2 % (c) The University of Glasgow 2006
3 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
10 -- The above warning supression flag is a temporary kludge.
11 -- While working on this module you are encouraged to remove it and fix
12 -- any warnings in the module. See
13 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
17 tcSimplifyInfer, tcSimplifyInferCheck,
18 tcSimplifyCheck, tcSimplifyRestricted,
19 tcSimplifyRuleLhs, tcSimplifyIPs,
20 tcSimplifySuperClasses,
21 tcSimplifyTop, tcSimplifyInteractive,
22 tcSimplifyBracket, tcSimplifyCheckPat,
24 tcSimplifyDeriv, tcSimplifyDefault,
30 #include "HsVersions.h"
32 import {-# SOURCE #-} TcUnify( unifyType )
73 %************************************************************************
77 %************************************************************************
79 --------------------------------------
80 Notes on functional dependencies (a bug)
81 --------------------------------------
88 instance D a b => C a b -- Undecidable
89 -- (Not sure if it's crucial to this eg)
90 f :: C a b => a -> Bool
93 g :: C a b => a -> Bool
96 Here f typechecks, but g does not!! Reason: before doing improvement,
97 we reduce the (C a b1) constraint from the call of f to (D a b1).
99 Here is a more complicated example:
101 | > class Foo a b | a->b
103 | > class Bar a b | a->b
107 | > instance Bar Obj Obj
109 | > instance (Bar a b) => Foo a b
111 | > foo:: (Foo a b) => a -> String
114 | > runFoo:: (forall a b. (Foo a b) => a -> w) -> w
120 | Could not deduce (Bar a b) from the context (Foo a b)
121 | arising from use of `foo' at <interactive>:1
123 | Add (Bar a b) to the expected type of an expression
124 | In the first argument of `runFoo', namely `foo'
125 | In the definition of `it': it = runFoo foo
127 | Why all of the sudden does GHC need the constraint Bar a b? The
128 | function foo didn't ask for that...
130 The trouble is that to type (runFoo foo), GHC has to solve the problem:
132 Given constraint Foo a b
133 Solve constraint Foo a b'
135 Notice that b and b' aren't the same. To solve this, just do
136 improvement and then they are the same. But GHC currently does
141 That is usually fine, but it isn't here, because it sees that Foo a b is
142 not the same as Foo a b', and so instead applies the instance decl for
143 instance Bar a b => Foo a b. And that's where the Bar constraint comes
146 The Right Thing is to improve whenever the constraint set changes at
147 all. Not hard in principle, but it'll take a bit of fiddling to do.
149 Note [Choosing which variables to quantify]
150 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
151 Suppose we are about to do a generalisation step. We have in our hand
154 T the type of the RHS
155 C the constraints from that RHS
157 The game is to figure out
159 Q the set of type variables over which to quantify
160 Ct the constraints we will *not* quantify over
161 Cq the constraints we will quantify over
163 So we're going to infer the type
167 and float the constraints Ct further outwards.
169 Here are the things that *must* be true:
171 (A) Q intersect fv(G) = EMPTY limits how big Q can be
172 (B) Q superset fv(Cq union T) \ oclose(fv(G),C) limits how small Q can be
174 (A) says we can't quantify over a variable that's free in the environment.
175 (B) says we must quantify over all the truly free variables in T, else
176 we won't get a sufficiently general type.
178 We do not *need* to quantify over any variable that is fixed by the
179 free vars of the environment G.
181 BETWEEN THESE TWO BOUNDS, ANY Q WILL DO!
183 Example: class H x y | x->y where ...
185 fv(G) = {a} C = {H a b, H c d}
188 (A) Q intersect {a} is empty
189 (B) Q superset {a,b,c,d} \ oclose({a}, C) = {a,b,c,d} \ {a,b} = {c,d}
191 So Q can be {c,d}, {b,c,d}
193 In particular, it's perfectly OK to quantify over more type variables
194 than strictly necessary; there is no need to quantify over 'b', since
195 it is determined by 'a' which is free in the envt, but it's perfectly
196 OK to do so. However we must not quantify over 'a' itself.
198 Other things being equal, however, we'd like to quantify over as few
199 variables as possible: smaller types, fewer type applications, more
200 constraints can get into Ct instead of Cq. Here's a good way to
203 Q = grow( fv(T), C ) \ oclose( fv(G), C )
205 That is, quantify over all variable that that MIGHT be fixed by the
206 call site (which influences T), but which aren't DEFINITELY fixed by
207 G. This choice definitely quantifies over enough type variables,
208 albeit perhaps too many.
210 Why grow( fv(T), C ) rather than fv(T)? Consider
212 class H x y | x->y where ...
217 If we used fv(T) = {c} we'd get the type
219 forall c. H c d => c -> b
221 And then if the fn was called at several different c's, each of
222 which fixed d differently, we'd get a unification error, because
223 d isn't quantified. Solution: quantify d. So we must quantify
224 everything that might be influenced by c.
226 Why not oclose( fv(T), C )? Because we might not be able to see
227 all the functional dependencies yet:
229 class H x y | x->y where ...
230 instance H x y => Eq (T x y) where ...
235 Now oclose(fv(T),C) = {c}, because the functional dependency isn't
236 apparent yet, and that's wrong. We must really quantify over d too.
238 There really isn't any point in quantifying over any more than
239 grow( fv(T), C ), because the call sites can't possibly influence
240 any other type variables.
244 -------------------------------------
246 -------------------------------------
248 It's very hard to be certain when a type is ambiguous. Consider
252 instance H x y => K (x,y)
254 Is this type ambiguous?
255 forall a b. (K (a,b), Eq b) => a -> a
257 Looks like it! But if we simplify (K (a,b)) we get (H a b) and
258 now we see that a fixes b. So we can't tell about ambiguity for sure
259 without doing a full simplification. And even that isn't possible if
260 the context has some free vars that may get unified. Urgle!
262 Here's another example: is this ambiguous?
263 forall a b. Eq (T b) => a -> a
264 Not if there's an insance decl (with no context)
265 instance Eq (T b) where ...
267 You may say of this example that we should use the instance decl right
268 away, but you can't always do that:
270 class J a b where ...
271 instance J Int b where ...
273 f :: forall a b. J a b => a -> a
275 (Notice: no functional dependency in J's class decl.)
276 Here f's type is perfectly fine, provided f is only called at Int.
277 It's premature to complain when meeting f's signature, or even
278 when inferring a type for f.
282 However, we don't *need* to report ambiguity right away. It'll always
283 show up at the call site.... and eventually at main, which needs special
284 treatment. Nevertheless, reporting ambiguity promptly is an excellent thing.
286 So here's the plan. We WARN about probable ambiguity if
288 fv(Cq) is not a subset of oclose(fv(T) union fv(G), C)
290 (all tested before quantification).
291 That is, all the type variables in Cq must be fixed by the the variables
292 in the environment, or by the variables in the type.
294 Notice that we union before calling oclose. Here's an example:
296 class J a b c | a b -> c
300 forall b c. (J a b c) => b -> b
302 Only if we union {a} from G with {b} from T before using oclose,
303 do we see that c is fixed.
305 It's a bit vague exactly which C we should use for this oclose call. If we
306 don't fix enough variables we might complain when we shouldn't (see
307 the above nasty example). Nothing will be perfect. That's why we can
308 only issue a warning.
311 Can we ever be *certain* about ambiguity? Yes: if there's a constraint
313 c in C such that fv(c) intersect (fv(G) union fv(T)) = EMPTY
315 then c is a "bubble"; there's no way it can ever improve, and it's
316 certainly ambiguous. UNLESS it is a constant (sigh). And what about
321 instance H x y => K (x,y)
323 Is this type ambiguous?
324 forall a b. (K (a,b), Eq b) => a -> a
326 Urk. The (Eq b) looks "definitely ambiguous" but it isn't. What we are after
327 is a "bubble" that's a set of constraints
329 Cq = Ca union Cq' st fv(Ca) intersect (fv(Cq') union fv(T) union fv(G)) = EMPTY
331 Hence another idea. To decide Q start with fv(T) and grow it
332 by transitive closure in Cq (no functional dependencies involved).
333 Now partition Cq using Q, leaving the definitely-ambiguous and probably-ok.
334 The definitely-ambiguous can then float out, and get smashed at top level
335 (which squashes out the constants, like Eq (T a) above)
338 --------------------------------------
339 Notes on principal types
340 --------------------------------------
345 f x = let g y = op (y::Int) in True
347 Here the principal type of f is (forall a. a->a)
348 but we'll produce the non-principal type
349 f :: forall a. C Int => a -> a
352 --------------------------------------
353 The need for forall's in constraints
354 --------------------------------------
356 [Exchange on Haskell Cafe 5/6 Dec 2000]
358 class C t where op :: t -> Bool
359 instance C [t] where op x = True
361 p y = (let f :: c -> Bool; f x = op (y >> return x) in f, y ++ [])
362 q y = (y ++ [], let f :: c -> Bool; f x = op (y >> return x) in f)
364 The definitions of p and q differ only in the order of the components in
365 the pair on their right-hand sides. And yet:
367 ghc and "Typing Haskell in Haskell" reject p, but accept q;
368 Hugs rejects q, but accepts p;
369 hbc rejects both p and q;
370 nhc98 ... (Malcolm, can you fill in the blank for us!).
372 The type signature for f forces context reduction to take place, and
373 the results of this depend on whether or not the type of y is known,
374 which in turn depends on which component of the pair the type checker
377 Solution: if y::m a, float out the constraints
378 Monad m, forall c. C (m c)
379 When m is later unified with [], we can solve both constraints.
382 --------------------------------------
383 Notes on implicit parameters
384 --------------------------------------
386 Note [Inheriting implicit parameters]
387 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
392 where f is *not* a top-level binding.
393 From the RHS of f we'll get the constraint (?y::Int).
394 There are two types we might infer for f:
398 (so we get ?y from the context of f's definition), or
400 f :: (?y::Int) => Int -> Int
402 At first you might think the first was better, becuase then
403 ?y behaves like a free variable of the definition, rather than
404 having to be passed at each call site. But of course, the WHOLE
405 IDEA is that ?y should be passed at each call site (that's what
406 dynamic binding means) so we'd better infer the second.
408 BOTTOM LINE: when *inferring types* you *must* quantify
409 over implicit parameters. See the predicate isFreeWhenInferring.
412 Note [Implicit parameters and ambiguity]
413 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
414 Only a *class* predicate can give rise to ambiguity
415 An *implicit parameter* cannot. For example:
416 foo :: (?x :: [a]) => Int
418 is fine. The call site will suppply a particular 'x'
420 Furthermore, the type variables fixed by an implicit parameter
421 propagate to the others. E.g.
422 foo :: (Show a, ?x::[a]) => Int
424 The type of foo looks ambiguous. But it isn't, because at a call site
426 let ?x = 5::Int in foo
427 and all is well. In effect, implicit parameters are, well, parameters,
428 so we can take their type variables into account as part of the
429 "tau-tvs" stuff. This is done in the function 'FunDeps.grow'.
432 Question 2: type signatures
433 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
434 BUT WATCH OUT: When you supply a type signature, we can't force you
435 to quantify over implicit parameters. For example:
439 This is perfectly reasonable. We do not want to insist on
441 (?x + 1) :: (?x::Int => Int)
443 That would be silly. Here, the definition site *is* the occurrence site,
444 so the above strictures don't apply. Hence the difference between
445 tcSimplifyCheck (which *does* allow implicit paramters to be inherited)
446 and tcSimplifyCheckBind (which does not).
448 What about when you supply a type signature for a binding?
449 Is it legal to give the following explicit, user type
450 signature to f, thus:
455 At first sight this seems reasonable, but it has the nasty property
456 that adding a type signature changes the dynamic semantics.
459 (let f x = (x::Int) + ?y
460 in (f 3, f 3 with ?y=5)) with ?y = 6
466 in (f 3, f 3 with ?y=5)) with ?y = 6
470 Indeed, simply inlining f (at the Haskell source level) would change the
473 Nevertheless, as Launchbury says (email Oct 01) we can't really give the
474 semantics for a Haskell program without knowing its typing, so if you
475 change the typing you may change the semantics.
477 To make things consistent in all cases where we are *checking* against
478 a supplied signature (as opposed to inferring a type), we adopt the
481 a signature does not need to quantify over implicit params.
483 [This represents a (rather marginal) change of policy since GHC 5.02,
484 which *required* an explicit signature to quantify over all implicit
485 params for the reasons mentioned above.]
487 But that raises a new question. Consider
489 Given (signature) ?x::Int
490 Wanted (inferred) ?x::Int, ?y::Bool
492 Clearly we want to discharge the ?x and float the ?y out. But
493 what is the criterion that distinguishes them? Clearly it isn't
494 what free type variables they have. The Right Thing seems to be
495 to float a constraint that
496 neither mentions any of the quantified type variables
497 nor any of the quantified implicit parameters
499 See the predicate isFreeWhenChecking.
502 Question 3: monomorphism
503 ~~~~~~~~~~~~~~~~~~~~~~~~
504 There's a nasty corner case when the monomorphism restriction bites:
508 The argument above suggests that we *must* generalise
509 over the ?y parameter, to get
510 z :: (?y::Int) => Int,
511 but the monomorphism restriction says that we *must not*, giving
513 Why does the momomorphism restriction say this? Because if you have
515 let z = x + ?y in z+z
517 you might not expect the addition to be done twice --- but it will if
518 we follow the argument of Question 2 and generalise over ?y.
521 Question 4: top level
522 ~~~~~~~~~~~~~~~~~~~~~
523 At the top level, monomorhism makes no sense at all.
526 main = let ?x = 5 in print foo
530 woggle :: (?x :: Int) => Int -> Int
533 We definitely don't want (foo :: Int) with a top-level implicit parameter
534 (?x::Int) becuase there is no way to bind it.
