figure: io
Simply Scheme: Introducing Computer Science ch 20: Input and Output
Simply Scheme: Introducing Computer Science 2/e Copyright (C) 1999 MIT

Chapter 20

Input and Output

cover photo
Brian Harvey
University of California, Berkeley
Matthew Wright
University of California, Santa Barbara

Download PDF version
Back to Table of Contents
BACK chapter thread NEXT
MIT Press web page for Simply Scheme

In the tic-tac-toe project in Chapter 10, we didn't write a complete game program. We wrote a function that took a board position and x or o as arguments, returning the next move. We noted at the time that a complete game program would also need to carry on a conversation with the user. Instead of computing and returning one single value, a conversational program must carry out a sequence of events in time, reading information from the keyboard and displaying other information on the screen.

Before we complete the tic-tac-toe project, we'll start by exploring Scheme's mechanisms for interactive programming.

Printing

Up until now, we've never told Scheme to print anything. The programs we've written have computed values and returned them; we've relied on the read-eval-print loop to print these values.[1]

But let's say we want to write a program to print out all of the words to "99 Bottles of Beer on the Wall." We could implement a function to produce a humongous list of the lines of the song, like this:

(define (bottles n)
  (if (= n 0)
      '()
      (append (verse n)
	      (bottles (- n 1)))))

(define (verse n)
  (list (cons n '(bottles of beer on the wall))
	(cons n '(bottles of beer))
	'(if one of those bottles should happen to fall)
	(cons (- n 1) '(bottles of beer on the wall))
	'()))

> (bottles 3)
((3 BOTTLES OF BEER ON THE WALL)
 (3 BOTTLES OF BEER)
 (IF ONE OF THOSE BOTTLES SHOULD HAPPEN TO FALL)
 (2 BOTTLES OF BEER ON THE WALL)
 ()
 (2 BOTTLES OF BEER ON THE WALL)
 (2 BOTTLES OF BEER)
 (IF ONE OF THOSE BOTTLES SHOULD HAPPEN TO FALL)
 (1 BOTTLES OF BEER ON THE WALL)
 ()
 (1 BOTTLES OF BEER ON THE WALL)
 (1 BOTTLES OF BEER)
 (IF ONE OF THOSE BOTTLES SHOULD HAPPEN TO FALL)
 (0 BOTTLES OF BEER ON THE WALL)
 ())

The problem is that we don't want a list. All we want is to print out the lines of the song; storing them in a data structure is unnecessary and inefficient. Also, some versions of Scheme would print the above list like this:

((3 BOTTLES OF BEER ON THE WALL) (3 BOTTLES OF BEER) (IF ONE OF
 THOSE BOTTLES SHOULD HAPPEN TO FALL) (2 BOTTLES OF BEER ON THE
 WALL) () (2 BOTTLES OF BEER ON THE WALL) (2 BOTTLES OF BEER) (IF
 ONE OF THOSE BOTTLES SHOULD HAPPEN TO FALL) (1 BOTTLES OF BEER ON
 THE WALL) () (1 BOTTLES OF BEER ON THE WALL) (1 BOTTLES OF BEER)
 (IF ONE OF THOSE BOTTLES SHOULD HAPPEN TO FALL) (0 BOTTLES OF BEER
 ON THE WALL) ())

or even all on one line. We can't rely on Scheme's mechanism for printing lists if we want to be sure of a particular arrangement on the screen.

Instead we'll write a program to print a verse, rather than return it in a list:

(define (bottles n)
  (if (= n 0)
      'burp
      (begin (verse n)
	     (bottles (- n 1)))))

(define (verse n)
  (show (cons n '(bottles of beer on the wall)))
  (show (cons n '(bottles of beer)))
  (show '(if one of those bottles should happen to fall))
  (show (cons (- n 1) '(bottles of beer on the wall)))
  (show '()))

> (bottles 3)
(3 BOTTLES OF BEER ON THE WALL)
(3 BOTTLES OF BEER)
(IF ONE OF THOSE BOTTLES SHOULD HAPPEN TO FALL)
(2 BOTTLES OF BEER ON THE WALL)
()
(2 BOTTLES OF BEER ON THE WALL)
(2 BOTTLES OF BEER)
(IF ONE OF THOSE BOTTLES SHOULD HAPPEN TO FALL)
(1 BOTTLES OF BEER ON THE WALL)
()
(1 BOTTLES OF BEER ON THE WALL)
(1 BOTTLES OF BEER)
(IF ONE OF THOSE BOTTLES SHOULD HAPPEN TO FALL)
(0 BOTTLES OF BEER ON THE WALL)
()
BURP

Notice that Scheme doesn't print an outer set of parentheses. Each line was printed separately; there isn't one big list containing all of them.[2]

Why was "burp" printed at the end? Just because we're printing things explicitly doesn't mean that the read-eval-print loop stops functioning. We typed the expression (bottles 3). In the course of evaluating that expression, Scheme printed several lines for us. But the value of the expression was the word burp, because that's what bottles returned.

