in the same way as before, but starting from irreds2. Why?
681 -- a) By now improvment may have taken place, and we must *not*
682 -- quantify over any variable free in the environment
683 -- tc137 (function h inside g) is an example
685 -- b) Do not quantify over constraints that *now* do not
686 -- mention quantified type variables, because they are
687 -- simply ambiguous (or might be bound further out). Example:
688 -- f :: Eq b => a -> (a, b)
690 -- From the RHS of g we get the MethodInst f77 :: alpha -> (alpha, beta)
691 -- We decide to quantify over 'alpha' alone, but free1 does not include f77
692 -- because f77 mentions 'alpha'. Then reducing leaves only the (ambiguous)
693 -- constraint (Eq beta), which we dump back into the free set
694 -- See test tcfail181
696 -- c) irreds may contain type variables not previously mentioned,
697 -- e.g. instance D a x => Foo [a]
699 -- Then after simplifying we'll get (D a x), and x is fresh
700 -- We must quantify over x else it'll be totally unbound
701 ; tau_tvs2 <- zonkTcTyVarsAndFV (varSetElems tau_tvs1)
702 ; gbl_tvs2 <- zonkTcTyVarsAndFV (varSetElems gbl_tvs1)
703 -- Note that we start from gbl_tvs1
704 -- We use tcGetGlobalTyVars, then oclose wrt preds2, because
705 -- we've already put some of the original preds1 into frees
706 -- E.g. wanteds = C a b (where a->b)
709 -- Then b is fixed by gbl_tvs, so (C a b) will be in free, and
710 -- irreds2 will be empty. But we don't want to generalise over b!
711 ; let preds2 = fdPredsOfInsts irreds2 -- irreds2 is zonked
712 qtvs = growInstsTyVars irreds2 tau_tvs2 `minusVarSet` oclose preds2 gbl_tvs2
713 ; let (free, irreds3) = partition (isFreeWhenInferring qtvs) irreds2
716 -- Turn the quantified meta-type variables into real type variables
717 ; qtvs2 <- zonkQuantifiedTyVars (varSetElems qtvs)
719 -- We can't abstract over any remaining unsolved
720 -- implications so instead just float them outwards. Ugh.
721 ; let (q_dicts0, implics) = partition isAbstractableInst irreds3
722 ; loc <- getInstLoc (ImplicOrigin doc)
723 ; implic_bind <- bindIrreds loc qtvs2 q_dicts0 implics
725 -- Prepare equality instances for quantification
726 ; let (q_eqs0,q_dicts) = partition isEqInst q_dicts0
727 ; q_eqs <- mapM finalizeEqInst q_eqs0
729 ; return (qtvs2, q_eqs ++ q_dicts, binds1 `unionBags` binds2 `unionBags` implic_bind) }
730 -- NB: when we are done, we might have some bindings, but
731 -- the final qtvs might be empty. See Note [NO TYVARS] below.
733 approximateImplications :: SDoc -> (Inst -> Bool) -> [Inst] -> TcM ([Inst], TcDictBinds)
734 -- Note [Inference and implication constraints]
735 -- Given a bunch of Dict and ImplicInsts, try to approximate the implications by
736 -- - fetching any dicts inside them that are free
737 -- - using those dicts as cruder constraints, to solve the implications
738 -- - returning the extra ones too
740 approximateImplications doc want_dict irreds
742 = return (irreds, emptyBag)
744 = do { extra_dicts' <- mapM cloneDict extra_dicts
745 ; tryHardCheckLoop doc (extra_dicts' ++ irreds) }
746 -- By adding extra_dicts', we make them
747 -- available to solve the implication constraints
749 extra_dicts = get_dicts (filter isImplicInst irreds)
751 get_dicts :: [Inst] -> [Inst] -- Returns only Dicts
752 -- Find the wanted constraints in implication constraints that satisfy
753 -- want_dict, and are not bound by forall's in the constraint itself
754 get_dicts ds = concatMap get_dict ds
756 get_dict d@(Dict {}) | want_dict d = [d]
758 get_dict (ImplicInst {tci_tyvars = tvs, tci_wanted = wanteds})
759 = [ d | let tv_set = mkVarSet tvs
760 , d <- get_dicts wanteds
761 , not (tyVarsOfInst d `intersectsVarSet` tv_set)]
762 get_dict i@(EqInst {}) | want_dict i = [i]
764 get_dict other = pprPanic "approximateImplications" (ppr other)
767 Note [Inference and implication constraints]
768 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
769 Suppose we have a wanted implication constraint (perhaps arising from
770 a nested pattern match) like
772 and we are now trying to quantify over 'a' when inferring the type for
773 a function. In principle it's possible that there might be an instance
774 instance (C a, E a) => D [a]
775 so the context (E a) would suffice. The Right Thing is to abstract over
776 the implication constraint, but we don't do that (a) because it'll be
777 surprising to programmers and (b) because we don't have the machinery to deal
778 with 'given' implications.
780 So our best approximation is to make (D [a]) part of the inferred
781 context, so we can use that to discharge the implication. Hence
782 the strange function get_dicts in approximateImplications.
784 The common cases are more clear-cut, when we have things like
786 Here, abstracting over (C b) is not an approximation at all -- but see
787 Note [Freeness and implications].
789 See Trac #1430 and test tc228.
793 -----------------------------------------------------------
794 -- tcSimplifyInferCheck is used when we know the constraints we are to simplify
795 -- against, but we don't know the type variables over which we are going to quantify.
796 -- This happens when we have a type signature for a mutually recursive group
799 -> TcTyVarSet -- fv(T)
802 -> TcM ([TyVar], -- Fully zonked, and quantified
803 TcDictBinds) -- Bindings
805 tcSimplifyInferCheck loc tau_tvs givens wanteds
806 = do { traceTc (text "tcSimplifyInferCheck <-" <+> ppr wanteds)
807 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
809 -- Figure out which type variables to quantify over
810 -- You might think it should just be the signature tyvars,
811 -- but in bizarre cases you can get extra ones
812 -- f :: forall a. Num a => a -> a
813 -- f x = fst (g (x, head [])) + 1
815 -- Here we infer g :: forall a b. a -> b -> (b,a)
816 -- We don't want g to be monomorphic in b just because
817 -- f isn't quantified over b.
818 ; let all_tvs = varSetElems (tau_tvs `unionVarSet` tyVarsOfInsts givens)
819 ; all_tvs <- zonkTcTyVarsAndFV all_tvs
820 ; gbl_tvs <- tcGetGlobalTyVars
821 ; let qtvs = varSetElems (all_tvs `minusVarSet` gbl_tvs)
822 -- We could close gbl_tvs, but its not necessary for
823 -- soundness, and it'll only affect which tyvars, not which
824 -- dictionaries, we quantify over
826 ; qtvs' <- zonkQuantifiedTyVars qtvs
828 -- Now we are back to normal (c.f. tcSimplCheck)
829 ; implic_bind <- bindIrreds loc qtvs' givens irreds
831 ; traceTc (text "tcSimplifyInferCheck ->" <+> ppr (implic_bind))
832 ; return (qtvs', binds `unionBags` implic_bind) }
835 Note [Squashing methods]
836 ~~~~~~~~~~~~~~~~~~~~~~~~~
837 Be careful if you want to float methods more:
838 truncate :: forall a. RealFrac a => forall b. Integral b => a -> b
839 From an application (truncate f i) we get
842 If we have also have a second occurrence of truncate, we get
845 When simplifying with i,f free, we might still notice that
846 t1=t3; but alas, the binding for t2 (which mentions t1)
847 may continue to float out!
852 class Y a b | a -> b where
855 instance Y [[a]] a where
858 k :: X a -> X a -> X a
860 g :: Num a => [X a] -> [X a]
863 h ys = ys ++ map (k (y [[0]])) xs
865 The excitement comes when simplifying the bindings for h. Initially
866 try to simplify {y @ [[t1]] t2, 0 @ t1}, with initial qtvs = {t2}.
867 From this we get t1~t2, but also various bindings. We can't forget
868 the bindings (because of [LOOP]), but in fact t1 is what g is
871 The net effect of [NO TYVARS]
874 isFreeWhenInferring :: TyVarSet -> Inst -> Bool
875 isFreeWhenInferring qtvs inst
876 = isFreeWrtTyVars qtvs inst -- Constrains no quantified vars
877 && isInheritableInst inst -- and no implicit parameter involved
878 -- see Note [Inheriting implicit parameters]
880 {- No longer used (with implication constraints)
881 isFreeWhenChecking :: TyVarSet -- Quantified tyvars
882 -> NameSet -- Quantified implicit parameters
884 isFreeWhenChecking qtvs ips inst
885 = isFreeWrtTyVars qtvs inst
886 && isFreeWrtIPs ips inst
889 isFreeWrtTyVars :: VarSet -> Inst -> Bool
890 isFreeWrtTyVars qtvs inst = tyVarsOfInst inst `disjointVarSet` qtvs
891 isFreeWrtIPs :: NameSet -> Inst -> Bool
892 isFreeWrtIPs ips inst = not (any (`elemNameSet` ips) (ipNamesOfInst inst))
896 %************************************************************************
898 \subsection{tcSimplifyCheck}
900 %************************************************************************
902 @tcSimplifyCheck@ is used when we know exactly the set of variables
903 we are going to quantify over. For example, a class or instance declaration.
906 -----------------------------------------------------------
907 -- tcSimplifyCheck is used when checking expression type signatures,
908 -- class decls, instance decls etc.
909 tcSimplifyCheck :: InstLoc
910 -> [TcTyVar] -- Quantify over these
913 -> TcM TcDictBinds -- Bindings
914 tcSimplifyCheck loc qtvs givens wanteds
915 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
916 do { traceTc (text "tcSimplifyCheck")
917 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
918 ; implic_bind <- bindIrreds loc qtvs givens irreds
919 ; return (binds `unionBags` implic_bind) }
921 -----------------------------------------------------------
922 -- tcSimplifyCheckPat is used for existential pattern match
923 tcSimplifyCheckPat :: InstLoc
924 -> [TcTyVar] -- Quantify over these
927 -> TcM TcDictBinds -- Bindings
928 tcSimplifyCheckPat loc qtvs givens wanteds
929 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
930 do { traceTc (text "tcSimplifyCheckPat")
931 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
932 ; implic_bind <- bindIrredsR loc qtvs givens irreds
933 ; return (binds `unionBags` implic_bind) }
935 -----------------------------------------------------------
936 bindIrreds :: InstLoc -> [TcTyVar]
939 bindIrreds loc qtvs givens irreds
940 = bindIrredsR loc qtvs givens irreds
942 bindIrredsR :: InstLoc -> [TcTyVar] -> [Inst] -> [Inst] -> TcM TcDictBinds
943 -- Make a binding that binds 'irreds', by generating an implication
944 -- constraint for them, *and* throwing the constraint into the LIE
945 bindIrredsR loc qtvs givens irreds
949 = do { let givens' = filter isAbstractableInst givens
950 -- The givens can (redundantly) include methods
951 -- We want to retain both EqInsts and Dicts
952 -- There should be no implicadtion constraints
953 -- See Note [Pruning the givens in an implication constraint]
955 -- If there are no 'givens', then it's safe to
956 -- partition the 'wanteds' by their qtvs, thereby trimming irreds
957 -- See Note [Freeness and implications]
958 ; irreds' <- if null givens'
960 { let qtv_set = mkVarSet qtvs
961 (frees, real_irreds) = partition (isFreeWrtTyVars qtv_set) irreds
963 ; return real_irreds }
966 ; (implics, bind) <- makeImplicationBind loc qtvs givens' irreds'
967 -- This call does the real work
968 -- If irreds' is empty, it does something sensible
973 makeImplicationBind :: InstLoc -> [TcTyVar]
975 -> TcM ([Inst], TcDictBinds)
976 -- Make a binding that binds 'irreds', by generating an implication
977 -- constraint for them.
