CS4212: Compiler Design

Week 10: Type Safety

Typing for Advanced Features

Ilya Sergey
[email protected]

ilyasergey.net/CS4212/
Simply-typed Lambda Calculus with Integers
• For the language in “stlc.ml” we have five inference rules:

INT VAR ADD

x :T ∈ G G ⊢ e1 : int G ⊢ e2 : int

G ⊢ i : int G ⊢ x :T G ⊢ e1 + e2 : int

FUN APP
G, x : T ⊢ e : S G ⊢ e1 : T -> S G ⊢ e2 : T

G ⊢ fun (x:T) -> e : T -> S G ⊢ e1 e2 : S

• Note how these rules correspond to the OCaml code.

 Implementing a Type Checker
for Lambda Calculus

See stlc.ml
 Exercise

• Implement the missing parts of the type-checker

 Notes about this Type Checker

• In the interpreter, we only evaluate the body of a function when it's applied.

• In the type checker, we always check the body of the function (even if it's never applied.)
– We assume the input has some type (say t1) and reflect this in the type of the function (t1 -> t2).

• Dually, at a call site (e1 e2), we don't know what closure we're going to get as e1.
– But we can calculate e1's type, check that e2 is an argument of the right type, and also determine
what type will (e1 e2) have.

• Question: Why is this a valid approximation of the dynamic program behaviour?

 Contexts and Inference Rules

• Need to keep track of contextual information.

– What variables are in scope?
– What are their types?
– What information doe we have about each syntactic construct?

• What relationships are there among the syntactic objects?

– e.g. is one type a subtype of another?

• How do we describe this information?

– In the compiler there's a mapping from variables to information we know about them – the "context".
– The compiler has a collection of (mutually recursive) functions that follow the structure of the syntax.
 Why Inference Rules?
• They are a compact, precise way of specifying language properties.
– E.g. ~20 pages for full Java vs. 100’s of pages of prose Java Language Spec.
– Check out oat-v1-typing.pdf

• Inference rules correspond closely to the recursive AST traversal that implements them

• Type checking (and type inference) is nothing more than attempting

to prove a judgment (G ⊢ e : t) by searching backwards through the rules.

• Strong mathematical foundations

– The “Curry-Howard correspondence”:
– Programming Language ~ Logic,
– Program ~ Proof,
– Type ~ Proposition

• Talk to me you’re interested in type systems!

Types and Type Safety
 Type Safety

Theorem: (type soundness of simply typed lambda calculus with integers)

If ⊢ e : t then there exists a value v such that e ⇓ v .

"Well typed programs do not go wrong."

– Robin Milner, 1978

• Note: this is a very strong property.

– Well-typed programs cannot "go wrong" by trying to execute undefined code
(such as 3 + (fun x -> 2))
– Simply-typed lambda calculus is guaranteed to terminate!
(i.e. it isn't Turing complete)
 Tuples
• ML-style tuples with statically known number of products:
• First: add a new type constructor: T1 * … * Tn

TUPLE
G ⊢ e1 : T1 … G ⊢ en : Tn

G ⊢ (e1, …, en) : T1 * … * Tn

PROJ
G ⊢ e : T1 * … * Tn 1≤i≤n

G ⊢ #i e : Ti
A note on Curry-Howard Correspondence

11
 Arrays
• Array constructs are not hard
• First: add a new type constructor: T[]

G ⊢ e1 : int G ⊢ e2 : T e1 is the size of the newly

NEW
allocated array. e2
initialises the elements of
G ⊢ new T[e1](e2) : T[] the array.

INDEX
G ⊢ e1 : T[] G ⊢ e2 : int

G ⊢ e1[e2] : T Note: These rules don’t

ensure that the array index
UPDATE
is in bounds – that should
G ⊢ e1 : T[] G ⊢ e2 : int G ⊢ e3 : T be checked dynamically.

G ⊢ e1[e2] = e3 ok
 References
• OCaml-style references (note that in OCaml references are expressions)
• First, add a new type constructor: T ref

REF G ⊢ e :T

G ⊢ ref e : T ref

DEREF
G ⊢ e : T ref

G ⊢ !e : T
ASSIGN Note the similarity with the
rules for arrays…
G ⊢ e1 : T ref E ⊢ e2 : T

G ⊢ e1 := e2 : unit
 Well-Formed Types
• In languages with type definitions, need additional rules to define well-formed types
• Judgements take the form H ⊢ t
– H is set of type names
⊢
– t is atype ⊢
– H⊢t
means
⊢
“Assuming
⊢ H lists well-formed types, t is a well-formed type”

⊢ 1 ⊢ 2 ∈
⊢ 1 ⊢ 2 ⊢ ∈
⊢ ⊢ ⊢ ⊢ ⊢⊢ 1 → ⊢
→2 1 2

• Note: also need to modify the typing rules and judgements. E.g.,

⊢ ⊢1 1 , Γ{ $→$→ 1 }
, Γ{ 1 }⊢⊢ : : 22
⊢⊢
, Γ, Γ ( ( : :1 )1 ) : : 11→→ 22
 Type-Checking Statements
• In languages with statements, need additional rules to define well-formed statements

• E.g.,judgements may take the form H;G;rt ⊢ s

– H maps type names to their definitions
– G is a type environment (variables -> types)
– rt is a type
– H;G;rt ⊢ s ; Γ; ⊢
means
Γ →
“with type definitions H, assuming type environment G, s is avalid statement within
the context
; Γ; ⊢ of a function that returns a value of type rt” Γ

