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 )
74 %************************************************************************
78 %************************************************************************
80 --------------------------------------
81 Notes on functional dependencies (a bug)
82 --------------------------------------
89 instance D a b => C a b -- Undecidable
90 -- (Not sure if it's crucial to this eg)
91 f :: C a b => a -> Bool
94 g :: C a b => a -> Bool
97 Here f typechecks, but g does not!! Reason: before doing improvement,
98 we reduce the (C a b1) constraint from the call of f to (D a b1).
100 Here is a more complicated example:
102 | > class Foo a b | a->b
104 | > class Bar a b | a->b
108 | > instance Bar Obj Obj
110 | > instance (Bar a b) => Foo a b
112 | > foo:: (Foo a b) => a -> String
115 | > runFoo:: (forall a b. (Foo a b) => a -> w) -> w
121 | Could not deduce (Bar a b) from the context (Foo a b)
122 | arising from use of `foo' at <interactive>:1
124 | Add (Bar a b) to the expected type of an expression
125 | In the first argument of `runFoo', namely `foo'
126 | In the definition of `it': it = runFoo foo
128 | Why all of the sudden does GHC need the constraint Bar a b? The
129 | function foo didn't ask for that...
131 The trouble is that to type (runFoo foo), GHC has to solve the problem:
133 Given constraint Foo a b
134 Solve constraint Foo a b'
136 Notice that b and b' aren't the same. To solve this, just do
137 improvement and then they are the same. But GHC currently does
142 That is usually fine, but it isn't here, because it sees that Foo a b is
143 not the same as Foo a b', and so instead applies the instance decl for
144 instance Bar a b => Foo a b. And that's where the Bar constraint comes
147 The Right Thing is to improve whenever the constraint set changes at
148 all. Not hard in principle, but it'll take a bit of fiddling to do.
150 Note [Choosing which variables to quantify]
151 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
152 Suppose we are about to do a generalisation step. We have in our hand
155 T the type of the RHS
156 C the constraints from that RHS
158 The game is to figure out
160 Q the set of type variables over which to quantify
161 Ct the constraints we will *not* quantify over
162 Cq the constraints we will quantify over
164 So we're going to infer the type
168 and float the constraints Ct further outwards.
170 Here are the things that *must* be true:
172 (A) Q intersect fv(G) = EMPTY limits how big Q can be
173 (B) Q superset fv(Cq union T) \ oclose(fv(G),C) limits how small Q can be
175 (A) says we can't quantify over a variable that's free in the environment.
176 (B) says we must quantify over all the truly free variables in T, else
177 we won't get a sufficiently general type.
179 We do not *need* to quantify over any variable that is fixed by the
180 free vars of the environment G.
182 BETWEEN THESE TWO BOUNDS, ANY Q WILL DO!
184 Example: class H x y | x->y where ...
186 fv(G) = {a} C = {H a b, H c d}
189 (A) Q intersect {a} is empty
190 (B) Q superset {a,b,c,d} \ oclose({a}, C) = {a,b,c,d} \ {a,b} = {c,d}
192 So Q can be {c,d}, {b,c,d}
194 In particular, it's perfectly OK to quantify over more type variables
195 than strictly necessary; there is no need to quantify over 'b', since
196 it is determined by 'a' which is free in the envt, but it's perfectly
197 OK to do so. However we must not quantify over 'a' itself.
199 Other things being equal, however, we'd like to quantify over as few
200 variables as possible: smaller types, fewer type applications, more
201 constraints can get into Ct instead of Cq. Here's a good way to
204 Q = grow( fv(T), C ) \ oclose( fv(G), C )
206 That is, quantify over all variable that that MIGHT be fixed by the
207 call site (which influences T), but which aren't DEFINITELY fixed by
208 G. This choice definitely quantifies over enough type variables,
209 albeit perhaps too many.
211 Why grow( fv(T), C ) rather than fv(T)? Consider
213 class H x y | x->y where ...
218 If we used fv(T) = {c} we'd get the type
220 forall c. H c d => c -> b
222 And then if the fn was called at several different c's, each of
223 which fixed d differently, we'd get a unification error, because
224 d isn't quantified. Solution: quantify d. So we must quantify
225 everything that might be influenced by c.
227 Why not oclose( fv(T), C )? Because we might not be able to see
228 all the functional dependencies yet:
230 class H x y | x->y where ...
231 instance H x y => Eq (T x y) where ...
236 Now oclose(fv(T),C) = {c}, because the functional dependency isn't
237 apparent yet, and that's wrong. We must really quantify over d too.
239 There really isn't any point in quantifying over any more than
240 grow( fv(T), C ), because the call sites can't possibly influence
241 any other type variables.
245 -------------------------------------
247 -------------------------------------
249 It's very hard to be certain when a type is ambiguous. Consider
253 instance H x y => K (x,y)
255 Is this type ambiguous?
256 forall a b. (K (a,b), Eq b) => a -> a
258 Looks like it! But if we simplify (K (a,b)) we get (H a b) and
259 now we see that a fixes b. So we can't tell about ambiguity for sure
260 without doing a full simplification. And even that isn't possible if
261 the context has some free vars that may get unified. Urgle!
263 Here's another example: is this ambiguous?
264 forall a b. Eq (T b) => a -> a
265 Not if there's an insance decl (with no context)
266 instance Eq (T b) where ...
268 You may say of this example that we should use the instance decl right
269 away, but you can't always do that:
271 class J a b where ...
272 instance J Int b where ...
274 f :: forall a b. J a b => a -> a
276 (Notice: no functional dependency in J's class decl.)
277 Here f's type is perfectly fine, provided f is only called at Int.
278 It's premature to complain when meeting f's signature, or even
279 when inferring a type for f.
283 However, we don't *need* to report ambiguity right away. It'll always
284 show up at the call site.... and eventually at main, which needs special
285 treatment. Nevertheless, reporting ambiguity promptly is an excellent thing.
287 So here's the plan. We WARN about probable ambiguity if
289 fv(Cq) is not a subset of oclose(fv(T) union fv(G), C)
291 (all tested before quantification).
292 That is, all the type variables in Cq must be fixed by the the variables
293 in the environment, or by the variables in the type.
295 Notice that we union before calling oclose. Here's an example:
297 class J a b c | a b -> c
301 forall b c. (J a b c) => b -> b
303 Only if we union {a} from G with {b} from T before using oclose,
304 do we see that c is fixed.
306 It's a bit vague exactly which C we should use for this oclose call. If we
307 don't fix enough variables we might complain when we shouldn't (see
308 the above nasty example). Nothing will be perfect. That's why we can
309 only issue a warning.
312 Can we ever be *certain* about ambiguity? Yes: if there's a constraint
314 c in C such that fv(c) intersect (fv(G) union fv(T)) = EMPTY
316 then c is a "bubble"; there's no way it can ever improve, and it's
317 certainly ambiguous. UNLESS it is a constant (sigh). And what about
322 instance H x y => K (x,y)
324 Is this type ambiguous?
325 forall a b. (K (a,b), Eq b) => a -> a
327 Urk. The (Eq b) looks "definitely ambiguous" but it isn't. What we are after
328 is a "bubble" that's a set of constraints
330 Cq = Ca union Cq' st fv(Ca) intersect (fv(Cq') union fv(T) union fv(G)) = EMPTY
332 Hence another idea. To decide Q start with fv(T) and grow it
333 by transitive closure in Cq (no functional dependencies involved).
334 Now partition Cq using Q, leaving the definitely-ambiguous and probably-ok.
335 The definitely-ambiguous can then float out, and get smashed at top level
336 (which squashes out the constants, like Eq (T a) above)
339 --------------------------------------
340 Notes on principal types
341 --------------------------------------
346 f x = let g y = op (y::Int) in True
348 Here the principal type of f is (forall a. a->a)
349 but we'll produce the non-principal type
350 f :: forall a. C Int => a -> a
353 --------------------------------------
354 The need for forall's in constraints
355 --------------------------------------
357 [Exchange on Haskell Cafe 5/6 Dec 2000]
359 class C t where op :: t -> Bool
360 instance C [t] where op x = True
362 p y = (let f :: c -> Bool; f x = op (y >> return x) in f, y ++ [])
363 q y = (y ++ [], let f :: c -> Bool; f x = op (y >> return x) in f)
365 The definitions of p and q differ only in the order of the components in
366 the pair on their right-hand sides. And yet:
368 ghc and "Typing Haskell in Haskell" reject p, but accept q;
369 Hugs rejects q, but accepts p;
370 hbc rejects both p and q;
371 nhc98 ... (Malcolm, can you fill in the blank for us!).
373 The type signature for f forces context reduction to take place, and
374 the results of this depend on whether or not the type of y is known,
375 which in turn depends on which component of the pair the type checker
378 Solution: if y::m a, float out the constraints
379 Monad m, forall c. C (m c)
380 When m is later unified with [], we can solve both constraints.
383 --------------------------------------
384 Notes on implicit parameters
385 --------------------------------------
387 Note [Inheriting implicit parameters]
388 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
393 where f is *not* a top-level binding.
394 From the RHS of f we'll get the constraint (?y::Int).
395 There are two types we might infer for f:
399 (so we get ?y from the context of f's definition), or
401 f :: (?y::Int) => Int -> Int
403 At first you might think the first was better, becuase then
404 ?y behaves like a free variable of the definition, rather than
405 having to be passed at each call site. But of course, the WHOLE
406 IDEA is that ?y should be passed at each call site (that's what
407 dynamic binding means) so we'd better infer the second.
409 BOTTOM LINE: when *inferring types* you *must* quantify
410 over implicit parameters. See the predicate isFreeWhenInferring.
413 Note [Implicit parameters and ambiguity]
414 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
415 Only a *class* predicate can give rise to ambiguity
416 An *implicit parameter* cannot. For example:
417 foo :: (?x :: [a]) => Int
419 is fine. The call site will suppply a particular 'x'
421 Furthermore, the type variables fixed by an implicit parameter
422 propagate to the others. E.g.
423 foo :: (Show a, ?x::[a]) => Int
425 The type of foo looks ambiguous. But it isn't, because at a call site
427 let ?x = 5::Int in foo
428 and all is well. In effect, implicit parameters are, well, parameters,
429 so we can take their type variables into account as part of the
430 "tau-tvs" stuff. This is done in the function 'FunDeps.grow'.
433 Question 2: type signatures
434 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
435 BUT WATCH OUT: When you supply a type signature, we can't force you
436 to quantify over implicit parameters. For example:
440 This is perfectly reasonable. We do not want to insist on
442 (?x + 1) :: (?x::Int => Int)
444 That would be silly. Here, the definition site *is* the occurrence site,
445 so the above strictures don't apply. Hence the difference between
446 tcSimplifyCheck (which *does* allow implicit paramters to be inherited)
447 and tcSimplifyCheckBind (which does not).
449 What about when you supply a type signature for a binding?
450 Is it legal to give the following explicit, user type
451 signature to f, thus:
456 At first sight this seems reasonable, but it has the nasty property
457 that adding a type signature changes the dynamic semantics.
460 (let f x = (x::Int) + ?y
461 in (f 3, f 3 with ?y=5)) with ?y = 6
467 in (f 3, f 3 with ?y=5)) with ?y = 6
471 Indeed, simply inlining f (at the Haskell source level) would change the
474 Nevertheless, as Launchbury says (email Oct 01) we can't really give the
475 semantics for a Haskell program without knowing its typing, so if you
476 change the typing you may change the semantics.
478 To make things consistent in all cases where we are *checking* against
479 a supplied signature (as opposed to inferring a type), we adopt the
482 a signature does not need to quantify over implicit params.
484 [This represents a (rather marginal) change of policy since GHC 5.02,
485 which *required* an explicit signature to quantify over all implicit
486 params for the reasons mentioned above.]
488 But that raises a new question. Consider
490 Given (signature) ?x::Int
491 Wanted (inferred) ?x::Int, ?y::Bool
493 Clearly we want to discharge the ?x and float the ?y out. But
494 what is the criterion that distinguishes them? Clearly it isn't
495 what free type variables they have. The Right Thing seems to be
496 to float a constraint that
497 neither mentions any of the quantified type variables
498 nor any of the quantified implicit parameters
500 See the predicate isFreeWhenChecking.
503 Question 3: monomorphism
504 ~~~~~~~~~~~~~~~~~~~~~~~~
505 There's a nasty corner case when the monomorphism restriction bites:
509 The argument above suggests that we *must* generalise
510 over the ?y parameter, to get
511 z :: (?y::Int) => Int,
512 but the monomorphism restriction says that we *must not*, giving
514 Why does the momomorphism restriction say this? Because if you have
516 let z = x + ?y in z+z
518 you might not expect the addition to be done twice --- but it will if
519 we follow the argument of Question 2 and generalise over ?y.
522 Question 4: top level
523 ~~~~~~~~~~~~~~~~~~~~~
524 At the top level, monomorhism makes no sense at all.
527 main = let ?x = 5 in print foo
531 woggle :: (?x :: Int) => Int -> Int
534 We definitely don't want (foo :: Int) with a top-level implicit parameter
535 (?x::Int) becuase there is no way to bind it.
540 (A) Always generalise over implicit parameters
541 Bindings that fall under the monomorphism restriction can't
545 * Inlining remains valid
546 * No unexpected loss of sharing
547 * But simple bindings like
549 will be rejected, unless you add an explicit type signature
550 (to avoid the monomorphism restriction)
551 z :: (?y::Int) => Int
553 This seems unacceptable
555 (B) Monomorphism restriction "wins"
556 Bindings that fall under the monomorphism restriction can't
558 Always generalise over implicit parameters *except* for bindings
559 that fall under the monomorphism restriction
562 * Inlining isn't valid in general
563 * No unexpected loss of sharing
564 * Simple bindings like
566 accepted (get value of ?y from binding site)
568 (C) Always generalise over implicit parameters
569 Bindings that fall under the monomorphism restriction can't
570 be generalised, EXCEPT for implicit parameters
572 * Inlining remains valid
573 * Unexpected loss of sharing (from the extra generalisation)
574 * Simple bindings like
576 accepted (get value of ?y from occurrence sites)
581 None of these choices seems very satisfactory. But at least we should
582 decide which we want to do.