539 (A) Always generalise over implicit parameters
540 Bindings that fall under the monomorphism restriction can't
544 * Inlining remains valid
545 * No unexpected loss of sharing
546 * But simple bindings like
548 will be rejected, unless you add an explicit type signature
549 (to avoid the monomorphism restriction)
550 z :: (?y::Int) => Int
552 This seems unacceptable
554 (B) Monomorphism restriction "wins"
555 Bindings that fall under the monomorphism restriction can't
557 Always generalise over implicit parameters *except* for bindings
558 that fall under the monomorphism restriction
561 * Inlining isn't valid in general
562 * No unexpected loss of sharing
563 * Simple bindings like
565 accepted (get value of ?y from binding site)
567 (C) Always generalise over implicit parameters
568 Bindings that fall under the monomorphism restriction can't
569 be generalised, EXCEPT for implicit parameters
571 * Inlining remains valid
572 * Unexpected loss of sharing (from the extra generalisation)
573 * Simple bindings like
575 accepted (get value of ?y from occurrence sites)
580 None of these choices seems very satisfactory. But at least we should
581 decide which we want to do.
583 It's really not clear what is the Right Thing To Do. If you see
587 would you expect the value of ?y to be got from the *occurrence sites*
588 of 'z', or from the valuue of ?y at the *definition* of 'z'? In the
589 case of function definitions, the answer is clearly the former, but
590 less so in the case of non-fucntion definitions. On the other hand,
591 if we say that we get the value of ?y from the definition site of 'z',
592 then inlining 'z' might change the semantics of the program.
594 Choice (C) really says "the monomorphism restriction doesn't apply
595 to implicit parameters". Which is fine, but remember that every
596 innocent binding 'x = ...' that mentions an implicit parameter in
597 the RHS becomes a *function* of that parameter, called at each
598 use of 'x'. Now, the chances are that there are no intervening 'with'
599 clauses that bind ?y, so a decent compiler should common up all
600 those function calls. So I think I strongly favour (C). Indeed,
601 one could make a similar argument for abolishing the monomorphism
602 restriction altogether.
604 BOTTOM LINE: we choose (B) at present. See tcSimplifyRestricted
608 %************************************************************************
610 \subsection{tcSimplifyInfer}
612 %************************************************************************
614 tcSimplify is called when we *inferring* a type. Here's the overall game plan:
616 1. Compute Q = grow( fvs(T), C )
618 2. Partition C based on Q into Ct and Cq. Notice that ambiguous
619 predicates will end up in Ct; we deal with them at the top level
621 3. Try improvement, using functional dependencies
623 4. If Step 3 did any unification, repeat from step 1
624 (Unification can change the result of 'grow'.)
626 Note: we don't reduce dictionaries in step 2. For example, if we have
627 Eq (a,b), we don't simplify to (Eq a, Eq b). So Q won't be different
628 after step 2. However note that we may therefore quantify over more
629 type variables than we absolutely have to.
631 For the guts, we need a loop, that alternates context reduction and
632 improvement with unification. E.g. Suppose we have
634 class C x y | x->y where ...
636 and tcSimplify is called with:
638 Then improvement unifies a with b, giving
641 If we need to unify anything, we rattle round the whole thing all over
648 -> TcTyVarSet -- fv(T); type vars
650 -> TcM ([TcTyVar], -- Tyvars to quantify (zonked and quantified)
651 [Inst], -- Dict Ids that must be bound here (zonked)
652 TcDictBinds) -- Bindings
653 -- Any free (escaping) Insts are tossed into the environment
658 tcSimplifyInfer doc tau_tvs wanted
659 = do { tau_tvs1 <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
660 ; wanted' <- mapM zonkInst wanted -- Zonk before deciding quantified tyvars
661 ; gbl_tvs <- tcGetGlobalTyVars
662 ; let preds1 = fdPredsOfInsts wanted'
663 gbl_tvs1 = oclose preds1 gbl_tvs
664 qtvs = grow preds1 tau_tvs1 `minusVarSet` gbl_tvs1
665 -- See Note [Choosing which variables to quantify]
667 -- To maximise sharing, remove from consideration any
668 -- constraints that don't mention qtvs at all
669 ; let (free, bound) = partition (isFreeWhenInferring qtvs) wanted'
672 -- To make types simple, reduce as much as possible
673 ; traceTc (text "infer" <+> (ppr preds1 $$ ppr (grow preds1 tau_tvs1) $$ ppr gbl_tvs $$
674 ppr gbl_tvs1 $$ ppr free $$ ppr bound))
675 ; (irreds1, binds1) <- tryHardCheckLoop doc bound
677 -- Note [Inference and implication constraints]
678 ; let want_dict d = tyVarsOfInst d `intersectsVarSet` qtvs
679 ; (irreds2, binds2) <- approximateImplications doc want_dict irreds1
681 -- Now work out all over again which type variables to quantify,
682 -- exactly in the same way as before, but starting from irreds2. Why?
683 -- a) By now improvment may have taken place, and we must *not*
684 -- quantify over any variable free in the environment
685 -- tc137 (function h inside g) is an example
687 -- b) Do not quantify over constraints that *now* do not
688 -- mention quantified type variables, because they are
689 -- simply ambiguous (or might be bound further out). Example:
690 -- f :: Eq b => a -> (a, b)
692 -- From the RHS of g we get the MethodInst f77 :: alpha -> (alpha, beta)
693 -- We decide to quantify over 'alpha' alone, but free1 does not include f77
694 -- because f77 mentions 'alpha'. Then reducing leaves only the (ambiguous)
695 -- constraint (Eq beta), which we dump back into the free set
696 -- See test tcfail181
698 -- c) irreds may contain type variables not previously mentioned,
699 -- e.g. instance D a x => Foo [a]
701 -- Then after simplifying we'll get (D a x), and x is fresh
702 -- We must quantify over x else it'll be totally unbound
703 ; tau_tvs2 <- zonkTcTyVarsAndFV (varSetElems tau_tvs1)
704 ; gbl_tvs2 <- zonkTcTyVarsAndFV (varSetElems gbl_tvs1)
705 -- Note that we start from gbl_tvs1
706 -- We use tcGetGlobalTyVars, then oclose wrt preds2, because
707 -- we've already put some of the original preds1 into frees
708 -- E.g. wanteds = C a b (where a->b)
711 -- Then b is fixed by gbl_tvs, so (C a b) will be in free, and
712 -- irreds2 will be empty. But we don't want to generalise over b!
713 ; let preds2 = fdPredsOfInsts irreds2 -- irreds2 is zonked
714 qtvs = grow preds2 tau_tvs2 `minusVarSet` oclose preds2 gbl_tvs2
715 ; let (free, irreds3) = partition (isFreeWhenInferring qtvs) irreds2
718 -- Turn the quantified meta-type variables into real type variables
719 ; qtvs2 <- zonkQuantifiedTyVars (varSetElems qtvs)
721 -- We can't abstract over any remaining unsolved
722 -- implications so instead just float them outwards. Ugh.
723 ; let (q_dicts0, implics) = partition isAbstractableInst irreds3
724 ; loc <- getInstLoc (ImplicOrigin doc)
725 ; implic_bind <- bindIrreds loc qtvs2 q_dicts0 implics
727 -- Prepare equality instances for quantification
728 ; let (q_eqs0,q_dicts) = partition isEqInst q_dicts0
729 ; q_eqs <- mapM finalizeEqInst q_eqs0
731 ; return (qtvs2, q_eqs ++ q_dicts, binds1 `unionBags` binds2 `unionBags` implic_bind) }
732 -- NB: when we are done, we might have some bindings, but
733 -- the final qtvs might be empty. See Note [NO TYVARS] below.
735 approximateImplications :: SDoc -> (Inst -> Bool) -> [Inst] -> TcM ([Inst], TcDictBinds)
736 -- Note [Inference and implication constraints]
737 -- Given a bunch of Dict and ImplicInsts, try to approximate the implications by
738 -- - fetching any dicts inside them that are free
739 -- - using those dicts as cruder constraints, to solve the implications
740 -- - returning the extra ones too
742 approximateImplications doc want_dict irreds
744 = return (irreds, emptyBag)
746 = do { extra_dicts' <- mapM cloneDict extra_dicts
747 ; tryHardCheckLoop doc (extra_dicts' ++ irreds) }
748 -- By adding extra_dicts', we make them
749 -- available to solve the implication constraints
751 extra_dicts = get_dicts (filter isImplicInst irreds)
753 get_dicts :: [Inst] -> [Inst] -- Returns only Dicts
754 -- Find the wanted constraints in implication constraints that satisfy
755 -- want_dict, and are not bound by forall's in the constraint itself
756 get_dicts ds = concatMap get_dict ds
758 get_dict d@(Dict {}) | want_dict d = [d]
760 get_dict (ImplicInst {tci_tyvars = tvs, tci_wanted = wanteds})
761 = [ d | let tv_set = mkVarSet tvs
762 , d <- get_dicts wanteds
763 , not (tyVarsOfInst d `intersectsVarSet` tv_set)]
764 get_dict i@(EqInst {}) | want_dict i = [i]
766 get_dict other = pprPanic "approximateImplications" (ppr other)
769 Note [Inference and implication constraints]
770 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
771 Suppose we have a wanted implication constraint (perhaps arising from
772 a nested pattern match) like
774 and we are now trying to quantify over 'a' when inferring the type for
775 a function. In principle it's possible that there might be an instance
776 instance (C a, E a) => D [a]
777 so the context (E a) would suffice. The Right Thing is to abstract over
778 the implication constraint, but we don't do that (a) because it'll be
779 surprising to programmers and (b) because we don't have the machinery to deal
780 with 'given' implications.
782 So our best approximation is to make (D [a]) part of the inferred
783 context, so we can use that to discharge the implication. Hence
784 the strange function get_dicts in approximateImplications.
786 The common cases are more clear-cut, when we have things like
788 Here, abstracting over (C b) is not an approximation at all -- but see
789 Note [Freeness and implications].
791 See Trac #1430 and test tc228.
795 -----------------------------------------------------------
796 -- tcSimplifyInferCheck is used when we know the constraints we are to simplify
797 -- against, but we don't know the type variables over which we are going to quantify.
798 -- This happens when we have a type signature for a mutually recursive group
801 -> TcTyVarSet -- fv(T)
804 -> TcM ([TyVar], -- Fully zonked, and quantified
805 TcDictBinds) -- Bindings
807 tcSimplifyInferCheck loc tau_tvs givens wanteds
808 = do { traceTc (text "tcSimplifyInferCheck <-" <+> ppr wanteds)
809 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
811 -- Figure out which type variables to quantify over
812 -- You might think it should just be the signature tyvars,
813 -- but in bizarre cases you can get extra ones
814 -- f :: forall a. Num a => a -> a
815 -- f x = fst (g (x, head [])) + 1
817 -- Here we infer g :: forall a b. a -> b -> (b,a)
818 -- We don't want g to be monomorphic in b just because
819 -- f isn't quantified over b.
820 ; let all_tvs = varSetElems (tau_tvs `unionVarSet` tyVarsOfInsts givens)
821 ; all_tvs <- zonkTcTyVarsAndFV all_tvs
822 ; gbl_tvs <- tcGetGlobalTyVars
823 ; let qtvs = varSetElems (all_tvs `minusVarSet` gbl_tvs)
824 -- We could close gbl_tvs, but its not necessary for
825 -- soundness, and it'll only affect which tyvars, not which
826 -- dictionaries, we quantify over
828 ; qtvs' <- zonkQuantifiedTyVars qtvs
830 -- Now we are back to normal (c.f. tcSimplCheck)
831 ; implic_bind <- bindIrreds loc qtvs' givens irreds
833 ; traceTc (text "tcSimplifyInferCheck ->" <+> ppr (implic_bind))
834 ; return (qtvs', binds `unionBags` implic_bind) }
837 Note [Squashing methods]
838 ~~~~~~~~~~~~~~~~~~~~~~~~~
839 Be careful if you want to float methods more:
840 truncate :: forall a. RealFrac a => forall b. Integral b => a -> b
841 From an application (truncate f i) we get
844 If we have also have a second occurrence of truncate, we get
847 When simplifying with i,f free, we might still notice that
848 t1=t3; but alas, the binding for t2 (which mentions t1)
849 may continue to float out!
854 class Y a b | a -> b where
857 instance Y [[a]] a where
860 k :: X a -> X a -> X a
862 g :: Num a => [X a] -> [X a]
865 h ys = ys ++ map (k (y [[0]])) xs
867 The excitement comes when simplifying the bindings for h. Initially
868 try to simplify {y @ [[t1]] t2, 0 @ t1}, with initial qtvs = {t2}.
