%
% (c) The University of Glasgow 2006
%
\begin{code}
module Unify (
tcMatchTy, tcMatchTys, tcMatchTyX,
ruleMatchTyX, tcMatchPreds,
MatchEnv(..), matchList,
typesCantMatch,
tcUnifyTy, tcUnifyTys, BindFlag(..),
niFixTvSubst, niSubstTvSet,
UnifyResultM(..), UnifyResult, tcUnifyTysFG
) where
#include "HsVersions.h"
import Var
import VarEnv
import VarSet
import Kind
import Type
import TyCon
import TypeRep
import Util
import Control.Monad (liftM, ap)
import Control.Applicative (Applicative(..))
\end{code}
%************************************************************************
%* *
Matching
%* *
%************************************************************************
Matching is much tricker than you might think.
1. The substitution we generate binds the *template type variables*
which are given to us explicitly.
2. We want to match in the presence of foralls;
e.g (forall a. t1) ~ (forall b. t2)
That is what the RnEnv2 is for; it does the alpha-renaming
that makes it as if a and b were the same variable.
Initialising the RnEnv2, so that it can generate a fresh
binder when necessary, entails knowing the free variables of
both types.
3. We must be careful not to bind a template type variable to a
locally bound variable. E.g.
(forall a. x) ~ (forall b. b)
where x is the template type variable. Then we do not want to
bind x to a/b! This is a kind of occurs check.
The necessary locals accumulate in the RnEnv2.
\begin{code}
data MatchEnv
= ME { me_tmpls :: VarSet
, me_env :: RnEnv2
}
tcMatchTy :: TyVarSet
-> Type
-> Type
-> Maybe TvSubst
tcMatchTy tmpls ty1 ty2
= case match menv emptyTvSubstEnv ty1 ty2 of
Just subst_env -> Just (TvSubst in_scope subst_env)
Nothing -> Nothing
where
menv = ME { me_tmpls = tmpls, me_env = mkRnEnv2 in_scope }
in_scope = mkInScopeSet (tmpls `unionVarSet` tyVarsOfType ty2)
tcMatchTys :: TyVarSet
-> [Type]
-> [Type]
-> Maybe TvSubst
tcMatchTys tmpls tys1 tys2
= case match_tys menv emptyTvSubstEnv tys1 tys2 of
Just subst_env -> Just (TvSubst in_scope subst_env)
Nothing -> Nothing
where
menv = ME { me_tmpls = tmpls, me_env = mkRnEnv2 in_scope }
in_scope = mkInScopeSet (tmpls `unionVarSet` tyVarsOfTypes tys2)
tcMatchTyX :: TyVarSet
-> TvSubst
-> Type
-> Type
-> Maybe TvSubst
tcMatchTyX tmpls (TvSubst in_scope subst_env) ty1 ty2
= case match menv subst_env ty1 ty2 of
Just subst_env -> Just (TvSubst in_scope subst_env)
Nothing -> Nothing
where
menv = ME {me_tmpls = tmpls, me_env = mkRnEnv2 in_scope}
tcMatchPreds
:: [TyVar]
-> [PredType] -> [PredType]
-> Maybe TvSubstEnv
tcMatchPreds tmpls ps1 ps2
= matchList (match menv) emptyTvSubstEnv ps1 ps2
where
menv = ME { me_tmpls = mkVarSet tmpls, me_env = mkRnEnv2 in_scope_tyvars }
in_scope_tyvars = mkInScopeSet (tyVarsOfTypes ps1 `unionVarSet` tyVarsOfTypes ps2)
ruleMatchTyX :: MatchEnv
-> TvSubstEnv
-> Type
-> Type
-> Maybe TvSubstEnv
ruleMatchTyX menv subst ty1 ty2 = match menv subst ty1 ty2
\end{code}
Now the internals of matching
\begin{code}
match :: MatchEnv
-> TvSubstEnv
-> Type -> Type
-> Maybe TvSubstEnv
match menv subst ty1 ty2 | Just ty1' <- coreView ty1 = match menv subst ty1' ty2
| Just ty2' <- coreView ty2 = match menv subst ty1 ty2'
