-- (c) The University of Glasgow 2006 -- -- FamInstEnv: Type checked family instance declarations {-# LANGUAGE CPP, GADTs, ScopedTypeVariables, BangPatterns, TupleSections, DeriveFunctor #-} {-# OPTIONS_GHC -Wno-incomplete-record-updates #-} module GHC.Core.FamInstEnv ( FamInst(..), FamFlavor(..), famInstAxiom, famInstTyCon, famInstRHS, famInstsRepTyCons, famInstRepTyCon_maybe, dataFamInstRepTyCon, pprFamInst, pprFamInsts, mkImportedFamInst, FamInstEnvs, FamInstEnv, emptyFamInstEnv, emptyFamInstEnvs, extendFamInstEnv, extendFamInstEnvList, famInstEnvElts, famInstEnvSize, familyInstances, -- * CoAxioms mkCoAxBranch, mkBranchedCoAxiom, mkUnbranchedCoAxiom, mkSingleCoAxiom, mkNewTypeCoAxiom, FamInstMatch(..), lookupFamInstEnv, lookupFamInstEnvConflicts, lookupFamInstEnvByTyCon, isDominatedBy, apartnessCheck, -- Injectivity InjectivityCheckResult(..), lookupFamInstEnvInjectivityConflicts, injectiveBranches, -- Normalisation topNormaliseType, topNormaliseType_maybe, normaliseType, normaliseTcApp, topReduceTyFamApp_maybe, reduceTyFamApp_maybe, -- Flattening flattenTys ) where #include "HsVersions.h" import GHC.Prelude import GHC.Core.Unify import GHC.Core.Type as Type import GHC.Core.TyCo.Rep import GHC.Core.TyCon import GHC.Core.Coercion import GHC.Core.Coercion.Axiom import GHC.Types.Var.Set import GHC.Types.Var.Env import GHC.Types.Name import GHC.Types.Unique.DFM import GHC.Utils.Outputable import GHC.Data.Maybe import GHC.Core.Map import GHC.Types.Unique import GHC.Utils.Misc import GHC.Types.Var import GHC.Types.SrcLoc import GHC.Data.FastString import Control.Monad import Data.List( mapAccumL ) import Data.Array( Array, assocs ) {- ************************************************************************ * * Type checked family instance heads * * ************************************************************************ Note [FamInsts and CoAxioms] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ * CoAxioms and FamInsts are just like DFunIds and ClsInsts * A CoAxiom is a System-FC thing: it can relate any two types * A FamInst is a Haskell source-language thing, corresponding to a type/data family instance declaration. - The FamInst contains a CoAxiom, which is the evidence for the instance - The LHS of the CoAxiom is always of form F ty1 .. tyn where F is a type family -} data FamInst -- See Note [FamInsts and CoAxioms] = FamInst { fi_axiom :: CoAxiom Unbranched -- The new coercion axiom -- introduced by this family -- instance -- INVARIANT: apart from freshening (see below) -- fi_tvs = cab_tvs of the (single) axiom branch -- fi_cvs = cab_cvs ...ditto... -- fi_tys = cab_lhs ...ditto... -- fi_rhs = cab_rhs ...ditto... , fi_flavor :: FamFlavor -- Everything below here is a redundant, -- cached version of the two things above -- except that the TyVars are freshened , fi_fam :: Name -- Family name -- Used for "rough matching"; same idea as for class instances -- See Note [Rough-match field] in GHC.Core.InstEnv , fi_tcs :: [Maybe Name] -- Top of type args -- INVARIANT: fi_tcs = roughMatchTcs fi_tys -- Used for "proper matching"; ditto , fi_tvs :: [TyVar] -- Template tyvars for full match , fi_cvs :: [CoVar] -- Template covars for full match -- Like ClsInsts, these variables are always fresh -- See Note [Template tyvars are fresh] in GHC.Core.InstEnv , fi_tys :: [Type] -- The LHS type patterns -- May be eta-reduced; see Note [Eta reduction for data families] -- in GHC.Core.Coercion.Axiom , fi_rhs :: Type -- the RHS, with its freshened vars } data FamFlavor = SynFamilyInst -- A synonym family | DataFamilyInst TyCon -- A data family, with its representation TyCon {- Note [Arity of data families] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Data family instances might legitimately be over- or under-saturated. Under-saturation has two potential causes: U1) Eta reduction. See Note [Eta reduction for data families] in GHC.Core.Coercion.Axiom. U2) When the user has specified a return kind instead of written out patterns. Example: data family Sing (a :: k) data instance Sing :: Bool -> Type The data family tycon Sing has an arity of 2, the k and the a. But the data instance has only one pattern, Bool (standing in for k). This instance is equivalent to `data instance Sing (a :: Bool)`, but without the last pattern, we have an under-saturated data family instance. On its own, this example is not compelling enough to add support for under-saturation, but U1 makes this feature more compelling. Over-saturation is also possible: O1) If the data family's return kind is a type variable (see also #12369), an instance might legitimately have more arguments than the family. Example: data family Fix :: (Type -> k) -> k data instance Fix f = MkFix1 (f (Fix f)) data instance Fix f x = MkFix2 (f (Fix f x) x) In the first instance here, the k in the data family kind is chosen to be Type. In the second, it's (Type -> Type). However, we require that any over-saturation is eta-reducible. That is, we require that any extra patterns be bare unrepeated type variables; see Note [Eta reduction for data families] in GHC.Core.Coercion.Axiom. Accordingly, the FamInst is never over-saturated. Why can we allow such flexibility for data families but not for type families? Because data families can be decomposed -- that is, they are generative and injective. A Type family is neither and so always must be applied to all its arguments. -} -- Obtain the axiom of a family instance famInstAxiom :: FamInst -> CoAxiom Unbranched famInstAxiom = fi_axiom -- Split the left-hand side of the FamInst famInstSplitLHS :: FamInst -> (TyCon, [Type]) famInstSplitLHS (FamInst { fi_axiom = axiom, fi_tys = lhs }) = (coAxiomTyCon axiom, lhs) -- Get the RHS of the FamInst famInstRHS :: FamInst -> Type famInstRHS = fi_rhs -- Get the family TyCon of the FamInst famInstTyCon :: FamInst -> TyCon famInstTyCon = coAxiomTyCon . famInstAxiom -- Return the representation TyCons introduced by data family instances, if any famInstsRepTyCons :: [FamInst] -> [TyCon] famInstsRepTyCons fis = [tc | FamInst { fi_flavor = DataFamilyInst tc } <- fis] -- Extracts the TyCon for this *data* (or newtype) instance famInstRepTyCon_maybe :: FamInst -> Maybe TyCon famInstRepTyCon_maybe fi = case fi_flavor fi of DataFamilyInst tycon -> Just tycon SynFamilyInst -> Nothing dataFamInstRepTyCon :: FamInst -> TyCon dataFamInstRepTyCon fi = case fi_flavor fi of DataFamilyInst tycon -> tycon SynFamilyInst -> pprPanic "dataFamInstRepTyCon" (ppr fi) {- ************************************************************************ * * Pretty printing * * ************************************************************************ -} instance NamedThing FamInst where getName = coAxiomName . fi_axiom instance Outputable FamInst where ppr = pprFamInst pprFamInst :: FamInst -> SDoc -- Prints the FamInst as a family instance declaration -- NB: This function, FamInstEnv.pprFamInst, is used only for internal, -- debug printing. See GHC.Core.Ppr.TyThing.pprFamInst for printing for the user pprFamInst (FamInst { fi_flavor = flavor, fi_axiom = ax , fi_tvs = tvs, fi_tys = tys, fi_rhs = rhs }) = hang (ppr_tc_sort <+> text "instance" <+> pprCoAxBranchUser (coAxiomTyCon ax) (coAxiomSingleBranch ax)) 2 (whenPprDebug debug_stuff) where ppr_tc_sort = case flavor of SynFamilyInst -> text "type" DataFamilyInst tycon | isDataTyCon tycon -> text "data" | isNewTyCon tycon -> text "newtype" | isAbstractTyCon tycon -> text "data" | otherwise -> text "WEIRD" <+> ppr tycon debug_stuff = vcat [ text "Coercion axiom:" <+> ppr ax , text "Tvs:" <+> ppr tvs , text "LHS:" <+> ppr tys , text "RHS:" <+> ppr rhs ] pprFamInsts :: [FamInst] -> SDoc pprFamInsts finsts = vcat (map pprFamInst finsts) {- Note [Lazy axiom match] ~~~~~~~~~~~~~~~~~~~~~~~ It is Vitally Important that mkImportedFamInst is *lazy* in its axiom parameter. The axiom is loaded lazily, via a forkM, in GHC.IfaceToCore. Sometime later, mkImportedFamInst is called using that axiom. However, the axiom may itself depend on entities which are not yet loaded as of the time of the mkImportedFamInst. Thus, if mkImportedFamInst eagerly looks at the axiom, a dependency loop spontaneously appears and GHC hangs. The solution is simply for mkImportedFamInst never, ever to look inside of the axiom until everything else is good and ready to do so. We can assume that this readiness has been achieved when some other code pulls on the axiom in the FamInst. Thus, we pattern match on the axiom lazily (in the where clause, not in the parameter list) and we assert the consistency of names there also. -} -- Make a family instance representation from the