979 -- The binding looks like
980 -- (ir1, .., irn) = f qtvs givens
981 -- where f is (evidence for) the new implication constraint
982 -- f :: forall qtvs. givens => (ir1, .., irn)
983 -- qtvs includes coercion variables
985 -- This binding must line up the 'rhs' in reduceImplication
986 makeImplicationBind loc all_tvs
987 givens -- Guaranteed all Dicts or EqInsts
989 | null irreds -- If there are no irreds, we are done
990 = return ([], emptyBag)
991 | otherwise -- Otherwise we must generate a binding
992 = do { uniq <- newUnique
993 ; span <- getSrcSpanM
994 ; let (eq_givens, dict_givens) = partition isEqInst givens
996 -- extract equality binders
997 eq_cotvs = map eqInstType eq_givens
999 -- make the implication constraint instance
1000 name = mkInternalName uniq (mkVarOcc "ic") span
1001 implic_inst = ImplicInst { tci_name = name,
1002 tci_tyvars = all_tvs,
1003 tci_given = eq_givens ++ dict_givens,
1004 -- same order as binders
1005 tci_wanted = irreds,
1008 -- create binders for the irreducible dictionaries
1009 dict_irreds = filter (not . isEqInst) irreds
1010 dict_irred_ids = map instToId dict_irreds
1011 lpat = mkBigLHsPatTup (map (L span . VarPat) dict_irred_ids)
1013 -- create the binding
1014 rhs = L span (mkHsWrap co (HsVar (instToId implic_inst)))
1015 co = mkWpApps (map instToId dict_givens)
1016 <.> mkWpTyApps eq_cotvs
1017 <.> mkWpTyApps (mkTyVarTys all_tvs)
1018 bind | [dict_irred_id] <- dict_irred_ids
1019 = VarBind dict_irred_id rhs
1021 = PatBind { pat_lhs = lpat
1022 , pat_rhs = unguardedGRHSs rhs
1023 , pat_rhs_ty = hsLPatType lpat
1024 , bind_fvs = placeHolderNames
1027 ; traceTc $ text "makeImplicationBind" <+> ppr implic_inst
1028 ; return ([implic_inst], unitBag (L span bind))
1031 -----------------------------------------------------------
1032 tryHardCheckLoop :: SDoc
1034 -> TcM ([Inst], TcDictBinds)
1036 tryHardCheckLoop doc wanteds
1037 = do { (irreds,binds) <- checkLoop (mkInferRedEnv doc try_me) wanteds
1038 ; return (irreds,binds)
1042 -- Here's the try-hard bit
1044 -----------------------------------------------------------
1045 gentleCheckLoop :: InstLoc
1048 -> TcM ([Inst], TcDictBinds)
1050 gentleCheckLoop inst_loc givens wanteds
1051 = do { (irreds,binds) <- checkLoop env wanteds
1052 ; return (irreds,binds)
1055 env = mkRedEnv (pprInstLoc inst_loc) try_me givens
1057 try_me inst | isMethodOrLit inst = ReduceMe
1059 -- When checking against a given signature
1060 -- we MUST be very gentle: Note [Check gently]
1062 gentleInferLoop :: SDoc -> [Inst]
1063 -> TcM ([Inst], TcDictBinds)
1064 gentleInferLoop doc wanteds
1065 = do { (irreds, binds) <- checkLoop env wanteds
1066 ; return (irreds, binds) }
1068 env = mkInferRedEnv doc try_me
1069 try_me inst | isMethodOrLit inst = ReduceMe
1074 ~~~~~~~~~~~~~~~~~~~~
1075 We have to very careful about not simplifying too vigorously
1080 f :: Show b => T b -> b
1081 f (MkT x) = show [x]
1083 Inside the pattern match, which binds (a:*, x:a), we know that
1085 Hence we have a dictionary for Show [a] available; and indeed we
1086 need it. We are going to build an implication contraint
1087 forall a. (b~[a]) => Show [a]
1088 Later, we will solve this constraint using the knowledge (Show b)
1090 But we MUST NOT reduce (Show [a]) to (Show a), else the whole
1091 thing becomes insoluble. So we simplify gently (get rid of literals
1092 and methods only, plus common up equal things), deferring the real
1093 work until top level, when we solve the implication constraint
1094 with tryHardCheckLooop.
1098 -----------------------------------------------------------
1101 -> TcM ([Inst], TcDictBinds)
1102 -- Precondition: givens are completely rigid
1103 -- Postcondition: returned Insts are zonked
1105 checkLoop env wanteds
1107 where go env wanteds
1108 = do { -- We do need to zonk the givens; cf Note [Zonking RedEnv]
1109 ; env' <- zonkRedEnv env
1110 ; wanteds' <- zonkInsts wanteds
1112 ; (improved, binds, irreds) <- reduceContext env' wanteds'
1114 ; if null irreds || not improved then
1115 return (irreds, binds)
1118 -- If improvement did some unification, we go round again.
1119 -- We start again with irreds, not wanteds
1120 -- Using an instance decl might have introduced a fresh type
1121 -- variable which might have been unified, so we'd get an
1122 -- infinite loop if we started again with wanteds!
1124 { (irreds1, binds1) <- go env' irreds
1125 ; return (irreds1, binds `unionBags` binds1) } }
1128 Note [Zonking RedEnv]
1129 ~~~~~~~~~~~~~~~~~~~~~
1130 It might appear as if the givens in RedEnv are always rigid, but that is not
1131 necessarily the case for programs involving higher-rank types that have class
1132 contexts constraining the higher-rank variables. An example from tc237 in the
1135 class Modular s a | s -> a
1137 wim :: forall a w. Integral a
1138 => a -> (forall s. Modular s a => M s w) -> w
1139 wim i k = error "urk"
1141 test5 :: (Modular s a, Integral a) => M s a
1144 test4 = wim 4 test4'
1146 Notice how the variable 'a' of (Modular s a) in the rank-2 type of wim is
1147 quantified further outside. When type checking test4, we have to check
1148 whether the signature of test5 is an instance of
1150 (forall s. Modular s a => M s w)
1152 Consequently, we will get (Modular s t_a), where t_a is a TauTv into the
1155 Given the FD of Modular in this example, class improvement will instantiate
1156 t_a to 'a', where 'a' is the skolem from test5's signatures (due to the
1157 Modular s a predicate in that signature). If we don't zonk (Modular s t_a) in
1158 the givens, we will get into a loop as improveOne uses the unification engine
1159 Unify.tcUnifyTys, which doesn't know about mutable type variables.
1164 class If b t e r | b t e -> r
1167 class Lte a b c | a b -> c where lte :: a -> b -> c
1169 instance (Lte a b l,If l b a c) => Max a b c
1171 Wanted: Max Z (S x) y
1173 Then we'll reduce using the Max instance to:
1174 (Lte Z (S x) l, If l (S x) Z y)
1175 and improve by binding l->T, after which we can do some reduction
1176 on both the Lte and If constraints. What we *can't* do is start again
1177 with (Max Z (S x) y)!
1181 %************************************************************************
1183 tcSimplifySuperClasses
1185 %************************************************************************
1187 Note [SUPERCLASS-LOOP 1]
1188 ~~~~~~~~~~~~~~~~~~~~~~~~
1189 We have to be very, very careful when generating superclasses, lest we
1190 accidentally build a loop. Here's an example:
1194 class S a => C a where { opc :: a -> a }
1195 class S b => D b where { opd :: b -> b }
1197 instance C Int where
1200 instance D Int where
1203 From (instance C Int) we get the constraint set {ds1:S Int, dd:D Int}
1204 Simplifying, we may well get:
1205 $dfCInt = :C ds1 (opd dd)
1208 Notice that we spot that we can extract ds1 from dd.
1210 Alas! Alack! We can do the same for (instance D Int):
1212 $dfDInt = :D ds2 (opc dc)
1216 And now we've defined the superclass in terms of itself.
1217 Two more nasty cases are in
1222 - Satisfy the superclass context *all by itself*
1223 (tcSimplifySuperClasses)
1224 - And do so completely; i.e. no left-over constraints
1225 to mix with the constraints arising from method declarations
1228 Note [Recursive instances and superclases]
1229 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1230 Consider this code, which arises in the context of "Scrap Your
1231 Boilerplate with Class".
1235 instance Sat (ctx Char) => Data ctx Char
1236 instance (Sat (ctx [a]), Data ctx a) => Data ctx [a]
1238 class Data Maybe a => Foo a
1240 instance Foo t => Sat (Maybe t)
1242 instance Data Maybe a => Foo a
1243 instance Foo a => Foo [a]
1246 In the instance for Foo [a], when generating evidence for the superclasses
1247 (ie in tcSimplifySuperClasses) we need a superclass (Data Maybe [a]).
1248 Using the instance for Data, we therefore need
1249 (Sat (Maybe [a], Data Maybe a)
1250 But we are given (Foo a), and hence its superclass (Data Maybe a).
1251 So that leaves (Sat (Maybe [a])). Using the instance for Sat means
1252 we need (Foo [a]). And that is the very dictionary we are bulding
1253 an instance for! So we must put that in the "givens". So in this
1255 Given: Foo a, Foo [a]
1256 Watend: Data Maybe [a]
1258 BUT we must *not not not* put the *superclasses* of (Foo [a]) in
1259 the givens, which is what 'addGiven' would normally do. Why? Because
1260 (Data Maybe [a]) is the superclass, so we'd "satisfy" the wanted
1261 by selecting a superclass from Foo [a], which simply makes a loop.
1263 On the other hand we *must* put the superclasses of (Foo a) in
1264 the givens, as you can see from the derivation described above.
1266 Conclusion: in the very special case of tcSimplifySuperClasses
1267 we have one 'given' (namely the "this" dictionary) whose superclasses
1268 must not be added to 'givens' by addGiven.
1270 There is a complication though. Suppose there are equalities
1271 instance (Eq a, a~b) => Num (a,b)
1272 Then we normalise the 'givens' wrt the equalities, so the original
1273 given "this" dictionary is cast to one of a different type. So it's a
1274 bit trickier than before to identify the "special" dictionary whose
1275 superclasses must not be added. See test
1276 indexed-types/should_run/EqInInstance
1278 We need a persistent property of the dictionary to record this
1279 special-ness. Current I'm using the InstLocOrigin (a bit of a hack,
1280 but cool), which is maintained by dictionary normalisation.
1281 Specifically, the InstLocOrigin is
1283 then the no-superclass thing kicks in. WATCH OUT if you fiddle
1287 tcSimplifySuperClasses
1289 -> Inst -- The dict whose superclasses
1290 -- are being figured out
1294 tcSimplifySuperClasses loc this givens sc_wanteds
1295 = do { traceTc (text "tcSimplifySuperClasses")
1297 -- Note [Recursive instances and superclases]
1298 ; no_sc_loc <- getInstLoc NoScOrigin
1299 ; let no_sc_this = setInstLoc this no_sc_loc
1301 ; let env = RedEnv { red_doc = pprInstLoc loc,
1302 red_try_me = try_me,
1303 red_givens = no_sc_this : givens,
1305 red_improve = False } -- No unification vars
1308 ; (irreds,binds1) <- checkLoop env sc_wanteds
1309 ; let (tidy_env, tidy_irreds) = tidyInsts irreds
1310 ; reportNoInstances tidy_env (Just (loc, givens)) [] tidy_irreds
1313 try_me _ = ReduceMe -- Try hard, so we completely solve the superclass
1314 -- constraints right here. See Note [SUPERCLASS-LOOP 1]
1318 %************************************************************************
1320 \subsection{tcSimplifyRestricted}
1322 %************************************************************************
1324 tcSimplifyRestricted infers which type variables to quantify for a
1325 group of restricted bindings. This isn't trivial.
1328 We want to quantify over a to get id :: forall a. a->a
1331 We do not want to quantify over a, because there's an Eq a
1332 constraint, so we get eq :: a->a->Bool (notice no forall)
1335 RHS has type 'tau', whose free tyvars are tau_tvs
1336 RHS has constraints 'wanteds'
1339 Quantify over (tau_tvs \ ftvs(wanteds))
1340 This is bad. The constraints may contain (Monad (ST s))
1341 where we have instance Monad (ST s) where...
1342 so there's no need to be monomorphic in s!
1344 Also the constraint might be a method constraint,
1345 whose type mentions a perfectly innocent tyvar:
1346 op :: Num a => a -> b -> a
1347 Here, b is unconstrained. A good example would be
1349 We want to infer the polymorphic type
1350 foo :: forall b. b -> b
1353 Plan B (cunning, used for a long time up to and including GHC 6.2)
1354 Step 1: Simplify the constraints as much as possible (to deal
1355 with Plan A's problem). Then set
1356 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1358 Step 2: Now simplify again, treating the constraint as 'free' if
1359 it does not mention qtvs, and trying to reduce it otherwise.
1360 The reasons for this is to maximise sharing.
1362 This fails for a very subtle reason. Suppose that in the Step 2
1363 a constraint (Foo (Succ Zero) (Succ Zero) b) gets thrown upstairs as 'free'.
1364 In the Step 1 this constraint might have been simplified, perhaps to
1365 (Foo Zero Zero b), AND THEN THAT MIGHT BE IMPROVED, to bind 'b' to 'T'.
1366 This won't happen in Step 2... but that in turn might prevent some other
1367 constraint (Baz [a] b) being simplified (e.g. via instance Baz [a] T where {..})
1368 and that in turn breaks the invariant that no constraints are quantified over.
1370 Test typecheck/should_compile/tc177 (which failed in GHC 6.2) demonstrates
1375 Step 1: Simplify the constraints as much as possible (to deal
1376 with Plan A's problem). Then set
1377 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1378 Return the bindings from Step 1.
