2 % (c) The University of Glasgow 2006
3 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
10 tcSimplifyInfer, tcSimplifyInferCheck,
11 tcSimplifyCheck, tcSimplifyRestricted,
12 tcSimplifyRuleLhs, tcSimplifyIPs,
13 tcSimplifySuperClasses,
14 tcSimplifyTop, tcSimplifyInteractive,
15 tcSimplifyBracket, tcSimplifyCheckPat,
17 tcSimplifyDeriv, tcSimplifyDefault,
18 bindInstsOfLocalFuns, bindIrreds,
21 #include "HsVersions.h"
23 import {-# SOURCE #-} TcUnify( unifyType )
59 %************************************************************************
63 %************************************************************************
65 --------------------------------------
66 Notes on functional dependencies (a bug)
67 --------------------------------------
74 instance D a b => C a b -- Undecidable
75 -- (Not sure if it's crucial to this eg)
76 f :: C a b => a -> Bool
79 g :: C a b => a -> Bool
82 Here f typechecks, but g does not!! Reason: before doing improvement,
83 we reduce the (C a b1) constraint from the call of f to (D a b1).
85 Here is a more complicated example:
87 | > class Foo a b | a->b
89 | > class Bar a b | a->b
93 | > instance Bar Obj Obj
95 | > instance (Bar a b) => Foo a b
97 | > foo:: (Foo a b) => a -> String
100 | > runFoo:: (forall a b. (Foo a b) => a -> w) -> w
106 | Could not deduce (Bar a b) from the context (Foo a b)
107 | arising from use of `foo' at <interactive>:1
109 | Add (Bar a b) to the expected type of an expression
110 | In the first argument of `runFoo', namely `foo'
111 | In the definition of `it': it = runFoo foo
113 | Why all of the sudden does GHC need the constraint Bar a b? The
114 | function foo didn't ask for that...
116 The trouble is that to type (runFoo foo), GHC has to solve the problem:
118 Given constraint Foo a b
119 Solve constraint Foo a b'
121 Notice that b and b' aren't the same. To solve this, just do
122 improvement and then they are the same. But GHC currently does
127 That is usually fine, but it isn't here, because it sees that Foo a b is
128 not the same as Foo a b', and so instead applies the instance decl for
129 instance Bar a b => Foo a b. And that's where the Bar constraint comes
132 The Right Thing is to improve whenever the constraint set changes at
133 all. Not hard in principle, but it'll take a bit of fiddling to do.
137 --------------------------------------
138 Notes on quantification
139 --------------------------------------
141 Suppose we are about to do a generalisation step.
145 T the type of the RHS
146 C the constraints from that RHS
148 The game is to figure out
150 Q the set of type variables over which to quantify
151 Ct the constraints we will *not* quantify over
152 Cq the constraints we will quantify over
154 So we're going to infer the type
158 and float the constraints Ct further outwards.
160 Here are the things that *must* be true:
162 (A) Q intersect fv(G) = EMPTY limits how big Q can be
163 (B) Q superset fv(Cq union T) \ oclose(fv(G),C) limits how small Q can be
165 (A) says we can't quantify over a variable that's free in the
166 environment. (B) says we must quantify over all the truly free
167 variables in T, else we won't get a sufficiently general type. We do
168 not *need* to quantify over any variable that is fixed by the free
169 vars of the environment G.
171 BETWEEN THESE TWO BOUNDS, ANY Q WILL DO!
173 Example: class H x y | x->y where ...
175 fv(G) = {a} C = {H a b, H c d}
178 (A) Q intersect {a} is empty
179 (B) Q superset {a,b,c,d} \ oclose({a}, C) = {a,b,c,d} \ {a,b} = {c,d}
181 So Q can be {c,d}, {b,c,d}
183 Other things being equal, however, we'd like to quantify over as few
184 variables as possible: smaller types, fewer type applications, more
185 constraints can get into Ct instead of Cq.
188 -----------------------------------------
191 fv(T) the free type vars of T
193 oclose(vs,C) The result of extending the set of tyvars vs
194 using the functional dependencies from C
196 grow(vs,C) The result of extend the set of tyvars vs
197 using all conceivable links from C.
199 E.g. vs = {a}, C = {H [a] b, K (b,Int) c, Eq e}
200 Then grow(vs,C) = {a,b,c}
202 Note that grow(vs,C) `superset` grow(vs,simplify(C))
203 That is, simplfication can only shrink the result of grow.
206 oclose is conservative one way: v `elem` oclose(vs,C) => v is definitely fixed by vs
207 grow is conservative the other way: if v might be fixed by vs => v `elem` grow(vs,C)
210 -----------------------------------------
214 Here's a good way to choose Q:
216 Q = grow( fv(T), C ) \ oclose( fv(G), C )
218 That is, quantify over all variable that that MIGHT be fixed by the
219 call site (which influences T), but which aren't DEFINITELY fixed by
220 G. This choice definitely quantifies over enough type variables,
221 albeit perhaps too many.
223 Why grow( fv(T), C ) rather than fv(T)? Consider
225 class H x y | x->y where ...
230 If we used fv(T) = {c} we'd get the type
232 forall c. H c d => c -> b
234 And then if the fn was called at several different c's, each of
235 which fixed d differently, we'd get a unification error, because
236 d isn't quantified. Solution: quantify d. So we must quantify
237 everything that might be influenced by c.
239 Why not oclose( fv(T), C )? Because we might not be able to see
240 all the functional dependencies yet:
242 class H x y | x->y where ...
243 instance H x y => Eq (T x y) where ...
248 Now oclose(fv(T),C) = {c}, because the functional dependency isn't
249 apparent yet, and that's wrong. We must really quantify over d too.
252 There really isn't any point in quantifying over any more than
253 grow( fv(T), C ), because the call sites can't possibly influence
254 any other type variables.
258 -------------------------------------
260 -------------------------------------
262 It's very hard to be certain when a type is ambiguous. Consider
266 instance H x y => K (x,y)
268 Is this type ambiguous?
269 forall a b. (K (a,b), Eq b) => a -> a
271 Looks like it! But if we simplify (K (a,b)) we get (H a b) and
272 now we see that a fixes b. So we can't tell about ambiguity for sure
273 without doing a full simplification. And even that isn't possible if
274 the context has some free vars that may get unified. Urgle!
276 Here's another example: is this ambiguous?
277 forall a b. Eq (T b) => a -> a
278 Not if there's an insance decl (with no context)
279 instance Eq (T b) where ...
281 You may say of this example that we should use the instance decl right
282 away, but you can't always do that:
284 class J a b where ...
285 instance J Int b where ...
287 f :: forall a b. J a b => a -> a
289 (Notice: no functional dependency in J's class decl.)
290 Here f's type is perfectly fine, provided f is only called at Int.
291 It's premature to complain when meeting f's signature, or even
292 when inferring a type for f.
296 However, we don't *need* to report ambiguity right away. It'll always
297 show up at the call site.... and eventually at main, which needs special
298 treatment. Nevertheless, reporting ambiguity promptly is an excellent thing.
300 So here's the plan. We WARN about probable ambiguity if
302 fv(Cq) is not a subset of oclose(fv(T) union fv(G), C)
304 (all tested before quantification).
305 That is, all the type variables in Cq must be fixed by the the variables
306 in the environment, or by the variables in the type.
308 Notice that we union before calling oclose. Here's an example:
310 class J a b c | a b -> c
314 forall b c. (J a b c) => b -> b
316 Only if we union {a} from G with {b} from T before using oclose,
317 do we see that c is fixed.
319 It's a bit vague exactly which C we should use for this oclose call. If we
320 don't fix enough variables we might complain when we shouldn't (see
321 the above nasty example). Nothing will be perfect. That's why we can
322 only issue a warning.
325 Can we ever be *certain* about ambiguity? Yes: if there's a constraint
327 c in C such that fv(c) intersect (fv(G) union fv(T)) = EMPTY
329 then c is a "bubble"; there's no way it can ever improve, and it's
330 certainly ambiguous. UNLESS it is a constant (sigh). And what about
335 instance H x y => K (x,y)
337 Is this type ambiguous?
338 forall a b. (K (a,b), Eq b) => a -> a
340 Urk. The (Eq b) looks "definitely ambiguous" but it isn't. What we are after
341 is a "bubble" that's a set of constraints
343 Cq = Ca union Cq' st fv(Ca) intersect (fv(Cq') union fv(T) union fv(G)) = EMPTY
345 Hence another idea. To decide Q start with fv(T) and grow it
346 by transitive closure in Cq (no functional dependencies involved).
347 Now partition Cq using Q, leaving the definitely-ambiguous and probably-ok.
348 The definitely-ambiguous can then float out, and get smashed at top level
349 (which squashes out the constants, like Eq (T a) above)
352 --------------------------------------
353 Notes on principal types
354 --------------------------------------
359 f x = let g y = op (y::Int) in True
361 Here the principal type of f is (forall a. a->a)
362 but we'll produce the non-principal type
363 f :: forall a. C Int => a -> a
366 --------------------------------------
367 The need for forall's in constraints
368 --------------------------------------
370 [Exchange on Haskell Cafe 5/6 Dec 2000]
372 class C t where op :: t -> Bool
373 instance C [t] where op x = True
375 p y = (let f :: c -> Bool; f x = op (y >> return x) in f, y ++ [])
376 q y = (y ++ [], let f :: c -> Bool; f x = op (y >> return x) in f)
378 The definitions of p and q differ only in the order of the components in
379 the pair on their right-hand sides. And yet:
381 ghc and "Typing Haskell in Haskell" reject p, but accept q;
382 Hugs rejects q, but accepts p;
383 hbc rejects both p and q;
384 nhc98 ... (Malcolm, can you fill in the blank for us!).
386 The type signature for f forces context reduction to take place, and
387 the results of this depend on whether or not the type of y is known,
388 which in turn depends on which component of the pair the type checker
391 Solution: if y::m a, float out the constraints
392 Monad m, forall c. C (m c)
393 When m is later unified with [], we can solve both constraints.
396 --------------------------------------
397 Notes on implicit parameters
398 --------------------------------------
400 Note [Inheriting implicit parameters]
401 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
406 where f is *not* a top-level binding.
407 From the RHS of f we'll get the constraint (?y::Int).
408 There are two types we might infer for f:
412 (so we get ?y from the context of f's definition), or
414 f :: (?y::Int) => Int -> Int
416 At first you might think the first was better, becuase then
417 ?y behaves like a free variable of the definition, rather than
418 having to be passed at each call site. But of course, the WHOLE
419 IDEA is that ?y should be passed at each call site (that's what
420 dynamic binding means) so we'd better infer the second.
422 BOTTOM LINE: when *inferring types* you *must* quantify
423 over implicit parameters. See the predicate isFreeWhenInferring.
426 Note [Implicit parameters and ambiguity]
427 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
428 What type should we infer for this?
429 f x = (show ?y, x::Int)
430 Since we must quantify over the ?y, the most plausible type is
431 f :: (Show a, ?y::a) => Int -> (String, Int)
432 But notice that the type of the RHS is (String,Int), with no type
433 varibables mentioned at all! The type of f looks ambiguous. But
434 it isn't, because at a call site we might have
435 let ?y = 5::Int in f 7
436 and all is well. In effect, implicit parameters are, well, parameters,
437 so we can take their type variables into account as part of the
438 "tau-tvs" stuff. This is done in the function 'FunDeps.grow'.
441 Question 2: type signatures
442 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
443 BUT WATCH OUT: When you supply a type signature, we can't force you
444 to quantify over implicit parameters. For example:
448 This is perfectly reasonable. We do not want to insist on
450 (?x + 1) :: (?x::Int => Int)
452 That would be silly. Here, the definition site *is* the occurrence site,
453 so the above strictures don't apply. Hence the difference between
454 tcSimplifyCheck (which *does* allow implicit paramters to be inherited)
455 and tcSimplifyCheckBind (which does not).
457 What about when you supply a type signature for a binding?
458 Is it legal to give the following explicit, user type
459 signature to f, thus:
464 At first sight this seems reasonable, but it has the nasty property
465 that adding a type signature changes the dynamic semantics.