869 From this we get t1:=:t2, but also various bindings. We can't forget
870 the bindings (because of [LOOP]), but in fact t1 is what g is
873 The net effect of [NO TYVARS]
876 isFreeWhenInferring :: TyVarSet -> Inst -> Bool
877 isFreeWhenInferring qtvs inst
878 = isFreeWrtTyVars qtvs inst -- Constrains no quantified vars
879 && isInheritableInst inst -- and no implicit parameter involved
880 -- see Note [Inheriting implicit parameters]
882 {- No longer used (with implication constraints)
883 isFreeWhenChecking :: TyVarSet -- Quantified tyvars
884 -> NameSet -- Quantified implicit parameters
886 isFreeWhenChecking qtvs ips inst
887 = isFreeWrtTyVars qtvs inst
888 && isFreeWrtIPs ips inst
891 isFreeWrtTyVars qtvs inst = tyVarsOfInst inst `disjointVarSet` qtvs
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
925 -> [TcTyVar] -- Quantify over these
928 -> TcM TcDictBinds -- Bindings
929 tcSimplifyCheckPat loc co_vars qtvs givens wanteds
930 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
931 do { traceTc (text "tcSimplifyCheckPat")
932 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
933 ; implic_bind <- bindIrredsR loc qtvs co_vars emptyRefinement
935 ; return (binds `unionBags` implic_bind) }
937 -----------------------------------------------------------
938 bindIrreds :: InstLoc -> [TcTyVar]
941 bindIrreds loc qtvs givens irreds
942 = bindIrredsR loc qtvs [] emptyRefinement givens irreds
944 bindIrredsR :: InstLoc -> [TcTyVar] -> [CoVar]
945 -> Refinement -> [Inst] -> [Inst]
947 -- Make a binding that binds 'irreds', by generating an implication
948 -- constraint for them, *and* throwing the constraint into the LIE
949 bindIrredsR loc qtvs co_vars reft givens irreds
953 = do { let givens' = filter isAbstractableInst givens
954 -- The givens can (redundantly) include methods
955 -- We want to retain both EqInsts and Dicts
956 -- There should be no implicadtion constraints
957 -- See Note [Pruning the givens in an implication constraint]
959 -- If there are no 'givens' *and* the refinement is empty
960 -- (the refinement is like more givens), then it's safe to
961 -- partition the 'wanteds' by their qtvs, thereby trimming irreds
962 -- See Note [Freeness and implications]
963 ; irreds' <- if null givens' && isEmptyRefinement reft
965 { let qtv_set = mkVarSet qtvs
966 (frees, real_irreds) = partition (isFreeWrtTyVars qtv_set) irreds
968 ; return real_irreds }
971 ; let all_tvs = qtvs ++ co_vars -- Abstract over all these
972 ; (implics, bind) <- makeImplicationBind loc all_tvs reft givens' irreds'
973 -- This call does the real work
974 -- If irreds' is empty, it does something sensible
979 makeImplicationBind :: InstLoc -> [TcTyVar] -> Refinement
981 -> TcM ([Inst], TcDictBinds)
982 -- Make a binding that binds 'irreds', by generating an implication
983 -- constraint for them, *and* throwing the constraint into the LIE
984 -- The binding looks like
985 -- (ir1, .., irn) = f qtvs givens
986 -- where f is (evidence for) the new implication constraint
987 -- f :: forall qtvs. {reft} givens => (ir1, .., irn)
988 -- qtvs includes coercion variables
990 -- This binding must line up the 'rhs' in reduceImplication
991 makeImplicationBind loc all_tvs reft
992 givens -- Guaranteed all Dicts
995 | null irreds -- If there are no irreds, we are done
996 = return ([], emptyBag)
997 | otherwise -- Otherwise we must generate a binding
998 = do { uniq <- newUnique
999 ; span <- getSrcSpanM
1000 ; let (eq_givens, dict_givens) = partition isEqInst givens
1001 eq_tyvar_cos = mkTyVarTys (varSetElems $ tyVarsOfTypes $ map eqInstType eq_givens)
1002 -- Urgh! See line 2187 or thereabouts. I believe that all these
1003 -- 'givens' must be a simple CoVar. This MUST be cleaned up.
1005 ; let name = mkInternalName uniq (mkVarOcc "ic") span
1006 implic_inst = ImplicInst { tci_name = name, tci_reft = reft,
1007 tci_tyvars = all_tvs,
1008 tci_given = (eq_givens ++ dict_givens),
1009 tci_wanted = irreds, tci_loc = loc }
1010 ; let -- only create binder for dict_irreds
1011 (eq_irreds, dict_irreds) = partition isEqInst irreds
1012 n_dict_irreds = length dict_irreds
1013 dict_irred_ids = map instToId dict_irreds
1014 tup_ty = mkTupleTy Boxed n_dict_irreds (map idType dict_irred_ids)
1015 pat = TuplePat (map nlVarPat dict_irred_ids) Boxed tup_ty
1016 rhs = L span (mkHsWrap co (HsVar (instToId implic_inst)))
1017 co = mkWpApps (map instToId dict_givens)
1018 <.> mkWpTyApps eq_tyvar_cos
1019 <.> mkWpTyApps (mkTyVarTys all_tvs)
1020 bind | [dict_irred_id] <- dict_irred_ids = VarBind dict_irred_id rhs
1021 | otherwise = PatBind { pat_lhs = L span pat,
1022 pat_rhs = unguardedGRHSs rhs,
1023 pat_rhs_ty = tup_ty,
1024 bind_fvs = placeHolderNames }
1025 ; traceTc $ text "makeImplicationBind" <+> ppr implic_inst
1026 ; return ([implic_inst], unitBag (L span bind))
1029 -----------------------------------------------------------
1030 tryHardCheckLoop :: SDoc
1032 -> TcM ([Inst], TcDictBinds)
1034 tryHardCheckLoop doc wanteds
1035 = do { (irreds,binds) <- checkLoop (mkRedEnv doc try_me []) wanteds
1036 ; return (irreds,binds)
1039 try_me inst = ReduceMe AddSCs
1040 -- Here's the try-hard bit
1042 -----------------------------------------------------------
1043 gentleCheckLoop :: InstLoc
1046 -> TcM ([Inst], TcDictBinds)
1048 gentleCheckLoop inst_loc givens wanteds
1049 = do { (irreds,binds) <- checkLoop env wanteds
1050 ; return (irreds,binds)
1053 env = mkRedEnv (pprInstLoc inst_loc) try_me givens
1055 try_me inst | isMethodOrLit inst = ReduceMe AddSCs
1057 -- When checking against a given signature
1058 -- we MUST be very gentle: Note [Check gently]
1060 gentleInferLoop :: SDoc -> [Inst]
1061 -> TcM ([Inst], TcDictBinds)
1062 gentleInferLoop doc wanteds
1063 = do { (irreds, binds) <- checkLoop env wanteds
1064 ; return (irreds, binds) }
1066 env = mkRedEnv doc try_me []
1067 try_me inst | isMethodOrLit inst = ReduceMe AddSCs
1072 ~~~~~~~~~~~~~~~~~~~~
1073 We have to very careful about not simplifying too vigorously
1078 f :: Show b => T b -> b
1079 f (MkT x) = show [x]
1081 Inside the pattern match, which binds (a:*, x:a), we know that
1083 Hence we have a dictionary for Show [a] available; and indeed we
1084 need it. We are going to build an implication contraint
1085 forall a. (b~[a]) => Show [a]
1086 Later, we will solve this constraint using the knowledge (Show b)
1088 But we MUST NOT reduce (Show [a]) to (Show a), else the whole
1089 thing becomes insoluble. So we simplify gently (get rid of literals
1090 and methods only, plus common up equal things), deferring the real
1091 work until top level, when we solve the implication constraint
1092 with tryHardCheckLooop.
1096 -----------------------------------------------------------
1099 -> TcM ([Inst], TcDictBinds)
1100 -- Precondition: givens are completely rigid
1101 -- Postcondition: returned Insts are zonked
1103 checkLoop env wanteds
1104 = go env wanteds (return ())
1105 where go env wanteds elim_skolems
1106 = do { -- We do need to zonk the givens; cf Note [Zonking RedEnv]
1107 ; env' <- zonkRedEnv env
1108 ; wanteds' <- zonkInsts wanteds
1110 ; (improved, binds, irreds, elim_more_skolems)
1111 <- reduceContext env' wanteds'
1112 ; let elim_skolems' = elim_skolems >> elim_more_skolems
1114 ; if not improved then
1115 elim_skolems' >> 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 elim_skolems'
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 TcGadt.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.
1218 Solution: never generate a superclass selectors at all when
1219 satisfying the superclass context of an instance declaration.
1221 Two more nasty cases are in
1226 tcSimplifySuperClasses
1231 tcSimplifySuperClasses loc givens sc_wanteds
1232 = do { traceTc (text "tcSimplifySuperClasses")
1233 ; (irreds,binds1) <- checkLoop env sc_wanteds
1234 ; let (tidy_env, tidy_irreds) = tidyInsts irreds
1235 ; reportNoInstances tidy_env (Just (loc, givens)) tidy_irreds
1238 env = mkRedEnv (pprInstLoc loc) try_me givens
1239 try_me inst = ReduceMe NoSCs
1240 -- Like tryHardCheckLoop, but with NoSCs
1244 %************************************************************************
1246 \subsection{tcSimplifyRestricted}
1248 %************************************************************************
1250 tcSimplifyRestricted infers which type variables to quantify for a
1251 group of restricted bindings. This isn't trivial.
1254 We want to quantify over a to get id :: forall a. a->a
1257 We do not want to quantify over a, because there's an Eq a
1258 constraint, so we get eq :: a->a->Bool (notice no forall)
1261 RHS has type 'tau', whose free tyvars are tau_tvs
1262 RHS has constraints 'wanteds'
1265 Quantify over (tau_tvs \ ftvs(wanteds))
1266 This is bad. The constraints may contain (Monad (ST s))
1267 where we have instance Monad (ST s) where...
1268 so there's no need to be monomorphic in s!
1270 Also the constraint might be a method constraint,
1271 whose type mentions a perfectly innocent tyvar:
1272 op :: Num a => a -> b -> a
1273 Here, b is unconstrained. A good example would be
1275 We want to infer the polymorphic type
1276 foo :: forall b. b -> b
1279 Plan B (cunning, used for a long time up to and including GHC 6.2)
1280 Step 1: Simplify the constraints as much as possible (to deal
1281 with Plan A's problem). Then set
1282 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1284 Step 2: Now simplify again, treating the constraint as 'free' if
1285 it does not mention qtvs, and trying to reduce it otherwise.
1286 The reasons for this is to maximise sharing.
1288 This fails for a very subtle reason. Suppose that in the Step 2
1289 a constraint (Foo (Succ Zero) (Succ Zero) b) gets thrown upstairs as 'free'.
1290 In the Step 1 this constraint might have been simplified, perhaps to
1291 (Foo Zero Zero b), AND THEN THAT MIGHT BE IMPROVED, to bind 'b' to 'T'.
1292 This won't happen in Step 2... but that in turn might prevent some other
1293 constraint (Baz [a] b) being simplified (e.g. via instance Baz [a] T where {..})
1294 and that in turn breaks the invariant that no constraints are quantified over.
1296 Test typecheck/should_compile/tc177 (which failed in GHC 6.2) demonstrates
1301 Step 1: Simplify the constraints as much as possible (to deal
1302 with Plan A's problem). Then set
1303 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1304 Return the bindings from Step 1.
1307 A note about Plan C (arising from "bug" reported by George Russel March 2004)
1310 instance (HasBinary ty IO) => HasCodedValue ty
1312 foo :: HasCodedValue a => String -> IO a
1314 doDecodeIO :: HasCodedValue a => () -> () -> IO a
1315 doDecodeIO codedValue view
1316 = let { act = foo "foo" } in act
1318 You might think this should work becuase the call to foo gives rise to a constraint
1319 (HasCodedValue t), which can be satisfied by the type sig for doDecodeIO. But the
1320 restricted binding act = ... calls tcSimplifyRestricted, and PlanC simplifies the
1321 constraint using the (rather bogus) instance declaration, and now we are stuffed.
1323 I claim this is not really a bug -- but it bit Sergey as well as George. So here's
1327 Plan D (a variant of plan B)
1328 Step 1: Simplify the constraints as much as possible (to deal
1329 with Plan A's problem), BUT DO NO IMPROVEMENT. Then set
1330 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1332 Step 2: Now simplify again, treating the constraint as 'free' if
1333 it does not mention qtvs, and trying to reduce it otherwise.
1335 The point here is that it's generally OK to have too few qtvs; that is,
1336 to make the thing more monomorphic than it could be. We don't want to
1337 do that in the common cases, but in wierd cases it's ok: the programmer
1338 can always add a signature.
1340 Too few qtvs => too many wanteds, which is what happens if you do less
1345 tcSimplifyRestricted -- Used for restricted binding groups
1346 -- i.e. ones subject to the monomorphism restriction
1349 -> [Name] -- Things bound in this group
1350 -> TcTyVarSet -- Free in the type of the RHSs
1351 -> [Inst] -- Free in the RHSs
1352 -> TcM ([TyVar], -- Tyvars to quantify (zonked and quantified)
1353 TcDictBinds) -- Bindings
1354 -- tcSimpifyRestricted returns no constraints to
1355 -- quantify over; by definition there are none.
1356 -- They are all thrown back in the LIE
1358 tcSimplifyRestricted doc top_lvl bndrs tau_tvs wanteds
1359 -- Zonk everything in sight
1360 = do { traceTc (text "tcSimplifyRestricted")
1361 ; wanteds' <- zonkInsts wanteds
1363 -- 'ReduceMe': Reduce as far as we can. Don't stop at
1364 -- dicts; the idea is to get rid of as many type
1365 -- variables as possible, and we don't want to stop
1366 -- at (say) Monad (ST s), because that reduces
1367 -- immediately, with no constraint on s.
1369 -- BUT do no improvement! See Plan D above
1370 -- HOWEVER, some unification may take place, if we instantiate
1371 -- a method Inst with an equality constraint
1372 ; let env = mkNoImproveRedEnv doc (\i -> ReduceMe AddSCs)
1373 ; (_imp, _binds, constrained_dicts, elim_skolems)
1374 <- reduceContext env wanteds'
1377 -- Next, figure out the tyvars we will quantify over
1378 ; tau_tvs' <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
1379 ; gbl_tvs' <- tcGetGlobalTyVars
1380 ; constrained_dicts' <- zonkInsts constrained_dicts
1382 ; let qtvs1 = tau_tvs' `minusVarSet` oclose (fdPredsOfInsts constrained_dicts) gbl_tvs'
1383 -- As in tcSimplifyInfer
1385 -- Do not quantify over constrained type variables:
1386 -- this is the monomorphism restriction
1387 constrained_tvs' = tyVarsOfInsts constrained_dicts'
1388 qtvs = qtvs1 `minusVarSet` constrained_tvs'
1389 pp_bndrs = pprWithCommas (quotes . ppr) bndrs
1392 ; warn_mono <- doptM Opt_WarnMonomorphism
1393 ; warnTc (warn_mono && (constrained_tvs' `intersectsVarSet` qtvs1))
1394 (vcat[ ptext SLIT("the Monomorphism Restriction applies to the binding")
1395 <> plural bndrs <+> ptext SLIT("for") <+> pp_bndrs,
1396 ptext SLIT("Consider giving a type signature for") <+> pp_bndrs])
1398 ; traceTc (text "tcSimplifyRestricted" <+> vcat [
1399 pprInsts wanteds, pprInsts constrained_dicts',
1401 ppr constrained_tvs', ppr tau_tvs', ppr qtvs ])
1403 -- The first step may have squashed more methods than
1404 -- necessary, so try again, this time more gently, knowing the exact
1405 -- set of type variables to quantify over.