1381 A note about Plan C (arising from "bug" reported by George Russel March 2004)
1384 instance (HasBinary ty IO) => HasCodedValue ty
1386 foo :: HasCodedValue a => String -> IO a
1388 doDecodeIO :: HasCodedValue a => () -> () -> IO a
1389 doDecodeIO codedValue view
1390 = let { act = foo "foo" } in act
1392 You might think this should work becuase the call to foo gives rise to a constraint
1393 (HasCodedValue t), which can be satisfied by the type sig for doDecodeIO. But the
1394 restricted binding act = ... calls tcSimplifyRestricted, and PlanC simplifies the
1395 constraint using the (rather bogus) instance declaration, and now we are stuffed.
1397 I claim this is not really a bug -- but it bit Sergey as well as George. So here's
1401 Plan D (a variant of plan B)
1402 Step 1: Simplify the constraints as much as possible (to deal
1403 with Plan A's problem), BUT DO NO IMPROVEMENT. Then set
1404 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1406 Step 2: Now simplify again, treating the constraint as 'free' if
1407 it does not mention qtvs, and trying to reduce it otherwise.
1409 The point here is that it's generally OK to have too few qtvs; that is,
1410 to make the thing more monomorphic than it could be. We don't want to
1411 do that in the common cases, but in wierd cases it's ok: the programmer
1412 can always add a signature.
1414 Too few qtvs => too many wanteds, which is what happens if you do less
1419 tcSimplifyRestricted -- Used for restricted binding groups
1420 -- i.e. ones subject to the monomorphism restriction
1423 -> [Name] -- Things bound in this group
1424 -> TcTyVarSet -- Free in the type of the RHSs
1425 -> [Inst] -- Free in the RHSs
1426 -> TcM ([TyVar], -- Tyvars to quantify (zonked and quantified)
1427 TcDictBinds) -- Bindings
1428 -- tcSimpifyRestricted returns no constraints to
1429 -- quantify over; by definition there are none.
1430 -- They are all thrown back in the LIE
1432 tcSimplifyRestricted doc top_lvl bndrs tau_tvs wanteds
1433 -- Zonk everything in sight
1434 = do { traceTc (text "tcSimplifyRestricted")
1435 ; wanteds_z <- zonkInsts wanteds
1437 -- 'ReduceMe': Reduce as far as we can. Don't stop at
1438 -- dicts; the idea is to get rid of as many type
1439 -- variables as possible, and we don't want to stop
1440 -- at (say) Monad (ST s), because that reduces
1441 -- immediately, with no constraint on s.
1443 -- BUT do no improvement! See Plan D above
1444 -- HOWEVER, some unification may take place, if we instantiate
1445 -- a method Inst with an equality constraint
1446 ; let env = mkNoImproveRedEnv doc (\_ -> ReduceMe)
1447 ; (_imp, _binds, constrained_dicts) <- reduceContext env wanteds_z
1449 -- Next, figure out the tyvars we will quantify over
1450 ; tau_tvs' <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
1451 ; gbl_tvs' <- tcGetGlobalTyVars
1452 ; constrained_dicts' <- zonkInsts constrained_dicts
1454 ; let qtvs1 = tau_tvs' `minusVarSet` oclose (fdPredsOfInsts constrained_dicts) gbl_tvs'
1455 -- As in tcSimplifyInfer
1457 -- Do not quantify over constrained type variables:
1458 -- this is the monomorphism restriction
1459 constrained_tvs' = tyVarsOfInsts constrained_dicts'
1460 qtvs = qtvs1 `minusVarSet` constrained_tvs'
1461 pp_bndrs = pprWithCommas (quotes . ppr) bndrs
1464 ; warn_mono <- doptM Opt_WarnMonomorphism
1465 ; warnTc (warn_mono && (constrained_tvs' `intersectsVarSet` qtvs1))
1466 (vcat[ ptext (sLit "the Monomorphism Restriction applies to the binding")
1467 <> plural bndrs <+> ptext (sLit "for") <+> pp_bndrs,
1468 ptext (sLit "Consider giving a type signature for") <+> pp_bndrs])
1470 ; traceTc (text "tcSimplifyRestricted" <+> vcat [
1471 pprInsts wanteds, pprInsts constrained_dicts',
1473 ppr constrained_tvs', ppr tau_tvs', ppr qtvs ])
1475 -- Zonk wanteds again! The first call to reduceContext may have
1476 -- instantiated some variables.
1477 -- FIXME: If red_improve would work, we could propagate that into
1478 -- the equality solver, too, to prevent instantating any
1480 ; wanteds_zz <- zonkInsts wanteds_z
1482 -- The first step may have squashed more methods than
1483 -- necessary, so try again, this time more gently, knowing the exact
1484 -- set of type variables to quantify over.
1486 -- We quantify only over constraints that are captured by qtvs;
1487 -- these will just be a subset of non-dicts. This in contrast
1488 -- to normal inference (using isFreeWhenInferring) in which we quantify over
1489 -- all *non-inheritable* constraints too. This implements choice
1490 -- (B) under "implicit parameter and monomorphism" above.
1492 -- Remember that we may need to do *some* simplification, to
1493 -- (for example) squash {Monad (ST s)} into {}. It's not enough
1494 -- just to float all constraints
1496 -- At top level, we *do* squash methods becuase we want to
1497 -- expose implicit parameters to the test that follows
1498 ; let is_nested_group = isNotTopLevel top_lvl
1499 try_me inst | isFreeWrtTyVars qtvs inst,
1500 (is_nested_group || isDict inst) = Stop
1501 | otherwise = ReduceMe
1502 env = mkNoImproveRedEnv doc try_me
1503 ; (_imp, binds, irreds) <- reduceContext env wanteds_zz
1505 -- See "Notes on implicit parameters, Question 4: top level"
1506 ; ASSERT( all (isFreeWrtTyVars qtvs) irreds ) -- None should be captured
1507 if is_nested_group then
1509 else do { let (bad_ips, non_ips) = partition isIPDict irreds
1510 ; addTopIPErrs bndrs bad_ips
1511 ; extendLIEs non_ips }
1513 ; qtvs' <- zonkQuantifiedTyVars (varSetElems qtvs)
1514 ; return (qtvs', binds) }
1518 %************************************************************************
1522 %************************************************************************
1524 On the LHS of transformation rules we only simplify methods and constants,
1525 getting dictionaries. We want to keep all of them unsimplified, to serve
1526 as the available stuff for the RHS of the rule.
1528 Example. Consider the following left-hand side of a rule
1530 f (x == y) (y > z) = ...
1532 If we typecheck this expression we get constraints
1534 d1 :: Ord a, d2 :: Eq a
1536 We do NOT want to "simplify" to the LHS
1538 forall x::a, y::a, z::a, d1::Ord a.
1539 f ((==) (eqFromOrd d1) x y) ((>) d1 y z) = ...
1543 forall x::a, y::a, z::a, d1::Ord a, d2::Eq a.
1544 f ((==) d2 x y) ((>) d1 y z) = ...
1546 Here is another example:
1548 fromIntegral :: (Integral a, Num b) => a -> b
1549 {-# RULES "foo" fromIntegral = id :: Int -> Int #-}
1551 In the rule, a=b=Int, and Num Int is a superclass of Integral Int. But
1552 we *dont* want to get
1554 forall dIntegralInt.
1555 fromIntegral Int Int dIntegralInt (scsel dIntegralInt) = id Int
1557 because the scsel will mess up RULE matching. Instead we want
1559 forall dIntegralInt, dNumInt.
1560 fromIntegral Int Int dIntegralInt dNumInt = id Int
1564 g (x == y) (y == z) = ..
1566 where the two dictionaries are *identical*, we do NOT WANT
1568 forall x::a, y::a, z::a, d1::Eq a
1569 f ((==) d1 x y) ((>) d1 y z) = ...
1571 because that will only match if the dict args are (visibly) equal.
1572 Instead we want to quantify over the dictionaries separately.
1574 In short, tcSimplifyRuleLhs must *only* squash LitInst and MethInts, leaving
1575 all dicts unchanged, with absolutely no sharing. It's simpler to do this
1576 from scratch, rather than further parameterise simpleReduceLoop etc.
1577 Simpler, maybe, but alas not simple (see Trac #2494)
1579 * Type errors may give rise to an (unsatisfiable) equality constraint
1581 * Applications of a higher-rank function on the LHS may give
1582 rise to an implication constraint, esp if there are unsatisfiable
1583 equality constraints inside.
1586 tcSimplifyRuleLhs :: [Inst] -> TcM ([Inst], TcDictBinds)
1587 tcSimplifyRuleLhs wanteds
1588 = do { wanteds' <- zonkInsts wanteds
1589 ; (irreds, binds) <- go [] emptyBag wanteds'
1590 ; let (dicts, bad_irreds) = partition isDict irreds
1591 ; traceTc (text "tcSimplifyrulelhs" <+> pprInsts bad_irreds)
1592 ; addNoInstanceErrs (nub bad_irreds)
1593 -- The nub removes duplicates, which has
1594 -- not happened otherwise (see notes above)
1595 ; return (dicts, binds) }
1597 go :: [Inst] -> TcDictBinds -> [Inst] -> TcM ([Inst], TcDictBinds)
1599 = return (irreds, binds)
1600 go irreds binds (w:ws)
1602 = go (w:irreds) binds ws
1603 | isImplicInst w -- Have a go at reducing the implication
1604 = do { (binds1, irreds1) <- reduceImplication red_env w
1605 ; let (bad_irreds, ok_irreds) = partition isImplicInst irreds1
1606 ; go (bad_irreds ++ irreds)
1607 (binds `unionBags` binds1)
1610 = do { w' <- zonkInst w -- So that (3::Int) does not generate a call
1611 -- to fromInteger; this looks fragile to me
1612 ; lookup_result <- lookupSimpleInst w'
1613 ; case lookup_result of
1614 NoInstance -> go (w:irreds) binds ws
1615 GenInst ws' rhs -> go irreds binds' (ws' ++ ws)
1617 binds' = addInstToDictBind binds w rhs
1620 -- Sigh: we need to reduce inside implications
1621 red_env = mkInferRedEnv doc try_me
1622 doc = ptext (sLit "Implication constraint in RULE lhs")
1623 try_me inst | isMethodOrLit inst = ReduceMe
1624 | otherwise = Stop -- Be gentle
1627 tcSimplifyBracket is used when simplifying the constraints arising from
1628 a Template Haskell bracket [| ... |]. We want to check that there aren't
1629 any constraints that can't be satisfied (e.g. Show Foo, where Foo has no
1630 Show instance), but we aren't otherwise interested in the results.
1631 Nor do we care about ambiguous dictionaries etc. We will type check
1632 this bracket again at its usage site.
1635 tcSimplifyBracket :: [Inst] -> TcM ()
1636 tcSimplifyBracket wanteds
1637 = do { tryHardCheckLoop doc wanteds
1640 doc = text "tcSimplifyBracket"
1644 %************************************************************************
1646 \subsection{Filtering at a dynamic binding}
1648 %************************************************************************
1653 we must discharge all the ?x constraints from B. We also do an improvement
1654 step; if we have ?x::t1 and ?x::t2 we must unify t1, t2.
1656 Actually, the constraints from B might improve the types in ?x. For example
1658 f :: (?x::Int) => Char -> Char
1661 then the constraint (?x::Int) arising from the call to f will
1662 force the binding for ?x to be of type Int.
1665 tcSimplifyIPs :: [Inst] -- The implicit parameters bound here
1668 -- We need a loop so that we do improvement, and then
1669 -- (next time round) generate a binding to connect the two
1671 -- Here the two ?x's have different types, and improvement
1672 -- makes them the same.
1674 tcSimplifyIPs given_ips wanteds
1675 = do { wanteds' <- zonkInsts wanteds
1676 ; given_ips' <- zonkInsts given_ips
1677 -- Unusually for checking, we *must* zonk the given_ips
1679 ; let env = mkRedEnv doc try_me given_ips'
1680 ; (improved, binds, irreds) <- reduceContext env wanteds'
1682 ; if null irreds || not improved then
1683 ASSERT( all is_free irreds )
1684 do { extendLIEs irreds
1687 -- If improvement did some unification, we go round again.
1688 -- We start again with irreds, not wanteds
1689 -- Using an instance decl might have introduced a fresh type
1690 -- variable which might have been unified, so we'd get an
1691 -- infinite loop if we started again with wanteds!
1693 { binds1 <- tcSimplifyIPs given_ips' irreds
1694 ; return $ binds `unionBags` binds1
1697 doc = text "tcSimplifyIPs" <+> ppr given_ips
1698 ip_set = mkNameSet (ipNamesOfInsts given_ips)
1699 is_free inst = isFreeWrtIPs ip_set inst
1701 -- Simplify any methods that mention the implicit parameter
1702 try_me inst | is_free inst = Stop
1703 | otherwise = ReduceMe
1707 %************************************************************************
1709 \subsection[binds-for-local-funs]{@bindInstsOfLocalFuns@}
1711 %************************************************************************
1713 When doing a binding group, we may have @Insts@ of local functions.
1714 For example, we might have...
1716 let f x = x + 1 -- orig local function (overloaded)
1717 f.1 = f Int -- two instances of f
1722 The point is: we must drop the bindings for @f.1@ and @f.2@ here,
1723 where @f@ is in scope; those @Insts@ must certainly not be passed
1724 upwards towards the top-level. If the @Insts@ were binding-ified up
1725 there, they would have unresolvable references to @f@.