468 (let f x = (x::Int) + ?y
469 in (f 3, f 3 with ?y=5)) with ?y = 6
475 in (f 3, f 3 with ?y=5)) with ?y = 6
479 Indeed, simply inlining f (at the Haskell source level) would change the
482 Nevertheless, as Launchbury says (email Oct 01) we can't really give the
483 semantics for a Haskell program without knowing its typing, so if you
484 change the typing you may change the semantics.
486 To make things consistent in all cases where we are *checking* against
487 a supplied signature (as opposed to inferring a type), we adopt the
490 a signature does not need to quantify over implicit params.
492 [This represents a (rather marginal) change of policy since GHC 5.02,
493 which *required* an explicit signature to quantify over all implicit
494 params for the reasons mentioned above.]
496 But that raises a new question. Consider
498 Given (signature) ?x::Int
499 Wanted (inferred) ?x::Int, ?y::Bool
501 Clearly we want to discharge the ?x and float the ?y out. But
502 what is the criterion that distinguishes them? Clearly it isn't
503 what free type variables they have. The Right Thing seems to be
504 to float a constraint that
505 neither mentions any of the quantified type variables
506 nor any of the quantified implicit parameters
508 See the predicate isFreeWhenChecking.
511 Question 3: monomorphism
512 ~~~~~~~~~~~~~~~~~~~~~~~~
513 There's a nasty corner case when the monomorphism restriction bites:
517 The argument above suggests that we *must* generalise
518 over the ?y parameter, to get
519 z :: (?y::Int) => Int,
520 but the monomorphism restriction says that we *must not*, giving
522 Why does the momomorphism restriction say this? Because if you have
524 let z = x + ?y in z+z
526 you might not expect the addition to be done twice --- but it will if
527 we follow the argument of Question 2 and generalise over ?y.
530 Question 4: top level
531 ~~~~~~~~~~~~~~~~~~~~~
532 At the top level, monomorhism makes no sense at all.
535 main = let ?x = 5 in print foo
539 woggle :: (?x :: Int) => Int -> Int
542 We definitely don't want (foo :: Int) with a top-level implicit parameter
543 (?x::Int) becuase there is no way to bind it.
548 (A) Always generalise over implicit parameters
549 Bindings that fall under the monomorphism restriction can't
553 * Inlining remains valid
554 * No unexpected loss of sharing
555 * But simple bindings like
557 will be rejected, unless you add an explicit type signature
558 (to avoid the monomorphism restriction)
559 z :: (?y::Int) => Int
561 This seems unacceptable
563 (B) Monomorphism restriction "wins"
564 Bindings that fall under the monomorphism restriction can't
566 Always generalise over implicit parameters *except* for bindings
567 that fall under the monomorphism restriction
570 * Inlining isn't valid in general
571 * No unexpected loss of sharing
572 * Simple bindings like
574 accepted (get value of ?y from binding site)
576 (C) Always generalise over implicit parameters
577 Bindings that fall under the monomorphism restriction can't
578 be generalised, EXCEPT for implicit parameters
580 * Inlining remains valid
581 * Unexpected loss of sharing (from the extra generalisation)
582 * Simple bindings like
584 accepted (get value of ?y from occurrence sites)
589 None of these choices seems very satisfactory. But at least we should
590 decide which we want to do.
592 It's really not clear what is the Right Thing To Do. If you see
596 would you expect the value of ?y to be got from the *occurrence sites*
597 of 'z', or from the valuue of ?y at the *definition* of 'z'? In the
598 case of function definitions, the answer is clearly the former, but
599 less so in the case of non-fucntion definitions. On the other hand,
600 if we say that we get the value of ?y from the definition site of 'z',
601 then inlining 'z' might change the semantics of the program.
603 Choice (C) really says "the monomorphism restriction doesn't apply
604 to implicit parameters". Which is fine, but remember that every
605 innocent binding 'x = ...' that mentions an implicit parameter in
606 the RHS becomes a *function* of that parameter, called at each
607 use of 'x'. Now, the chances are that there are no intervening 'with'
608 clauses that bind ?y, so a decent compiler should common up all
609 those function calls. So I think I strongly favour (C). Indeed,
610 one could make a similar argument for abolishing the monomorphism
611 restriction altogether.
613 BOTTOM LINE: we choose (B) at present. See tcSimplifyRestricted
617 %************************************************************************
619 \subsection{tcSimplifyInfer}
621 %************************************************************************
623 tcSimplify is called when we *inferring* a type. Here's the overall game plan:
625 1. Compute Q = grow( fvs(T), C )
627 2. Partition C based on Q into Ct and Cq. Notice that ambiguous
628 predicates will end up in Ct; we deal with them at the top level
630 3. Try improvement, using functional dependencies
632 4. If Step 3 did any unification, repeat from step 1
633 (Unification can change the result of 'grow'.)
635 Note: we don't reduce dictionaries in step 2. For example, if we have
636 Eq (a,b), we don't simplify to (Eq a, Eq b). So Q won't be different
637 after step 2. However note that we may therefore quantify over more
638 type variables than we absolutely have to.
640 For the guts, we need a loop, that alternates context reduction and
641 improvement with unification. E.g. Suppose we have
643 class C x y | x->y where ...
645 and tcSimplify is called with:
647 Then improvement unifies a with b, giving
650 If we need to unify anything, we rattle round the whole thing all over
657 -> TcTyVarSet -- fv(T); type vars
659 -> TcM ([TcTyVar], -- Tyvars to quantify (zonked and quantified)
660 [Inst], -- Dict Ids that must be bound here (zonked)
661 TcDictBinds) -- Bindings
662 -- Any free (escaping) Insts are tossed into the environment
667 tcSimplifyInfer doc tau_tvs wanted
668 = do { tau_tvs' <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
669 ; wanted' <- mappM zonkInst wanted -- Zonk before deciding quantified tyvars
670 ; gbl_tvs <- tcGetGlobalTyVars
671 ; let preds = fdPredsOfInsts wanted'
672 qtvs = grow preds tau_tvs' `minusVarSet` oclose preds gbl_tvs
673 (free, bound) = partition (isFreeWhenInferring qtvs) wanted'
674 ; traceTc (text "infer" <+> (ppr preds $$ ppr (grow preds tau_tvs') $$ ppr gbl_tvs $$ ppr (oclose preds gbl_tvs) $$ ppr free $$ ppr bound))
677 -- To make types simple, reduce as much as possible
678 ; let try_me inst = ReduceMe AddSCs
679 ; (irreds, binds) <- checkLoop (mkRedEnv doc try_me []) bound
681 ; qtvs' <- zonkQuantifiedTyVars (varSetElems qtvs)
683 -- We can't abstract over implications
684 ; let (dicts, implics) = partition isDict irreds
685 ; loc <- getInstLoc (ImplicOrigin doc)
686 ; implic_bind <- bindIrreds loc qtvs' dicts implics
688 ; return (qtvs', dicts, binds `unionBags` implic_bind) }
689 -- NB: when we are done, we might have some bindings, but
690 -- the final qtvs might be empty. See Note [NO TYVARS] below.
694 -----------------------------------------------------------
695 -- tcSimplifyInferCheck is used when we know the constraints we are to simplify
696 -- against, but we don't know the type variables over which we are going to quantify.
697 -- This happens when we have a type signature for a mutually recursive group
700 -> TcTyVarSet -- fv(T)
703 -> TcM ([TyVar], -- Fully zonked, and quantified
704 TcDictBinds) -- Bindings
706 tcSimplifyInferCheck loc tau_tvs givens wanteds
707 = do { (irreds, binds) <- innerCheckLoop loc givens wanteds
709 -- Figure out which type variables to quantify over
710 -- You might think it should just be the signature tyvars,
711 -- but in bizarre cases you can get extra ones
712 -- f :: forall a. Num a => a -> a
713 -- f x = fst (g (x, head [])) + 1
715 -- Here we infer g :: forall a b. a -> b -> (b,a)
716 -- We don't want g to be monomorphic in b just because
717 -- f isn't quantified over b.
718 ; let all_tvs = varSetElems (tau_tvs `unionVarSet` tyVarsOfInsts givens)
719 ; all_tvs <- zonkTcTyVarsAndFV all_tvs
720 ; gbl_tvs <- tcGetGlobalTyVars
721 ; let qtvs = varSetElems (all_tvs `minusVarSet` gbl_tvs)
722 -- We could close gbl_tvs, but its not necessary for
723 -- soundness, and it'll only affect which tyvars, not which
724 -- dictionaries, we quantify over
726 ; qtvs' <- zonkQuantifiedTyVars qtvs
728 -- Now we are back to normal (c.f. tcSimplCheck)
729 ; implic_bind <- bindIrreds loc qtvs' givens irreds
731 ; return (qtvs', binds `unionBags` implic_bind) }
734 Note [Squashing methods]
735 ~~~~~~~~~~~~~~~~~~~~~~~~~
736 Be careful if you want to float methods more:
737 truncate :: forall a. RealFrac a => forall b. Integral b => a -> b
738 From an application (truncate f i) we get
741 If we have also have a second occurrence of truncate, we get
744 When simplifying with i,f free, we might still notice that
745 t1=t3; but alas, the binding for t2 (which mentions t1)
746 may continue to float out!
751 class Y a b | a -> b where
754 instance Y [[a]] a where
757 k :: X a -> X a -> X a
759 g :: Num a => [X a] -> [X a]
762 h ys = ys ++ map (k (y [[0]])) xs
764 The excitement comes when simplifying the bindings for h. Initially
765 try to simplify {y @ [[t1]] t2, 0 @ t1}, with initial qtvs = {t2}.
766 From this we get t1:=:t2, but also various bindings. We can't forget
767 the bindings (because of [LOOP]), but in fact t1 is what g is
770 The net effect of [NO TYVARS]
773 isFreeWhenInferring :: TyVarSet -> Inst -> Bool
774 isFreeWhenInferring qtvs inst
775 = isFreeWrtTyVars qtvs inst -- Constrains no quantified vars
776 && isInheritableInst inst -- and no implicit parameter involved
777 -- see Note [Inheriting implicit parameters]
779 {- No longer used (with implication constraints)
780 isFreeWhenChecking :: TyVarSet -- Quantified tyvars
781 -> NameSet -- Quantified implicit parameters
783 isFreeWhenChecking qtvs ips inst
784 = isFreeWrtTyVars qtvs inst
785 && isFreeWrtIPs ips inst
788 isFreeWrtTyVars qtvs inst = tyVarsOfInst inst `disjointVarSet` qtvs
789 isFreeWrtIPs ips inst = not (any (`elemNameSet` ips) (ipNamesOfInst inst))
793 %************************************************************************
795 \subsection{tcSimplifyCheck}
797 %************************************************************************
799 @tcSimplifyCheck@ is used when we know exactly the set of variables
800 we are going to quantify over. For example, a class or instance declaration.
803 -----------------------------------------------------------
804 -- tcSimplifyCheck is used when checking expression type signatures,
805 -- class decls, instance decls etc.