1407 -- We quantify only over constraints that are captured by qtvs;
1408 -- these will just be a subset of non-dicts. This in contrast
1409 -- to normal inference (using isFreeWhenInferring) in which we quantify over
1410 -- all *non-inheritable* constraints too. This implements choice
1411 -- (B) under "implicit parameter and monomorphism" above.
1413 -- Remember that we may need to do *some* simplification, to
1414 -- (for example) squash {Monad (ST s)} into {}. It's not enough
1415 -- just to float all constraints
1417 -- At top level, we *do* squash methods becuase we want to
1418 -- expose implicit parameters to the test that follows
1419 ; let is_nested_group = isNotTopLevel top_lvl
1420 try_me inst | isFreeWrtTyVars qtvs inst,
1421 (is_nested_group || isDict inst) = Stop
1422 | otherwise = ReduceMe AddSCs
1423 env = mkNoImproveRedEnv doc try_me
1424 ; (_imp, binds, irreds, elim_skolems) <- reduceContext env wanteds'
1427 -- See "Notes on implicit parameters, Question 4: top level"
1428 ; ASSERT( all (isFreeWrtTyVars qtvs) irreds ) -- None should be captured
1429 if is_nested_group then
1431 else do { let (bad_ips, non_ips) = partition isIPDict irreds
1432 ; addTopIPErrs bndrs bad_ips
1433 ; extendLIEs non_ips }
1435 ; qtvs' <- zonkQuantifiedTyVars (varSetElems qtvs)
1436 ; return (qtvs', binds) }
1440 %************************************************************************
1444 %************************************************************************
1446 On the LHS of transformation rules we only simplify methods and constants,
1447 getting dictionaries. We want to keep all of them unsimplified, to serve
1448 as the available stuff for the RHS of the rule.
1450 Example. Consider the following left-hand side of a rule
1452 f (x == y) (y > z) = ...
1454 If we typecheck this expression we get constraints
1456 d1 :: Ord a, d2 :: Eq a
1458 We do NOT want to "simplify" to the LHS
1460 forall x::a, y::a, z::a, d1::Ord a.
1461 f ((==) (eqFromOrd d1) x y) ((>) d1 y z) = ...
1465 forall x::a, y::a, z::a, d1::Ord a, d2::Eq a.
1466 f ((==) d2 x y) ((>) d1 y z) = ...
1468 Here is another example:
1470 fromIntegral :: (Integral a, Num b) => a -> b
1471 {-# RULES "foo" fromIntegral = id :: Int -> Int #-}
1473 In the rule, a=b=Int, and Num Int is a superclass of Integral Int. But
1474 we *dont* want to get
1476 forall dIntegralInt.
1477 fromIntegral Int Int dIntegralInt (scsel dIntegralInt) = id Int
1479 because the scsel will mess up RULE matching. Instead we want
1481 forall dIntegralInt, dNumInt.
1482 fromIntegral Int Int dIntegralInt dNumInt = id Int
1486 g (x == y) (y == z) = ..
1488 where the two dictionaries are *identical*, we do NOT WANT
1490 forall x::a, y::a, z::a, d1::Eq a
1491 f ((==) d1 x y) ((>) d1 y z) = ...
1493 because that will only match if the dict args are (visibly) equal.
1494 Instead we want to quantify over the dictionaries separately.
1496 In short, tcSimplifyRuleLhs must *only* squash LitInst and MethInts, leaving
1497 all dicts unchanged, with absolutely no sharing. It's simpler to do this
1498 from scratch, rather than further parameterise simpleReduceLoop etc
1501 tcSimplifyRuleLhs :: [Inst] -> TcM ([Inst], TcDictBinds)
1502 tcSimplifyRuleLhs wanteds
1503 = go [] emptyBag wanteds
1506 = return (dicts, binds)
1507 go dicts binds (w:ws)
1509 = go (w:dicts) binds ws
1511 = do { w' <- zonkInst w -- So that (3::Int) does not generate a call
1512 -- to fromInteger; this looks fragile to me
1513 ; lookup_result <- lookupSimpleInst w'
1514 ; case lookup_result of
1516 go dicts (addInstToDictBind binds w rhs) (ws' ++ ws)
1517 NoInstance -> pprPanic "tcSimplifyRuleLhs" (ppr w)
1521 tcSimplifyBracket is used when simplifying the constraints arising from
1522 a Template Haskell bracket [| ... |]. We want to check that there aren't
1523 any constraints that can't be satisfied (e.g. Show Foo, where Foo has no
1524 Show instance), but we aren't otherwise interested in the results.
1525 Nor do we care about ambiguous dictionaries etc. We will type check
1526 this bracket again at its usage site.
1529 tcSimplifyBracket :: [Inst] -> TcM ()
1530 tcSimplifyBracket wanteds
1531 = do { tryHardCheckLoop doc wanteds
1534 doc = text "tcSimplifyBracket"
1538 %************************************************************************
1540 \subsection{Filtering at a dynamic binding}
1542 %************************************************************************
1547 we must discharge all the ?x constraints from B. We also do an improvement
1548 step; if we have ?x::t1 and ?x::t2 we must unify t1, t2.
1550 Actually, the constraints from B might improve the types in ?x. For example
1552 f :: (?x::Int) => Char -> Char
1555 then the constraint (?x::Int) arising from the call to f will
1556 force the binding for ?x to be of type Int.
1559 tcSimplifyIPs :: [Inst] -- The implicit parameters bound here
1562 -- We need a loop so that we do improvement, and then
1563 -- (next time round) generate a binding to connect the two
1565 -- Here the two ?x's have different types, and improvement
1566 -- makes them the same.
1568 tcSimplifyIPs given_ips wanteds
1569 = do { wanteds' <- zonkInsts wanteds
1570 ; given_ips' <- zonkInsts given_ips
1571 -- Unusually for checking, we *must* zonk the given_ips
1573 ; let env = mkRedEnv doc try_me given_ips'
1574 ; (improved, binds, irreds, elim_skolems) <- reduceContext env wanteds'
1577 ; if not improved then
1578 ASSERT( all is_free irreds )
1579 do { extendLIEs irreds
1582 tcSimplifyIPs given_ips wanteds }
1584 doc = text "tcSimplifyIPs" <+> ppr given_ips
1585 ip_set = mkNameSet (ipNamesOfInsts given_ips)
1586 is_free inst = isFreeWrtIPs ip_set inst
1588 -- Simplify any methods that mention the implicit parameter
1589 try_me inst | is_free inst = Stop
1590 | otherwise = ReduceMe NoSCs
1594 %************************************************************************
1596 \subsection[binds-for-local-funs]{@bindInstsOfLocalFuns@}
1598 %************************************************************************
1600 When doing a binding group, we may have @Insts@ of local functions.
1601 For example, we might have...
1603 let f x = x + 1 -- orig local function (overloaded)
1604 f.1 = f Int -- two instances of f
1609 The point is: we must drop the bindings for @f.1@ and @f.2@ here,
1610 where @f@ is in scope; those @Insts@ must certainly not be passed
1611 upwards towards the top-level. If the @Insts@ were binding-ified up
1612 there, they would have unresolvable references to @f@.
1614 We pass in an @init_lie@ of @Insts@ and a list of locally-bound @Ids@.
1615 For each method @Inst@ in the @init_lie@ that mentions one of the
1616 @Ids@, we create a binding. We return the remaining @Insts@ (in an
1617 @LIE@), as well as the @HsBinds@ generated.
1620 bindInstsOfLocalFuns :: [Inst] -> [TcId] -> TcM TcDictBinds
1621 -- Simlifies only MethodInsts, and generate only bindings of form
1623 -- We're careful not to even generate bindings of the form
1625 -- You'd think that'd be fine, but it interacts with what is
1626 -- arguably a bug in Match.tidyEqnInfo (see notes there)
1628 bindInstsOfLocalFuns wanteds local_ids
1629 | null overloaded_ids = do
1632 return emptyLHsBinds
1635 = do { (irreds, binds) <- gentleInferLoop doc for_me
1636 ; extendLIEs not_for_me
1640 doc = text "bindInsts" <+> ppr local_ids
1641 overloaded_ids = filter is_overloaded local_ids
1642 is_overloaded id = isOverloadedTy (idType id)
1643 (for_me, not_for_me) = partition (isMethodFor overloaded_set) wanteds
1645 overloaded_set = mkVarSet overloaded_ids -- There can occasionally be a lot of them
1646 -- so it's worth building a set, so that
1647 -- lookup (in isMethodFor) is faster
1651 %************************************************************************
1653 \subsection{Data types for the reduction mechanism}
1655 %************************************************************************
1657 The main control over context reduction is here
1661 = RedEnv { red_doc :: SDoc -- The context
1662 , red_try_me :: Inst -> WhatToDo
1663 , red_improve :: Bool -- True <=> do improvement
1664 , red_givens :: [Inst] -- All guaranteed rigid
1666 -- but see Note [Rigidity]
1667 , red_reft :: Refinement -- The refinement to apply to the 'givens'
1668 -- You should think of it as 'given equalities'
1669 , red_stack :: (Int, [Inst]) -- Recursion stack (for err msg)
1670 -- See Note [RedStack]
1674 -- The red_givens are rigid so far as cmpInst is concerned.
1675 -- There is one case where they are not totally rigid, namely in tcSimplifyIPs
1676 -- let ?x = e in ...
1677 -- Here, the given is (?x::a), where 'a' is not necy a rigid type
1678 -- But that doesn't affect the comparison, which is based only on mame.
1681 -- The red_stack pair (n,insts) pair is just used for error reporting.
1682 -- 'n' is always the depth of the stack.
1683 -- The 'insts' is the stack of Insts being reduced: to produce X
1684 -- I had to produce Y, to produce Y I had to produce Z, and so on.
1687 mkRedEnv :: SDoc -> (Inst -> WhatToDo) -> [Inst] -> RedEnv
1688 mkRedEnv doc try_me givens
1689 = RedEnv { red_doc = doc, red_try_me = try_me,
1690 red_givens = givens,
1691 red_reft = emptyRefinement,
1693 red_improve = True }
1695 mkNoImproveRedEnv :: SDoc -> (Inst -> WhatToDo) -> RedEnv
1696 -- Do not do improvement; no givens
1697 mkNoImproveRedEnv doc try_me
1698 = RedEnv { red_doc = doc, red_try_me = try_me,
1699 red_givens = [], red_reft = emptyRefinement,
1701 red_improve = True }
1704 = ReduceMe WantSCs -- Try to reduce this
1705 -- If there's no instance, add the inst to the
1706 -- irreductible ones, but don't produce an error
1707 -- message of any kind.
1708 -- It might be quite legitimate such as (Eq a)!
1710 | Stop -- Return as irreducible unless it can
1711 -- be reduced to a constant in one step
1712 -- Do not add superclasses; see
1714 data WantSCs = NoSCs | AddSCs -- Tells whether we should add the superclasses
1715 -- of a predicate when adding it to the avails
1716 -- The reason for this flag is entirely the super-class loop problem
1717 -- Note [SUPER-CLASS LOOP 1]
1719 zonkRedEnv :: RedEnv -> TcM RedEnv
1721 = do { givens' <- mapM zonkInst (red_givens env)
1722 ; return $ env {red_givens = givens'}
1727 %************************************************************************
1729 \subsection[reduce]{@reduce@}
1731 %************************************************************************
1733 Note [Ancestor Equalities]
1734 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1735 During context reduction, we add to the wanted equalities also those
1736 equalities that (transitively) occur in superclass contexts of wanted
1737 class constraints. Consider the following code
1739 class a ~ Int => C a
1742 If (C a) is wanted, we want to add (a ~ Int), which will be discharged by
1743 substituting Int for a. Hence, we ultimately want (C Int), which we
1744 discharge with the explicit instance.
1747 reduceContext :: RedEnv
1749 -> TcM (ImprovementDone,
1750 TcDictBinds, -- Dictionary bindings
1751 [Inst], -- Irreducible
1752 TcM ()) -- Undo skolems from SkolemOccurs
1754 reduceContext env wanteds
1755 = do { traceTc (text "reduceContext" <+> (vcat [
1756 text "----------------------",
1758 text "given" <+> ppr (red_givens env),
1759 text "wanted" <+> ppr wanteds,
1760 text "----------------------"
1764 ; let givens = red_givens env
1765 (given_eqs0, given_dicts0) = partition isEqInst givens
1766 (wanted_eqs0, wanted_non_eqs) = partition isEqInst wanteds
1767 (wanted_implics0, wanted_dicts0) = partition isImplicInst wanted_non_eqs
1769 -- We want to add as wanted equalities those that (transitively)
1770 -- occur in superclass contexts of wanted class constraints.
1771 -- See Note [Ancestor Equalities]
1772 ; ancestor_eqs <- ancestorEqualities wanted_dicts0
1773 ; let wanted_eqs = wanted_eqs0 ++ ancestor_eqs
1774 ; traceTc $ text "reduceContext: ancestor eqs" <+> ppr ancestor_eqs
1776 -- 1. Normalise the *given* *equality* constraints
1777 ; (given_eqs, eliminate_skolems) <- normaliseGivenEqs given_eqs0
1779 -- 2. Normalise the *given* *dictionary* constraints
1780 -- wrt. the toplevel and given equations
1781 ; (given_dicts, given_binds) <- normaliseGivenDicts given_eqs
1784 -- 5. Build the Avail mapping from "given_dicts"
1785 -- Add dicts refined by the current type refinement
1786 ; (init_state, extra_givens) <- getLIE $ do
1787 { init_state <- foldlM addGiven emptyAvails given_dicts
1788 ; let reft = red_reft env
1789 ; if isEmptyRefinement reft then return init_state
1790 else foldlM (addRefinedGiven reft)
1791 init_state given_dicts }
1793 -- *** ToDo: what to do with the "extra_givens"? For the
1794 -- moment I'm simply discarding them, which is probably wrong
1796 -- 7. Normalise the *wanted* *dictionary* constraints
1797 -- wrt. the toplevel and given equations
1798 -- NB: normalisation includes zonking as part of what it does
1799 -- so it's important to do it after any unifications
1800 -- that happened as a result of the addGivens
1801 ; (wanted_dicts,normalise_binds1) <- normaliseWantedDicts given_eqs wanted_dicts0
1803 -- 6. Solve the *wanted* *dictionary* constraints (not implications)
1804 -- This may expose some further equational constraints...