match menv subst (TyVarTy tv1) ty2
| Just ty1' <- lookupVarEnv subst tv1'
= if eqTypeX (nukeRnEnvL rn_env) ty1' ty2
then Just subst
else Nothing
| tv1' `elemVarSet` me_tmpls menv
= if any (inRnEnvR rn_env) (varSetElems (tyVarsOfType ty2))
then Nothing
else do { subst1 <- match_kind menv subst (tyVarKind tv1) (typeKind ty2)
; return (extendVarEnv subst1 tv1' ty2) }
| otherwise
= case ty2 of
TyVarTy tv2 | tv1' == rnOccR rn_env tv2 -> Just subst
_ -> Nothing
where
rn_env = me_env menv
tv1' = rnOccL rn_env tv1
match menv subst (ForAllTy tv1 ty1) (ForAllTy tv2 ty2)
= do { subst' <- match_kind menv subst (tyVarKind tv1) (tyVarKind tv2)
; match menv' subst' ty1 ty2 }
where
menv' = menv { me_env = rnBndr2 (me_env menv) tv1 tv2 }
match menv subst (TyConApp tc1 tys1) (TyConApp tc2 tys2)
| tc1 == tc2 = match_tys menv subst tys1 tys2
match menv subst (FunTy ty1a ty1b) (FunTy ty2a ty2b)
= do { subst' <- match menv subst ty1a ty2a
; match menv subst' ty1b ty2b }
match menv subst (AppTy ty1a ty1b) ty2
| Just (ty2a, ty2b) <- repSplitAppTy_maybe ty2
= do { subst' <- match menv subst ty1a ty2a
; match menv subst' ty1b ty2b }
match _ subst (LitTy x) (LitTy y) | x == y = return subst
match _ _ _ _
= Nothing
match_kind :: MatchEnv -> TvSubstEnv -> Kind -> Kind -> Maybe TvSubstEnv
match_kind menv subst k1 k2
| k2 `isSubKind` k1
= return subst
| otherwise
= match menv subst k1 k2
match_tys :: MatchEnv -> TvSubstEnv -> [Type] -> [Type] -> Maybe TvSubstEnv
match_tys menv subst tys1 tys2 = matchList (match menv) subst tys1 tys2
matchList :: (env -> a -> b -> Maybe env)
-> env -> [a] -> [b] -> Maybe env
matchList _ subst [] [] = Just subst
matchList fn subst (a:as) (b:bs) = do { subst' <- fn subst a b
; matchList fn subst' as bs }
matchList _ _ _ _ = Nothing
\end{code}
%************************************************************************
%* *
GADTs
%* *
%************************************************************************
Note [Pruning dead case alternatives]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider data T a where
T1 :: T Int
T2 :: T a
newtype X = MkX Int
newtype Y = MkY Char
type family F a
type instance F Bool = Int
Now consider case x of { T1 -> e1; T2 -> e2 }
The question before the house is this: if I know something about the type
of x, can I prune away the T1 alternative?
Suppose x::T Char. It's impossible to construct a (T Char) using T1,
Answer = YES we can prune the T1 branch (clearly)
Suppose x::T (F a), where 'a' is in scope. Then 'a' might be instantiated
to 'Bool', in which case x::T Int, so
ANSWER = NO (clearly)
Suppose x::T X. Then *in Haskell* it's impossible to construct a (non-bottom)
value of type (T X) using T1. But *in FC* it's quite possible. The newtype
gives a coercion
CoX :: X ~ Int
So (T CoX) :: T X ~ T Int; hence (T1 `cast` sym (T CoX)) is a non-bottom value
of type (T X) constructed with T1. Hence
ANSWER = NO we can't prune the T1 branch (surprisingly)
Furthermore, this can even happen; see Trac #1251. GHC's newtype-deriving
mechanism uses a cast, just as above, to move from one dictionary to another,
in effect giving the programmer access to CoX.