806 tcSimplifyCheck :: InstLoc
807 -> [TcTyVar] -- Quantify over these
810 -> TcM TcDictBinds -- Bindings
811 tcSimplifyCheck loc qtvs givens wanteds
812 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
813 do { (irreds, binds) <- innerCheckLoop loc givens wanteds
814 ; implic_bind <- bindIrreds loc qtvs givens irreds
815 ; return (binds `unionBags` implic_bind) }
817 -----------------------------------------------------------
818 -- tcSimplifyCheckPat is used for existential pattern match
819 tcSimplifyCheckPat :: InstLoc
820 -> [CoVar] -> Refinement
821 -> [TcTyVar] -- Quantify over these
824 -> TcM TcDictBinds -- Bindings
825 tcSimplifyCheckPat loc co_vars reft qtvs givens wanteds
826 = ASSERT( all isTcTyVar qtvs && all isSkolemTyVar qtvs )
827 do { (irreds, binds) <- innerCheckLoop loc givens wanteds
828 ; implic_bind <- bindIrredsR loc qtvs co_vars reft
830 ; return (binds `unionBags` implic_bind) }
832 -----------------------------------------------------------
833 bindIrreds :: InstLoc -> [TcTyVar]
836 bindIrreds loc qtvs givens irreds
837 = bindIrredsR loc qtvs [] emptyRefinement givens irreds
839 bindIrredsR :: InstLoc -> [TcTyVar] -> [CoVar]
840 -> Refinement -> [Inst] -> [Inst]
842 -- Make a binding that binds 'irreds', by generating an implication
843 -- constraint for them, *and* throwing the constraint into the LIE
844 bindIrredsR loc qtvs co_vars reft givens irreds
848 = do { let givens' = filter isDict givens
849 -- The givens can include methods
850 -- See Note [Pruning the givens in an implication constraint]
852 -- If there are no 'givens' *and* the refinement is empty
853 -- (the refinement is like more givens), then it's safe to
854 -- partition the 'wanteds' by their qtvs, thereby trimming irreds
855 -- See Note [Freeness and implications]
856 ; irreds' <- if null givens' && isEmptyRefinement reft
858 { let qtv_set = mkVarSet qtvs
859 (frees, real_irreds) = partition (isFreeWrtTyVars qtv_set) irreds
861 ; return real_irreds }
864 ; let all_tvs = qtvs ++ co_vars -- Abstract over all these
865 ; (implics, bind) <- makeImplicationBind loc all_tvs reft givens' irreds'
866 -- This call does the real work
867 -- If irreds' is empty, it does something sensible
872 makeImplicationBind :: InstLoc -> [TcTyVar] -> Refinement
874 -> TcM ([Inst], TcDictBinds)
875 -- Make a binding that binds 'irreds', by generating an implication
876 -- constraint for them, *and* throwing the constraint into the LIE
877 -- The binding looks like
878 -- (ir1, .., irn) = f qtvs givens
879 -- where f is (evidence for) the new implication constraint
880 -- f :: forall qtvs. {reft} givens => (ir1, .., irn)
881 -- qtvs includes coercion variables
883 -- This binding must line up the 'rhs' in reduceImplication
884 makeImplicationBind loc all_tvs reft
885 givens -- Guaranteed all Dicts
887 | null irreds -- If there are no irreds, we are done
888 = return ([], emptyBag)
889 | otherwise -- Otherwise we must generate a binding
890 = do { uniq <- newUnique
891 ; span <- getSrcSpanM
892 ; let name = mkInternalName uniq (mkVarOcc "ic") (srcSpanStart span)
893 implic_inst = ImplicInst { tci_name = name, tci_reft = reft,
894 tci_tyvars = all_tvs,
896 tci_wanted = irreds, tci_loc = loc }
898 ; let n_irreds = length irreds
899 irred_ids = map instToId irreds
900 tup_ty = mkTupleTy Boxed n_irreds (map idType irred_ids)
901 pat = TuplePat (map nlVarPat irred_ids) Boxed tup_ty
902 rhs = L span (mkHsWrap co (HsVar (instToId implic_inst)))
903 co = mkWpApps (map instToId givens) <.> mkWpTyApps (mkTyVarTys all_tvs)
904 bind | n_irreds==1 = VarBind (head irred_ids) rhs
905 | otherwise = PatBind { pat_lhs = L span pat,
906 pat_rhs = unguardedGRHSs rhs,
908 bind_fvs = placeHolderNames }
909 ; -- pprTrace "Make implic inst" (ppr implic_inst) $
910 return ([implic_inst], unitBag (L span bind)) }
912 -----------------------------------------------------------
915 -> TcM ([Inst], TcDictBinds)
917 topCheckLoop doc wanteds
918 = checkLoop (mkRedEnv doc try_me []) wanteds
920 try_me inst = ReduceMe AddSCs
922 -----------------------------------------------------------
923 innerCheckLoop :: InstLoc
926 -> TcM ([Inst], TcDictBinds)
928 innerCheckLoop inst_loc givens wanteds
929 = checkLoop env wanteds
931 env = mkRedEnv (pprInstLoc inst_loc) try_me givens
933 try_me inst | isMethodOrLit inst = ReduceMe AddSCs
935 -- When checking against a given signature
936 -- we MUST be very gentle: Note [Check gently]
941 We have to very careful about not simplifying too vigorously
946 f :: Show b => T b -> b
949 Inside the pattern match, which binds (a:*, x:a), we know that
951 Hence we have a dictionary for Show [a] available; and indeed we
952 need it. We are going to build an implication contraint
953 forall a. (b~[a]) => Show [a]
954 Later, we will solve this constraint using the knowledge (Show b)
956 But we MUST NOT reduce (Show [a]) to (Show a), else the whole
957 thing becomes insoluble. So we simplify gently (get rid of literals
958 and methods only, plus common up equal things), deferring the real
959 work until top level, when we solve the implication constraint
964 -----------------------------------------------------------
967 -> TcM ([Inst], TcDictBinds)
968 -- Precondition: givens are completely rigid
970 checkLoop env wanteds
971 = do { -- Givens are skolems, so no need to zonk them
972 wanteds' <- mappM zonkInst wanteds
974 ; (improved, binds, irreds) <- reduceContext env wanteds'
976 ; if not improved then
977 return (irreds, binds)
980 -- If improvement did some unification, we go round again.
981 -- We start again with irreds, not wanteds
982 -- Using an instance decl might have introduced a fresh type variable
983 -- which might have been unified, so we'd get an infinite loop
984 -- if we started again with wanteds! See Note [LOOP]
985 { (irreds1, binds1) <- checkLoop env irreds
986 ; return (irreds1, binds `unionBags` binds1) } }
991 class If b t e r | b t e -> r
994 class Lte a b c | a b -> c where lte :: a -> b -> c
996 instance (Lte a b l,If l b a c) => Max a b c
998 Wanted: Max Z (S x) y
1000 Then we'll reduce using the Max instance to:
1001 (Lte Z (S x) l, If l (S x) Z y)
1002 and improve by binding l->T, after which we can do some reduction
1003 on both the Lte and If constraints. What we *can't* do is start again
1004 with (Max Z (S x) y)!
1008 %************************************************************************
1010 tcSimplifySuperClasses
1012 %************************************************************************
1014 Note [SUPERCLASS-LOOP 1]
1015 ~~~~~~~~~~~~~~~~~~~~~~~~
1016 We have to be very, very careful when generating superclasses, lest we
1017 accidentally build a loop. Here's an example:
1021 class S a => C a where { opc :: a -> a }
1022 class S b => D b where { opd :: b -> b }
1024 instance C Int where
1027 instance D Int where
1030 From (instance C Int) we get the constraint set {ds1:S Int, dd:D Int}
1031 Simplifying, we may well get:
1032 $dfCInt = :C ds1 (opd dd)
1035 Notice that we spot that we can extract ds1 from dd.
1037 Alas! Alack! We can do the same for (instance D Int):
1039 $dfDInt = :D ds2 (opc dc)
1043 And now we've defined the superclass in terms of itself.
1045 Solution: never generate a superclass selectors at all when
1046 satisfying the superclass context of an instance declaration.
1048 Two more nasty cases are in
1053 tcSimplifySuperClasses
1058 tcSimplifySuperClasses loc givens sc_wanteds
1059 = do { (irreds, binds1) <- checkLoop env sc_wanteds
1060 ; let (tidy_env, tidy_irreds) = tidyInsts irreds
1061 ; reportNoInstances tidy_env (Just (loc, givens)) tidy_irreds
1064 env = mkRedEnv (pprInstLoc loc) try_me givens
1065 try_me inst = ReduceMe NoSCs
1066 -- Like topCheckLoop, but with NoSCs
1070 %************************************************************************
1072 \subsection{tcSimplifyRestricted}
1074 %************************************************************************
1076 tcSimplifyRestricted infers which type variables to quantify for a
1077 group of restricted bindings. This isn't trivial.
1080 We want to quantify over a to get id :: forall a. a->a
1083 We do not want to quantify over a, because there's an Eq a
1084 constraint, so we get eq :: a->a->Bool (notice no forall)
1087 RHS has type 'tau', whose free tyvars are tau_tvs
1088 RHS has constraints 'wanteds'
1091 Quantify over (tau_tvs \ ftvs(wanteds))
1092 This is bad. The constraints may contain (Monad (ST s))
1093 where we have instance Monad (ST s) where...
1094 so there's no need to be monomorphic in s!
1096 Also the constraint might be a method constraint,
1097 whose type mentions a perfectly innocent tyvar:
1098 op :: Num a => a -> b -> a
1099 Here, b is unconstrained. A good example would be
1101 We want to infer the polymorphic type
1102 foo :: forall b. b -> b
1105 Plan B (cunning, used for a long time up to and including GHC 6.2)
1106 Step 1: Simplify the constraints as much as possible (to deal
1107 with Plan A's problem). Then set
1108 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1110 Step 2: Now simplify again, treating the constraint as 'free' if
1111 it does not mention qtvs, and trying to reduce it otherwise.
1112 The reasons for this is to maximise sharing.
1114 This fails for a very subtle reason. Suppose that in the Step 2
1115 a constraint (Foo (Succ Zero) (Succ Zero) b) gets thrown upstairs as 'free'.
1116 In the Step 1 this constraint might have been simplified, perhaps to
1117 (Foo Zero Zero b), AND THEN THAT MIGHT BE IMPROVED, to bind 'b' to 'T'.
1118 This won't happen in Step 2... but that in turn might prevent some other
1119 constraint (Baz [a] b) being simplified (e.g. via instance Baz [a] T where {..})
1120 and that in turn breaks the invariant that no constraints are quantified over.
1122 Test typecheck/should_compile/tc177 (which failed in GHC 6.2) demonstrates
1127 Step 1: Simplify the constraints as much as possible (to deal
1128 with Plan A's problem). Then set
1129 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1130 Return the bindings from Step 1.
1133 A note about Plan C (arising from "bug" reported by George Russel March 2004)
1136 instance (HasBinary ty IO) => HasCodedValue ty
1138 foo :: HasCodedValue a => String -> IO a
1140 doDecodeIO :: HasCodedValue a => () -> () -> IO a
1141 doDecodeIO codedValue view
1142 = let { act = foo "foo" } in act
1144 You might think this should work becuase the call to foo gives rise to a constraint
1145 (HasCodedValue t), which can be satisfied by the type sig for doDecodeIO. But the
1146 restricted binding act = ... calls tcSimplifyRestricted, and PlanC simplifies the
1147 constraint using the (rather bogus) instance declaration, and now we are stuffed.
1149 I claim this is not really a bug -- but it bit Sergey as well as George. So here's
1153 Plan D (a variant of plan B)
1154 Step 1: Simplify the constraints as much as possible (to deal
1155 with Plan A's problem), BUT DO NO IMPROVEMENT. Then set
1156 qtvs = tau_tvs \ ftvs( simplify( wanteds ) )
1158 Step 2: Now simplify again, treating the constraint as 'free' if
1159 it does not mention qtvs, and trying to reduce it otherwise.
1161 The point here is that it's generally OK to have too few qtvs; that is,
1162 to make the thing more monomorphic than it could be. We don't want to
1163 do that in the common cases, but in wierd cases it's ok: the programmer
1164 can always add a signature.
1166 Too few qtvs => too many wanteds, which is what happens if you do less
1171 tcSimplifyRestricted -- Used for restricted binding groups
1172 -- i.e. ones subject to the monomorphism restriction
1175 -> [Name] -- Things bound in this group
1176 -> TcTyVarSet -- Free in the type of the RHSs
1177 -> [Inst] -- Free in the RHSs
1178 -> TcM ([TyVar], -- Tyvars to quantify (zonked and quantified)
1179 TcDictBinds) -- Bindings
1180 -- tcSimpifyRestricted returns no constraints to
1181 -- quantify over; by definition there are none.
