# Types and Programming Languages

17 Jan 2015Note: This is a summary of the book *Types and Programming Languages*.

## Table of Contents

- Operational Semantics
- Lambda Calculus
- Simply Typed Lambda Calculus
- Reference Type
- Universal Type
- Type Reconstruction
- Existential Type
- Recursive Type
- Subtyping
- Object
- Featherweight Java
- Bounded Quantification
- Kinding and Higher-Order Polymorphism
- Higher-Order Bounded Polymorphism

## Operational Semantics

An abstract machine consists of:

- a set of states
- a transition relation on states

A state records all the information in the machine at a given moment.

A normal form is a term that cannot be evaluated any further.

## Lambda Calculus

Lambda calculus is the foundation for many programming languages. It’s more widely used than the Turing machine model in programming language studies due to its nice mathematical properties.

- Turing complete
- Everything is a function
- variables always denote functions
- function always take functions as parameters

Syntactically, “everything is a function” doesn’t work, we need variables. Ontologically it works, all closed lambda terms denote standalone functions.

There are several evaluation strategies for lambda calculus, call by value is the most widely used.

- Call by value(most languages)
- Call by name(Haskell)
- Normal order(left-most/outer-most)
- Full-beta

Call by value specification:

SYNTAX t ::= terms x variable λx.t abstraction t t application EVALUATION (λx.t) v ⟶ [x ⟼ v]t (E-AppAbs) t1 ⟶ t1' ------------------ (E-App1) t1 t2 ⟶ t1' t2 t2 ⟶ t2' ------------------ (E-App2) v1 t2 ⟶ v1 t2'

We can represent numbers, booleans, pairs, recursion functions, etc based on lambda calculus:

tru = λt. λf. t fls = λt. λf. f and = λb. λc. b c fls or = λb. λc. b tru c not = λb. b fls tru pair = λf.λs.λb. b f s fst = λp. p tru snd = λp. p fls 0 = λs. λz. z 1 = λs. λz.s z 2 = λs. λz.s (s z) 3 = λs. λz.s (s (s z)) ... scc= λn. λs. λz.s(n s z) plus= λm. λn. λs. λz.ms(n s z) times = λm. λn. λs. λz. m (n s) z power = λm. λn. n (times m m) fix = λf. (λx. f (λy. x x y)) (λx. f (λy. x x y)) fact = fix ( λfct. λn. if iszero n then 1 else n * (fct (pred n)) )

### behavioral equivalence

Two lambda terms are observationally equivalent if they both diverge or reduce to a value.

Two lambda terms are behaviorally equivalent if they are observationally equivalent to all possible applications.

## Simply Typed Lambda Calculus

Typing helps exclude certain bad terms – sometimes with the price of excluding some meaningful terms.

### Structural Types VS. Nominal Types

For *structural type systems*, what matters about a type (for typing, subtyping, etc.) is just
its structure. Names are just convenient (but inessential) abbreviations.

For *nominal type systems*:

- Types are always named.
- Typechecker mostly manipulates names, not structures.
- Subtyping is declared explicitly by programmer.

Structural types are simpler, cleaner, and more elegant, while nominal types make *recursive types* easier, make typechecking more efficient, and make it possible to use type name in runtime.

### Type Safety

A type relation is sound if it satisfied preservation and progress.

- preservation. If t:T and t -> t’, then t’:T.
- progress. If t:T, either t is a value or there exists t’ such that t -> t’.

To define and verify a type system, you must:

- Define types
- Specify typing rules
- Prove soundness: progress and preservation

### Specification of STLC

The main usage of types is to exclude invalid terms. Type checking happens at compile time, they have nothing to do with evaluation, thus can be erased before evaluation.

The pure STLC is not interesting, as it only has one type ->. Variants of STLC are augmented with primary types like Bool or Nat.

SYNTAX t ::= terms x variable λx:T.t abstraction t t application true constant true false constant false if t then t else t conditional v ::= values x.t abstraction value true true value false false value T ::= types Bool type of booleans T → T types of functions Γ ::= contexts ∅ empty context Γ, x:T term variable binding EVALUATION (λx.t) v ⟶ [x ⟼ v]t (E-AppAbs) t1 ⟶ t1' ------------------ (E-App1) t1 t2 ⟶ t1' t2 t2 ⟶ t2' ------------------ (E-App2) v1 t2 ⟶ v1 t2' TYPING true : Bool (T-True) false : Bool (T-False) Γ ⊢ t1: Bool, t2: T, t3: T ---------------------------- (T-If) Γ ⊢ if t1 then t2 else t3: T Γ, x: T1 ⊢ t: T2 --------------------- (T-Abs) Γ ⊢ λx:T1.t : T1 → T2 x:T ∈ Γ ---------- (T-Var) Γ ⊢ x : T Γ ⊢ t1: T → T', t2: T ----------------------- (T-App) Γ ⊢ t1 t2: T

### Ascription

SYNTAX: t ::= ... terms t as T ascription EVALUATION v1 as T ⟶ v1 (E-Ascribe) t1 ⟶ t1' -------------------- (E-Ascribe1) t1 as T ⟶ t1' as T TYPING Γ ⊢ t1: T -------------------- (T-Ascribe) Γ ⊢ t1 as T: T

### Let

SYNTAX t ::= ... terms let x=t in let binding EVALUATION let x=v1 in t2 ⟶ [x1 ⟼ v1]t2 (E-LetV) t1 ⟶ t1' ----------------------------------- (E-Let) let x=t1 in t2 ⟶ let x=t1' in t2 TYPING Γ ⊢ t1: T1, Γ, x: T1 ⊢ t2: T2 ---------------------------------- (T-Let) Γ ⊢ let x=t1 in t2: T2

### Tuple

SYNTAX t ::= ... terms { t(i) | i ∈ 1..n } tuple t.i projection v ::= ... values { v(i) | i ∈ 1..n } tuple value T ::= ... types { T(i) | i ∈ 1..n } tuple type EVALUATION { v(i) | i ∈ 1..n }.j ⟶ v(j) (E-ProjTuple) t1 ⟶ t1' ------------- (E-Proj) t1.j ⟶ t1'.j t(j) ⟶ t'(j) --------------------------------------------- (E-Tuple) { v(i), t(j), t(k) | i ∈ 1..j-1, k ∈ j+1..n } ⟶ { v(i), t'(j), t(k) | i ∈ 1..j-1, k ∈ j+1..n } TYPING TYPING for each i Γ ⊢ t(i) : T(i) --------------------------------------------- (T-Tuple) Γ ⊢ { t(i) | i ∈ 1..n }: { T(i) | i ∈ 1..n } Γ ⊢ t1: { T(i) | i ∈ 1..n } ------------------------------ (T-Proj) Γ ⊢ t1.j : T(j)