information found in an -- interface file. In particular, we get the rough match info from the iface -- (instead of computing it here). mkImportedFamInst :: Name -- Name of the family -> [Maybe Name] -- Rough match info -> CoAxiom Unbranched -- Axiom introduced -> FamInst -- Resulting family instance mkImportedFamInst fam mb_tcs axiom = FamInst { fi_fam = fam, fi_tcs = mb_tcs, fi_tvs = tvs, fi_cvs = cvs, fi_tys = tys, fi_rhs = rhs, fi_axiom = axiom, fi_flavor = flavor } where -- See Note [Lazy axiom match] ~(CoAxBranch { cab_lhs = tys , cab_tvs = tvs , cab_cvs = cvs , cab_rhs = rhs }) = coAxiomSingleBranch axiom -- Derive the flavor for an imported FamInst rather disgustingly -- Maybe we should store it in the IfaceFamInst? flavor = case splitTyConApp_maybe rhs of Just (tc, _) | Just ax' <- tyConFamilyCoercion_maybe tc , ax' == axiom -> DataFamilyInst tc _ -> SynFamilyInst {- ************************************************************************ * * FamInstEnv * * ************************************************************************ Note [FamInstEnv] ~~~~~~~~~~~~~~~~~ A FamInstEnv maps a family name to the list of known instances for that family. The same FamInstEnv includes both 'data family' and 'type family' instances. Type families are reduced during type inference, but not data families; the user explains when to use a data family instance by using constructors and pattern matching. Nevertheless it is still useful to have data families in the FamInstEnv: - For finding overlaps and conflicts - For finding the representation type...see FamInstEnv.topNormaliseType and its call site in GHC.Core.Opt.Simplify - In standalone deriving instance Eq (T [Int]) we need to find the representation type for T [Int] Note [Varying number of patterns for data family axioms] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ For data families, the number of patterns may vary between instances. For example data family T a b data instance T Int a = T1 a | T2 data instance T Bool [a] = T3 a Then we get a data type for each instance, and an axiom: data TInt a = T1 a | T2 data TBoolList a = T3 a axiom ax7 :: T Int ~ TInt -- Eta-reduced axiom ax8 a :: T Bool [a] ~ TBoolList a These two axioms for T, one with one pattern, one with two; see Note [Eta reduction for data families] in GHC.Core.Coercion.Axiom Note [FamInstEnv determinism] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ We turn FamInstEnvs into a list in some places that don't directly affect the ABI. That happens in family consistency checks and when producing output for `:info`. Unfortunately that nondeterminism is nonlocal and it's hard to tell what it affects without following a chain of functions. It's also easy to accidentally make that nondeterminism affect the ABI. Furthermore the envs should be relatively small, so it should be free to use deterministic maps here. Testing with nofib and validate detected no difference between UniqFM and UniqDFM. See Note [Deterministic UniqFM]. -} -- Internally we sometimes index by Name instead of TyCon despite -- of what the type says. This is safe since -- getUnique (tyCon) == getUniqe (tcName tyCon) type FamInstEnv = UniqDFM TyCon FamilyInstEnv -- Maps a family to its instances -- See Note [FamInstEnv] -- See Note [FamInstEnv determinism] type FamInstEnvs = (FamInstEnv, FamInstEnv) -- External package inst-env, Home-package inst-env newtype FamilyInstEnv = FamIE [FamInst] -- The instances for a particular family, in any order instance Outputable FamilyInstEnv where ppr (FamIE fs) = text "FamIE" <+> vcat (map ppr fs) -- | Index a FamInstEnv by the tyCons name. toNameInstEnv :: FamInstEnv -> UniqDFM Name FamilyInstEnv toNameInstEnv = unsafeCastUDFMKey -- | Create a FamInstEnv from Name indices. fromNameInstEnv :: UniqDFM Name FamilyInstEnv -> FamInstEnv fromNameInstEnv = unsafeCastUDFMKey -- INVARIANTS: -- * The fs_tvs are distinct in each FamInst -- of a range value of the map (so we can safely unify them) emptyFamInstEnvs :: (FamInstEnv, FamInstEnv) emptyFamInstEnvs = (emptyFamInstEnv, emptyFamInstEnv) emptyFamInstEnv :: FamInstEnv emptyFamInstEnv = emptyUDFM famInstEnvElts :: FamInstEnv -> [FamInst] famInstEnvElts fi = [elt | FamIE elts <- eltsUDFM fi, elt <- elts] -- See Note [FamInstEnv determinism] famInstEnvSize :: FamInstEnv -> Int famInstEnvSize = nonDetStrictFoldUDFM (\(FamIE elt) sum -> sum + length elt) 0 -- It's OK to use nonDetStrictFoldUDFM here since we're just computing the -- size. familyInstances :: (FamInstEnv, FamInstEnv) -> TyCon -> [FamInst] familyInstances (pkg_fie, home_fie) fam = get home_fie ++ get pkg_fie where get env = case lookupUDFM env fam of Just (FamIE insts) -> insts Nothing -> [] extendFamInstEnvList :: FamInstEnv -> [FamInst] -> FamInstEnv extendFamInstEnvList inst_env fis = foldl' extendFamInstEnv inst_env fis extendFamInstEnv :: FamInstEnv -> FamInst -> FamInstEnv extendFamInstEnv inst_env ins_item@(FamInst {fi_fam = cls_nm}) = fromNameInstEnv $ addToUDFM_C add (toNameInstEnv inst_env) cls_nm (FamIE [ins_item]) where add (FamIE items) _ = FamIE (ins_item:items) {- ************************************************************************ * * Compatibility * * ************************************************************************ Note [Apartness] ~~~~~~~~~~~~~~~~ In dealing with closed type families, we must be able to check that one type will never reduce to another. This check is called /apartness/. The check is always between a target (which may be an arbitrary type) and a pattern. Here is how we do it: apart(target, pattern) = not (unify(flatten(target), pattern)) where flatten (implemented in flattenTys, below) converts all type-family applications into fresh variables. (See Note [Flattening].) Note [Compatibility] ~~~~~~~~~~~~~~~~~~~~ Two patterns are /compatible/ if either of the following conditions hold: 1) The patterns are apart. 2) The patterns unify with a substitution S, and their right hand sides equal under that substitution. For open type families, only compatible instances are allowed. For closed type families, the story is slightly more complicated. Consider the following: type family F a where F Int = Bool F a = Int g :: Show a => a -> F a g x = length (show x) Should that type-check? No. We need to allow for the possibility that 'a' might be Int and therefore 'F a' should be Bool. We can simplify 'F a' to Int only when we can be sure that 'a' is not Int. To achieve this, after finding a possible match within the equations, we have to go back to all previous equations and check that, under the substitution induced by the match, other branches are surely apart. (See Note [Apartness].) This is similar to what happens with class instance selection, when we need to guarantee that there is only a match and no unifiers. The exact algorithm is different here because the potentially-overlapping group is closed. As another example, consider this: type family G x where G Int = Bool G a = Double type family H y -- no instances Now, we want to simplify (G (H Char)). We can't, because (H Char) might later simplify to be Int. So, (G (H Char)) is stuck, for now. While everything above is quite sound, it isn't as expressive as we'd like. Consider this: type family J a where J Int = Int J a = a Can we simplify (J b) to b? Sure we can. Yes, the first equation matches if b is instantiated with Int, but the RHSs coincide there, so it's all OK. So, the rule is this: when looking up a branch in a closed type family, we find a branch that matches the target, but then we make sure that the target is apart from every previous *incompatible* branch. We don't check the branches that are compatible with the matching branch, because they are either irrelevant (clause 1 of compatible) or benign (clause 2 of compatible). Note [Compatibility of eta-reduced axioms] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ In newtype instances of data families we eta-reduce the axioms, See Note [Eta reduction for data families] in GHC.Core.Coercion.Axiom. This means that we sometimes need to test compatibility of two axioms that were eta-reduced to different degrees, e.g.: data family D a b c newtype instance D a Int c = DInt (Maybe a) -- D a Int ~ Maybe -- lhs = [a, Int] newtype instance D Bool Int Char = DIntChar Float -- D Bool Int Char ~ Float -- lhs = [Bool, Int, Char] These are obviously incompatible. We could detect this by saturating (eta-expanding) the shorter LHS with fresh tyvars until the lists are of equal length, but instead we can just remove the tail of the longer list, as those types will simply unify with the freshly introduced