CS 2104 Type checking / Polymorphism Reading: Chapters 3.1, 4.3, 4.4.1 Dr. Abhik Roychoudhury Adapted from Goh Aik Hui’s lecture notes. 1 Overview Types Motivation for typed languages Issues in Type Checking How to type check? How to cater for Polymorphism Type Equivalence When to type check? Strong and Weak typed languages 2 1. Motivation for typed languages Untyped Languages: perform any operation on any data. Example: Assembly movi 5 r0 addf 3.6 r0 // Move integer 5 (2’s complement) to r0 // Treat bit representation in r0 as a // Floating point representation and add 3.6 // to it. Result? You can be sure that r0 does not contain 8.6! (+) Flexibility : “I can do anything I want to and you can’t stop me” (–) Ease of Error Checking. (programs are prone to errors, especially huge ones). “I am human, my brain is limited, I can’t remember and monitor everything.” 3 1. Motivation for typed languages Typed Languages: A type represents a set of values. Programs / procedures / operators are functions from an input type to an output type. Type Checking is the activity of ensuring that the operands / arguments of an operator / procedure are of compatible type This is done by using a set of rules for associating a type with every expression in the language. (These rules are known as the type system). A type error results when an operator is applied to an operand of inappropriate/incompatible type. Output of a type system: There are type-errors (wrt type system) => Program is NOT type-safe. There are no type-errors (wrt type system) => Program is type-safe. 4 1. Motivation for typed languages TC says that there are type errors TC says that there are no type errors Program really has errors Program really does not have errors Usually true Possible TC errs on the conservative side ????? Program MAY still have errors 1. It may still have type errors due to unsafe features of a language. This is due to bad type system design. 2. It may have logic errors. This serves to show that type errors is but one of the many errors you encounter. 5 1. Motivation for typed languages TC says that there are type errors Program really has errors Program really does not have errors Usually true Possible TC errs on the conservative side Typed Languages (+) Error Detection (+) Program documentation (–) Loss of Flexibility (but it’s ok, I don’t lose much freedom anyway since I don’t usually program in that way in the first place. I gain more than what I lose). 6 2. Issues in Type Checking How to type-check? How to cater for polymorphism? What is your definition of “compatible type”? When to perform type checking? Is your language strongly or weakly typed? 7 2.1 How to type-check? Definition: Type statements are of the form: <expr> : <type> meaning that an expression <expr> ‘is-of-thetype’ (the ‘:’ symbol) <type>. Examples: 3 : int 3+4 : int 3.14 : real “abc” : String while (x < 5) {x++;} : Stmt 8 2.1 How to type-check? Definition: Type rules are of the form: e1 : t1 e2 : t2 … en : tn f e1 e2 …en : t (rule name) where each ei : ti is a type statement, n 0. The rule is interpreted as “IF e1 is of type t1 and … and en is of type tn THEN f e1 e2 …en is of type t.” This is similar to the rules we saw when we studied semantics. 9 2.1 How to type-check? Examples of type rules: Rule for constants: 1 : int 2 : int 3 : int Rule for addition: E1 : int E2 : int (+) E1 + E2 : int Rule for boolean comparison: E1 : int E2 : int (==) E1 == E2 : bool 10 2.1 How to type-check? Examples of type rules: Rule for assignment statement: x:T E:T (:=) x := E; : Stmt Rule for if-statment: E1 : Bool S1 : Stmt S2 : Stmt (if) if (E1) {S1} else {S2} : Stmt 11 2.1 How to type-check? Given the program: …And Given the rules: 1 : int int x; … x := x+1; … 2 : int E1 : int E2 : int E1 + E2 : int E1 : int E2 : int E1 == E2 : bool x:T A program/expression is typesafe if we can construct a derivation tree to give a type for that program/expression. x : int E:T x := E; : Stmt E1 : Bool S1 : Stmt 3 : int (+) (==) (:=) S2 : Stmt (if) if (E1) {S1} else {S2} : Stmt x : int 1 : int (+) x+1 : int x:=x+1; : Stmt (:=) 12 2.1 How to type-check? Given the program: int x; float y; … if (x == 3) { y := x; } else { x := x+1; } … …And Given the rules: 1 : int 2 : int E1 : int E2 : int E1 : int E2 : int E:T (==) ??? y:=x; : Stmt (:=) x := E; : Stmt E1 : Bool S1 : Stmt S2 : Stmt (if) if (E1) {S1} else {S2} : Stmt Follow the rules! Try to build tree. Cannot build tree => Not type safe x==3 : Bool (==) E1 == E2 : bool A program/expression is typesafe if we can construct a derivation tree to give a type for that program/expression. 3 : int (+) E1 + E2 : int x:T x : int 3 : int (:=) if (x==3) {y:=x;} else {x:=x+1;} : Stmt x : int x : int 1 : int (+) x+1 : int x:=x+1; : Stmt (:=) (if) 13 Issues in Type Checking How to type-check? How to cater for polymorphism? What is your definition of “compatible type”? When to perform type checking? Is your language strongly or weakly typed? 14 2.2 How to cater for Polymorphism Polymorphism = poly (many) + morph (form) Polymorphism is the ability of a data object to that can take on or assume many different forms. Polymorphism can be categorised into 2 types Ad-hoc Polymorphism ( discussed now) Universal Polymorphism Parametric (discussed with Functional Programming) Inclusion (discussed later in the lecture) 15 2.2 How to cater for Polymorphism Cardelli and Wegner’s classification (1985) Polymorphism Ad-Hoc Coercion Overloading Ad-Hoc polymorphism is obtained when a function works, or appears to work on several different types (which may not exhibit a common structure) and may behave in unrelated ways for each type. Universal Parametric Inclusion Universal polymorphism is obtained when a function works uniformly on a range of types; these types normally exhibit some common structure. 16 2.2 Polymorphism – Coercion COERCION A coercion is a operation that converts the type of an expression to another type. It is done automatically by the language compiler. (If the programmer manually forces a type conversion, it’s called casting) E : int (Int-Float Coercion) E : float int x; float y; ... y := x; ... 17 2.2 Polymorphism – Coercion Example of the use of COERCION int x; float y; … if (x == 3) { y := x; } else { x := x+1; } … 1 : int 2 : int E1 : int E2 : int E1 : int E2 : int (==) E1 == E2 : bool Add in new rule… E : float (+) E1 + E2 : int x:T E : int 3 : int E:T (:=) x := E; : Stmt (Int-Float Coercion) E1 : Bool S1 : Stmt S2 : Stmt (if) if (E1) {S1} else {S2} : Stmt x : int x : int 3 : int x==3 : Bool (==) y : float x : float y:=x; : Stmt (Coercion) (:=) if (x==3) {y:=x;} else {x:=x+1;} : Stmt x : int x : int 1 : int (+) x+1 : int x:=x+1; : Stmt (:=) (if) 18 2.2 Polymorphism – Coercion Coercion Widening Narrowing Widening coercion converts a value to a type that can include (at least approximations of) all of the values of the original type. Narrowing coercion converts a value to a type that cannot store (even approximations of) all of the values of the original type. Widening is safe most of the time. It can be unsafe in certain cases. Narrowing is unsafe. Information is lost during conversion of type. int <- Widening float Narrowing -> int float 19 2.2 Polymorphism – Coercion Coercions (+) Increase flexibility in programming Example: float x,y,z; int a,b,c; If I have no coercions, and I intend to add y and a and store in x, then writing… x = y + ((float) a); …is too much of a hassle. Therefore coercion can be useful. 20 2.2 Polymorphism – Coercion Coercions (–) Decrease Reliability (error detection) Example: float x,y,z; int a,b,c; If I have coercions and I intend to add x and y and store in z, but I accidentally write… z = x + a; …then my error will go undetected because the compiler will simply coerce the a to a float. Therefore coercion can lead to problems. 