Γ⊢ : Γ( ) Γ⊢ : Γ⊢ : ; Γ{ #→ }; ⊢ 2
; Γ; ⊢ := ; Γ; ⊢ ; Γ; ⊢ = ; 2
 Type Safety For General Languages
Theorem: (Type Safety)

If ⊢ P : t is a well-typed program, then either:

(a) the program terminates in a well-defined way, or
(b) the program continues computing forever

• Well-defined termination could include:

– halting with a return value
– raising an exception
• Type safety rules out undefined behaviours:
– abusing "unsafe" casts: converting pointers to integers, etc.
– treating non-code values as code (and vice-versa)
– breaking the type abstractions of the language (e.g., via Java/Ruby reflection)
• What is "defined" depends on the language semantics…
A good place for a break

17
Types as Sets
 What are types, anyway?
• A type is just a predicate on the set of values in a system.
– For example, the type “int” can be thought of as a boolean function that returns “true”
on integers and “false” otherwise.
– Equivalently, we can think of a type as just a subset of all values.

• For efficiency and tractability, the predicates are usually taken to be very simple.
– Types are an abstraction mechanism

• We can easily add new types that distinguish different subsets of values:

type tp =
| IntT (* type of integers *)
| PosT | NegT | ZeroT (* refinements of ints *)
| BoolT (* type of booleans *)
| TrueT | FalseT (* subsets of booleans *)
| AnyT (* any value *)
 Modifying the typing rules
• We need to refine the typing rules too…
• Some easy cases:
– Just split up the integers into their more refined cases:

P-INT N-INT ZERO

i>0 i<0

G ⊢ i : Pos G ⊢ i : Neg G ⊢ 0 : Zero

• Same for booleans:

TRUE FALSE

G ⊢ true : True G ⊢ false : False

 What about “if”?
• Two cases are easy:

IF-T IF-F G ⊢ e1 : False E ⊢ e3 : T

G ⊢ e1 : True G ⊢ e2 : T

G ⊢ if (e1) e2 else e3 : T G ⊢ if (e1) e2 else e3 : T

• What happens when we don’t know statically which branch will be taken?
• Consider the type checking problem:

x:bool ⊢ if (x) 3 else -1 : ?

• The true branch has type Pos and the false branch has type Neg.
– What should be the result type of the whole if?
 Downcasting
• What happens if we have an Int but need something of type Pos?
– At compile time, we don’t know whether the Int is greater than zero.
– At run time, we do.

• Add a “checked downcast”

G ⊢ e1 : Int G, x : Pos ⊢ e2 : T2 G ⊢ e3 : T3

G ⊢ ifPos (x = e1) e2 else e3 : LUB(T2, T3)

• At runtime, ifPos checks whether e1 is > 0.

If so, branches to e2 and otherwise branches to e3.
• Inside the expression e2, x is the name for e1’s value, which is known to be
strictly positive because of the dynamic check.
• Note that such rules force the programmer to add the appropriate checks
– We could give integer division the type: Int → NonZero → Int
 Subtyping and Upper Bounds
• If we think of types as sets of values, we have a natural inclusion relation: Pos ⊆ Int
• This subset relation gives rise to a subtype relation: Pos <: Int (sometimes also typeset as ≤)
• Such inclusions give rise to a subtyping hierarchy:

Any :>
<:

Int Bool
<: :> : :>
<

:>
Neg Zero Pos True False

• Given any two types T1 and T2, we can calculate their least upper bound (LUB)
according to the hierarchy.
– Example: LUB(True, False) = Bool, LUB(Int, Bool) = Any
– Note: might want to add types for “NonZero”, “NonNegative”, and “NonPositive” so that set
union on values corresponds to taking LUBs on types.
 “If” Typing Rule Revisited
• For statically unknown conditionals, we want the return value to be the LUB of
the types of the branches:

IF-BOOL
G ⊢ e1 : bool E ⊢ e2 : T1 G ⊢ e3 : T2

G ⊢ if (e1) e2 else e3 : LUB(T1,T2)

• Note that LUB(T1, T2) is the most precise type (according to the hierarchy)
that is able to describe any value that has either type T1 or type T2.

• In math notation, LUB(T1, T2) is sometimes written T1 ⋁ T2

• LUB is also called the join operation.

 Subtyping Hierarchy

• A subtyping hierarchy:

Any :>
<:

Int Bool
<: :> : :>
<

:>
Neg Zero Pos True False

• The subtyping relation is a partial order:

– Reflexive: T <: T for any type T
– Transitive: T1 <: T2 and T2 <: T3 then T1 <: T3
– Antisymmetric: It T1 <: T2 and T2 <: T1 then T1 = T2
 Soundness of Subtyping Relations
• We don’t have to treat every subset of the integers as a type.
– e.g., we left out the type NonNeg

• A subtyping relation T1 <: T2 is sound if it approximates the underlying semantic subset relation.

• Formally: write ⟦T⟧ for the subset of (closed) values of type T

– i.e. ⟦T⟧ = {v | ⊢ v : T}
– e.g. ⟦Zero⟧ = {0}, ⟦Pos⟧ = {1, 2, 3, …}

• If T1 <: T2 implies ⟦T1⟧ ⊆ ⟦T2⟧, then T1 <: T2 is sound.