1805 ; (avails, extra_eqs) <- getLIE (reduceList env wanted_dicts init_state)
1806 ; (dict_binds, bound_dicts, dict_irreds) <- extractResults avails wanted_dicts
1807 ; traceTc $ text "reduceContext extractresults" <+> vcat
1808 [ppr avails,ppr wanted_dicts,ppr dict_binds]
1810 -- *** ToDo: what to do with the "extra_eqs"? For the
1811 -- moment I'm simply discarding them, which is probably wrong
1813 -- Solve the wanted *implications*. In doing so, we can provide
1814 -- as "given" all the dicts that were originally given,
1815 -- *or* for which we now have bindings,
1816 -- *or* which are now irreds
1817 ; let implic_env = env { red_givens = givens ++ bound_dicts ++ dict_irreds }
1818 ; (implic_binds_s, implic_irreds_s) <- mapAndUnzipM (reduceImplication implic_env) wanted_implics0
1819 ; let implic_binds = unionManyBags implic_binds_s
1820 implic_irreds = concat implic_irreds_s
1822 -- 3. Solve the *wanted* *equation* constraints
1823 ; eq_irreds0 <- solveWantedEqs given_eqs wanted_eqs
1825 -- 4. Normalise the *wanted* equality constraints with respect to
1827 ; eq_irreds <- normaliseWantedEqs eq_irreds0
1829 -- 8. Substitute the wanted *equations* in the wanted *dictionaries*
1830 ; let irreds = dict_irreds ++ implic_irreds
1831 ; (norm_irreds, normalise_binds2) <- substEqInDictInsts True {-wanted-}
1834 -- 9. eliminate the artificial skolem constants introduced in 1.
1835 -- ; eliminate_skolems
1837 -- Figure out whether we should go round again
1838 -- My current plan is to see if any of the mutable tyvars in
1839 -- givens or irreds has been filled in by improvement.
1840 -- If so, there is merit in going around again, because
1841 -- we may make further progress
1843 -- ToDo: is it only mutable stuff? We may have exposed new
1844 -- equality constraints and should probably go round again
1845 -- then as well. But currently we are dropping them on the
1848 ; let all_irreds = norm_irreds ++ eq_irreds
1849 ; improved <- anyM isFilledMetaTyVar $ varSetElems $
1850 tyVarsOfInsts (givens ++ all_irreds)
1852 -- The old plan (fragile)
1853 -- improveed = availsImproved avails
1854 -- || (not $ isEmptyBag normalise_binds1)
1855 -- || (not $ isEmptyBag normalise_binds2)
1856 -- || (any isEqInst irreds)
1858 ; traceTc (text "reduceContext end" <+> (vcat [
1859 text "----------------------",
1861 text "given" <+> ppr givens,
1862 text "given_eqs" <+> ppr given_eqs,
1863 text "wanted" <+> ppr wanteds,
1864 text "wanted_dicts" <+> ppr wanted_dicts,
1866 text "avails" <+> pprAvails avails,
1867 text "improved =" <+> ppr improved,
1868 text "(all) irreds = " <+> ppr all_irreds,
1869 text "dict-binds = " <+> ppr dict_binds,
1870 text "implic-binds = " <+> ppr implic_binds,
1871 text "----------------------"
1875 given_binds `unionBags` normalise_binds1
1876 `unionBags` normalise_binds2
1877 `unionBags` dict_binds
1878 `unionBags` implic_binds,
1883 tcImproveOne :: Avails -> Inst -> TcM ImprovementDone
1884 tcImproveOne avails inst
1885 | not (isDict inst) = return False
1887 = do { inst_envs <- tcGetInstEnvs
1888 ; let eqns = improveOne (classInstances inst_envs)
1889 (dictPred inst, pprInstArising inst)
1890 [ (dictPred p, pprInstArising p)
1891 | p <- availsInsts avails, isDict p ]
1892 -- Avails has all the superclasses etc (good)
1893 -- It also has all the intermediates of the deduction (good)
1894 -- It does not have duplicates (good)
1895 -- NB that (?x::t1) and (?x::t2) will be held separately in avails
1896 -- so that improve will see them separate
1897 ; traceTc (text "improveOne" <+> ppr inst)
1900 unifyEqns :: [(Equation,(PredType,SDoc),(PredType,SDoc))]
1901 -> TcM ImprovementDone
1902 unifyEqns [] = return False
1904 = do { traceTc (ptext SLIT("Improve:") <+> vcat (map pprEquationDoc eqns))
1908 unify ((qtvs, pairs), what1, what2)
1909 = addErrCtxtM (mkEqnMsg what1 what2) $ do
1910 (_, _, tenv) <- tcInstTyVars (varSetElems qtvs)
1911 mapM_ (unif_pr tenv) pairs
1912 unif_pr tenv (ty1,ty2) = unifyType (substTy tenv ty1) (substTy tenv ty2)
1914 pprEquationDoc (eqn, (p1,w1), (p2,w2)) = vcat [pprEquation eqn, nest 2 (ppr p1), nest 2 (ppr p2)]
1916 mkEqnMsg (pred1,from1) (pred2,from2) tidy_env
1917 = do { pred1' <- zonkTcPredType pred1; pred2' <- zonkTcPredType pred2
1918 ; let { pred1'' = tidyPred tidy_env pred1'; pred2'' = tidyPred tidy_env pred2' }
1919 ; let msg = vcat [ptext SLIT("When using functional dependencies to combine"),
1920 nest 2 (sep [ppr pred1'' <> comma, nest 2 from1]),
1921 nest 2 (sep [ppr pred2'' <> comma, nest 2 from2])]
1922 ; return (tidy_env, msg) }
1925 The main context-reduction function is @reduce@. Here's its game plan.
1928 reduceList :: RedEnv -> [Inst] -> Avails -> TcM Avails
1929 reduceList env@(RedEnv {red_stack = (n,stk)}) wanteds state
1930 = do { traceTc (text "reduceList " <+> (ppr wanteds $$ ppr state))
1934 dumpTcRn (hang (ptext SLIT("Interesting! Context reduction stack depth") <+> int n)
1935 2 (ifPprDebug (nest 2 (pprStack stk))))
1938 ; if n >= ctxtStkDepth dopts then
1939 failWithTc (reduceDepthErr n stk)
1943 go [] state = return state
1944 go (w:ws) state = do { state' <- reduce (env {red_stack = (n+1, w:stk)}) w state
1947 -- Base case: we're done!
1948 reduce env wanted avails
1949 -- It's the same as an existing inst, or a superclass thereof
1950 | Just avail <- findAvail avails wanted
1951 = do { traceTc (text "reduce: found " <+> ppr wanted)
1956 = do { traceTc (text "reduce" <+> ppr wanted $$ ppr avails)
1957 ; case red_try_me env wanted of {
1958 Stop -> try_simple (addIrred NoSCs);
1959 -- See Note [No superclasses for Stop]
1961 ReduceMe want_scs -> do -- It should be reduced
1962 { (avails, lookup_result) <- reduceInst env avails wanted
1963 ; case lookup_result of
1964 NoInstance -> addIrred want_scs avails wanted
1965 -- Add it and its superclasses
1967 GenInst [] rhs -> addWanted want_scs avails wanted rhs []
1969 GenInst wanteds' rhs
1970 -> do { avails1 <- addIrred NoSCs avails wanted
1971 ; avails2 <- reduceList env wanteds' avails1
1972 ; addWanted want_scs avails2 wanted rhs wanteds' } }
1973 -- Temporarily do addIrred *before* the reduceList,
1974 -- which has the effect of adding the thing we are trying
1975 -- to prove to the database before trying to prove the things it
1976 -- needs. See note [RECURSIVE DICTIONARIES]
1977 -- NB: we must not do an addWanted before, because that adds the
1978 -- superclasses too, and that can lead to a spurious loop; see
1979 -- the examples in [SUPERCLASS-LOOP]
1980 -- So we do an addIrred before, and then overwrite it afterwards with addWanted
1983 -- First, see if the inst can be reduced to a constant in one step
1984 -- Works well for literals (1::Int) and constant dictionaries (d::Num Int)
1985 -- Don't bother for implication constraints, which take real work
1986 try_simple do_this_otherwise
1987 = do { res <- lookupSimpleInst wanted
1989 GenInst [] rhs -> addWanted AddSCs avails wanted rhs []
1990 other -> do_this_otherwise avails wanted }
1994 Note [SUPERCLASS-LOOP 2]
1995 ~~~~~~~~~~~~~~~~~~~~~~~~
1996 But the above isn't enough. Suppose we are *given* d1:Ord a,
1997 and want to deduce (d2:C [a]) where
1999 class Ord a => C a where
2000 instance Ord [a] => C [a] where ...
2002 Then we'll use the instance decl to deduce C [a] from Ord [a], and then add the
2003 superclasses of C [a] to avails. But we must not overwrite the binding
2004 for Ord [a] (which is obtained from Ord a) with a superclass selection or we'll just
2007 Here's another variant, immortalised in tcrun020
2008 class Monad m => C1 m
2009 class C1 m => C2 m x
2010 instance C2 Maybe Bool
2011 For the instance decl we need to build (C1 Maybe), and it's no good if
2012 we run around and add (C2 Maybe Bool) and its superclasses to the avails
2013 before we search for C1 Maybe.
2015 Here's another example
2016 class Eq b => Foo a b
2017 instance Eq a => Foo [a] a
2021 we'll first deduce that it holds (via the instance decl). We must not
2022 then overwrite the Eq t constraint with a superclass selection!
2024 At first I had a gross hack, whereby I simply did not add superclass constraints
2025 in addWanted, though I did for addGiven and addIrred. This was sub-optimal,
2026 becuase it lost legitimate superclass sharing, and it still didn't do the job:
2027 I found a very obscure program (now tcrun021) in which improvement meant the
2028 simplifier got two bites a the cherry... so something seemed to be an Stop
2029 first time, but reducible next time.
2031 Now we implement the Right Solution, which is to check for loops directly
2032 when adding superclasses. It's a bit like the occurs check in unification.
2035 Note [RECURSIVE DICTIONARIES]
2036 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2038 data D r = ZeroD | SuccD (r (D r));
2040 instance (Eq (r (D r))) => Eq (D r) where
2041 ZeroD == ZeroD = True
2042 (SuccD a) == (SuccD b) = a == b
2045 equalDC :: D [] -> D [] -> Bool;
2048 We need to prove (Eq (D [])). Here's how we go:
2052 by instance decl, holds if
2056 by instance decl of Eq, holds if
2058 where d2 = dfEqList d3
2061 But now we can "tie the knot" to give
2067 and it'll even run! The trick is to put the thing we are trying to prove
2068 (in this case Eq (D []) into the database before trying to prove its
2069 contributing clauses.
2072 %************************************************************************
2074 Reducing a single constraint
2076 %************************************************************************
2079 ---------------------------------------------
2080 reduceInst :: RedEnv -> Avails -> Inst -> TcM (Avails, LookupInstResult)
2081 reduceInst env avails other_inst
2082 = do { result <- lookupSimpleInst other_inst
2083 ; return (avails, result) }
2086 Note [Equational Constraints in Implication Constraints]
2087 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2089 An implication constraint is of the form
2091 where Given and Wanted may contain both equational and dictionary
2092 constraints. The delay and reduction of these two kinds of constraints
2095 -) In the generated code, wanted Dictionary constraints are wrapped up in an
2096 implication constraint that is created at the code site where the wanted
2097 dictionaries can be reduced via a let-binding. This let-bound implication
2098 constraint is deconstructed at the use-site of the wanted dictionaries.
2100 -) While the reduction of equational constraints is also delayed, the delay
2101 is not manifest in the generated code. The required evidence is generated
2102 in the code directly at the use-site. There is no let-binding and deconstruction
2103 necessary. The main disadvantage is that we cannot exploit sharing as the
2104 same evidence may be generated at multiple use-sites. However, this disadvantage
2105 is limited because it only concerns coercions which are erased.
2107 The different treatment is motivated by the different in representation. Dictionary
2108 constraints require manifest runtime dictionaries, while equations require coercions
2112 ---------------------------------------------
2113 reduceImplication :: RedEnv
2115 -> TcM (TcDictBinds, [Inst])
2118 Suppose we are simplifying the constraint
2119 forall bs. extras => wanted
2120 in the context of an overall simplification problem with givens 'givens',
2121 and refinment 'reft'.
2124 * The refinement is often empty
2126 * The 'extra givens' need not mention any of the quantified type variables
2127 e.g. forall {}. Eq a => Eq [a]
2128 forall {}. C Int => D (Tree Int)
2130 This happens when you have something like
2132 T1 :: Eq a => a -> T a
2135 f x = ...(case x of { T1 v -> v==v })...
2138 -- ToDo: should we instantiate tvs? I think it's not necessary
2140 -- Note on coercion variables:
2142 -- The extra given coercion variables are bound at two different sites:
2143 -- -) in the creation context of the implication constraint
2144 -- the solved equational constraints use these binders
2146 -- -) at the solving site of the implication constraint
2147 -- the solved dictionaries use these binders
2148 -- these binders are generated by reduceImplication
2150 reduceImplication env
2151 orig_implic@(ImplicInst { tci_name = name, tci_loc = inst_loc,
2152 tci_tyvars = tvs, tci_reft = reft,
2153 tci_given = extra_givens, tci_wanted = wanteds })
2154 = do { -- Add refined givens, and the extra givens
2156 -- (refined_red_givens,refined_avails)
2157 -- <- if isEmptyRefinement reft then return (red_givens env,orig_avails)
2158 -- else foldlM (addRefinedGiven reft) ([],orig_avails) (red_givens env)
2159 -- Commented out SLPJ Sept 07; see comment with extractLocalResults below
2160 let refined_red_givens = []
2162 -- Solve the sub-problem
2163 ; let try_me inst = ReduceMe AddSCs -- Note [Freeness and implications]
2164 env' = env { red_givens = extra_givens ++ red_givens env
2166 , red_doc = sep [ptext SLIT("reduceImplication for") <+> ppr name,
2167 nest 2 (parens $ ptext SLIT("within") <+> red_doc env)]
2168 , red_try_me = try_me }
2170 ; traceTc (text "reduceImplication" <+> vcat
2171 [ ppr (red_givens env), ppr extra_givens,
2172 ppr reft, ppr wanteds])
2173 ; (irreds, binds) <- checkLoop env' wanteds
2174 ; let (extra_eq_givens, extra_dict_givens) = partition isEqInst extra_givens
2175 -- SLPJ Sept 07: I think this is bogus; currently
2176 -- there are no Eqinsts in extra_givens
2177 dict_ids = map instToId extra_dict_givens
2179 -- Note [Reducing implication constraints]
2180 -- Tom -- update note, put somewhere!