584 It's really not clear what is the Right Thing To Do. If you see
588 would you expect the value of ?y to be got from the *occurrence sites*
589 of 'z', or from the valuue of ?y at the *definition* of 'z'? In the
590 case of function definitions, the answer is clearly the former, but
591 less so in the case of non-fucntion definitions. On the other hand,
592 if we say that we get the value of ?y from the definition site of 'z',
593 then inlining 'z' might change the semantics of the program.
595 Choice (C) really says "the monomorphism restriction doesn't apply
596 to implicit parameters". Which is fine, but remember that every
597 innocent binding 'x = ...' that mentions an implicit parameter in
598 the RHS becomes a *function* of that parameter, called at each
599 use of 'x'. Now, the chances are that there are no intervening 'with'
600 clauses that bind ?y, so a decent compiler should common up all
601 those function calls. So I think I strongly favour (C). Indeed,
602 one could make a similar argument for abolishing the monomorphism
603 restriction altogether.
605 BOTTOM LINE: we choose (B) at present. See tcSimplifyRestricted
609 %************************************************************************
611 \subsection{tcSimplifyInfer}
613 %************************************************************************
615 tcSimplify is called when we *inferring* a type. Here's the overall game plan:
617 1. Compute Q = grow( fvs(T), C )
619 2. Partition C based on Q into Ct and Cq. Notice that ambiguous
620 predicates will end up in Ct; we deal with them at the top level
622 3. Try improvement, using functional dependencies
624 4. If Step 3 did any unification, repeat from step 1
625 (Unification can change the result of 'grow'.)
627 Note: we don't reduce dictionaries in step 2. For example, if we have
628 Eq (a,b), we don't simplify to (Eq a, Eq b). So Q won't be different
629 after step 2. However note that we may therefore quantify over more
630 type variables than we absolutely have to.
632 For the guts, we need a loop, that alternates context reduction and
633 improvement with unification. E.g. Suppose we have
635 class C x y | x->y where ...
637 and tcSimplify is called with:
639 Then improvement unifies a with b, giving
642 If we need to unify anything, we rattle round the whole thing all over
649 -> TcTyVarSet -- fv(T); type vars
651 -> TcM ([TcTyVar], -- Tyvars to quantify (zonked and quantified)
652 [Inst], -- Dict Ids that must be bound here (zonked)
653 TcDictBinds) -- Bindings
654 -- Any free (escaping) Insts are tossed into the environment
659 tcSimplifyInfer doc tau_tvs wanted
660 = do { tau_tvs1 <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
661 ; wanted' <- mapM zonkInst wanted -- Zonk before deciding quantified tyvars
662 ; gbl_tvs <- tcGetGlobalTyVars
663 ; let preds1 = fdPredsOfInsts wanted'
664 gbl_tvs1 = oclose preds1 gbl_tvs
665 qtvs = grow preds1 tau_tvs1 `minusVarSet` gbl_tvs1
666 -- See Note [Choosing which variables to quantify]
668 -- To maximise sharing, remove from consideration any
669 -- constraints that don't mention qtvs at all
670 ; let (free, bound) = partition (isFreeWhenInferring qtvs) wanted'
673 -- To make types simple, reduce as much as possible
674 ; traceTc (text "infer" <+> (ppr preds1 $$ ppr (grow preds1 tau_tvs1) $$ ppr gbl_tvs $$
675 ppr gbl_tvs1 $$ ppr free $$ ppr bound))
676 ; (irreds1, binds1) <- tryHardCheckLoop doc bound
678 -- Note [Inference and implication constraints]
679 ; let want_dict d = tyVarsOfInst d `intersectsVarSet` qtvs
680 ; (irreds2, binds2) <- approximateImplications doc want_dict irreds1
682 -- Now work out all over again which type variables to quantify,
683 -- exactly in the same way as before, but starting from irreds2. Why?
684 -- a) By now improvment may have taken place, and we must *not*
685 -- quantify over any variable free in the environment
686 -- tc137 (function h inside g) is an example
688 -- b) Do not quantify over constraints that *now* do not
689 -- mention quantified type variables, because they are
690 -- simply ambiguous (or might be bound further out). Example:
691 -- f :: Eq b => a -> (a, b)
693 -- From the RHS of g we get the MethodInst f77 :: alpha -> (alpha, beta)
694 -- We decide to quantify over 'alpha' alone, but free1 does not include f77
695 -- because f77 mentions 'alpha'. Then reducing leaves only the (ambiguous)
696 -- constraint (Eq beta), which we dump back into the free set
697 -- See test tcfail181
699 -- c) irreds may contain type variables not previously mentioned,
700 -- e.g. instance D a x => Foo [a]
702 -- Then after simplifying we'll get (D a x), and x is fresh
703 -- We must quantify over x else it'll be totally unbound
704 ; tau_tvs2 <- zonkTcTyVarsAndFV (varSetElems tau_tvs1)
705 ; gbl_tvs2 <- zonkTcTyVarsAndFV (varSetElems gbl_tvs1)
706 -- Note that we start from gbl_tvs1
707 -- We use tcGetGlobalTyVars, then oclose wrt preds2, because
708 -- we've already put some of the original preds1 into frees
709 -- E.g. wanteds = C a b (where a->b)
712 -- Then b is fixed by gbl_tvs, so (C a b) will be in free, and
713 -- irreds2 will be empty. But we don't want to generalise over b!
714 ; let preds2 = fdPredsOfInsts irreds2 -- irreds2 is zonked
715 qtvs = grow preds2 tau_tvs2 `minusVarSet` oclose preds2 gbl_tvs2
716 ; let (free, irreds3) = partition (isFreeWhenInferring qtvs) irreds2
719 -- Turn the quantified meta-type variables into real type variables
720 ; qtvs2 <- zonkQuantifiedTyVars (varSetElems qtvs)
722 -- We can't abstract over any remaining unsolved
723 -- implications so instead just float them outwards. Ugh.
724 ; let (q_dicts0, implics) = partition isAbstractableInst irreds3
725 ; loc <- getInstLoc (ImplicOrigin doc)
726 ; implic_bind <- bindIrreds loc qtvs2 q_dicts0 implics
728 -- Prepare equality instances for quantification
729 ; let (q_eqs0,q_dicts) = partition isEqInst q_dicts0
730 ; q_eqs <- mapM finalizeEqInst q_eqs0
732 ; return (qtvs2, q_eqs ++ q_dicts, binds1 `unionBags` binds2 `unionBags` implic_bind) }
733 -- NB: when we are done, we might have some bindings, but
734 -- the final qtvs might be empty. See Note [NO TYVARS] below.
736 approximateImplications :: SDoc -> (Inst -> Bool) -> [Inst] -> TcM ([Inst], TcDictBinds)
737 -- Note [Inference and implication constraints]
738 -- Given a bunch of Dict and ImplicInsts, try to approximate the implications by
739 -- - fetching any dicts inside them that are free
740 -- - using those dicts as cruder constraints, to solve the implications
741 -- - returning the extra ones too
743 approximateImplications doc want_dict irreds
745 = return (irreds, emptyBag)
747 = do { extra_dicts' <- mapM cloneDict extra_dicts
748 ; tryHardCheckLoop doc (extra_dicts' ++ irreds) }
749 -- By adding extra_dicts', we make them
750 -- available to solve the implication constraints
752 extra_dicts = get_dicts (filter isImplicInst irreds)
754 get_dicts :: [Inst] -> [Inst] -- Returns only Dicts
755 -- Find the wanted constraints in implication constraints that satisfy
756 -- want_dict, and are not bound by forall's in the constraint itself
757 get_dicts ds = concatMap get_dict ds
759 get_dict d@(Dict {}) | want_dict d = [d]
761 get_dict (ImplicInst {tci_tyvars = tvs, tci_wanted = wanteds})
762 = [ d | let tv_set = mkVarSet tvs
763 , d <- get_dicts wanteds
764 , not (tyVarsOfInst d `intersectsVarSet` tv_set)]
765 get_dict i@(EqInst {}) | want_dict i = [i]
767 get_dict other = pprPanic "approximateImplications" (ppr other)
770 Note [Inference and implication constraints]
771 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
772 Suppose we have a wanted implication constraint (perhaps arising from
773 a nested pattern match) like
775 and we are now trying to quantify over 'a' when inferring the type for
776 a function. In principle it's possible that there might be an instance
777 instance (C a, E a) => D [a]
778 so the context (E a) would suffice. The Right Thing is to abstract over
779 the implication constraint, but we don't do that (a) because it'll be
780 surprising to programmers and (b) because we don't have the machinery to deal
781 with 'given' implications.
783 So our best approximation is to make (D [a]) part of the inferred
784 context, so we can use that to discharge the implication. Hence
785 the strange function get_dicts in approximateImplications.
787 The common cases are more clear-cut, when we have things like
789 Here, abstracting over (C b) is not an approximation at all -- but see
790 Note [Freeness and implications].
792 See Trac #1430 and test tc228.
796 -----------------------------------------------------------
797 -- tcSimplifyInferCheck is used when we know the constraints we are to simplify
798 -- against, but we don't know the type variables over which we are going to quantify.
799 -- This happens when we have a type signature for a mutually recursive group
802 -> TcTyVarSet -- fv(T)
805 -> TcM ([TyVar], -- Fully zonked, and quantified
806 TcDictBinds) -- Bindings
808 tcSimplifyInferCheck loc tau_tvs givens wanteds
809 = do { traceTc (text "tcSimplifyInferCheck <-" <+> ppr wanteds)
810 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
812 -- Figure out which type variables to quantify over
813 -- You might think it should just be the signature tyvars,
814 -- but in bizarre cases you can get extra ones
815 -- f :: forall a. Num a => a -> a
816 -- f x = fst (g (x, head [])) + 1
818 -- Here we infer g :: forall a b. a -> b -> (b,a)
819 -- We don't want g to be monomorphic in b just because
820 -- f isn't quantified over b.
821 ; let all_tvs = varSetElems (tau_tvs `unionVarSet` tyVarsOfInsts givens)
822 ; all_tvs <- zonkTcTyVarsAndFV all_tvs
823 ; gbl_tvs <- tcGetGlobalTyVars
824 ; let qtvs = varSetElems (all_tvs `minusVarSet` gbl_tvs)
825 -- We could close gbl_tvs, but its not necessary for
826 -- soundness, and it'll only affect which tyvars, not which
827 -- dictionaries, we quantify over
829 ; qtvs' <- zonkQuantifiedTyVars qtvs
831 -- Now we are back to normal (c.f. tcSimplCheck)
832 ; implic_bind <- bindIrreds loc qtvs' givens irreds
834 ; traceTc (text "tcSimplifyInferCheck ->" <+> ppr (implic_bind))
835 ; return (qtvs', binds `unionBags` implic_bind) }
838 Note [Squashing methods]
839 ~~~~~~~~~~~~~~~~~~~~~~~~~
840 Be careful if you want to float methods more:
841 truncate :: forall a. RealFrac a => forall b. Integral b => a -> b
842 From an application (truncate f i) we get
845 If we have also have a second occurrence of truncate, we get
848 When simplifying with i,f free, we might still notice that
849 t1=t3; but alas, the binding for t2 (which mentions t1)
850 may continue to float out!
855 class Y a b | a -> b where
858 instance Y [[a]] a where
861 k :: X a -> X a -> X a
863 g :: Num a => [X a] -> [X a]
866 h ys = ys ++ map (k (y [[0]])) xs
868 The excitement comes when simplifying the bindings for h. Initially
869 try to simplify {y @ [[t1]] t2, 0 @ t1}, with initial qtvs = {t2}.
870 From this we get t1:=:t2, but also various bindings. We can't forget
871 the bindings (because of [LOOP]), but in fact t1 is what g is
874 The net effect of [NO TYVARS]
877 isFreeWhenInferring :: TyVarSet -> Inst -> Bool
878 isFreeWhenInferring qtvs inst
879 = isFreeWrtTyVars qtvs inst -- Constrains no quantified vars
880 && isInheritableInst inst -- and no implicit parameter involved
881 -- see Note [Inheriting implicit parameters]
883 {- No longer used (with implication constraints)
884 isFreeWhenChecking :: TyVarSet -- Quantified tyvars
885 -> NameSet -- Quantified implicit parameters
887 isFreeWhenChecking qtvs ips inst
888 = isFreeWrtTyVars qtvs inst
889 && isFreeWrtIPs ips inst
892 isFreeWrtTyVars qtvs inst = tyVarsOfInst inst `disjointVarSet` qtvs
893 isFreeWrtIPs ips inst = not (any (`elemNameSet` ips) (ipNamesOfInst inst))
897 %************************************************************************
899 \subsection{tcSimplifyCheck}
901 %************************************************************************
903 @tcSimplifyCheck@ is used when we know exactly the set of variables
904 we are going to quantify over. For example, a class or instance declaration.
907 -----------------------------------------------------------
908 -- tcSimplifyCheck is used when checking expression type signatures,
909 -- class decls, instance decls etc.