– e.g. Pos <: Int is sound, since {1,2,3,…} ⊆ {…,-3,-2,-1,0,1,2,3,...}
– e.g. Int <: Pos is not sound, since it is not the case that
{…,-3,-2,-1,0,1,2,3,...}⊆ {1,2,3,…}
 Subsumption Rule
• When we add subtyping judgments of the form T <: S we can uniformly
integrate it into the type system generically:

SUBSUMPTION
G ⊢ e : T T <: S

G⊢e:S

• Subsumption allows any value of type T to be treated as an S whenever T <: S.

• Adding this rule makes the search for typing derivations more difficult – this rule
can be applied anywhere, since T <: T.
– But careful engineering of the typing system can incorporate the subsumption rule into
a deterministic algorithm. (See, e.g., the Oat type system)
Subtyping in the Wild
 Extending Subtyping to Other Types

• What about subtyping for tuples?

– Intuition: whenever a program expects T1 <: S1 T2 <: S2
something of type S1 * S2, it is sound
to give it a T1 * T2. (T1 * T2) <: (S1 * S2)
– Example: (Pos * Neg) <: (Int * Int)

• What about functions?

• When is T1 → T2 <: S1 → S2 ?
 Subtyping for Function Types
• One way to see it:

Expected function

S1 T1 Actual function T2 S2

• Need to convert an S1 to a T1 and T2 to S2, so the argument type is

contravariant and the output type is covariant.

S1 <: T1 T2 <: S2

(T1 → T2) <: (S1 → S2)

 Immutable Records
• Record type: {lab1:T1; lab2:T2; … ; labn:Tn}
– Each labi is a label drawn from a set of identifiers.

RECORD
G ⊢ e1 : T1 G ⊢ e2 : T2 … G ⊢ en : Tn

G ⊢ {lab1 = e1; lab2 = e2; … ; labn = en} : {lab1:T1; lab2:T2; … ; labn:Tn}

PROJECTION
G ⊢ e : {lab1:T1; lab2:T2; … ; labn:Tn}

G ⊢ e.labi : Ti
 Immutable Record Subtyping
• Depth subtyping:
– Corresponding fields may be subtypes

DEPTH
T1 <: U1 T2 <: U2 … Tn <: Un

{lab1:T1; lab2:T2; … ; labn:Tn} <: {lab1:U1; lab2:U2; … ; labn:Un}

• Width subtyping:
– Subtype record may have more fields:

WIDTH
m≤n

{lab1:T1; lab2:T2; … ; labn:Tn} <: {lab1:T1; lab2:T2; … ; labm:Tm}

Mutability and Subtyping
 NULL
• What is the type of null?

• Consider:
int[] a = null; // OK?
int x = null; // OK? (nope)
string s = null; // OK?
NULL
G ⊢ null : r
• Null has any reference type
– Null is generic

• What about type safety?

– Requires defined behavior when dereferencing null
e.g. Java's NullPointerException
– Requires a safety check for every dereference operation
 Subtyping and References
• What is the proper subtyping relationship for references and arrays?

• Suppose we have NonZero as a type and the division operation has type:
Int → NonZero → Int
– Recall that NonZero <: Int

• Should (NonZero ref) <: (Int ref) ?

• Consider this program:

Int bad(NonZero ref r) {

Int ref a = r; ( * OK because (NonZero ref <: Int ref*)
a := 0; (* OK because 0 : Zero <: Int *)
return (42 / !r) (* OK because !r has type NonZero *)
}
 Mutable Structures are Invariant
• Covariant reference types are unsound (well-typed programs do go wrong)
– As demonstrated in the previous example
• Contravariant reference types are also unsound
– i.e. If T1 <: T2 then ref T2 <: ref T1 is also unsound
– Exercise: construct a program that breaks contravariant references.
• Moral: Mutable structures are invariant:
T1 ref <: T2 ref implies T1 = T2
⊢ 1 <: 1 ... ⊢ <: ⊢ <: ⊢ <:
• Same holds for arrays,
⊢ 1 ∗ ... ∗ <: 1 ∗ ... ∗ OCaml-style
⊢ mutable
<: records, object fields,
⊢ etc.
<:
– Note: Java and C# get this wrong. They allows covariant array subtyping, but then
compensate by adding a dynamic check on every array update!
Γ = [ #→ ]

<:
Γ⊢ : <:
Γ⊢ : Γ⊢0: Γ ⊢ −1 :
Γ ⊢ [0] := −1
Reminder: Subtyping for Function Types
• One way to see it:

Expected function

S1 T1 Actual function T2 S2

• Need to convert an S1 to a T1 and T2 to S2, so the argument type is

contravariant and the output type is covariant.

S1 <: T1 T2 <: S2

(T1 → T2) <: (S1 → S2)

 Another Way to See It
• We can think of a reference cell as an immutable record (object) with two
functions (methods) and some hidden state:
T ref ≃ {get: unit → T; set: T → unit}
– get returns the value hidden in the state.
– set updates the value hidden in the state.

• When is T ref <: S ref?