2182 ; traceTc (text "reduceImplication result" <+> vcat
2183 [ppr irreds, ppr binds])
2185 ; -- extract superclass binds
2186 -- (sc_binds,_) <- extractResults avails []
2187 -- ; traceTc (text "reduceImplication sc_binds" <+> vcat
2188 -- [ppr sc_binds, ppr avails])
2191 -- SLPJ Sept 07: what if improvement happened inside the checkLoop?
2192 -- Then we must iterate the outer loop too!
2194 ; traceTc (text "reduceImplication condition" <+> ppr ((isEmptyLHsBinds binds) || (null irreds)))
2196 -- Progress is no longer measered by the number of bindings
2197 ; if (isEmptyLHsBinds binds) && (not $ null irreds) then -- No progress
2198 -- If there are any irreds, we back off and do nothing
2199 return (emptyBag, [orig_implic])
2201 { (simpler_implic_insts, bind) <- makeImplicationBind inst_loc tvs reft extra_givens irreds
2202 -- This binding is useless if the recursive simplification
2203 -- made no progress; but currently we don't try to optimise that
2204 -- case. After all, we only try hard to reduce at top level, or
2205 -- when inferring types.
2207 ; let dict_wanteds = filter (not . isEqInst) wanteds
2208 -- TOMDO: given equational constraints bug!
2209 -- we need a different evidence for given
2210 -- equations depending on whether we solve
2211 -- dictionary constraints or equational constraints
2213 eq_tyvars = varSetElems $ tyVarsOfTypes $ map eqInstType extra_eq_givens
2214 -- SLPJ Sept07: this looks Utterly Wrong to me, but I think
2215 -- that current extra_givens has no EqInsts, so
2216 -- it makes no difference
2217 co = wrap_inline -- Note [Always inline implication constraints]
2219 <.> mkWpLams eq_tyvars
2220 <.> mkWpLams dict_ids
2221 <.> WpLet (binds `unionBags` bind)
2222 wrap_inline | null dict_ids = idHsWrapper
2223 | otherwise = WpInline
2224 rhs = mkHsWrap co payload
2225 loc = instLocSpan inst_loc
2226 payload | [dict_wanted] <- dict_wanteds = HsVar (instToId dict_wanted)
2227 | otherwise = ExplicitTuple (map (L loc . HsVar . instToId) dict_wanteds) Boxed
2230 ; traceTc (vcat [text "reduceImplication" <+> ppr name,
2231 ppr simpler_implic_insts,
2232 text "->" <+> ppr rhs])
2233 ; return (unitBag (L loc (VarBind (instToId orig_implic) (L loc rhs))),
2234 simpler_implic_insts)
2239 Note [Always inline implication constraints]
2240 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2241 Suppose an implication constraint floats out of an INLINE function.
2242 Then although the implication has a single call site, it won't be
2243 inlined. And that is bad because it means that even if there is really
2244 *no* overloading (type signatures specify the exact types) there will
2245 still be dictionary passing in the resulting code. To avert this,
2246 we mark the implication constraints themselves as INLINE, at least when
2247 there is no loss of sharing as a result.
2249 Note [Freeness and implications]
2250 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2251 It's hard to say when an implication constraint can be floated out. Consider
2252 forall {} Eq a => Foo [a]
2253 The (Foo [a]) doesn't mention any of the quantified variables, but it
2254 still might be partially satisfied by the (Eq a).
2256 There is a useful special case when it *is* easy to partition the
2257 constraints, namely when there are no 'givens'. Consider
2258 forall {a}. () => Bar b
2259 There are no 'givens', and so there is no reason to capture (Bar b).
2260 We can let it float out. But if there is even one constraint we
2261 must be much more careful:
2262 forall {a}. C a b => Bar (m b)
2263 because (C a b) might have a superclass (D b), from which we might
2264 deduce (Bar [b]) when m later gets instantiated to []. Ha!
2266 Here is an even more exotic example
2268 Now consider the constraint
2269 forall b. D Int b => C Int
2270 We can satisfy the (C Int) from the superclass of D, so we don't want
2271 to float the (C Int) out, even though it mentions no type variable in
2274 Note [Pruning the givens in an implication constraint]
2275 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2276 Suppose we are about to form the implication constraint
2277 forall tvs. Eq a => Ord b
2278 The (Eq a) cannot contribute to the (Ord b), because it has no access to
2279 the type variable 'b'. So we could filter out the (Eq a) from the givens.
2281 Doing so would be a bit tidier, but all the implication constraints get
2282 simplified away by the optimiser, so it's no great win. So I don't take
2283 advantage of that at the moment.
2285 If you do, BE CAREFUL of wobbly type variables.
2288 %************************************************************************
2290 Avails and AvailHow: the pool of evidence
2292 %************************************************************************
2296 data Avails = Avails !ImprovementDone !AvailEnv
2298 type ImprovementDone = Bool -- True <=> some unification has happened
2299 -- so some Irreds might now be reducible
2300 -- keys that are now
2302 type AvailEnv = FiniteMap Inst AvailHow
2304 = IsIrred -- Used for irreducible dictionaries,
2305 -- which are going to be lambda bound
2307 | Given Inst -- Used for dictionaries for which we have a binding
2308 -- e.g. those "given" in a signature
2310 | Rhs -- Used when there is a RHS
2311 (LHsExpr TcId) -- The RHS
2312 [Inst] -- Insts free in the RHS; we need these too
2314 instance Outputable Avails where
2317 pprAvails (Avails imp avails)
2318 = vcat [ ptext SLIT("Avails") <> (if imp then ptext SLIT("[improved]") else empty)
2320 vcat [ sep [ppr inst, nest 2 (equals <+> ppr avail)]
2321 | (inst,avail) <- fmToList avails ]]
2323 instance Outputable AvailHow where
2326 -------------------------
2327 pprAvail :: AvailHow -> SDoc
2328 pprAvail IsIrred = text "Irred"
2329 pprAvail (Given x) = text "Given" <+> ppr x
2330 pprAvail (Rhs rhs bs) = sep [text "Rhs" <+> ppr bs,
2333 -------------------------
2334 extendAvailEnv :: AvailEnv -> Inst -> AvailHow -> AvailEnv
2335 extendAvailEnv env inst avail = addToFM env inst avail
2337 findAvailEnv :: AvailEnv -> Inst -> Maybe AvailHow
2338 findAvailEnv env wanted = lookupFM env wanted
2339 -- NB 1: the Ord instance of Inst compares by the class/type info
2340 -- *not* by unique. So
2341 -- d1::C Int == d2::C Int
2343 emptyAvails :: Avails
2344 emptyAvails = Avails False emptyFM
2346 findAvail :: Avails -> Inst -> Maybe AvailHow
2347 findAvail (Avails _ avails) wanted = findAvailEnv avails wanted
2349 elemAvails :: Inst -> Avails -> Bool
2350 elemAvails wanted (Avails _ avails) = wanted `elemFM` avails
2352 extendAvails :: Avails -> Inst -> AvailHow -> TcM Avails
2354 extendAvails avails@(Avails imp env) inst avail
2355 = do { imp1 <- tcImproveOne avails inst -- Do any improvement
2356 ; return (Avails (imp || imp1) (extendAvailEnv env inst avail)) }
2358 availsInsts :: Avails -> [Inst]
2359 availsInsts (Avails _ avails) = keysFM avails
2361 availsImproved (Avails imp _) = imp
2363 updateImprovement :: Avails -> Avails -> Avails
2364 -- (updateImprovement a1 a2) sets a1's improvement flag from a2
2365 updateImprovement (Avails _ avails1) (Avails imp2 _) = Avails imp2 avails1
2368 Extracting the bindings from a bunch of Avails.
2369 The bindings do *not* come back sorted in dependency order.
2370 We assume that they'll be wrapped in a big Rec, so that the
2371 dependency analyser can sort them out later
2374 type DoneEnv = FiniteMap Inst [Id]
2375 -- Tracks which things we have evidence for
2377 extractResults :: Avails
2379 -> TcM (TcDictBinds, -- Bindings
2380 [Inst], -- The insts bound by the bindings
2381 [Inst]) -- Irreducible ones
2382 -- Note [Reducing implication constraints]
2384 extractResults (Avails _ avails) wanteds
2385 = go emptyBag [] [] emptyFM wanteds
2387 go :: TcDictBinds -- Bindings for dicts
2388 -> [Inst] -- Bound by the bindings
2390 -> DoneEnv -- Has an entry for each inst in the above three sets
2392 -> TcM (TcDictBinds, [Inst], [Inst])
2393 go binds bound_dicts irreds done []
2394 = return (binds, bound_dicts, irreds)
2396 go binds bound_dicts irreds done (w:ws)
2397 | Just done_ids@(done_id : rest_done_ids) <- lookupFM done w
2398 = if w_id `elem` done_ids then
2399 go binds bound_dicts irreds done ws
2401 go (add_bind (nlHsVar done_id)) bound_dicts irreds
2402 (addToFM done w (done_id : w_id : rest_done_ids)) ws
2404 | otherwise -- Not yet done
2405 = case findAvailEnv avails w of
2406 Nothing -> pprTrace "Urk: extractResults" (ppr w) $
2407 go binds bound_dicts irreds done ws
2409 Just IsIrred -> go binds bound_dicts (w:irreds) done' ws
2411 Just (Rhs rhs ws') -> go (add_bind rhs) (w:bound_dicts) irreds done' (ws' ++ ws)
2413 Just (Given g) -> go binds' bound_dicts irreds (addToFM done w [g_id]) ws
2416 binds' | w_id == g_id = binds
2417 | otherwise = add_bind (nlHsVar g_id)
2420 done' = addToFM done w [w_id]
2421 add_bind rhs = addInstToDictBind binds w rhs
2425 Note [No superclasses for Stop]
2426 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2427 When we decide not to reduce an Inst -- the 'WhatToDo' --- we still
2428 add it to avails, so that any other equal Insts will be commoned up
2429 right here. However, we do *not* add superclasses. If we have
2432 but a is not bound here, then we *don't* want to derive dn from df
2433 here lest we lose sharing.
2436 addWanted :: WantSCs -> Avails -> Inst -> LHsExpr TcId -> [Inst] -> TcM Avails
2437 addWanted want_scs avails wanted rhs_expr wanteds
2438 = addAvailAndSCs want_scs avails wanted avail
2440 avail = Rhs rhs_expr wanteds
2442 addGiven :: Avails -> Inst -> TcM Avails
2443 addGiven avails given = addAvailAndSCs AddSCs avails given (Given given)
2444 -- Always add superclasses for 'givens'
2446 -- No ASSERT( not (given `elemAvails` avails) ) because in an instance
2447 -- decl for Ord t we can add both Ord t and Eq t as 'givens',
2448 -- so the assert isn't true
2450 addRefinedGiven :: Refinement -> Avails -> Inst -> TcM Avails
2451 addRefinedGiven reft avails given
2452 | isDict given -- We sometimes have 'given' methods, but they
2453 -- are always optional, so we can drop them
2454 , let pred = dictPred given
2455 , isRefineablePred pred -- See Note [ImplicInst rigidity]
2456 , Just (co, pred) <- refinePred reft pred
2457 = do { new_given <- newDictBndr (instLoc given) pred
2458 ; let rhs = L (instSpan given) $
2459 HsWrap (WpCo co) (HsVar (instToId given))
2460 ; addAvailAndSCs AddSCs avails new_given (Rhs rhs [given]) }
2461 -- ToDo: the superclasses of the original given all exist in Avails
2462 -- so we could really just cast them, but it's more awkward to do,
2463 -- and hopefully the optimiser will spot the duplicated work
2468 Note [ImplicInst rigidity]
2469 ~~~~~~~~~~~~~~~~~~~~~~~~~~
2471 C :: forall ab. (Eq a, Ord b) => b -> T a
2473 ...(case x of C v -> <body>)...
2475 From the case (where x::T ty) we'll get an implication constraint
2476 forall b. (Eq ty, Ord b) => <body-constraints>
2477 Now suppose <body-constraints> itself has an implication constraint
2479 forall c. <reft> => <payload>
2480 Then, we can certainly apply the refinement <reft> to the Ord b, becuase it is
2481 existential, but we probably should not apply it to the (Eq ty) because it may
2482 be wobbly. Hence the isRigidInst
2484 @Insts@ are ordered by their class/type info, rather than by their
2485 unique. This allows the context-reduction mechanism to use standard finite
2486 maps to do their stuff. It's horrible that this code is here, rather
2487 than with the Avails handling stuff in TcSimplify
2490 addIrred :: WantSCs -> Avails -> Inst -> TcM Avails
2491 addIrred want_scs avails irred = ASSERT2( not (irred `elemAvails` avails), ppr irred $$ ppr avails )
2492 addAvailAndSCs want_scs avails irred IsIrred
2494 addAvailAndSCs :: WantSCs -> Avails -> Inst -> AvailHow -> TcM Avails
2495 addAvailAndSCs want_scs avails inst avail
2496 | not (isClassDict inst) = extendAvails avails inst avail
2497 | NoSCs <- want_scs = extendAvails avails inst avail
2498 | otherwise = do { traceTc (text "addAvailAndSCs" <+> vcat [ppr inst, ppr deps])
2499 ; avails' <- extendAvails avails inst avail
2500 ; addSCs is_loop avails' inst }
2502 is_loop pred = any (`tcEqType` mkPredTy pred) dep_tys
2503 -- Note: this compares by *type*, not by Unique
2504 deps = findAllDeps (unitVarSet (instToVar inst)) avail
2505 dep_tys = map idType (varSetElems deps)
2507 findAllDeps :: IdSet -> AvailHow -> IdSet
2508 -- Find all the Insts that this one depends on
2509 -- See Note [SUPERCLASS-LOOP 2]
2510 -- Watch out, though. Since the avails may contain loops
2511 -- (see Note [RECURSIVE DICTIONARIES]), so we need to track the ones we've seen so far
2512 findAllDeps so_far (Rhs _ kids) = foldl find_all so_far kids
2513 findAllDeps so_far other = so_far
2515 find_all :: IdSet -> Inst -> IdSet
2517 | isEqInst kid = so_far
2518 | kid_id `elemVarSet` so_far = so_far
2519 | Just avail <- findAvail avails kid = findAllDeps so_far' avail
2520 | otherwise = so_far'
2522 so_far' = extendVarSet so_far kid_id -- Add the new kid to so_far
2523 kid_id = instToId kid
2525 addSCs :: (TcPredType -> Bool) -> Avails -> Inst -> TcM Avails
2526 -- Add all the superclasses of the Inst to Avails
2527 -- The first param says "don't do this because the original thing
2528 -- depends on this one, so you'd build a loop"
2529 -- Invariant: the Inst is already in Avails.