910 tcSimplifyCheck :: InstLoc
911 -> [TcTyVar] -- Quantify over these
914 -> TcM TcDictBinds -- Bindings
915 tcSimplifyCheck loc qtvs givens wanteds
916 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
917 do { traceTc (text "tcSimplifyCheck")
918 ; (irreds, binds) <- gentleCheckLoop loc givens wanteds
919 ; implic_bind <- bindIrreds loc qtvs givens irreds
920 ; return (binds `unionBags` implic_bind) }
922 -----------------------------------------------------------
923 -- tcSimplifyCheckPat is used for existential pattern match
924 tcSimplifyCheckPat :: InstLoc
925 -> [TcTyVar] -- Quantify over these
928 -> TcM TcDictBinds -- Bindings
929 tcSimplifyCheckPat loc 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 givens irreds
934 ; return (binds `unionBags` implic_bind) }
936 -----------------------------------------------------------
937 bindIrreds :: InstLoc -> [TcTyVar]
940 bindIrreds loc qtvs givens irreds
941 = bindIrredsR loc qtvs givens irreds
943 bindIrredsR :: InstLoc -> [TcTyVar] -> [Inst] -> [Inst] -> TcM TcDictBinds
944 -- Make a binding that binds 'irreds', by generating an implication
945 -- constraint for them, *and* throwing the constraint into the LIE
946 bindIrredsR loc qtvs givens irreds
950 = do { let givens' = filter isAbstractableInst givens
951 -- The givens can (redundantly) include methods
952 -- We want to retain both EqInsts and Dicts
953 -- There should be no implicadtion constraints
954 -- See Note [Pruning the givens in an implication constraint]
956 -- If there are no 'givens', then it's safe to
957 -- partition the 'wanteds' by their qtvs, thereby trimming irreds
958 -- See Note [Freeness and implications]
959 ; irreds' <- if null givens'
961 { let qtv_set = mkVarSet qtvs
962 (frees, real_irreds) = partition (isFreeWrtTyVars qtv_set) irreds
964 ; return real_irreds }
967 ; (implics, bind) <- makeImplicationBind loc qtvs givens' irreds'
968 -- This call does the real work
969 -- If irreds' is empty, it does something sensible
974 makeImplicationBind :: InstLoc -> [TcTyVar]
976 -> TcM ([Inst], TcDictBinds)
977 -- Make a binding that binds 'irreds', by generating an implication
978 -- constraint for them, *and* throwing the constraint into the LIE
979 -- The binding looks like
980 -- (ir1, .., irn) = f qtvs givens
981 -- where f is (evidence for) the new implication constraint
982 -- f :: forall qtvs. {reft} givens => (ir1, .., irn)
983 -- qtvs includes coercion variables
985 -- This binding must line up the 'rhs' in reduceImplication
986 makeImplicationBind loc all_tvs
987 givens -- Guaranteed all Dicts
990 | null irreds -- If there are no irreds, we are done
991 = return ([], emptyBag)
992 | otherwise -- Otherwise we must generate a binding
993 = do { uniq <- newUnique
994 ; span <- getSrcSpanM
995 ; let (eq_givens, dict_givens) = partition isEqInst givens
996 eq_tyvar_cos = mkTyVarTys (varSetElems $ tyVarsOfTypes $ map eqInstType eq_givens)
997 -- Urgh! See line 2187 or thereabouts. I believe that all these
998 -- 'givens' must be a simple CoVar. This MUST be cleaned up.
1000 ; let name = mkInternalName uniq (mkVarOcc "ic") span
1001 implic_inst = ImplicInst { tci_name = name,
1002 tci_tyvars = all_tvs,
1003 tci_given = (eq_givens ++ dict_givens),
1004 tci_wanted = irreds, tci_loc = loc }
1005 ; let -- only create binder for dict_irreds
1006 (eq_irreds, dict_irreds) = partition isEqInst irreds
1007 n_dict_irreds = length dict_irreds
1008 dict_irred_ids = map instToId dict_irreds
1009 tup_ty = mkTupleTy Boxed n_dict_irreds (map idType dict_irred_ids)
1010 pat = TuplePat (map nlVarPat dict_irred_ids) Boxed tup_ty
1011 rhs = L span (mkHsWrap co (HsVar (instToId implic_inst)))
1012 co = mkWpApps (map instToId dict_givens)
1013 <.> mkWpTyApps eq_tyvar_cos
1014 <.> mkWpTyApps (mkTyVarTys all_tvs)
1015 bind | [dict_irred_id] <- dict_irred_ids = VarBind dict_irred_id rhs
1016 | otherwise = PatBind { pat_lhs = L span pat,
1017 pat_rhs = unguardedGRHSs rhs,
1018 pat_rhs_ty = tup_ty,
1019 bind_fvs = placeHolderNames }
1020 ; traceTc $ text "makeImplicationBind" <+> ppr implic_inst
1021 ; return ([implic_inst], unitBag (L span bind))
1024 -----------------------------------------------------------
1025 tryHardCheckLoop :: SDoc
1027 -> TcM ([Inst], TcDictBinds)
1029 tryHardCheckLoop doc wanteds
1030 = do { (irreds,binds) <- checkLoop (mkRedEnv doc try_me []) wanteds
1031 ; return (irreds,binds)
1034 try_me inst = ReduceMe AddSCs
1035 -- Here's the try-hard bit
1037 -----------------------------------------------------------
1038 gentleCheckLoop :: InstLoc
1041 -> TcM ([Inst], TcDictBinds)
1043 gentleCheckLoop inst_loc givens wanteds
1044 = do { (irreds,binds) <- checkLoop env wanteds
1045 ; return (irreds,binds)
1048 env = mkRedEnv (pprInstLoc inst_loc) try_me givens
1050 try_me inst | isMethodOrLit inst = ReduceMe AddSCs
1052 -- When checking against a given signature
1053 -- we MUST be very gentle: Note [Check gently]
1055 gentleInferLoop :: SDoc -> [Inst]
1056 -> TcM ([Inst], TcDictBinds)
1057 gentleInferLoop doc wanteds
1058 = do { (irreds, binds) <- checkLoop env wanteds
1059 ; return (irreds, binds) }
1061 env = mkRedEnv doc try_me []
1062 try_me inst | isMethodOrLit inst = ReduceMe AddSCs
1067 ~~~~~~~~~~~~~~~~~~~~
1068 We have to very careful about not simplifying too vigorously
1073 f :: Show b => T b -> b
1074 f (MkT x) = show [x]
1076 Inside the pattern match, which binds (a:*, x:a), we know that
1078 Hence we have a dictionary for Show [a] available; and indeed we
1079 need it. We are going to build an implication contraint
1080 forall a. (b~[a]) => Show [a]
1081 Later, we will solve this constraint using the knowledge (Show b)
1083 But we MUST NOT reduce (Show [a]) to (Show a), else the whole
1084 thing becomes insoluble. So we simplify gently (get rid of literals
1085 and methods only, plus common up equal things), deferring the real
1086 work until top level, when we solve the implication constraint
1087 with tryHardCheckLooop.
1091 -----------------------------------------------------------
1094 -> TcM ([Inst], TcDictBinds)
1095 -- Precondition: givens are completely rigid
1096 -- Postcondition: returned Insts are zonked
1098 checkLoop env wanteds
1099 = go env wanteds (return ())
1100 where go env wanteds elim_skolems
1101 = do { -- We do need to zonk the givens; cf Note [Zonking RedEnv]
1102 ; env' <- zonkRedEnv env
1103 ; wanteds' <- zonkInsts wanteds
1105 ; (improved, binds, irreds, elim_more_skolems)
1106 <- reduceContext env' wanteds'
1107 ; let elim_skolems' = elim_skolems >> elim_more_skolems
1109 ; if not improved then
1110 elim_skolems' >> return (irreds, binds)
1113 -- If improvement did some unification, we go round again.
1114 -- We start again with irreds, not wanteds
1115 -- Using an instance decl might have introduced a fresh type
1116 -- variable which might have been unified, so we'd get an
1117 -- infinite loop if we started again with wanteds!
1119 { (irreds1, binds1) <- go env' irreds elim_skolems'
1120 ; return (irreds1, binds `unionBags` binds1) } }
1123 Note [Zonking RedEnv]
1124 ~~~~~~~~~~~~~~~~~~~~~
1125 It might appear as if the givens in RedEnv are always rigid, but that is not
1126 necessarily the case for programs involving higher-rank types that have class
1127 contexts constraining the higher-rank variables. An example from tc237 in the
1130 class Modular s a | s -> a
1132 wim :: forall a w. Integral a
1133 => a -> (forall s. Modular s a => M s w) -> w
1134 wim i k = error "urk"
1136 test5 :: (Modular s a, Integral a) => M s a
1139 test4 = wim 4 test4'
1141 Notice how the variable 'a' of (Modular s a) in the rank-2 type of wim is
1142 quantified further outside. When type checking test4, we have to check
1143 whether the signature of test5 is an instance of
1145 (forall s. Modular s a => M s w)
1147 Consequently, we will get (Modular s t_a), where t_a is a TauTv into the
1150 Given the FD of Modular in this example, class improvement will instantiate
1151 t_a to 'a', where 'a' is the skolem from test5's signatures (due to the
1152 Modular s a predicate in that signature). If we don't zonk (Modular s t_a) in
1153 the givens, we will get into a loop as improveOne uses the unification engine
1154 Unify.tcUnifyTys, which doesn't know about mutable type variables.
1159 class If b t e r | b t e -> r
1162 class Lte a b c | a b -> c where lte :: a -> b -> c
1164 instance (Lte a b l,If l b a c) => Max a b c
1166 Wanted: Max Z (S x) y
1168 Then we'll reduce using the Max instance to:
1169 (Lte Z (S x) l, If l (S x) Z y)
1170 and improve by binding l->T, after which we can do some reduction
1171 on both the Lte and If constraints. What we *can't* do is start again
1172 with (Max Z (S x) y)!
1176 %************************************************************************
1178 tcSimplifySuperClasses
1180 %************************************************************************
1182 Note [SUPERCLASS-LOOP 1]
1183 ~~~~~~~~~~~~~~~~~~~~~~~~
1184 We have to be very, very careful when generating superclasses, lest we
1185 accidentally build a loop. Here's an example:
1189 class S a => C a where { opc :: a -> a }
1190 class S b => D b where { opd :: b -> b }
1192 instance C Int where
1195 instance D Int where
1198 From (instance C Int) we get the constraint set {ds1:S Int, dd:D Int}
1199 Simplifying, we may well get:
1200 $dfCInt = :C ds1 (opd dd)
1203 Notice that we spot that we can extract ds1 from dd.
1205 Alas! Alack! We can do the same for (instance D Int):
1207 $dfDInt = :D ds2 (opc dc)
1211 And now we've defined the superclass in terms of itself.
1213 Solution: never generate a superclass selectors at all when
1214 satisfying the superclass context of an instance declaration.
1216 Two more nasty cases are in
1221 tcSimplifySuperClasses
1226 tcSimplifySuperClasses loc givens sc_wanteds
1227 = do { traceTc (text "tcSimplifySuperClasses")
1228 ; (irreds,binds1) <- checkLoop env sc_wanteds
1229 ; let (tidy_env, tidy_irreds) = tidyInsts irreds
1230 ; reportNoInstances tidy_env (Just (loc, givens)) tidy_irreds
1233 env = mkRedEnv (pprInstLoc loc) try_me givens
1234 try_me inst = ReduceMe NoSCs
1235 -- Like tryHardCheckLoop, but with NoSCs
1239 %************************************************************************
1241 \subsection{tcSimplifyRestricted}
1243 %************************************************************************
1245 tcSimplifyRestricted infers which type variables to quantify for a
1246 group of restricted bindings. This isn't trivial.
1249 We want to quantify over a to get id :: forall a. a->a
1252 We do not want to quantify over a, because there's an Eq a
1253 constraint, so we get eq :: a->a->Bool (notice no forall)
1256 RHS has type 'tau', whose free tyvars are tau_tvs
1257 RHS has constraints 'wanteds'
1260 Quantify over (tau_tvs \ ftvs(wanteds))
1261 This is bad. The constraints may contain (Monad (ST s))
1262 where we have instance Monad (ST s) where...
1263 so there's no need to be monomorphic in s!
1265 Also the constraint might be a method constraint,
1266 whose type mentions a perfectly innocent tyvar:
1267 op :: Num a => a -> b -> a
1268 Here, b is unconstrained. A good example would be
1270 We want to infer the polymorphic type
1271 foo :: forall b. b -> b
1274 Plan B (cunning, used for a long time up to and including GHC 6.2)
1275 Step 1: Simplify the constraints as much as possible (to deal
1276 with Plan A's problem). Then set
1277 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1279 Step 2: Now simplify again, treating the constraint as 'free' if
1280 it does not mention qtvs, and trying to reduce it otherwise.
1281 The reasons for this is to maximise sharing.
1283 This fails for a very subtle reason. Suppose that in the Step 2
1284 a constraint (Foo (Succ Zero) (Succ Zero) b) gets thrown upstairs as 'free'.
1285 In the Step 1 this constraint might have been simplified, perhaps to
1286 (Foo Zero Zero b), AND THEN THAT MIGHT BE IMPROVED, to bind 'b' to 'T'.
1287 This won't happen in Step 2... but that in turn might prevent some other
1288 constraint (Baz [a] b) being simplified (e.g. via instance Baz [a] T where {..})
1289 and that in turn breaks the invariant that no constraints are quantified over.
1291 Test typecheck/should_compile/tc177 (which failed in GHC 6.2) demonstrates
1296 Step 1: Simplify the constraints as much as possible (to deal
1297 with Plan A's problem). Then set
1298 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1299 Return the bindings from Step 1.
1302 A note about Plan C (arising from "bug" reported by George Russel March 2004)
1305 instance (HasBinary ty IO) => HasCodedValue ty
1307 foo :: HasCodedValue a => String -> IO a
1309 doDecodeIO :: HasCodedValue a => () -> () -> IO a
1310 doDecodeIO codedValue view
1311 = let { act = foo "foo" } in act
1313 You might think this should work becuase the call to foo gives rise to a constraint
1314 (HasCodedValue t), which can be satisfied by the type sig for doDecodeIO. But the
1315 restricted binding act = ... calls tcSimplifyRestricted, and PlanC simplifies the
1316 constraint using the (rather bogus) instance declaration, and now we are stuffed.
1318 I claim this is not really a bug -- but it bit Sergey as well as George. So here's
1322 Plan D (a variant of plan B)
1323 Step 1: Simplify the constraints as much as possible (to deal
1324 with Plan A's problem), BUT DO NO IMPROVEMENT. Then set
1325 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1327 Step 2: Now simplify again, treating the constraint as 'free' if
1328 it does not mention qtvs, and trying to reduce it otherwise.
1330 The point here is that it's generally OK to have too few qtvs; that is,
1331 to make the thing more monomorphic than it could be. We don't want to
1332 do that in the common cases, but in wierd cases it's ok: the programmer
1333 can always add a signature.