1727 We pass in an @init_lie@ of @Insts@ and a list of locally-bound @Ids@.
1728 For each method @Inst@ in the @init_lie@ that mentions one of the
1729 @Ids@, we create a binding. We return the remaining @Insts@ (in an
1730 @LIE@), as well as the @HsBinds@ generated.
1733 bindInstsOfLocalFuns :: [Inst] -> [TcId] -> TcM TcDictBinds
1734 -- Simlifies only MethodInsts, and generate only bindings of form
1736 -- We're careful not to even generate bindings of the form
1738 -- You'd think that'd be fine, but it interacts with what is
1739 -- arguably a bug in Match.tidyEqnInfo (see notes there)
1741 bindInstsOfLocalFuns wanteds local_ids
1742 | null overloaded_ids = do
1745 return emptyLHsBinds
1748 = do { (irreds, binds) <- gentleInferLoop doc for_me
1749 ; extendLIEs not_for_me
1753 doc = text "bindInsts" <+> ppr local_ids
1754 overloaded_ids = filter is_overloaded local_ids
1755 is_overloaded id = isOverloadedTy (idType id)
1756 (for_me, not_for_me) = partition (isMethodFor overloaded_set) wanteds
1758 overloaded_set = mkVarSet overloaded_ids -- There can occasionally be a lot of them
1759 -- so it's worth building a set, so that
1760 -- lookup (in isMethodFor) is faster
1764 %************************************************************************
1766 \subsection{Data types for the reduction mechanism}
1768 %************************************************************************
1770 The main control over context reduction is here
1774 = RedEnv { red_doc :: SDoc -- The context
1775 , red_try_me :: Inst -> WhatToDo
1776 , red_improve :: Bool -- True <=> do improvement
1777 , red_givens :: [Inst] -- All guaranteed rigid
1778 -- Always dicts & equalities
1779 -- but see Note [Rigidity]
1781 , red_stack :: (Int, [Inst]) -- Recursion stack (for err msg)
1782 -- See Note [RedStack]
1786 -- The red_givens are rigid so far as cmpInst is concerned.
1787 -- There is one case where they are not totally rigid, namely in tcSimplifyIPs
1788 -- let ?x = e in ...
1789 -- Here, the given is (?x::a), where 'a' is not necy a rigid type
1790 -- But that doesn't affect the comparison, which is based only on mame.
1793 -- The red_stack pair (n,insts) pair is just used for error reporting.
1794 -- 'n' is always the depth of the stack.
1795 -- The 'insts' is the stack of Insts being reduced: to produce X
1796 -- I had to produce Y, to produce Y I had to produce Z, and so on.
1799 mkRedEnv :: SDoc -> (Inst -> WhatToDo) -> [Inst] -> RedEnv
1800 mkRedEnv doc try_me givens
1801 = RedEnv { red_doc = doc, red_try_me = try_me,
1802 red_givens = givens,
1804 red_improve = True }
1806 mkInferRedEnv :: SDoc -> (Inst -> WhatToDo) -> RedEnv
1808 mkInferRedEnv doc try_me
1809 = RedEnv { red_doc = doc, red_try_me = try_me,
1812 red_improve = True }
1814 mkNoImproveRedEnv :: SDoc -> (Inst -> WhatToDo) -> RedEnv
1815 -- Do not do improvement; no givens
1816 mkNoImproveRedEnv doc try_me
1817 = RedEnv { red_doc = doc, red_try_me = try_me,
1820 red_improve = True }
1823 = ReduceMe -- Try to reduce this
1824 -- If there's no instance, add the inst to the
1825 -- irreductible ones, but don't produce an error
1826 -- message of any kind.
1827 -- It might be quite legitimate such as (Eq a)!
1829 | Stop -- Return as irreducible unless it can
1830 -- be reduced to a constant in one step
1831 -- Do not add superclasses; see
1833 data WantSCs = NoSCs | AddSCs -- Tells whether we should add the superclasses
1834 -- of a predicate when adding it to the avails
1835 -- The reason for this flag is entirely the super-class loop problem
1836 -- Note [SUPER-CLASS LOOP 1]
1838 zonkRedEnv :: RedEnv -> TcM RedEnv
1840 = do { givens' <- mapM zonkInst (red_givens env)
1841 ; return $ env {red_givens = givens'}
1846 %************************************************************************
1848 \subsection[reduce]{@reduce@}
1850 %************************************************************************
1852 Note [Ancestor Equalities]
1853 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1854 During context reduction, we add to the wanted equalities also those
1855 equalities that (transitively) occur in superclass contexts of wanted
1856 class constraints. Consider the following code
1858 class a ~ Int => C a
1861 If (C a) is wanted, we want to add (a ~ Int), which will be discharged by
1862 substituting Int for a. Hence, we ultimately want (C Int), which we
1863 discharge with the explicit instance.
1866 reduceContext :: RedEnv
1868 -> TcM (ImprovementDone,
1869 TcDictBinds, -- Dictionary bindings
1870 [Inst]) -- Irreducible
1872 reduceContext env wanteds0
1873 = do { traceTc (text "reduceContext" <+> (vcat [
1874 text "----------------------",
1876 text "given" <+> ppr (red_givens env),
1877 text "wanted" <+> ppr wanteds0,
1878 text "----------------------"
1881 -- We want to add as wanted equalities those that (transitively)
1882 -- occur in superclass contexts of wanted class constraints.
1883 -- See Note [Ancestor Equalities]
1884 ; ancestor_eqs <- ancestorEqualities wanteds0
1885 ; traceTc $ text "reduceContext: ancestor eqs" <+> ppr ancestor_eqs
1887 -- Normalise and solve all equality constraints as far as possible
1888 -- and normalise all dictionary constraints wrt to the reduced
1889 -- equalities. The returned wanted constraints include the
1890 -- irreducible wanted equalities.
1891 ; let wanteds = wanteds0 ++ ancestor_eqs
1892 givens = red_givens env
1896 eq_improved) <- tcReduceEqs givens wanteds
1897 ; traceTc $ text "reduceContext: tcReduceEqs result" <+> vcat
1898 [ppr givens', ppr wanteds', ppr normalise_binds]
1900 -- Build the Avail mapping from "given_dicts"
1901 ; (init_state, _) <- getLIE $ do
1902 { init_state <- foldlM addGiven emptyAvails givens'
1906 -- Solve the *wanted* *dictionary* constraints (not implications)
1907 -- This may expose some further equational constraints in the course
1908 -- of improvement due to functional dependencies if any of the
1909 -- involved unifications gets deferred.
1910 ; let (wanted_implics, wanted_dicts) = partition isImplicInst wanteds'
1911 ; (avails, extra_eqs) <- getLIE (reduceList env wanted_dicts init_state)
1912 -- The getLIE is reqd because reduceList does improvement
1913 -- (via extendAvails) which may in turn do unification
1916 dict_irreds) <- extractResults avails wanted_dicts
1917 ; traceTc $ text "reduceContext: extractResults" <+> vcat
1918 [ppr avails, ppr wanted_dicts, ppr dict_binds]
1920 -- Solve the wanted *implications*. In doing so, we can provide
1921 -- as "given" all the dicts that were originally given,
1922 -- *or* for which we now have bindings,
1923 -- *or* which are now irreds
1924 -- NB: Equality irreds need to be converted, as the recursive
1925 -- invocation of the solver will still treat them as wanteds
1927 ; let implic_env = env { red_givens
1928 = givens ++ bound_dicts ++
1929 map wantedToLocalEqInst dict_irreds }
1930 ; (implic_binds_s, implic_irreds_s)
1931 <- mapAndUnzipM (reduceImplication implic_env) wanted_implics
1932 ; let implic_binds = unionManyBags implic_binds_s
1933 implic_irreds = concat implic_irreds_s
1935 -- Collect all irreducible instances, and determine whether we should
1936 -- go round again. We do so in either of two cases:
1937 -- (1) If dictionary reduction or equality solving led to
1938 -- improvement (i.e., instantiated type variables).
1939 -- (2) If we reduced dictionaries (i.e., got dictionary bindings),
1940 -- they may have exposed further opportunities to normalise
1941 -- family applications. See Note [Dictionary Improvement]
1943 -- NB: We do *not* go around for new extra_eqs. Morally, we should,
1944 -- but we can't without risking non-termination (see #2688). By
1945 -- not going around, we miss some legal programs mixing FDs and
1946 -- TFs, but we never claimed to support such programs in the
1947 -- current implementation anyway.
1949 ; let all_irreds = dict_irreds ++ implic_irreds ++ extra_eqs
1950 avails_improved = availsImproved avails
1951 improvedFlexible = avails_improved || eq_improved
1952 reduced_dicts = not (isEmptyBag dict_binds)
1953 improved = improvedFlexible || reduced_dicts
1955 improvedHint = (if avails_improved then " [AVAILS]" else "") ++
1956 (if eq_improved then " [EQ]" else "")
1958 ; traceTc (text "reduceContext end" <+> (vcat [
1959 text "----------------------",
1961 text "given" <+> ppr givens,
1962 text "wanted" <+> ppr wanteds0,
1964 text "avails" <+> pprAvails avails,
1965 text "improved =" <+> ppr improved <+> text improvedHint,
1966 text "(all) irreds = " <+> ppr all_irreds,
1967 text "dict-binds = " <+> ppr dict_binds,
1968 text "implic-binds = " <+> ppr implic_binds,
1969 text "----------------------"
1973 normalise_binds `unionBags` dict_binds
1974 `unionBags` implic_binds,
1978 tcImproveOne :: Avails -> Inst -> TcM ImprovementDone
1979 tcImproveOne avails inst
1980 | not (isDict inst) = return False
1982 = do { inst_envs <- tcGetInstEnvs
1983 ; let eqns = improveOne (classInstances inst_envs)
1984 (dictPred inst, pprInstArising inst)
1985 [ (dictPred p, pprInstArising p)
1986 | p <- availsInsts avails, isDict p ]
1987 -- Avails has all the superclasses etc (good)
1988 -- It also has all the intermediates of the deduction (good)
1989 -- It does not have duplicates (good)
1990 -- NB that (?x::t1) and (?x::t2) will be held separately in
1991 -- avails so that improve will see them separate
1992 ; traceTc (text "improveOne" <+> ppr inst)
1995 unifyEqns :: [(Equation, (PredType, SDoc), (PredType, SDoc))]
1996 -> TcM ImprovementDone
1997 unifyEqns [] = return False
1999 = do { traceTc (ptext (sLit "Improve:") <+> vcat (map pprEquationDoc eqns))
2000 ; improved <- mapM unify eqns
2001 ; return $ or improved
2004 unify ((qtvs, pairs), what1, what2)
2005 = addErrCtxtM (mkEqnMsg what1 what2) $
2006 do { let freeTyVars = unionVarSets (map tvs_pr pairs)
2008 ; (_, _, tenv) <- tcInstTyVars (varSetElems qtvs)
2009 ; mapM_ (unif_pr tenv) pairs
2010 ; anyM isFilledMetaTyVar $ varSetElems freeTyVars
2013 unif_pr tenv (ty1, ty2) = unifyType (substTy tenv ty1) (substTy tenv ty2)
2015 tvs_pr (ty1, ty2) = tyVarsOfType ty1 `unionVarSet` tyVarsOfType ty2
2017 pprEquationDoc :: (Equation, (PredType, SDoc), (PredType, SDoc)) -> SDoc
2018 pprEquationDoc (eqn, (p1, _), (p2, _))
2019 = vcat [pprEquation eqn, nest 2 (ppr p1), nest 2 (ppr p2)]
2021 mkEqnMsg :: (TcPredType, SDoc) -> (TcPredType, SDoc) -> TidyEnv
2022 -> TcM (TidyEnv, SDoc)
2023 mkEqnMsg (pred1,from1) (pred2,from2) tidy_env
2024 = do { pred1' <- zonkTcPredType pred1
2025 ; pred2' <- zonkTcPredType pred2
2026 ; let { pred1'' = tidyPred tidy_env pred1'
2027 ; pred2'' = tidyPred tidy_env pred2' }
2028 ; let msg = vcat [ptext (sLit "When using functional dependencies to combine"),
2029 nest 2 (sep [ppr pred1'' <> comma, nest 2 from1]),
2030 nest 2 (sep [ppr pred2'' <> comma, nest 2 from2])]
2031 ; return (tidy_env, msg) }
2034 Note [Dictionary Improvement]
2035 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2036 In reduceContext, we first reduce equalities and then class constraints.
