⏴ System F Pure Type Systems ⏵
Contents

Type operators

In Haskell, Map Integer String describes a map of integers to strings. Thus Map is an example of a type operator, because it takes 2 types and returns a type.

GHC has an syntax sugar extension called "type operators". We use the term differently; for us, a type operator is a type-level function.

We introduce simply-typed lambda calculus at the level of types. We have operator abstractions and operator applications. We say kind for the type of a type-level lambda expression, and define the base kind * for proper types that is, the types of (term-level) lambda expressions.

For example, the Map type constructor has kind * -> * -> *. No term has type Map. The Integer and String types both have kind *, so Map Integer String has kind * and it is therefore a proper type. Another example of a proper type is (String -> Int) -> String.

When type operators are added to System F, we obtain System Fω.


Definitions

Our Type and Term data types both have their own variables, abstractions, and applications. The new Kind data type holds typing information for Type values, and as before, Type holds typing information for Term values.

Because we’re extending System F, we also have Forall, TLam, and TApp for functions that take types and return terms; without these, we obtain a system known as \(\lambda\underline{\omega}\). [I don’t know much about \(\lambda\underline{\omega}\), but because types and terms undergo beta reduction in their own separate worlds, I sense it’s only a minor upgrade for simply-typed lambda calculus.]

The kinding ::* is common, so we elide it.

data Kind = Star | Kind :=> Kind deriving Eq
data Type = TV String | Forall (String, Kind) Type | Type :-> Type
 | OLam (String, Kind) Type | OApp Type Type
data Term = Var String | App Term Term | Lam (String, Type) Term
 | Let String Term Term
 | TLam (String, Kind) Term | TApp Term Type
instance Show Kind where
 show Star = "*"
 show (a :=> b) = showA ++ "->" ++ show b where
 showA = case a of
 _ :=> _ -> "(" ++ show a ++ ")"
 _ -> show a
showK Star = ""
showK k = "::" ++ show k
instance Show Type where
 show ty = case ty of
 TV s -> s
 Forall (s, k) t -> "∀" ++ s ++ showK k ++ "." ++ show t
 t :-> u -> showL ++ " -> " ++ showR where
 showL = case t of
 Forall _ _ -> "(" ++ show t ++ ")"
 _ :-> _ -> "(" ++ show t ++ ")"
 _ -> show t
 showR = case u of
 Forall _ _ -> "(" ++ show u ++ ")"
 _ -> show u
 OLam (s, k) t -> "λ" ++ s ++ showK k ++ "." ++ show t
 OApp t u -> showL ++ showR where
 showL = case t of
 TV _ -> show t
 OApp _ _ -> show t
 _ -> "(" ++ show t ++ ")"
 showR = case u of
 TV _ -> ' ':show u
 _ -> "(" ++ show u ++ ")"
instance Show Term where
 show (Lam (x, t) y) = "λ" ++ x ++ showT t ++ showB y where
 showB (Lam (x, t) y) = " " ++ x ++ showT t ++ showB y
 showB expr = '.':show expr
 showT (TV "_") = ""
 showT t = ':':show t
 show (TLam (s, k) t) = "λ" ++ s ++ showK k ++ showB t where
 showB (TLam (s, k) t) = " " ++ s ++ showK k ++ showB t
 showB expr = '.':show expr
 show (Var s) = s
 show (App x y) = showL x ++ showR y where
 showL (Lam _ _) = "(" ++ show x ++ ")"
 showL _ = show x
 showR (Var s) = ' ':s
 showR _ = "(" ++ show y ++ ")"
 show (TApp x y) = showL x ++ "[" ++ show y ++ "]" where
 showL (Lam _ _) = "(" ++ show x ++ ")"
 showL _ = show x
 show (Let x y z) =
 "let " ++ x ++ " = " ++ show y ++ " in " ++ show z
instance Eq Type where
 t1 == t2 = f [] t1 t2 where
 f alpha (TV s) (TV t)
 | Just t' <- lookup s alpha = t' == t
 | Just _ <- lookup t ((\(a,b) -> (b,a)) <$> alpha) = False
 | otherwise = s == t
 f alpha (Forall (s, ks) x) (Forall (t, kt) y)
 | ks /= kt = False
 | s == t = f alpha x y
 | otherwise = f ((s, t):alpha) x y
 f alpha (a :-> b) (c :-> d) = f alpha a c && f alpha b d
 f alpha _ _ = False

Parsing

With 3 different abstractions, we must tread carefully. Different conventions exist for denoting them:

Term → Term

\(\lambda x:T\)

\(\lambda x:T\)

Type → Term

\(\lambda X::K\)

\(\Lambda t:K\)

Type → Type

\(\lambda X::K\)

\(\lambda t:K\)

We use the notation in first column to avoid the uppercase lambda.