Finally, suppose x::T Y. Then *even in FC* we can't construct a
non-bottom value of type (T Y) using T1. That's because we can get
from Y to Char, but not to Int.
Here's a related question. data Eq a b where EQ :: Eq a a
Consider
case x of { EQ -> ... }
Suppose x::Eq Int Char. Is the alternative dead? Clearly yes.
What about x::Eq Int a, in a context where we have evidence that a~Char.
Then again the alternative is dead.
Summary
We are really doing a test for unsatisfiability of the type
constraints implied by the match. And that is clearly, in general, a
hard thing to do.
However, since we are simply dropping dead code, a conservative test
suffices. There is a continuum of tests, ranging from easy to hard, that
drop more and more dead code.
For now we implement a very simple test: type variables match
anything, type functions (incl newtypes) match anything, and only
distinct data types fail to match. We can elaborate later.
\begin{code}
typesCantMatch :: [(Type,Type)] -> Bool
typesCantMatch prs = any (\(s,t) -> cant_match s t) prs
where
cant_match :: Type -> Type -> Bool
cant_match t1 t2
| Just t1' <- coreView t1 = cant_match t1' t2
| Just t2' <- coreView t2 = cant_match t1 t2'
cant_match (FunTy a1 r1) (FunTy a2 r2)
= cant_match a1 a2 || cant_match r1 r2
cant_match (TyConApp tc1 tys1) (TyConApp tc2 tys2)
| isDistinctTyCon tc1 && isDistinctTyCon tc2
= tc1 /= tc2 || typesCantMatch (zipEqual "typesCantMatch" tys1 tys2)
cant_match (FunTy {}) (TyConApp tc _) = isDistinctTyCon tc
cant_match (TyConApp tc _) (FunTy {}) = isDistinctTyCon tc
cant_match (AppTy f1 a1) ty2
| Just (f2, a2) <- repSplitAppTy_maybe ty2
= cant_match f1 f2 || cant_match a1 a2
cant_match ty1 (AppTy f2 a2)
| Just (f1, a1) <- repSplitAppTy_maybe ty1
= cant_match f1 f2 || cant_match a1 a2
cant_match (LitTy x) (LitTy y) = x /= y
cant_match _ _ = False
\end{code}
%************************************************************************
%* *
Unification
%* *
%************************************************************************
Note [Fine-grained unification]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Do the types (x, x) and ([y], y) unify? The answer is seemingly "no" --
no substitution to finite types makes these match. But, a substitution to
*infinite* types can unify these two types: [x |-> [[[...]]], y |-> [[[...]]] ].
Why do we care? Consider these two type family instances:
type instance F x x = Int
type instance F [y] y = Bool
If we also have
type instance Looper = [Looper]
then the instances potentially overlap. The solution is to use unification
over infinite terms. This is possible (see [1] for lots of gory details), but
a full algorithm is a little more power than we need. Instead, we make a
conservative approximation and just omit the occurs check.
[1]: http://research.microsoft.com/en-us/um/people/simonpj/papers/ext-f/axioms-extended.pdf
tcUnifyTys considers an occurs-check problem as the same as general unification
failure.
tcUnifyTysFG ("fine-grained") returns one of three results: success, occurs-check
failure ("MaybeApart"), or general failure ("SurelyApart").
See also Trac #8162.
It's worth noting that unification in the presence of infinite types is not
complete. This means that, sometimes, a closed type family does not reduce
when it should. See test case indexed-types/should_fail/Overlap15 for an
example.
Note [The substitution in MaybeApart]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The constructor MaybeApart carries data with it, typically a TvSubstEnv. Why?
Because consider unifying these:
(a, a, Int) ~ (b, [b], Bool)
If we go left-to-right, we start with [a |-> b]. Then, on the middle terms, we
apply the subst we have so far and discover that we need [b |-> [b]]. Because
this fails the occurs check, we say that the types are MaybeApart (see above
Note [Fine-grained unification]). But, we can't stop there! Because if we
continue, we discover that Int is SurelyApart from Bool, and therefore the
types are apart. This has practical consequences for the ability for closed
type family applications to reduce. See test case
indexed-types/should_compile/Overlap14.