2037 However, the letter may expose further opportunities for the former. Hence,
2038 we need to go around again if dictionary reduction produced any dictionary
2039 bindings. The following example demonstrated the point:
2041 data EX _x _y (p :: * -> *)
2046 class Base (Def p) => Prop p where
2050 instance Prop () where
2053 instance (Base (Def (p ANY))) => Base (EX _x _y p)
2054 instance (Prop (p ANY)) => Prop (EX _x _y p) where
2055 type Def (EX _x _y p) = EX _x _y p
2058 instance Prop (FOO x) where
2059 type Def (FOO x) = ()
2062 instance Prop BAR where
2063 type Def BAR = EX () () FOO
2065 During checking the last instance declaration, we need to check the superclass
2066 cosntraint Base (Def BAR), which family normalisation reduced to
2067 Base (EX () () FOO). Chasing the instance for Base (EX _x _y p), gives us
2068 Base (Def (FOO ANY)), which again requires family normalisation of Def to
2069 Base () before we can finish.
2072 The main context-reduction function is @reduce@. Here's its game plan.
2075 reduceList :: RedEnv -> [Inst] -> Avails -> TcM Avails
2076 reduceList env@(RedEnv {red_stack = (n,stk)}) wanteds state
2077 = do { traceTc (text "reduceList " <+> (ppr wanteds $$ ppr state))
2079 ; when (debugIsOn && (n > 8)) $ do
2080 debugDumpTcRn (hang (ptext (sLit "Interesting! Context reduction stack depth") <+> int n)
2081 2 (ifPprDebug (nest 2 (pprStack stk))))
2082 ; if n >= ctxtStkDepth dopts then
2083 failWithTc (reduceDepthErr n stk)
2087 go [] state = return state
2088 go (w:ws) state = do { state' <- reduce (env {red_stack = (n+1, w:stk)}) w state
2091 -- Base case: we're done!
2092 reduce :: RedEnv -> Inst -> Avails -> TcM Avails
2093 reduce env wanted avails
2095 -- We don't reduce equalities here (and they must not end up as irreds
2100 -- It's the same as an existing inst, or a superclass thereof
2101 | Just _ <- findAvail avails wanted
2102 = do { traceTc (text "reduce: found " <+> ppr wanted)
2107 = do { traceTc (text "reduce" <+> ppr wanted $$ ppr avails)
2108 ; case red_try_me env wanted of {
2109 Stop -> try_simple (addIrred NoSCs);
2110 -- See Note [No superclasses for Stop]
2112 ReduceMe -> do -- It should be reduced
2113 { (avails, lookup_result) <- reduceInst env avails wanted
2114 ; case lookup_result of
2115 NoInstance -> addIrred AddSCs avails wanted
2116 -- Add it and its superclasses
2118 GenInst [] rhs -> addWanted AddSCs avails wanted rhs []
2120 GenInst wanteds' rhs
2121 -> do { avails1 <- addIrred NoSCs avails wanted
2122 ; avails2 <- reduceList env wanteds' avails1
2123 ; addWanted AddSCs avails2 wanted rhs wanteds' } }
2124 -- Temporarily do addIrred *before* the reduceList,
2125 -- which has the effect of adding the thing we are trying
2126 -- to prove to the database before trying to prove the things it
2127 -- needs. See note [RECURSIVE DICTIONARIES]
2128 -- NB: we must not do an addWanted before, because that adds the
2129 -- superclasses too, and that can lead to a spurious loop; see
2130 -- the examples in [SUPERCLASS-LOOP]
2131 -- So we do an addIrred before, and then overwrite it afterwards with addWanted
2134 -- First, see if the inst can be reduced to a constant in one step
2135 -- Works well for literals (1::Int) and constant dictionaries (d::Num Int)
2136 -- Don't bother for implication constraints, which take real work
2137 try_simple do_this_otherwise
2138 = do { res <- lookupSimpleInst wanted
2140 GenInst [] rhs -> addWanted AddSCs avails wanted rhs []
2141 _ -> do_this_otherwise avails wanted }
2145 Note [RECURSIVE DICTIONARIES]
2146 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2148 data D r = ZeroD | SuccD (r (D r));
2150 instance (Eq (r (D r))) => Eq (D r) where
2151 ZeroD == ZeroD = True
2152 (SuccD a) == (SuccD b) = a == b
2155 equalDC :: D [] -> D [] -> Bool;
2158 We need to prove (Eq (D [])). Here's how we go:
2162 by instance decl, holds if
2166 by instance decl of Eq, holds if
2168 where d2 = dfEqList d3
2171 But now we can "tie the knot" to give
2177 and it'll even run! The trick is to put the thing we are trying to prove
2178 (in this case Eq (D []) into the database before trying to prove its
2179 contributing clauses.
2181 Note [SUPERCLASS-LOOP 2]
2182 ~~~~~~~~~~~~~~~~~~~~~~~~
2183 We need to be careful when adding "the constaint we are trying to prove".
2184 Suppose we are *given* d1:Ord a, and want to deduce (d2:C [a]) where
2186 class Ord a => C a where
2187 instance Ord [a] => C [a] where ...
2189 Then we'll use the instance decl to deduce C [a] from Ord [a], and then add the
2190 superclasses of C [a] to avails. But we must not overwrite the binding
2191 for Ord [a] (which is obtained from Ord a) with a superclass selection or we'll just
2194 Here's another variant, immortalised in tcrun020
2195 class Monad m => C1 m
2196 class C1 m => C2 m x
2197 instance C2 Maybe Bool
2198 For the instance decl we need to build (C1 Maybe), and it's no good if
2199 we run around and add (C2 Maybe Bool) and its superclasses to the avails
2200 before we search for C1 Maybe.
2202 Here's another example
2203 class Eq b => Foo a b
2204 instance Eq a => Foo [a] a
2208 we'll first deduce that it holds (via the instance decl). We must not
2209 then overwrite the Eq t constraint with a superclass selection!
2211 At first I had a gross hack, whereby I simply did not add superclass constraints
2212 in addWanted, though I did for addGiven and addIrred. This was sub-optimal,
2213 becuase it lost legitimate superclass sharing, and it still didn't do the job:
2214 I found a very obscure program (now tcrun021) in which improvement meant the
2215 simplifier got two bites a the cherry... so something seemed to be an Stop
2216 first time, but reducible next time.
2218 Now we implement the Right Solution, which is to check for loops directly
2219 when adding superclasses. It's a bit like the occurs check in unification.
2223 %************************************************************************
2225 Reducing a single constraint
2227 %************************************************************************
2230 ---------------------------------------------
2231 reduceInst :: RedEnv -> Avails -> Inst -> TcM (Avails, LookupInstResult)
2232 reduceInst _ avails other_inst
2233 = do { result <- lookupSimpleInst other_inst
2234 ; return (avails, result) }
2237 Note [Equational Constraints in Implication Constraints]
2238 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2240 An implication constraint is of the form
2242 where Given and Wanted may contain both equational and dictionary
2243 constraints. The delay and reduction of these two kinds of constraints
2246 -) In the generated code, wanted Dictionary constraints are wrapped up in an
2247 implication constraint that is created at the code site where the wanted
2248 dictionaries can be reduced via a let-binding. This let-bound implication
2249 constraint is deconstructed at the use-site of the wanted dictionaries.
2251 -) While the reduction of equational constraints is also delayed, the delay
2252 is not manifest in the generated code. The required evidence is generated
2253 in the code directly at the use-site. There is no let-binding and deconstruction
2254 necessary. The main disadvantage is that we cannot exploit sharing as the
2255 same evidence may be generated at multiple use-sites. However, this disadvantage
2256 is limited because it only concerns coercions which are erased.
2258 The different treatment is motivated by the different in representation. Dictionary
2259 constraints require manifest runtime dictionaries, while equations require coercions
2263 ---------------------------------------------
2264 reduceImplication :: RedEnv
2266 -> TcM (TcDictBinds, [Inst])
2269 Suppose we are simplifying the constraint
2270 forall bs. extras => wanted
2271 in the context of an overall simplification problem with givens 'givens'.
2274 * The 'givens' need not mention any of the quantified type variables
2275 e.g. forall {}. Eq a => Eq [a]
2276 forall {}. C Int => D (Tree Int)
2278 This happens when you have something like
2280 T1 :: Eq a => a -> T a
2283 f x = ...(case x of { T1 v -> v==v })...
2286 -- ToDo: should we instantiate tvs? I think it's not necessary
2288 -- Note on coercion variables:
2290 -- The extra given coercion variables are bound at two different
2293 -- -) in the creation context of the implication constraint
2294 -- the solved equational constraints use these binders
2296 -- -) at the solving site of the implication constraint
2297 -- the solved dictionaries use these binders;
2298 -- these binders are generated by reduceImplication
2300 -- Note [Binders for equalities]
2301 -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2302 -- To reuse the binders of local/given equalities in the binders of
2303 -- implication constraints, it is crucial that these given equalities
2304 -- always have the form
2306 -- where cotv is a simple coercion type variable (and not a more
2307 -- complex coercion term). We require that the extra_givens always
2308 -- have this form and exploit the special form when generating binders.
2309 reduceImplication env
2310 orig_implic@(ImplicInst { tci_name = name, tci_loc = inst_loc,
2312 tci_given = extra_givens, tci_wanted = wanteds
2314 = do { -- Solve the sub-problem
2315 ; let try_me _ = ReduceMe -- Note [Freeness and implications]
2316 env' = env { red_givens = extra_givens ++ red_givens env
2317 , red_doc = sep [ptext (sLit "reduceImplication for")
2319 nest 2 (parens $ ptext (sLit "within")
2321 , red_try_me = try_me }
2323 ; traceTc (text "reduceImplication" <+> vcat
2324 [ ppr (red_givens env), ppr extra_givens,
2326 ; (irreds, binds) <- checkLoop env' wanteds
2328 ; traceTc (text "reduceImplication result" <+> vcat
2329 [ppr irreds, ppr binds])
2331 ; -- extract superclass binds
2332 -- (sc_binds,_) <- extractResults avails []
2333 -- ; traceTc (text "reduceImplication sc_binds" <+> vcat
2334 -- [ppr sc_binds, ppr avails])
2337 -- SLPJ Sept 07: what if improvement happened inside the checkLoop?
2338 -- Then we must iterate the outer loop too!
2340 ; didntSolveWantedEqs <- allM wantedEqInstIsUnsolved wanteds
2341 -- we solve wanted eqs by side effect!
2343 -- Progress is no longer measered by the number of bindings
2344 -- If there are any irreds, but no bindings and no solved
2345 -- equalities, we back off and do nothing
2346 ; let backOff = isEmptyLHsBinds binds && -- no new bindings
2347 (not $ null irreds) && -- but still some irreds
2348 didntSolveWantedEqs -- no instantiated cotv
2350 ; if backOff then -- No progress
2351 return (emptyBag, [orig_implic])
2353 { (simpler_implic_insts, bind)
2354 <- makeImplicationBind inst_loc tvs extra_givens irreds
2355 -- This binding is useless if the recursive simplification
2356 -- made no progress; but currently we don't try to optimise that
2357 -- case. After all, we only try hard to reduce at top level, or
2358 -- when inferring types.
2360 ; let -- extract Id binders for dicts and CoTyVar binders for eqs;
2361 -- see Note [Binders for equalities]
2362 (extra_eq_givens, extra_dict_givens) = partition isEqInst
2364 eq_cotvs = map instToVar extra_eq_givens
2365 dict_ids = map instToId extra_dict_givens
2367 -- Note [Always inline implication constraints]
2368 wrap_inline | null dict_ids = idHsWrapper
2369 | otherwise = WpInline
2372 <.> mkWpTyLams eq_cotvs
2373 <.> mkWpLams dict_ids
2374 <.> WpLet (binds `unionBags` bind)
2375 rhs = mkLHsWrap co payload
2376 loc = instLocSpan inst_loc
2377 -- wanted equalities are solved by updating their
2378 -- cotv; we don't generate bindings for them
2379 dict_bndrs = map (L loc . HsVar . instToId)
2380 . filter (not . isEqInst)
2382 payload = mkBigLHsTup dict_bndrs
2385 ; traceTc (vcat [text "reduceImplication" <+> ppr name,
2386 ppr simpler_implic_insts,
2387 text "->" <+> ppr rhs])
2388 ; return (unitBag (L loc (VarBind (instToId orig_implic) rhs)),
2389 simpler_implic_insts)
2392 reduceImplication _ i = pprPanic "reduceImplication" (ppr i)
2395 Note [Always inline implication constraints]
2396 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2397 Suppose an implication constraint floats out of an INLINE function.
2398 Then although the implication has a single call site, it won't be
2399 inlined. And that is bad because it means that even if there is really
2400 *no* overloading (type signatures specify the exact types) there will
2401 still be dictionary passing in the resulting code. To avert this,
2402 we mark the implication constraints themselves as INLINE, at least when
2403 there is no loss of sharing as a result.
2405 Note [Freeness and implications]
2406 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2407 It's hard to say when an implication constraint can be floated out. Consider
2408 forall {} Eq a => Foo [a]
2409 The (Foo [a]) doesn't mention any of the quantified variables, but it
2410 still might be partially satisfied by the (Eq a).