Writing \x:X y. was previously equivalent to \x:X.\y. but now X y is parsed as an operator application. One solution is write more lambdas.

We add the typo expression, which is a type-level let expression.

jsEval "curl_module('../compiler/Charser.ob')"
import Charser
data FOmegaLine = Blank | Typo String Type
 | TopLet String Term | Run Term deriving Show
line = between ws eof $ typo <|>
 TopLet <$> v <*> (str "=" *> term) <|> Run <$> term <|> pure Blank where
 typo = Typo <$> between (str "typo") (str "=") v <*> typ
 term = letx <|> lam <|> app
 letx = Let <$> (str "let" *> v) <*> (str "=" *> term)
 <*> (str "in" *> term)
 lam0 = str "\\" <|> str "λ"
 lam1 = str "."
 lam = flip (foldr ($)) <$> between lam0 lam1 (some bind) <*> term where
 bind = (&) <$> v <*>
 ( str "::" *> ((\k s -> TLam (s, k)) <$> kin)
 <|> str ":" *> ((\t s -> Lam (s, t)) <$> typ)
 <|> pure (\s -> TLam (s, Star))
 )
 typ = olam <|> fun
 olam = flip (foldr OLam) <$> between lam0 lam1 (some vk) <*> typ
 fun = oapp `chainr1` (const (:->) <$> str "->")
 oapp = foldl1 OApp <$> some (forallt <|> (TV <$> v)
 <|> between (str "(") (str ")") typ)
 forallt = flip (foldr Forall) <$> between fa0 fa1 (some vk) <*> typ where
 fa0 = str "forall" <|> str "∀"
 fa1 = str "."
 vk = (,) <$> v <*> (str "::" *> kin <|> pure Star)
 kin = ((str "*" *> pure Star) <|> between (str "(") (str ")") kin)
 `chainr1` (const (:=>) <$> str "->")
 app = termArg >>= moreArg
 termArg = (Var <$> v) <|> between (str "(") (str ")") term
 moreArg t = ((App t <$> termArg
 <|> TApp t <$> between (str "[") (str "]") typ) >>= moreArg) <|> pure t
 v = do
 s <- some alphaNumChar
 when (s `elem` words "let in forall typo") $ Charser $ const $ Left $ "unexpected " ++ s
 ws
 pure s
 str = (<* ws) . string
 ws = space *> (string "--" *> many (sat $ const True) *> pure () <|> pure ())
chainr1 p op = go id where
 go d = do
 x <- p
 (op >>= \f -> go (d . (f x:))) <|> pure (foldr ($) x $ d [])

Type-level lambda calculus

In System F, for type-checking, we needed a beta-reduction which substitued a given type variable with a given type value.

This time, this routine is used to build a type-level evaluation function that returns the weak head normal form of a type expression, which in turn is used to compute its normal form.

newName x ys = head $ filter (`notElem` ys) $ (s ++) . show <$> [1..] where
 s = dropWhileEnd (\c -> '0' <= c && c <= '9') x
tBeta (s, a) t = rec t where
 rec (TV v) | s == v = a
 | otherwise = TV v
 rec (Forall (u, k) v)
 | s == u = Forall (u, k) v
 | u `elem` fvs = let u1 = newName u fvs in
 Forall (u1, k) $ rec $ tRename u u1 v
 | otherwise = Forall (u, k) $ rec v
 rec (m :-> n) = rec m :-> rec n
 rec (OLam (u, ku) v)
 | s == u = OLam (u, ku) v
 | u `elem` fvs = let u1 = newName u fvs in
 OLam (u1, ku) $ rec $ tRename u u1 v
 | otherwise = OLam (u, ku) $ rec v
 rec (OApp m n) = OApp (rec m) (rec n)
 fvs = tfv [] a
tEval env (OApp m a) = let m' = tEval env m in case m' of
 OLam (s, _) f -> tEval env $ tBeta (s, a) f where
 _ -> OApp m' a
tEval env term@(TV v) | Just x <- lookup v (fst env) = case x of
 TV _ -> x
 _ -> tEval env x
tEval _ ty = ty
tNorm env ty = case tEval env ty of
 TV _ -> ty
 m :-> n -> rec m :-> rec n
 Forall sk t -> Forall sk (rec t)
 OApp m n -> OApp (rec m) (rec n)
 OLam sk t -> OLam sk (rec t)
 where rec = tNorm env
tfv vs (TV s) | s `elem` vs = []
 | otherwise = [s]
tfv vs (x :-> y) = tfv vs x `union` tfv vs y
tfv vs (Forall (s, _) t) = tfv (s:vs) t
tfv vs (OLam (s, _) t) = tfv (s:vs) t
tfv vs (OApp x y) = tfv vs x `union` tfv vs y
tRename x x1 ty = case ty of
 TV s | s == x -> TV x1
 | otherwise -> ty
 Forall (s, k) t
 | s == x -> ty
 | otherwise -> Forall (s, k) (rec t)
 OLam (s, k) t
 | s == x -> ty
 | otherwise -> OLam (s, k) (rec t)
 a :-> b -> rec a :-> rec b
 OApp a b -> OApp (rec a) (rec b)
 where rec = tRename x x1