• Records are like tuples: subtyping extends pointwise over each component.
• {get: unit → T; set: T → unit} <: {get: unit → S; set: S → unit}
– get components are subtypes: unit → T <: unit → S
set components are subtypes: T → unit <: S → unit
• From get, we must have T <: S (covariant return)
• From set, we must have S <: T (contravariant arg.)
• From T <: S and S <: T we conclude T = S.
 Demo: Arrays in Java

• Check out https://github.com/cs4212/week-10-java-arrays

• The code shows the run-time issue with covariant array subtyping

39
Structural vs. Nominal Subtyping
 Structural vs. Nominal Typing
• Is type equality / subsumption defined by the structure of the data or the name of the data?
• Example: type abbreviations (OCaml) vs. “newtypes” (a la Haskell)

(* OCaml: *)
type cents = int (* cents = int in this scope *)
type age = int

let foo (x:cents) (y:age) = x + y

(* Haskell: *)
newtype Cents = Cents Integer (* Integer and Cents are isomorphic, not identical. *)
newtype Age = Age Integer

foo :: Cents -> Age -> Int

foo x y = x + y (* Ill typed! *)

• Type abbreviations are treated “structurally” Newtypes are treated “by name”.
 Nominal Subtyping in Java
• In Java, Classes and Interfaces must be named and their relationships explicitly declared:

(* Java: *)
interface Foo {
int foo();
}

class C { /* Does not implement the Foo interface */

int foo() {return 2;}
}

class D implements Foo {

int foo() {return 4230;}
}

• Similarly for inheritance: the subclass relation must be declared via the “extends” keyword.
– Typechecker still checks that the classes are structurally compatible
Oat’s Type System
 Oat’s Treatment of Types
• Primitive (non-reference) types:
– int, bool
• Definitely non-null reference types: R
– (named) mutable structs with (right-oriented) width subtyping
– string
– arrays (including length information, per HW4)
• Possibly-null reference types: R?
– Subtyping: R <: R?
int sum(int[]? arr) {
– Checked downcast syntax if?:
var z = 0;
if? (int[] a = arr) {
for(var i = 0; i<length(a); i = i + 1;) {
z = z + a[i];
}
}
return z;
}
 Full Oat Features
• Named structure types with mutable fields
– but using structural, width subtyping

• Typed function pointers

• Polymorphic operations: length and == / !=

– need special case handling in the typechecker

• Type-annotated null values: t null always has type t?

• Definitely-not-null values means we need an "atomic" array initialization syntax

