CS 4110 Programming Languages & Logics. Lecture 16 Programming in the λ-calculus

Transcription

1 CS 4110 Programming Languages & Logics Lecture 16 Programming in the λ-calculus 30 September 2016

2 Review: Church Booleans 2 We can encode TRUE, FALSE, and IF, as: TRUE λx. λy. x FALSE λx. λy. y IF λb. λt. λf. b t f This way, IF behaves how it ought to: IF TRUE v t v f v t IF FALSE v t v f v f

3 Review: Church Numerals 3 Church numerals encode a number n as a function that takes f and x, and applies f to x n times. 0 λf. λx. x 1 λf. λx. f x 2 λf. λx. f (f x) We can define other functions on integers: SUCC λn. λf. λx. f (n f x)

4 Review: Church Numerals 3 Church numerals encode a number n as a function that takes f and x, and applies f to x n times. 0 λf. λx. x 1 λf. λx. f x 2 λf. λx. f (f x) We can define other functions on integers: SUCC λn. λf. λx. f (n f x) PLUS λn 1. λn 2. n 1 SUCC n 2

5 Review: Church Numerals 3 Church numerals encode a number n as a function that takes f and x, and applies f to x n times. 0 λf. λx. x 1 λf. λx. f x 2 λf. λx. f (f x) We can define other functions on integers: SUCC λn. λf. λx. f (n f x) PLUS λn 1. λn 2. n 1 SUCC n 2 TIMES λn 1. λn 2. n 1 (PLUS n 2 ) 0

6 Review: Church Numerals 3 Church numerals encode a number n as a function that takes f and x, and applies f to x n times. 0 λf. λx. x 1 λf. λx. f x 2 λf. λx. f (f x) We can define other functions on integers: SUCC λn. λf. λx. f (n f x) PLUS λn 1. λn 2. n 1 SUCC n 2 TIMES λn 1. λn 2. n 1 (PLUS n 2 ) 0 ISZERO λn. n (λz. FALSE) TRUE

7 Recursive Functions 4 How would we write recursive functions like factorial?

8 Recursive Functions 4 How would we write recursive functions like factorial? We d like to write it like this... FACT λn. IF (ISZERO n) 1 (TIMES n (FACT (PRED n)))

9 Recursive Functions 4 How would we write recursive functions like factorial? We d like to write it like this... FACT λn. IF (ISZERO n) 1 (TIMES n (FACT (PRED n))) In slightly more readable notation this is... FACT λn. if n = 0 then 1 else n FACT (n 1)...but this is an equation, not a definition!

10 Recursion removal trick 5 We can perform a trick to define a function FACT that satisfies the recursive equation on the previous slide.

11 Recursion removal trick 5 We can perform a trick to define a function FACT that satisfies the recursive equation on the previous slide. Define a new function FACT : FACT λf. λn. if n = 0 then 1 else n (f f (n 1))

12 Recursion removal trick 5 We can perform a trick to define a function FACT that satisfies the recursive equation on the previous slide. Define a new function FACT : FACT λf. λn. if n = 0 then 1 else n (f f (n 1)) Then define FACT as FACT applied to itself: FACT FACT FACT

13 Example 6 Let s try evaluating FACT on 3... FACT 3

14 Example 6 Let s try evaluating FACT on 3... FACT 3 = (FACT FACT ) 3

15 Example 6 Let s try evaluating FACT on 3... FACT 3 = (FACT FACT ) 3 = ((λf. λn. if n = 0 then 1 else n (f f (n 1))) FACT ) 3

16 Example 6 Let s try evaluating FACT on 3... FACT 3 = (FACT FACT ) 3 = ((λf. λn. if n = 0 then 1 else n (f f (n 1))) FACT ) 3 (λn. if n = 0 then 1 else n (FACT FACT (n 1))) 3

17 Example 6 Let s try evaluating FACT on 3... FACT 3 = (FACT FACT ) 3 = ((λf. λn. if n = 0 then 1 else n (f f (n 1))) FACT ) 3 (λn. if n = 0 then 1 else n (FACT FACT (n 1))) 3 if 3 = 0 then 1 else 3 (FACT FACT (3 1))