tyvars. By doing this, in case the LHS are unifiable, the yielded substitution won't mention the tyvars that appear in the tail we dropped off, and we might try to test equality RHSes of different kinds, but that's fine since this case occurs only for data families, where the RHS is a unique tycon and the equality fails anyway. -} -- See Note [Compatibility] compatibleBranches :: CoAxBranch -> CoAxBranch -> Bool compatibleBranches (CoAxBranch { cab_lhs = lhs1, cab_rhs = rhs1 }) (CoAxBranch { cab_lhs = lhs2, cab_rhs = rhs2 }) = let (commonlhs1, commonlhs2) = zipAndUnzip lhs1 lhs2 -- See Note [Compatibility of eta-reduced axioms] in case tcUnifyTysFG (const BindMe) commonlhs1 commonlhs2 of SurelyApart -> True Unifiable subst | Type.substTyAddInScope subst rhs1 `eqType` Type.substTyAddInScope subst rhs2 -> True _ -> False -- | Result of testing two type family equations for injectiviy. data InjectivityCheckResult = InjectivityAccepted -- ^ Either RHSs are distinct or unification of RHSs leads to unification of -- LHSs | InjectivityUnified CoAxBranch CoAxBranch -- ^ RHSs unify but LHSs don't unify under that substitution. Relevant for -- closed type families where equation after unification might be -- overlpapped (in which case it is OK if they don't unify). Constructor -- stores axioms after unification. -- | Check whether two type family axioms don't violate injectivity annotation. injectiveBranches :: [Bool] -> CoAxBranch -> CoAxBranch -> InjectivityCheckResult injectiveBranches injectivity ax1@(CoAxBranch { cab_lhs = lhs1, cab_rhs = rhs1 }) ax2@(CoAxBranch { cab_lhs = lhs2, cab_rhs = rhs2 }) -- See Note [Verifying injectivity annotation], case 1. = let getInjArgs = filterByList injectivity in case tcUnifyTyWithTFs True rhs1 rhs2 of -- True = two-way pre-unification Nothing -> InjectivityAccepted -- RHS are different, so equations are injective. -- This is case 1A from Note [Verifying injectivity annotation] Just subst -> -- RHS unify under a substitution let lhs1Subst = Type.substTys subst (getInjArgs lhs1) lhs2Subst = Type.substTys subst (getInjArgs lhs2) -- If LHSs are equal under the substitution used for RHSs then this pair -- of equations does not violate injectivity annotation. If LHSs are not -- equal under that substitution then this pair of equations violates -- injectivity annotation, but for closed type families it still might -- be the case that one LHS after substitution is unreachable. in if eqTypes lhs1Subst lhs2Subst -- check case 1B1 from Note. then InjectivityAccepted else InjectivityUnified ( ax1 { cab_lhs = Type.substTys subst lhs1 , cab_rhs = Type.substTy subst rhs1 }) ( ax2 { cab_lhs = Type.substTys subst lhs2 , cab_rhs = Type.substTy subst rhs2 }) -- payload of InjectivityUnified used only for check 1B2, only -- for closed type families -- takes a CoAxiom with unknown branch incompatibilities and computes -- the compatibilities -- See Note [Storing compatibility] in GHC.Core.Coercion.Axiom computeAxiomIncomps :: [CoAxBranch] -> [CoAxBranch] computeAxiomIncomps branches = snd (mapAccumL go [] branches) where go :: [CoAxBranch] -> CoAxBranch -> ([CoAxBranch], CoAxBranch) go prev_brs cur_br = (cur_br : prev_brs, new_br) where new_br = cur_br { cab_incomps = mk_incomps prev_brs cur_br } mk_incomps :: [CoAxBranch] -> CoAxBranch -> [CoAxBranch] mk_incomps prev_brs cur_br = filter (not . compatibleBranches cur_br) prev_brs {- ************************************************************************ * * Constructing axioms These functions are here because tidyType / tcUnifyTysFG are not available in GHC.Core.Coercion.Axiom Also computeAxiomIncomps is too sophisticated for CoAxiom * * ************************************************************************ Note [Tidy axioms when we build them] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Like types and classes, we build axioms fully quantified over all their variables, and tidy them when we build them. For example, we print out axioms and don't want to print stuff like F k k a b = ... Instead we must tidy those kind variables. See #7524. We could instead tidy when we print, but that makes it harder to get things like injectivity errors to come out right. Danger of Type family equation violates injectivity annotation. Kind variable ‘k’ cannot be inferred from the right-hand side. In the type family equation: PolyKindVars @[k1] @[k2] ('[] @k1) = '[] @k2 Note [Always number wildcard types in CoAxBranch] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Consider the following example (from the DataFamilyInstanceLHS test case): data family Sing (a :: k) data instance Sing (_ :: MyKind) where SingA :: Sing A SingB :: Sing B If we're not careful during tidying, then when this program is compiled with -ddump-types, we'll get the following information: COERCION AXIOMS axiom DataFamilyInstanceLHS.D:R:SingMyKind_0 :: Sing _ = DataFamilyInstanceLHS.R:SingMyKind_ _ It's misleading to have a wildcard type appearing on the RHS like that. To avoid this issue, when building a CoAxiom (which is what eventually gets printed above), we tidy all the variables in an env that already contains '_'. Thus, any variable named '_' will be renamed, giving us the nicer output here: COERCION AXIOMS axiom DataFamilyInstanceLHS.D:R:SingMyKind_0 :: Sing _1 = DataFamilyInstanceLHS.R:SingMyKind_ _1 Which is at least legal syntax. See also Note [CoAxBranch type variables] in GHC.Core.Coercion.Axiom; note that we are tidying (changing OccNames only), not freshening, in accordance with that Note. -} -- all axiom roles are Nominal, as this is only used with type families mkCoAxBranch :: [TyVar] -- original, possibly stale, tyvars -> [TyVar] -- Extra eta tyvars -> [CoVar] -- possibly stale covars -> [Type] -- LHS patterns -> Type -- RHS -> [Role] -> SrcSpan -> CoAxBranch mkCoAxBranch tvs eta_tvs cvs lhs rhs roles loc = CoAxBranch { cab_tvs = tvs' , cab_eta_tvs = eta_tvs' , cab_cvs = cvs' , cab_lhs = tidyTypes env lhs , cab_roles = roles , cab_rhs = tidyType env rhs , cab_loc = loc , cab_incomps = placeHolderIncomps } where (env1, tvs') = tidyVarBndrs init_tidy_env tvs (env2, eta_tvs') = tidyVarBndrs env1 eta_tvs (env, cvs') = tidyVarBndrs env2 cvs -- See Note [Tidy axioms when we build them] -- See also Note [CoAxBranch type variables] in GHC.Core.Coercion.Axiom init_occ_env = initTidyOccEnv [mkTyVarOcc "_"] init_tidy_env = mkEmptyTidyEnv init_occ_env -- See Note [Always number wildcard types in CoAxBranch] -- all of the following code is here to avoid mutual dependencies with -- Coercion mkBranchedCoAxiom :: Name -> TyCon -> [CoAxBranch] -> CoAxiom Branched mkBranchedCoAxiom ax_name fam_tc branches = CoAxiom { co_ax_unique = nameUnique ax_name , co_ax_name = ax_name , co_ax_tc = fam_tc , co_ax_role = Nominal , co_ax_implicit = False , co_ax_branches = manyBranches (computeAxiomIncomps branches) } mkUnbranchedCoAxiom :: Name -> TyCon -> CoAxBranch -> CoAxiom Unbranched mkUnbranchedCoAxiom ax_name fam_tc branch = CoAxiom { co_ax_unique = nameUnique ax_name , co_ax_name = ax_name , co_ax_tc = fam_tc , co_ax_role = Nominal , co_ax_implicit = False , co_ax_branches = unbranched (branch { cab_incomps = [] }) } mkSingleCoAxiom :: Role -> Name -> [TyVar] -> [TyVar] -> [CoVar] -> TyCon -> [Type] -> Type -> CoAxiom Unbranched -- Make a single-branch CoAxiom, including making the branch itself -- Used for both type family (Nominal) and data family (Representational) -- axioms, hence passing in the Role mkSingleCoAxiom role ax_name tvs eta_tvs cvs fam_tc lhs_tys rhs_ty = CoAxiom { co_ax_unique = nameUnique ax_name , co_ax_name = ax_name , co_ax_tc = fam_tc , co_ax_role = role , co_ax_implicit = False , co_ax_branches = unbranched (branch { cab_incomps = [] }) } where branch = mkCoAxBranch tvs eta_tvs cvs lhs_tys rhs_ty (map (const Nominal) tvs) (getSrcSpan ax_name) -- | Create a coercion constructor (axiom) suitable for the given -- newtype 'TyCon'. The 'Name' should be that of a new coercion -- 'CoAxiom', the 'TyVar's the arguments expected by the @newtype@ and -- the type the appropriate right hand side of the @newtype@, with -- the free variables a subset of those 'TyVar's. mkNewTypeCoAxiom :: Name -> TyCon -> [TyVar] -> [Role] -> Type -> CoAxiom Unbranched mkNewTypeCoAxiom name tycon tvs roles rhs_ty = CoAxiom { co_ax_unique = nameUnique name , co_ax_name = name , co_ax_implicit = True -- See Note [Implicit axioms] in GHC.Core.TyCon , co_ax_role = Representational , co_ax_tc = tycon , co_ax_branches = unbranched (branch { cab_incomps = [] }) } where branch = mkCoAxBranch tvs [] [] (mkTyVarTys tvs) rhs_ty roles (getSrcSpan