– for example, null is not allowed as a value of type int[], so to construct a record
containing a field of type int[], we need to initialize it
– subtlety: int[][] cannot be initialized by default, but int[] can be
 H;G;L;rt ` block1 ;r1
H;G;L;rt ` block2 ;r2 H;G;L0 ` vdecl1 ) L1 . . . H;G;Ln 1 ` vdecli ) Ln
Oat "Returns" Analysis
lhs not a global function id typ_if typ_vdecls
H; G; L; rt ` if(exp)
H; G; L `block lhs 1: elset block2 ) L; r1 ^ r2 H;G;L0 ` vdecl1 , .. ,vdecln ) Ln
0 H;G;L H;G;L 1 ` vdecl
1 ` vdecl
) L2 ) L2
H; G; L 0 ` exp : t H; G; L1 ; rt ` stmt ) L2 ; returns
H; G; L ` exp : ref 0 ?
0 • HType-safe,
` t  t statement-oriented imperative languages H;likeG; L Oat ` exp (or: Java)t x 62 L
H ` ref  ref typ_assn H;function
G; L ` : t x 62 L
H; G; L; rt ` lhs = exp; ) L; ? lhs not a global id exptyp_decl
H;G;L, x : ref ;rt must
` block ensure
1 ;r1 H;G;L;rtthat a function ` block2 ;r (always)
2 returns a value H;G;L H; of
G; ` Lthe
var
` appropriate
x
lhs = : exp
t ) L, x type.
: t ty
H;G;L ` vdecl )1 Lelse typ_ifq H;G;L 0 ` var x = exp ) L, x : t
L; rt ` if?(ref x = 1exp) block
– Does the returned expression's 2 block 2 )
typ_stmtdecl
L; r ^
type
1 r 2 match the one declaredH; G; L
by ` exp
the : t
function?
H; G; L1 ; rt ` vdecl; ) L2 ; ? H;G;L 0 ` vdecls ) L i H ` t0  t
H;
– G; DoL all ` exppaths : bool
through the H;G;L code return`appropriately? typ_assn
exp : (t1 , ..`, tblock;r
H; G; L ` H;G;L;rt n ) -> void 0 vdecls0 )
H;G;L Li H;1 )
` vdecl G; L;L1rt `. .lhs . =H;G;L exp; ) L; ?
n 1 ` vdecli ) Ln
H; G; L ` • Oat’s
exp 1 : t 0
statement
.. H; G; L checking
` exp n : t judgment
0 typ_while typ_vdecls
H; G;0 L; rt ` while(exp) 1
0 block ) L; ? n H;G;L H;G;L
0 ` vdecl1 ` 1vdecl ) L2 n ) Ln
, .. ,vdecl
H;G;L vdecl L . . typ_stmtdecl
. void H;G;L n 1 ` vdecl
H ` t1  t1 .. H ` tn  tn ` )
– takes the expected return type typ_scall as input: what type should H; G; 0 L1the
; rt `statement 1
vdecl; ) return
1L2 ; ? (or if none)
H; G; L; H;G;L 1 ` vdecls
rt ` exp(exp ) Ln2); ) L;H;
1 , .. ,exp ? G; L1 ; rt ` stmt ) L2 ; returns
H;–G; produces
L2 ` exp :a boolean bool flag as output: does the statement exp :H;G;L
H; G; L ` definitely (t1 , return?
.. , tn )0 -> ` vdecl1 , .. ,vdecln ) Ln
void
H; G; L ` exp : bool H; G; L ` exp : t 0 .. H; G; L ` exp : t 0
H; G; L 2 ; rt ` stmt ) L 3 ; ? lhs not
1 a global 1 function id n n
H;G;L;rt ` block1 ;r1 H t 0  t .. H t 0  t
H;G;L 2 ;rt
H;G;L;rt ` block;r
` block2 ;r2 H; G; L ; rt `
1 typ_for stmt `
) 1 LH;2 ; returns
1G; L ` `
lhs n: t n
typ_scall
typ_if H; G; L; rt
H; `G; exp(exp
L ` exp1 , ..
: ,exp
t 0 n ); ) L; ?
L1 ;G;
G; H; rt L;
` rtfor(vdecls;
` if(exp) block expopt 1
; stmt
else block
opt ) 2
block
) L;) r 1 ^L r
12
; ?
H ` H; t0 G;  Lt` exp : bool
H; G; L H; ` G;expL :`refexp 0 0
? : t H`t  t
0 lhs not a global function
typ_assn id
typ_retT H; G; L; H;G;L;rt
rt ` lhs = ` block
exp; )
1 ;r 1L; ?
H;` G; blockL2 ;r`2 lhs : t
0
H ` refH;  reft ` return exp; ) L; >
G; L; H;G;L;rt
H;G;L ` vdecl ) L 0 typ_if
H;G;L, x : ref ;rt ` block1 ;r1 H;G;L;rt ` block2 ;r2
typ_ifq
1
H; G; L; rt ` if(exp) block H; G;
1 else L
2 `
block 2 exp
) L;:r1 t^ r2
typ_stmtdecl
G; L; rtH; typ_retVoid H; G; L1 ; rt ` 0 vdecl; ) 0L2 ; ?
` G; L; void ` return;1 ) L; > 2 ) L; r1 ^ r2
if?(ref x = exp) block else block
H; G; L ` exp : ref ? H`t  t
H; G; L ` exp : bool H;
H G; L 0` exp
` ref ref : (t1 , .. , tn ) -> void t
H;G;L;rt ` block;r H; G; L exp : t 0H;..G;H; L; G;rt L ` lhs
exp = : exp;
t 0 ) L; ?
H;G;L, x : ref ;rt
` 1 ` block1 1 ;r1 H;G;L;rt `n blockn2 ;r2
`
typ_while 0 0 typ_ifq
H; G; L; rt ` while(exp) block ) L; ? H; G;HL;`rtt1`  t1 ..x = H
if?(ref exp) ` tblock
n 1 else tn block2 ) L; r1 ^ r2
H;G;L1 ` vdecl ) Ltyp_scall 2
H;G;L1 ` vdecls ) L2 H; G; L; rt `H;exp(exp G; L ` 1exp , .. ,exp
: bool n ); ) L; ? typ_
Example: Typing in Oat
 Checking Derivations
• A derivation or proof tree has (instances of) judgments as its nodes and edges that connect
premises to a conclusion according to an inference rule.
• Leaves of the tree are axioms (i.e. rules with no premises)
– Example: the INT rule is an axiom
• Goal of the type checker: verify that such a tree exists.
• Example 1: Find a tree for the following program using the inference rules in Oat specification
var x1 = 0;
var x2 = x1 + x1;
x1 = x1 – x2;
return(x1);

Example 2: There is no tree for this ill-typed program:

int f() {
var x = int[] null;
x = new int[] {3,4};
return x[0];
}
 CIS341 – Steve Zdancewic
Example Derivation
October 7, 2020
H;G;L;rt ` block;returns

H;G;L0 ;rt `ss stmt1 .. stmtn ) Ln ;r

This is an example derivation for the following OAT statements. Note that it is not a complete OAT
typ_block
H;G;L0 ;rt ` {stmt1 .. stmtn } ;r
program (it’s technically not even a block because it isn’t inside of braces).
Let S be the following OAT program: H;G;L0 ;rt `ss stmt1 .. stmtn ) Ln ;returns
H; G; L0 ; rt ` stmt1 ) L1 ; ?
var x1 = 0; D1 ...
var x1 = 0;
var x2 = x1 + x1; D2 H; G; Ln 2 ; rt ` stmtn 1 ) Ln 1 ; ?
var x2 = x1 + x1; H; G; Ln 1 ; rt ` stmtn ) Ln ; r
x1 = x1 - x2; D2 typ_stmts
x1 = x1 – x2; H;G;L0 ;rt `ss stmt1 .. stmtn 1 stmtn ) Ln ;r
return x1;
return(x1); D4
H;G `s tdecl
The following
H ` t1 .. H ` ti x1 .. xi distinct
typ_tdeclok
H;G `s struct S{ t1 x1 ; .. ;ti xi }
D1 D2 H;GD`3 f fdeclD4
typ_stmts
H;G; · ;int `ss var x1 = 0; var x2 = x1 + x2 ; x1 = x1 - x2 ; return x 1 ; ) x 1 : int, x 2 : int, · ; >
H;G;x : t , .. , x : t ;rt ` block;> x .. x distinct
1 1 i i 1 i
typ_fdeclok
H;G ` f rt f (t1 x1 , .. , ti xi ) block