### Record

SYNTAX t ::= ... { l(i) = t(i) | i ∈ 1..n } record t.l projection v ::= ... { l(i) = v(i) | i ∈ 1..n } record value T ::= ... { l(i):T(i) | i ∈ 1..n } type of records EVALUATION { l(i) = v(i) | i ∈ 1..n }.l(j) ⟶ v(j) (E-ProjRcd) t1 ⟶ t1' ----------------- (E-Proj) t1.l ⟶ t1'.l t(j) ⟶ t'(j) ------------------------------------------ (E-Rcd) { l(i) = v(i), l(j) = t(j), l(k) = t(k) | i ∈ 1..j-1, k ∈ j+1..n } ⟶ { l(i) = v(i), l(j) = t'(j), l(k) = t(k) | i ∈ 1..j-1, k ∈ j+1..n } TYPING for each i Γ ⊢ t(i) : T(i) ----------------------------------------------------------- (T-Rcd) Γ ⊢ { l(i) = t(i) | i ∈ 1..n }: { l(i) : T(i) | i ∈ 1..n } Γ ⊢ t1: { l(i) : T(i) | i ∈ 1..n } ---------------------------------- (T-Proj) Γ ⊢ t1.l(j) : T(j)

### Sum

SYNTAX t ::= ... inl t as T inr t as T v ::= ... inl v as T inr v as T TYPING Γ ⊢ t1 : T1 --------------------------- (T-Inl) Γ ⊢ inl t1 as T1+T2 : T1+T2 Γ ⊢ t1 : T1 --------------------------- (T-Inr) Γ ⊢ inr t1 as T1+T2 : T1+T2 EVALUATION case (inl v0 as T0) of inl x1⟹t1 | inr x2⟹t2 ⟶ [x1 ⟼ v0]t1 (E-CaseInl) case (inr v0 as T0) of inl x1⟹t1 | inr x2⟹t2 ⟶ [x2 ⟼ v0]t2 (E-CaseInr) t1 -> t1' ------------------------------- (E-Inl) inl t1 as T2 ⟶ inl t1' as T2 t1 -> t1' ------------------------------- (E-Inr) inr t1 as T2 ⟶ inr t1' as T2

### Variants

SYNTAX: t ::= ... Terms < l=t > as T tagging case t of < l(i) = x(i) > ⟹ t(i) case T ::= ... Types < l(i): T(i) | i ∈ 1..n > types of variants EVALUATION case (< l(j) = v(j) > as T) of (E-CaseVariant) < l(1) = x(1) > ⟹ t(1) < l(2) = x(2) > ⟹ t(2) ... ⟶ [x(j) ⟼ v(j)]t(j) t0 ⟶ t0' ------------------------- (E-Case) case t0 of < l(1) = x(1) > ⟹ t(1) < l(2) = x(2) > ⟹ t(2) ... ⟶ case t0' of < l(1) = x(1) > ⟹ t(1) < l(2) = x(2) > ⟹ t(2) ... t(i) ⟶ t'(i) --------------------- (E-Variant) < l(i)=t(i) > as T ⟶ < l(i)=t'(i) > as T TYPING Γ ⊢ t(j) : T(j) ----------------------------------------------- (T-Variant) Γ ⊢ < l(i)=t(i) > as < l(i): T(i) | i ∈ 1..n > : < l(i): T(i) | i ∈ 1..n > Γ ⊢ t0: < l(i): T(i) | i ∈ 1..n > for each i Γ, x(i): T(i) ⊢ t(j) : T ------------------------------------------- (T-Case) Γ ⊢ case t0 of < l(1) = x(1) > ⟹ t(1) < l(2) = x(2) > ⟹ t(2) ... : T

Example:

Addr = < physical:PhysicalAddr, virtual:VirtualAddr >; a = < physical=pa > as Addr; getName = λa:Addr. case a of < physical=x > => x.firstlast | < virtual=y > => y.name

Options and enumerations are special cases of variants.

### Recursion

In STLC, all terms terminate. It’s impossible to implement recursion using the fix combinator as in untyped lambda calculus. However, it’s possible to introduce the *fix* combinator into the language to restore the ability to do recursive computation.

SYNTAX t ::= ... terms fix t fix point of t EVALUATION fix (λx:T1. t2) ⟶ (E-FixBeta) [x ⟼ (fix (λx:T1. t2))]t2 t ⟶ t' --------------- (E-Fix) fix t ⟶ fix t' TYPING Γ ⊢ t: T → T ------------- (T-Fix) Γ ⊢ fix t: T

To make the usage of fix combinator easier, the syntactic sugar *letrec* is introduced:

letrec x:T1=t1 in t2 ⟹ let x = fix (λx:T1.t1) in t2 letrec iseven : Nat!Bool = λx:Nat. if iszero x then true else if iszero (pred x) then false else iseven (pred (pred x)) in iseven 7

## Reference Type

Mutable state is a popular programming model. How can we model mutable states? Mutable states doesn’t only bring changes to the model, but also things that don’t change. As change is only meaningful if it can happen to something that doesn’t change. A common approach to model mutable state is to represent the unchanged as a location or cell, so that the changes can be modeled as the things that the location or cell holds.

Though mutable state makes a lot of things natural and simple, it introduces the problem of aliasing. With the introduction of storage cells, it’s possible that there are multiple references to the same cell in different places of a program under different names. Without some careful coordination or well-defined protocol, this could lead to a chaos.

Another issue with mutable state is that now the order of evaluation matters. Lazy evaluation doesn’t make sense in the context of mutable state.

