cse130

Roadmap

Past three weeks:

How do we use a functional language?

Rest of the class:

How do we implement a functional language?
… in a functional language (of course)

This week: Interpreter

How do we evaluate a program given its abstract syntax tree (AST)?

The Nano Language

Features of Nano:

Arithmetic
Variables
Let-bindings
Functions
Recursion

1. Nano: Arithmetic

A grammar of arithmetic expressions:

e ::= n
    | e1 + e2
    | e1 - e2
    | e1 * e2

Examples:

Expression		Value
`4`	`==>`	4
`4 + 12`	`==>`	16
`(4 + 12) - 5`	`==>`	11

Representing Expressions and Values

Let’s represent arithmetic expressions as a type:

data Expr = Num Int
          | Add Expr Expr
          | Sub Expr Expr
          | Mul Expr Expr

Let’s represent arithmetic values as a type:

type Value = Int

Evaluating Arithmetic Expressions

We can now write a Haskell function to evaluate an expression:

eval :: Expr -> Value
eval (Num n)     = n
eval (Add e1 e2) = eval e1 + eval e2
eval (Sub e1 e2) = eval e1 - eval e2
eval (Mul e1 e2) = eval e1 * eval e2

Alternative representation

Let’s factor out the operators into a separate type

data Binop = Add | Sub | Mul

data Expr = Num Int              -- number
          | Bin Binop Expr Expr  -- binary expression

QUIZ

Evaluator for alternative representation:

eval :: Expr -> Value
eval (Num n)        = n
eval (Bin op e1 e2) = evalOp op (eval e1) (eval e2)

What’s a suitable type for evalOp?

(A) Binop -> Value

(B) Binop -> Value -> Value -> Value

(C) Binop -> Expr -> Expr -> Value

(D) Binop -> Expr -> Expr -> Expr

(E) Binop -> Expr -> Value

The Nano Language

Features of Nano:

Arithmetic [done]
Variables
Let-bindings
Functions
Recursion

2. Nano: Variables

Let’s add variables!

e ::= n        -- OLD
    | e1 + e2
    | e1 - e2
    | e1 * e2
    | x        -- NEW

Let’s extend the representation of expressions:

data Expr = Num Int              -- number
          | Bin Binop Expr Expr  -- binary expression
          | ???                  -- variable

type Id = String

data Expr = Num Int              -- number
          | Bin Binop Expr Expr  -- binary expression
          | Var Id               -- variable

Now let’s extend the evaluation function!

QUIZ

What should the following expression evaluate to?

x + 1

(A) 0

(B) 1

(C) Runtime error

Environment

An expression is evaluated in an environment

It’s a dictionary that maps variables to values

["x" := 0, "y" := 12, ...]

We can represent an environment using the following type:

type Env = [(Id, Value)]

Evaluation in an Environment

We write

eval env expr  ==> value

To mean that evaluating expr in the environment env returns value

QUIZ

What should the result of?

eval ["x" := 0, "y" := 12, ...] (x + 1)

(A) 0

(B) 1

(C) Runtime error

To evaluate a variable, look up its value in the environment!

Environment	Expression		Value
`["x" := 5]`	`x`	`==>`	5
`["x" := 5]`	`x + 12`	`==>`	17
`["x" := 5]`	`y - 5`	`==>`	error

Evaluating Variables

We need to update our evaluation function to take the environment as an argument:

eval :: Env -> Expr -> Value
eval env (Num n)        = ???
eval env (Bin op e1 e2) = ???
eval env (Var x)        = ???

eval :: Env -> Expr -> Value
eval env (Num n)        = n
eval env (Bin op e1 e2) = evalOp op (eval env e1) (eval env e2)
eval env (Var x)        = lookup x env

But how do variables get into the environment?