2531 addSCs is_loop avails dict
2532 = ASSERT( isDict dict )
2533 do { sc_dicts <- newDictBndrs (instLoc dict) sc_theta'
2534 ; foldlM add_sc avails (zipEqual "add_scs" sc_dicts sc_sels) }
2536 (clas, tys) = getDictClassTys dict
2537 (tyvars, sc_theta, sc_sels, _) = classBigSig clas
2538 sc_theta' = filter (not . isEqPred) $
2539 substTheta (zipTopTvSubst tyvars tys) sc_theta
2541 add_sc avails (sc_dict, sc_sel)
2542 | is_loop (dictPred sc_dict) = return avails -- See Note [SUPERCLASS-LOOP 2]
2543 | is_given sc_dict = return avails
2544 | otherwise = do { avails' <- extendAvails avails sc_dict (Rhs sc_sel_rhs [dict])
2545 ; addSCs is_loop avails' sc_dict }
2547 sc_sel_rhs = L (instSpan dict) (HsWrap co_fn (HsVar sc_sel))
2548 co_fn = WpApp (instToVar dict) <.> mkWpTyApps tys
2550 is_given :: Inst -> Bool
2551 is_given sc_dict = case findAvail avails sc_dict of
2552 Just (Given _) -> True -- Given is cheaper than superclass selection
2555 -- From the a set of insts obtain all equalities that (transitively) occur in
2556 -- superclass contexts of class constraints (aka the ancestor equalities).
2558 ancestorEqualities :: [Inst] -> TcM [Inst]
2560 = mapM mkWantedEqInst -- turn only equality predicates..
2561 . filter isEqPred -- ..into wanted equality insts
2563 . addAEsToBag emptyBag -- collect the superclass constraints..
2564 . map dictPred -- ..of all predicates in a bag
2565 . filter isClassDict
2567 addAEsToBag :: Bag PredType -> [PredType] -> Bag PredType
2568 addAEsToBag bag [] = bag
2569 addAEsToBag bag (pred:preds)
2570 | pred `elemBag` bag = addAEsToBag bag preds
2571 | isEqPred pred = addAEsToBag bagWithPred preds
2572 | isClassPred pred = addAEsToBag bagWithPred predsWithSCs
2573 | otherwise = addAEsToBag bag preds
2575 bagWithPred = bag `snocBag` pred
2576 predsWithSCs = preds ++ substTheta (zipTopTvSubst tyvars tys) sc_theta
2578 (tyvars, sc_theta, _, _) = classBigSig clas
2579 (clas, tys) = getClassPredTys pred
2583 %************************************************************************
2585 \section{tcSimplifyTop: defaulting}
2587 %************************************************************************
2590 @tcSimplifyTop@ is called once per module to simplify all the constant
2591 and ambiguous Insts.
2593 We need to be careful of one case. Suppose we have
2595 instance Num a => Num (Foo a b) where ...
2597 and @tcSimplifyTop@ is given a constraint (Num (Foo x y)). Then it'll simplify
2598 to (Num x), and default x to Int. But what about y??
2600 It's OK: the final zonking stage should zap y to (), which is fine.
2604 tcSimplifyTop, tcSimplifyInteractive :: [Inst] -> TcM TcDictBinds
2605 tcSimplifyTop wanteds
2606 = tc_simplify_top doc False wanteds
2608 doc = text "tcSimplifyTop"
2610 tcSimplifyInteractive wanteds
2611 = tc_simplify_top doc True wanteds
2613 doc = text "tcSimplifyInteractive"
2615 -- The TcLclEnv should be valid here, solely to improve
2616 -- error message generation for the monomorphism restriction
2617 tc_simplify_top doc interactive wanteds
2618 = do { dflags <- getDOpts
2619 ; wanteds <- zonkInsts wanteds
2620 ; mapM_ zonkTopTyVar (varSetElems (tyVarsOfInsts wanteds))
2622 ; traceTc (text "tc_simplify_top 0: " <+> ppr wanteds)
2623 ; (irreds1, binds1) <- tryHardCheckLoop doc1 wanteds
2624 -- ; (irreds1, binds1) <- gentleInferLoop doc1 wanteds
2625 ; traceTc (text "tc_simplify_top 1: " <+> ppr irreds1)
2626 ; (irreds2, binds2) <- approximateImplications doc2 (\d -> True) irreds1
2627 ; traceTc (text "tc_simplify_top 2: " <+> ppr irreds2)
2629 -- Use the defaulting rules to do extra unification
2630 -- NB: irreds2 are already zonked
2631 ; (irreds3, binds3) <- disambiguate doc3 interactive dflags irreds2
2633 -- Deal with implicit parameters
2634 ; let (bad_ips, non_ips) = partition isIPDict irreds3
2635 (ambigs, others) = partition isTyVarDict non_ips
2637 ; topIPErrs bad_ips -- Can arise from f :: Int -> Int
2639 ; addNoInstanceErrs others
2640 ; addTopAmbigErrs ambigs
2642 ; return (binds1 `unionBags` binds2 `unionBags` binds3) }
2644 doc1 = doc <+> ptext SLIT("(first round)")
2645 doc2 = doc <+> ptext SLIT("(approximate)")
2646 doc3 = doc <+> ptext SLIT("(disambiguate)")
2649 If a dictionary constrains a type variable which is
2650 * not mentioned in the environment
2651 * and not mentioned in the type of the expression
2652 then it is ambiguous. No further information will arise to instantiate
2653 the type variable; nor will it be generalised and turned into an extra
2654 parameter to a function.
2656 It is an error for this to occur, except that Haskell provided for
2657 certain rules to be applied in the special case of numeric types.
2659 * at least one of its classes is a numeric class, and
2660 * all of its classes are numeric or standard
2661 then the type variable can be defaulted to the first type in the
2662 default-type list which is an instance of all the offending classes.
2664 So here is the function which does the work. It takes the ambiguous
2665 dictionaries and either resolves them (producing bindings) or
2666 complains. It works by splitting the dictionary list by type
2667 variable, and using @disambigOne@ to do the real business.
2669 @disambigOne@ assumes that its arguments dictionaries constrain all
2670 the same type variable.
2672 ADR Comment 20/6/94: I've changed the @CReturnable@ case to default to
2673 @()@ instead of @Int@. I reckon this is the Right Thing to do since
2674 the most common use of defaulting is code like:
2676 _ccall_ foo `seqPrimIO` bar
2678 Since we're not using the result of @foo@, the result if (presumably)
2682 disambiguate :: SDoc -> Bool -> DynFlags -> [Inst] -> TcM ([Inst], TcDictBinds)
2683 -- Just does unification to fix the default types
2684 -- The Insts are assumed to be pre-zonked
2685 disambiguate doc interactive dflags insts
2687 = return (insts, emptyBag)
2689 | null defaultable_groups
2690 = do { traceTc (text "disambigutate, no defaultable groups" <+> vcat [ppr unaries, ppr insts, ppr bad_tvs, ppr defaultable_groups])
2691 ; return (insts, emptyBag) }
2694 = do { -- Figure out what default types to use
2695 default_tys <- getDefaultTys extended_defaulting ovl_strings
2697 ; traceTc (text "disambiguate1" <+> vcat [ppr insts, ppr unaries, ppr bad_tvs, ppr defaultable_groups])
2698 ; mapM_ (disambigGroup default_tys) defaultable_groups
2700 -- disambigGroup does unification, hence try again
2701 ; tryHardCheckLoop doc insts }
2704 extended_defaulting = interactive || dopt Opt_ExtendedDefaultRules dflags
2705 ovl_strings = dopt Opt_OverloadedStrings dflags
2707 unaries :: [(Inst, Class, TcTyVar)] -- (C tv) constraints
2708 bad_tvs :: TcTyVarSet -- Tyvars mentioned by *other* constraints
2709 (unaries, bad_tvs_s) = partitionWith find_unary insts
2710 bad_tvs = unionVarSets bad_tvs_s
2712 -- Finds unary type-class constraints
2713 find_unary d@(Dict {tci_pred = ClassP cls [ty]})
2714 | Just tv <- tcGetTyVar_maybe ty = Left (d,cls,tv)
2715 find_unary inst = Right (tyVarsOfInst inst)
2717 -- Group by type variable
2718 defaultable_groups :: [[(Inst,Class,TcTyVar)]]
2719 defaultable_groups = filter defaultable_group (equivClasses cmp_tv unaries)
2720 cmp_tv (_,_,tv1) (_,_,tv2) = tv1 `compare` tv2
2722 defaultable_group :: [(Inst,Class,TcTyVar)] -> Bool
2723 defaultable_group ds@((_,_,tv):_)
2724 = isTyConableTyVar tv -- Note [Avoiding spurious errors]
2725 && not (tv `elemVarSet` bad_tvs)
2726 && defaultable_classes [c | (_,c,_) <- ds]
2727 defaultable_group [] = panic "defaultable_group"
2729 defaultable_classes clss
2730 | extended_defaulting = any isInteractiveClass clss
2731 | otherwise = all is_std_class clss && (any is_num_class clss)
2733 -- In interactive mode, or with -fextended-default-rules,
2734 -- we default Show a to Show () to avoid graututious errors on "show []"
2735 isInteractiveClass cls
2736 = is_num_class cls || (classKey cls `elem` [showClassKey, eqClassKey, ordClassKey])
2738 is_num_class cls = isNumericClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2739 -- is_num_class adds IsString to the standard numeric classes,
2740 -- when -foverloaded-strings is enabled
2742 is_std_class cls = isStandardClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2743 -- Similarly is_std_class
2745 -----------------------
2746 disambigGroup :: [Type] -- The default types
2747 -> [(Inst,Class,TcTyVar)] -- All standard classes of form (C a)
2748 -> TcM () -- Just does unification, to fix the default types
2750 disambigGroup default_tys dicts
2751 = try_default default_tys
2753 (_,_,tyvar) = ASSERT(not (null dicts)) head dicts -- Should be non-empty
2754 classes = [c | (_,c,_) <- dicts]
2756 try_default [] = return ()
2757 try_default (default_ty : default_tys)
2758 = tryTcLIE_ (try_default default_tys) $
2759 do { tcSimplifyDefault [mkClassPred clas [default_ty] | clas <- classes]
2760 -- This may fail; then the tryTcLIE_ kicks in
2761 -- Failure here is caused by there being no type in the
2762 -- default list which can satisfy all the ambiguous classes.
2763 -- For example, if Real a is reqd, but the only type in the
2764 -- default list is Int.
2766 -- After this we can't fail
2767 ; warnDefault dicts default_ty
2768 ; unifyType default_ty (mkTyVarTy tyvar)
2769 ; return () -- TOMDO: do something with the coercion
2773 -----------------------
2774 getDefaultTys :: Bool -> Bool -> TcM [Type]
2775 getDefaultTys extended_deflts ovl_strings
2776 = do { mb_defaults <- getDeclaredDefaultTys
2777 ; case mb_defaults of {
2778 Just tys -> return tys ; -- User-supplied defaults
2781 -- No use-supplied default
2782 -- Use [Integer, Double], plus modifications
2783 { integer_ty <- tcMetaTy integerTyConName
2784 ; checkWiredInTyCon doubleTyCon
2785 ; string_ty <- tcMetaTy stringTyConName
2786 ; return (opt_deflt extended_deflts unitTy
2787 -- Note [Default unitTy]
2789 [integer_ty,doubleTy]
2791 opt_deflt ovl_strings string_ty) } } }
2793 opt_deflt True ty = [ty]
2794 opt_deflt False ty = []
2797 Note [Default unitTy]
2798 ~~~~~~~~~~~~~~~~~~~~~
2799 In interative mode (or with -fextended-default-rules) we add () as the first type we
2800 try when defaulting. This has very little real impact, except in the following case.
2802 Text.Printf.printf "hello"
2803 This has type (forall a. IO a); it prints "hello", and returns 'undefined'. We don't
2804 want the GHCi repl loop to try to print that 'undefined'. The neatest thing is to
2805 default the 'a' to (), rather than to Integer (which is what would otherwise happen;
2806 and then GHCi doesn't attempt to print the (). So in interactive mode, we add
2807 () to the list of defaulting types. See Trac #1200.
2809 Note [Avoiding spurious errors]
2810 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2811 When doing the unification for defaulting, we check for skolem
2812 type variables, and simply don't default them. For example:
2813 f = (*) -- Monomorphic
2814 g :: Num a => a -> a
2816 Here, we get a complaint when checking the type signature for g,
2817 that g isn't polymorphic enough; but then we get another one when
2818 dealing with the (Num a) context arising from f's definition;
2819 we try to unify a with Int (to default it), but find that it's
2820 already been unified with the rigid variable from g's type sig
2823 %************************************************************************
2825 \subsection[simple]{@Simple@ versions}
2827 %************************************************************************
2829 Much simpler versions when there are no bindings to make!
2831 @tcSimplifyThetas@ simplifies class-type constraints formed by
2832 @deriving@ declarations and when specialising instances. We are
2833 only interested in the simplified bunch of class/type constraints.