SYNTAX t ::= ... terms ref t reference creation !t dereference t:=t assignment l store location v ::= ... values l store location EVALUATION t1 | μ ⟶ t1' | μ' ------------------------------- (E-Assign1) t1 := t2 | μ ⟶ t1' := t2 | μ' t2 | μ ⟶ t2' | μ' ------------------------------- (E-Assign2) v1 := t2 | μ ⟶ v1 := t2' | μ' l := v2 | μ ⟶ unit | [l ⟼ v2]μ (E-Assign) t | μ ⟶ t' | μ' ------------------------- (E-Ref) ref t | μ ⟶ ref t' | μ' l ∉ dom(μ) ---------------------------- (E-RefV) ref v | μ ⟶ l | (μ, l ⟼ v) t | μ ⟶ t' | μ' -------------------- (E-DeRef) !t | μ ⟶ !t' | μ' μ(l) = v ----------------- (E-DerefLoc) !l | μ ⟶ v | μ (λx.t) v | μ ⟶ [x ⟼ v]t | μ (E-AppAbs) t1 | μ ⟶ t1' | μ' -------------------------- (E-App1) t1 t2 | μ ⟶ t1' t2 | μ' t2 | μ ⟶ t2' | μ' -------------------------- (E-App2) v1 t2 | μ ⟶ v1 t2' | μ' TYPING Σ(l) = T ------------------- (T-Loc) Γ | Σ ⊢ l: Ref T Γ|Σ ⊢ t: T ---------------------- (T-Ref) Γ|Σ ⊢ ref t: Ref T Γ|Σ ⊢ t: Ref T -------------- (T-Deref) Γ|Σ ⊢ !t: T Γ|Σ ⊢ t1: Ref T Γ|Σ ⊢ t2: Ref T ----------------------------------- (T-Assign) Γ|Σ ⊢ t1:=t2: Unit

Note that in the above, we augment each evaluation rule with the evaluation environment μ in order to track state of the program. In the typing rules, the store typing Σ is introduced to speed up type checking, instead of computing it from *ref t* expressions.

## Universal Type

Universal type can model polymorphism in programming languages. The logic theory behind universal type is *System F*, which is introduced by Jean-Yves Girard (1972). It is also sometimes called the second-order lambda-calculus, because it corresponds, via the Curry-Howard correspondence, to second-order intuitionistic logic, which allows quantification not only over individuals [terms], but also over predicates [types].

SYNTAX t ::= ... terms λX.t type abstraction t [T] type application v ::= ... values λX.t type abstraction value T ::= ... types X type variable ∀X.T universal type Γ ::= contexts ∅ empty context Γ, x:T term variable binding Γ, X type variable binding EVALUATION t1 ⟶ t1' ------------------- (E-TApp) t1 [T2] ⟶ t1' [T2] (λX.t) [T] ⟶ [X ⟼ T]t (E-TappTabs) TYPING Γ , X ⊢ t : T ---------------- (T-TAbs) Γ ⊢ λX.t : ∀X.T Γ ⊢ t : ∀X.T ------------------------- (T-TApp) Γ ⊢ t [T1] : [X ⟼ T1]T

In simply typed lambda calculus, there’s no way to type the term “λx. x x”. However, it’s possible in System F:

```
term: λx:∀X.X→X. x [∀X.X→X] x
type: (∀X. X→X) → (∀X. X → X)
```

It’s possible to type Church encodings in System F:

CBool = ∀X.X→X→X true = λX. λt:X. λf:X. t false = λX. λt:X. λf:X. f CNat = ∀X. (X→X) → X → X 0 = λX. λs:X→X. λz:X. z 1 = λX. λs:X→X. λz:X. s z

## Type Reconstruction

- Type checking: Given Γ, t and T, check whether Γ ⊢ t : T.
- Type reconstruction: Given Γ and t, find a type T such that Γ ⊢ t : T.

In general, a type reconstruction algorithm A assigns to an environment !and a term t a set of types 𝒜(Γ, t).

The algorithm is **sound** if for every type T ∈ 𝒜(Γ, t) we can prove the judgment Γ ⊢ t : T.

The algorithm is **complete** if for every provable judgment Γ ⊢ t : T we have that T ∈ 𝒜(Γ, t).

Type reconstruction with subtyping and universal types are difficult. Here we only concern us with simple types. The type reconstruction process first generates constraint equations from terms, then use the *unification* algorithm to solve the equations.

Following are the rules for generating constraint equations:

x: T ∈ Γ -------------- (CT-VAR) Γ ⊢ x: T |∅ {} Γ, x:T1 ⊢ t2: T2 |𝓍 C ------------------------- (CT-Abs) Γ ⊢ λx:T1.t2 : T1→T2 |𝓍 C Γ ⊢ t1 : T1 |𝓍1 C1 Γ ⊢ t2 : T2 |𝓍2 C2 𝓍1 ∩ 𝓍2 = 𝓍1 ∩ FV(T2) = 𝓍2 ∩ FV(T1) = ∅ X ∉ 𝓍1, 𝓍2, T1, T2, C1, C2, Γ, t1, or t2 C′ = C1 ∪ C2 ∪ {T1 = T2→X} ----------------------------------------- (CT-App) Γ ⊢ t1 t2 : X |𝓍1 ∪ 𝓍2 ∪ {X} C′ Γ ⊢ true : Bool |∅ {} (CT-True) Γ ⊢ false : Bool |∅ {} (CT-False) Γ ⊢ t1 : T1 |𝓍1 C1 Γ ⊢ t2 : T2 |𝓍2 C2 Γ ⊢ t3 : T3 |𝓍3 C3 𝓍1, 𝓍2, 𝓍3 nonoverlapping C′ = C1 ∪ C2 ∪ C3 ∪ {T1 = Bool, T2 = T3} ----------------------------------------------- (CT-If) Γ ⊢ if t1 then t2 else t3 : T2 |𝓍1 ∪ 𝓍2 ∪ 𝓍3 C′

The unification algorithm is due to Robinson(1965), its usage in type reconstruction is due to Hindley(1969) and Milner (1978).

unify(C) = if C = ∅, then [] else let {S = T} ∪ C′ = C in if S = T then unify(C′) else if S = X and X ∉ FV(T) then unify([X ⟼ T]C′) ◦ [X ⟼ T] else if T = X and X ∉ FV(S) then unify([X ⟼ S]C′) ◦ [X ⟼ S] else if S = S1→S2 and T = T1→T2 then unify(C′ ∪ {S1 = T1, S2 = T2}) else fail

Type reconstruction with universal type is hard. However, type reconstruction for a fragment of System F has been successfully implemented in ML by Milner(1978) under the name *let-polymorphism*.

