2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
4 \section[InstEnv]{Utilities for typechecking instance declarations}
6 The bits common to TcInstDcls and TcDeriv.
10 DFunId, ClsInstEnv, InstEnv,
12 emptyInstEnv, extendInstEnv,
13 lookupInstEnv, InstLookupResult(..),
14 classInstEnv, simpleDFunClassTyCon
17 #include "HsVersions.h"
19 import Class ( Class )
21 import VarSet ( TyVarSet, unionVarSet, mkVarSet )
22 import VarEnv ( TyVarSubstEnv )
23 import Maybes ( MaybeErr(..), returnMaB, failMaB, thenMaB, maybeToBool )
24 import Name ( getSrcLoc )
25 import Type ( Type, tyConAppTyCon,
26 splitSigmaTy, splitDFunTy, tyVarsOfTypes
29 import TyCon ( TyCon )
31 import Unify ( matchTys, unifyTyListsX )
32 import UniqFM ( UniqFM, lookupWithDefaultUFM, addToUFM, emptyUFM )
34 import ErrUtils ( Message )
39 %************************************************************************
41 \subsection{The key types}
43 %************************************************************************
48 type InstEnv = UniqFM ClsInstEnv -- Maps Class to instances for that class
50 type ClsInstEnv = [(TyVarSet, [Type], DFunId)] -- The instances for a particular class
52 simpleDFunClassTyCon :: DFunId -> (Class, TyCon)
53 simpleDFunClassTyCon dfun
56 (_,_,clas,[ty]) = splitDFunTy (idType dfun)
57 tycon = tyConAppTyCon ty
60 %************************************************************************
62 \subsection{Instance environments: InstEnv and ClsInstEnv}
64 %************************************************************************
66 The actual type declarations are in HscTypes.
69 emptyInstEnv :: InstEnv
70 emptyInstEnv = emptyUFM
72 classInstEnv :: InstEnv -> Class -> ClsInstEnv
73 classInstEnv env cls = lookupWithDefaultUFM env [] cls
76 A @ClsInstEnv@ lives inside a class, and identifies all the instances
77 of that class. The @Id@ inside a ClsInstEnv mapping is the dfun for
80 If class C maps to a list containing the item ([a,b], [t1,t2,t3], dfun), then
82 forall a b, C t1 t2 t3 can be constructed by dfun
84 or, to put it another way, we have
86 instance (...) => C t1 t2 t3, witnessed by dfun
88 There is an important consistency constraint in the elements of a ClsInstEnv:
90 * [a,b] must be a superset of the free vars of [t1,t2,t3]
92 * The dfun must itself be quantified over [a,b]
94 Thus, the @ClassInstEnv@ for @Eq@ might contain the following entry:
95 [a] ===> dfun_Eq_List :: forall a. Eq a => Eq [a]
96 The "a" in the pattern must be one of the forall'd variables in
101 Notes on overlapping instances
102 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
103 In some ClsInstEnvs, overlap is prohibited; that is, no pair of templates unify.
105 In others, overlap is permitted, but only in such a way that one can make
106 a unique choice when looking up. That is, overlap is only permitted if
107 one template matches the other, or vice versa. So this is ok:
115 If overlap is permitted, the list is kept most specific first, so that
116 the first lookup is the right choice.
119 For now we just use association lists.
121 \subsection{Avoiding a problem with overlapping}
123 Consider this little program:
126 class C a where c :: a
127 class C a => D a where d :: a
129 instance C Int where c = 17
130 instance D Int where d = 13
132 instance C a => C [a] where c = [c]
133 instance ({- C [a], -} D a) => D [a] where d = c
135 instance C [Int] where c = [37]
137 main = print (d :: [Int])
140 What do you think `main' prints (assuming we have overlapping instances, and
141 all that turned on)? Well, the instance for `D' at type `[a]' is defined to
142 be `c' at the same type, and we've got an instance of `C' at `[Int]', so the
143 answer is `[37]', right? (the generic `C [a]' instance shouldn't apply because
144 the `C [Int]' instance is more specific).
146 Ghc-4.04 gives `[37]', while ghc-4.06 gives `[17]', so 4.06 is wrong. That
147 was easy ;-) Let's just consult hugs for good measure. Wait - if I use old
148 hugs (pre-September99), I get `[17]', and stranger yet, if I use hugs98, it
149 doesn't even compile! What's going on!?
151 What hugs complains about is the `D [a]' instance decl.
154 ERROR "mj.hs" (line 10): Cannot build superclass instance
156 *** Context supplied : D a
157 *** Required superclass : C [a]
160 You might wonder what hugs is complaining about. It's saying that you
161 need to add `C [a]' to the context of the `D [a]' instance (as appears
162 in comments). But there's that `C [a]' instance decl one line above
163 that says that I can reduce the need for a `C [a]' instance to the
164 need for a `C a' instance, and in this case, I already have the
165 necessary `C a' instance (since we have `D a' explicitly in the
166 context, and `C' is a superclass of `D').
168 Unfortunately, the above reasoning indicates a premature commitment to the
169 generic `C [a]' instance. I.e., it prematurely rules out the more specific
170 instance `C [Int]'. This is the mistake that ghc-4.06 makes. The fix is to
171 add the context that hugs suggests (uncomment the `C [a]'), effectively
172 deferring the decision about which instance to use.
174 Now, interestingly enough, 4.04 has this same bug, but it's covered up
175 in this case by a little known `optimization' that was disabled in
176 4.06. Ghc-4.04 silently inserts any missing superclass context into
177 an instance declaration. In this case, it silently inserts the `C
178 [a]', and everything happens to work out.
