blog.poucet.org Rotating Header Image

TypeChecking Lambda Calculus using TyperMonad (courtesy of Saizan)

While I had planned to show how to use the typechecker monad presented in in an earlier blog post by showing how to typecheck MiniML, I was positively surprised to find out that people had actually started using due to it’s simplicity. More specifically, Saizan in the #haskell channel on Freenode had written a type-checker for a simple typed lambda-calculus. With his permission, I have presented it here along with some modifications. His original code was a bit shorter, which goes to show how easy the monad is, but I’ve introduced a few helper functions that I should’ve introduced in the original blog post but didn’t because at the time I felt like they would confuse the issue too much. As such, the contents of the actual typechecking is almost the same, except for a few details which I shall mention in the body of this blogpost.


I’m just going to walk straight through the code as it’s rather straightforward.

> module STLC where
> import TypeChecker
> import qualified Data.Map as M
> import Control.Monad
> import MonadScope
> import MonadEnv
> import MonadSupply

First we introduce a function that replaces type variables with other type variables in types, this is rather straight forward. Given an environment, replace variables when you find them.

> ------------------------------------------------------------------------------
> -- substVar : Replaces variables with other variables within a type
> ------------------------------------------------------------------------------
> substVar :: [(Var, Var)] -> MonoType -> MonoType
> substVar subst tx@(TyVar x) =
> case x `lookup` subst of
> Nothing -> tx
> Just y -> TyVar y
> substVar subst (TyConst n ts) =
> TyConst n . map (substVar subst) $ ts
> substVar subst (TyFun ta tb) =
> TyFun (substVar subst ta) (substVar subst tb)

Next is an important function, and something that Saizan missed in his original post. This is actually part of my TypeChecker.hs file which I trimmed down for the previous blogpost. When you have a polytype (some type that says \forall variables. type) and you want to actually use it, you have to make a monotype out of it. Now, a polytype contains a monotype, but it can not simply be used as is. This is something that was missing in the original code, though in the case of this specific code, it would not have made a difference so it. It is important to instantiate a polytype’s free-variables with fresh type-variables, or you can end up over-specifying a type. More specifically, I think this is what makes or breaks the Monomorphism Restriction, though if I’m wrong, feel free to point that out. Just to give you an idea, assume the following code (in -some- functional language)


let id x = x in
(id 'b', id 1)

What will happen, when typechecking this code is that in the scope of the body of that let-clause, the identifier ‘id’ will refer to a polytype (Namely: forall a. a -> a). Now, if this polytype’s forall bit is simply dropped and the mono-type is used, we’ll end up with the ‘a’ being unified both with Char and Integer. Hence, it is important to instantiate each use of ‘id’ to make sure that fresh variables are used instead of ‘a’. The following function does that, using the functionality of substVar.

> ------------------------------------------------------------------------------
> -- instantiate : Instantiates a polytype, making sure all free type-variables
> -- are fresh
> ------------------------------------------------------------------------------
> instantiate :: (TyperMonad m) => PolyType -> m MonoType
> instantiate (TyPoly is t) = do
> is' <- mapM (\x -> supply) is
> return $ substVar (zip is is') t

Because we have scoping, we want to find the type of a variable. The following code finds the poly-type of a let-bound (or lambda-bound) variable in the scope, and then makes sure to instantiate it properly so that we do not get the monomorphism restriction.

> ------------------------------------------------------------------------------
> -- findType : Finds the type of a variable within the current scope
> ------------------------------------------------------------------------------
> findType :: (TyperMonad m) => String -> m MonoType
> findType x = find x >>= (maybe (error $ "Variable not in scope: " ++ x) instantiate) >>= normalize

Obviously, since we’re dealing with scopes, we also have to create new scopes.
The following function comes directly from Saizan’s code, except that I’ve taken out the dual of instantiate, namely generalize. Additionally, sometimes it’s useful to get a poly-type without generalizing any of the free-variables in the monotype (for instance patterns, though at the moment I can’t recall why). Actually, Saizan corrected me after this post went public, lambda-bound variables should be non-generalized (which looking back at my old-code makes sense, since they’re also patterns).

> ------------------------------------------------------------------------------
> -- generalize : Generalizes a monotype to a polytype by generalizing over all
> -- of its free variables
> ------------------------------------------------------------------------------
> generalize :: (TyperMonad m) => MonoType -> m PolyType
> generalize = liftM toPoly . normalize

> ------------------------------------------------------------------------------
> -- nogeneralize : Turns a type into a polytype without generalizing any of
> -- it's free variables (and thus when instantiated, this can lead to the
> -- monomorphism restriction)
> ------------------------------------------------------------------------------
> nogeneralize :: (TyperMonad m) => MonoType -> m PolyType
> nogeneralize = liftM (TyPoly []) . normalize

> ------------------------------------------------------------------------------
> -- withinScope : Applies a TyperMonad action within a nested scope with a new
> -- variable with a given type
> ------------------------------------------------------------------------------
> withinScope :: (TyperMonad m) => String -> MonoType -> m a -> m a
> withinScope x t ma = do
> pt <- nogeneralize t
> scope (M.insert x pt) ma

A last little helper function to generate unique typevariables…

> supplyVar :: TyperMonad m => m MonoType
> supplyVar = liftM TyVar supply

… and we get to the actual lambdacalculus. It’s a very simple calculus, with only three forms: identifiers (variables), lambda abstractions and function application.

> ------------------------------------------------------------------------------
> -- Simple Typed Lambda Calculus
> ------------------------------------------------------------------------------
> type Id = String
> data Expr = Var Id
> | Lambda Id Expr
> | App Expr Expr

Here the code differs just -slightly- with Saizan’s original code. Instead of normalizing after checking each sub-expression, it is perfectly fine to normalize at the end. Even if you were decorating the expression-tree, you’d still have to wait until the end, because only then could all types be normalized. So there’s no point, really, in normalizing before typechecking (and thus unification) is finished.

> typeCheck :: (TyperMonad m) => Expr -> m MonoType
> typeCheck e = check e >>= normalize
> where check (Var x) = findType x
> check (Lambda x b) = do
> tx <- supplyVar
> tb <- withinScope x tx $ check b
> return $ TyFun tx tb
> check (App a b) = do
> ta <- check a
> tb <- check b
> te <- supplyVar
> unify ta (TyFun tb te)
> return te

There’s a few alterations, but the gist of Saizan’s core which prompted this blog-post is all there, and I want to thank him for this great illustration of how to use the typermonad, since I did not think about it myself at all. Notice that the only code that would change in the case of a different type-checker is the code that came after the definition of the lambda-calculus. All the code that came before that were generic helper functions that I did not introduce in the prior blog post as then they would’ve not been clear as to why they are useful, but they -should- go into TypeChecker.hs.

Finally, let’s show some examples that Saizan originally showed, and see how nicely they typecheck :)

> id' = Lambda "x" (Var "x")
> flip' = Lambda "f" (Lambda "a" (Lambda "x" (App (App (Var "f") < font color="Green">(Var "x")) (Var "a"))))
> run m = evalTyperT M.empty m
>
> -- *STLC> run $ typeCheck (App flip' id')
> -- b -> (b -> g) -> g

As usual, this blogpost is a literate haskell file, and if you have any comments, questions or suggestions, feel free to leave a comment on my blog :)

Be Sociable, Share!

Leave a Reply

Your email address will not be published. Required fields are marked *

*

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>