The essence of *let-polymorphism* is to allow multiple different type bindings to the same function. This idea can be easily implemented as follows:

Γ ⊢ [x ⟼ t1]t2 : T2 Γ ⊢ t1 : T1 ------------------------------------- (T-LetPoly) Γ ⊢ let x=t1 in t2 : T2

However, if there’re many occurrences of x in t1, the type checker would need to do a lot of duplicate work.

A more efficient approach is to bind x to a *type scheme*, and instantiate the type scheme in each usage of x.

## Existential Type

Existential type provides an elegant foundation for data abstraction and information hiding. Existential type corresponds to the existential quantification in logic.

SYNTAX t ::= ... terms {*T,t} as T packing let {X,x}=t in t unpacking v ::= ... values {*T,v} as T package value T ::= ... types ∃X.T existential type EVALUATION let {X,x}=({*T11,v12} as T1) in t2 (E-UnpackPack) ⟶ [X ⟼ T11][x ⟼ v12]t2 t ⟶ t' ------------------------------------- (E-Pack) {*T11, t} as T1 ⟶ {*T11, t'} as T1 t1 ⟶ t1' ------------------------------------- (E-Unpack) let {X,x}=t1 in t2 ⟶ let {X,x}=t1' in t2 TYPING Γ ⊢ t2 : [X ⟼ U]T2 ------------------------------- (T-Pack) Γ ⊢ {*U,t2} as ∃X.T2: ∃X.T2 Γ ⊢ t1 : ∃X.T12 Γ, X, x:T12 ⊢ t2 : T2 ---------------------------------------- (T-Unpack) Γ ⊢ let {X,x}=t1 in t2 : T2

Special notes in TAPL(Chapter 24):

(1) In the rule *T-Unpack*, the type variable X appears in the context
in which t2’s type is calculated, but does not appear in the context
of the rule’s conclusion. This means that the result type T2 cannot
contain X free, since any free occurrences of X will be out of scope
in the conclusion.

(2) In the rule *T-Pack*, the type U doesn’t appear in the type
signature. It means that packages with different hidden representation
types can inhabit the same existential type.

Following is an example of information hiding based on existential type.

ADT counter = type Counter representation Nat signature new : Counter, get : Counter→Nat, inc : Counter→Counter; operations new = 1, get = λi:Nat. i, inc = λi:Nat. succ(i) counter.get (counter.inc counter.new)

The code above can be encoded as follows:

counterADT = {*Nat, { new = 1, get = λi:Nat. i, inc = λi:Nat. succ(i) } } as ∃Counter. { new: Counter, get: Counter→Nat, inc: Counter→Counter } }; let {Counter,counter} = counterADT in counter.get (counter.inc counter.new);

It’s possible to encode existential type in universal type:

∃X.T = ∀Y. (∀X. T→Y) → Y {*S,t} as ∃X.T = λY. λf:(∀X.T→Y). f [S] t let {X,x}=t1 in t2 = t1 [T2] (λX. λx:T11. t2)

## Recursive Type

Recursive types are everywhere, lists and trees are the most common. Ideally, we can define a recursive type NatList as follows:

```
NatList = <nil:Unit, cons:{Nat,NatList}>
```

If we’re working with nominal types, then everything is simple and straight-forward. If we limit ourself to structural types, then we have to eliminate the names in the type:

```
NatList = μX. <nil:Unit, cons:{Nat,X}>
```

There are generally two approaches to recursive type depending on how they answer following question:

What is the relation between the type μX.T and its one-step unfolding?

For example, what is the relation between NatList and < nil:Unit,cons:{Nat,NatList} >?

- The
**equi-recursive**approach takes these two type expressions as definitionally equal - The
**iso-recursive**approach takes a recursive type and its unfolding as different, but isomorphic

In a system with iso-recursive types, we have to introduce, for each recursive type μX.T, a pair of functions:

```
unfold[μX.T] : μX.T → [X ⟼ μX.T] T
fold[μX.T] : [X ⟼ μX.T] T → μX.T
```

Both approaches are widely used in theoretical studies and practical language design. The equi-recursive approach is more intuitive but difficult to work with bounded quantification and type operators. The iso-recursive approach is heavier in notation, but in practical languages they can be done implicitly by the compiler.

The specification for iso-recursive types (λμ) is listed below:

SYNTAX t ::= ... terms fold [T] t folding unfold [T] t unfolding v ::= ... values fold [T] v folding T ::= ... types X type variable μX.T recursive type EVALUATION unfold [T] (fold [T] v) ⟶ v (E-UnfldFld) t ⟶ t' ------------------------- (E-Fld) fold [T] t ⟶ fold [T] t' t ⟶ t' ----------------------------- (E-Unfld) unfold [T] t ⟶ unfold [T] t' TYPING U = μX.T1 Γ ⊢ t1:[X ⟼ U]T1 ------------------------------ (T-Fld) Γ ⊢ fold [U] t1 : U U = μX.T1 Γ ⊢ t1:U ------------------------------ (T-Unfld) Γ ⊢ unfold [U] t1 : [X ⟼ U]T1

## Subtyping

Why we need subtyping? See the example below:

```
(λr:{x:Nat}. r.x) {x=0,y=1}
```

The term doesn’t type check, however we would like it to type check.

SUBTYPING Γ ⊢ t : S S ＜: T --------------------- (T-Sub) Γ ⊢ t : T S ＜: S (S-Refl) S ＜: U U ＜: T ------------------ (S-Trans) S ＜: T {l(i):T(i) | i ∈ 1..n+k} ＜: {l(i):T(i) | i ∈ 1..n} (S-RcdWidth) for each i Si ＜: Ti --------------------------------------------------- (S-RcdDepth) {l(i):S(i) | i ∈ 1..n} ＜: {l(i):T(i) | i ∈ 1..n} {k(j):S(j) | j ∈ 1..n} is a permutation of {l(i):T(i) | i ∈ 1..n} ----------------------------------------------------------------- (S-RcdPerm) {k(j):S(j) | j ∈ 1..n} ＜: {l(i):T(i) | i ∈ 1..n} T1 ＜: S1 S2 ＜: T2 ----------------------- (S-Arrow) S1 → S2 ＜: T1 → T2 S ＜: Top (S-Top)

We can also introduce a bottom type Bot in addition to the top type Top, to make the typing relation a lattice. Note that in the S-Arrow rule, the parameter is contra-variant, while the return value is covariant.