21 2.2 Polymorphism – Overloading OVERLOADING An overloaded operation has different meanings, and different types, in different contexts. E1 : int E2 : int (+-int) E1 + E2 : int E1 : float E2 : float (+-float) E1 + E2 : float 22 2.2 Polymorphism – Overloading Example of the use of Overloading int x,y,z; float a,b,c; … if (x == 3) { x := y + z; } else { a := b + c; } … 1 : int 2 : int E1 : int E1 : int x : int x==3 : Bool (==) E2 : int (==) E:T (:=) x := E; : Stmt E2 : float 3 : int (+) E1 == E2 : bool Add in new rule… E1 + E2 : float E2 : int E1 + E2 : int x:T E1 : float 3 : int (+-float) E1 : Bool S1 : Stmt S2 : Stmt (if) if (E1) {S1} else {S2} : Stmt x : int y:int z:int (+) y+z : int x:=y+z; : Stmt (:=) if (x==3) {x:=y+z;} else {a:=b+c;} : Stmt b:float c:float (+ -float) a : float b+c : float a:=b+c; : Stmt (:=) (if) 23 2.2 Polymorphism – Overloading Overloading (+) Increase flexibility in programming Examples are when user wants to use an operator to express similar ideas. Example: int int int a = p = x = a,b,c; p[10], q[10], r[10]; x[10][10], y[10][10], z[10][10]; b * c; // integer multiplication a * q; // Scalar multiplication y * z; // Matrix multiplication Therefore overloading is good. 24 2.2 Polymorphism – Overloading Overloading (–) Decrease Reliability (error detection) Examples are when user intends to use the operator in one context, but accidentally uses it in another. Example In many languages, the minus sign is overloaded to both unary and binary uses. x = z–y and x = -y will both compile. What if I intend to do the first, but accidentally leave out the ‘z’? Similarly, in C, we can have a situation when x = z&y and x = &y will both compile. Is overloading good? 25 2.2 Polymorphism – Overloading Overloading (–) Decrease Reliability (error detection) Even for common operations, overloading may not be good. Example int sum, count; float average; ... average = sum / count; Since sum and count are integers, integer division is performed first before result is coerced to float. That’s why Pascal has div for integer division and / for floating point division. 26 2.2 How to cater for Polymorphism Cardelli and Wegner’s classification (1985) Polymorphism Ad-Hoc Coercion Overloading Ad-Hoc polymorphism is obtained when a function works, or appears to work on several different types (which may not exhibit a common structure) and may behave in unrelated ways for each type. Universal Parametric Inclusion Universal polymorphism is obtained when a function works uniformly on a range of types; these types normally exhibit some common structure. 27 2.2 Inclusion Polymorphism Q: Is the subclass regarded as a subtype of the parent class? Yes – Inclusion Polymorphism (Sub-typing) class A {…} class B extends A {…} Note that B A (Inclusion) A a = new B(); A a = new A(); Polymorphism 28 2.2 Inclusion Polymorphism Q: Is the subclass regarded as a subtype of the parent class? Yes – Inclusion Polymorphism (Sub-typing) Some people call it the IS-A relationship between parent and derived class. “class Table extends Furniture” Table IS-A Furniture. Table Furniture 29 2.2 Inclusion Polymorphism Variables are polymorphic – since they can refer to the declared class and to subclasses too. Requirement (Do you know why?): Subclass must INHERIT EVERYTHING from the base class. Subclass must NOT MODIFY ACCESS CONTROL of the base class methods/data. That’s why C++ Inclusion Polymorphism definition adds a ‘public’ to the derived class since a private derived class modifies access control of base class methods/data. 30 Issues in Type Checking How to type-check? How to cater for polymorphism? What is your definition of “compatible type”? When to perform type checking? Is your language strongly or weakly typed? 