D1 = H;G ` prog
typ_int x1 6 2 ·
H; G; · ` 0 : int typ_dempty
typ_decl H;G ` e
H;G;· ` var x1 = 0 ) ·, x1 : int
typ_stmtdecl
H;G ` prog
H; G; ·; int ` var x1 = 0; ) ·, x1 : int; ? typ_dgdecl
 lhs not a global function id
program
H; G; L (it’s technically
` lhs : t not even a block because it isn’t inside of braces).
; G; L ` expLet
: tS be the following
H; G; L ` exp : t0 Example Derivation
OAT program:
H`t  t 0
var x1 = 0; H ` ref
typ_assn
H; G; L; rt ` lhs = exp; ) L; ?H;G;L1 ` vdecl ) L2 typ_null
var x2 = x1 + x1; H; G; L ` ref null : ref ?
x1H;G;L
= x1 1-`x2; vdecl ) L2 H; G; L ` exp : t x 62 L
typ_stmtdecl typ_bool_true typ_decl
H;return ` vdecl; ) H;
G; L1 ; rt x1; ; ?L ` true : bool
L2G; H;G;L ` var x = exp ) L, x : t

H; G; L ` exp : (t ,
The following
1 .. , t n ) -> void H;G;L 0 ` vdecls ) L typ_bool_false
i
var x1 = 0; 0 H; G; L ` false 0 : bool
H; G; L ` exp1 : t1 .. H; G; L ` expn : tn
var x2 = x1 + x1; H;G;L0 ` vdecl1 ) L1 . . . H;G;Ln 1 ` vdecli ) Ln
0
H ` t1  t1 .. H ` tn  tn 0 typ_int typ_vdec
x1 = x1 – x2; H; G; L D` integer D : int
typ_scall D H;G;L
D 0 ` vdecl1 , .. ,vdecl n ) Ln
G; L; rt ` exp(exp1 , .. ,expn ); ) L; ?
H; return(x1); 1 2 3 4
typ_stmt
H;G; · ;int `ss var x1 = 0; G; Lx
H;var 1 ;2rt=`x1stmt
+ x2 ;) 1 - x2 ; return x1 ; ) x1 : int, x2 : int, ·;>
x1 L=2 ;typ_string
xreturns
H; G; L ` exp : H; boolG; L ` string : string
H;G;L;rt ` block1 ;r1 lhs not a global function id
id : t 2 L
D1H;G;L;rt
= ` block2 ;r2 typ_local H; G; L ` lhs : t
H; G; L ` id : t typ_if H; G; L ` exp : t 0
; G; L; rt ` if(exp) block1 else block2 ) L; r1 ^ typ_int r2 x1 6 2 ·
H; G; · ` 0 : int 0  t
H ` ttyp_decl
0 id 62 L id : t 2 G typ_assn
; G; L ` exp : ref ? H;G;· ` var x1 = 0 ) ·typ_global , x1 : intH; G; L; rt ` lhs = exp; ) L; ?
0 H; G; L ` id : t typ_stmtdecl
` ref  ref H; G; ·; int ` var x = 0; ) ·, x : int; ?
1 1
H;G;L 1 ` vdecl ) L2
;G;L, x : ref ;rt ` blockH 1 t H;G;L;rt ` block2 ;r2
1 ;r` typ_stmtdecl
H; G; L1 ; rt ` vdecl; ) L2 ; ?
typ_ifq
L; rt ` if?(refD2 x== exp)H;block
G; L1 `else 1 : t21 ).. L;H;
expblock ^ rL2 ` expn : tn
r1 G;
H ` t1  t .. H ` tn  t H; G; L ` exp : (t1 , .. , tn ) -> void
` exp : bool
H; G; Ltyp_intOps H; G; L ` exp1 typ_carr
0
: t1 .. H; G; L ` expn : tn 0
H; G; L ` new t[]{exp , .. , exp } : t[] 0 0
 lhs not a global function id typ_null
var x1 = 0;
H; G;varL `x2lhs : t H; G; L ` ref null : ref ?
; G; L ` exp : t
H; G;x1 L`
= x1 + x1;
: t0
exp- x2;
= x1 Example Derivation typ_bool_true
0
H ` t  tx1;
return H; G; L ` true : bool
Htyp_assn
` ref
H; G; L; rt `Thelhs = exp; )
following L; ?H;G;L1 ` vdecl ) L2 typ_null typ_bool_false
H; G; L ` ref null : ref ? H; G; L ` false : bool
H;G;L1 ` vdecl ) L2 H; G; L ` exp : t x 62 L
typ_stmtdecl
D1 D2 D3 typ_bool_true
D4 typ_int
typ_decl
H; G; L1 ; rt H;G;
` vdecl; ) H;
L G;
; ? L ` true : bool H;G;L ` H;
var G;
x L= `expinteger
) L, x:: tint
typ_stmts
· ;int ` ss var2 x 1 = 0; var x2 = x1 + x2 ; x1 = x1 - x2 ; return x1 ; ) x1 : int, x2 : int, ·;>

H; G; L ` exp : (t1 , .. , tn ) -> void H;G;L ` vdecls ) L typ_bool_false typ_string