2412 There is a useful special case when it *is* easy to partition the
2413 constraints, namely when there are no 'givens'. Consider
2414 forall {a}. () => Bar b
2415 There are no 'givens', and so there is no reason to capture (Bar b).
2416 We can let it float out. But if there is even one constraint we
2417 must be much more careful:
2418 forall {a}. C a b => Bar (m b)
2419 because (C a b) might have a superclass (D b), from which we might
2420 deduce (Bar [b]) when m later gets instantiated to []. Ha!
2422 Here is an even more exotic example
2424 Now consider the constraint
2425 forall b. D Int b => C Int
2426 We can satisfy the (C Int) from the superclass of D, so we don't want
2427 to float the (C Int) out, even though it mentions no type variable in
2430 One more example: the constraint
2432 instance (C a, E c) => E (a,c)
2434 constraint: forall b. D Int b => E (Int,c)
2436 You might think that the (D Int b) can't possibly contribute
2437 to solving (E (Int,c)), since the latter mentions 'c'. But
2438 in fact it can, because solving the (E (Int,c)) constraint needs
2441 and the (C Int) can be satisfied from the superclass of (D Int b).
2442 So we must still not float (E (Int,c)) out.
2444 To think about: special cases for unary type classes?
2446 Note [Pruning the givens in an implication constraint]
2447 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2448 Suppose we are about to form the implication constraint
2449 forall tvs. Eq a => Ord b
2450 The (Eq a) cannot contribute to the (Ord b), because it has no access to
2451 the type variable 'b'. So we could filter out the (Eq a) from the givens.
2452 But BE CAREFUL of the examples above in [Freeness and implications].
2454 Doing so would be a bit tidier, but all the implication constraints get
2455 simplified away by the optimiser, so it's no great win. So I don't take
2456 advantage of that at the moment.
2458 If you do, BE CAREFUL of wobbly type variables.
2461 %************************************************************************
2463 Avails and AvailHow: the pool of evidence
2465 %************************************************************************
2469 data Avails = Avails !ImprovementDone !AvailEnv
2471 type ImprovementDone = Bool -- True <=> some unification has happened
2472 -- so some Irreds might now be reducible
2473 -- keys that are now
2475 type AvailEnv = FiniteMap Inst AvailHow
2477 = IsIrred -- Used for irreducible dictionaries,
2478 -- which are going to be lambda bound
2480 | Given Inst -- Used for dictionaries for which we have a binding
2481 -- e.g. those "given" in a signature
2483 | Rhs -- Used when there is a RHS
2484 (LHsExpr TcId) -- The RHS
2485 [Inst] -- Insts free in the RHS; we need these too
2487 instance Outputable Avails where
2490 pprAvails :: Avails -> SDoc
2491 pprAvails (Avails imp avails)
2492 = vcat [ ptext (sLit "Avails") <> (if imp then ptext (sLit "[improved]") else empty)
2494 vcat [ sep [ppr inst, nest 2 (equals <+> ppr avail)]
2495 | (inst,avail) <- fmToList avails ]]
2497 instance Outputable AvailHow where
2500 -------------------------
2501 pprAvail :: AvailHow -> SDoc
2502 pprAvail IsIrred = text "Irred"
2503 pprAvail (Given x) = text "Given" <+> ppr x
2504 pprAvail (Rhs rhs bs) = sep [text "Rhs" <+> ppr bs,
2507 -------------------------
2508 extendAvailEnv :: AvailEnv -> Inst -> AvailHow -> AvailEnv
2509 extendAvailEnv env inst avail = addToFM env inst avail
2511 findAvailEnv :: AvailEnv -> Inst -> Maybe AvailHow
2512 findAvailEnv env wanted = lookupFM env wanted
2513 -- NB 1: the Ord instance of Inst compares by the class/type info
2514 -- *not* by unique. So
2515 -- d1::C Int == d2::C Int
2517 emptyAvails :: Avails
2518 emptyAvails = Avails False emptyFM
2520 findAvail :: Avails -> Inst -> Maybe AvailHow
2521 findAvail (Avails _ avails) wanted = findAvailEnv avails wanted
2523 elemAvails :: Inst -> Avails -> Bool
2524 elemAvails wanted (Avails _ avails) = wanted `elemFM` avails
2526 extendAvails :: Avails -> Inst -> AvailHow -> TcM Avails
2528 extendAvails avails@(Avails imp env) inst avail
2529 = do { imp1 <- tcImproveOne avails inst -- Do any improvement
2530 ; return (Avails (imp || imp1) (extendAvailEnv env inst avail)) }
2532 availsInsts :: Avails -> [Inst]
2533 availsInsts (Avails _ avails) = keysFM avails
2535 availsImproved :: Avails -> ImprovementDone
2536 availsImproved (Avails imp _) = imp
2539 Extracting the bindings from a bunch of Avails.
2540 The bindings do *not* come back sorted in dependency order.
2541 We assume that they'll be wrapped in a big Rec, so that the
2542 dependency analyser can sort them out later
2545 type DoneEnv = FiniteMap Inst [Id]
2546 -- Tracks which things we have evidence for
2548 extractResults :: Avails
2550 -> TcM (TcDictBinds, -- Bindings
2551 [Inst], -- The insts bound by the bindings
2552 [Inst]) -- Irreducible ones
2553 -- Note [Reducing implication constraints]
2555 extractResults (Avails _ avails) wanteds
2556 = go emptyBag [] [] emptyFM wanteds
2558 go :: TcDictBinds -- Bindings for dicts
2559 -> [Inst] -- Bound by the bindings
2561 -> DoneEnv -- Has an entry for each inst in the above three sets
2563 -> TcM (TcDictBinds, [Inst], [Inst])
2564 go binds bound_dicts irreds _ []
2565 = return (binds, bound_dicts, irreds)
2567 go binds bound_dicts irreds done (w:ws)
2569 = go binds bound_dicts (w:irreds) done' ws
2571 | Just done_ids@(done_id : rest_done_ids) <- lookupFM done w
2572 = if w_id `elem` done_ids then
2573 go binds bound_dicts irreds done ws
2575 go (add_bind (nlHsVar done_id)) bound_dicts irreds
2576 (addToFM done w (done_id : w_id : rest_done_ids)) ws
2578 | otherwise -- Not yet done
2579 = case findAvailEnv avails w of
2580 Nothing -> pprTrace "Urk: extractResults" (ppr w) $
2581 go binds bound_dicts irreds done ws
2583 Just IsIrred -> go binds bound_dicts (w:irreds) done' ws
2585 Just (Rhs rhs ws') -> go (add_bind rhs) (w:bound_dicts) irreds done' (ws' ++ ws)
2587 Just (Given g) -> go binds' bound_dicts irreds (addToFM done w [g_id]) ws
2590 binds' | w_id == g_id = binds
2591 | otherwise = add_bind (nlHsVar g_id)
2594 done' = addToFM done w [w_id]
2595 add_bind rhs = addInstToDictBind binds w rhs
2599 Note [No superclasses for Stop]
2600 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2601 When we decide not to reduce an Inst -- the 'WhatToDo' --- we still
2602 add it to avails, so that any other equal Insts will be commoned up
2603 right here. However, we do *not* add superclasses. If we have
2606 but a is not bound here, then we *don't* want to derive dn from df
2607 here lest we lose sharing.
2610 addWanted :: WantSCs -> Avails -> Inst -> LHsExpr TcId -> [Inst] -> TcM Avails
2611 addWanted want_scs avails wanted rhs_expr wanteds
2612 = addAvailAndSCs want_scs avails wanted avail
2614 avail = Rhs rhs_expr wanteds
2616 addGiven :: Avails -> Inst -> TcM Avails
2617 addGiven avails given
2618 = addAvailAndSCs want_scs avails given (Given given)
2620 want_scs = case instLocOrigin (instLoc given) of
2623 -- Conditionally add superclasses for 'given'
2624 -- See Note [Recursive instances and superclases]
2626 -- No ASSERT( not (given `elemAvails` avails) ) because in an
2627 -- instance decl for Ord t we can add both Ord t and Eq t as
2628 -- 'givens', so the assert isn't true
2632 addIrred :: WantSCs -> Avails -> Inst -> TcM Avails
2633 addIrred want_scs avails irred = ASSERT2( not (irred `elemAvails` avails), ppr irred $$ ppr avails )
2634 addAvailAndSCs want_scs avails irred IsIrred
2636 addAvailAndSCs :: WantSCs -> Avails -> Inst -> AvailHow -> TcM Avails
2637 addAvailAndSCs want_scs avails inst avail
2638 | not (isClassDict inst) = extendAvails avails inst avail
2639 | NoSCs <- want_scs = extendAvails avails inst avail
2640 | otherwise = do { traceTc (text "addAvailAndSCs" <+> vcat [ppr inst, ppr deps])
2641 ; avails' <- extendAvails avails inst avail
2642 ; addSCs is_loop avails' inst }
2644 is_loop pred = any (`tcEqType` mkPredTy pred) dep_tys
2645 -- Note: this compares by *type*, not by Unique
2646 deps = findAllDeps (unitVarSet (instToVar inst)) avail
2647 dep_tys = map idType (varSetElems deps)
2649 findAllDeps :: IdSet -> AvailHow -> IdSet
2650 -- Find all the Insts that this one depends on
2651 -- See Note [SUPERCLASS-LOOP 2]
2652 -- Watch out, though. Since the avails may contain loops
2653 -- (see Note [RECURSIVE DICTIONARIES]), so we need to track the ones we've seen so far
2654 findAllDeps so_far (Rhs _ kids) = foldl find_all so_far kids
2655 findAllDeps so_far _ = so_far
2657 find_all :: IdSet -> Inst -> IdSet
2659 | isEqInst kid = so_far
2660 | kid_id `elemVarSet` so_far = so_far
2661 | Just avail <- findAvail avails kid = findAllDeps so_far' avail
2662 | otherwise = so_far'
2664 so_far' = extendVarSet so_far kid_id -- Add the new kid to so_far
2665 kid_id = instToId kid
2667 addSCs :: (TcPredType -> Bool) -> Avails -> Inst -> TcM Avails
2668 -- Add all the superclasses of the Inst to Avails
2669 -- The first param says "don't do this because the original thing
2670 -- depends on this one, so you'd build a loop"
2671 -- Invariant: the Inst is already in Avails.
2673 addSCs is_loop avails dict
2674 = ASSERT( isDict dict )
2675 do { sc_dicts <- newDictBndrs (instLoc dict) sc_theta'
2676 ; foldlM add_sc avails (zipEqual "add_scs" sc_dicts sc_sels) }
2678 (clas, tys) = getDictClassTys dict
2679 (tyvars, sc_theta, sc_sels, _) = classBigSig clas
2680 sc_theta' = filter (not . isEqPred) $
2681 substTheta (zipTopTvSubst tyvars tys) sc_theta
2683 add_sc avails (sc_dict, sc_sel)
2684 | is_loop (dictPred sc_dict) = return avails -- See Note [SUPERCLASS-LOOP 2]
2685 | is_given sc_dict = return avails
2686 | otherwise = do { avails' <- extendAvails avails sc_dict (Rhs sc_sel_rhs [dict])
2687 ; addSCs is_loop avails' sc_dict }
2689 sc_sel_rhs = L (instSpan dict) (HsWrap co_fn (HsVar sc_sel))
2690 co_fn = WpApp (instToVar dict) <.> mkWpTyApps tys
2692 is_given :: Inst -> Bool
2693 is_given sc_dict = case findAvail avails sc_dict of
2694 Just (Given _) -> True -- Given is cheaper than superclass selection
2697 -- From the a set of insts obtain all equalities that (transitively) occur in
2698 -- superclass contexts of class constraints (aka the ancestor equalities).
2700 ancestorEqualities :: [Inst] -> TcM [Inst]
2702 = mapM mkWantedEqInst -- turn only equality predicates..
2703 . filter isEqPred -- ..into wanted equality insts
2705 . addAEsToBag emptyBag -- collect the superclass constraints..
2706 . map dictPred -- ..of all predicates in a bag
2707 . filter isClassDict
2709 addAEsToBag :: Bag PredType -> [PredType] -> Bag PredType
2710 addAEsToBag bag [] = bag
2711 addAEsToBag bag (pred:preds)
2712 | pred `elemBag` bag = addAEsToBag bag preds
2713 | isEqPred pred = addAEsToBag bagWithPred preds
2714 | isClassPred pred = addAEsToBag bagWithPred predsWithSCs
2715 | otherwise = addAEsToBag bag preds
2717 bagWithPred = bag `snocBag` pred
2718 predsWithSCs = preds ++ substTheta (zipTopTvSubst tyvars tys) sc_theta
2720 (tyvars, sc_theta, _, _) = classBigSig clas
2721 (clas, tys) = getClassPredTys pred
2725 %************************************************************************
2727 \section{tcSimplifyTop: defaulting}
2729 %************************************************************************
2732 @tcSimplifyTop@ is called once per module to simplify all the constant
2733 and ambiguous Insts.