2835 It simplifies to constraints of the form (C a b c) where
2836 a,b,c are type variables. This is required for the context of
2837 instance declarations.
2840 tcSimplifyDeriv :: InstOrigin
2842 -> ThetaType -- Wanted
2843 -> TcM ThetaType -- Needed
2844 -- Given instance (wanted) => C inst_ty
2845 -- Simplify 'wanted' as much as possible
2847 tcSimplifyDeriv orig tyvars theta
2848 = do { (tvs, _, tenv) <- tcInstTyVars tyvars
2849 -- The main loop may do unification, and that may crash if
2850 -- it doesn't see a TcTyVar, so we have to instantiate. Sigh
2851 -- ToDo: what if two of them do get unified?
2852 ; wanteds <- newDictBndrsO orig (substTheta tenv theta)
2853 ; (irreds, _) <- tryHardCheckLoop doc wanteds
2855 ; let (tv_dicts, others) = partition ok irreds
2856 ; addNoInstanceErrs others
2857 -- See Note [Exotic derived instance contexts] in TcMType
2859 ; let rev_env = zipTopTvSubst tvs (mkTyVarTys tyvars)
2860 simpl_theta = substTheta rev_env (map dictPred tv_dicts)
2861 -- This reverse-mapping is a pain, but the result
2862 -- should mention the original TyVars not TcTyVars
2864 ; return simpl_theta }
2866 doc = ptext SLIT("deriving classes for a data type")
2868 ok dict | isDict dict = validDerivPred (dictPred dict)
2873 @tcSimplifyDefault@ just checks class-type constraints, essentially;
2874 used with \tr{default} declarations. We are only interested in
2875 whether it worked or not.
2878 tcSimplifyDefault :: ThetaType -- Wanted; has no type variables in it
2881 tcSimplifyDefault theta = do
2882 wanteds <- newDictBndrsO DefaultOrigin theta
2883 (irreds, _) <- tryHardCheckLoop doc wanteds
2884 addNoInstanceErrs irreds
2888 traceTc (ptext SLIT("tcSimplifyDefault failing")) >> failM
2890 doc = ptext SLIT("default declaration")
2894 %************************************************************************
2896 \section{Errors and contexts}
2898 %************************************************************************
2900 ToDo: for these error messages, should we note the location as coming
2901 from the insts, or just whatever seems to be around in the monad just
2905 groupErrs :: ([Inst] -> TcM ()) -- Deal with one group
2906 -> [Inst] -- The offending Insts
2908 -- Group together insts with the same origin
2909 -- We want to report them together in error messages
2911 groupErrs report_err []
2913 groupErrs report_err (inst:insts)
2914 = do { do_one (inst:friends)
2915 ; groupErrs report_err others }
2917 -- (It may seem a bit crude to compare the error messages,
2918 -- but it makes sure that we combine just what the user sees,
2919 -- and it avoids need equality on InstLocs.)
2920 (friends, others) = partition is_friend insts
2921 loc_msg = showSDoc (pprInstLoc (instLoc inst))
2922 is_friend friend = showSDoc (pprInstLoc (instLoc friend)) == loc_msg
2923 do_one insts = addInstCtxt (instLoc (head insts)) (report_err insts)
2924 -- Add location and context information derived from the Insts
2926 -- Add the "arising from..." part to a message about bunch of dicts
2927 addInstLoc :: [Inst] -> Message -> Message
2928 addInstLoc insts msg = msg $$ nest 2 (pprInstArising (head insts))
2930 addTopIPErrs :: [Name] -> [Inst] -> TcM ()
2931 addTopIPErrs bndrs []
2933 addTopIPErrs bndrs ips
2934 = do { dflags <- getDOpts
2935 ; addErrTcM (tidy_env, mk_msg dflags tidy_ips) }
2937 (tidy_env, tidy_ips) = tidyInsts ips
2939 = vcat [sep [ptext SLIT("Implicit parameters escape from"),
2940 nest 2 (ptext SLIT("the monomorphic top-level binding")
2941 <> plural bndrs <+> ptext SLIT("of")
2942 <+> pprBinders bndrs <> colon)],
2943 nest 2 (vcat (map ppr_ip ips)),
2944 monomorphism_fix dflags]
2945 ppr_ip ip = pprPred (dictPred ip) <+> pprInstArising ip
2947 topIPErrs :: [Inst] -> TcM ()
2949 = groupErrs report tidy_dicts
2951 (tidy_env, tidy_dicts) = tidyInsts dicts
2952 report dicts = addErrTcM (tidy_env, mk_msg dicts)
2953 mk_msg dicts = addInstLoc dicts (ptext SLIT("Unbound implicit parameter") <>
2954 plural tidy_dicts <+> pprDictsTheta tidy_dicts)
2956 addNoInstanceErrs :: [Inst] -- Wanted (can include implications)
2958 addNoInstanceErrs insts
2959 = do { let (tidy_env, tidy_insts) = tidyInsts insts
2960 ; reportNoInstances tidy_env Nothing tidy_insts }
2964 -> Maybe (InstLoc, [Inst]) -- Context
2965 -- Nothing => top level
2966 -- Just (d,g) => d describes the construct
2968 -> [Inst] -- What is wanted (can include implications)
2971 reportNoInstances tidy_env mb_what insts
2972 = groupErrs (report_no_instances tidy_env mb_what) insts
2974 report_no_instances tidy_env mb_what insts
2975 = do { inst_envs <- tcGetInstEnvs
2976 ; let (implics, insts1) = partition isImplicInst insts
2977 (insts2, overlaps) = partitionWith (check_overlap inst_envs) insts1
2978 (eqInsts, insts3) = partition isEqInst insts2
2979 ; traceTc (text "reportNoInstances" <+> vcat
2980 [ppr insts, ppr implics, ppr insts1, ppr insts2])
2981 ; mapM_ complain_implic implics
2982 ; mapM_ (\doc -> addErrTcM (tidy_env, doc)) overlaps
2983 ; groupErrs complain_no_inst insts3
2984 ; mapM_ (addErrTcM . mk_eq_err) eqInsts
2987 complain_no_inst insts = addErrTcM (tidy_env, mk_no_inst_err insts)
2989 complain_implic inst -- Recurse!
2990 = reportNoInstances tidy_env
2991 (Just (tci_loc inst, tci_given inst))
2994 check_overlap :: (InstEnv,InstEnv) -> Inst -> Either Inst SDoc
2995 -- Right msg => overlap message
2996 -- Left inst => no instance
2997 check_overlap inst_envs wanted
2998 | not (isClassDict wanted) = Left wanted
3000 = case lookupInstEnv inst_envs clas tys of
3001 -- The case of exactly one match and no unifiers means a
3002 -- successful lookup. That can't happen here, because dicts
3003 -- only end up here if they didn't match in Inst.lookupInst
3005 ([m],[]) -> pprPanic "reportNoInstance" (ppr wanted)
3007 ([], _) -> Left wanted -- No match
3008 res -> Right (mk_overlap_msg wanted res)
3010 (clas,tys) = getDictClassTys wanted
3012 mk_overlap_msg dict (matches, unifiers)
3013 = ASSERT( not (null matches) )
3014 vcat [ addInstLoc [dict] ((ptext SLIT("Overlapping instances for")
3015 <+> pprPred (dictPred dict))),
3016 sep [ptext SLIT("Matching instances") <> colon,
3017 nest 2 (vcat [pprInstances ispecs, pprInstances unifiers])],
3018 if not (isSingleton matches)
3019 then -- Two or more matches
3021 else -- One match, plus some unifiers
3022 ASSERT( not (null unifiers) )
3023 parens (vcat [ptext SLIT("The choice depends on the instantiation of") <+>
3024 quotes (pprWithCommas ppr (varSetElems (tyVarsOfInst dict))),
3025 ptext SLIT("To pick the first instance above, use -fallow-incoherent-instances"),
3026 ptext SLIT("when compiling the other instance declarations")])]
3028 ispecs = [ispec | (ispec, _) <- matches]
3030 mk_eq_err :: Inst -> (TidyEnv, SDoc)
3031 mk_eq_err inst = misMatchMsg tidy_env (eqInstTys inst)
3033 mk_no_inst_err insts
3034 | null insts = empty
3036 | Just (loc, givens) <- mb_what, -- Nested (type signatures, instance decls)
3037 not (isEmptyVarSet (tyVarsOfInsts insts))
3038 = vcat [ addInstLoc insts $
3039 sep [ ptext SLIT("Could not deduce") <+> pprDictsTheta insts
3040 , nest 2 $ ptext SLIT("from the context") <+> pprDictsTheta givens]
3041 , show_fixes (fix1 loc : fixes2) ]
3043 | otherwise -- Top level
3044 = vcat [ addInstLoc insts $
3045 ptext SLIT("No instance") <> plural insts
3046 <+> ptext SLIT("for") <+> pprDictsTheta insts
3047 , show_fixes fixes2 ]
3050 fix1 loc = sep [ ptext SLIT("add") <+> pprDictsTheta insts
3051 <+> ptext SLIT("to the context of"),
3052 nest 2 (ppr (instLocOrigin loc)) ]
3053 -- I'm not sure it helps to add the location
3054 -- nest 2 (ptext SLIT("at") <+> ppr (instLocSpan loc)) ]
3056 fixes2 | null instance_dicts = []
3057 | otherwise = [sep [ptext SLIT("add an instance declaration for"),
3058 pprDictsTheta instance_dicts]]
3059 instance_dicts = [d | d <- insts, isClassDict d, not (isTyVarDict d)]
3060 -- Insts for which it is worth suggesting an adding an instance declaration
3061 -- Exclude implicit parameters, and tyvar dicts
3063 show_fixes :: [SDoc] -> SDoc
3064 show_fixes [] = empty
3065 show_fixes (f:fs) = sep [ptext SLIT("Possible fix:"),
3066 nest 2 (vcat (f : map (ptext SLIT("or") <+>) fs))]
3068 addTopAmbigErrs dicts
3069 -- Divide into groups that share a common set of ambiguous tyvars
3070 = ifErrsM (return ()) $ -- Only report ambiguity if no other errors happened
3071 -- See Note [Avoiding spurious errors]
3072 mapM_ report (equivClasses cmp [(d, tvs_of d) | d <- tidy_dicts])
3074 (tidy_env, tidy_dicts) = tidyInsts dicts
3076 tvs_of :: Inst -> [TcTyVar]
3077 tvs_of d = varSetElems (tyVarsOfInst d)
3078 cmp (_,tvs1) (_,tvs2) = tvs1 `compare` tvs2
3080 report :: [(Inst,[TcTyVar])] -> TcM ()
3081 report pairs@((inst,tvs) : _) = do -- The pairs share a common set of ambiguous tyvars
3082 (tidy_env, mono_msg) <- mkMonomorphismMsg tidy_env tvs
3083 setSrcSpan (instSpan inst) $
3084 -- the location of the first one will do for the err message
3085 addErrTcM (tidy_env, msg $$ mono_msg)
3087 dicts = map fst pairs
3088 msg = sep [text "Ambiguous type variable" <> plural tvs <+>
3089 pprQuotedList tvs <+> in_msg,
3090 nest 2 (pprDictsInFull dicts)]
3091 in_msg = text "in the constraint" <> plural dicts <> colon
3092 report [] = panic "addTopAmbigErrs"
3095 mkMonomorphismMsg :: TidyEnv -> [TcTyVar] -> TcM (TidyEnv, Message)
3096 -- There's an error with these Insts; if they have free type variables
3097 -- it's probably caused by the monomorphism restriction.
3098 -- Try to identify the offending variable
3099 -- ASSUMPTION: the Insts are fully zonked
3100 mkMonomorphismMsg tidy_env inst_tvs
3101 = do { dflags <- getDOpts
3102 ; (tidy_env, docs) <- findGlobals (mkVarSet inst_tvs) tidy_env
3103 ; return (tidy_env, mk_msg dflags docs) }
3105 mk_msg _ _ | any isRuntimeUnk inst_tvs
3106 = vcat [ptext SLIT("Cannot resolve unknown runtime types:") <+>
3107 (pprWithCommas ppr inst_tvs),
3108 ptext SLIT("Use :print or :force to determine these types")]
3109 mk_msg _ [] = ptext SLIT("Probable fix: add a type signature that fixes these type variable(s)")
3110 -- This happens in things like
3111 -- f x = show (read "foo")
3112 -- where monomorphism doesn't play any role
3114 = vcat [ptext SLIT("Possible cause: the monomorphism restriction applied to the following:"),
3116 monomorphism_fix dflags]
3118 monomorphism_fix :: DynFlags -> SDoc
3119 monomorphism_fix dflags
3120 = ptext SLIT("Probable fix:") <+> vcat
3121 [ptext SLIT("give these definition(s) an explicit type signature"),
3122 if dopt Opt_MonomorphismRestriction dflags
3123 then ptext SLIT("or use -fno-monomorphism-restriction")
3124 else empty] -- Only suggest adding "-fno-monomorphism-restriction"
3125 -- if it is not already set!
3127 warnDefault ups default_ty = do
3128 warn_flag <- doptM Opt_WarnTypeDefaults
3129 addInstCtxt (instLoc (head (dicts))) (warnTc warn_flag warn_msg)
3131 dicts = [d | (d,_,_) <- ups]
3134 (_, tidy_dicts) = tidyInsts dicts
3135 warn_msg = vcat [ptext SLIT("Defaulting the following constraint(s) to type") <+>
3136 quotes (ppr default_ty),
3137 pprDictsInFull tidy_dicts]
3139 reduceDepthErr n stack
3140 = vcat [ptext SLIT("Context reduction stack overflow; size =") <+> int n,
3141 ptext SLIT("Use -fcontext-stack=N to increase stack size to N"),
3142 nest 4 (pprStack stack)]
3144 pprStack stack = vcat (map pprInstInFull stack)