The Nano Language

Features of Nano:

Arithmetic [done]
Variables [done]
Let-bindings
Functions
Recursion

3. Nano: Let Bindings

Let’s add let bindings!

e ::= n                -- OLD
    | e1 + e2
    | e1 - e2
    | e1 * e2
    | x
    | let x = e1 in e2 -- NEW

Example:

Environment	Expression		Value
`[]`	`let x = 2 + 3 in x * 2`	`==>`	10

Let’s extend the representation of expressions:

data Expr = Num Int              -- number
          | Bin Binop Expr Expr  -- binary expression
          | Var x                -- variable
          | ???                  -- let binding

data Expr = Num Int              -- number
          | Bin Binop Expr Expr  -- binary expression
          | Var Id               -- variable
          | Let Id Expr Expr     -- let binding

Now let’s extend the evaluation function!

eval :: Env -> Expr -> Value
eval env (Num n)          = n
eval env (Bin op e1 e2)   = evalOp op (eval env e1) (eval env e2)
eval env (Var x)          = lookup x env
eval env (Let x def body) = ???

Let’s develop intuition with examples!

QUIZ

What should this evaluate to?

let x = 2 + 3
in
  x + 1

(A) 1

(B) 5

(C) 6

(D) Error: unbound variable x

(E) Error: unbound variable y

QUIZ

What should this evaluate to?

let x = 5
in
  let y = x + 1
  in
    x * y

(A) 5

(B) 6

(C) 30

(D) Error: unbound variable x

(E) Error: unbound variable y

QUIZ

What should this evaluate to?

let x = 0
in
  (let x = 100
   in
     x + 1
  ) + x

(A) 1

(B) 101

(C) 201

(D) 2

(E) Error: multiple definitions of x

Principle: Static (Lexical) Scoping

Every variable use (occurrence) gets its value from the most local definition (binding)

in a pure language, the value never changes once defined
easy to tell by looking at the program, where a variable’s value came from!

Implementing Lexical Scoping

Example 1:

               -- environment:
let x = 2 + 3  -- []
in             --   [x := 5]
  x + 1        --   |

Note: first evaluate 2 + 3 and bind x to the resulting value in the environment

Example 2:

                 -- environment:
let x = 5        -- []
in               --   [x := 5]
  let y = x + 1  --   |
  in             --   | [y := 6, x := 5]
    x * y        --   | |

Note: [y := 6] got added to the environment

Example 3:

                 -- environment:
let x = 0        -- []
in               --   [x := 0]
  (let x = 100   --   |
   in            --   | [x := 100, x := 0]
     x + 1       --   | |
  )              --   | |
  + x            --   |

Note: [x := 100] was only added for the inner scope

Evaluating let Expressions

To evaluate let x = e1 in e2 in env:

Evaluate e1 in env to val
Extend env with a mapping ["x" := val]
Evaluate e2 in this extended environment

eval :: Env -> Expr -> Value
eval env (Num n)          = n
eval env (Bin op e1 e2)   = evalOp op (eval e1) (eval e2)
eval env (Var x)          = lookup x env
eval env (Let x e1 e2)    = eval env' e2
  where
    v    = eval env e1
    env' = (x, v) : env

The Nano Language

Features of Nano:

Arithmetic [done]
Variables [done]
Let binding [done]
Functions
Recursion

4. Nano: Functions

Let’s add:

lambda abstraction (aka function definitions)
applications (aka function calls)

e ::= n                -- OLD
    | e1 + e2
    | e1 - e2
    | e1 * e2
    | x
    | let x = e1 in e2
                       -- NEW
    | \x -> e  -- abstraction
    | e1 e2    -- application

Example:

let inc = \x -> x + 1 in
inc 10

QUIZ

What should this evaluate to?

let inc = \x -> x + 1 in
inc 10

(A) Undefined variable x

(B) Undefined variable inc

(C) 1

(D) 10

(E) 11

Representing functions

Let’s extend the representation of expressions:

data Expr = Num Int              -- number
          | Bin Binop Expr Expr  -- binary expression
          | Var Id               -- variable
          | Let Id Expr Expr     -- let expression
          | ???                  -- abstraction
          | ???                  -- application

data Expr = Num Int              -- number
          | Bin Binop Expr Expr  -- binary expression
          | Var Id               -- variable
          | Let Id Expr Expr     -- let expression
          | Lam Id Expr          -- abstraction: formal + body
          | App Expr Expr        -- application: function + actual