31 2.3 Type Equivalence type // type definitions Q = array [1..10] of integer; S = array [1..10] of integer; T = S; type // Queue = Stack = Tree = var var a b c d : : : : // variable declarations Q; S; T; array [1..10] of integer; a b c d : : : : type definitions array [1..10] of integer; array [1..10] of integer; Stack; // variable declarations Queue; Stack; Tree; array [1..10] of integer; begin a := b; // Is this allowed? // Meaning to say “Is a and b // the same type?” begin a := b; // Is this allowed? // Meaning to say “Is a and b // the same type?” a := c; // Is this allowed? a := d; // Is this allowed? b := c; // Is this allowed? end. a := c; // Is this allowed? a := d; // Is this allowed? b := c; // Is this allowed? end. If you had said “yes” to most of it, chances are that you are adopting structural equivalence. If you had said “no” most of the time, then it is likely you are adopting name equivalence. 32 2.3 Type Equivalence Difference between type names and anonymous type names. The type of a variable is either described through: A type name: (1) those names defined using a type definition command. (eg. ‘type’ for Pascal, ‘typedef’ for C.), or… (2) the primitive numeric types (eg. int, float) Or directly through a type constructor (eg. array-of, record-of, pointer-to). In this case, the variable has an anonymous type name. 33 2.3 Type Equivalence Example type // type definitions Q = array [1..10] of integer; S = array [1..10] of integer; T = S; var a b c d : : : : // variable declarations Q; S; T; array [1..10] of integer; Q,S,T are type names d has a type, but d does not have a type name. begin a := b; // Is this allowed? // Meaning to say “Is a and b // the same type?” a := c; // Is this allowed? a := d; // Is this allowed? b := c; // Is this allowed? end. 34 2.3 Type Equivalence When are two types equivalent ()? Rule 1: For any type name T, T T. Rule 2: If C is a type constructor and T1 T2, then CT1 CT2 . Rule 3: If it is declared that type name = T, then name T. Rule 4 (Symmetry): If T1 T2,then T2 T1. Rule 5 (Transitivity): If T1 T2 and T2 T3, then T1 T3. What rules do you want to use? 35 2.3 Type Equivalence When are two types equivalent ()? Rule 1: For any type name T, T T. Rule 2: If C is a type constructor and T1 T2, then CT1 CT2 . Rule 3: If it is declared that type name = T, then name T. Rule 4 (Symmetry): If T1 T2,then T2 T1. Rule 5 (Transitivity): If T1 T2 and T2 T3, then T1 T3. Structural Equivalence will use all the rules to check for type equivalence. 36 2.3 Type Equivalence When are two types equivalent ()? Rule 1: For any type name T, T T. Rule 2: If C is a type constructor and T1 T2, then CT1 CT2 . Rule 3: If it is declared that type name = T, then name T. Rule 4 (Symmetry): If T1 T2,then T2 T1. Rule 5 (Transitivity): If T1 T2 and T2 T3, then T1 T3. (Pure) Name Equivalence will use only the first rule. Unless the two variables have the same type name, they will be treated as different type 37 2.3 Type Equivalence When are two types equivalent ()? Rule 1: For any type name T, T T. Rule 2: If C is a type constructor and T1 T2, then CT1 CT2 . Rule 3: If it is declared that type name = T, then name T. Rule 4 (Symmetry): If T1 T2,then T2 T1. Rule 5 (Transitivity): If T1 T2 and T2 T3, then T1 T3. Declarative Equivalence will leave out the second rule. 38 2.3 Type Equivalence Example type // type definitions Q = array [1..10] of integer; S = array [1..10] of integer; T = S; var a,x b : c : d : e : // variable declarations : Q; S; T; array [1..10] of integer; array [1..10] of integer; begin a := x; // Is this allowed? // Meaning to say “Is a and b // the same type?” a := b; // Is this allowed? a := c; // Is this allowed? a := d; // Is this allowed? b := c; // Is this allowed? d := e; // Is this allowed? end. R1: For any type name T, T T. R2: If C is a type constructor and T1 T2, then CT1 CT2 . R3: If it is declared that type name = T, then name T. R4 (Symmetry): If T1 T2,then T2 T1. R5 (Transitivity): If T1 T2 and T2 T3, then T1 T3. SE NE DE yes yes yes yes yes yes yes yes no no no no no no no no yes no 39 2.3 Type Equivalence Name Equivalence Easy to implement checking, since we need only compare the name. Very restrictive, inflexible. type idxtype = 1..100; var count : integer; index : idxtype; Structure Equivalence Harder to implement since entire structures must be compared. Other issues to consider: eg. arrays with same sizes but different subscripts – are they the same type? (similar for records and enumerations) More flexible, yet the flexibility can be bad too. type celsius = real; fahrenheit = real; var x : celsius; y : fahrenheit; ...x := y; // Allowed? 