2735 We need to be careful of one case. Suppose we have
2737 instance Num a => Num (Foo a b) where ...
2739 and @tcSimplifyTop@ is given a constraint (Num (Foo x y)). Then it'll simplify
2740 to (Num x), and default x to Int. But what about y??
2742 It's OK: the final zonking stage should zap y to (), which is fine.
2746 tcSimplifyTop, tcSimplifyInteractive :: [Inst] -> TcM TcDictBinds
2747 tcSimplifyTop wanteds
2748 = tc_simplify_top doc False wanteds
2750 doc = text "tcSimplifyTop"
2752 tcSimplifyInteractive wanteds
2753 = tc_simplify_top doc True wanteds
2755 doc = text "tcSimplifyInteractive"
2757 -- The TcLclEnv should be valid here, solely to improve
2758 -- error message generation for the monomorphism restriction
2759 tc_simplify_top :: SDoc -> Bool -> [Inst] -> TcM (Bag (LHsBind TcId))
2760 tc_simplify_top doc interactive wanteds
2761 = do { dflags <- getDOpts
2762 ; wanteds <- zonkInsts wanteds
2763 ; mapM_ zonkTopTyVar (varSetElems (tyVarsOfInsts wanteds))
2765 ; traceTc (text "tc_simplify_top 0: " <+> ppr wanteds)
2766 ; (irreds1, binds1) <- tryHardCheckLoop doc1 wanteds
2767 -- ; (irreds1, binds1) <- gentleInferLoop doc1 wanteds
2768 ; traceTc (text "tc_simplify_top 1: " <+> ppr irreds1)
2769 ; (irreds2, binds2) <- approximateImplications doc2 (\_ -> True) irreds1
2770 ; traceTc (text "tc_simplify_top 2: " <+> ppr irreds2)
2772 -- Use the defaulting rules to do extra unification
2773 -- NB: irreds2 are already zonked
2774 ; (irreds3, binds3) <- disambiguate doc3 interactive dflags irreds2
2776 -- Deal with implicit parameters
2777 ; let (bad_ips, non_ips) = partition isIPDict irreds3
2778 (ambigs, others) = partition isTyVarDict non_ips
2780 ; topIPErrs bad_ips -- Can arise from f :: Int -> Int
2782 ; addNoInstanceErrs others
2783 ; addTopAmbigErrs ambigs
2785 ; return (binds1 `unionBags` binds2 `unionBags` binds3) }
2787 doc1 = doc <+> ptext (sLit "(first round)")
2788 doc2 = doc <+> ptext (sLit "(approximate)")
2789 doc3 = doc <+> ptext (sLit "(disambiguate)")
2792 If a dictionary constrains a type variable which is
2793 * not mentioned in the environment
2794 * and not mentioned in the type of the expression
2795 then it is ambiguous. No further information will arise to instantiate
2796 the type variable; nor will it be generalised and turned into an extra
2797 parameter to a function.
2799 It is an error for this to occur, except that Haskell provided for
2800 certain rules to be applied in the special case of numeric types.
2802 * at least one of its classes is a numeric class, and
2803 * all of its classes are numeric or standard
2804 then the type variable can be defaulted to the first type in the
2805 default-type list which is an instance of all the offending classes.
2807 So here is the function which does the work. It takes the ambiguous
2808 dictionaries and either resolves them (producing bindings) or
2809 complains. It works by splitting the dictionary list by type
2810 variable, and using @disambigOne@ to do the real business.
2812 @disambigOne@ assumes that its arguments dictionaries constrain all
2813 the same type variable.
2815 ADR Comment 20/6/94: I've changed the @CReturnable@ case to default to
2816 @()@ instead of @Int@. I reckon this is the Right Thing to do since
2817 the most common use of defaulting is code like:
2819 _ccall_ foo `seqPrimIO` bar
2821 Since we're not using the result of @foo@, the result if (presumably)
2825 disambiguate :: SDoc -> Bool -> DynFlags -> [Inst] -> TcM ([Inst], TcDictBinds)
2826 -- Just does unification to fix the default types
2827 -- The Insts are assumed to be pre-zonked
2828 disambiguate doc interactive dflags insts
2830 = return (insts, emptyBag)
2832 | null defaultable_groups
2833 = do { traceTc (text "disambigutate, no defaultable groups" <+> vcat [ppr unaries, ppr insts, ppr bad_tvs, ppr defaultable_groups])
2834 ; return (insts, emptyBag) }
2837 = do { -- Figure out what default types to use
2838 default_tys <- getDefaultTys extended_defaulting ovl_strings
2840 ; traceTc (text "disambiguate1" <+> vcat [ppr insts, ppr unaries, ppr bad_tvs, ppr defaultable_groups])
2841 ; mapM_ (disambigGroup default_tys) defaultable_groups
2843 -- disambigGroup does unification, hence try again
2844 ; tryHardCheckLoop doc insts }
2847 extended_defaulting = interactive || dopt Opt_ExtendedDefaultRules dflags
2848 ovl_strings = dopt Opt_OverloadedStrings dflags
2850 unaries :: [(Inst, Class, TcTyVar)] -- (C tv) constraints
2851 bad_tvs :: TcTyVarSet -- Tyvars mentioned by *other* constraints
2852 (unaries, bad_tvs_s) = partitionWith find_unary insts
2853 bad_tvs = unionVarSets bad_tvs_s
2855 -- Finds unary type-class constraints
2856 find_unary d@(Dict {tci_pred = ClassP cls [ty]})
2857 | Just tv <- tcGetTyVar_maybe ty = Left (d,cls,tv)
2858 find_unary inst = Right (tyVarsOfInst inst)
2860 -- Group by type variable
2861 defaultable_groups :: [[(Inst,Class,TcTyVar)]]
2862 defaultable_groups = filter defaultable_group (equivClasses cmp_tv unaries)
2863 cmp_tv (_,_,tv1) (_,_,tv2) = tv1 `compare` tv2
2865 defaultable_group :: [(Inst,Class,TcTyVar)] -> Bool
2866 defaultable_group ds@((_,_,tv):_)
2867 = isTyConableTyVar tv -- Note [Avoiding spurious errors]
2868 && not (tv `elemVarSet` bad_tvs)
2869 && defaultable_classes [c | (_,c,_) <- ds]
2870 defaultable_group [] = panic "defaultable_group"
2872 defaultable_classes clss
2873 | extended_defaulting = any isInteractiveClass clss
2874 | otherwise = all is_std_class clss && (any is_num_class clss)
2876 -- In interactive mode, or with -XExtendedDefaultRules,
2877 -- we default Show a to Show () to avoid graututious errors on "show []"
2878 isInteractiveClass cls
2879 = is_num_class cls || (classKey cls `elem` [showClassKey, eqClassKey, ordClassKey])
2881 is_num_class cls = isNumericClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2882 -- is_num_class adds IsString to the standard numeric classes,
2883 -- when -foverloaded-strings is enabled
2885 is_std_class cls = isStandardClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2886 -- Similarly is_std_class
2888 -----------------------
2889 disambigGroup :: [Type] -- The default types
2890 -> [(Inst,Class,TcTyVar)] -- All standard classes of form (C a)
2891 -> TcM () -- Just does unification, to fix the default types
2893 disambigGroup default_tys dicts
2894 = try_default default_tys
2896 (_,_,tyvar) = ASSERT(not (null dicts)) head dicts -- Should be non-empty
2897 classes = [c | (_,c,_) <- dicts]
2899 try_default [] = return ()
2900 try_default (default_ty : default_tys)
2901 = tryTcLIE_ (try_default default_tys) $
2902 do { tcSimplifyDefault [mkClassPred clas [default_ty] | clas <- classes]
2903 -- This may fail; then the tryTcLIE_ kicks in
2904 -- Failure here is caused by there being no type in the
2905 -- default list which can satisfy all the ambiguous classes.
2906 -- For example, if Real a is reqd, but the only type in the
2907 -- default list is Int.
2909 -- After this we can't fail
2910 ; warnDefault dicts default_ty
2911 ; unifyType default_ty (mkTyVarTy tyvar)
2912 ; return () -- TOMDO: do something with the coercion
2916 -----------------------
2917 getDefaultTys :: Bool -> Bool -> TcM [Type]
2918 getDefaultTys extended_deflts ovl_strings
2919 = do { mb_defaults <- getDeclaredDefaultTys
2920 ; case mb_defaults of {
2921 Just tys -> return tys ; -- User-supplied defaults
2924 -- No use-supplied default
2925 -- Use [Integer, Double], plus modifications
2926 { integer_ty <- tcMetaTy integerTyConName
2927 ; checkWiredInTyCon doubleTyCon
2928 ; string_ty <- tcMetaTy stringTyConName
2929 ; return (opt_deflt extended_deflts unitTy
2930 -- Note [Default unitTy]
2932 [integer_ty,doubleTy]
2934 opt_deflt ovl_strings string_ty) } } }
2936 opt_deflt True ty = [ty]
2937 opt_deflt False _ = []
2940 Note [Default unitTy]
2941 ~~~~~~~~~~~~~~~~~~~~~
2942 In interative mode (or with -XExtendedDefaultRules) we add () as the first type we
2943 try when defaulting. This has very little real impact, except in the following case.
2945 Text.Printf.printf "hello"
2946 This has type (forall a. IO a); it prints "hello", and returns 'undefined'. We don't
2947 want the GHCi repl loop to try to print that 'undefined'. The neatest thing is to
2948 default the 'a' to (), rather than to Integer (which is what would otherwise happen;
2949 and then GHCi doesn't attempt to print the (). So in interactive mode, we add
2950 () to the list of defaulting types. See Trac #1200.
2952 Note [Avoiding spurious errors]
2953 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2954 When doing the unification for defaulting, we check for skolem
2955 type variables, and simply don't default them. For example:
2956 f = (*) -- Monomorphic
2957 g :: Num a => a -> a
2959 Here, we get a complaint when checking the type signature for g,
2960 that g isn't polymorphic enough; but then we get another one when
2961 dealing with the (Num a) context arising from f's definition;
2962 we try to unify a with Int (to default it), but find that it's
2963 already been unified with the rigid variable from g's type sig
2966 %************************************************************************
2968 \subsection[simple]{@Simple@ versions}
2970 %************************************************************************
2972 Much simpler versions when there are no bindings to make!
2974 @tcSimplifyThetas@ simplifies class-type constraints formed by
2975 @deriving@ declarations and when specialising instances. We are
2976 only interested in the simplified bunch of class/type constraints.
2978 It simplifies to constraints of the form (C a b c) where
2979 a,b,c are type variables. This is required for the context of
2980 instance declarations.
2983 tcSimplifyDeriv :: InstOrigin
2985 -> ThetaType -- Wanted
2986 -> TcM ThetaType -- Needed
2987 -- Given instance (wanted) => C inst_ty
2988 -- Simplify 'wanted' as much as possible
2990 tcSimplifyDeriv orig tyvars theta
2991 = do { (tvs, _, tenv) <- tcInstTyVars tyvars
2992 -- The main loop may do unification, and that may crash if
2993 -- it doesn't see a TcTyVar, so we have to instantiate. Sigh
2994 -- ToDo: what if two of them do get unified?
2995 ; wanteds <- newDictBndrsO orig (substTheta tenv theta)
2996 ; (irreds, _) <- tryHardCheckLoop doc wanteds
2998 ; let (tv_dicts, others) = partition ok irreds
2999 (tidy_env, tidy_insts) = tidyInsts others
3000 ; reportNoInstances tidy_env Nothing [alt_fix] tidy_insts
3001 -- See Note [Exotic derived instance contexts] in TcMType
3003 ; let rev_env = zipTopTvSubst tvs (mkTyVarTys tyvars)
3004 simpl_theta = substTheta rev_env (map dictPred tv_dicts)
3005 -- This reverse-mapping is a pain, but the result
3006 -- should mention the original TyVars not TcTyVars
3008 ; return simpl_theta }
3010 doc = ptext (sLit "deriving classes for a data type")
3012 ok dict | isDict dict = validDerivPred (dictPred dict)
3014 alt_fix = vcat [ptext (sLit "use a standalone 'deriving instance' declaration instead,"),
3015 ptext (sLit "so you can specify the instance context yourself")]
3019 @tcSimplifyDefault@ just checks class-type constraints, essentially;
3020 used with \tr{default} declarations. We are only interested in
3021 whether it worked or not.
3024 tcSimplifyDefault :: ThetaType -- Wanted; has no type variables in it
3027 tcSimplifyDefault theta = do
3028 wanteds <- newDictBndrsO DefaultOrigin theta
3029 (irreds, _) <- tryHardCheckLoop doc wanteds
3030 addNoInstanceErrs irreds
3034 traceTc (ptext (sLit "tcSimplifyDefault failing")) >> failM
3036 doc = ptext (sLit "default declaration")
3039 @tcSimplifyStagedExpr@ performs a simplification but does so at a new
3040 stage. This is used when typechecking annotations and splices.