Note [Unifying with skolems]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
If we discover that two types unify if and only if a skolem variable is
substituted, we can't properly unify the types. But, that skolem variable
may later be instantiated with a unifyable type. So, we return maybeApart
in these cases.
\begin{code}
tcUnifyTy :: Type -> Type
-> Maybe TvSubst
tcUnifyTy ty1 ty2
= case initUM (const BindMe) (unify emptyTvSubstEnv ty1 ty2) of
Unifiable subst_env -> Just (niFixTvSubst subst_env)
_other -> Nothing
tcUnifyTys :: (TyVar -> BindFlag)
-> [Type] -> [Type]
-> Maybe TvSubst
tcUnifyTys bind_fn tys1 tys2
= case tcUnifyTysFG bind_fn tys1 tys2 of
Unifiable subst -> Just subst
_ -> Nothing
type UnifyResult = UnifyResultM TvSubst
data UnifyResultM a = Unifiable a
| MaybeApart a
| SurelyApart
tcUnifyTysFG :: (TyVar -> BindFlag)
-> [Type] -> [Type]
-> UnifyResult
tcUnifyTysFG bind_fn tys1 tys2
= initUM bind_fn $
do { subst <- unifyList emptyTvSubstEnv tys1 tys2
; return (niFixTvSubst subst) }
\end{code}
%************************************************************************
%* *
Non-idempotent substitution
%* *
%************************************************************************
Note [Non-idempotent substitution]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
During unification we use a TvSubstEnv that is
(a) non-idempotent
(b) loop-free; ie repeatedly applying it yields a fixed point
\begin{code}
niFixTvSubst :: TvSubstEnv -> TvSubst
niFixTvSubst env = f env
where
f e | not_fixpoint = f (mapVarEnv (substTy subst) e)
| otherwise = subst
where
range_tvs = foldVarEnv (unionVarSet . tyVarsOfType) emptyVarSet e
subst = mkTvSubst (mkInScopeSet range_tvs) e
not_fixpoint = foldVarSet ((||) . in_domain) False range_tvs
in_domain tv = tv `elemVarEnv` e
niSubstTvSet :: TvSubstEnv -> TyVarSet -> TyVarSet
niSubstTvSet subst tvs
= foldVarSet (unionVarSet . get) emptyVarSet tvs
where
get tv = case lookupVarEnv subst tv of
Nothing -> unitVarSet tv
Just ty -> niSubstTvSet subst (tyVarsOfType ty)
\end{code}
%************************************************************************
%* *
The workhorse
%* *
%************************************************************************
\begin{code}
unify :: TvSubstEnv
-> Type -> Type
-> UM TvSubstEnv
unify subst (TyVarTy tv1) ty2 = uVar subst tv1 ty2
unify subst ty1 (TyVarTy tv2) = uVar subst tv2 ty1
unify subst ty1 ty2 | Just ty1' <- tcView ty1 = unify subst ty1' ty2
unify subst ty1 ty2 | Just ty2' <- tcView ty2 = unify subst ty1 ty2'
unify subst (TyConApp tyc1 tys1) (TyConApp tyc2 tys2)
| tyc1 == tyc2
= unify_tys subst tys1 tys2
unify subst (FunTy ty1a ty1b) (FunTy ty2a ty2b)
= do { subst' <- unify subst ty1a ty2a
; unify subst' ty1b ty2b }
unify subst (AppTy ty1a ty1b) ty2
| Just (ty2a, ty2b) <- repSplitAppTy_maybe ty2
= do { subst' <- unify subst ty1a ty2a
; unify subst' ty1b ty2b }
unify subst ty1 (AppTy ty2a ty2b)