40 2.3 Type Equivalence Different Languages adopt different rules. And the rules may change for one language (people can change their minds too!) Pascal Before 1982 – unknown. ISO1982 – Declarative Equivalence. ISO1990 – Structural Eqivalence. C : Structural Equivalence, except for structs and unions, for which C uses declarative equivalence. If the two structs are in different files, then C goes back to structural equivalence. C++ : Name Equivalence Haskell/SML : Structural Equivalence. 41 Issues in Type Checking How to type-check? How to cater for polymorphism? What is your definition of “compatible type”? When to perform type checking? Is your language strongly or weakly typed? 42 2.4 When to perform Type Checking? When is the variable bound to the type? When can I type check? Compile-Time Run-Time (Static Type Binding) (Dynamic Type Binding) In theory, you can choose to type check at compile time or run-time. No choice but to do dynamic type checking. In practice, languages try to do it as much statically as possible. Eg. SML, Pascal Eg. JavaScript, APL 43 2.4 When to perform Type Checking? Static Type Checking – done at compile time. (+) Done only once (+) Earlier detection of errors (–) Less Program Flexibility (Fewer shortcuts and tricks) 44 2.4 When to perform Type Checking? Dynamic Type Checking – done at run time. (–) Done many times (–) Late detection of errors (–) More memory needed, since we need to maintain type information of all the current values in their respective memory cells. (–) Slows down overall execution time, since extra code is inserted into the program to detect type error. (+) Program Flexibility (Allows you to ‘hack’ dirty code.) 45 Issues in Type Checking How to type-check? How to cater for polymorphism? What is your definition of “compatible type”? When to perform type checking? Is your language strongly or weakly typed? 46 2.5 Strong Type Systems A programming language is defined to be strongly typed if type errors are always detected STATICALLY. A language with a strong-type system only allows typesafe programs to be successfully compiled into executables. (Otherwise, language is said to have a weak type system). Programs of strong-type systems are guaranteed to be executed without type-error. (The only error left to contend with is logic error). 47 2.5 Strong Type Systems Language Strongly Typed? Fortran No Ada No Modula-3 No Allows variable of one type to refer to value of another type through EQUIVALENCE keyword. Library function UNCHECKED_CONVERSION suspends type checking. Same as Ada through use of keyword LOOPHOLE C, C++ No 1. Forced conversion of type through type casting Why? 2. Union Types can compromise type safety Java No Type Casting Pascal Almost Variant Records can compromise type safety SML Yes Haskell Yes All variables have STATIC TYPE BINDING. 48 2.5 Weak-Type Systems: Variant Recs Variant Records in C (via union keyword) compromises Type Safety ... typedef union { int X; float Y; char Z[4];} B; ... B P; Variant part all have overlapping (same) L-value!!! Problems can occur. What happens to the code below? P.X = 142; printf(“%O\n”, P.Z[3]) All 3 data objects have same L-value and occupy same storage. No enforcement of type checking. Poor language and type system design 49 2.5 Weak-Type Systems: Variant Recs Variant Records in Pascal tries to overcome C’s deficiency. They have a tagged union type. type whichtype = (inttype, realtype); type uniontype = record case V : whichtype of inttype : (X: integer); realtype: (Y: real); end But the compiler usually doesn’t check the consistency between the variant and the tag. So we can ‘subvert’ the tagged field: var P: uniontype P.V = inttype; P.X = 142; P.V = realtype; // type safety compromised 50