1182 -- They are all thrown back in the LIE
1184 tcSimplifyRestricted doc top_lvl bndrs tau_tvs wanteds
1185 -- Zonk everything in sight
1186 = do { wanteds' <- mappM zonkInst wanteds
1188 -- 'ReduceMe': Reduce as far as we can. Don't stop at
1189 -- dicts; the idea is to get rid of as many type
1190 -- variables as possible, and we don't want to stop
1191 -- at (say) Monad (ST s), because that reduces
1192 -- immediately, with no constraint on s.
1194 -- BUT do no improvement! See Plan D above
1195 -- HOWEVER, some unification may take place, if we instantiate
1196 -- a method Inst with an equality constraint
1197 ; let env = mkNoImproveRedEnv doc (\i -> ReduceMe AddSCs)
1198 ; (_imp, _binds, constrained_dicts) <- reduceContext env wanteds'
1200 -- Next, figure out the tyvars we will quantify over
1201 ; tau_tvs' <- zonkTcTyVarsAndFV (varSetElems tau_tvs)
1202 ; gbl_tvs' <- tcGetGlobalTyVars
1203 ; constrained_dicts' <- mappM zonkInst constrained_dicts
1205 ; let qtvs1 = tau_tvs' `minusVarSet` oclose (fdPredsOfInsts constrained_dicts) gbl_tvs'
1206 -- As in tcSimplifyInfer
1208 -- Do not quantify over constrained type variables:
1209 -- this is the monomorphism restriction
1210 constrained_tvs' = tyVarsOfInsts constrained_dicts'
1211 qtvs = qtvs1 `minusVarSet` constrained_tvs'
1212 pp_bndrs = pprWithCommas (quotes . ppr) bndrs
1215 ; warn_mono <- doptM Opt_WarnMonomorphism
1216 ; warnTc (warn_mono && (constrained_tvs' `intersectsVarSet` qtvs1))
1217 (vcat[ ptext SLIT("the Monomorphism Restriction applies to the binding")
1218 <> plural bndrs <+> ptext SLIT("for") <+> pp_bndrs,
1219 ptext SLIT("Consider giving a type signature for") <+> pp_bndrs])
1221 ; traceTc (text "tcSimplifyRestricted" <+> vcat [
1222 pprInsts wanteds, pprInsts constrained_dicts',
1224 ppr constrained_tvs', ppr tau_tvs', ppr qtvs ])
1226 -- The first step may have squashed more methods than
1227 -- necessary, so try again, this time more gently, knowing the exact
1228 -- set of type variables to quantify over.
1230 -- We quantify only over constraints that are captured by qtvs;
1231 -- these will just be a subset of non-dicts. This in contrast
1232 -- to normal inference (using isFreeWhenInferring) in which we quantify over
1233 -- all *non-inheritable* constraints too. This implements choice
1234 -- (B) under "implicit parameter and monomorphism" above.
1236 -- Remember that we may need to do *some* simplification, to
1237 -- (for example) squash {Monad (ST s)} into {}. It's not enough
1238 -- just to float all constraints
1240 -- At top level, we *do* squash methods becuase we want to
1241 -- expose implicit parameters to the test that follows
1242 ; let is_nested_group = isNotTopLevel top_lvl
1243 try_me inst | isFreeWrtTyVars qtvs inst,
1244 (is_nested_group || isDict inst) = Stop
1245 | otherwise = ReduceMe AddSCs
1246 env = mkNoImproveRedEnv doc try_me
1247 ; (_imp, binds, irreds) <- reduceContext env wanteds'
1249 -- See "Notes on implicit parameters, Question 4: top level"
1250 ; ASSERT( all (isFreeWrtTyVars qtvs) irreds ) -- None should be captured
1251 if is_nested_group then
1253 else do { let (bad_ips, non_ips) = partition isIPDict irreds
1254 ; addTopIPErrs bndrs bad_ips
1255 ; extendLIEs non_ips }
1257 ; qtvs' <- zonkQuantifiedTyVars (varSetElems qtvs)
1258 ; return (qtvs', binds) }
1262 %************************************************************************
1266 %************************************************************************
1268 On the LHS of transformation rules we only simplify methods and constants,
1269 getting dictionaries. We want to keep all of them unsimplified, to serve
1270 as the available stuff for the RHS of the rule.
1272 Example. Consider the following left-hand side of a rule
1274 f (x == y) (y > z) = ...
1276 If we typecheck this expression we get constraints
1278 d1 :: Ord a, d2 :: Eq a
1280 We do NOT want to "simplify" to the LHS
1282 forall x::a, y::a, z::a, d1::Ord a.
1283 f ((==) (eqFromOrd d1) x y) ((>) d1 y z) = ...
1287 forall x::a, y::a, z::a, d1::Ord a, d2::Eq a.
1288 f ((==) d2 x y) ((>) d1 y z) = ...
1290 Here is another example:
1292 fromIntegral :: (Integral a, Num b) => a -> b
1293 {-# RULES "foo" fromIntegral = id :: Int -> Int #-}
1295 In the rule, a=b=Int, and Num Int is a superclass of Integral Int. But
1296 we *dont* want to get
1298 forall dIntegralInt.
1299 fromIntegral Int Int dIntegralInt (scsel dIntegralInt) = id Int
1301 because the scsel will mess up RULE matching. Instead we want
1303 forall dIntegralInt, dNumInt.
1304 fromIntegral Int Int dIntegralInt dNumInt = id Int
1308 g (x == y) (y == z) = ..
1310 where the two dictionaries are *identical*, we do NOT WANT
1312 forall x::a, y::a, z::a, d1::Eq a
1313 f ((==) d1 x y) ((>) d1 y z) = ...
1315 because that will only match if the dict args are (visibly) equal.
1316 Instead we want to quantify over the dictionaries separately.
1318 In short, tcSimplifyRuleLhs must *only* squash LitInst and MethInts, leaving
1319 all dicts unchanged, with absolutely no sharing. It's simpler to do this
1320 from scratch, rather than further parameterise simpleReduceLoop etc
1323 tcSimplifyRuleLhs :: [Inst] -> TcM ([Inst], TcDictBinds)
1324 tcSimplifyRuleLhs wanteds
1325 = go [] emptyBag wanteds
1328 = return (dicts, binds)
1329 go dicts binds (w:ws)
1331 = go (w:dicts) binds ws
1333 = do { w' <- zonkInst w -- So that (3::Int) does not generate a call
1334 -- to fromInteger; this looks fragile to me
1335 ; lookup_result <- lookupSimpleInst w'
1336 ; case lookup_result of
1337 GenInst ws' rhs -> go dicts (addBind binds w rhs) (ws' ++ ws)
1338 NoInstance -> pprPanic "tcSimplifyRuleLhs" (ppr w)
1342 tcSimplifyBracket is used when simplifying the constraints arising from
1343 a Template Haskell bracket [| ... |]. We want to check that there aren't
1344 any constraints that can't be satisfied (e.g. Show Foo, where Foo has no
1345 Show instance), but we aren't otherwise interested in the results.
1346 Nor do we care about ambiguous dictionaries etc. We will type check
1347 this bracket again at its usage site.
1350 tcSimplifyBracket :: [Inst] -> TcM ()
1351 tcSimplifyBracket wanteds
1352 = do { topCheckLoop doc wanteds
1355 doc = text "tcSimplifyBracket"
1359 %************************************************************************
1361 \subsection{Filtering at a dynamic binding}
1363 %************************************************************************
1368 we must discharge all the ?x constraints from B. We also do an improvement
1369 step; if we have ?x::t1 and ?x::t2 we must unify t1, t2.
1371 Actually, the constraints from B might improve the types in ?x. For example
1373 f :: (?x::Int) => Char -> Char
1376 then the constraint (?x::Int) arising from the call to f will
1377 force the binding for ?x to be of type Int.
1380 tcSimplifyIPs :: [Inst] -- The implicit parameters bound here
1383 -- We need a loop so that we do improvement, and then
1384 -- (next time round) generate a binding to connect the two
1386 -- Here the two ?x's have different types, and improvement
1387 -- makes them the same.
1389 tcSimplifyIPs given_ips wanteds
1390 = do { wanteds' <- mappM zonkInst wanteds
1391 ; given_ips' <- mappM zonkInst given_ips
1392 -- Unusually for checking, we *must* zonk the given_ips
1394 ; let env = mkRedEnv doc try_me given_ips'
1395 ; (improved, binds, irreds) <- reduceContext env wanteds'
1397 ; if not improved then
1398 ASSERT( all is_free irreds )
1399 do { extendLIEs irreds
1402 tcSimplifyIPs given_ips wanteds }
1404 doc = text "tcSimplifyIPs" <+> ppr given_ips
1405 ip_set = mkNameSet (ipNamesOfInsts given_ips)
1406 is_free inst = isFreeWrtIPs ip_set inst
1408 -- Simplify any methods that mention the implicit parameter
1409 try_me inst | is_free inst = Stop
1410 | otherwise = ReduceMe NoSCs
1414 %************************************************************************
1416 \subsection[binds-for-local-funs]{@bindInstsOfLocalFuns@}
1418 %************************************************************************
1420 When doing a binding group, we may have @Insts@ of local functions.
1421 For example, we might have...
1423 let f x = x + 1 -- orig local function (overloaded)
1424 f.1 = f Int -- two instances of f
1429 The point is: we must drop the bindings for @f.1@ and @f.2@ here,
1430 where @f@ is in scope; those @Insts@ must certainly not be passed
1431 upwards towards the top-level. If the @Insts@ were binding-ified up
1432 there, they would have unresolvable references to @f@.
1434 We pass in an @init_lie@ of @Insts@ and a list of locally-bound @Ids@.
1435 For each method @Inst@ in the @init_lie@ that mentions one of the
1436 @Ids@, we create a binding. We return the remaining @Insts@ (in an
1437 @LIE@), as well as the @HsBinds@ generated.
1440 bindInstsOfLocalFuns :: [Inst] -> [TcId] -> TcM TcDictBinds
1441 -- Simlifies only MethodInsts, and generate only bindings of form
1443 -- We're careful not to even generate bindings of the form
1445 -- You'd think that'd be fine, but it interacts with what is
1446 -- arguably a bug in Match.tidyEqnInfo (see notes there)
1448 bindInstsOfLocalFuns wanteds local_ids
1449 | null overloaded_ids
1451 = extendLIEs wanteds `thenM_`
1452 returnM emptyLHsBinds
1455 = do { (irreds, binds) <- checkLoop env for_me
1456 ; extendLIEs not_for_me
1460 env = mkRedEnv doc try_me []
1461 doc = text "bindInsts" <+> ppr local_ids
1462 overloaded_ids = filter is_overloaded local_ids
1463 is_overloaded id = isOverloadedTy (idType id)
1464 (for_me, not_for_me) = partition (isMethodFor overloaded_set) wanteds
1466 overloaded_set = mkVarSet overloaded_ids -- There can occasionally be a lot of them
1467 -- so it's worth building a set, so that
1468 -- lookup (in isMethodFor) is faster
1469 try_me inst | isMethod inst = ReduceMe NoSCs
1474 %************************************************************************
1476 \subsection{Data types for the reduction mechanism}
1478 %************************************************************************
1480 The main control over context reduction is here
1484 = RedEnv { red_doc :: SDoc -- The context
1485 , red_try_me :: Inst -> WhatToDo
1486 , red_improve :: Bool -- True <=> do improvement
1487 , red_givens :: [Inst] -- All guaranteed rigid
1489 -- but see Note [Rigidity]
1490 , red_stack :: (Int, [Inst]) -- Recursion stack (for err msg)
1491 -- See Note [RedStack]
1495 -- The red_givens are rigid so far as cmpInst is concerned.
1496 -- There is one case where they are not totally rigid, namely in tcSimplifyIPs
1497 -- let ?x = e in ...
1498 -- Here, the given is (?x::a), where 'a' is not necy a rigid type
1499 -- But that doesn't affect the comparison, which is based only on mame.
1502 -- The red_stack pair (n,insts) pair is just used for error reporting.
1503 -- 'n' is always the depth of the stack.
1504 -- The 'insts' is the stack of Insts being reduced: to produce X
1505 -- I had to produce Y, to produce Y I had to produce Z, and so on.