18 Example 6 Let s try evaluating FACT on 3... FACT 3 = (FACT FACT ) 3 = ((λf. λn. if n = 0 then 1 else n (f f (n 1))) FACT ) 3 (λn. if n = 0 then 1 else n (FACT FACT (n 1))) 3 if 3 = 0 then 1 else 3 (FACT FACT (3 1)) 3 (FACT FACT (3 1))

19 Example 6 Let s try evaluating FACT on 3... FACT 3 = (FACT FACT ) 3 = ((λf. λn. if n = 0 then 1 else n (f f (n 1))) FACT ) 3 (λn. if n = 0 then 1 else n (FACT FACT (n 1))) 3 if 3 = 0 then 1 else 3 (FACT FACT (3 1)) 3 (FACT FACT (3 1)) = 3 (FACT (3 1))

20 Example 6 Let s try evaluating FACT on 3... FACT 3 = (FACT FACT ) 3 = ((λf. λn. if n = 0 then 1 else n (f f (n 1))) FACT ) 3 (λn. if n = 0 then 1 else n (FACT FACT (n 1))) 3 if 3 = 0 then 1 else 3 (FACT FACT (3 1)) 3 (FACT FACT (3 1)) = 3 (FACT (3 1))

21 Example 6 Let s try evaluating FACT on 3... FACT 3 = (FACT FACT ) 3 = ((λf. λn. if n = 0 then 1 else n (f f (n 1))) FACT ) 3 (λn. if n = 0 then 1 else n (FACT FACT (n 1))) 3 if 3 = 0 then 1 else 3 (FACT FACT (3 1)) 3 (FACT FACT (3 1)) = 3 (FACT (3 1))

22 Example 6 Let s try evaluating FACT on 3... FACT 3 = (FACT FACT ) 3 = ((λf. λn. if n = 0 then 1 else n (f f (n 1))) FACT ) 3 (λn. if n = 0 then 1 else n (FACT FACT (n 1))) 3 if 3 = 0 then 1 else 3 (FACT FACT (3 1)) 3 (FACT FACT (3 1)) = 3 (FACT (3 1))

23 Example Let s try evaluating FACT on 3... FACT 3 = (FACT FACT ) 3 = ((λf. λn. if n = 0 then 1 else n (f f (n 1))) FACT ) 3 (λn. if n = 0 then 1 else n (FACT FACT (n 1))) 3 if 3 = 0 then 1 else 3 (FACT FACT (3 1)) 3 (FACT FACT (3 1)) = 3 (FACT (3 1)) So we have a technique for writing recursive functions: write a function f that takes itself as an argument and define f as f f. 6

24 Fixed point combinators 7 There is another way: fixed points!

25 Fixed point combinators 7 There is another way: fixed points! Consider factorial again. It is a fixed point of the following: G λf. λn. if n = 0 then 1 else n (f (n 1))

26 Fixed point combinators 7 There is another way: fixed points! Consider factorial again. It is a fixed point of the following: G λf. λn. if n = 0 then 1 else n (f (n 1)) Recall that if g if a fixed point of G, then G g = g. To see that any fixed point g is a real factorial function, try evaluating it: g 5

27 Fixed point combinators 7 There is another way: fixed points! Consider factorial again. It is a fixed point of the following: G λf. λn. if n = 0 then 1 else n (f (n 1)) Recall that if g if a fixed point of G, then G g = g. To see that any fixed point g is a real factorial function, try evaluating it: g 5 = (G g) 5

28 Fixed point combinators 7 There is another way: fixed points! Consider factorial again. It is a fixed point of the following: G λf. λn. if n = 0 then 1 else n (f (n 1)) Recall that if g if a fixed point of G, then G g = g. To see that any fixed point g is a real factorial function, try evaluating it: g 5 = (G g) 5 5 (g 4)

29 Fixed point combinators 7 There is another way: fixed points! Consider factorial again. It is a fixed point of the following: G λf. λn. if n = 0 then 1 else n (f (n 1)) Recall that if g if a fixed point of G, then G g = g. To see that any fixed point g is a real factorial function, try evaluating it: g 5 = (G g) 5 5 (g 4) = 5 ((G g) 4)

30 Fixed point combinators 8 How can we generate the fixed point of G? In denotational semantics, finding fixed points took a lot of math. In the λ-calculus, we just need a suitable combinator...