H; G; L ` false 0
: bool i H; G; L ` string : string
H; G;varL ` x1 exp=1 :0;tD10 1 =.. H; G; L ` expn : tn0
var
0 x2 = x1 + x1;0 typ_int H;G;L · vdecl1 ) L1 . . . id
x1062`typ_int H;G;L
: t 2 Ln 1 ` vdecli ) Ln
H ` t1  t1 .. H ` tn  tnH; G; · ` 0 : int typ_local typ_vdec
x1 = x1 – x2; H; G; L ` integer
H;G;· ` var x1 = 0 ) ·, x1 : int
:
typ_scallint typ_decl
H;G;L 0 ` vdecl H;1 ,G; .. ,vdecl
L ` idn ) : tLn
G; L; rt ` exp(exp1 , .. ,expn ); ) L; ?
H; return(x1); typ_stmtdecl
H; G; ·; int ` var x1 = 0; ) ·, x1 : int; ? id 62 L id : t 2 G
H; G; L1 ; rt ` stmt ) L2 ;typ_string returns
H; G; L ` exp : H; bool G; L ` string : string typ_global
H; G; L ` id : t
H;G;L;rtD2 = ` block ;r
1 1 lhs not a global function id
id : t 2 L H ` t
H;G;L;rttyp_intOps ` block2 ;r2 typ_local H; G; L ` lhs : t
` +:(int, int) -> int
H; G; L ` id : t typ_if H; G;H; L G;
` Lexp` 1 exp: t1: t0.. H; G; L ` expn : tn
; G; L; rt `x1 : int
if(exp) block1 else
2 ·, x1 : int block
typ_local 2 ) x1 :L;
intr 2 ^
1
·, xr1 : int
2 H `H
typ_local t1 `t0 t ..t H ` tn  t
H; G; ·, x1 : int ` x1 : int H; G; ·, x1 : int ` x1 : int typ_car
id 6 2 L id : t 2 G typ_bop x 6 2 · , x : int
, .. ,L; typ_assn
:
H;H;
G;G; L; L lhs =t[]exp; exp t[]
0 2 1
; G; L ` exp : ref ? H; G; ·, x1 : int ` x1 + x1 : int typ_global rt `` new {exp1) ?n }
typ_decl
0 H;G;· H; ` varG; x2L= ` x1 ): ·,tx1 : int, x2 : int
x1 +id
` ref  ref H; G; ·, x1 : int; int ` var x2 = x1 + x1 ; ) ·, x1 : int,H;G;L H ` t typ_stmtdecl
x2 : int; ?1 ` vdecl ) L2
;G;L, x : ref ;rt ` blockH 1 ;r` 1 t H;G;L;rt ` block2 ;r2 H; G; L ` exp1 : int typ_stmtdecl
typ_ifqH; G; L1 ; rt ` vdecl; ) L2 ; ?
L; rt ` if?(ref x = exp)H;block G; L1 `else 1 : t21 ).. L;H;
expblock r1 G; ^ rL2 ` expn : tn t 2 {int, bool, r?} typ_newarray
H ` t1  t .. H ` tn  t H; G; L ` exp : (t ,
H; G; L01` new.. , t ) -> void
n t[exp1 ] : t[]
H; G; L ` exp : bool H; G; L ` exp1 typ_carr: t1 .. H; G; L ` expn : tn 0
H; G; L ` new t[]{exp , .. , exp } : t[]
0 0
 lhs not a global function id typ_null
H; G; L ` lhs : t H; G; L ` ref null : ref ?
; G; L ` exp : t
H; G; L ` exp : t0 Example Derivation typ_bool_true
H`t  t 0 H; G; L ` true : bool
Htyp_assn
` ref
H; G; L; rt ` lhs = exp; ) L; ?H;G;L1 ` vdecl ) L2 typ_null typ_bool_false
H; G; L ` ref null : ref ? H; G; L ` false : bool
H;G;L1 ` vdecl ) L2 H; G; L ` exp : t x 62 L
typ_stmtdecl typ_bool_true typ_int
typ_decl
H; G;
H; G; L1 ; rt ` vdecl; ) L2 ; ? L ` true : bool H;G;L ` H;
varG;
x L= ` integer
exp ) L, x:: tint

H; G; L ` exp : (t1 , .. , tn ) -> void H;G;L ` vdecls ) L typ_bool_false typ_string