The subtype relation in the above is structural: that’s to decide whether S is a subtype of T by examining the structure of S and T. In real world, most languages uses nominal typing. Nominal typing makes recursive types easier, while structural typing is more flexible.

Ref is neither covariant nor contra-variant. That’s the same with Array due to side effects. Java makes Array covariant, which is a mistake in design!

One technique to make Ref work with subtyping is to split its type for read and write, and make read covariant, make write contra-variant.

The intricacy related to reference is caused by the ability to change the type of the cell during execution, similar to the problem in universal type.

## Object

Most of class-based object-oriented programming features can be represented with STLC with extensions like reference, records, recursion and subtyping. We’ll show how to represent following object features:

- dynamic dispatch
- encapsulation of state
- inheritance
- this
- super

### Encapsulation of State

// Java class Counter { protected int x = 1; // Hidden state int get() { return x; } void inc() { x++; } } void inc3(Counter c) { c.inc(); c.inc(); c.inc(); } // STLC Counter c = new Counter(); inc3(c); inc3(c); c.get(); newCounter = λ_:Unit. let r = {x=ref 1} in { get = λ_:Unit. !(r.x), inc = λ_:Unit. r.x:=succ(!(r.x)) };

Note that encapsulation of state is a form of information hiding. A different approach of information hiding is *Abstract Data Type*(ADT).

An ADT comprises:

- A hidden representation type X
- A collection of operations for creating and manipulating elements of type X.

The theory behind ADT is *existential type*. In the OO community, the term “abstract data type” is used as a synonym for “object type”, this usage is very different from its standard usage in academics.

### Subtyping and Inheritance

// Java class Counter { protected int x = 1; int get() { return x; } void inc() { x++; } } class ResetCounter extends Counter { void reset() { x = 1; } } class BackupCounter extends ResetCounter { protected int b = 1; void backup() { b = x; } void reset() { x = b; } } // STLC counterClass = λr:CounterRep. { get = λ_:Unit. !(r.x), inc = λ_:Unit. r.x:=succ(!(r.x)) }; newCounter = λ_:Unit. let r = {x=ref 1} in counterClass r; resetCounterClass = λr:CounterRep. let super = counterClass r in { get = super.get, inc = super.inc, reset = λ_:Unit. r.x:=1 }; newResetCounter = λ_:Unit. let r = { x=ref 1 } in resetCounterClass r; BackupCounter = { get:Unit→Nat, inc:Unit→Unit, reset:Unit→Unit, backup: Unit→Unit }; BackupCounterRep = { x: Ref Nat, b: Ref Nat }; backupCounterClass = λr:BackupCounterRep. let super = resetCounterClass r in { get = super.get, inc = super.inc, reset = λ_:Unit. r.x:=!(r.b), backup = λ_:Unit. r.b:=!(r.x) }

### this

In Java, we can call between class methods:

class SetCounter { protected int x = 0; int get () { return x; } void set (int i) { x = i; } void inc () { this.set( this.get() + 1 ); } }

First try:

setCounterClass = λr:CounterRep. fix (λthis: SetCounter. { get = λ_:Unit. !(r.x), set = λi:Nat. r.x:=i, inc = λ_:Unit. this.set (succ (this.get unit)) });

The problem with the implementation above is that the binding of *this* is closed. However, most OO languages support late binding of *this*, also known as *open recursion through this*. Open recursion through this allows the methods of a superclass to call the methods of a subclass, even though the subclass does not exist when the superclass is being defined.

We can solve the problem by moving *fix* out of *setCounterClass* to *newSetCounter*.

setCounterClass = λr:CounterRep. λthis: SetCounter. { get = λ_:Unit. !(r.x), set = λi:Nat. r.x:=i, inc = λ_:Unit. this.set (succ(this.get unit)) } newSetCounter = λ_:Unit. let r = {x=ref 1} in fix (setCounterClass r) InstrCounterRep = {x: Ref Nat, a: Ref Nat} InstrCounter = { get:Unit→Nat, set:Nat→Unit, inc:Unit→Unit, accesses:Unit→Nat} instrCounterClass = λr:InstrCounterRep. λthis: InstrCounter. let super = setCounterClass r this in { get = super.get, set = λi:Nat. (r.a:=succ(!(r.a)); super.set i), inc = super.inc, accesses = λ_:Unit. !(r.a) } newInstrCounter = λ_:Unit. let r = {x=ref 1, a=ref 0} in fix (instrCounterClass r)

A problem with the version above is that *newInstrCounter unit* will diverge – we’re using *this* too early in creating *super*. An inelegant solution to the problem is to delay the usage of *this* in `λ_:Unit t`

.

setCounterClass = λr:CounterRep. λthis: Unit→SetCounter. λ_:Unit. { get = λ_:Unit. !(r.x), set = λi:Nat. r.x:=i, inc = λ_:Unit. (this unit).set(succ((this unit).get unit)) } newSetCounter = λ_:Unit. let r = {x=ref 1} in fix (setCounterClass r) unit instrCounterClass = λr:InstrCounterRep. λthis: Unit→InstrCounter. λ_:Unit. let super = setCounterClass r this unit in { get = super.get, set = λi:Nat. (r.a:=succ(!(r.a)); super.set i), inc = super.inc, accesses = λ_:Unit. !(r.a) } newInstrCounter = λ_:Unit. let r = {x=ref 1, a=ref 0} in fix (instrCounterClass r) unit

This solution is ugly because *this* has become *this unit* in every place. A more elegant approach is to use reference.

setCounterClass = λr:CounterRep. λthis: Source SetCounter. { get = λ_:Unit. !(r.x), set = λi:Nat. r.x:=i, inc = λ_:Unit. (!this).set (succ ((!this).get unit)) } dummySetCounter = { get = λ_:Unit. 0, set = λi:Nat. unit, inc = λ_:Unit. unit } newSetCounter = λ_:Unit. let r = {x=ref 1} in let cAux = ref dummySetCounter in (cAux := (setCounterClass r cAux); !cAux)

Note that in the above we use `Source SetCounter`

instead of `Ref SetCounter`

to enable subtyping with reference. `Source`

makes read covariant, `Sink`

makes write contravariant.