31 Y Combinator 9 The (infamous) Y combinator is defined as Y λf. (λx. f (x x)) (λx. f (x x)) We say that Y is a fixed point combinator because Y f is a fixed point of f (for any lambda term f).

32 Y Combinator 9 The (infamous) Y combinator is defined as Y λf. (λx. f (x x)) (λx. f (x x)) We say that Y is a fixed point combinator because Y f is a fixed point of f (for any lambda term f). What happens when we evaluate Y G under CBV?

33 Z Combinator 10 To avoid this issue, we ll use a slight variant of the Y combinator, called Z, which is easier to use under CBV.

34 Z Combinator 10 To avoid this issue, we ll use a slight variant of the Y combinator, called Z, which is easier to use under CBV. Z λf. (λx. f (λy. x x y)) (λx. f (λy. x x y))

35 Example 11 Let s see Z in action, on our function G. FACT

36 Example 11 Let s see Z in action, on our function G. FACT = Z G

37 Example 11 Let s see Z in action, on our function G. FACT = Z G = (λf. (λx. f (λy. x x y)) (λx. f (λy. x x y))) G

38 Example 11 Let s see Z in action, on our function G. FACT = Z G = (λf. (λx. f (λy. x x y)) (λx. f (λy. x x y))) G (λx. G (λy. x x y)) (λx. G (λy. x x y))

39 Example 11 Let s see Z in action, on our function G. FACT = Z G = (λf. (λx. f (λy. x x y)) (λx. f (λy. x x y))) G (λx. G (λy. x x y)) (λx. G (λy. x x y)) G (λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y)

40 Example 11 Let s see Z in action, on our function G. FACT = Z G = (λf. (λx. f (λy. x x y)) (λx. f (λy. x x y))) G (λx. G (λy. x x y)) (λx. G (λy. x x y)) G (λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) = (λf. λn. if n = 0 then 1 else n (f (n 1))) (λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y)

41 Example 11 Let s see Z in action, on our function G. FACT = Z G = (λf. (λx. f (λy. x x y)) (λx. f (λy. x x y))) G (λx. G (λy. x x y)) (λx. G (λy. x x y)) G (λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) = (λf. λn. if n = 0 then 1 else n (f (n 1))) (λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) λn. if n = 0 then 1 else n ((λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) (n 1))

42 Example 11 Let s see Z in action, on our function G. FACT = Z G = (λf. (λx. f (λy. x x y)) (λx. f (λy. x x y))) G (λx. G (λy. x x y)) (λx. G (λy. x x y)) G (λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) = (λf. λn. if n = 0 then 1 else n (f (n 1))) (λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) λn. if n = 0 then 1 else n ((λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) (n 1)) = β λn. if n = 0 then 1 else n (λy. (Z G) y) (n 1)

43 Example 11 Let s see Z in action, on our function G. FACT = Z G = (λf. (λx. f (λy. x x y)) (λx. f (λy. x x y))) G (λx. G (λy. x x y)) (λx. G (λy. x x y)) G (λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) = (λf. λn. if n = 0 then 1 else n (f (n 1))) (λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) λn. if n = 0 then 1 else n ((λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) (n 1)) = β λn. if n = 0 then 1 else n (λy. (Z G) y) (n 1) = β λn. if n = 0 then 1 else n (Z G (n 1))