1335 Too few qtvs => too many wanteds, which is what happens if you do less
1340 tcSimplifyRestricted -- Used for restricted binding groups
1341 -- i.e. ones subject to the monomorphism restriction
1344 -> [Name] -- Things bound in this group
1345 -> TcTyVarSet -- Free in the type of the RHSs
1346 -> [Inst] -- Free in the RHSs
1347 -> TcM ([TyVar], -- Tyvars to quantify (zonked and quantified)
1348 TcDictBinds) -- Bindings
1349 -- tcSimpifyRestricted returns no constraints to
1350 -- quantify over; by definition there are none.
1351 -- They are all thrown back in the LIE
1353 tcSimplifyRestricted doc top_lvl bndrs tau_tvs wanteds
1354 -- Zonk everything in sight
1355 = do { traceTc (text "tcSimplifyRestricted")
1356 ; wanteds' <- zonkInsts wanteds
1358 -- 'ReduceMe': Reduce as far as we can. Don't stop at
1359 -- dicts; the idea is to get rid of as many type
1360 -- variables as possible, and we don't want to stop
1361 -- at (say) Monad (ST s), because that reduces
1362 -- immediately, with no constraint on s.
1364 -- BUT do no improvement! See Plan D above
1365 -- HOWEVER, some unification may take place, if we instantiate
1366 -- a method Inst with an equality constraint
1367 ; let env = mkNoImproveRedEnv doc (\i -> ReduceMe AddSCs)
1368 ; (_imp, _binds, constrained_dicts, elim_skolems)
1369 <- reduceContext env wanteds'
1372 -- Next, figure out the tyvars we will quantify over
1373 ; tau_tvs' <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
1374 ; gbl_tvs' <- tcGetGlobalTyVars
1375 ; constrained_dicts' <- zonkInsts constrained_dicts
1377 ; let qtvs1 = tau_tvs' `minusVarSet` oclose (fdPredsOfInsts constrained_dicts) gbl_tvs'
1378 -- As in tcSimplifyInfer
1380 -- Do not quantify over constrained type variables:
1381 -- this is the monomorphism restriction
1382 constrained_tvs' = tyVarsOfInsts constrained_dicts'
1383 qtvs = qtvs1 `minusVarSet` constrained_tvs'
1384 pp_bndrs = pprWithCommas (quotes . ppr) bndrs
1387 ; warn_mono <- doptM Opt_WarnMonomorphism
1388 ; warnTc (warn_mono && (constrained_tvs' `intersectsVarSet` qtvs1))
1389 (vcat[ ptext (sLit "the Monomorphism Restriction applies to the binding")
1390 <> plural bndrs <+> ptext (sLit "for") <+> pp_bndrs,
1391 ptext (sLit "Consider giving a type signature for") <+> pp_bndrs])
1393 ; traceTc (text "tcSimplifyRestricted" <+> vcat [
1394 pprInsts wanteds, pprInsts constrained_dicts',
1396 ppr constrained_tvs', ppr tau_tvs', ppr qtvs ])
1398 -- The first step may have squashed more methods than
1399 -- necessary, so try again, this time more gently, knowing the exact
1400 -- set of type variables to quantify over.
1402 -- We quantify only over constraints that are captured by qtvs;
1403 -- these will just be a subset of non-dicts. This in contrast
1404 -- to normal inference (using isFreeWhenInferring) in which we quantify over
1405 -- all *non-inheritable* constraints too. This implements choice
1406 -- (B) under "implicit parameter and monomorphism" above.
1408 -- Remember that we may need to do *some* simplification, to
1409 -- (for example) squash {Monad (ST s)} into {}. It's not enough
1410 -- just to float all constraints
1412 -- At top level, we *do* squash methods becuase we want to
1413 -- expose implicit parameters to the test that follows
1414 ; let is_nested_group = isNotTopLevel top_lvl
1415 try_me inst | isFreeWrtTyVars qtvs inst,
1416 (is_nested_group || isDict inst) = Stop
1417 | otherwise = ReduceMe AddSCs
1418 env = mkNoImproveRedEnv doc try_me
1419 ; (_imp, binds, irreds, elim_skolems) <- reduceContext env wanteds'
1422 -- See "Notes on implicit parameters, Question 4: top level"
1423 ; ASSERT( all (isFreeWrtTyVars qtvs) irreds ) -- None should be captured
1424 if is_nested_group then
1426 else do { let (bad_ips, non_ips) = partition isIPDict irreds
1427 ; addTopIPErrs bndrs bad_ips
1428 ; extendLIEs non_ips }
1430 ; qtvs' <- zonkQuantifiedTyVars (varSetElems qtvs)
1431 ; return (qtvs', binds) }
1435 %************************************************************************
1439 %************************************************************************
1441 On the LHS of transformation rules we only simplify methods and constants,
1442 getting dictionaries. We want to keep all of them unsimplified, to serve
1443 as the available stuff for the RHS of the rule.
1445 Example. Consider the following left-hand side of a rule
1447 f (x == y) (y > z) = ...
1449 If we typecheck this expression we get constraints
1451 d1 :: Ord a, d2 :: Eq a
1453 We do NOT want to "simplify" to the LHS
1455 forall x::a, y::a, z::a, d1::Ord a.
1456 f ((==) (eqFromOrd d1) x y) ((>) d1 y z) = ...
1460 forall x::a, y::a, z::a, d1::Ord a, d2::Eq a.
1461 f ((==) d2 x y) ((>) d1 y z) = ...
1463 Here is another example:
1465 fromIntegral :: (Integral a, Num b) => a -> b
1466 {-# RULES "foo" fromIntegral = id :: Int -> Int #-}
1468 In the rule, a=b=Int, and Num Int is a superclass of Integral Int. But
1469 we *dont* want to get
1471 forall dIntegralInt.
1472 fromIntegral Int Int dIntegralInt (scsel dIntegralInt) = id Int
1474 because the scsel will mess up RULE matching. Instead we want
1476 forall dIntegralInt, dNumInt.
1477 fromIntegral Int Int dIntegralInt dNumInt = id Int
1481 g (x == y) (y == z) = ..
1483 where the two dictionaries are *identical*, we do NOT WANT
1485 forall x::a, y::a, z::a, d1::Eq a
1486 f ((==) d1 x y) ((>) d1 y z) = ...
1488 because that will only match if the dict args are (visibly) equal.
1489 Instead we want to quantify over the dictionaries separately.
1491 In short, tcSimplifyRuleLhs must *only* squash LitInst and MethInts, leaving
1492 all dicts unchanged, with absolutely no sharing. It's simpler to do this
1493 from scratch, rather than further parameterise simpleReduceLoop etc
1496 tcSimplifyRuleLhs :: [Inst] -> TcM ([Inst], TcDictBinds)
1497 tcSimplifyRuleLhs wanteds
1498 = go [] emptyBag wanteds
1501 = return (dicts, binds)
1502 go dicts binds (w:ws)
1504 = go (w:dicts) binds ws
1506 = do { w' <- zonkInst w -- So that (3::Int) does not generate a call
1507 -- to fromInteger; this looks fragile to me
1508 ; lookup_result <- lookupSimpleInst w'
1509 ; case lookup_result of
1511 go dicts (addInstToDictBind binds w rhs) (ws' ++ ws)
1512 NoInstance -> pprPanic "tcSimplifyRuleLhs" (ppr w)
1516 tcSimplifyBracket is used when simplifying the constraints arising from
1517 a Template Haskell bracket [| ... |]. We want to check that there aren't
1518 any constraints that can't be satisfied (e.g. Show Foo, where Foo has no
1519 Show instance), but we aren't otherwise interested in the results.
1520 Nor do we care about ambiguous dictionaries etc. We will type check
1521 this bracket again at its usage site.
1524 tcSimplifyBracket :: [Inst] -> TcM ()
1525 tcSimplifyBracket wanteds
1526 = do { tryHardCheckLoop doc wanteds
1529 doc = text "tcSimplifyBracket"
1533 %************************************************************************
1535 \subsection{Filtering at a dynamic binding}
1537 %************************************************************************
1542 we must discharge all the ?x constraints from B. We also do an improvement
1543 step; if we have ?x::t1 and ?x::t2 we must unify t1, t2.
1545 Actually, the constraints from B might improve the types in ?x. For example
1547 f :: (?x::Int) => Char -> Char
1550 then the constraint (?x::Int) arising from the call to f will
1551 force the binding for ?x to be of type Int.
1554 tcSimplifyIPs :: [Inst] -- The implicit parameters bound here
1557 -- We need a loop so that we do improvement, and then
1558 -- (next time round) generate a binding to connect the two
1560 -- Here the two ?x's have different types, and improvement
1561 -- makes them the same.
1563 tcSimplifyIPs given_ips wanteds
1564 = do { wanteds' <- zonkInsts wanteds
1565 ; given_ips' <- zonkInsts given_ips
1566 -- Unusually for checking, we *must* zonk the given_ips
1568 ; let env = mkRedEnv doc try_me given_ips'
1569 ; (improved, binds, irreds, elim_skolems) <- reduceContext env wanteds'
1572 ; if not improved then
1573 ASSERT( all is_free irreds )
1574 do { extendLIEs irreds
1577 tcSimplifyIPs given_ips wanteds }
1579 doc = text "tcSimplifyIPs" <+> ppr given_ips
1580 ip_set = mkNameSet (ipNamesOfInsts given_ips)
1581 is_free inst = isFreeWrtIPs ip_set inst
1583 -- Simplify any methods that mention the implicit parameter
1584 try_me inst | is_free inst = Stop
1585 | otherwise = ReduceMe NoSCs
1589 %************************************************************************
1591 \subsection[binds-for-local-funs]{@bindInstsOfLocalFuns@}
1593 %************************************************************************
1595 When doing a binding group, we may have @Insts@ of local functions.
1596 For example, we might have...
1598 let f x = x + 1 -- orig local function (overloaded)
1599 f.1 = f Int -- two instances of f
1604 The point is: we must drop the bindings for @f.1@ and @f.2@ here,
1605 where @f@ is in scope; those @Insts@ must certainly not be passed
1606 upwards towards the top-level. If the @Insts@ were binding-ified up
1607 there, they would have unresolvable references to @f@.
1609 We pass in an @init_lie@ of @Insts@ and a list of locally-bound @Ids@.
1610 For each method @Inst@ in the @init_lie@ that mentions one of the
1611 @Ids@, we create a binding. We return the remaining @Insts@ (in an
1612 @LIE@), as well as the @HsBinds@ generated.
1615 bindInstsOfLocalFuns :: [Inst] -> [TcId] -> TcM TcDictBinds
1616 -- Simlifies only MethodInsts, and generate only bindings of form
1618 -- We're careful not to even generate bindings of the form
1620 -- You'd think that'd be fine, but it interacts with what is
1621 -- arguably a bug in Match.tidyEqnInfo (see notes there)
1623 bindInstsOfLocalFuns wanteds local_ids
1624 | null overloaded_ids = do
1627 return emptyLHsBinds
1630 = do { (irreds, binds) <- gentleInferLoop doc for_me
1631 ; extendLIEs not_for_me
1635 doc = text "bindInsts" <+> ppr local_ids
1636 overloaded_ids = filter is_overloaded local_ids
1637 is_overloaded id = isOverloadedTy (idType id)
1638 (for_me, not_for_me) = partition (isMethodFor overloaded_set) wanteds
1640 overloaded_set = mkVarSet overloaded_ids -- There can occasionally be a lot of them
1641 -- so it's worth building a set, so that
1642 -- lookup (in isMethodFor) is faster
1646 %************************************************************************
1648 \subsection{Data types for the reduction mechanism}
1650 %************************************************************************
1652 The main control over context reduction is here
1656 = RedEnv { red_doc :: SDoc -- The context
1657 , red_try_me :: Inst -> WhatToDo
1658 , red_improve :: Bool -- True <=> do improvement
1659 , red_givens :: [Inst] -- All guaranteed rigid
1661 -- but see Note [Rigidity]
1662 , red_stack :: (Int, [Inst]) -- Recursion stack (for err msg)
1663 -- See Note [RedStack]
1667 -- The red_givens are rigid so far as cmpInst is concerned.
1668 -- There is one case where they are not totally rigid, namely in tcSimplifyIPs
1669 -- let ?x = e in ...
1670 -- Here, the given is (?x::a), where 'a' is not necy a rigid type
1671 -- But that doesn't affect the comparison, which is based only on mame.
1674 -- The red_stack pair (n,insts) pair is just used for error reporting.
1675 -- 'n' is always the depth of the stack.
1676 -- The 'insts' is the stack of Insts being reduced: to produce X
1677 -- I had to produce Y, to produce Y I had to produce Z, and so on.
1680 mkRedEnv :: SDoc -> (Inst -> WhatToDo) -> [Inst] -> RedEnv
1681 mkRedEnv doc try_me givens
1682 = RedEnv { red_doc = doc, red_try_me = try_me,
1683 red_givens = givens,
1685 red_improve = True }
1687 mkNoImproveRedEnv :: SDoc -> (Inst -> WhatToDo) -> RedEnv
1688 -- Do not do improvement; no givens
1689 mkNoImproveRedEnv doc try_me
1690 = RedEnv { red_doc = doc, red_try_me = try_me,
1693 red_improve = True }
1696 = ReduceMe WantSCs -- Try to reduce this
1697 -- If there's no instance, add the inst to the
1698 -- irreductible ones, but don't produce an error
1699 -- message of any kind.
1700 -- It might be quite legitimate such as (Eq a)!
1702 | Stop -- Return as irreducible unless it can
1703 -- be reduced to a constant in one step
1704 -- Do not add superclasses; see
1706 data WantSCs = NoSCs | AddSCs -- Tells whether we should add the superclasses
1707 -- of a predicate when adding it to the avails
1708 -- The reason for this flag is entirely the super-class loop problem
1709 -- Note [SUPER-CLASS LOOP 1]
1711 zonkRedEnv :: RedEnv -> TcM RedEnv
1713 = do { givens' <- mapM zonkInst (red_givens env)
1714 ; return $ env {red_givens = givens'}
1719 %************************************************************************
1721 \subsection[reduce]{@reduce@}
1723 %************************************************************************
1725 Note [Ancestor Equalities]
1726 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1727 During context reduction, we add to the wanted equalities also those
1728 equalities that (transitively) occur in superclass contexts of wanted
1729 class constraints. Consider the following code
1731 class a ~ Int => C a
1734 If (C a) is wanted, we want to add (a ~ Int), which will be discharged by
1735 substituting Int for a. Hence, we ultimately want (C Int), which we
1736 discharge with the explicit instance.