name) {- ************************************************************************ * * Looking up a family instance * * ************************************************************************ @lookupFamInstEnv@ looks up in a @FamInstEnv@, using a one-way match. Multiple matches are only possible in case of type families (not data families), and then, it doesn't matter which match we choose (as the instances are guaranteed confluent). We return the matching family instances and the type instance at which it matches. For example, if we lookup 'T [Int]' and have a family instance data instance T [a] = .. desugared to data :R42T a = .. coe :Co:R42T a :: T [a] ~ :R42T a we return the matching instance '(FamInst{.., fi_tycon = :R42T}, Int)'. -} -- when matching a type family application, we get a FamInst, -- and the list of types the axiom should be applied to data FamInstMatch = FamInstMatch { fim_instance :: FamInst , fim_tys :: [Type] , fim_cos :: [Coercion] } -- See Note [Over-saturated matches] instance Outputable FamInstMatch where ppr (FamInstMatch { fim_instance = inst , fim_tys = tys , fim_cos = cos }) = text "match with" <+> parens (ppr inst) <+> ppr tys <+> ppr cos lookupFamInstEnvByTyCon :: FamInstEnvs -> TyCon -> [FamInst] lookupFamInstEnvByTyCon (pkg_ie, home_ie) fam_tc = get pkg_ie ++ get home_ie where get ie = case lookupUDFM ie fam_tc of Nothing -> [] Just (FamIE fis) -> fis lookupFamInstEnv :: FamInstEnvs -> TyCon -> [Type] -- What we are looking for -> [FamInstMatch] -- Successful matches -- Precondition: the tycon is saturated (or over-saturated) lookupFamInstEnv = lookup_fam_inst_env match where match _ _ tpl_tys tys = tcMatchTys tpl_tys tys lookupFamInstEnvConflicts :: FamInstEnvs -> FamInst -- Putative new instance -> [FamInstMatch] -- Conflicting matches (don't look at the fim_tys field) -- E.g. when we are about to add -- f : type instance F [a] = a->a -- we do (lookupFamInstConflicts f [b]) -- to find conflicting matches -- -- Precondition: the tycon is saturated (or over-saturated) lookupFamInstEnvConflicts envs fam_inst@(FamInst { fi_axiom = new_axiom }) = lookup_fam_inst_env my_unify envs fam tys where (fam, tys) = famInstSplitLHS fam_inst -- In example above, fam tys' = F [b] my_unify (FamInst { fi_axiom = old_axiom }) tpl_tvs tpl_tys _ = ASSERT2( tyCoVarsOfTypes tys `disjointVarSet` tpl_tvs, (ppr fam <+> ppr tys) $$ (ppr tpl_tvs <+> ppr tpl_tys) ) -- Unification will break badly if the variables overlap -- They shouldn't because we allocate separate uniques for them if compatibleBranches (coAxiomSingleBranch old_axiom) new_branch then Nothing else Just noSubst -- Note [Family instance overlap conflicts] noSubst = panic "lookupFamInstEnvConflicts noSubst" new_branch = coAxiomSingleBranch new_axiom -------------------------------------------------------------------------------- -- Type family injectivity checking bits -- -------------------------------------------------------------------------------- {- Note [Verifying injectivity annotation] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Injectivity means that the RHS of a type family uniquely determines the LHS (see Note [Type inference for type families with injectivity]). The user informs us about injectivity using an injectivity annotation and it is GHC's task to verify that this annotation is correct w.r.t. type family equations. Whenever we see a new equation of a type family we need to make sure that adding this equation to the already known equations of a type family does not violate the injectivity annotation supplied by the user (see Note [Injectivity annotation]). Of course if the type family has no injectivity annotation then no check is required. But if a type family has injectivity annotation we need to make sure that the following conditions hold: 1. For each pair of *different* equations of a type family, one of the following conditions holds: A: RHSs are different. (Check done in GHC.Core.FamInstEnv.injectiveBranches) B1: OPEN TYPE FAMILIES: If the RHSs can be unified under some substitution then it must be possible to unify the LHSs under the same substitution. Example: type family FunnyId a = r | r -> a type instance FunnyId Int = Int type instance FunnyId a = a RHSs of these two equations unify under [ a |-> Int ] substitution. Under this substitution LHSs are equal therefore these equations don't violate injectivity annotation. (Check done in GHC.Core.FamInstEnv.injectiveBranches) B2: CLOSED TYPE FAMILIES: If the RHSs can be unified under some substitution then either the LHSs unify under the same substitution or the LHS of the latter equation is overlapped by earlier equations. Example 1: type family SwapIntChar a = r | r -> a where SwapIntChar Int = Char SwapIntChar Char = Int SwapIntChar a = a Say we are checking the last two equations. RHSs unify under [ a |-> Int ] substitution but LHSs don't. So we apply the substitution to LHS of last equation and check whether it is overlapped by any of previous equations. Since it is overlapped by the first equation we conclude that pair of last two equations does not violate injectivity annotation. (Check done in GHC.Tc.Validity.checkValidCoAxiom#gather_conflicts) A special case of B is when RHSs unify with an empty substitution ie. they are identical. If any of the above two conditions holds we conclude that the pair of equations does not violate injectivity annotation. But if we find a pair of equations where neither of the above holds we report that this pair violates injectivity annotation because for a given RHS we don't have a unique LHS. (Note that (B) actually implies (A).) Note that we only take into account these LHS patterns that were declared as injective. 2. If an RHS of a type family equation is a bare type variable then all LHS variables (including implicit kind variables) also have to be bare. In other words, this has to be a sole equation of that type family and it has to cover all possible patterns. So for example this definition will be rejected: type family W1 a = r | r -> a type instance W1 [a] = a If it were accepted we could call `W1 [W1 Int]`, which would reduce to `W1 Int` and then by injectivity we could conclude that `[W1 Int] ~ Int`, which is bogus. Checked FamInst.bareTvInRHSViolated. 3. If the RHS of a type family equation is a type family application then the type family is rejected as not injective. This is checked by FamInst.isTFHeaded. 4. If a LHS type variable that is declared as injective is not mentioned in an injective position in the RHS then the type family is rejected as not injective. "Injective position" means either an argument to a type constructor or argument to a type family on injective position. There are subtleties here. See Note [Coverage condition for injective type families] in GHC.Tc.Instance.Family. Check (1) must be done for all family instances (transitively) imported. Other checks (2-4) should be done just for locally written equations, as they are checks involving just a single equation, not about interactions. Doing the other checks for imported equations led to #17405, as the behavior of check (4) depends on -XUndecidableInstances (see Note [Coverage condition for injective type families] in FamInst), which may vary between modules. See also Note [Injective type families] in GHC.Core.TyCon -} -- | Check whether an open type family equation can be added to already existing -- instance environment without causing conflicts with supplied injectivity -- annotations. Returns list of conflicting axioms (type instance -- declarations). lookupFamInstEnvInjectivityConflicts :: [Bool] -- injectivity annotation for this type family instance -- INVARIANT: list contains at least one True value -> FamInstEnvs -- all type instances seens so far -> FamInst -- new type instance that we're checking -> [CoAxBranch] -- conflicting instance declarations lookupFamInstEnvInjectivityConflicts injList (pkg_ie, home_ie) fam_inst@(FamInst { fi_axiom = new_axiom }) -- See Note [Verifying injectivity annotation]. This function implements -- check (1.B1) for open type families described there. = lookup_inj_fam_conflicts home_ie ++ lookup_inj_fam_conflicts pkg_ie where fam = famInstTyCon fam_inst new_branch = coAxiomSingleBranch new_axiom -- filtering function used by `lookup_inj_fam_conflicts` to check whether -- a pair of equations conflicts with the injectivity annotation. isInjConflict (FamInst { fi_axiom = old_axiom }) | InjectivityAccepted <- injectiveBranches injList (coAxiomSingleBranch old_axiom) new_branch = False -- no conflict | otherwise = True lookup_inj_fam_conflicts ie | isOpenFamilyTyCon fam, Just (FamIE insts) <- lookupUDFM ie fam = map (coAxiomSingleBranch . fi_axiom) $ filter isInjConflict insts | otherwise = [] -------------------------------------------------------------------------------- -- Type family overlap checking bits -- -------------------------------------------------------------------------------- {- Note [Family instance overlap conflicts] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ - In the case of data family instances, any overlap is fundamentally a conflict (as these instances imply injective type mappings). - In the case of type family instances, overlap is admitted as long as the right-hand sides of the overlapping rules coincide under the overlap substitution. eg type instance F a Int = a type instance F Int b = b These two overlap on (F Int Int) but then both RHSs are Int, so all is well. We require that they are syntactically equal; anything else would be difficult to test for at this stage. -} ------------------------------------------------------------ -- Might be a one-way match or a unifier type MatchFun = FamInst -- The FamInst template -> TyVarSet -> [Type] -- fi_tvs, fi_tys of that FamInst -> [Type] -- Target to match against -> Maybe TCvSubst lookup_fam_inst_env' -- The worker, local to this module :: MatchFun -> FamInstEnv -> TyCon -> [Type] -- What we are looking for -> [FamInstMatch] lookup_fam_inst_env' match_fun ie fam match_tys | isOpenFamilyTyCon fam , Just (FamIE insts) <- lookupUDFM ie fam = find insts -- The common case | otherwise = [] where find [] = [] find (item@(FamInst { fi_tcs = mb_tcs, fi_tvs = tpl_tvs, fi_cvs = tpl_cvs , fi_tys = tpl_tys }) : rest) -- Fast check for no match, uses the "rough match" fields | instanceCantMatch rough_tcs mb_tcs = find rest -- Proper check | Just subst <- match_fun item (mkVarSet tpl_tvs) tpl_tys match_tys1 = (FamInstMatch { fim_instance = item , fim_tys = substTyVars subst tpl_tvs `chkAppend` match_tys2 , fim_cos = ASSERT( all (isJust . lookupCoVar subst) tpl_cvs ) substCoVars subst tpl_cvs }) : find rest -- No match => try next | otherwise = find rest where (rough_tcs, match_tys1, match_tys2) = split_tys tpl_tys -- Precondition: the tycon is saturated (or over-saturated) -- Deal with over-saturation -- See Note [Over-saturated matches] split_tys tpl_tys | isTypeFamilyTyCon fam = pre_rough_split_tys | otherwise = let (match_tys1, match_tys2) = splitAtList tpl_tys match_tys rough_tcs = roughMatchTcs match_tys1 in (rough_tcs, match_tys1, match_tys2) (pre_match_tys1, pre_match_tys2) = splitAt (tyConArity fam) match_tys pre_rough_split_tys = (roughMatchTcs pre_match_tys1, pre_match_tys1, pre_match_tys2) lookup_fam_inst_env -- The worker, local to this module :: MatchFun -> FamInstEnvs -> TyCon -> [Type] -- What we are looking for -> [FamInstMatch] -- Successful matches -- Precondition: the tycon is saturated (or over-saturated) lookup_fam_inst_env match_fun (pkg_ie, home_ie) fam tys = lookup_fam_inst_env' match_fun home_ie fam tys ++ lookup_fam_inst_env' match_fun pkg_ie fam tys {- Note [Over-saturated matches] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ It's ok to look up an over-saturated type constructor. E.g. type family F a :: * -> * type instance F (a,b) = Either (a->b) The type instance gives rise to a newtype TyCon (at a higher kind which you can't do in Haskell!): newtype FPair a b = FP (Either (a->b)) Then looking up (F (Int,Bool) Char) will return a FamInstMatch (FPair, [Int,Bool,Char]) The "extra" type argument [Char] just stays on the end. We handle data families and type families separately here: * For type families, all instances of a type family must have the same arity, so we can precompute the split between the match_tys and the overflow tys. This is done in pre_rough_split_tys. * For data family instances, though, we need to re-split for each instance, because the breakdown might be different for each instance. Why? Because of eta reduction; see Note [Eta reduction for data families] in GHC.Core.Coercion.Axiom. -} -- checks if one LHS is dominated by a list of other branches -- in other words, if an application would match the first LHS, it is guaranteed -- to match at least one of the others. The RHSs are ignored. -- This algorithm is conservative: -- True -> the LHS is definitely covered by the others -- False -> no information -- It is currently (Oct 2012) used only for generating errors for -- inaccessible branches. If these errors go unreported, no harm done. -- This is defined here to avoid a dependency from CoAxiom to Unify isDominatedBy :: CoAxBranch -> [CoAxBranch] -> Bool isDominatedBy branch branches = or $ map match branches where lhs = coAxBranchLHS branch match (CoAxBranch { cab_lhs = tys }) = isJust $ tcMatchTys tys lhs {- ************************************************************************ * * Choosing an axiom application * * ************************************************************************ The lookupFamInstEnv function does a nice job for *open* type families, but we also need to handle closed ones when normalising a type: -} reduceTyFamApp_maybe :: FamInstEnvs -> Role -- Desired role of result coercion -> TyCon -> [Type] -> Maybe (Coercion, Type) -- Attempt to do a *one-step* reduction of a type-family application -- but *not* newtypes -- Works on type-synonym families always; data-families only if -- the role we seek is representational -- It does *not* normalise the type arguments first, so this may not -- go as far as you want. If you want normalised type arguments, -- use topReduceTyFamApp_maybe -- -- The TyCon can be oversaturated. -- Works on both open and closed families -- -- Always returns a *homogeneous* coercion -- type family reductions are always -- homogeneous reduceTyFamApp_maybe envs role tc tys | Phantom <- role = Nothing | case role of Representational -> isOpenFamilyTyCon tc _ -> isOpenTypeFamilyTyCon tc -- If we seek a representational coercion -- (e.g. the call in topNormaliseType_maybe) then we can -- unwrap data families as well as type-synonym families; -- otherwise only type-synonym families , FamInstMatch { fim_instance = FamInst { fi_axiom = ax } , fim_tys = inst_tys , fim_cos = inst_cos } : _ <- lookupFamInstEnv envs tc tys -- NB: Allow multiple matches because of compatible overlap = let co = mkUnbranchedAxInstCo role ax inst_tys inst_cos ty = coercionRKind co in Just (co, ty) | Just ax <- isClosedSynFamilyTyConWithAxiom_maybe tc , Just (ind, inst_tys, inst_cos) <- chooseBranch ax tys = let co = mkAxInstCo role ax ind inst_tys inst_cos ty = coercionRKind co in Just (co, ty) | Just ax <- isBuiltInSynFamTyCon_maybe tc , Just (coax,ts,ty) <- sfMatchFam ax tys = let co = mkAxiomRuleCo coax (zipWith mkReflCo (coaxrAsmpRoles coax) ts) in Just (co, ty) | otherwise = Nothing -- The axiom can be oversaturated. (Closed families only.) chooseBranch :: CoAxiom Branched -> [Type] -> Maybe (BranchIndex, [Type], [Coercion]) -- found match, with args chooseBranch axiom tys = do { let num_pats = coAxiomNumPats axiom (target_tys, extra_tys) = splitAt num_pats tys branches = coAxiomBranches axiom ; (ind, inst_tys, inst_cos) <- findBranch (unMkBranches branches) target_tys ; return ( ind, inst_tys `chkAppend` extra_tys, inst_cos ) } -- The axiom must *not* be oversaturated findBranch :: Array BranchIndex CoAxBranch -> [Type] -> Maybe (BranchIndex, [Type], [Coercion]) -- coercions relate requested types to returned axiom LHS at role N findBranch branches target_tys = foldr go Nothing (assocs branches) where go :: (BranchIndex, CoAxBranch) -> Maybe (BranchIndex, [Type], [Coercion]) -> Maybe (BranchIndex, [Type], [Coercion]) go (index, branch) other = let (CoAxBranch { cab_tvs = tpl_tvs, cab_cvs = tpl_cvs , cab_lhs = tpl_lhs , cab_incomps = incomps }) = branch in_scope = mkInScopeSet (unionVarSets $ map (tyCoVarsOfTypes . coAxBranchLHS) incomps) -- See Note [Flattening] below flattened_target = flattenTys in_scope target_tys in case tcMatchTys tpl_lhs target_tys of Just subst -- matching worked. now, check for apartness. | apartnessCheck flattened_target branch -> -- matching worked & we're apart from all incompatible branches. -- success ASSERT( all (isJust . lookupCoVar subst) tpl_cvs ) Just (index, substTyVars subst tpl_tvs, substCoVars subst tpl_cvs) -- failure. keep looking _ -> other -- | Do an apartness check, as described in the "Closed Type Families" paper -- (POPL '14). This should be used when determining if an equation -- ('CoAxBranch') of a closed type family can be used to reduce a certain target -- type family application. apartnessCheck :: [Type] -- ^ /flattened/ target arguments. Make sure -- they're flattened! See Note [Flattening]. -- (NB: This "flat" is a different -- "flat" than is used in GHC.Tc.Solver.Flatten.) -> CoAxBranch -- ^ the candidate equation we wish to use -- Precondition: this matches the target -> Bool -- ^ True <=> equation can fire apartnessCheck flattened_target (CoAxBranch { cab_incomps = incomps }) = all (isSurelyApart . tcUnifyTysFG (const BindMe) flattened_target . coAxBranchLHS) incomps where isSurelyApart SurelyApart = True isSurelyApart _ = False {- ************************************************************************ * * Looking up a family instance * * ************************************************************************ Note [Normalising types] ~~~~~~~~~~~~~~~~~~~~~~~~ The topNormaliseType function removes all occurrences of type families and newtypes from the top-level structure of a type. normaliseTcApp does the type family lookup and is fairly straightforward. normaliseType is a little more involved. The complication comes from the fact that a type family might be used in the kind of a variable bound in a forall. We wish to remove this type family application, but that means coming up with a fresh variable (with the new kind). Thus, we need a substitution to be built up as we recur through the type. However, an ordinary TCvSubst just won't do: when we hit a type variable whose kind has changed during normalisation, we need both the new type variable *and* the coercion. We could conjure up a new VarEnv with just this property, but a usable substitution environment already exists: LiftingContexts from the liftCoSubst family of functions, defined in GHC.Core.Coercion. A LiftingContext maps a type variable to a coercion and a coercion variable to a pair of coercions. Let's ignore coercion variables for now. Because the coercion a type variable maps to contains the destination type (via coercionKind), we don't need to store that destination type separately. Thus, a LiftingContext has what we need: a map from type variables to (Coercion, Type) pairs. We also benefit because we can piggyback on the liftCoSubstVarBndr function to deal with binders. However, I had to modify that function to work with this application. Thus, we now have liftCoSubstVarBndrUsing, which takes a function used to process the kind of the binder. We don't wish to lift the kind, but instead normalise it. So, we pass in a callback function that processes the kind of the binder. After that brilliant explanation of all this, I'm sure you've forgotten the dangling reference to coercion variables. What do we do with those? Nothing at all. The point of normalising types is to remove type family applications, but there's no sense in removing these from coercions. We would just get back a new coercion witnessing the equality between the same types as the original coercion. Because coercions are irrelevant anyway, there is no point in doing this. So, whenever we encounter a coercion, we just say that it won't change. That's what the CoercionTy case is doing within normalise_type. Note [Normalisation and type synonyms] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ We need to be a bit careful about normalising in the presence of type synonyms (#13035). Suppose S is a type synonym, and we have S t1 t2 If S is family-free (on its RHS) we can just normalise t1 and t2 and reconstruct (S t1' t2'). Expanding S could not reveal any new redexes because type families are saturated. But if S has a type family on its RHS we expand /before/ normalising the args t1, t2. If we normalise t1, t2 first, we'll re-normalise them after expansion, and that can lead to /exponential/ behaviour; see #13035. Notice, though, that expanding first can in principle duplicate t1,t2, which might contain redexes. I'm sure you could conjure up an exponential case by that route too, but it hasn't happened in practice yet! -} topNormaliseType :: FamInstEnvs -> Type -> Type topNormaliseType env ty = case topNormaliseType_maybe env ty of Just (_co, ty') -> ty' Nothing -> ty topNormaliseType_maybe :: FamInstEnvs -> Type -> Maybe (Coercion, Type) -- ^ Get rid of *outermost* (or toplevel) -- * type function redex -- * data family redex -- * newtypes -- returning an appropriate Representational coercion. Specifically, if -- topNormaliseType_maybe env ty = Just (co, ty') -- then -- (a) co :: ty ~R ty' -- (b) ty' is not a newtype, and is not a type-family or data-family redex -- -- However, ty' can be something like (Maybe (F ty)), where -- (F ty) is a redex. -- -- Always operates homogeneously: the returned type has the same kind as the -- original type, and the returned coercion is always homogeneous. topNormaliseType_maybe env ty = do { ((co, mkind_co), nty) <- topNormaliseTypeX stepper combine ty ; return $ case mkind_co of MRefl -> (co, nty) MCo kind_co -> let nty_casted = nty `mkCastTy` mkSymCo kind_co final_co = mkCoherenceRightCo Representational nty (mkSymCo kind_co) co in (final_co, nty_casted) } where stepper = unwrapNewTypeStepper' `composeSteppers` tyFamStepper combine (c1, mc1) (c2, mc2) = (c1 `mkTransCo` c2, mc1 `mkTransMCo` mc2) unwrapNewTypeStepper' :: NormaliseStepper (Coercion, MCoercionN) unwrapNewTypeStepper' rec_nts tc tys = mapStepResult (, MRefl) $ unwrapNewTypeStepper rec_nts tc tys -- second coercion below is the kind coercion relating the original type's kind -- to the normalised type's kind tyFamStepper :: NormaliseStepper (Coercion, MCoercionN) tyFamStepper rec_nts tc tys -- Try to step a type/data family = case topReduceTyFamApp_maybe env tc tys of Just (co, rhs, res_co) -> NS_Step rec_nts rhs (co, MCo res_co) _ -> NS_Done --------------- normaliseTcApp :: FamInstEnvs -> Role -> TyCon -> [Type] -> (Coercion, Type) -- See comments on normaliseType for the arguments of this function normaliseTcApp env role tc tys = initNormM env role (tyCoVarsOfTypes tys) $ normalise_tc_app tc tys -- See Note [Normalising types] about the LiftingContext normalise_tc_app :: TyCon -> [Type] -> NormM (Coercion, Type) normalise_tc_app tc tys | Just (tenv, rhs, tys') <- expandSynTyCon_maybe tc tys , not (isFamFreeTyCon tc) -- Expand and try again = -- A synonym with type families in the RHS -- Expand and try again -- See Note [Normalisation and type synonyms] normalise_type (mkAppTys (substTy (mkTvSubstPrs tenv) rhs) tys') | isFamilyTyCon tc = -- A type-family application do { env <- getEnv ; role <- getRole ; (args_co, ntys, res_co) <- normalise_tc_args tc tys ; case reduceTyFamApp_maybe env role tc ntys of Just (first_co, ty') -> do { (rest_co,nty) <- normalise_type ty' ; return (assemble_result role nty (args_co `mkTransCo` first_co `mkTransCo` rest_co) res_co) } _ -> -- No unique matching family instance exists; -- we do not do anything return (assemble_result role (mkTyConApp tc ntys) args_co res_co) } | otherwise = -- A synonym with no type families in the RHS; or data type etc -- Just normalise the arguments and rebuild do { (args_co, ntys, res_co) <- normalise_tc_args tc tys ; role <- getRole ; return (assemble_result role (mkTyConApp tc ntys) args_co res_co) } where assemble_result :: Role -- r, ambient role in NormM monad -> Type -- nty, result type, possibly of changed kind -> Coercion -- orig_ty ~r nty, possibly heterogeneous -> CoercionN -- typeKind(orig_ty) ~N typeKind(nty) -> (Coercion, Type) -- (co :: orig_ty ~r nty_casted, nty_casted) -- where nty_casted has same kind as orig_ty assemble_result r nty orig_to_nty kind_co = ( final_co, nty_old_kind ) where nty_old_kind = nty `mkCastTy` mkSymCo kind_co final_co = mkCoherenceRightCo r nty (mkSymCo kind_co) orig_to_nty --------------- -- | Try to simplify a type-family application, by *one* step -- If topReduceTyFamApp_maybe env r F tys = Just (co, rhs, res_co) -- then co :: F tys ~R# rhs -- res_co :: typeKind(F tys) ~ typeKind(rhs) -- Type families and data families; always Representational role topReduceTyFamApp_maybe :: FamInstEnvs -> TyCon -> [Type] -> Maybe (Coercion, Type, Coercion) topReduceTyFamApp_maybe envs fam_tc arg_tys | isFamilyTyCon fam_tc -- type families and data families , Just (co, rhs) <- reduceTyFamApp_maybe envs role fam_tc ntys = Just (args_co `mkTransCo` co, rhs, res_co) | otherwise = Nothing where role = Representational (args_co, ntys, res_co) = initNormM envs role (tyCoVarsOfTypes arg_tys) $ normalise_tc_args fam_tc arg_tys normalise_tc_args :: TyCon -> [Type] -- tc tys -> NormM (Coercion, [Type], CoercionN) -- (co, new_tys), where -- co :: tc tys ~ tc new_tys; might not be homogeneous -- res_co :: typeKind(tc tys) ~N typeKind(tc new_tys) normalise_tc_args tc