44 Example 11 Let s see Z in action, on our function G. FACT = Z G = (λf. (λx. f (λy. x x y)) (λx. f (λy. x x y))) G (λx. G (λy. x x y)) (λx. G (λy. x x y)) G (λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) = (λf. λn. if n = 0 then 1 else n (f (n 1))) (λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) λn. if n = 0 then 1 else n ((λy. (λx. G (λy. x x y)) (λx. G (λy. x x y)) y) (n 1)) = β λn. if n = 0 then 1 else n (λy. (Z G) y) (n 1) = β λn. if n = 0 then 1 else n (Z G (n 1)) = λn. if n = 0 then 1 else n (FACT (n 1))

45 Other fixed point combinators 12 There are many (indeed infinitely many) fixed-point combinators. Here s a cute one: where Y k (L L L L L L L L L L L L L L L L L L L L L L L L L L) L λabcdefghijklmnopqstuvwxyzr. (r (t h i s i s a f i x e d p o i n t c o m b i n a t o r))

46 Turing s Fixed Point Combinator 13 To gain some more intuition for fixed point combinators, let s derive a combinator Θ originally discovered by Turing.

47 Turing s Fixed Point Combinator 13 To gain some more intuition for fixed point combinators, let s derive a combinator Θ originally discovered by Turing. We know that Θ f is a fixed point of f, so we have Θ f = f (Θ f).

48 Turing s Fixed Point Combinator 13 To gain some more intuition for fixed point combinators, let s derive a combinator Θ originally discovered by Turing. We know that Θ f is a fixed point of f, so we have Θ f = f (Θ f). We can write the following recursive equation: Θ = λf. f (Θ f).

49 Turing s Fixed Point Combinator 13 To gain some more intuition for fixed point combinators, let s derive a combinator Θ originally discovered by Turing. We know that Θ f is a fixed point of f, so we have Θ f = f (Θ f). We can write the following recursive equation: Θ = λf. f (Θ f). Now use the recursion removal trick: Θ λt. λf. f (t t f) Θ Θ Θ

50 θ Example 14 FACT = Θ G

51 θ Example 14 FACT = Θ G = ((λt. λf. f (t t f)) (λt. λf. f (t t f))) G

52 θ Example 14 FACT = Θ G = ((λt. λf. f (t t f)) (λt. λf. f (t t f))) G (λf. f ((λt. λf. f (t t f)) (λt. λf. f (t t f)) f)) G

53 θ Example 14 FACT = Θ G = ((λt. λf. f (t t f)) (λt. λf. f (t t f))) G (λf. f ((λt. λf. f (t t f)) (λt. λf. f (t t f)) f)) G G ((λt. λf. f (t t f)) (λt. λf. f (t t f)) G)

54 θ Example 14 FACT = Θ G = ((λt. λf. f (t t f)) (λt. λf. f (t t f))) G (λf. f ((λt. λf. f (t t f)) (λt. λf. f (t t f)) f)) G G ((λt. λf. f (t t f)) (λt. λf. f (t t f)) G) = G (Θ G)

55 θ Example 14 FACT = Θ G = ((λt. λf. f (t t f)) (λt. λf. f (t t f))) G (λf. f ((λt. λf. f (t t f)) (λt. λf. f (t t f)) f)) G G ((λt. λf. f (t t f)) (λt. λf. f (t t f)) G) = G (Θ G) = (λf. λn. if n = 0 then 1 else n (f (n 1))) (Θ G) λn. if n = 0 then 1 else n ((Θ G) (n 1)) = λn. if n = 0 then 1 else n (FACT (n 1))

56 Definitional Translation 15 We know how to encode Booleans, conditionals, natural numbers, and recursion in λ-calculus. Can we define a real programming language by translating everything in it into the λ-calculus?

57 Definitional Translation 15 We know how to encode Booleans, conditionals, natural numbers, and recursion in λ-calculus. Can we define a real programming language by translating everything in it into the λ-calculus? In definitional translation, we define a denotational semantics where the target is a simpler programming language instead of mathematical objects.