1739 reduceContext :: RedEnv
1741 -> TcM (ImprovementDone,
1742 TcDictBinds, -- Dictionary bindings
1743 [Inst], -- Irreducible
1744 TcM ()) -- Undo skolems from SkolemOccurs
1746 reduceContext env wanteds
1747 = do { traceTc (text "reduceContext" <+> (vcat [
1748 text "----------------------",
1750 text "given" <+> ppr (red_givens env),
1751 text "wanted" <+> ppr wanteds,
1752 text "----------------------"
1756 ; let givens = red_givens env
1757 (given_eqs0, given_dicts0) = partition isEqInst givens
1758 (wanted_eqs0, wanted_non_eqs) = partition isEqInst wanteds
1759 (wanted_implics0, wanted_dicts) = partition isImplicInst wanted_non_eqs
1761 -- We want to add as wanted equalities those that (transitively)
1762 -- occur in superclass contexts of wanted class constraints.
1763 -- See Note [Ancestor Equalities]
1764 ; ancestor_eqs <- ancestorEqualities wanted_dicts
1765 ; let wanted_eqs = wanted_eqs0 ++ ancestor_eqs
1766 ; traceTc $ text "reduceContext: ancestor eqs" <+> ppr ancestor_eqs
1768 -- 1. Normalise the *given* *equality* constraints
1769 ; (given_eqs, eliminate_skolems) <- normaliseGivenEqs given_eqs0
1771 -- 2. Normalise the *given* *dictionary* constraints
1772 -- wrt. the toplevel and given equations
1773 ; (given_dicts, given_binds) <- normaliseGivenDicts given_eqs
1776 -- 5. Build the Avail mapping from "given_dicts"
1777 ; (init_state, extra_givens) <- getLIE $ do
1778 { init_state <- foldlM addGiven emptyAvails given_dicts
1782 -- *** ToDo: what to do with the "extra_givens"? For the
1783 -- moment I'm simply discarding them, which is probably wrong
1785 -- 6. Solve the *wanted* *dictionary* constraints (not implications)
1786 -- This may expose some further equational constraints...
1787 ; (avails, extra_eqs) <- getLIE (reduceList env wanted_dicts init_state)
1788 ; (dict_binds, bound_dicts, dict_irreds)
1789 <- extractResults avails wanted_dicts
1790 ; traceTc $ text "reduceContext extractresults" <+> vcat
1791 [ppr avails, ppr wanted_dicts, ppr dict_binds]
1793 -- Solve the wanted *implications*. In doing so, we can provide
1794 -- as "given" all the dicts that were originally given,
1795 -- *or* for which we now have bindings,
1796 -- *or* which are now irreds
1797 ; let implic_env = env { red_givens = givens ++ bound_dicts
1799 ; (implic_binds_s, implic_irreds_s)
1800 <- mapAndUnzipM (reduceImplication implic_env) wanted_implics0
1801 ; let implic_binds = unionManyBags implic_binds_s
1802 implic_irreds = concat implic_irreds_s
1804 -- Normalise the wanted equality constraints
1805 ; eq_irreds <- normaliseWantedEqs given_eqs (wanted_eqs ++ extra_eqs)
1807 -- Normalise the wanted dictionaries
1808 ; let irreds = dict_irreds ++ implic_irreds
1809 eqs = eq_irreds ++ given_eqs
1810 ; (norm_irreds, normalise_binds) <- normaliseWantedDicts eqs irreds
1812 -- Figure out whether we should go round again. We do so in either
1814 -- (1) If any of the mutable tyvars in givens or irreds has been
1815 -- filled in by improvement, there is merit in going around
1816 -- again, because we may make further progress.
1817 -- (2) If we managed to normalise any dicts, there is merit in going
1818 -- around gain, because reduceList may be able to get further.
1820 -- ToDo: We may have exposed new
1821 -- equality constraints and should probably go round again
1822 -- then as well. But currently we are dropping them on the
1825 ; let all_irreds = norm_irreds ++ eq_irreds
1826 ; improvedMetaTy <- anyM isFilledMetaTyVar $ varSetElems $
1827 tyVarsOfInsts (givens ++ all_irreds)
1828 ; let improvedDicts = not $ isEmptyBag normalise_binds
1829 improved = improvedMetaTy || improvedDicts
1831 -- The old plan (fragile)
1832 -- improveed = availsImproved avails
1833 -- || (not $ isEmptyBag normalise_binds1)
1834 -- || (not $ isEmptyBag normalise_binds2)
1835 -- || (any isEqInst irreds)
1837 ; traceTc (text "reduceContext end" <+> (vcat [
1838 text "----------------------",
1840 text "given" <+> ppr givens,
1841 text "given_eqs" <+> ppr given_eqs,
1842 text "wanted" <+> ppr wanteds,
1843 text "wanted_dicts" <+> ppr wanted_dicts,
1845 text "avails" <+> pprAvails avails,
1846 text "improved =" <+> ppr improved,
1847 text "(all) irreds = " <+> ppr all_irreds,
1848 text "dict-binds = " <+> ppr dict_binds,
1849 text "implic-binds = " <+> ppr implic_binds,
1850 text "----------------------"
1854 given_binds `unionBags` normalise_binds
1855 `unionBags` dict_binds
1856 `unionBags` implic_binds,
1861 tcImproveOne :: Avails -> Inst -> TcM ImprovementDone
1862 tcImproveOne avails inst
1863 | not (isDict inst) = return False
1865 = do { inst_envs <- tcGetInstEnvs
1866 ; let eqns = improveOne (classInstances inst_envs)
1867 (dictPred inst, pprInstArising inst)
1868 [ (dictPred p, pprInstArising p)
1869 | p <- availsInsts avails, isDict p ]
1870 -- Avails has all the superclasses etc (good)
1871 -- It also has all the intermediates of the deduction (good)
1872 -- It does not have duplicates (good)
1873 -- NB that (?x::t1) and (?x::t2) will be held separately in avails
1874 -- so that improve will see them separate
1875 ; traceTc (text "improveOne" <+> ppr inst)
1878 unifyEqns :: [(Equation,(PredType,SDoc),(PredType,SDoc))]
1879 -> TcM ImprovementDone
1880 unifyEqns [] = return False
1882 = do { traceTc (ptext (sLit "Improve:") <+> vcat (map pprEquationDoc eqns))
1886 unify ((qtvs, pairs), what1, what2)
1887 = addErrCtxtM (mkEqnMsg what1 what2) $ do
1888 (_, _, tenv) <- tcInstTyVars (varSetElems qtvs)
1889 mapM_ (unif_pr tenv) pairs
1890 unif_pr tenv (ty1,ty2) = unifyType (substTy tenv ty1) (substTy tenv ty2)
1892 pprEquationDoc (eqn, (p1,w1), (p2,w2)) = vcat [pprEquation eqn, nest 2 (ppr p1), nest 2 (ppr p2)]
1894 mkEqnMsg (pred1,from1) (pred2,from2) tidy_env
1895 = do { pred1' <- zonkTcPredType pred1; pred2' <- zonkTcPredType pred2
1896 ; let { pred1'' = tidyPred tidy_env pred1'; pred2'' = tidyPred tidy_env pred2' }
1897 ; let msg = vcat [ptext (sLit "When using functional dependencies to combine"),
1898 nest 2 (sep [ppr pred1'' <> comma, nest 2 from1]),
1899 nest 2 (sep [ppr pred2'' <> comma, nest 2 from2])]
1900 ; return (tidy_env, msg) }
1903 The main context-reduction function is @reduce@. Here's its game plan.
1906 reduceList :: RedEnv -> [Inst] -> Avails -> TcM Avails
1907 reduceList env@(RedEnv {red_stack = (n,stk)}) wanteds state
1908 = do { traceTc (text "reduceList " <+> (ppr wanteds $$ ppr state))
1910 ; when (debugIsOn && (n > 8)) $ do
1911 dumpTcRn (hang (ptext (sLit "Interesting! Context reduction stack depth") <+> int n)
1912 2 (ifPprDebug (nest 2 (pprStack stk))))
1913 ; if n >= ctxtStkDepth dopts then
1914 failWithTc (reduceDepthErr n stk)
1918 go [] state = return state
1919 go (w:ws) state = do { state' <- reduce (env {red_stack = (n+1, w:stk)}) w state
1922 -- Base case: we're done!
1923 reduce env wanted avails
1924 -- It's the same as an existing inst, or a superclass thereof
1925 | Just avail <- findAvail avails wanted
1926 = do { traceTc (text "reduce: found " <+> ppr wanted)
1931 = do { traceTc (text "reduce" <+> ppr wanted $$ ppr avails)
1932 ; case red_try_me env wanted of {
1933 Stop -> try_simple (addIrred NoSCs);
1934 -- See Note [No superclasses for Stop]
1936 ReduceMe want_scs -> do -- It should be reduced
1937 { (avails, lookup_result) <- reduceInst env avails wanted
1938 ; case lookup_result of
1939 NoInstance -> addIrred want_scs avails wanted
1940 -- Add it and its superclasses
1942 GenInst [] rhs -> addWanted want_scs avails wanted rhs []
1944 GenInst wanteds' rhs
1945 -> do { avails1 <- addIrred NoSCs avails wanted
1946 ; avails2 <- reduceList env wanteds' avails1
1947 ; addWanted want_scs avails2 wanted rhs wanteds' } }
1948 -- Temporarily do addIrred *before* the reduceList,
1949 -- which has the effect of adding the thing we are trying
1950 -- to prove to the database before trying to prove the things it
1951 -- needs. See note [RECURSIVE DICTIONARIES]
1952 -- NB: we must not do an addWanted before, because that adds the
1953 -- superclasses too, and that can lead to a spurious loop; see
1954 -- the examples in [SUPERCLASS-LOOP]
1955 -- So we do an addIrred before, and then overwrite it afterwards with addWanted
1958 -- First, see if the inst can be reduced to a constant in one step
1959 -- Works well for literals (1::Int) and constant dictionaries (d::Num Int)
1960 -- Don't bother for implication constraints, which take real work
1961 try_simple do_this_otherwise
1962 = do { res <- lookupSimpleInst wanted
1964 GenInst [] rhs -> addWanted AddSCs avails wanted rhs []
1965 other -> do_this_otherwise avails wanted }
1969 Note [SUPERCLASS-LOOP 2]
1970 ~~~~~~~~~~~~~~~~~~~~~~~~
1971 But the above isn't enough. Suppose we are *given* d1:Ord a,
1972 and want to deduce (d2:C [a]) where
1974 class Ord a => C a where
1975 instance Ord [a] => C [a] where ...
1977 Then we'll use the instance decl to deduce C [a] from Ord [a], and then add the
1978 superclasses of C [a] to avails. But we must not overwrite the binding
1979 for Ord [a] (which is obtained from Ord a) with a superclass selection or we'll just
1982 Here's another variant, immortalised in tcrun020
1983 class Monad m => C1 m
1984 class C1 m => C2 m x
1985 instance C2 Maybe Bool
1986 For the instance decl we need to build (C1 Maybe), and it's no good if
1987 we run around and add (C2 Maybe Bool) and its superclasses to the avails
1988 before we search for C1 Maybe.
1990 Here's another example
1991 class Eq b => Foo a b
1992 instance Eq a => Foo [a] a
1996 we'll first deduce that it holds (via the instance decl). We must not
1997 then overwrite the Eq t constraint with a superclass selection!
1999 At first I had a gross hack, whereby I simply did not add superclass constraints
2000 in addWanted, though I did for addGiven and addIrred. This was sub-optimal,
2001 becuase it lost legitimate superclass sharing, and it still didn't do the job:
2002 I found a very obscure program (now tcrun021) in which improvement meant the
2003 simplifier got two bites a the cherry... so something seemed to be an Stop
2004 first time, but reducible next time.
2006 Now we implement the Right Solution, which is to check for loops directly
2007 when adding superclasses. It's a bit like the occurs check in unification.
2010 Note [RECURSIVE DICTIONARIES]
2011 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2013 data D r = ZeroD | SuccD (r (D r));
2015 instance (Eq (r (D r))) => Eq (D r) where
2016 ZeroD == ZeroD = True
2017 (SuccD a) == (SuccD b) = a == b
2020 equalDC :: D [] -> D [] -> Bool;
2023 We need to prove (Eq (D [])). Here's how we go:
2027 by instance decl, holds if
2031 by instance decl of Eq, holds if
2033 where d2 = dfEqList d3
2036 But now we can "tie the knot" to give
2042 and it'll even run! The trick is to put the thing we are trying to prove
2043 (in this case Eq (D []) into the database before trying to prove its
2044 contributing clauses.
2047 %************************************************************************
2049 Reducing a single constraint
2051 %************************************************************************
2054 ---------------------------------------------
2055 reduceInst :: RedEnv -> Avails -> Inst -> TcM (Avails, LookupInstResult)
2056 reduceInst env avails other_inst
2057 = do { result <- lookupSimpleInst other_inst
2058 ; return (avails, result) }
2061 Note [Equational Constraints in Implication Constraints]
2062 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2064 An implication constraint is of the form
2066 where Given and Wanted may contain both equational and dictionary
2067 constraints. The delay and reduction of these two kinds of constraints
2070 -) In the generated code, wanted Dictionary constraints are wrapped up in an
2071 implication constraint that is created at the code site where the wanted
2072 dictionaries can be reduced via a let-binding. This let-bound implication
2073 constraint is deconstructed at the use-site of the wanted dictionaries.