Kind checking

We require type lambda expressions to be well-kinded to guarantee strong normalization. Much of the code is similar to type checking for simply typed lambda calculus. A few checks verify that proper types have base type *.

kindOf :: ([(String, Type)], [(String, Kind)]) -> Type -> Either String Kind
kindOf gamma t = case t of
 TV s | Just k <- lookup s (snd gamma) -> pure k
 | otherwise -> Left $ "undefined " ++ s
 t :-> u -> do
 kt <- kindOf gamma t
 when (kt /= Star) $ Left $ "Arr left: " ++ show t
 ku <- kindOf gamma u
 when (ku /= Star) $ Left $ "Arr right: " ++ show u
 pure Star
 Forall (s, k) t -> do
 k' <- kindOf (second ((s, k):) gamma) t
 when (k' /= Star) $ Left $ "Forall: " ++ show k'
 pure Star
 OApp t u -> do
 kt <- kindOf gamma t
 ku <- kindOf gamma u
 case kt of
 kx :=> ky -> if ku /= kx then Left ("OApp " ++ show ku ++ " /= " ++ show kx) else pure ky
 _ -> Left $ "OApp left " ++ show t
 OLam (s, k) t -> (k :=>) <$> kindOf (second ((s, k):) gamma) t

Type checking

For App and TApp, we find the weak head normal form of the first argument to check it is a suitable abstraction. In the case of App, we compare the normal form of the type of the abstraction binding against the normal form of the type of the second argument to check that the application can proceed.

typeOf :: ([(String, Type)], [(String, Kind)]) -> Term -> Either String Type
typeOf gamma t = case t of
 Var s | Just t <- lookup s (fst gamma) -> pure t
 | otherwise -> Left $ "undefined " ++ s
 App x y -> do
 tx <- rec x
 ty <- rec y
 case tEval gamma tx of
 ty' :-> tz | tNorm gamma ty == tNorm gamma ty' -> pure tz
 _ -> Left $ "App: " ++ show tx ++ " to " ++ show ty
 Lam (x, t) y -> do
 k <- kindOf gamma t
 if k == Star then (t :->) <$> typeOf (first ((x, t):) gamma) y else
 Left $ "Lam: " ++ show t ++ " has kind " ++ show k
 TLam (s, k) t -> Forall (s, k) <$> typeOf (second ((s, k):) gamma) t
 TApp x y -> do
 tx <- tEval gamma <$> rec x
 case tx of
 Forall (s, k) t -> do
 k' <- kindOf gamma y
 when (k /= k') $ Left $ "TApp: " ++ show k ++ " /= " ++ show k'
 pure $ tBeta (s, y) t
 _ -> Left $ "TApp " ++ show tx
 Let s t u -> do
 tt <- rec t
 typeOf (first ((s, tt):) gamma) u
 where rec = typeOf gamma

Evaluation

We again erase types as we lazily evaluate a given term.