1508 mkRedEnv :: SDoc -> (Inst -> WhatToDo) -> [Inst] -> RedEnv
1509 mkRedEnv doc try_me givens
1510 = RedEnv { red_doc = doc, red_try_me = try_me,
1511 red_givens = givens, red_stack = (0,[]),
1512 red_improve = True }
1514 mkNoImproveRedEnv :: SDoc -> (Inst -> WhatToDo) -> RedEnv
1515 -- Do not do improvement; no givens
1516 mkNoImproveRedEnv doc try_me
1517 = RedEnv { red_doc = doc, red_try_me = try_me,
1518 red_givens = [], red_stack = (0,[]),
1519 red_improve = True }
1522 = ReduceMe WantSCs -- Try to reduce this
1523 -- If there's no instance, add the inst to the
1524 -- irreductible ones, but don't produce an error
1525 -- message of any kind.
1526 -- It might be quite legitimate such as (Eq a)!
1528 | Stop -- Return as irreducible unless it can
1529 -- be reduced to a constant in one step
1530 -- Do not add superclasses; see
1532 data WantSCs = NoSCs | AddSCs -- Tells whether we should add the superclasses
1533 -- of a predicate when adding it to the avails
1534 -- The reason for this flag is entirely the super-class loop problem
1535 -- Note [SUPER-CLASS LOOP 1]
1538 %************************************************************************
1540 \subsection[reduce]{@reduce@}
1542 %************************************************************************
1546 reduceContext :: RedEnv
1548 -> TcM (ImprovementDone,
1549 TcDictBinds, -- Dictionary bindings
1550 [Inst]) -- Irreducible
1552 reduceContext env wanteds
1553 = do { traceTc (text "reduceContext" <+> (vcat [
1554 text "----------------------",
1556 text "given" <+> ppr (red_givens env),
1557 text "wanted" <+> ppr wanteds,
1558 text "----------------------"
1561 -- Build the Avail mapping from "givens"
1562 ; init_state <- foldlM addGiven emptyAvails (red_givens env)
1565 ; avails <- reduceList env wanteds init_state
1567 ; let improved = availsImproved avails
1568 ; (binds, irreds) <- extractResults avails wanteds
1570 ; traceTc (text "reduceContext end" <+> (vcat [
1571 text "----------------------",
1573 text "given" <+> ppr (red_givens env),
1574 text "wanted" <+> ppr wanteds,
1576 text "avails" <+> pprAvails avails,
1577 text "improved =" <+> ppr improved,
1578 text "----------------------"
1581 ; return (improved, binds, irreds) }
1583 tcImproveOne :: Avails -> Inst -> TcM ImprovementDone
1584 tcImproveOne avails inst
1585 | not (isDict inst) = return False
1587 = do { inst_envs <- tcGetInstEnvs
1588 ; let eqns = improveOne (classInstances inst_envs)
1589 (dictPred inst, pprInstArising inst)
1590 [ (dictPred p, pprInstArising p)
1591 | p <- availsInsts avails, isDict p ]
1592 -- Avails has all the superclasses etc (good)
1593 -- It also has all the intermediates of the deduction (good)
1594 -- It does not have duplicates (good)
1595 -- NB that (?x::t1) and (?x::t2) will be held separately in avails
1596 -- so that improve will see them separate
1597 ; traceTc (text "improveOne" <+> ppr inst)
1600 unifyEqns :: [(Equation,(PredType,SDoc),(PredType,SDoc))]
1601 -> TcM ImprovementDone
1602 unifyEqns [] = return False
1604 = do { traceTc (ptext SLIT("Improve:") <+> vcat (map pprEquationDoc eqns))
1608 unify ((qtvs, pairs), what1, what2)
1609 = addErrCtxtM (mkEqnMsg what1 what2) $
1610 tcInstTyVars (varSetElems qtvs) `thenM` \ (_, _, tenv) ->
1611 mapM_ (unif_pr tenv) pairs
1612 unif_pr tenv (ty1,ty2) = unifyType (substTy tenv ty1) (substTy tenv ty2)
1614 pprEquationDoc (eqn, (p1,w1), (p2,w2)) = vcat [pprEquation eqn, nest 2 (ppr p1), nest 2 (ppr p2)]
1616 mkEqnMsg (pred1,from1) (pred2,from2) tidy_env
1617 = do { pred1' <- zonkTcPredType pred1; pred2' <- zonkTcPredType pred2
1618 ; let { pred1'' = tidyPred tidy_env pred1'; pred2'' = tidyPred tidy_env pred2' }
1619 ; let msg = vcat [ptext SLIT("When using functional dependencies to combine"),
1620 nest 2 (sep [ppr pred1'' <> comma, nest 2 from1]),
1621 nest 2 (sep [ppr pred2'' <> comma, nest 2 from2])]
1622 ; return (tidy_env, msg) }
1625 The main context-reduction function is @reduce@. Here's its game plan.
1628 reduceList :: RedEnv -> [Inst] -> Avails -> TcM Avails
1629 reduceList env@(RedEnv {red_stack = (n,stk)}) wanteds state
1630 = do { dopts <- getDOpts
1633 dumpTcRn (hang (ptext SLIT("Interesting! Context reduction stack depth") <+> int n)
1634 2 (ifPprDebug (nest 2 (pprStack stk))))
1637 ; if n >= ctxtStkDepth dopts then
1638 failWithTc (reduceDepthErr n stk)
1642 go [] state = return state
1643 go (w:ws) state = do { state' <- reduce (env {red_stack = (n+1, w:stk)}) w state
1646 -- Base case: we're done!
1647 reduce env wanted avails
1648 -- It's the same as an existing inst, or a superclass thereof
1649 | Just avail <- findAvail avails wanted
1653 = case red_try_me env wanted of {
1654 ; Stop -> try_simple (addIrred NoSCs) -- See Note [No superclasses for Stop]
1656 ; ReduceMe want_scs -> -- It should be reduced
1657 reduceInst env avails wanted `thenM` \ (avails, lookup_result) ->
1658 case lookup_result of
1659 NoInstance -> -- No such instance!
1660 -- Add it and its superclasses
1661 addIrred want_scs avails wanted
1663 GenInst [] rhs -> addWanted want_scs avails wanted rhs []
1665 GenInst wanteds' rhs -> do { avails1 <- addIrred NoSCs avails wanted
1666 ; avails2 <- reduceList env wanteds' avails1
1667 ; addWanted want_scs avails2 wanted rhs wanteds' }
1668 -- Temporarily do addIrred *before* the reduceList,
1669 -- which has the effect of adding the thing we are trying
1670 -- to prove to the database before trying to prove the things it
1671 -- needs. See note [RECURSIVE DICTIONARIES]
1672 -- NB: we must not do an addWanted before, because that adds the
1673 -- superclasses too, and thaat can lead to a spurious loop; see
1674 -- the examples in [SUPERCLASS-LOOP]
1675 -- So we do an addIrred before, and then overwrite it afterwards with addWanted
1679 -- First, see if the inst can be reduced to a constant in one step
1680 -- Works well for literals (1::Int) and constant dictionaries (d::Num Int)
1681 -- Don't bother for implication constraints, which take real work
1682 try_simple do_this_otherwise
1683 = do { res <- lookupSimpleInst wanted
1685 GenInst [] rhs -> addWanted AddSCs avails wanted rhs []
1686 other -> do_this_otherwise avails wanted }
1690 Note [SUPERCLASS-LOOP 2]
1691 ~~~~~~~~~~~~~~~~~~~~~~~~
1692 But the above isn't enough. Suppose we are *given* d1:Ord a,
1693 and want to deduce (d2:C [a]) where
1695 class Ord a => C a where
1696 instance Ord [a] => C [a] where ...
1698 Then we'll use the instance decl to deduce C [a] from Ord [a], and then add the
1699 superclasses of C [a] to avails. But we must not overwrite the binding
1700 for Ord [a] (which is obtained from Ord a) with a superclass selection or we'll just
1703 Here's another variant, immortalised in tcrun020
1704 class Monad m => C1 m
1705 class C1 m => C2 m x
1706 instance C2 Maybe Bool
1707 For the instance decl we need to build (C1 Maybe), and it's no good if
1708 we run around and add (C2 Maybe Bool) and its superclasses to the avails
1709 before we search for C1 Maybe.
1711 Here's another example
1712 class Eq b => Foo a b
1713 instance Eq a => Foo [a] a
1717 we'll first deduce that it holds (via the instance decl). We must not
1718 then overwrite the Eq t constraint with a superclass selection!
1720 At first I had a gross hack, whereby I simply did not add superclass constraints
1721 in addWanted, though I did for addGiven and addIrred. This was sub-optimal,
1722 becuase it lost legitimate superclass sharing, and it still didn't do the job:
1723 I found a very obscure program (now tcrun021) in which improvement meant the
1724 simplifier got two bites a the cherry... so something seemed to be an Stop
1725 first time, but reducible next time.
1727 Now we implement the Right Solution, which is to check for loops directly
1728 when adding superclasses. It's a bit like the occurs check in unification.
1731 Note [RECURSIVE DICTIONARIES]
1732 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1734 data D r = ZeroD | SuccD (r (D r));
1736 instance (Eq (r (D r))) => Eq (D r) where
1737 ZeroD == ZeroD = True
1738 (SuccD a) == (SuccD b) = a == b
1741 equalDC :: D [] -> D [] -> Bool;
1744 We need to prove (Eq (D [])). Here's how we go:
1748 by instance decl, holds if
1752 by instance decl of Eq, holds if
1754 where d2 = dfEqList d3
1757 But now we can "tie the knot" to give
1763 and it'll even run! The trick is to put the thing we are trying to prove
1764 (in this case Eq (D []) into the database before trying to prove its
1765 contributing clauses.
1768 %************************************************************************
1770 Reducing a single constraint
1772 %************************************************************************
1775 ---------------------------------------------
1776 reduceInst :: RedEnv -> Avails -> Inst -> TcM (Avails, LookupInstResult)
1777 reduceInst env avails (ImplicInst { tci_tyvars = tvs, tci_reft = reft, tci_loc = loc,
1778 tci_given = extra_givens, tci_wanted = wanteds })
1779 = reduceImplication env avails reft tvs extra_givens wanteds loc
1781 reduceInst env avails other_inst
1782 = do { result <- lookupSimpleInst other_inst
1783 ; return (avails, result) }
1787 ---------------------------------------------
1788 reduceImplication :: RedEnv
1790 -> Refinement -- May refine the givens; often empty
1791 -> [TcTyVar] -- Quantified type variables; all skolems
1792 -> [Inst] -- Extra givens; all rigid
1795 -> TcM (Avails, LookupInstResult)
1798 Suppose we are simplifying the constraint
1799 forall bs. extras => wanted
1800 in the context of an overall simplification problem with givens 'givens',
1801 and refinment 'reft'.
1804 * The refinement is often empty
1806 * The 'extra givens' need not mention any of the quantified type variables
1807 e.g. forall {}. Eq a => Eq [a]
1808 forall {}. C Int => D (Tree Int)
1810 This happens when you have something like
1812 T1 :: Eq a => a -> T a
1815 f x = ...(case x of { T1 v -> v==v })...
1818 -- ToDo: should we instantiate tvs? I think it's not necessary
1820 -- ToDo: what about improvement? There may be some improvement
1821 -- exposed as a result of the simplifications done by reduceList
1822 -- which are discarded if we back off.