2075 -) While the reduction of equational constraints is also delayed, the delay
2076 is not manifest in the generated code. The required evidence is generated
2077 in the code directly at the use-site. There is no let-binding and deconstruction
2078 necessary. The main disadvantage is that we cannot exploit sharing as the
2079 same evidence may be generated at multiple use-sites. However, this disadvantage
2080 is limited because it only concerns coercions which are erased.
2082 The different treatment is motivated by the different in representation. Dictionary
2083 constraints require manifest runtime dictionaries, while equations require coercions
2087 ---------------------------------------------
2088 reduceImplication :: RedEnv
2090 -> TcM (TcDictBinds, [Inst])
2093 Suppose we are simplifying the constraint
2094 forall bs. extras => wanted
2095 in the context of an overall simplification problem with givens 'givens'.
2098 * The 'givens' need not mention any of the quantified type variables
2099 e.g. forall {}. Eq a => Eq [a]
2100 forall {}. C Int => D (Tree Int)
2102 This happens when you have something like
2104 T1 :: Eq a => a -> T a
2107 f x = ...(case x of { T1 v -> v==v })...
2110 -- ToDo: should we instantiate tvs? I think it's not necessary
2112 -- Note on coercion variables:
2114 -- The extra given coercion variables are bound at two different sites:
2115 -- -) in the creation context of the implication constraint
2116 -- the solved equational constraints use these binders
2118 -- -) at the solving site of the implication constraint
2119 -- the solved dictionaries use these binders
2120 -- these binders are generated by reduceImplication
2122 reduceImplication env
2123 orig_implic@(ImplicInst { tci_name = name, tci_loc = inst_loc,
2125 tci_given = extra_givens, tci_wanted = wanteds })
2126 = do { -- Solve the sub-problem
2127 ; let try_me inst = ReduceMe AddSCs -- Note [Freeness and implications]
2128 env' = env { red_givens = extra_givens ++ red_givens env
2129 , red_doc = sep [ptext (sLit "reduceImplication for")
2131 nest 2 (parens $ ptext (sLit "within")
2133 , red_try_me = try_me }
2135 ; traceTc (text "reduceImplication" <+> vcat
2136 [ ppr (red_givens env), ppr extra_givens,
2138 ; (irreds, binds) <- checkLoop env' wanteds
2139 ; let (extra_eq_givens, extra_dict_givens) = partition isEqInst extra_givens
2140 -- SLPJ Sept 07: I think this is bogus; currently
2141 -- there are no Eqinsts in extra_givens
2142 dict_ids = map instToId extra_dict_givens
2144 -- Note [Reducing implication constraints]
2145 -- Tom -- update note, put somewhere!
2147 ; traceTc (text "reduceImplication result" <+> vcat
2148 [ppr irreds, ppr binds])
2150 ; -- extract superclass binds
2151 -- (sc_binds,_) <- extractResults avails []
2152 -- ; traceTc (text "reduceImplication sc_binds" <+> vcat
2153 -- [ppr sc_binds, ppr avails])
2156 -- SLPJ Sept 07: what if improvement happened inside the checkLoop?
2157 -- Then we must iterate the outer loop too!
2159 ; traceTc (text "reduceImplication condition" <+> ppr ((isEmptyLHsBinds binds) || (null irreds)))
2161 -- Progress is no longer measered by the number of bindings
2162 ; if (isEmptyLHsBinds binds) && (not $ null irreds) then -- No progress
2163 -- If there are any irreds, we back off and do nothing
2164 return (emptyBag, [orig_implic])
2166 { (simpler_implic_insts, bind)
2167 <- makeImplicationBind inst_loc tvs extra_givens irreds
2168 -- This binding is useless if the recursive simplification
2169 -- made no progress; but currently we don't try to optimise that
2170 -- case. After all, we only try hard to reduce at top level, or
2171 -- when inferring types.
2173 ; let dict_wanteds = filter (not . isEqInst) wanteds
2174 -- TOMDO: given equational constraints bug!
2175 -- we need a different evidence for given
2176 -- equations depending on whether we solve
2177 -- dictionary constraints or equational constraints
2179 eq_tyvars = varSetElems $ tyVarsOfTypes $ map eqInstType extra_eq_givens
2180 -- SLPJ Sept07: this looks Utterly Wrong to me, but I think
2181 -- that current extra_givens has no EqInsts, so
2182 -- it makes no difference
2183 co = wrap_inline -- Note [Always inline implication constraints]
2185 <.> mkWpLams eq_tyvars
2186 <.> mkWpLams dict_ids
2187 <.> WpLet (binds `unionBags` bind)
2188 wrap_inline | null dict_ids = idHsWrapper
2189 | otherwise = WpInline
2190 rhs = mkHsWrap co payload
2191 loc = instLocSpan inst_loc
2192 payload | [dict_wanted] <- dict_wanteds = HsVar (instToId dict_wanted)
2193 | otherwise = ExplicitTuple (map (L loc . HsVar . instToId) dict_wanteds) Boxed
2196 ; traceTc (vcat [text "reduceImplication" <+> ppr name,
2197 ppr simpler_implic_insts,
2198 text "->" <+> ppr rhs])
2199 ; return (unitBag (L loc (VarBind (instToId orig_implic) (L loc rhs))),
2200 simpler_implic_insts)
2205 Note [Always inline implication constraints]
2206 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2207 Suppose an implication constraint floats out of an INLINE function.
2208 Then although the implication has a single call site, it won't be
2209 inlined. And that is bad because it means that even if there is really
2210 *no* overloading (type signatures specify the exact types) there will
2211 still be dictionary passing in the resulting code. To avert this,
2212 we mark the implication constraints themselves as INLINE, at least when
2213 there is no loss of sharing as a result.
2215 Note [Freeness and implications]
2216 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2217 It's hard to say when an implication constraint can be floated out. Consider
2218 forall {} Eq a => Foo [a]
2219 The (Foo [a]) doesn't mention any of the quantified variables, but it
2220 still might be partially satisfied by the (Eq a).
2222 There is a useful special case when it *is* easy to partition the
2223 constraints, namely when there are no 'givens'. Consider
2224 forall {a}. () => Bar b
2225 There are no 'givens', and so there is no reason to capture (Bar b).
2226 We can let it float out. But if there is even one constraint we
2227 must be much more careful:
2228 forall {a}. C a b => Bar (m b)
2229 because (C a b) might have a superclass (D b), from which we might
2230 deduce (Bar [b]) when m later gets instantiated to []. Ha!
2232 Here is an even more exotic example
2234 Now consider the constraint
2235 forall b. D Int b => C Int
2236 We can satisfy the (C Int) from the superclass of D, so we don't want
2237 to float the (C Int) out, even though it mentions no type variable in
2240 Note [Pruning the givens in an implication constraint]
2241 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2242 Suppose we are about to form the implication constraint
2243 forall tvs. Eq a => Ord b
2244 The (Eq a) cannot contribute to the (Ord b), because it has no access to
2245 the type variable 'b'. So we could filter out the (Eq a) from the givens.
2247 Doing so would be a bit tidier, but all the implication constraints get
2248 simplified away by the optimiser, so it's no great win. So I don't take
2249 advantage of that at the moment.
2251 If you do, BE CAREFUL of wobbly type variables.
2254 %************************************************************************
2256 Avails and AvailHow: the pool of evidence
2258 %************************************************************************
2262 data Avails = Avails !ImprovementDone !AvailEnv
2264 type ImprovementDone = Bool -- True <=> some unification has happened
2265 -- so some Irreds might now be reducible
2266 -- keys that are now
2268 type AvailEnv = FiniteMap Inst AvailHow
2270 = IsIrred -- Used for irreducible dictionaries,
2271 -- which are going to be lambda bound
2273 | Given Inst -- Used for dictionaries for which we have a binding
2274 -- e.g. those "given" in a signature
2276 | Rhs -- Used when there is a RHS
2277 (LHsExpr TcId) -- The RHS
2278 [Inst] -- Insts free in the RHS; we need these too
2280 instance Outputable Avails where
2283 pprAvails (Avails imp avails)
2284 = vcat [ ptext (sLit "Avails") <> (if imp then ptext (sLit "[improved]") else empty)
2286 vcat [ sep [ppr inst, nest 2 (equals <+> ppr avail)]
2287 | (inst,avail) <- fmToList avails ]]
2289 instance Outputable AvailHow where
2292 -------------------------
2293 pprAvail :: AvailHow -> SDoc
2294 pprAvail IsIrred = text "Irred"
2295 pprAvail (Given x) = text "Given" <+> ppr x
2296 pprAvail (Rhs rhs bs) = sep [text "Rhs" <+> ppr bs,
2299 -------------------------
2300 extendAvailEnv :: AvailEnv -> Inst -> AvailHow -> AvailEnv
2301 extendAvailEnv env inst avail = addToFM env inst avail
2303 findAvailEnv :: AvailEnv -> Inst -> Maybe AvailHow
2304 findAvailEnv env wanted = lookupFM env wanted
2305 -- NB 1: the Ord instance of Inst compares by the class/type info
2306 -- *not* by unique. So
2307 -- d1::C Int == d2::C Int
2309 emptyAvails :: Avails
2310 emptyAvails = Avails False emptyFM
2312 findAvail :: Avails -> Inst -> Maybe AvailHow
2313 findAvail (Avails _ avails) wanted = findAvailEnv avails wanted
2315 elemAvails :: Inst -> Avails -> Bool
2316 elemAvails wanted (Avails _ avails) = wanted `elemFM` avails
2318 extendAvails :: Avails -> Inst -> AvailHow -> TcM Avails
2320 extendAvails avails@(Avails imp env) inst avail
2321 = do { imp1 <- tcImproveOne avails inst -- Do any improvement
2322 ; return (Avails (imp || imp1) (extendAvailEnv env inst avail)) }
2324 availsInsts :: Avails -> [Inst]
2325 availsInsts (Avails _ avails) = keysFM avails
2327 availsImproved (Avails imp _) = imp
2329 updateImprovement :: Avails -> Avails -> Avails
2330 -- (updateImprovement a1 a2) sets a1's improvement flag from a2
2331 updateImprovement (Avails _ avails1) (Avails imp2 _) = Avails imp2 avails1
2334 Extracting the bindings from a bunch of Avails.
2335 The bindings do *not* come back sorted in dependency order.
2336 We assume that they'll be wrapped in a big Rec, so that the
2337 dependency analyser can sort them out later
2340 type DoneEnv = FiniteMap Inst [Id]
2341 -- Tracks which things we have evidence for
2343 extractResults :: Avails
2345 -> TcM (TcDictBinds, -- Bindings
2346 [Inst], -- The insts bound by the bindings
2347 [Inst]) -- Irreducible ones
2348 -- Note [Reducing implication constraints]
2350 extractResults (Avails _ avails) wanteds
2351 = go emptyBag [] [] emptyFM wanteds
2353 go :: TcDictBinds -- Bindings for dicts
2354 -> [Inst] -- Bound by the bindings
2356 -> DoneEnv -- Has an entry for each inst in the above three sets
2358 -> TcM (TcDictBinds, [Inst], [Inst])
2359 go binds bound_dicts irreds done []
2360 = return (binds, bound_dicts, irreds)
2362 go binds bound_dicts irreds done (w:ws)
2363 | Just done_ids@(done_id : rest_done_ids) <- lookupFM done w
2364 = if w_id `elem` done_ids then
2365 go binds bound_dicts irreds done ws
2367 go (add_bind (nlHsVar done_id)) bound_dicts irreds
2368 (addToFM done w (done_id : w_id : rest_done_ids)) ws
2370 | otherwise -- Not yet done
2371 = case findAvailEnv avails w of
2372 Nothing -> pprTrace "Urk: extractResults" (ppr w) $
2373 go binds bound_dicts irreds done ws
2375 Just IsIrred -> go binds bound_dicts (w:irreds) done' ws
2377 Just (Rhs rhs ws') -> go (add_bind rhs) (w:bound_dicts) irreds done' (ws' ++ ws)
2379 Just (Given g) -> go binds' bound_dicts irreds (addToFM done w [g_id]) ws
2382 binds' | w_id == g_id = binds
2383 | otherwise = add_bind (nlHsVar g_id)
2386 done' = addToFM done w [w_id]
2387 add_bind rhs = addInstToDictBind binds w rhs
2391 Note [No superclasses for Stop]
2392 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2393 When we decide not to reduce an Inst -- the 'WhatToDo' --- we still
2394 add it to avails, so that any other equal Insts will be commoned up
2395 right here. However, we do *not* add superclasses. If we have
2398 but a is not bound here, then we *don't* want to derive dn from df
2399 here lest we lose sharing.
2402 addWanted :: WantSCs -> Avails -> Inst -> LHsExpr TcId -> [Inst] -> TcM Avails
2403 addWanted want_scs avails wanted rhs_expr wanteds
2404 = addAvailAndSCs want_scs avails wanted avail
2406 avail = Rhs rhs_expr wanteds
2408 addGiven :: Avails -> Inst -> TcM Avails
2409 addGiven avails given = addAvailAndSCs AddSCs avails given (Given given)
2410 -- Always add superclasses for 'givens'
2412 -- No ASSERT( not (given `elemAvails` avails) ) because in an instance
2413 -- decl for Ord t we can add both Ord t and Eq t as 'givens',
2414 -- so the assert isn't true
2418 addIrred :: WantSCs -> Avails -> Inst -> TcM Avails
2419 addIrred want_scs avails irred = ASSERT2( not (irred `elemAvails` avails), ppr irred $$ ppr avails )
2420 addAvailAndSCs want_scs avails irred IsIrred
2422 addAvailAndSCs :: WantSCs -> Avails -> Inst -> AvailHow -> TcM Avails
2423 addAvailAndSCs want_scs avails inst avail
2424 | not (isClassDict inst) = extendAvails avails inst avail
2425 | NoSCs <- want_scs = extendAvails avails inst avail
2426 | otherwise = do { traceTc (text "addAvailAndSCs" <+> vcat [ppr inst, ppr deps])
2427 ; avails' <- extendAvails avails inst avail
2428 ; addSCs is_loop avails' inst }
2430 is_loop pred = any (`tcEqType` mkPredTy pred) dep_tys
2431 -- Note: this compares by *type*, not by Unique
2432 deps = findAllDeps (unitVarSet (instToVar inst)) avail
2433 dep_tys = map idType (varSetElems deps)
2435 findAllDeps :: IdSet -> AvailHow -> IdSet
2436 -- Find all the Insts that this one depends on
2437 -- See Note [SUPERCLASS-LOOP 2]
2438 -- Watch out, though. Since the avails may contain loops
2439 -- (see Note [RECURSIVE DICTIONARIES]), so we need to track the ones we've seen so far
2440 findAllDeps so_far (Rhs _ kids) = foldl find_all so_far kids
2441 findAllDeps so_far other = so_far
2443 find_all :: IdSet -> Inst -> IdSet
2445 | isEqInst kid = so_far
2446 | kid_id `elemVarSet` so_far = so_far
2447 | Just avail <- findAvail avails kid = findAllDeps so_far' avail
2448 | otherwise = so_far'
2450 so_far' = extendVarSet so_far kid_id -- Add the new kid to so_far
2451 kid_id = instToId kid
2453 addSCs :: (TcPredType -> Bool) -> Avails -> Inst -> TcM Avails
2454 -- Add all the superclasses of the Inst to Avails
2455 -- The first param says "don't do this because the original thing
2456 -- depends on this one, so you'd build a loop"
2457 -- Invariant: the Inst is already in Avails.