| Just (ty1a, ty1b) <- repSplitAppTy_maybe ty1
= do { subst' <- unify subst ty1a ty2a
; unify subst' ty1b ty2b }
unify subst (LitTy x) (LitTy y) | x == y = return subst
unify _ _ _ = surelyApart
unify_tys :: TvSubstEnv -> [Type] -> [Type] -> UM TvSubstEnv
unify_tys subst xs ys = unifyList subst xs ys
unifyList :: TvSubstEnv -> [Type] -> [Type] -> UM TvSubstEnv
unifyList subst orig_xs orig_ys
= go subst orig_xs orig_ys
where
go subst [] [] = return subst
go subst (x:xs) (y:ys) = do { subst' <- unify subst x y
; go subst' xs ys }
go _ _ _ = surelyApart
uVar :: TvSubstEnv
-> TyVar
-> Type
-> UM TvSubstEnv
uVar subst tv1 ty
=
case (lookupVarEnv subst tv1) of
Just ty' -> unify subst ty' ty
Nothing -> uUnrefined subst
tv1 ty ty
uUnrefined :: TvSubstEnv
-> TyVar
-> Type
-> Type
-> UM TvSubstEnv
uUnrefined subst tv1 ty2 ty2'
| Just ty2'' <- tcView ty2'
= uUnrefined subst tv1 ty2 ty2''
uUnrefined subst tv1 ty2 (TyVarTy tv2)
| tv1 == tv2
= return subst
| Just ty' <- lookupVarEnv subst tv2
= uUnrefined subst tv1 ty' ty'
| otherwise
= do {
; subst' <- unify subst (tyVarKind tv1) (tyVarKind tv2)
; b1 <- tvBindFlag tv1
; b2 <- tvBindFlag tv2
; let ty1 = TyVarTy tv1
; case (b1, b2) of
(Skolem, Skolem) -> maybeApart subst'
(BindMe, _) -> return (extendVarEnv subst' tv1 ty2)
(_, BindMe) -> return (extendVarEnv subst' tv2 ty1) }
uUnrefined subst tv1 ty2 ty2'
| tv1 `elemVarSet` niSubstTvSet subst (tyVarsOfType ty2')
= maybeApart subst
| otherwise
= do { subst' <- unify subst k1 k2
; bindTv subst' tv1 ty2 }
where
k1 = tyVarKind tv1
k2 = typeKind ty2'
bindTv :: TvSubstEnv -> TyVar -> Type -> UM TvSubstEnv
bindTv subst tv ty
= do { b <- tvBindFlag tv
; case b of
Skolem -> maybeApart subst
BindMe -> return $ extendVarEnv subst tv ty
}
\end{code}
%************************************************************************
%* *
Binding decisions
%* *
%************************************************************************
\begin{code}
data BindFlag
= BindMe
| Skolem
\end{code}
%************************************************************************
%* *
Unification monad
%* *
%************************************************************************
\begin{code}
newtype UM a = UM { unUM :: (TyVar -> BindFlag)
-> UnifyResultM a }
instance Functor UM where
fmap = liftM
instance Applicative UM where
pure = return
(<*>) = ap
instance Monad UM where
return a = UM (\_tvs -> Unifiable a)
fail _ = UM (\_tvs -> SurelyApart)
m >>= k = UM (\tvs -> case unUM m tvs of
Unifiable v -> unUM (k v) tvs
MaybeApart v ->
case unUM (k v) tvs of
Unifiable v' -> MaybeApart v'
other -> other
SurelyApart -> SurelyApart)
initUM :: (TyVar -> BindFlag) -> UM a -> UnifyResultM a
initUM badtvs um = unUM um badtvs
tvBindFlag :: TyVar -> UM BindFlag
tvBindFlag tv = UM (\tv_fn -> Unifiable (tv_fn tv))
maybeApart :: TvSubstEnv -> UM TvSubstEnv
maybeApart subst = UM (\_tv_fn -> MaybeApart subst)
surelyApart :: UM a
surelyApart = UM (\_tv_fn -> SurelyApart)
\end{code}