tys = do { role <- getRole ; (args_cos, nargs, res_co) <- normalise_args (tyConKind tc) (tyConRolesX role tc) tys ; return (mkTyConAppCo role tc args_cos, nargs, res_co) } --------------- normaliseType :: FamInstEnvs -> Role -- desired role of coercion -> Type -> (Coercion, Type) normaliseType env role ty = initNormM env role (tyCoVarsOfType ty) $ normalise_type ty normalise_type :: Type -- old type -> NormM (Coercion, Type) -- (coercion, new type), where -- co :: old-type ~ new_type -- Normalise the input type, by eliminating *all* type-function redexes -- but *not* newtypes (which are visible to the programmer) -- Returns with Refl if nothing happens -- Does nothing to newtypes -- The returned coercion *must* be *homogeneous* -- See Note [Normalising types] -- Try not to disturb type synonyms if possible normalise_type ty = go ty where go (TyConApp tc tys) = normalise_tc_app tc tys go ty@(LitTy {}) = do { r <- getRole ; return (mkReflCo r ty, ty) } go (AppTy ty1 ty2) = go_app_tys ty1 [ty2] go ty@(FunTy { ft_mult = w, ft_arg = ty1, ft_res = ty2 }) = do { (co1, nty1) <- go ty1 ; (co2, nty2) <- go ty2 ; (wco, wty) <- withRole Nominal $ go w ; r <- getRole ; return (mkFunCo r wco co1 co2, ty { ft_mult = wty, ft_arg = nty1, ft_res = nty2 }) } go (ForAllTy (Bndr tcvar vis) ty) = do { (lc', tv', h, ki') <- normalise_var_bndr tcvar ; (co, nty) <- withLC lc' $ normalise_type ty ; let tv2 = setTyVarKind tv' ki' ; return (mkForAllCo tv' h co, ForAllTy (Bndr tv2 vis) nty) } go (TyVarTy tv) = normalise_tyvar tv go (CastTy ty co) = do { (nco, nty) <- go ty ; lc <- getLC ; let co' = substRightCo lc co ; return (castCoercionKind2 nco Nominal ty nty co co' , mkCastTy nty co') } go (CoercionTy co) = do { lc <- getLC ; r <- getRole ; let right_co = substRightCo lc co ; return ( mkProofIrrelCo r (liftCoSubst Nominal lc (coercionType co)) co right_co , mkCoercionTy right_co ) } go_app_tys :: Type -- function -> [Type] -- args -> NormM (Coercion, Type) -- cf. GHC.Tc.Solver.Flatten.flatten_app_ty_args go_app_tys (AppTy ty1 ty2) tys = go_app_tys ty1 (ty2 : tys) go_app_tys fun_ty arg_tys = do { (fun_co, nfun) <- go fun_ty ; case tcSplitTyConApp_maybe nfun of Just (tc, xis) -> do { (second_co, nty) <- go (mkTyConApp tc (xis ++ arg_tys)) -- flatten_app_ty_args avoids redundantly processing the xis, -- but that's a much more performance-sensitive function. -- This type normalisation is not called in a loop. ; return (mkAppCos fun_co (map mkNomReflCo arg_tys) `mkTransCo` second_co, nty) } Nothing -> do { (args_cos, nargs, res_co) <- normalise_args (typeKind nfun) (repeat Nominal) arg_tys ; role <- getRole ; let nty = mkAppTys nfun nargs nco = mkAppCos fun_co args_cos nty_casted = nty `mkCastTy` mkSymCo res_co final_co = mkCoherenceRightCo role nty (mkSymCo res_co) nco ; return (final_co, nty_casted) } } normalise_args :: Kind -- of the function -> [Role] -- roles at which to normalise args -> [Type] -- args -> NormM ([Coercion], [Type], Coercion) -- returns (cos, xis, res_co), where each xi is the normalised -- version of the corresponding type, each co is orig_arg ~ xi, -- and the res_co :: kind(f orig_args) ~ kind(f xis) -- NB: The xis might *not* have the same kinds as the input types, -- but the resulting application *will* be well-kinded -- cf. GHC.Tc.Solver.Flatten.flatten_args_slow normalise_args fun_ki roles args = do { normed_args <- zipWithM normalise1 roles args ; let (xis, cos, res_co) = simplifyArgsWorker ki_binders inner_ki fvs roles normed_args ; return (map mkSymCo cos, xis, mkSymCo res_co) } where (ki_binders, inner_ki) = splitPiTys fun_ki fvs = tyCoVarsOfTypes args -- flattener conventions are different from ours impedance_match :: NormM (Coercion, Type) -> NormM (Type, Coercion) impedance_match action = do { (co, ty) <- action ; return (ty, mkSymCo co) } normalise1 role ty = impedance_match $ withRole role $ normalise_type ty normalise_tyvar :: TyVar -> NormM (Coercion, Type) normalise_tyvar tv = ASSERT( isTyVar tv ) do { lc <- getLC ; r <- getRole ; return $ case liftCoSubstTyVar lc r tv of Just co -> (co, coercionRKind co) Nothing -> (mkReflCo r ty, ty) } where ty = mkTyVarTy tv normalise_var_bndr :: TyCoVar -> NormM (LiftingContext, TyCoVar, Coercion, Kind) normalise_var_bndr tcvar -- works for both tvar and covar = do { lc1 <- getLC ; env <- getEnv ; let callback lc ki = runNormM (normalise_type ki) env lc Nominal ; return $ liftCoSubstVarBndrUsing callback lc1 tcvar } -- | a monad for the normalisation functions, reading 'FamInstEnvs', -- a 'LiftingContext', and a 'Role'. newtype NormM a = NormM { runNormM :: FamInstEnvs -> LiftingContext -> Role -> a } deriving (Functor) initNormM :: FamInstEnvs -> Role -> TyCoVarSet -- the in-scope variables -> NormM a -> a initNormM env role vars (NormM thing_inside) = thing_inside env lc role where in_scope = mkInScopeSet vars lc = emptyLiftingContext in_scope getRole :: NormM Role getRole = NormM (\ _ _ r -> r) getLC :: NormM LiftingContext getLC = NormM (\ _ lc _ -> lc) getEnv :: NormM FamInstEnvs getEnv = NormM (\ env _ _ -> env) withRole :: Role -> NormM a -> NormM a withRole r thing = NormM $ \ envs lc _old_r -> runNormM thing envs lc r withLC :: LiftingContext -> NormM a -> NormM a withLC lc thing = NormM $ \ envs _old_lc r -> runNormM thing envs lc r instance Monad NormM where ma >>= fmb = NormM $ \env lc r -> let a = runNormM ma env lc r in runNormM (fmb a) env lc r instance Applicative NormM where pure x = NormM $ \ _ _ _ -> x (<*>) = ap {- ************************************************************************ * * Flattening * * ************************************************************************ Note [Flattening] ~~~~~~~~~~~~~~~~~ As described in "Closed type families with overlapping equations" http://research.microsoft.com/en-us/um/people/simonpj/papers/ext-f/axioms-extended.pdf we need to flatten core types before unifying them, when checking for "surely-apart" against earlier equations of a closed type family. Flattening means replacing all top-level uses of type functions with fresh variables, *taking care to preserve sharing*. That is, the type (Either (F a b) (F a b)) should flatten to (Either c c), never (Either c d). Here is a nice example of why it's all necessary: type family F a b where F Int Bool = Char F a b = Double type family G a -- open, no instances How do we reduce (F (G Float) (G Float))? The first equation clearly doesn't match, while the second equation does. But, before reducing, we must make sure that the target can never become (F Int Bool). Well, no matter what G Float becomes, it certainly won't become *both* Int and Bool, so indeed we're safe reducing (F (G Float) (G Float)) to Double. This is necessary not only to get more reductions (which we might be willing to give up on), but for substitutivity. If we have (F x x), we can see that (F x x) can reduce to Double. So, it had better be the case that (F blah blah) can reduce to Double, no matter what (blah) is! Flattening as done below ensures this. The algorithm works by building up a TypeMap TyVar, mapping type family applications to fresh variables. This mapping must be threaded through all the function calls, as any entry in the mapping must be propagated to all future nodes in the tree. The algorithm also must track the set of in-scope variables, in order to make fresh variables as it flattens. (We are far from a source of fresh Uniques.) See Wrinkle 2, below. There are wrinkles, of course: 1. The flattening algorithm must account for the possibility of inner `forall`s. (A `forall` seen here can happen only because of impredicativity. However, the flattening operation is an algorithm in Core, which is impredicative.) Suppose we have (forall b. F b) -> (forall b. F b). Of course, those two bs are entirely unrelated, and so we should certainly not flatten the two calls F b to the same variable. Instead, they must be treated separately. We thus carry a substitution that freshens variables; we must apply this substitution (in `coreFlattenTyFamApp`) before looking up an application in the environment. Note that the range of the substitution contains only TyVars, never anything else. For the sake of efficiency, we only apply this substitution when absolutely necessary. Namely: * We do not perform the substitution at all if it is empty. * We only need to worry about the arguments of a type family that are within the arity of said type family, so we can get away with not applying the substitution to any oversaturated type family arguments. * Importantly, we do /not/ achieve this substitution by recursively flattening the arguments, as this would be wrong. Consider `F (G a)`, where F and G are type families. We might decide that `F (G a)` flattens to `beta`. Later, the substitution is non-empty (but does not map `a`) and so we flatten `G a` to `gamma` and try to flatten `F gamma`. Of course, `F gamma` is unknown, and so we flatten it to `delta`, but it really should have been `beta`! Argh! Moral of the story: instead of flattening the arguments, just substitute them directly. 