2459 addSCs is_loop avails dict
2460 = ASSERT( isDict dict )
2461 do { sc_dicts <- newDictBndrs (instLoc dict) sc_theta'
2462 ; foldlM add_sc avails (zipEqual "add_scs" sc_dicts sc_sels) }
2464 (clas, tys) = getDictClassTys dict
2465 (tyvars, sc_theta, sc_sels, _) = classBigSig clas
2466 sc_theta' = filter (not . isEqPred) $
2467 substTheta (zipTopTvSubst tyvars tys) sc_theta
2469 add_sc avails (sc_dict, sc_sel)
2470 | is_loop (dictPred sc_dict) = return avails -- See Note [SUPERCLASS-LOOP 2]
2471 | is_given sc_dict = return avails
2472 | otherwise = do { avails' <- extendAvails avails sc_dict (Rhs sc_sel_rhs [dict])
2473 ; addSCs is_loop avails' sc_dict }
2475 sc_sel_rhs = L (instSpan dict) (HsWrap co_fn (HsVar sc_sel))
2476 co_fn = WpApp (instToVar dict) <.> mkWpTyApps tys
2478 is_given :: Inst -> Bool
2479 is_given sc_dict = case findAvail avails sc_dict of
2480 Just (Given _) -> True -- Given is cheaper than superclass selection
2483 -- From the a set of insts obtain all equalities that (transitively) occur in
2484 -- superclass contexts of class constraints (aka the ancestor equalities).
2486 ancestorEqualities :: [Inst] -> TcM [Inst]
2488 = mapM mkWantedEqInst -- turn only equality predicates..
2489 . filter isEqPred -- ..into wanted equality insts
2491 . addAEsToBag emptyBag -- collect the superclass constraints..
2492 . map dictPred -- ..of all predicates in a bag
2493 . filter isClassDict
2495 addAEsToBag :: Bag PredType -> [PredType] -> Bag PredType
2496 addAEsToBag bag [] = bag
2497 addAEsToBag bag (pred:preds)
2498 | pred `elemBag` bag = addAEsToBag bag preds
2499 | isEqPred pred = addAEsToBag bagWithPred preds
2500 | isClassPred pred = addAEsToBag bagWithPred predsWithSCs
2501 | otherwise = addAEsToBag bag preds
2503 bagWithPred = bag `snocBag` pred
2504 predsWithSCs = preds ++ substTheta (zipTopTvSubst tyvars tys) sc_theta
2506 (tyvars, sc_theta, _, _) = classBigSig clas
2507 (clas, tys) = getClassPredTys pred
2511 %************************************************************************
2513 \section{tcSimplifyTop: defaulting}
2515 %************************************************************************
2518 @tcSimplifyTop@ is called once per module to simplify all the constant
2519 and ambiguous Insts.
2521 We need to be careful of one case. Suppose we have
2523 instance Num a => Num (Foo a b) where ...
2525 and @tcSimplifyTop@ is given a constraint (Num (Foo x y)). Then it'll simplify
2526 to (Num x), and default x to Int. But what about y??
2528 It's OK: the final zonking stage should zap y to (), which is fine.
2532 tcSimplifyTop, tcSimplifyInteractive :: [Inst] -> TcM TcDictBinds
2533 tcSimplifyTop wanteds
2534 = tc_simplify_top doc False wanteds
2536 doc = text "tcSimplifyTop"
2538 tcSimplifyInteractive wanteds
2539 = tc_simplify_top doc True wanteds
2541 doc = text "tcSimplifyInteractive"
2543 -- The TcLclEnv should be valid here, solely to improve
2544 -- error message generation for the monomorphism restriction
2545 tc_simplify_top doc interactive wanteds
2546 = do { dflags <- getDOpts
2547 ; wanteds <- zonkInsts wanteds
2548 ; mapM_ zonkTopTyVar (varSetElems (tyVarsOfInsts wanteds))
2550 ; traceTc (text "tc_simplify_top 0: " <+> ppr wanteds)
2551 ; (irreds1, binds1) <- tryHardCheckLoop doc1 wanteds
2552 -- ; (irreds1, binds1) <- gentleInferLoop doc1 wanteds
2553 ; traceTc (text "tc_simplify_top 1: " <+> ppr irreds1)
2554 ; (irreds2, binds2) <- approximateImplications doc2 (\d -> True) irreds1
2555 ; traceTc (text "tc_simplify_top 2: " <+> ppr irreds2)
2557 -- Use the defaulting rules to do extra unification
2558 -- NB: irreds2 are already zonked
2559 ; (irreds3, binds3) <- disambiguate doc3 interactive dflags irreds2
2561 -- Deal with implicit parameters
2562 ; let (bad_ips, non_ips) = partition isIPDict irreds3
2563 (ambigs, others) = partition isTyVarDict non_ips
2565 ; topIPErrs bad_ips -- Can arise from f :: Int -> Int
2567 ; addNoInstanceErrs others
2568 ; addTopAmbigErrs ambigs
2570 ; return (binds1 `unionBags` binds2 `unionBags` binds3) }
2572 doc1 = doc <+> ptext (sLit "(first round)")
2573 doc2 = doc <+> ptext (sLit "(approximate)")
2574 doc3 = doc <+> ptext (sLit "(disambiguate)")
2577 If a dictionary constrains a type variable which is
2578 * not mentioned in the environment
2579 * and not mentioned in the type of the expression
2580 then it is ambiguous. No further information will arise to instantiate
2581 the type variable; nor will it be generalised and turned into an extra
2582 parameter to a function.
2584 It is an error for this to occur, except that Haskell provided for
2585 certain rules to be applied in the special case of numeric types.
2587 * at least one of its classes is a numeric class, and
2588 * all of its classes are numeric or standard
2589 then the type variable can be defaulted to the first type in the
2590 default-type list which is an instance of all the offending classes.
2592 So here is the function which does the work. It takes the ambiguous
2593 dictionaries and either resolves them (producing bindings) or
2594 complains. It works by splitting the dictionary list by type
2595 variable, and using @disambigOne@ to do the real business.
2597 @disambigOne@ assumes that its arguments dictionaries constrain all
2598 the same type variable.
2600 ADR Comment 20/6/94: I've changed the @CReturnable@ case to default to
2601 @()@ instead of @Int@. I reckon this is the Right Thing to do since
2602 the most common use of defaulting is code like:
2604 _ccall_ foo `seqPrimIO` bar
2606 Since we're not using the result of @foo@, the result if (presumably)
2610 disambiguate :: SDoc -> Bool -> DynFlags -> [Inst] -> TcM ([Inst], TcDictBinds)
2611 -- Just does unification to fix the default types
2612 -- The Insts are assumed to be pre-zonked
2613 disambiguate doc interactive dflags insts
2615 = return (insts, emptyBag)
2617 | null defaultable_groups
2618 = do { traceTc (text "disambigutate, no defaultable groups" <+> vcat [ppr unaries, ppr insts, ppr bad_tvs, ppr defaultable_groups])
2619 ; return (insts, emptyBag) }
2622 = do { -- Figure out what default types to use
2623 default_tys <- getDefaultTys extended_defaulting ovl_strings
2625 ; traceTc (text "disambiguate1" <+> vcat [ppr insts, ppr unaries, ppr bad_tvs, ppr defaultable_groups])
2626 ; mapM_ (disambigGroup default_tys) defaultable_groups
2628 -- disambigGroup does unification, hence try again
2629 ; tryHardCheckLoop doc insts }
2632 extended_defaulting = interactive || dopt Opt_ExtendedDefaultRules dflags
2633 ovl_strings = dopt Opt_OverloadedStrings dflags
2635 unaries :: [(Inst, Class, TcTyVar)] -- (C tv) constraints
2636 bad_tvs :: TcTyVarSet -- Tyvars mentioned by *other* constraints
2637 (unaries, bad_tvs_s) = partitionWith find_unary insts
2638 bad_tvs = unionVarSets bad_tvs_s
2640 -- Finds unary type-class constraints
2641 find_unary d@(Dict {tci_pred = ClassP cls [ty]})
2642 | Just tv <- tcGetTyVar_maybe ty = Left (d,cls,tv)
2643 find_unary inst = Right (tyVarsOfInst inst)
2645 -- Group by type variable
2646 defaultable_groups :: [[(Inst,Class,TcTyVar)]]
2647 defaultable_groups = filter defaultable_group (equivClasses cmp_tv unaries)
2648 cmp_tv (_,_,tv1) (_,_,tv2) = tv1 `compare` tv2
2650 defaultable_group :: [(Inst,Class,TcTyVar)] -> Bool
2651 defaultable_group ds@((_,_,tv):_)
2652 = isTyConableTyVar tv -- Note [Avoiding spurious errors]
2653 && not (tv `elemVarSet` bad_tvs)
2654 && defaultable_classes [c | (_,c,_) <- ds]
2655 defaultable_group [] = panic "defaultable_group"
2657 defaultable_classes clss
2658 | extended_defaulting = any isInteractiveClass clss
2659 | otherwise = all is_std_class clss && (any is_num_class clss)
2661 -- In interactive mode, or with -fextended-default-rules,
2662 -- we default Show a to Show () to avoid graututious errors on "show []"
2663 isInteractiveClass cls
2664 = is_num_class cls || (classKey cls `elem` [showClassKey, eqClassKey, ordClassKey])
2666 is_num_class cls = isNumericClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2667 -- is_num_class adds IsString to the standard numeric classes,
2668 -- when -foverloaded-strings is enabled
2670 is_std_class cls = isStandardClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2671 -- Similarly is_std_class
2673 -----------------------
2674 disambigGroup :: [Type] -- The default types
2675 -> [(Inst,Class,TcTyVar)] -- All standard classes of form (C a)
2676 -> TcM () -- Just does unification, to fix the default types
2678 disambigGroup default_tys dicts
2679 = try_default default_tys
2681 (_,_,tyvar) = ASSERT(not (null dicts)) head dicts -- Should be non-empty
2682 classes = [c | (_,c,_) <- dicts]
2684 try_default [] = return ()
2685 try_default (default_ty : default_tys)
2686 = tryTcLIE_ (try_default default_tys) $
2687 do { tcSimplifyDefault [mkClassPred clas [default_ty] | clas <- classes]
2688 -- This may fail; then the tryTcLIE_ kicks in
2689 -- Failure here is caused by there being no type in the
2690 -- default list which can satisfy all the ambiguous classes.
2691 -- For example, if Real a is reqd, but the only type in the
2692 -- default list is Int.
2694 -- After this we can't fail
2695 ; warnDefault dicts default_ty
2696 ; unifyType default_ty (mkTyVarTy tyvar)
2697 ; return () -- TOMDO: do something with the coercion
2701 -----------------------
2702 getDefaultTys :: Bool -> Bool -> TcM [Type]
2703 getDefaultTys extended_deflts ovl_strings
2704 = do { mb_defaults <- getDeclaredDefaultTys
2705 ; case mb_defaults of {
2706 Just tys -> return tys ; -- User-supplied defaults
2709 -- No use-supplied default
2710 -- Use [Integer, Double], plus modifications
2711 { integer_ty <- tcMetaTy integerTyConName
2712 ; checkWiredInTyCon doubleTyCon
2713 ; string_ty <- tcMetaTy stringTyConName
2714 ; return (opt_deflt extended_deflts unitTy
2715 -- Note [Default unitTy]
2717 [integer_ty,doubleTy]
2719 opt_deflt ovl_strings string_ty) } } }
2721 opt_deflt True ty = [ty]
2722 opt_deflt False ty = []
2725 Note [Default unitTy]
2726 ~~~~~~~~~~~~~~~~~~~~~
2727 In interative mode (or with -fextended-default-rules) we add () as the first type we
2728 try when defaulting. This has very little real impact, except in the following case.
2730 Text.Printf.printf "hello"
2731 This has type (forall a. IO a); it prints "hello", and returns 'undefined'. We don't
2732 want the GHCi repl loop to try to print that 'undefined'. The neatest thing is to
2733 default the 'a' to (), rather than to Integer (which is what would otherwise happen;
2734 and then GHCi doesn't attempt to print the (). So in interactive mode, we add
2735 () to the list of defaulting types. See Trac #1200.
2737 Note [Avoiding spurious errors]
2738 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2739 When doing the unification for defaulting, we check for skolem
2740 type variables, and simply don't default them. For example:
2741 f = (*) -- Monomorphic
2742 g :: Num a => a -> a
2744 Here, we get a complaint when checking the type signature for g,
2745 that g isn't polymorphic enough; but then we get another one when
2746 dealing with the (Num a) context arising from f's definition;
2747 we try to unify a with Int (to default it), but find that it's
2748 already been unified with the rigid variable from g's type sig
2751 %************************************************************************
2753 \subsection[simple]{@Simple@ versions}
2755 %************************************************************************
2757 Much simpler versions when there are no bindings to make!
2759 @tcSimplifyThetas@ simplifies class-type constraints formed by
2760 @deriving@ declarations and when specialising instances. We are
2761 only interested in the simplified bunch of class/type constraints.