Example:

let inc = \x -> x + 1 in
inc 10

represented as:

Let "inc"
  (Lam "x" (Bin Add (Var "x") (Num 1)))
  (App (Var "inc") (Num 10))

Evaluating Functions

                      -- environment
let inc = \x -> x + 1
in                    -- [inc := ???]
  inc 10              -- use the value of inc to evaluate this

What is the value of inc???

Rethinking our values

Until now: a program evaluates to an integer (or fails)

type Value = Int

type Env = [(Id, Value)]

eval :: Env -> Expr -> Value

What do these programs evaluate to?

(1)
\x -> x + 1
==> ???

(2)
let f = \x y -> x + y in
f 1
==> ???

Now: a program evaluates to an integer or a function (or fails)

Remember: functions are first-class values

Let’s change our definition of values!

data Value = VNum Int
           | VFun ??? -- What info do we need to store?

-- Other types stay the same
type Env = [(Id, Value)]

eval :: Env -> Expr -> Value

Function values

How should we represent a function value?

let inc = \x -> x + 1 in
inc 10

We need to store enough information about inc so that we can later evaluate any application of inc (like inc 0, inc 5, inc 10, inc (factorial 100))

The value of a function is its code!

Representing Function Values (First Attempt)

Grammar for values:

v ::= n       -- OLD: number
    | <x, e>  -- NEW: formal + body

Haskell representation:

data Value = VNum Int
           | VFun Id Expr -- formal + body

Let’s try this!

                      -- environment
let inc = \x -> x + 1
in                    -- [inc := <x, x + 1>]
  inc 10              -- how do we evaluate this?

Evaluating applications

                      -- environment
let inc = \x -> x + 1
in                    -- [inc := <x, x + 1>]
  inc 10              -- how do we evaluate this?

To evaluate inc 10:

Evaluate inc, get <x, x + 1>
Evaluate 10, get 10
Evaluate x + 1 in an environment extended with [x := 10]

Let’s extend our eval function!

eval :: Env -> Expr -> Value
eval env (Num n)          = ???
eval env (Bin op e1 e2)   = ???
eval env (Var x)          = ???
eval env (Let x e1 e2)    = ???
eval env (Lam x e)        = ???
eval env (App e1 e2)      = ???

eval :: Env -> Expr -> Value
eval env (Num n)        = VNum n
eval env (Var x)        = lookup x env
eval env (Bin op e1 e2) = evalOp op (eval env e1) (eval env e1)
eval env (Let x e1 e2)  = eval env' e2
  where
    v = eval env e1
    env' = (x, v) : env
eval env (Lam x body)   = VFun x body
eval env (App fun arg)  = evalApp (eval env e1) (eval env e1)
  where
    evalApp :: Value -> Value -> Value
    evalApp (VFun x body) vArg = eval env' body
      where
        env' = (x, vArg) : env
    evalApp _ _                = error "applying a non-function"

QUIZ

What should this evaluate to?

let c = 1
in
  let inc = \x -> x + c
  in
    inc 10

(A) Undefined variable x

(B) Undefined variable c

(C) 1

(D) 10

(E) 11

QUIZ

And what should this evaluate to?

let c = 1
in
  let inc = \x -> x + c
  in
    let c = 100
    in
      inc 10

(A) Error: multiple definitions of c

(B) 11

(C) 110

Reminder: Referential Transparency

The same expression must always evaluate to the same value

In particular: a function must always return the same output for a given input

Why?

> myFunc 10
11

> myFunc 10
110

Oh no! How do I find the bug???

Is it in myFunc?
Is it in a global variable?
Is it in a library somewhere else?

My worst debugging nightmare!