180 (See `basicTypes/MkId:mkDictFunId' for the code in question. Search for
181 `Mark Jones', although Mark claims no credit for the `optimization' in
182 question, and would rather it stopped being called the `Mark Jones
185 So, what's the fix? I think hugs has it right. Here's why. Let's try
186 something else out with ghc-4.04. Let's add the following line:
191 Everyone raise their hand who thinks that `d :: [Int]' should give a
192 different answer from `d' :: [Int]'. Well, in ghc-4.04, it does. The
193 `optimization' only applies to instance decls, not to regular
194 bindings, giving inconsistent behavior.
196 Old hugs had this same bug. Here's how we fixed it: like GHC, the
197 list of instances for a given class is ordered, so that more specific
198 instances come before more generic ones. For example, the instance
199 list for C might contain:
200 ..., C Int, ..., C a, ...
201 When we go to look for a `C Int' instance we'll get that one first.
202 But what if we go looking for a `C b' (`b' is unconstrained)? We'll
203 pass the `C Int' instance, and keep going. But if `b' is
204 unconstrained, then we don't know yet if the more specific instance
205 will eventually apply. GHC keeps going, and matches on the generic `C
206 a'. The fix is to, at each step, check to see if there's a reverse
207 match, and if so, abort the search. This prevents hugs from
208 prematurely chosing a generic instance when a more specific one
214 @lookupInstEnv@ looks up in a @InstEnv@, using a one-way match. Since
215 the env is kept ordered, the first match must be the only one. The
216 thing we are looking up can have an arbitrary "flexi" part.
219 lookupInstEnv :: InstEnv -- The envt
220 -> Class -> [Type] -- Key
223 data InstLookupResult
224 = FoundInst -- There is a (template,substitution) pair
225 -- that makes the template match the key,
226 -- and no template is an instance of the key
229 | NoMatch Bool -- Boolean is true iff there is at least one
230 -- template that matches the key.
231 -- (but there are other template(s) that are
232 -- instances of the key, so we don't report
234 -- The NoMatch True case happens when we look up
236 -- in an InstEnv that has entries for
239 -- Then which we choose would depend on the way in which 'a'
240 -- is instantiated. So we say there is no match, but identify
241 -- it as ambiguous case in the hope of giving a better error msg.
242 -- See the notes above from Jeff Lewis
244 lookupInstEnv env key_cls key_tys
245 = find (classInstEnv env key_cls)
247 key_vars = tyVarsOfTypes key_tys
249 find [] = NoMatch False
250 find ((tpl_tyvars, tpl, val) : rest)
251 = case matchTys tpl_tyvars tpl key_tys of
253 case matchTys key_vars key_tys tpl of
255 Just (_, _) -> NoMatch (any_match rest)
256 Just (subst, leftovers) -> ASSERT( null leftovers )
259 any_match rest = or [ maybeToBool (matchTys tvs tpl key_tys)
260 | (tvs,tpl,_) <- rest
264 @addToClsInstEnv@ extends a @ClsInstEnv@, checking for overlaps.
266 A boolean flag controls overlap reporting.
268 True => overlap is permitted, but only if one template matches the other;
269 not if they unify but neither is
272 extendInstEnv :: DynFlags -> InstEnv -> [DFunId] -> (InstEnv, [Message])
273 -- Similar, but all we have is the DFuns
274 extendInstEnv dflags env infos
277 go env msgs [] = (env, msgs)
278 go env msgs (dfun:dfuns) = case addToInstEnv dflags env dfun of
279 Succeeded new_env -> go new_env msgs dfuns
280 Failed dfun' -> go env (msg:msgs) dfuns
282 msg = dupInstErr dfun dfun'
285 dupInstErr dfun1 dfun2
286 -- Overlapping/duplicate instances for given class; msg could be more glamourous
287 = hang (ptext SLIT("Duplicate or overlapping instance declarations:"))
288 2 (ppr_dfun dfun1 $$ ppr_dfun dfun2)
290 ppr_dfun dfun = ppr (getSrcLoc dfun) <> colon <+> ppr tau
292 (_,_,tau) = splitSigmaTy (idType dfun)
294 addToInstEnv :: DynFlags
296 -> MaybeErr InstEnv -- Success...
297 DFunId -- Failure: Offending overlap
299 addToInstEnv dflags inst_env dfun_id
300 = case insert_into (classInstEnv inst_env clas) of
301 Failed stuff -> Failed stuff
302 Succeeded new_env -> Succeeded (addToUFM inst_env clas new_env)
305 (ins_tvs, _, clas, ins_tys) = splitDFunTy (idType dfun_id)
307 ins_tv_set = mkVarSet ins_tvs
308 ins_item = (ins_tv_set, ins_tys, dfun_id)
310 insert_into [] = returnMaB [ins_item]
311 insert_into env@(cur_item@(tpl_tvs, tpl_tys, val) : rest)
314 -- (a) they are the same, or
315 -- (b) they unify, and any sort of overlap is prohibited,
316 -- (c) they unify but neither is more specific than t'other
318 || (unifiable && not (dopt Opt_AllowOverlappingInstances dflags))
319 || (unifiable && not (ins_item_more_specific || cur_item_more_specific))
322 -- New item is an instance of current item, so drop it here
323 | ins_item_more_specific = returnMaB (ins_item : env)
325 -- Otherwise carry on
326 | otherwise = insert_into rest `thenMaB` \ rest' ->
327 returnMaB (cur_item : rest')
329 unifiable = maybeToBool (unifyTyListsX (ins_tv_set `unionVarSet` tpl_tvs) tpl_tys ins_tys)
330 ins_item_more_specific = maybeToBool (matchTys tpl_tvs tpl_tys ins_tys)
331 cur_item_more_specific = maybeToBool (matchTys ins_tv_set ins_tys tpl_tys)
332 identical = ins_item_more_specific && cur_item_more_specific