The implementation is based on the trick that we can implement recursive functions with reference:

factref = ref λx:Nat. 0 fact = λx:Nat. if iszero(x) then 1 else x*(!factref)(x-1) factref := fact

## Featherweight Java

Featherweight Java was introduced by Atsushi Igarashi(2001) to model “core OO features” and their types. It removed from Java many features:

- Reflection, concurrency, class loading, inner classes, …
- Exceptions, loops, …
- Interfaces, overloading, …
- Assignment

The features that are supported:

- Classes and objects
- Methods and method invocation
- Fields and field access
- Inheritance (including open recursion through this)
- Casting

Syntax:

t ::= terms x variables t.f field access t.m([t]) method invocation new C([t]) object creation (C) t cast v ::= values new C([v]) K ::= constructor declarations C(C [f]) {super([f]); [this.f=f;]} M ::= method declarations C m([C x]) {return t;} CL ::= class declarations class C extends C {[C f;] K [M]}

Auxiliary methods:

FIELDS LOOKUP fields(Object) = ∅; CT(C) = class C extends D {[C f;] K [M]} fields(D) = [D g] ---------------------------------------- fields(C) = [D g], [C f] METHOD TYPE LOOKUP CT(C) = class C extends D {[C f;] K [M]} B m ([B x]) {return t;} ∈ [M] ----------------------------------------- mtype(m, C) = [B] → B CT(C) = class C extends D {[C f;] K [M]} m is not defined in [M] ---------------------------------------- mtype(m, C) = mtype(m, D) METHOD BODY LOOKUP CT(C) = class C extends D {[C f;] K [M]} B m ([B x]) {return t;} ∈ [M] ---------------------------------------- mbody(m, C) = ([x], t) CT(C) = class C extends D {[C f;] K [M]} m is not defined in [M] ---------------------------------------- mbody(m, C) = mbody(m, D) VALID METHOD OVERRIDING mtype(m, D) = [D]→D0 implies [C] = [D] and C0 = D0 -------------------------------------------------- override(m, D, [C]→C0)

Evaluation rules:

fields(C) = [C f] ------------------------- (E-ProjNew) (new C([v])).f(i) ⟶ v(i) mbody(m, C) = ([x], t) ------------------------------- (E-InvkNew) (new C([v])).m([u]) ⟶[[x]⟼[u], this⟼new C([v])]t C ＜: D ----------------------------- (E-CastNew) (D)(new C([v])) ⟶ new C([v]) t ⟶ t' ----------- (E-Field) t.f ⟶ t'.f t ⟶ t' --------------------- (E-Invk-Recv) t.m([t]) ⟶ t'.m([t]) t ⟶ t' ---------------------- (E-Invk-Arg) v.m([v], t, [t]) ⟶ v.m([v], t', [t]) t ⟶ t' ---------------------- (E-New-Arg) new C([v], t, [t]) ⟶new C([v], t', [t]) t ⟶ t' ---------------- (E-Cast) (C)t ⟶ (C)t'

Typing rules:

CT(C) = class C extends D {...} ------------------------------- C ＜: D C ＜: C C ＜: D D ＜: E ------------------ C ＜: E x:C ∈ Γ --------- (T-Var) Γ ⊢ x : C Γ ⊢ t : C0 fields(C0) = [C f] -------------------------------- (T-Field) Γ ⊢ t.f(i) : C(i) Γ ⊢ t: D D ＜: C --------------------- (T-UCast) Γ ⊢ (C)t : C Γ ⊢ t: D C ＜: D C ≠ D --------------------------- (T-DCast) Γ ⊢ (C)t : C Γ ⊢ t: C0 mtype(m, C0) = [D] → C Γ ⊢ [t: C] [C ＜: D] ----------------------- (T-Invk) Γ ⊢ t.m([t]) : C [x:C], this:C ⊢ t0: E E ＜: C0 CT(C) = class C extends D {[C f;] K [M]} override(m, D, [C]→C0) ---------------------------------------- C0 m ([C x]) {return t0;} OK in C K = C([D g], [C f]) {super([g]); [this.f = f;]} fields(D) = [D g] [M] OK in C ----------------------------------------------- class C extends D {[C f;] K [M]} OK fields(C) = [D f] Γ ⊢ [t : C] C ＜: D ---------------------- (T-New) Γ ⊢ new C([t]) : C

Note that a well-typed FJ could get stuck in casting, thus the progress property has to be weakened. Also, to make the preservation property holds, we’ve to add a stupid cast rule:

Γ ⊢ t:D C ≮: D D ≮: C stupid warning ------------------------- (T-SCast) Γ ⊢ (C)t : C

## Bounded Quantification

Polymorphism combined with subtyping can greatly increase the expressive power of the typing system. There are two possible ways to define bounded quantification F<: typing rules, depending on how they formulate the rule for comparing bounded quantifiers.

**kernel F＜:**is more tractable but less flexible**full F＜:**is more expressive but somewhat technically problematic

The specification of kernel F＜: is as follows:

SYNTAX t ::= terms x variable λx:T.t abstraction t t application λX＜:T.t type abstraction t [T] type application v ::= values λx:T.t abstraction value λX＜:T.t type abstraction value T ::= types X type variable Top maximum type T→T type of functions ∀X＜:T.T universal type Γ ::= contexts ∅ empty context Γ, x:T term variable binding Γ, X＜:Ttype variable binding EVALUATION t1 ⟶ t1' ------------------ (E-App1) t1 t2 ⟶ t1' t2 t2 ⟶ t2' ------------------ (E-App2) v1 t2 ⟶ v1 t2' t1 ⟶ t1' ------------------- (E-TApp) t1 [T2] ⟶ t1' [T2] (λx:T.t) v ⟶ [x ⟼ v]t (E-AppAbs) (λX＜:T.t) [T] ⟶ [X ⟼ T]t (E-TappTabs) SYPTYPINGΓ ⊢S ＜: S (S-Refl)Γ ⊢S ＜: UΓ ⊢U ＜: T ----------------------------- (S-Trans)Γ ⊢S ＜: TΓ ⊢S ＜: Top (S-Top) X＜:T ∈ Γ -----------(S-TVar)Γ ⊢ X ＜: TΓ ⊢T1 ＜: S1Γ ⊢S2 ＜: T2 ---------------------------------- (S-Arrow)Γ ⊢S1 → S2 ＜: T1 → T2 Γ , X＜:U1 ⊢ S2 ＜: T2 ------------------------------(S-All)Γ ⊢ ∀X＜:U1.S2 ＜: ∀X＜:U1.T2 TYPING Γ, x: T1 ⊢ t: T2 --------------------- (T-Abs) Γ ⊢ λx:T1.t : T1 → T2 x:T ∈ Γ ---------- (T-Var) Γ ⊢ x : T Γ ⊢ t1: T → T', t2: T ----------------------- (T-App) Γ ⊢ t1 t2: T Γ , X＜:T1⊢ t : T -------------------------------- (T-TAbs) Γ ⊢ λX＜:T1.t : ∀X＜:T1.T Γ ⊢ t : ∀X＜:T1.TΓ ⊢ T2 ＜: T1-------------------------------------(T-TApp) Γ ⊢ t [T1] : [X ⟼ T2]T Γ ⊢ t : SΓ ⊢S ＜: T --------------------------- (T-Sub) Γ ⊢ t : T