Static vs Dynamic Scoping

What we want:

let c = 1               -- <-------------------
in                      --                    \
  let inc = \x -> x + c -- refers to this def \
  in
    let c = 100
    in
      inc 10

==> 11

Lexical (or static) scoping:

each occurrence of a variable refers to the most recent binding in the program text
definition of each variable is unique and known statically
guarantees referential transparency:

let c = 1               -- <-------------------
in                      --                    \
  let inc = \x -> x + c -- refers to this def \
  in
    let c = 100
    in
      let res1 = inc 10      -- ==> 11
      in
        let c = 200
        in
           let res2 = inc 10 -- ==> 11
           in res1 == res2   -- ==> True

What we don’t want:

let c = 1
in
  let inc = \x -> x + c -- refers to this def \
  in                    --                    \
    let c = 100         -- <-------------------
    in
      inc 10

==> 110

Dynamic scoping:

each occurrence of a variable refers to the most recent binding during program execution
can’t tell where a variable is defined just by looking at the function body
violates referential transparency:

let c = 1
in
  let inc = \x -> x + c     -- refers to this def \  \
  in                        --                    \  \
    let c = 100             -- <-------------------  \
    in                      --                       \
      let res1 = inc 10     -- ==> 110               \
      in                    --                       \
        let c = 200         -- <----------------------
        in
          let res2 = inc 10 -- ==> 210!!!
          in res1 == res2   -- ==> False

QUIZ

Which scoping does our eval function implement?

...
eval env (Lam x body)   = VFun x body
eval env (App fun arg)  = evalApp (eval env e1) (eval env e1)
  where
    evalApp (VFun x body) vArg = eval ((x,vArg):env) body

(A) Static

(B) Dynamic

(C) Neither

Let’s find out!

                         -- env:
let c = 1                --                                          []
in                       --                                  ["c" := 1]
  let inc = \x -> x + c
  in                     --             ["inc" := <x, x + c>, "c" := 1]
    let c = 100
    in                   -- ["c" := 100, "inc" := <x, x + c>, "c" := 1]
      inc             10

-- 1. ==> <x, x + c>

-- 2.                 ==> 10

-- 3. x + c      ["x" := 10, "c" := 100, "inc" := <x, x + c>, "c" := 1]

--    ==> 110

Ouch.

What went wrong?

let c = 1
in                       --                                  ["c" := 1]
  let inc = \x -> x + c  -- we want this "c" to ALWAYS mean 1!
  in                     --             ["inc" := <x, x + c>, "c" := 1]
    let c = 100
    in                   -- ["c" := 100, "inc" := <x, x + c>, "c" := 1]
      inc 10       -- but now it means 100 because we are in a new env!

Lesson learned: need to remember what c was bound to when inc was defined!

i.e. “freeze” the environment at the point of function definition

Implementing Static Scoping

Key ideas:

At definition: Freeze the environment in the function’s value
At call: Use the frozen environment to evaluate the body
- instead of the current environment

                         -- env:
let c = 1                --                                          []
in                       --                                  ["c" := 1]
  let inc = \x -> x + c
  in                     --        ["inc" := <fro, x, x + c>, "c" := 1]
                         --             where fro = ["c" := 1]
    let c = 100
    in                   -- ["c" := 100, "inc" := <fro, x, x + c>, ...]
      inc             10

-- 1. ==> <fro, x, x + c>

-- 2.                 ==> 10
--               add "x" to fro instead of env:
-- 3. x + c      ["x" := 10, "c" := 1]

--    ==> 11

Tada!