3044 tcSimplifyStagedExpr :: ThStage -> TcM a -> TcM (a, TcDictBinds)
3045 -- Type check an expression that runs at a top level stage as if
3046 -- it were going to be spliced and then simplify it
3047 tcSimplifyStagedExpr stage tc_action
3048 = setStage stage $ do {
3049 -- Typecheck the expression
3050 (thing', lie) <- getLIE tc_action
3052 -- Solve the constraints
3053 ; const_binds <- tcSimplifyTop lie
3055 ; return (thing', const_binds) }
3060 %************************************************************************
3062 \section{Errors and contexts}
3064 %************************************************************************
3066 ToDo: for these error messages, should we note the location as coming
3067 from the insts, or just whatever seems to be around in the monad just
3071 groupErrs :: ([Inst] -> TcM ()) -- Deal with one group
3072 -> [Inst] -- The offending Insts
3074 -- Group together insts with the same origin
3075 -- We want to report them together in error messages
3079 groupErrs report_err (inst:insts)
3080 = do { do_one (inst:friends)
3081 ; groupErrs report_err others }
3083 -- (It may seem a bit crude to compare the error messages,
3084 -- but it makes sure that we combine just what the user sees,
3085 -- and it avoids need equality on InstLocs.)
3086 (friends, others) = partition is_friend insts
3087 loc_msg = showSDoc (pprInstLoc (instLoc inst))
3088 is_friend friend = showSDoc (pprInstLoc (instLoc friend)) == loc_msg
3089 do_one insts = addInstCtxt (instLoc (head insts)) (report_err insts)
3090 -- Add location and context information derived from the Insts
3092 -- Add the "arising from..." part to a message about bunch of dicts
3093 addInstLoc :: [Inst] -> Message -> Message
3094 addInstLoc insts msg = msg $$ nest 2 (pprInstArising (head insts))
3096 addTopIPErrs :: [Name] -> [Inst] -> TcM ()
3099 addTopIPErrs bndrs ips
3100 = do { dflags <- getDOpts
3101 ; addErrTcM (tidy_env, mk_msg dflags tidy_ips) }
3103 (tidy_env, tidy_ips) = tidyInsts ips
3105 = vcat [sep [ptext (sLit "Implicit parameters escape from"),
3106 nest 2 (ptext (sLit "the monomorphic top-level binding")
3107 <> plural bndrs <+> ptext (sLit "of")
3108 <+> pprBinders bndrs <> colon)],
3109 nest 2 (vcat (map ppr_ip ips)),
3110 monomorphism_fix dflags]
3111 ppr_ip ip = pprPred (dictPred ip) <+> pprInstArising ip
3113 topIPErrs :: [Inst] -> TcM ()
3115 = groupErrs report tidy_dicts
3117 (tidy_env, tidy_dicts) = tidyInsts dicts
3118 report dicts = addErrTcM (tidy_env, mk_msg dicts)
3119 mk_msg dicts = addInstLoc dicts (ptext (sLit "Unbound implicit parameter") <>
3120 plural tidy_dicts <+> pprDictsTheta tidy_dicts)
3122 addNoInstanceErrs :: [Inst] -- Wanted (can include implications)
3124 addNoInstanceErrs insts
3125 = do { let (tidy_env, tidy_insts) = tidyInsts insts
3126 ; reportNoInstances tidy_env Nothing [] tidy_insts }
3130 -> Maybe (InstLoc, [Inst]) -- Context
3131 -- Nothing => top level
3132 -- Just (d,g) => d describes the construct
3134 -> [SDoc] -- Alternative fix for no-such-instance
3135 -> [Inst] -- What is wanted (can include implications)
3138 reportNoInstances tidy_env mb_what alt_fix insts
3139 = groupErrs (report_no_instances tidy_env mb_what alt_fix) insts
3141 report_no_instances :: TidyEnv -> Maybe (InstLoc, [Inst]) -> [SDoc] -> [Inst] -> TcM ()
3142 report_no_instances tidy_env mb_what alt_fixes insts
3143 = do { inst_envs <- tcGetInstEnvs
3144 ; let (implics, insts1) = partition isImplicInst insts
3145 (insts2, overlaps) = partitionWith (check_overlap inst_envs) insts1
3146 (eqInsts, insts3) = partition isEqInst insts2
3147 ; traceTc (text "reportNoInstances" <+> vcat
3148 [ppr insts, ppr implics, ppr insts1, ppr insts2])
3149 ; mapM_ complain_implic implics
3150 ; mapM_ (\doc -> addErrTcM (tidy_env, doc)) overlaps
3151 ; groupErrs complain_no_inst insts3
3152 ; mapM_ (addErrTcM . mk_eq_err) eqInsts
3155 complain_no_inst insts = addErrTcM (tidy_env, mk_no_inst_err insts)
3157 complain_implic inst -- Recurse!
3158 = reportNoInstances tidy_env
3159 (Just (tci_loc inst, tci_given inst))
3160 alt_fixes (tci_wanted inst)
3162 check_overlap :: (InstEnv,InstEnv) -> Inst -> Either Inst SDoc
3163 -- Right msg => overlap message
3164 -- Left inst => no instance
3165 check_overlap inst_envs wanted
3166 | not (isClassDict wanted) = Left wanted
3168 = case lookupInstEnv inst_envs clas tys of
3169 ([], _) -> Left wanted -- No match
3170 -- The case of exactly one match and no unifiers means a
3171 -- successful lookup. That can't happen here, because dicts
3172 -- only end up here if they didn't match in Inst.lookupInst
3174 | debugIsOn -> pprPanic "reportNoInstance" (ppr wanted)
3175 res -> Right (mk_overlap_msg wanted res)
3177 (clas,tys) = getDictClassTys wanted
3179 mk_overlap_msg dict (matches, unifiers)
3180 = ASSERT( not (null matches) )
3181 vcat [ addInstLoc [dict] ((ptext (sLit "Overlapping instances for")
3182 <+> pprPred (dictPred dict))),
3183 sep [ptext (sLit "Matching instances") <> colon,
3184 nest 2 (vcat [pprInstances ispecs, pprInstances unifiers])],
3185 if not (isSingleton matches)
3186 then -- Two or more matches
3188 else -- One match, plus some unifiers
3189 ASSERT( not (null unifiers) )
3190 parens (vcat [ptext (sLit "The choice depends on the instantiation of") <+>
3191 quotes (pprWithCommas ppr (varSetElems (tyVarsOfInst dict))),
3192 ptext (sLit "To pick the first instance above, use -XIncoherentInstances"),
3193 ptext (sLit "when compiling the other instance declarations")])]
3195 ispecs = [ispec | (ispec, _) <- matches]
3197 mk_eq_err :: Inst -> (TidyEnv, SDoc)
3198 mk_eq_err inst = misMatchMsg tidy_env (eqInstTys inst)
3200 mk_no_inst_err insts
3201 | null insts = empty
3203 | Just (loc, givens) <- mb_what, -- Nested (type signatures, instance decls)
3204 not (isEmptyVarSet (tyVarsOfInsts insts))
3205 = vcat [ addInstLoc insts $
3206 sep [ ptext (sLit "Could not deduce") <+> pprDictsTheta insts
3207 , nest 2 $ ptext (sLit "from the context") <+> pprDictsTheta givens]
3208 , show_fixes (fix1 loc : fixes2 ++ alt_fixes) ]
3210 | otherwise -- Top level
3211 = vcat [ addInstLoc insts $
3212 ptext (sLit "No instance") <> plural insts
3213 <+> ptext (sLit "for") <+> pprDictsTheta insts
3214 , show_fixes (fixes2 ++ alt_fixes) ]
3217 fix1 loc = sep [ ptext (sLit "add") <+> pprDictsTheta insts
3218 <+> ptext (sLit "to the context of"),
3219 nest 2 (ppr (instLocOrigin loc)) ]
3220 -- I'm not sure it helps to add the location
3221 -- nest 2 (ptext (sLit "at") <+> ppr (instLocSpan loc)) ]
3223 fixes2 | null instance_dicts = []
3224 | otherwise = [sep [ptext (sLit "add an instance declaration for"),
3225 pprDictsTheta instance_dicts]]
3226 instance_dicts = [d | d <- insts, isClassDict d, not (isTyVarDict d)]
3227 -- Insts for which it is worth suggesting an adding an instance declaration
3228 -- Exclude implicit parameters, and tyvar dicts
3230 show_fixes :: [SDoc] -> SDoc
3231 show_fixes [] = empty
3232 show_fixes (f:fs) = sep [ptext (sLit "Possible fix:"),
3233 nest 2 (vcat (f : map (ptext (sLit "or") <+>) fs))]
3235 addTopAmbigErrs :: [Inst] -> TcRn ()
3236 addTopAmbigErrs dicts
3237 -- Divide into groups that share a common set of ambiguous tyvars
3238 = ifErrsM (return ()) $ -- Only report ambiguity if no other errors happened
3239 -- See Note [Avoiding spurious errors]
3240 mapM_ report (equivClasses cmp [(d, tvs_of d) | d <- tidy_dicts])
3242 (tidy_env, tidy_dicts) = tidyInsts dicts
3244 tvs_of :: Inst -> [TcTyVar]
3245 tvs_of d = varSetElems (tyVarsOfInst d)
3246 cmp (_,tvs1) (_,tvs2) = tvs1 `compare` tvs2
3248 report :: [(Inst,[TcTyVar])] -> TcM ()
3249 report pairs@((inst,tvs) : _) = do -- The pairs share a common set of ambiguous tyvars
3250 (tidy_env, mono_msg) <- mkMonomorphismMsg tidy_env tvs
3251 setSrcSpan (instSpan inst) $
3252 -- the location of the first one will do for the err message
3253 addErrTcM (tidy_env, msg $$ mono_msg)
3255 dicts = map fst pairs
3256 msg = sep [text "Ambiguous type variable" <> plural tvs <+>
3257 pprQuotedList tvs <+> in_msg,
3258 nest 2 (pprDictsInFull dicts)]
3259 in_msg = text "in the constraint" <> plural dicts <> colon
3260 report [] = panic "addTopAmbigErrs"
3263 mkMonomorphismMsg :: TidyEnv -> [TcTyVar] -> TcM (TidyEnv, Message)
3264 -- There's an error with these Insts; if they have free type variables
3265 -- it's probably caused by the monomorphism restriction.
3266 -- Try to identify the offending variable
3267 -- ASSUMPTION: the Insts are fully zonked
3268 mkMonomorphismMsg tidy_env inst_tvs
3269 = do { dflags <- getDOpts
3270 ; (tidy_env, docs) <- findGlobals (mkVarSet inst_tvs) tidy_env
3271 ; return (tidy_env, mk_msg dflags docs) }
3273 mk_msg _ _ | any isRuntimeUnk inst_tvs
3274 = vcat [ptext (sLit "Cannot resolve unknown runtime types:") <+>
3275 (pprWithCommas ppr inst_tvs),
3276 ptext (sLit "Use :print or :force to determine these types")]
3277 mk_msg _ [] = ptext (sLit "Probable fix: add a type signature that fixes these type variable(s)")
3278 -- This happens in things like
3279 -- f x = show (read "foo")
3280 -- where monomorphism doesn't play any role
3282 = vcat [ptext (sLit "Possible cause: the monomorphism restriction applied to the following:"),
3284 monomorphism_fix dflags]
3286 monomorphism_fix :: DynFlags -> SDoc
3287 monomorphism_fix dflags
3288 = ptext (sLit "Probable fix:") <+> vcat
3289 [ptext (sLit "give these definition(s) an explicit type signature"),
3290 if dopt Opt_MonomorphismRestriction dflags
3291 then ptext (sLit "or use -XNoMonomorphismRestriction")
3292 else empty] -- Only suggest adding "-XNoMonomorphismRestriction"
3293 -- if it is not already set!
3295 warnDefault :: [(Inst, Class, Var)] -> Type -> TcM ()
3296 warnDefault ups default_ty = do
3297 warn_flag <- doptM Opt_WarnTypeDefaults
3298 addInstCtxt (instLoc (head (dicts))) (warnTc warn_flag warn_msg)
3300 dicts = [d | (d,_,_) <- ups]
3303 (_, tidy_dicts) = tidyInsts dicts
3304 warn_msg = vcat [ptext (sLit "Defaulting the following constraint(s) to type") <+>
3305 quotes (ppr default_ty),
3306 pprDictsInFull tidy_dicts]
3308 reduceDepthErr :: Int -> [Inst] -> SDoc
3309 reduceDepthErr n stack
3310 = vcat [ptext (sLit "Context reduction stack overflow; size =") <+> int n,
3311 ptext (sLit "Use -fcontext-stack=N to increase stack size to N"),
3312 nest 4 (pprStack stack)]
3314 pprStack :: [Inst] -> SDoc
3315 pprStack stack = vcat (map pprInstInFull stack)