Note that in the specification above, all quantifications are bounded. Unbounded quantification can be easily restored using following transformation:

```
∀X.T ＝ ∀X＜:Top.T
```

The full F＜: only differs from kernel F＜: in the S-ALL rule. The full F＜: variant of S-ALL is as follows:

Γ ⊢ T1 ＜: S1 Γ, X＜:T1 ⊢ S2 ＜: T2 --------------------------------------- (S-All) Γ ⊢ ∀X＜:S1.S2 ＜: ∀X＜:T1.T2

Subtyping can also be combined with existential type to form bounded existential quantification. Bounded existential quantification allows the modeling of access control.

c = {*{x:Nat, private:Bool}, { state = {x=5, private=false}, methods = { get = λs:{x:Nat}. s.x, inc = λs:{x:Nat,private:Bool}. {x=succ(s.x), private=s.private}}}} as ∃X＜:{x:Nat}.{state:X, methods: {get:X→Nat, inc:X→X}}}

As it’s shown in above, the field *x* is visible outside, but the field *private* is protected.

## Kinding and Higher-Order Polymorphism

In universal type we’ve seen that a type can be a variable, then it’s possible to define functions that accept types as parameter and return types as value. These kind of functions are called *type operators*. Kinding is an approach to formalize type operators.

With the introduction of type operators, the concept of equivalence of types has to be extended, as now it’s possible for two seemingly different types to be equal as shown in the example below:

```
Nat → Bool Nat → Id Bool Id Nat → Id Bool
Id Nat → Bool Id (Nat → Bool) Id (Id (Id Nat → Bool))
```

To formalize this intuition, we have to introduce a definitional equivalence relation on types, written S ≡ T. Following is a rule of the relation:

```
(λX::K11.T12) T2 ≡ [X ⟼ T2]T12
```

As now we can apply one type to another, it’s important to defend against meaningless expressions like “Bool Nat” just like we forbid “true 6”. This can be achieved by introducing the concept **kinds**, which classify type expressions according to their arity, just as arrow types tell us about the arities of terms.

* the kind of proper types (like Bool and Bool→Bool) *⟹* the kind of type operators (i.e., functions from proper types to proper types) *⟹*⟹* the kind of functions from proper types to type operators (i.e., two-argument operators) (*⟹*)⟹* the kind of functions from type operators to proper types

Intuitively, kinds are “the types of types.” Following diagram illustrates the relationship among kinds, types and terms(TAPL, P444).

In theory, the hierarchy can go to infinity. But in practice, three levels is sufficient for programming languages.

The specification of higher-order polymorphic lambda calculus (Fω) is as follows:

SYNTAX t ::= terms x variable λx:T.t abstraction t t application λX::K.t type abstraction t [T] type application v ::= values λx:T.t abstraction value λX.t type abstraction value T ::= types X type variable T→T type of functions ∀X::K.T universal typeλX::K.Toperator abstractionT Toperator application Γ ::= contexts ∅ empty context Γ , x:T term variable binding Γ , X::Ktype variable binding K ::= kinds * kind of proper types K⟹K kind of operators EVALUATION t1 ⟶ t1' ------------------ (E-App1) t1 t2 ⟶ t1' t2 t2 ⟶ t2' ------------------ (E-App2) v1 t2 ⟶ v1 t2' (λx:T.t) v ⟶ [x ⟼ v]t (E-AppAbs) t1 ⟶ t1' ------------------- (E-TApp) t1 [T2] ⟶ t1' [T2] (λX::K.t) [T] ⟶ [X ⟼ T]t (E-TappTabs) KINDING X::K ∈ Γ ---------- (K-TVar) Γ ⊢ X :: K Γ, X::K1 ⊢ T2::K2 ----------------------- (K-Abs) Γ ⊢ λX::K1.T2 :: K1⟹K2 Γ ⊢ T1:: K11⟹K12 Γ ⊢ T2::K11 -------------------------------- (K-App) Γ ⊢ T1 T2 :: K12 Γ ⊢ T1 :: * Γ ⊢ T2 :: * --------------------------- (K-Arrow) Γ ⊢ T1→T2 :: * Γ, X::K1 ⊢ T2 :: * -------------------- (K-All) Γ ⊢ ∀X::K1.T2 :: * TYPE EQUIVALENCE T ≡ T (Q-Refl) T ≡ S ----- (Q-Symm) S ≡ T S ≡ U U ≡ T ------------- (Q-Trans) S ≡ T S1 ≡ T1 S2 ≡ T2 ----------------- (Q-Arrow) S1→S2 ≡ T1→T2 S2 ≡ T2 ----------------------- (Q-All) ∀X::K1 .S2 ≡ ∀X::K1 .T2 S2 ≡ T2 --------------------- (Q-Abs) λX::K1.S2 ≡ λX::K1.T2 S1 ≡ T1 S2 ≡ T2 ------------------ (Q-App) S1 S2 ≡ T1 T2 (λX::K11.T12) T2 ≡ [X ⟼ T2 ]T12 (Q-AppAbs) TYPING x:T ∈ Γ ---------- (T-Var) Γ ⊢ x : T Γ ⊢ T1::* Γ, x:T1 ⊢ t:T2 --------------------------- (T-Abs) Γ ⊢ λx:T1.t : T1 → T2 Γ ⊢ t1: T → T', t2: T ----------------------- (T-App) Γ ⊢ t1 t2: T Γ, X::K⊢ t : T --------------------- (T-TAbs) Γ ⊢ λX::K.t : ∀X::K.T Γ ⊢ t : ∀X::K.TΓ ⊢ T1::K------------------------- (T-TApp) Γ ⊢ t [T1] : [X ⟼ T1]T Γ ⊢ t:S S ≡ T Γ ⊢ T::* -------------------------(T-Eq)Γ ⊢ t : T

## Higher-Order Bounded Polymorphism

We’ve seen that kinding can be combined with System F, it’s also possible to combine kinding with System F with subtyping.