Function Values as Closures

To implement lexical scoping, we will represent function values as closures

Closure = lambda abstraction (formal + body) + environment at function definition

Updated grammar for values:

v ::= n
    | <env, x, e>  -- NEW: frozen env + formal + body

env ::= []
      | (x := v) : env

Updated Haskell representation:

data Value = VNum Int
           | VClos Env Id Expr -- frozen env + formal + body

Evaluating function definitions

How should we modify our eval for Lam?

data Value = VNum Int
           | VClos Env Id Expr -- env + formal + body

eval :: Env -> Expr -> Value
eval env (Lam x body) = ??? -- construct a closure

Recall: At definition: Freeze the environment in the function’s value

Exact code for you to figure out in HW4

Evaluating function calls

How should we modify our eval for App?

data Value = VNum Int
           | VClos Env Id Expr -- env + formal + body

eval :: Env -> Expr -> Value
eval env (App e1 e2) = ??? -- apply the closure

Recall: At call: Use the frozen environment to evaluate the body

Exact code for you to figure out in HW4

Hint: Recall evaluating inc 10:

Evaluate inc to get <fro, x, x + c>
Evaluate 10 to get 10
Evaluate x + c in (x := 10) : fro

Let’s generalize to e1 e2:

Evaluate e1 to get <fro, param, body>
Evaluate e2 to get v2
Evaluate body in (param := v2) : fro

Advanced Features of Functions

Functions returning functions
- aka partial applications
Functions taking functions as arguments
- aka higher-order functions
Recursion

Does our eval support this?

QUIZ

What should the following evaluate to?

let add = \x y -> x + y
in
  let add1 = add 1
  in
    let add10 = add 10
    in
      add1 100 + add10 1000

(A) Runtime error

(B) 1102

(C) 1120

(D) 1111

Partial Applications Achieved!

Closures support functions returning functions!

let add = \x -> (\y -> x + y) --                           env0 = []
in                         -- env1 = ["add" := <[], x, \y -> x + y>]
  let add1 =
        add 1 -- eval ("x" := 1 : env0) (\y -> x + y)
              --   ==> <["x" := 1], y, x + y>
  in       -- env2 = ["add1" := <["x" := 1], y, x + y>, "add" := ...]
    let add10 =
      add 10 -- eval ("x" := 10 : env0) (\y -> x + y)
             --   ==> <["x" := 10], y, x + y>
    in  -- env3 = ["add10" := <["x" := 10], y, x + y>, "add1" := ...]
      add1 100 -- eval ["y" := 100, "x" := 1] (x + y)
               --   ==> 101
      +
      add10 1000 -- eval ["y" := 1000, "x" := 10] (x + y)
                 --  ==> 1010

==> 1111

QUIZ

What should the following evaluate to?

let inc = \x -> x + 1
in
  let doTwice = \f -> (\x -> f (f x))
  in
    doTwice inc 10

(A) Runtime error

(B) 11

(C) 12

Higher-order Functions Achieved

Closures support functions taking functions as arguments!

let inc = \x -> x + 1                                -- env0 = []
in                              -- env1 = [inc := <[], x, x + 1>]
  let doTwice = \f -> (\x -> f (f x))
  in  -- env2 = [doTwice := <env1, f, \x -> f (f x)>, inc := ...]
    ((doTwice inc) -- eval ("f" := <[],x,x + 1> : env1) (\x -> f (f x))
                   -- ==> <("f" := <[],x,x + 1> : env1), x, f (f x)>

      10)          -- eval ["x" := 10, "f" := <[],x,x + 1>, ...] f (f x)

-- f   ==> <[], x, x + 1>
-- x   ==> 10
-- <[], x, x + 1> 10 ==> eval ["x" := 10] x + 1 ==> 11
-- <[], x, x + 1> 11 ==> eval ["x" := 11] x + 1 ==> 12

==> 12

QUIZ

What does this evaluate to?

let f = \n -> n * f (n - 1)
in
  f 5

(A) 120

(B) Evaluation does not terminate

(C) Error: unbound variable f

let f = \n -> n * f (n - 1)
in -- [f := <[], n, n * f (n - 1)>]
  f 5 -- eval [n := 5] (n * f (n - 1))
      --  ==> unbound variable f!!!

Lesson learned: to support recursion, you need to put the function itself back into its closure environment before the body gets evaluated!

The Nano Language

Features of Nano:

Arithmetic [done]
Variables [done]
Let bindings [done]
Functions [done]
Recursion [you figure it out in HW4]

That’s all folks!