Because this system is getting complex, it may be better to treat type substitutions as part of the computation to verify our code works as intended. For now, we leave this as an exercise.

eval env (Let x y z) = eval env $ beta (x, y) z
eval env (App m a) = let m' = eval env m in case m' of
 Lam (v, _) f -> eval env $ beta (v, a) f
 _ -> App m' a
eval env (TApp m _) = eval env m
eval env (TLam _ t) = eval env t
eval env term@(Var v) | Just x <- lookup v (fst env) = case x of
 Var v' | v == v' -> x
 _ -> eval env x
eval _ term = term
beta (v, a) f = case f of
 Var s | s == v -> a
 | otherwise -> Var s
 Lam (s, _) m
 | s == v -> Lam (s, TV "_") m
 | s `elem` fvs -> let s1 = newName s fvs in
 Lam (s1, TV "_") $ rec $ rename s s1 m
 | otherwise -> Lam (s, TV "_") (rec m)
 App m n -> App (rec m) (rec n)
 TLam s t -> TLam s (rec t)
 TApp t ty -> TApp (rec t) ty
 Let x y z -> Let x (rec y) (rec z)
 where
 fvs = fv [] a
 rec = beta (v, a)
fv vs (Var s) | s `elem` vs = []
 | otherwise = [s]
fv vs (Lam (s, _) f) = fv (s:vs) f
fv vs (App x y) = fv vs x `union` fv vs y
fv vs (Let _ x y) = fv vs x `union` fv vs y
fv vs (TLam _ t) = fv vs t
fv vs (TApp x _) = fv vs x
rename x x1 term = case term of
 Var s | s == x -> Var x1
 | otherwise -> term
 Lam (s, t) b
 | s == x -> term
 | otherwise -> Lam (s, t) (rec b)
 App a b -> App (rec a) (rec b)
 Let a b c -> Let a (rec b) (rec c)
 TLam s t -> TLam s (rec t)
 TApp a b -> TApp (rec a) b
 where rec = rename x x1
norm env@(lets, gamma) term = case eval env term of
 Var v -> Var v
 -- Record abstraction variable to avoid clashing with let definitions.
 Lam (v, _) m -> Lam (v, TV "_") (norm ((v, Var v):lets, gamma) m)
 App m n -> App (rec m) (rec n)
 Let x y z -> Let x (rec y) (rec z)
 TApp m _ -> rec m
 TLam _ t -> rec t
 where rec = norm env

User Interface

Our user interface code grows uglier still, because to support let expressions, we now must maintain three association lists in the environment: one for terms, one for types, and one for kinds.

eval = do
 es <- map (parse line "") . lines <$> jsEval "input.value;"
 jsEval $ "output.value = `" ++ run [] [] [] es "" ++ "`;"
 where
 run lets types kinds = \case
 [] -> id
 h:t -> let next = run lets types kinds t in case h of
 Left err -> ("parse error: "++) . shows err . ("\n"++) . next
 Right m -> case m of
 Blank -> next
 Run term -> case typeOf (types, kinds) term of
 Left m -> ("type error: "++) . shows term . (": "++) .
 (m++) . ("\n"++) . next
 Right _ -> shows (norm (lets, types) term) . ("\n"++) . next
 Typo s typo -> case kindOf (types, kinds) typo of
 Left m -> ("kind error: "++) . shows typo . (": "++) .
 (m++) . ("\n"++) . next
 Right k -> ("["++) . (s++) . (":"++) .
 shows (tNorm (types, kinds) typo) . ("]\n"++) .
 run lets ((s, typo):types) ((s, k):kinds) t
 TopLet s term -> case typeOf (types, kinds) term of
 Left m -> ("type error: "++) . shows term . (": "++) .
 (m++) . ("\n"++) . next
 Right x -> ("["++) . (s++) . (":"++) . shows x . ("]\n"++) .
 run ((s, term):lets) ((s, x):types) kinds t
jsEval [r|
 evalB.addEventListener("click", (ev) => { repl.run("chat", ["Main"], "eval"); });
|]

Applications

Type operators make System F less unbearable, though in our example the savings are miniscule. We do get to write List X once, which is nice.

Haskell’s type constructors are a restricted form of type operators. In practice, the full power of type operators is rarely needed, so we limit them to simplify type checking.

Above, we saw 3 sorts of abstraction. We’re only missing a way of feeding a term to a function and getting a type, namely dependent types. We can add these while still preserving decidable type checking and strong normalization.

However, real programming languages often support unrestricted recursion and hence it is undecidable whether a term normalizes. Adding dependent types to such a language would lead to undecidable type checking. System Fω is about as far as we can go if we want unrestricted recursion and decidable type checking.

⏴ System F Pure Type Systems ⏵
Contents

Ben Lynn blynn@cs.stanford.edu 💡

AltStyle によって変換されたページ (->オリジナル) /