1823 -- This is almost certainly Wrong, but we'll fix it when dealing
1824 -- better with equality constraints
1825 reduceImplication env orig_avails reft tvs extra_givens wanteds inst_loc
1826 = do { -- Add refined givens, and the extra givens
1827 (refined_red_givens, avails)
1828 <- if isEmptyRefinement reft then return (red_givens env, orig_avails)
1829 else foldlM (addRefinedGiven reft) ([], orig_avails) (red_givens env)
1830 ; avails <- foldlM addGiven avails extra_givens
1832 -- Solve the sub-problem
1833 ; let try_me inst = ReduceMe AddSCs -- Note [Freeness and implications]
1834 env' = env { red_givens = refined_red_givens ++ extra_givens
1835 , red_try_me = try_me }
1837 ; traceTc (text "reduceImplication" <+> vcat
1839 ppr (red_givens env), ppr extra_givens,
1840 ppr reft, ppr wanteds, ppr avails ])
1841 ; avails <- reduceList env' wanteds avails
1843 -- Extract the binding
1844 ; (binds, irreds) <- extractResults avails wanteds
1846 -- We always discard the extra avails we've generated;
1847 -- but we remember if we have done any (global) improvement
1848 ; let ret_avails = updateImprovement orig_avails avails
1850 ; if isEmptyLHsBinds binds then -- No progress
1851 return (ret_avails, NoInstance)
1853 { (implic_insts, bind) <- makeImplicationBind inst_loc tvs reft extra_givens irreds
1854 -- This binding is useless if the recursive simplification
1855 -- made no progress; but currently we don't try to optimise that
1856 -- case. After all, we only try hard to reduce at top level, or
1857 -- when inferring types.
1859 ; let dict_ids = map instToId extra_givens
1860 co = mkWpTyLams tvs <.> mkWpLams dict_ids <.> WpLet (binds `unionBags` bind)
1861 rhs = mkHsWrap co payload
1862 loc = instLocSpan inst_loc
1863 payload | isSingleton wanteds = HsVar (instToId (head wanteds))
1864 | otherwise = ExplicitTuple (map (L loc . HsVar . instToId) wanteds) Boxed
1866 -- If there are any irreds, we back off and return NoInstance
1867 ; return (ret_avails, GenInst implic_insts (L loc rhs))
1871 Note [Freeness and implications]
1872 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1873 It's hard to say when an implication constraint can be floated out. Consider
1874 forall {} Eq a => Foo [a]
1875 The (Foo [a]) doesn't mention any of the quantified variables, but it
1876 still might be partially satisfied by the (Eq a).
1878 There is a useful special case when it *is* easy to partition the
1879 constraints, namely when there are no 'givens'. Consider
1880 forall {a}. () => Bar b
1881 There are no 'givens', and so there is no reason to capture (Bar b).
1882 We can let it float out. But if there is even one constraint we
1883 must be much more careful:
1884 forall {a}. C a b => Bar (m b)
1885 because (C a b) might have a superclass (D b), from which we might
1886 deduce (Bar [b]) when m later gets instantiated to []. Ha!
1888 Here is an even more exotic example
1890 Now consider the constraint
1891 forall b. D Int b => C Int
1892 We can satisfy the (C Int) from the superclass of D, so we don't want
1893 to float the (C Int) out, even though it mentions no type variable in
1896 Note [Pruning the givens in an implication constraint]
1897 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1898 Suppose we are about to form the implication constraint
1899 forall tvs. Eq a => Ord b
1900 The (Eq a) cannot contribute to the (Ord b), because it has no access to
1901 the type variable 'b'. So we could filter out the (Eq a) from the givens.
1903 Doing so would be a bit tidier, but all the implication constraints get
1904 simplified away by the optimiser, so it's no great win. So I don't take
1905 advantage of that at the moment.
1907 If you do, BE CAREFUL of wobbly type variables.
1910 %************************************************************************
1912 Avails and AvailHow: the pool of evidence
1914 %************************************************************************
1918 data Avails = Avails !ImprovementDone !AvailEnv
1920 type ImprovementDone = Bool -- True <=> some unification has happened
1921 -- so some Irreds might now be reducible
1922 -- keys that are now
1924 type AvailEnv = FiniteMap Inst AvailHow
1926 = IsIrred -- Used for irreducible dictionaries,
1927 -- which are going to be lambda bound
1929 | Given TcId -- Used for dictionaries for which we have a binding
1930 -- e.g. those "given" in a signature
1932 | Rhs -- Used when there is a RHS
1933 (LHsExpr TcId) -- The RHS
1934 [Inst] -- Insts free in the RHS; we need these too
1936 instance Outputable Avails where
1939 pprAvails (Avails imp avails)
1940 = vcat [ ptext SLIT("Avails") <> (if imp then ptext SLIT("[improved]") else empty)
1941 , nest 2 (vcat [sep [ppr inst, nest 2 (equals <+> ppr avail)]
1942 | (inst,avail) <- fmToList avails ])]
1944 instance Outputable AvailHow where
1947 -------------------------
1948 pprAvail :: AvailHow -> SDoc
1949 pprAvail IsIrred = text "Irred"
1950 pprAvail (Given x) = text "Given" <+> ppr x
1951 pprAvail (Rhs rhs bs) = text "Rhs" <+> ppr rhs <+> braces (ppr bs)
1953 -------------------------
1954 extendAvailEnv :: AvailEnv -> Inst -> AvailHow -> AvailEnv
1955 extendAvailEnv env inst avail = addToFM env inst avail
1957 findAvailEnv :: AvailEnv -> Inst -> Maybe AvailHow
1958 findAvailEnv env wanted = lookupFM env wanted
1959 -- NB 1: the Ord instance of Inst compares by the class/type info
1960 -- *not* by unique. So
1961 -- d1::C Int == d2::C Int
1963 emptyAvails :: Avails
1964 emptyAvails = Avails False emptyFM
1966 findAvail :: Avails -> Inst -> Maybe AvailHow
1967 findAvail (Avails _ avails) wanted = findAvailEnv avails wanted
1969 elemAvails :: Inst -> Avails -> Bool
1970 elemAvails wanted (Avails _ avails) = wanted `elemFM` avails
1972 extendAvails :: Avails -> Inst -> AvailHow -> TcM Avails
1974 extendAvails avails@(Avails imp env) inst avail
1975 = do { imp1 <- tcImproveOne avails inst -- Do any improvement
1976 ; return (Avails (imp || imp1) (extendAvailEnv env inst avail)) }
1978 availsInsts :: Avails -> [Inst]
1979 availsInsts (Avails _ avails) = keysFM avails
1981 availsImproved (Avails imp _) = imp
1983 updateImprovement :: Avails -> Avails -> Avails
1984 -- (updateImprovement a1 a2) sets a1's improvement flag from a2
1985 updateImprovement (Avails _ avails1) (Avails imp2 _) = Avails imp2 avails1
1988 Extracting the bindings from a bunch of Avails.
1989 The bindings do *not* come back sorted in dependency order.
1990 We assume that they'll be wrapped in a big Rec, so that the
1991 dependency analyser can sort them out later
1994 extractResults :: Avails
1996 -> TcM ( TcDictBinds, -- Bindings
1997 [Inst]) -- Irreducible ones
1999 extractResults (Avails _ avails) wanteds
2000 = go avails emptyBag [] wanteds
2002 go :: AvailEnv -> TcDictBinds -> [Inst] -> [Inst]
2003 -> TcM (TcDictBinds, [Inst])
2004 go avails binds irreds []
2005 = returnM (binds, irreds)
2007 go avails binds irreds (w:ws)
2008 = case findAvailEnv avails w of
2009 Nothing -> pprTrace "Urk: extractResults" (ppr w) $
2010 go avails binds irreds ws
2012 Just IsIrred -> go (add_given avails w) binds (w:irreds) ws
2016 -> go avails binds irreds ws
2017 -- The sought Id can be one of the givens, via a superclass chain
2018 -- and then we definitely don't want to generate an x=x binding!
2021 -> go avails (addBind binds w (nlHsVar id)) irreds ws
2023 Just (Rhs rhs ws') -> go (add_given avails w) new_binds irreds (ws' ++ ws)
2025 new_binds = addBind binds w rhs
2027 add_given avails w = extendAvailEnv avails w (Given (instToId w))
2029 addBind binds inst rhs = binds `unionBags` unitBag (L (instSpan inst)
2030 (VarBind (instToId inst) rhs))
2034 Note [No superclasses for Stop]
2035 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2036 When we decide not to reduce an Inst -- the 'WhatToDo' --- we still
2037 add it to avails, so that any other equal Insts will be commoned up
2038 right here. However, we do *not* add superclasses. If we have
2041 but a is not bound here, then we *don't* want to derive dn from df
2042 here lest we lose sharing.
2045 addWanted :: WantSCs -> Avails -> Inst -> LHsExpr TcId -> [Inst] -> TcM Avails
2046 addWanted want_scs avails wanted rhs_expr wanteds
2047 = addAvailAndSCs want_scs avails wanted avail
2049 avail = Rhs rhs_expr wanteds
2051 addGiven :: Avails -> Inst -> TcM Avails
2052 addGiven avails given = addAvailAndSCs AddSCs avails given (Given (instToId given))
2053 -- Always add superclasses for 'givens'
2055 -- No ASSERT( not (given `elemAvails` avails) ) because in an instance
2056 -- decl for Ord t we can add both Ord t and Eq t as 'givens',
2057 -- so the assert isn't true
2059 addRefinedGiven :: Refinement -> ([Inst], Avails) -> Inst -> TcM ([Inst], Avails)
2060 addRefinedGiven reft (refined_givens, avails) given
2061 | isDict given -- We sometimes have 'given' methods, but they
2062 -- are always optional, so we can drop them
2063 , let pred = dictPred given
2064 , isRefineablePred pred -- See Note [ImplicInst rigidity]
2065 , Just (co, pred) <- refinePred reft pred
2066 = do { new_given <- newDictBndr (instLoc given) pred
2067 ; let rhs = L (instSpan given) $
2068 HsWrap (WpCo co) (HsVar (instToId given))
2069 ; avails <- addAvailAndSCs AddSCs avails new_given (Rhs rhs [given])
2070 ; return (new_given:refined_givens, avails) }
2071 -- ToDo: the superclasses of the original given all exist in Avails
2072 -- so we could really just cast them, but it's more awkward to do,
2073 -- and hopefully the optimiser will spot the duplicated work
2075 = return (refined_givens, avails)
2078 Note [ImplicInst rigidity]
2079 ~~~~~~~~~~~~~~~~~~~~~~~~~~
2081 C :: forall ab. (Eq a, Ord b) => b -> T a
2083 ...(case x of C v -> <body>)...
2085 From the case (where x::T ty) we'll get an implication constraint
2086 forall b. (Eq ty, Ord b) => <body-constraints>
2087 Now suppose <body-constraints> itself has an implication constraint
2089 forall c. <reft> => <payload>
2090 Then, we can certainly apply the refinement <reft> to the Ord b, becuase it is
2091 existential, but we probably should not apply it to the (Eq ty) because it may
2092 be wobbly. Hence the isRigidInst
2094 @Insts@ are ordered by their class/type info, rather than by their
2095 unique. This allows the context-reduction mechanism to use standard finite
2096 maps to do their stuff. It's horrible that this code is here, rather
2097 than with the Avails handling stuff in TcSimplify
2100 addIrred :: WantSCs -> Avails -> Inst -> TcM Avails
2101 addIrred want_scs avails irred = ASSERT2( not (irred `elemAvails` avails), ppr irred $$ ppr avails )
2102 addAvailAndSCs want_scs avails irred IsIrred
2104 addAvailAndSCs :: WantSCs -> Avails -> Inst -> AvailHow -> TcM Avails
2105 addAvailAndSCs want_scs avails inst avail
2106 | not (isClassDict inst) = extendAvails avails inst avail
2107 | NoSCs <- want_scs = extendAvails avails inst avail
2108 | otherwise = do { traceTc (text "addAvailAndSCs" <+> vcat [ppr inst, ppr deps])
2109 ; avails' <- extendAvails avails inst avail
2110 ; addSCs is_loop avails' inst }
2112 is_loop pred = any (`tcEqType` mkPredTy pred) dep_tys
2113 -- Note: this compares by *type*, not by Unique
2114 deps = findAllDeps (unitVarSet (instToId inst)) avail
2115 dep_tys = map idType (varSetElems deps)
2117 findAllDeps :: IdSet -> AvailHow -> IdSet
2118 -- Find all the Insts that this one depends on
2119 -- See Note [SUPERCLASS-LOOP 2]
2120 -- Watch out, though. Since the avails may contain loops
2121 -- (see Note [RECURSIVE DICTIONARIES]), so we need to track the ones we've seen so far
2122 findAllDeps so_far (Rhs _ kids) = foldl find_all so_far kids
2123 findAllDeps so_far other = so_far
2125 find_all :: IdSet -> Inst -> IdSet
2127 | kid_id `elemVarSet` so_far = so_far
2128 | Just avail <- findAvail avails kid = findAllDeps so_far' avail
2129 | otherwise = so_far'
2131 so_far' = extendVarSet so_far kid_id -- Add the new kid to so_far
2132 kid_id = instToId kid
2134 addSCs :: (TcPredType -> Bool) -> Avails -> Inst -> TcM Avails
2135 -- Add all the superclasses of the Inst to Avails
2136 -- The first param says "dont do this because the original thing
2137 -- depends on this one, so you'd build a loop"
2138 -- Invariant: the Inst is already in Avails.