2. There are two different reasons we might add a variable to the in-scope set as we work: A. We have just invented a new flattening variable. B. We have entered a `forall`. Annoying here is that in-scope variable source (A) must be threaded through the calls. For example, consider (F b -> forall c. F c). Suppose that, when flattening F b, we invent a fresh variable c. Now, when we encounter (forall c. F c), we need to know c is already in scope so that we locally rename c to c'. However, if we don't thread through the in-scope set from one argument of (->) to the other, we won't know this and might get very confused. In contrast, source (B) increases only as we go deeper, as in-scope sets normally do. However, even here we must be careful. The TypeMap TyVar that contains mappings from type family applications to freshened variables will be threaded through both sides of (forall b. F b) -> (forall b. F b). We thus must make sure that the two `b`s don't get renamed to the same b1. (If they did, then looking up `F b1` would yield the same flatten var for each.) So, even though `forall`-bound variables should really be in the in-scope set only when they are in scope, we retain these variables even outside of their scope. This ensures that, if we encounter a fresh `forall`-bound b, we will rename it to b2, not b1. Note that keeping a larger in-scope set than strictly necessary is always OK, as in-scope sets are only ever used to avoid collisions. Sadly, the freshening substitution described in (1) really mustn't bind variables outside of their scope: note that its domain is the *unrenamed* variables. This means that the substitution gets "pushed down" (like a reader monad) while the in-scope set gets threaded (like a state monad). Because a TCvSubst contains its own in-scope set, we don't carry a TCvSubst; instead, we just carry a TvSubstEnv down, tying it to the InScopeSet traveling separately as necessary. 3. Consider `F ty_1 ... ty_n`, where F is a type family with arity k: type family F ty_1 ... ty_k :: res_k It's tempting to just flatten `F ty_1 ... ty_n` to `alpha`, where alpha is a flattening skolem. But we must instead flatten it to `alpha ty_(k+1) ... ty_n`—that is, by only flattening up to the arity of the type family. Why is this better? Consider the following concrete example from #16995: type family Param :: Type -> Type type family LookupParam (a :: Type) :: Type where LookupParam (f Char) = Bool LookupParam x = Int foo :: LookupParam (Param ()) foo = 42 In order for `foo` to typecheck, `LookupParam (Param ())` must reduce to `Int`. But if we flatten `Param ()` to `alpha`, then GHC can't be sure if `alpha` is apart from `f Char`, so it won't fall through to the second equation. But since the `Param` type family has arity 0, we can instead flatten `Param ()` to `alpha ()`, about which GHC knows with confidence is apart from `f Char`, permitting the second equation to be reached. Not only does this allow more programs to be accepted, it's also important for correctness. Not doing this was the root cause of the Core Lint error in #16995. flattenTys is defined here because of module dependencies. -} data FlattenEnv = FlattenEnv { fe_type_map :: TypeMap TyVar -- domain: exactly-saturated type family applications -- range: fresh variables , fe_in_scope :: InScopeSet } -- See Note [Flattening] emptyFlattenEnv :: InScopeSet -> FlattenEnv emptyFlattenEnv in_scope = FlattenEnv { fe_type_map = emptyTypeMap , fe_in_scope = in_scope } updateInScopeSet :: FlattenEnv -> (InScopeSet -> InScopeSet) -> FlattenEnv updateInScopeSet env upd = env { fe_in_scope = upd (fe_in_scope env) } flattenTys :: InScopeSet -> [Type] -> [Type] -- See Note [Flattening] -- NB: the returned types may mention fresh type variables, -- arising from the flattening. We don't return the -- mapping from those fresh vars to the ty-fam -- applications they stand for (we could, but no need) flattenTys in_scope tys = snd $ coreFlattenTys emptyTvSubstEnv (emptyFlattenEnv in_scope) tys coreFlattenTys :: TvSubstEnv -> FlattenEnv -> [Type] -> (FlattenEnv, [Type]) coreFlattenTys subst = mapAccumL (coreFlattenTy subst) coreFlattenTy :: TvSubstEnv -> FlattenEnv -> Type -> (FlattenEnv, Type) coreFlattenTy subst = go where go env ty | Just ty' <- coreView ty = go env ty' go env (TyVarTy tv) | Just ty <- lookupVarEnv subst tv = (env, ty) | otherwise = let (env', ki) = go env (tyVarKind tv) in (env', mkTyVarTy $ setTyVarKind tv ki) go env (AppTy ty1 ty2) = let (env1, ty1') = go env ty1 (env2, ty2') = go env1 ty2 in (env2, AppTy ty1' ty2') go env (TyConApp tc tys) -- NB: Don't just check if isFamilyTyCon: this catches *data* families, -- which are generative and thus can be preserved during flattening | not (isGenerativeTyCon tc Nominal) = coreFlattenTyFamApp subst env tc tys | otherwise = let (env', tys') = coreFlattenTys subst env tys in (env', mkTyConApp tc tys') go env ty@(FunTy { ft_mult = mult, ft_arg = ty1, ft_res = ty2 }) = let (env1, ty1') = go env ty1 (env2, ty2') = go env1 ty2 (env3, mult') = go env2 mult in (env3, ty { ft_mult = mult', ft_arg = ty1', ft_res = ty2' }) go env (ForAllTy (Bndr tv vis) ty) = let (env1, subst', tv') = coreFlattenVarBndr subst env tv (env2, ty') = coreFlattenTy subst' env1 ty in (env2, ForAllTy (Bndr tv' vis) ty') go env ty@(LitTy {}) = (env, ty) go env (CastTy ty co) = let (env1, ty') = go env ty (env2, co') = coreFlattenCo subst env1 co in (env2, CastTy ty' co') go env (CoercionTy co) = let (env', co') = coreFlattenCo subst env co in (env', CoercionTy co') -- when flattening, we don't care about the contents of coercions. -- so, just return a fresh variable of the right (flattened) type coreFlattenCo :: TvSubstEnv -> FlattenEnv -> Coercion -> (FlattenEnv, Coercion) coreFlattenCo subst env co = (env2, mkCoVarCo covar) where (env1, kind') = coreFlattenTy subst env (coercionType co) covar = mkFlattenFreshCoVar (fe_in_scope env1) kind' -- Add the covar to the FlattenEnv's in-scope set. -- See Note [Flattening], wrinkle 2A. env2 = updateInScopeSet env1 (flip extendInScopeSet covar) coreFlattenVarBndr :: TvSubstEnv -> FlattenEnv -> TyCoVar -> (FlattenEnv, TvSubstEnv, TyVar) coreFlattenVarBndr subst env tv = (env2, subst', tv') where -- See Note [Flattening], wrinkle 2B. kind = varType tv (env1, kind') = coreFlattenTy subst env kind tv' = uniqAway (fe_in_scope env1) (setVarType tv kind') subst' = extendVarEnv subst tv (mkTyVarTy tv') env2 = updateInScopeSet env1 (flip extendInScopeSet tv') coreFlattenTyFamApp :: TvSubstEnv -> FlattenEnv -> TyCon -- type family tycon -> [Type] -- args, already flattened -> (FlattenEnv, Type) coreFlattenTyFamApp tv_subst env fam_tc fam_args = case lookupTypeMap type_map fam_ty of Just tv -> (env', mkAppTys (mkTyVarTy tv) leftover_args') Nothing -> let tyvar_name = mkFlattenFreshTyName fam_tc tv = uniqAway in_scope $ mkTyVar tyvar_name (typeKind fam_ty) ty' = mkAppTys (mkTyVarTy tv) leftover_args' env'' = env' { fe_type_map = extendTypeMap type_map fam_ty tv , fe_in_scope = extendInScopeSet in_scope tv } in (env'', ty') where arity = tyConArity fam_tc tcv_subst = TCvSubst (fe_in_scope env) tv_subst emptyVarEnv (sat_fam_args, leftover_args) = ASSERT( arity <= length fam_args ) splitAt arity fam_args -- Apply the substitution before looking up an application in the -- environment. See Note [Flattening], wrinkle 1. -- NB: substTys short-cuts the common case when the substitution is empty. sat_fam_args' = substTys tcv_subst sat_fam_args (env', leftover_args') = coreFlattenTys tv_subst env leftover_args -- `fam_tc` may be over-applied to `fam_args` (see Note [Flattening], -- wrinkle 3), so we split it into the arguments needed to saturate it -- (sat_fam_args') and the rest (leftover_args') fam_ty = mkTyConApp fam_tc sat_fam_args' FlattenEnv { fe_type_map = type_map , fe_in_scope = in_scope } = env' mkFlattenFreshTyName :: Uniquable a => a -> Name mkFlattenFreshTyName unq = mkSysTvName (getUnique unq) (fsLit "flt") mkFlattenFreshCoVar :: InScopeSet -> Kind -> CoVar mkFlattenFreshCoVar in_scope kind = let uniq = unsafeGetFreshLocalUnique in_scope name = mkSystemVarName uniq (fsLit "flc") in mkCoVar name kind