H; G; L ` false 0
: bool i H; G; L ` string : string
H; G;varL ` x1 exp=1 :0;t10 .. H; G; L ` expn : tn0
var x2 = x1 + x1; H;G;L0 ` vdecl1 ) L1 . . . idH;G;L : t 2 L n 1 ` vdecli ) Ln
0
H ` t1  t1 .. H ` tn  tn 0 typ_int typ_local typ_vdec
x1 = x1 – x2; H; G; L ` integer : int
typ_scall H;G;L0 ` vdeclH; 1 ,G;.. ,vdecl
L ` idn ) : tLn
G; L; rt ` exp(exp1 , .. ,expn ); ) L; ?
H; return(x1);
H; G; L1 ; rt ` stmt ) L2 ;typ_string returns id 62 L id : t 2 G
H; G; L ` exp : H; bool G; L ` string : string typ_global
D5 = H; G; L ` id : t
H;G;L;rt ` block1 ;r1 lhs not a global function id
id : t 2 L
typ_intOps H ` t
H;G;L;rt ` -:(int, block
`int) ->2int;r2 typ_local H; G; L ` lhs : t
x1 : int 2 ·, x1 : int, xH; : G;
int L ` id : t typ_if x : H;
int G;
2 H;·L, x`
G;: L exp
int, ` x1 :
exp
: int t 1: t 0.. H; G; L ` expn : tn
; G; L; rt ` if(exp) block1 else block22 ) L; r1typ_local ^ r2 2
H ` t
1
 0 t
2
.. H ` t typ_local
 t
H; G; ·, x1 : int, x2 : int ` x1 : int H; G; ·, x1 : int, H `2 t  t2
1 x : int ` x : int n
typ_car
id 6 2 L id : t 2 G , .. ,L;
typ_bop
typ_assn
?n } : t[]
H; G; ·, x1 : int, x2 : int ` x1 - x1 :H;
H; G;G; L; L lhs =t[]exp; exp
0 int
; G; L ` exp : ref ? typ_global rt `` new {exp1)
0
` ref  ref H; G; L ` id : t
H;G;L H `` tvdecl ) L
D3 = 1 2
;G;L, x : ref ;rt ` blockH 1 ;r` 1 t H;G;L;rt ` block2 ;r2 H; G; L ` exp1 : int typ_stmtdecl
x1 : int 2 ·, x1 : int, x2 : int typ_ifq H; G; L1 ; rt ` vdecl; ) L2 ; ?
L; rt ` if?(ref
G ` x1 not axglobal
= exp) H;block
function G;
id L1 ` else expblock
1 : t21 ).. L;H; r 1 G;
^ rL `
typ_local
2 exp n : t n t 2 { int,
sub_sub_int bool, r?
D }
H; G; ·, x1 : int, x2 : int ` x1 : int
H; G; L H ` int  int
` exp : (t , .. , t ) -> void
5
typ_newarray
HH;`G;t1·, x : intt, x ..
: int ; H
int `
` t
x n = 
x - t
x ; ) · , x : int , x : int H;
; ? G; L 1 ` new n t[exp1 ] :typ_assn t[]
H; G; L ` exp : bool 1 2 1 1 2
H; G; L ` exp1 typ_carr
1 2 0
: t1 .. H; G; L ` expn : tn 0
H; G; L ` new t[]{exp , .. , exp } : t[]
0 0
 H; G; L2 ` exp : bool typ_null
H; G; L ` ref null : ref ?
H; G; L2 ; rt ` stmt ) L3 ; ?
H;G;L
D5 =2 ;rt ` block;r
2 Subtyping Example Derivation
Rules typ_bool_true
typ_for H; G; L ` true : bool
1 ; rt ` for(vdecls; expopt ; stmt opt )
H`t  t
block ) L1 ; ?
typ_intOps
1 2 typ_bool_false
` -:(int, int) 0-> int 0 H; G; L ` false : bool
H; G; L `x1 :exp int 2 : ·t, x1 : H `xt2 : int
int,  t x2 : int 2 ·, x1 : int, x2 : int
typ_retT
typ_local typ_local
sub_sub_int typ_int
H; G; · , x
H; G; L; t ` return1 : int, x 2 : int ` x
exp; )1 L; > : int H; G; · , x1 : H
int,
` xint
2 : int ` x
int2 : int
H; G; L ` integer : int typ_bop
H; G; ·, x1 : int, x2 : int ` x1 - x1 : int
typ_retVoid sub_sub_bool
typ_string
H; G; L; void ` return; ) L; > H ` bool  bool
H; G; L ` string : string
var x1 = 0;
D3 =
var x2 = x1 + x1; H ` r ref 1  ref
id : t2 2 L sub_sub_nref
H ` ref 1 ? H;  G;refL2`? id : t typ_local
x1 = x1 – x2; x1 : int 2 ·, x1 : int, x2 : int
1 not areturn(x1);
global function id typ_local sub_sub_int D5
H; G; ·, x1 : int, x2 : int ` x1 : int H ` int  int
H `r refid 1 L refid2 : t 2 G
62 typ_assn
H; G; ·, x1 : int, x2 : int; int ` x1 = x1 - x2 ; ) ·, x1 : int, x2 : int; ? typ_global
sub_sub_ref
H ` ref 1H;G; ref L `2 id : t
H `Ht ` ref  ref
D4 = r 1 2
9 H; G; L ` exp1 : t1 .. H; sub_sub_nrref
G; L ` expn : tn
H ` ref  ref
H ` t1  t .. H ` tn  t
1 2 ?
x1 : int 2 ·, x1 : int, x2 : int typ_car
typ_local H; G;L `int sub_sub_int
new t[]{exp1 , .. , expn } : t[]
H; G; ·, x1 : int, x2 : intH``rx1ref: 1 int
 ref 2 H ` int
typ_retT
H; G; ·, x1 : int, x2 : int; int ` return x1 ; ) ·, x1 : int,Hx2`: int;
t >
H; G; L ` exp1 : int sub_subrstring
H `r string  string
t 2 {int, bool, r?}
typ_newarray
H; G; L ` new t[exp sub_subrarray
1 ] : t[]
H `r t[]  t[]
 Example: Ill-Typed Oat Program

int f() {
var x = int[] null;
x = new int[] {3,4};
return x[0];
}
 Next in this Module

• Making our programs faster