2140 addSCs is_loop avails dict
2141 = ASSERT( isDict dict )
2142 do { sc_dicts <- newDictBndrs (instLoc dict) sc_theta'
2143 ; foldlM add_sc avails (zipEqual "add_scs" sc_dicts sc_sels) }
2145 (clas, tys) = getDictClassTys dict
2146 (tyvars, sc_theta, sc_sels, _) = classBigSig clas
2147 sc_theta' = substTheta (zipTopTvSubst tyvars tys) sc_theta
2149 add_sc avails (sc_dict, sc_sel)
2150 | is_loop (dictPred sc_dict) = return avails -- See Note [SUPERCLASS-LOOP 2]
2151 | is_given sc_dict = return avails
2152 | otherwise = do { avails' <- extendAvails avails sc_dict (Rhs sc_sel_rhs [dict])
2153 ; addSCs is_loop avails' sc_dict }
2155 sc_sel_rhs = L (instSpan dict) (HsWrap co_fn (HsVar sc_sel))
2156 co_fn = WpApp (instToId dict) <.> mkWpTyApps tys
2158 is_given :: Inst -> Bool
2159 is_given sc_dict = case findAvail avails sc_dict of
2160 Just (Given _) -> True -- Given is cheaper than superclass selection
2164 %************************************************************************
2166 \section{tcSimplifyTop: defaulting}
2168 %************************************************************************
2171 @tcSimplifyTop@ is called once per module to simplify all the constant
2172 and ambiguous Insts.
2174 We need to be careful of one case. Suppose we have
2176 instance Num a => Num (Foo a b) where ...
2178 and @tcSimplifyTop@ is given a constraint (Num (Foo x y)). Then it'll simplify
2179 to (Num x), and default x to Int. But what about y??
2181 It's OK: the final zonking stage should zap y to (), which is fine.
2185 tcSimplifyTop, tcSimplifyInteractive :: [Inst] -> TcM TcDictBinds
2186 tcSimplifyTop wanteds
2187 = tc_simplify_top doc False wanteds
2189 doc = text "tcSimplifyTop"
2191 tcSimplifyInteractive wanteds
2192 = tc_simplify_top doc True wanteds
2194 doc = text "tcSimplifyInteractive"
2196 -- The TcLclEnv should be valid here, solely to improve
2197 -- error message generation for the monomorphism restriction
2198 tc_simplify_top doc interactive wanteds
2199 = do { wanteds <- mapM zonkInst wanteds
2200 ; mapM_ zonkTopTyVar (varSetElems (tyVarsOfInsts wanteds))
2202 ; (irreds1, binds1) <- topCheckLoop doc wanteds
2204 ; if null irreds1 then
2207 -- OK, so there are some errors
2208 { -- Use the defaulting rules to do extra unification
2209 -- NB: irreds are already zonked
2210 ; dflags <- getDOpts
2211 ; disambiguate interactive dflags irreds1 -- Does unification
2212 ; (irreds2, binds2) <- topCheckLoop doc irreds1
2214 -- Deal with implicit parameter
2215 ; let (bad_ips, non_ips) = partition isIPDict irreds2
2216 (ambigs, others) = partition isTyVarDict non_ips
2218 ; topIPErrs bad_ips -- Can arise from f :: Int -> Int
2220 ; addNoInstanceErrs others
2221 ; addTopAmbigErrs ambigs
2223 ; return (binds1 `unionBags` binds2) }}
2226 If a dictionary constrains a type variable which is
2227 * not mentioned in the environment
2228 * and not mentioned in the type of the expression
2229 then it is ambiguous. No further information will arise to instantiate
2230 the type variable; nor will it be generalised and turned into an extra
2231 parameter to a function.
2233 It is an error for this to occur, except that Haskell provided for
2234 certain rules to be applied in the special case of numeric types.
2236 * at least one of its classes is a numeric class, and
2237 * all of its classes are numeric or standard
2238 then the type variable can be defaulted to the first type in the
2239 default-type list which is an instance of all the offending classes.
2241 So here is the function which does the work. It takes the ambiguous
2242 dictionaries and either resolves them (producing bindings) or
2243 complains. It works by splitting the dictionary list by type
2244 variable, and using @disambigOne@ to do the real business.
2246 @disambigOne@ assumes that its arguments dictionaries constrain all
2247 the same type variable.
2249 ADR Comment 20/6/94: I've changed the @CReturnable@ case to default to
2250 @()@ instead of @Int@. I reckon this is the Right Thing to do since
2251 the most common use of defaulting is code like:
2253 _ccall_ foo `seqPrimIO` bar
2255 Since we're not using the result of @foo@, the result if (presumably)
2259 disambiguate :: Bool -> DynFlags -> [Inst] -> TcM ()
2260 -- Just does unification to fix the default types
2261 -- The Insts are assumed to be pre-zonked
2262 disambiguate interactive dflags insts
2263 | null defaultable_groups
2264 = do { traceTc (text "disambigutate" <+> vcat [ppr unaries, ppr bad_tvs, ppr defaultable_groups])
2267 = do { -- Figure out what default types to use
2268 ; default_tys <- getDefaultTys extended_defaulting ovl_strings
2270 ; traceTc (text "disambigutate" <+> vcat [ppr unaries, ppr bad_tvs, ppr defaultable_groups])
2271 ; mapM_ (disambigGroup default_tys) defaultable_groups }
2273 extended_defaulting = interactive || dopt Opt_ExtendedDefaultRules dflags
2274 ovl_strings = dopt Opt_OverloadedStrings dflags
2276 unaries :: [(Inst,Class, TcTyVar)] -- (C tv) constraints
2277 bad_tvs :: TcTyVarSet -- Tyvars mentioned by *other* constraints
2278 (unaries, bad_tvs) = getDefaultableDicts insts
2280 -- Group by type variable
2281 defaultable_groups :: [[(Inst,Class,TcTyVar)]]
2282 defaultable_groups = filter defaultable_group (equivClasses cmp_tv unaries)
2283 cmp_tv (_,_,tv1) (_,_,tv2) = tv1 `compare` tv2
2285 defaultable_group :: [(Inst,Class,TcTyVar)] -> Bool
2286 defaultable_group ds@((_,_,tv):_)
2287 = isTyConableTyVar tv -- Note [Avoiding spurious errors]
2288 && not (tv `elemVarSet` bad_tvs)
2289 && defaultable_classes [c | (_,c,_) <- ds]
2290 defaultable_group [] = panic "defaultable_group"
2292 defaultable_classes clss
2293 | extended_defaulting = any isInteractiveClass clss
2294 | otherwise = all is_std_class clss && (any is_num_class clss)
2296 -- In interactive mode, or with -fextended-default-rules,
2297 -- we default Show a to Show () to avoid graututious errors on "show []"
2298 isInteractiveClass cls
2299 = is_num_class cls || (classKey cls `elem` [showClassKey, eqClassKey, ordClassKey])
2301 is_num_class cls = isNumericClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2302 -- is_num_class adds IsString to the standard numeric classes,
2303 -- when -foverloaded-strings is enabled
2305 is_std_class cls = isStandardClass cls || (ovl_strings && (cls `hasKey` isStringClassKey))
2306 -- Similarly is_std_class
2308 disambigGroup :: [Type] -- The default types
2309 -> [(Inst,Class,TcTyVar)] -- All standard classes of form (C a)
2310 -> TcM () -- Just does unification, to fix the default types
2312 disambigGroup default_tys dicts
2313 = try_default default_tys
2315 (_,_,tyvar) = head dicts -- Should be non-empty
2316 classes = [c | (_,c,_) <- dicts]
2318 try_default [] = return ()
2319 try_default (default_ty : default_tys)
2320 = tryTcLIE_ (try_default default_tys) $
2321 do { tcSimplifyDefault [mkClassPred clas [default_ty] | clas <- classes]
2322 -- This may fail; then the tryTcLIE_ kicks in
2323 -- Failure here is caused by there being no type in the
2324 -- default list which can satisfy all the ambiguous classes.
2325 -- For example, if Real a is reqd, but the only type in the
2326 -- default list is Int.
2328 -- After this we can't fail
2329 ; warnDefault dicts default_ty
2330 ; unifyType default_ty (mkTyVarTy tyvar) }
2333 getDefaultTys :: Bool -> Bool -> TcM [Type]
2334 getDefaultTys extended_deflts ovl_strings
2335 = do { mb_defaults <- getDeclaredDefaultTys
2336 ; case mb_defaults of {
2337 Just tys -> return tys ; -- User-supplied defaults
2340 -- No use-supplied default
2341 -- Use [Integer, Double], plus modifications
2342 { integer_ty <- tcMetaTy integerTyConName
2343 ; checkWiredInTyCon doubleTyCon
2344 ; string_ty <- tcMetaTy stringTyConName
2345 ; return (opt_deflt extended_deflts unitTy
2346 -- Note [Default unitTy]
2348 [integer_ty,doubleTy]
2350 opt_deflt ovl_strings string_ty) } } }
2352 opt_deflt True ty = [ty]
2353 opt_deflt False ty = []
2356 Note [Default unitTy]
2357 ~~~~~~~~~~~~~~~~~~~~~
2358 In interative mode (or with -fextended-default-rules) we add () as the first type we
2359 try when defaulting. This has very little real impact, except in the following case.
2361 Text.Printf.printf "hello"
2362 This has type (forall a. IO a); it prints "hello", and returns 'undefined'. We don't
2363 want the GHCi repl loop to try to print that 'undefined'. The neatest thing is to
2364 default the 'a' to (), rather than to Integer (which is what would otherwise happen;
2365 and then GHCi doesn't attempt to print the (). So in interactive mode, we add
2366 () to the list of defaulting types. See Trac #1200.
2368 Note [Avoiding spurious errors]
2369 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2370 When doing the unification for defaulting, we check for skolem
2371 type variables, and simply don't default them. For example:
2372 f = (*) -- Monomorphic
2373 g :: Num a => a -> a
2375 Here, we get a complaint when checking the type signature for g,
2376 that g isn't polymorphic enough; but then we get another one when
2377 dealing with the (Num a) context arising from f's definition;
2378 we try to unify a with Int (to default it), but find that it's
2379 already been unified with the rigid variable from g's type sig
2382 %************************************************************************
2384 \subsection[simple]{@Simple@ versions}
2386 %************************************************************************
2388 Much simpler versions when there are no bindings to make!
2390 @tcSimplifyThetas@ simplifies class-type constraints formed by
2391 @deriving@ declarations and when specialising instances. We are
2392 only interested in the simplified bunch of class/type constraints.
2394 It simplifies to constraints of the form (C a b c) where
2395 a,b,c are type variables. This is required for the context of
2396 instance declarations.
2399 tcSimplifyDeriv :: InstOrigin
2401 -> ThetaType -- Wanted
2402 -> TcM ThetaType -- Needed
2403 -- Given instance (wanted) => C inst_ty
2404 -- Simplify 'wanted' as much as possible
2405 -- The inst_ty is needed only for the termination check
2407 tcSimplifyDeriv orig tyvars theta
2408 = do { (tvs, _, tenv) <- tcInstTyVars tyvars
2409 -- The main loop may do unification, and that may crash if
2410 -- it doesn't see a TcTyVar, so we have to instantiate. Sigh
2411 -- ToDo: what if two of them do get unified?