There are two important decisions to be made on how to extend bounded polymorphism with kinding:

- Whether type operators like λX::K1.T2 should be generalized to bounded type operators of the form λX<:T1.T2?
- How to extend subtype relation to include type operators?

For the first question, if *bounded* type operators like λX<:T1.T2 were introduced, then we’ll have to extend kinds with things like ∀X<:T1.K2, which greatly complicates the system, though it’s theoretically possible. For simplicity, here we allow type operators to be unbounded.

For the second question, the simplest solution is to lift the subtype relation on proper types point-wise to type operators. For abstractions, we say that λX.S is a subtype of λX.T whenever applying both to any argument U yields types that are in the subtype relation.

Equivalently, we can say that λX.S is a subtype of λX.T if S is a subtype of T when we hold X abstract, making no assumptions about its subtypes and supertypes.

A useful side effect of this definition is that the subtype relations for higher kinds all have maximal elements. If we let Top[*] = Top and define(maximal elements of higher kinds)

```
Top[K1⇒K2] = λX::K1.Top[K2],
```

then a simple induction shows that Γ, S::K ⊢ S <: Top[K].

The specification of higher-order bounded polymorphism is as follows:

SYNTAX t ::= terms x variable λx:T.t abstraction t t application λX＜:T.t type abstraction t [T] type application v ::= values λx:T.t abstraction value λX＜:T.t type abstraction value T ::= ... types X type variable T→T type of functions ∀X＜:T.T universal type λX::K.T operator abstraction T T operator application Γ ::= contexts ∅ empty context Γ , x:T term variable binding Γ , X＜:Ttype variable binding K ::= kinds * kind of proper types K⟹K kind of operators EVALUATION t1 ⟶ t1' ------------------ (E-App1) t1 t2 ⟶ t1' t2 t2 ⟶ t2' ------------------ (E-App2) v1 t2 ⟶ v1 t2' (λx:T.t) v ⟶ [x ⟼ v]t (E-AppAbs) t1 ⟶ t1' ------------------- (E-TApp) t1 [T2] ⟶ t1' [T2] (λX＜:T.t) [U] ⟶ [X ⟼ U]t (E-TappTabs) KINDING Γ ⊢ Top :: *(K-Top)X＜:T ∈ Γ Γ ⊢ T :: K ------------------------- (K-TVar) Γ ⊢ X :: K Γ, X＜:Top[K1]⊢ T2::K2 ----------------------- (K-Abs) Γ ⊢ λX::K1.T2 :: K1⟹K2 Γ ⊢ T1:: K11⟹K12 Γ ⊢ T2::K11 -------------------------------- (K-App) Γ ⊢ T1 T2 :: K12 Γ ⊢ T1 :: * Γ ⊢ T2 :: * --------------------------- (K-Arrow) Γ ⊢ T1→T2 :: * Γ, X＜:Top[K1]⊢ T2 :: * -------------------- (K-All) Γ ⊢ ∀＜:Top[K1].T2 :: * TYPE EQUIVALENCE T ≡ T (Q-Refl) T ≡ S ----- (Q-Symm) S ≡ T S ≡ U U ≡ T ------------- (Q-Trans) S ≡ T S1 ≡ T1 S2 ≡ T2 ----------------- (Q-Arrow) S1→S2 ≡ T1→T2S1 ≡ T1S2 ≡ T2 ----------------------- (Q-All) ∀X＜:S1.S2 ≡ ∀X＜:T1.T2 S2 ≡ T2 --------------------- (Q-Abs) λX::K1.S2 ≡ λX::K1.T2 S1 ≡ T1 S2 ≡ T2 ------------------ (Q-App) S1 S2 ≡ T1 T2 (λX::K11.T12) T2 ≡ [X ⟼ T2]T12 (Q-AppAbs) SUBTYPING Γ ⊢ S＜:U Γ ⊢ U＜:TΓ ⊢ U::K---------------------------------- (S-Trans) Γ ⊢ S ＜: T Γ ⊢ S :: * ------------- (S-Top) Γ ⊢ S ＜: Top X＜:T ∈ Γ ----------- (S-TVar) Γ ⊢ X ＜: T Γ ⊢ T1 ＜: S1 Γ ⊢ S2 ＜: T2 ---------------------------------- (S-Arrow) Γ ⊢ S1 → S2 ＜: T1 → T2Γ ⊢ U1::K1Γ, X＜:U1 ⊢ S2＜:T2 ------------------------------------ (S-All) Γ ⊢ ∀X＜:U1.S2 ＜: ∀X＜:U1.T2 Γ, X＜:Top[K1] ⊢ S2＜:T2 ---------------------------(S-Abs)Γ ⊢ λX::K1.S2 ＜: λX::K1.T2 Γ ⊢ S1 ＜: T1 ------------------(S-App)Γ ⊢ S1 U ＜: T1 U Γ ⊢ S::K Γ ⊢ T::K S ≡ T -------------------------------(S-Eq)Γ ⊢ S ＜: T TYPING x:T ∈ Γ ---------- (T-Var) Γ ⊢ x : T Γ ⊢ T1::* Γ, x:T1 ⊢ t:T2 --------------------------- (T-Abs) Γ ⊢ λx:T1.t : T1 → T2 Γ ⊢ t1: T → T', t2: T ----------------------- (T-App) Γ ⊢ t1 t2: T Γ, X＜:T⊢ t : T --------------------- (T-TAbs) Γ ⊢ λX＜:T.t : ∀X＜:T.T Γ ⊢ t : ∀X＜:T1.TΓ ⊢ T2 ＜: T1------------------------- (T-TApp) Γ ⊢ t [T2] : [X ⟼ T2]T Γ ⊢ t : S S ＜: TΓ ⊢ T :: *----------------------------------- (T-Sub) Γ ⊢ t : T