2763 It simplifies to constraints of the form (C a b c) where
2764 a,b,c are type variables. This is required for the context of
2765 instance declarations.
2768 tcSimplifyDeriv :: InstOrigin
2770 -> ThetaType -- Wanted
2771 -> TcM ThetaType -- Needed
2772 -- Given instance (wanted) => C inst_ty
2773 -- Simplify 'wanted' as much as possible
2775 tcSimplifyDeriv orig tyvars theta
2776 = do { (tvs, _, tenv) <- tcInstTyVars tyvars
2777 -- The main loop may do unification, and that may crash if
2778 -- it doesn't see a TcTyVar, so we have to instantiate. Sigh
2779 -- ToDo: what if two of them do get unified?
2780 ; wanteds <- newDictBndrsO orig (substTheta tenv theta)
2781 ; (irreds, _) <- tryHardCheckLoop doc wanteds
2783 ; let (tv_dicts, others) = partition ok irreds
2784 ; addNoInstanceErrs others
2785 -- See Note [Exotic derived instance contexts] in TcMType
2787 ; let rev_env = zipTopTvSubst tvs (mkTyVarTys tyvars)
2788 simpl_theta = substTheta rev_env (map dictPred tv_dicts)
2789 -- This reverse-mapping is a pain, but the result
2790 -- should mention the original TyVars not TcTyVars
2792 ; return simpl_theta }
2794 doc = ptext (sLit "deriving classes for a data type")
2796 ok dict | isDict dict = validDerivPred (dictPred dict)
2801 @tcSimplifyDefault@ just checks class-type constraints, essentially;
2802 used with \tr{default} declarations. We are only interested in
2803 whether it worked or not.
2806 tcSimplifyDefault :: ThetaType -- Wanted; has no type variables in it
2809 tcSimplifyDefault theta = do
2810 wanteds <- newDictBndrsO DefaultOrigin theta
2811 (irreds, _) <- tryHardCheckLoop doc wanteds
2812 addNoInstanceErrs irreds
2816 traceTc (ptext (sLit "tcSimplifyDefault failing")) >> failM
2818 doc = ptext (sLit "default declaration")
2822 %************************************************************************
2824 \section{Errors and contexts}
2826 %************************************************************************
2828 ToDo: for these error messages, should we note the location as coming
2829 from the insts, or just whatever seems to be around in the monad just
2833 groupErrs :: ([Inst] -> TcM ()) -- Deal with one group
2834 -> [Inst] -- The offending Insts
2836 -- Group together insts with the same origin
2837 -- We want to report them together in error messages
2839 groupErrs report_err []
2841 groupErrs report_err (inst:insts)
2842 = do { do_one (inst:friends)
2843 ; groupErrs report_err others }
2845 -- (It may seem a bit crude to compare the error messages,
2846 -- but it makes sure that we combine just what the user sees,
2847 -- and it avoids need equality on InstLocs.)
2848 (friends, others) = partition is_friend insts
2849 loc_msg = showSDoc (pprInstLoc (instLoc inst))
2850 is_friend friend = showSDoc (pprInstLoc (instLoc friend)) == loc_msg
2851 do_one insts = addInstCtxt (instLoc (head insts)) (report_err insts)
2852 -- Add location and context information derived from the Insts
2854 -- Add the "arising from..." part to a message about bunch of dicts
2855 addInstLoc :: [Inst] -> Message -> Message
2856 addInstLoc insts msg = msg $$ nest 2 (pprInstArising (head insts))
2858 addTopIPErrs :: [Name] -> [Inst] -> TcM ()
2859 addTopIPErrs bndrs []
2861 addTopIPErrs bndrs ips
2862 = do { dflags <- getDOpts
2863 ; addErrTcM (tidy_env, mk_msg dflags tidy_ips) }
2865 (tidy_env, tidy_ips) = tidyInsts ips
2867 = vcat [sep [ptext (sLit "Implicit parameters escape from"),
2868 nest 2 (ptext (sLit "the monomorphic top-level binding")
2869 <> plural bndrs <+> ptext (sLit "of")
2870 <+> pprBinders bndrs <> colon)],
2871 nest 2 (vcat (map ppr_ip ips)),
2872 monomorphism_fix dflags]
2873 ppr_ip ip = pprPred (dictPred ip) <+> pprInstArising ip
2875 topIPErrs :: [Inst] -> TcM ()
2877 = groupErrs report tidy_dicts
2879 (tidy_env, tidy_dicts) = tidyInsts dicts
2880 report dicts = addErrTcM (tidy_env, mk_msg dicts)
2881 mk_msg dicts = addInstLoc dicts (ptext (sLit "Unbound implicit parameter") <>
2882 plural tidy_dicts <+> pprDictsTheta tidy_dicts)
2884 addNoInstanceErrs :: [Inst] -- Wanted (can include implications)
2886 addNoInstanceErrs insts
2887 = do { let (tidy_env, tidy_insts) = tidyInsts insts
2888 ; reportNoInstances tidy_env Nothing tidy_insts }
2892 -> Maybe (InstLoc, [Inst]) -- Context
2893 -- Nothing => top level
2894 -- Just (d,g) => d describes the construct
2896 -> [Inst] -- What is wanted (can include implications)
2899 reportNoInstances tidy_env mb_what insts
2900 = groupErrs (report_no_instances tidy_env mb_what) insts
2902 report_no_instances tidy_env mb_what insts
2903 = do { inst_envs <- tcGetInstEnvs
2904 ; let (implics, insts1) = partition isImplicInst insts
2905 (insts2, overlaps) = partitionWith (check_overlap inst_envs) insts1
2906 (eqInsts, insts3) = partition isEqInst insts2
2907 ; traceTc (text "reportNoInstances" <+> vcat
2908 [ppr insts, ppr implics, ppr insts1, ppr insts2])
2909 ; mapM_ complain_implic implics
2910 ; mapM_ (\doc -> addErrTcM (tidy_env, doc)) overlaps
2911 ; groupErrs complain_no_inst insts3
2912 ; mapM_ (addErrTcM . mk_eq_err) eqInsts
2915 complain_no_inst insts = addErrTcM (tidy_env, mk_no_inst_err insts)
2917 complain_implic inst -- Recurse!
2918 = reportNoInstances tidy_env
2919 (Just (tci_loc inst, tci_given inst))
2922 check_overlap :: (InstEnv,InstEnv) -> Inst -> Either Inst SDoc
2923 -- Right msg => overlap message
2924 -- Left inst => no instance
2925 check_overlap inst_envs wanted
2926 | not (isClassDict wanted) = Left wanted
2928 = case lookupInstEnv inst_envs clas tys of
2929 ([], _) -> Left wanted -- No match
2930 -- The case of exactly one match and no unifiers means a
2931 -- successful lookup. That can't happen here, because dicts
2932 -- only end up here if they didn't match in Inst.lookupInst
2934 | debugIsOn -> pprPanic "reportNoInstance" (ppr wanted)
2935 res -> Right (mk_overlap_msg wanted res)
2937 (clas,tys) = getDictClassTys wanted
2939 mk_overlap_msg dict (matches, unifiers)
2940 = ASSERT( not (null matches) )
2941 vcat [ addInstLoc [dict] ((ptext (sLit "Overlapping instances for")
2942 <+> pprPred (dictPred dict))),
2943 sep [ptext (sLit "Matching instances") <> colon,
2944 nest 2 (vcat [pprInstances ispecs, pprInstances unifiers])],
2945 if not (isSingleton matches)
2946 then -- Two or more matches
2948 else -- One match, plus some unifiers
2949 ASSERT( not (null unifiers) )
2950 parens (vcat [ptext (sLit "The choice depends on the instantiation of") <+>
2951 quotes (pprWithCommas ppr (varSetElems (tyVarsOfInst dict))),
2952 ptext (sLit "To pick the first instance above, use -fallow-incoherent-instances"),
2953 ptext (sLit "when compiling the other instance declarations")])]
2955 ispecs = [ispec | (ispec, _) <- matches]
2957 mk_eq_err :: Inst -> (TidyEnv, SDoc)
2958 mk_eq_err inst = misMatchMsg tidy_env (eqInstTys inst)
2960 mk_no_inst_err insts
2961 | null insts = empty
2963 | Just (loc, givens) <- mb_what, -- Nested (type signatures, instance decls)
2964 not (isEmptyVarSet (tyVarsOfInsts insts))
2965 = vcat [ addInstLoc insts $
2966 sep [ ptext (sLit "Could not deduce") <+> pprDictsTheta insts
2967 , nest 2 $ ptext (sLit "from the context") <+> pprDictsTheta givens]
2968 , show_fixes (fix1 loc : fixes2) ]
2970 | otherwise -- Top level
2971 = vcat [ addInstLoc insts $
2972 ptext (sLit "No instance") <> plural insts
2973 <+> ptext (sLit "for") <+> pprDictsTheta insts
2974 , show_fixes fixes2 ]
2977 fix1 loc = sep [ ptext (sLit "add") <+> pprDictsTheta insts
2978 <+> ptext (sLit "to the context of"),
2979 nest 2 (ppr (instLocOrigin loc)) ]
2980 -- I'm not sure it helps to add the location
2981 -- nest 2 (ptext (sLit "at") <+> ppr (instLocSpan loc)) ]
2983 fixes2 | null instance_dicts = []
2984 | otherwise = [sep [ptext (sLit "add an instance declaration for"),
2985 pprDictsTheta instance_dicts]]
2986 instance_dicts = [d | d <- insts, isClassDict d, not (isTyVarDict d)]
2987 -- Insts for which it is worth suggesting an adding an instance declaration
2988 -- Exclude implicit parameters, and tyvar dicts
2990 show_fixes :: [SDoc] -> SDoc
2991 show_fixes [] = empty
2992 show_fixes (f:fs) = sep [ptext (sLit "Possible fix:"),
2993 nest 2 (vcat (f : map (ptext (sLit "or") <+>) fs))]
2995 addTopAmbigErrs dicts
2996 -- Divide into groups that share a common set of ambiguous tyvars
2997 = ifErrsM (return ()) $ -- Only report ambiguity if no other errors happened
2998 -- See Note [Avoiding spurious errors]
2999 mapM_ report (equivClasses cmp [(d, tvs_of d) | d <- tidy_dicts])
3001 (tidy_env, tidy_dicts) = tidyInsts dicts
3003 tvs_of :: Inst -> [TcTyVar]
3004 tvs_of d = varSetElems (tyVarsOfInst d)
3005 cmp (_,tvs1) (_,tvs2) = tvs1 `compare` tvs2
3007 report :: [(Inst,[TcTyVar])] -> TcM ()
3008 report pairs@((inst,tvs) : _) = do -- The pairs share a common set of ambiguous tyvars
3009 (tidy_env, mono_msg) <- mkMonomorphismMsg tidy_env tvs
3010 setSrcSpan (instSpan inst) $
3011 -- the location of the first one will do for the err message
3012 addErrTcM (tidy_env, msg $$ mono_msg)
3014 dicts = map fst pairs
3015 msg = sep [text "Ambiguous type variable" <> plural tvs <+>
3016 pprQuotedList tvs <+> in_msg,
3017 nest 2 (pprDictsInFull dicts)]
3018 in_msg = text "in the constraint" <> plural dicts <> colon
3019 report [] = panic "addTopAmbigErrs"
3022 mkMonomorphismMsg :: TidyEnv -> [TcTyVar] -> TcM (TidyEnv, Message)
3023 -- There's an error with these Insts; if they have free type variables
3024 -- it's probably caused by the monomorphism restriction.
3025 -- Try to identify the offending variable
3026 -- ASSUMPTION: the Insts are fully zonked
3027 mkMonomorphismMsg tidy_env inst_tvs
3028 = do { dflags <- getDOpts
3029 ; (tidy_env, docs) <- findGlobals (mkVarSet inst_tvs) tidy_env
3030 ; return (tidy_env, mk_msg dflags docs) }
3032 mk_msg _ _ | any isRuntimeUnk inst_tvs
3033 = vcat [ptext (sLit "Cannot resolve unknown runtime types:") <+>
3034 (pprWithCommas ppr inst_tvs),
3035 ptext (sLit "Use :print or :force to determine these types")]
3036 mk_msg _ [] = ptext (sLit "Probable fix: add a type signature that fixes these type variable(s)")
3037 -- This happens in things like
3038 -- f x = show (read "foo")
3039 -- where monomorphism doesn't play any role
3041 = vcat [ptext (sLit "Possible cause: the monomorphism restriction applied to the following:"),
3043 monomorphism_fix dflags]
3045 monomorphism_fix :: DynFlags -> SDoc
3046 monomorphism_fix dflags
3047 = ptext (sLit "Probable fix:") <+> vcat
3048 [ptext (sLit "give these definition(s) an explicit type signature"),
3049 if dopt Opt_MonomorphismRestriction dflags
3050 then ptext (sLit "or use -fno-monomorphism-restriction")
3051 else empty] -- Only suggest adding "-fno-monomorphism-restriction"
3052 -- if it is not already set!
3054 warnDefault ups default_ty = do
3055 warn_flag <- doptM Opt_WarnTypeDefaults
3056 addInstCtxt (instLoc (head (dicts))) (warnTc warn_flag warn_msg)
3058 dicts = [d | (d,_,_) <- ups]
3061 (_, tidy_dicts) = tidyInsts dicts
3062 warn_msg = vcat [ptext (sLit "Defaulting the following constraint(s) to type") <+>
3063 quotes (ppr default_ty),
3064 pprDictsInFull tidy_dicts]
3066 reduceDepthErr n stack
3067 = vcat [ptext (sLit "Context reduction stack overflow; size =") <+> int n,
3068 ptext (sLit "Use -fcontext-stack=N to increase stack size to N"),
3069 nest 4 (pprStack stack)]
3071 pprStack stack = vcat (map pprInstInFull stack)