58 Review: Call-by-Value Here are the syntax and CBV semantics of λ-calculus: e ::= x λx. e e 1 e 2 v ::= λx. e e 1 e 1 e e e 1 e 2 e 1 e 2 v e v e (λx. e) v e{v/x} β There are two kinds of rules: congruence rules that specify evaluation order and computation rules that specify the interesting reductions. 16

59 Evaluation Contexts 17 Evaluation contexts let us separate out these two kinds of rules.

60 Evaluation Contexts 17 Evaluation contexts let us separate out these two kinds of rules. An evaluation context E is an expression with a hole in it: a single occurrence of the special symbol [ ] in place of a subexpression. E ::= [ ] E e v E

61 Evaluation Contexts 17 Evaluation contexts let us separate out these two kinds of rules. An evaluation context E is an expression with a hole in it: a single occurrence of the special symbol [ ] in place of a subexpression. E ::= [ ] E e v E We write E[e] to mean the evaluation context E where the hole has been replaced with the expression e.

62 Examples 18 E 1 = [ ] (λx. x) E 1 [λy. y y] = (λy. y y) λx. x

63 Examples 18 E 1 = [ ] (λx. x) E 1 [λy. y y] = (λy. y y) λx. x E 2 = (λz. z z) [ ] E 2 [λx. λy. x] = (λz. z z) (λx. λy. x)

64 Examples 18 E 1 = [ ] (λx. x) E 1 [λy. y y] = (λy. y y) λx. x E 2 = (λz. z z) [ ] E 2 [λx. λy. x] = (λz. z z) (λx. λy. x) E 3 = ([ ] λx. x x) ((λy. y) (λy. y)) E 3 [λf. λg. f g] = ((λf. λg. f g) λx. x x) ((λy. y) (λy. y))

65 CBV With Evaluation Contexts 19 With evaluation contexts, we can define the evaluation semantics for the CBV λ-calculus with just two rules: one for evaluation contexts, and one for β-reduction.

66 CBV With Evaluation Contexts (λx. e) v e{v/x} β 19 With evaluation contexts, we can define the evaluation semantics for the CBV λ-calculus with just two rules: one for evaluation contexts, and one for β-reduction. With this syntax: The small-step rules are: E ::= [ ] E e v E e e E[e] E[e ]

67 CBN With Evaluation Contexts 20 We can also define the semantics of CBN λ-calculus with evaluation contexts.

68 CBN With Evaluation Contexts 20 We can also define the semantics of CBN λ-calculus with evaluation contexts. For call-by-name, the syntax for evaluation contexts is different: E ::= [ ] E e

69 CBN With Evaluation Contexts We can also define the semantics of CBN λ-calculus with evaluation contexts. For call-by-name, the syntax for evaluation contexts is different: E ::= [ ] E e But the small-step rules are the same: e e E[e] E[e ] (λx. e) e e{e /x} β 20

70 Multi-Argument λ-calculus 21 Let s define a version of the λ-calculus that allows functions to take multiple arguments. e ::= x λx 1,..., x n. e e 0 e 1... e n

71 Multi-Argument λ-calculus 22 We can define a CBV operational semantics: E ::= [ ] v 0... v i 1 E e i+1... e n e e E[e] E[e ] (λx 1,..., x n. e 0 ) v 1... v n e 0 {v 1 /x 1 }{v 2 /x 2 }... {v n /x n } β The evaluation contexts ensure that we evaluate multi-argument applications e 0 e 1... e n from left to right.

72 Definitional Translation 23 The multi-argument λ-calculus isn t any more expressive that the pure λ-calculus.

73 Definitional Translation 23 The multi-argument λ-calculus isn t any more expressive that the pure λ-calculus. We can define a translation T [[ ]] that takes an expression in the multi-argument λ-calculus and returns an equivalent expression in the pure λ-calculus.

74 Definitional Translation 23 The multi-argument λ-calculus isn t any more expressive that the pure λ-calculus. We can define a translation T [[ ]] that takes an expression in the multi-argument λ-calculus and returns an equivalent expression in the pure λ-calculus. T [[x]] = x T [[λx 1,..., x n. e]] = λx λx n. T [[e]] T [[e 0 e 1 e 2... e n ]] = (... ((T [[e 0 ]] T [[e 1 ]]) T [[e 2 ]])... T [[e n ]]) This translation curries the multi-argument λ-calculus.