2412 ; wanteds <- newDictBndrsO orig (substTheta tenv theta)
2413 ; (irreds, _) <- topCheckLoop doc wanteds
2415 ; let rev_env = zipTopTvSubst tvs (mkTyVarTys tyvars)
2416 simpl_theta = substTheta rev_env (map dictPred irreds)
2417 -- This reverse-mapping is a pain, but the result
2418 -- should mention the original TyVars not TcTyVars
2420 -- NB: the caller will further check the tv_dicts for
2421 -- legal instance-declaration form
2423 ; return simpl_theta }
2425 doc = ptext SLIT("deriving classes for a data type")
2430 @tcSimplifyDefault@ just checks class-type constraints, essentially;
2431 used with \tr{default} declarations. We are only interested in
2432 whether it worked or not.
2435 tcSimplifyDefault :: ThetaType -- Wanted; has no type variables in it
2438 tcSimplifyDefault theta
2439 = newDictBndrsO DefaultOrigin theta `thenM` \ wanteds ->
2440 topCheckLoop doc wanteds `thenM` \ (irreds, _) ->
2441 addNoInstanceErrs irreds `thenM_`
2447 doc = ptext SLIT("default declaration")
2451 %************************************************************************
2453 \section{Errors and contexts}
2455 %************************************************************************
2457 ToDo: for these error messages, should we note the location as coming
2458 from the insts, or just whatever seems to be around in the monad just
2462 groupErrs :: ([Inst] -> TcM ()) -- Deal with one group
2463 -> [Inst] -- The offending Insts
2465 -- Group together insts with the same origin
2466 -- We want to report them together in error messages
2468 groupErrs report_err []
2470 groupErrs report_err (inst:insts)
2471 = do_one (inst:friends) `thenM_`
2472 groupErrs report_err others
2475 -- (It may seem a bit crude to compare the error messages,
2476 -- but it makes sure that we combine just what the user sees,
2477 -- and it avoids need equality on InstLocs.)
2478 (friends, others) = partition is_friend insts
2479 loc_msg = showSDoc (pprInstLoc (instLoc inst))
2480 is_friend friend = showSDoc (pprInstLoc (instLoc friend)) == loc_msg
2481 do_one insts = addInstCtxt (instLoc (head insts)) (report_err insts)
2482 -- Add location and context information derived from the Insts
2484 -- Add the "arising from..." part to a message about bunch of dicts
2485 addInstLoc :: [Inst] -> Message -> Message
2486 addInstLoc insts msg = msg $$ nest 2 (pprInstArising (head insts))
2488 addTopIPErrs :: [Name] -> [Inst] -> TcM ()
2489 addTopIPErrs bndrs []
2491 addTopIPErrs bndrs ips
2492 = addErrTcM (tidy_env, mk_msg tidy_ips)
2494 (tidy_env, tidy_ips) = tidyInsts ips
2495 mk_msg ips = vcat [sep [ptext SLIT("Implicit parameters escape from"),
2496 nest 2 (ptext SLIT("the monomorphic top-level binding")
2497 <> plural bndrs <+> ptext SLIT("of")
2498 <+> pprBinders bndrs <> colon)],
2499 nest 2 (vcat (map ppr_ip ips)),
2501 ppr_ip ip = pprPred (dictPred ip) <+> pprInstArising ip
2503 topIPErrs :: [Inst] -> TcM ()
2505 = groupErrs report tidy_dicts
2507 (tidy_env, tidy_dicts) = tidyInsts dicts
2508 report dicts = addErrTcM (tidy_env, mk_msg dicts)
2509 mk_msg dicts = addInstLoc dicts (ptext SLIT("Unbound implicit parameter") <>
2510 plural tidy_dicts <+> pprDictsTheta tidy_dicts)
2512 addNoInstanceErrs :: [Inst] -- Wanted (can include implications)
2514 addNoInstanceErrs insts
2515 = do { let (tidy_env, tidy_insts) = tidyInsts insts
2516 ; reportNoInstances tidy_env Nothing tidy_insts }
2520 -> Maybe (InstLoc, [Inst]) -- Context
2521 -- Nothing => top level
2522 -- Just (d,g) => d describes the construct
2524 -> [Inst] -- What is wanted (can include implications)
2527 reportNoInstances tidy_env mb_what insts
2528 = groupErrs (report_no_instances tidy_env mb_what) insts
2530 report_no_instances tidy_env mb_what insts
2531 = do { inst_envs <- tcGetInstEnvs
2532 ; let (implics, insts1) = partition isImplicInst insts
2533 (insts2, overlaps) = partitionWith (check_overlap inst_envs) insts1
2534 ; traceTc (text "reportNoInstnces" <+> vcat
2535 [ppr implics, ppr insts1, ppr insts2])
2536 ; mapM_ complain_implic implics
2537 ; mapM_ (\doc -> addErrTcM (tidy_env, doc)) overlaps
2538 ; groupErrs complain_no_inst insts2 }
2540 complain_no_inst insts = addErrTcM (tidy_env, mk_no_inst_err insts)
2542 complain_implic inst -- Recurse!
2543 = reportNoInstances tidy_env
2544 (Just (tci_loc inst, tci_given inst))
2547 check_overlap :: (InstEnv,InstEnv) -> Inst -> Either Inst SDoc
2548 -- Right msg => overlap message
2549 -- Left inst => no instance
2550 check_overlap inst_envs wanted
2551 | not (isClassDict wanted) = Left wanted
2553 = case lookupInstEnv inst_envs clas tys of
2554 -- The case of exactly one match and no unifiers means
2555 -- a successful lookup. That can't happen here, becuase
2556 -- dicts only end up here if they didn't match in Inst.lookupInst
2558 ([m],[]) -> pprPanic "reportNoInstance" (ppr wanted)
2560 ([], _) -> Left wanted -- No match
2561 res -> Right (mk_overlap_msg wanted res)
2563 (clas,tys) = getDictClassTys wanted
2565 mk_overlap_msg dict (matches, unifiers)
2566 = vcat [ addInstLoc [dict] ((ptext SLIT("Overlapping instances for")
2567 <+> pprPred (dictPred dict))),
2568 sep [ptext SLIT("Matching instances") <> colon,
2569 nest 2 (vcat [pprInstances ispecs, pprInstances unifiers])],
2570 ASSERT( not (null matches) )
2571 if not (isSingleton matches)
2572 then -- Two or more matches
2574 else -- One match, plus some unifiers
2575 ASSERT( not (null unifiers) )
2576 parens (vcat [ptext SLIT("The choice depends on the instantiation of") <+>
2577 quotes (pprWithCommas ppr (varSetElems (tyVarsOfInst dict))),
2578 ptext SLIT("Use -fallow-incoherent-instances to use the first choice above")])]
2580 ispecs = [ispec | (_, ispec) <- matches]
2582 mk_no_inst_err insts
2583 | null insts = empty
2585 | Just (loc, givens) <- mb_what, -- Nested (type signatures, instance decls)
2586 not (isEmptyVarSet (tyVarsOfInsts insts))
2587 = vcat [ addInstLoc insts $
2588 sep [ ptext SLIT("Could not deduce") <+> pprDictsTheta insts
2589 , nest 2 $ ptext SLIT("from the context") <+> pprDictsTheta givens]
2590 , show_fixes (fix1 loc : fixes2) ]
2592 | otherwise -- Top level
2593 = vcat [ addInstLoc insts $
2594 ptext SLIT("No instance") <> plural insts
2595 <+> ptext SLIT("for") <+> pprDictsTheta insts
2596 , show_fixes fixes2 ]
2599 fix1 loc = sep [ ptext SLIT("add") <+> pprDictsTheta insts
2600 <+> ptext SLIT("to the context of"),
2601 nest 2 (ppr (instLocOrigin loc)) ]
2602 -- I'm not sure it helps to add the location
2603 -- nest 2 (ptext SLIT("at") <+> ppr (instLocSpan loc)) ]
2605 fixes2 | null instance_dicts = []
2606 | otherwise = [sep [ptext SLIT("add an instance declaration for"),
2607 pprDictsTheta instance_dicts]]
2608 instance_dicts = [d | d <- insts, isClassDict d, not (isTyVarDict d)]
2609 -- Insts for which it is worth suggesting an adding an instance declaration
2610 -- Exclude implicit parameters, and tyvar dicts
2612 show_fixes :: [SDoc] -> SDoc
2613 show_fixes [] = empty
2614 show_fixes (f:fs) = sep [ptext SLIT("Possible fix:"),
2615 nest 2 (vcat (f : map (ptext SLIT("or") <+>) fs))]
2617 addTopAmbigErrs dicts
2618 -- Divide into groups that share a common set of ambiguous tyvars
2619 = ifErrsM (return ()) $ -- Only report ambiguity if no other errors happened
2620 -- See Note [Avoiding spurious errors]
2621 mapM_ report (equivClasses cmp [(d, tvs_of d) | d <- tidy_dicts])
2623 (tidy_env, tidy_dicts) = tidyInsts dicts
2625 tvs_of :: Inst -> [TcTyVar]
2626 tvs_of d = varSetElems (tyVarsOfInst d)
2627 cmp (_,tvs1) (_,tvs2) = tvs1 `compare` tvs2
2629 report :: [(Inst,[TcTyVar])] -> TcM ()
2630 report pairs@((inst,tvs) : _) -- The pairs share a common set of ambiguous tyvars
2631 = mkMonomorphismMsg tidy_env tvs `thenM` \ (tidy_env, mono_msg) ->
2632 setSrcSpan (instSpan inst) $
2633 -- the location of the first one will do for the err message
2634 addErrTcM (tidy_env, msg $$ mono_msg)
2636 dicts = map fst pairs
2637 msg = sep [text "Ambiguous type variable" <> plural tvs <+>
2638 pprQuotedList tvs <+> in_msg,
2639 nest 2 (pprDictsInFull dicts)]
2640 in_msg = text "in the constraint" <> plural dicts <> colon
2641 report [] = panic "addTopAmbigErrs"
2644 mkMonomorphismMsg :: TidyEnv -> [TcTyVar] -> TcM (TidyEnv, Message)
2645 -- There's an error with these Insts; if they have free type variables
2646 -- it's probably caused by the monomorphism restriction.
2647 -- Try to identify the offending variable
2648 -- ASSUMPTION: the Insts are fully zonked
2649 mkMonomorphismMsg tidy_env inst_tvs
2650 = findGlobals (mkVarSet inst_tvs) tidy_env `thenM` \ (tidy_env, docs) ->
2651 returnM (tidy_env, mk_msg docs)
2653 mk_msg [] = ptext SLIT("Probable fix: add a type signature that fixes these type variable(s)")
2654 -- This happens in things like
2655 -- f x = show (read "foo")
2656 -- where monomorphism doesn't play any role
2657 mk_msg docs = vcat [ptext SLIT("Possible cause: the monomorphism restriction applied to the following:"),
2661 monomorphism_fix :: SDoc
2662 monomorphism_fix = ptext SLIT("Probable fix:") <+>
2663 (ptext SLIT("give these definition(s) an explicit type signature")
2664 $$ ptext SLIT("or use -fno-monomorphism-restriction"))
2666 warnDefault ups default_ty
2667 = doptM Opt_WarnTypeDefaults `thenM` \ warn_flag ->
2668 addInstCtxt (instLoc (head (dicts))) (warnTc warn_flag warn_msg)
2670 dicts = [d | (d,_,_) <- ups]
2673 (_, tidy_dicts) = tidyInsts dicts
2674 warn_msg = vcat [ptext SLIT("Defaulting the following constraint(s) to type") <+>
2675 quotes (ppr default_ty),
2676 pprDictsInFull tidy_dicts]
2678 reduceDepthErr n stack
2679 = vcat [ptext SLIT("Context reduction stack overflow; size =") <+> int n,
2680 ptext SLIT("Use -fcontext-stack=N to increase stack size to N"),
2681 nest 4 (pprStack stack)]
2683 pprStack stack = vcat (map pprInstInFull stack)