Side Effects and Sequencing

How does our program work? There are two new ideas here: side effects and sequencing.

Until now, whenever we've invoked a procedure, our only goal has been to get a return value. The procedures we've used compute and return a value, and do nothing else. Show is different. Although every Scheme procedure returns a value, the Scheme language standard doesn't specify what value the printing procedures should return.[3] Instead, we are interested in their side effects. In other words, we invoke show because we want it to do something, namely, print its argument on the screen.

What exactly do we mean by "side effect"? The kinds of procedures that we've used before this chapter can compute values, invoke helper procedures, provide arguments to the helper procedures, and return a value. There may be a lot of activity going on within the procedure, but the procedure affects the world outside of itself only by returning a value that some other procedure might use. Show affects the world outside of itself by putting something on the screen. After show has finished its work, someone who looks at the screen can tell that show was used.[4]

Here's an example to illustrate the difference between values and effects:

(define (effect x)
  (show x)
  'done)

(define (value x)
  x)

> (effect '(oh! darling))
(OH! DARLING)
DONE

> (value '(oh! darling))
(OH! DARLING)

> (bf (effect '(oh! darling)))
(OH! DARLING)
ONE
> (bf (value '(oh! darling)))
(DARLING)

> (define (lots-of-effect x)
    (effect x)
    (effect x)
    (effect x))

> (define (lots-of-value x)
    (value x)
    (value x)
    (value x))

> (lots-of-effect '(oh! darling))
(OH! DARLING)
(OH! DARLING)
(OH! DARLING)
DONE

> (lots-of-value '(oh! darling))
(OH! DARLING)

This example also demonstrates the second new idea, sequencing: Each of effect, lots-of-effect, and lots-of-value contains more than one expression in its body. When you invoke such a procedure, Scheme evaluates all the expressions in the body, in order, and returns the value of the last one.[5] This also works in the body of a let, which is really the body of a procedure, and in each clause of a cond.[6]

When we invoked lots-of-value, Scheme invoked value three times; it discarded the values returned by the first two invocations, and returned the value from the third invocation. Similarly, when we invoked lots-of-effect, Scheme invoked effect three times and returned the value from the third invocation. But each invocation of effect caused its argument to be printed by invoking show.

The Begin Special Form

The lots-of-effect procedure accomplished sequencing by having more than one expression in its body. This works fine if the sequence of events that you want to perform is the entire body of a procedure. But in bottles we wanted to include a sequence as one of the alternatives in an if construction. We couldn't just say

(define (bottles n)                          ;; wrong
  (if (= n 0)
      '()
      (verse n)
      (bottles (- n 1))))

because if must have exactly three arguments. Otherwise, how would if know whether we meant (verse n) to be the second expression in the true case, or the first expression in the false case?

Instead, to turn the sequence of expressions into a single expression, we use the special form begin. It takes any number of arguments, evaluates them from left to right, and returns the value of the last one.

(define bottles n)
  (if (= n 0)
      'burp
      (begin (verse n)
	     (bottles (- n 1)))))

(One way to think about sequences in procedure bodies is that every procedure body has an invisible begin surrounding it.)

This Isn't Functional Programming

Sequencing and side effects are radical departures from the idea of functional programming. In fact, we'd like to reserve the name function for something that computes and returns one value, with no side effects. "Procedure" is the general term for the thing that lambda returns—an embodiment of an algorithm. If the algorithm is the kind that computes and returns a single value without side effects, then we say that the procedure implements a function.[7]

There is a certain kind of sequencing even in functional programming. If you say

(* (+ 3 4) (- 92 15))

it's clear that the addition has to happen before the multiplication, because the result of the addition provides one of the arguments to the multiplication. What's new in the sequential programming style is the emphasis on sequence, and the fact that the expressions in the sequence are independent instead of contributing values to each other. In this multiplication problem, for example, we don't care whether the addition happens before or after the subtraction. If the addition and subtraction were in a sequence, we'd be using them for independent purposes:

(begin
 (show (+ 3 4))
 (show (- 92 15)))

This is what we mean by being independent. Neither expression helps in computing the other. And the order matters because we can see the order in which the results are printed.

Not Moving to the Next Line

Each invocation of show prints a separate line. What if we want a program that prints several things on the same line, like this:

> (begin (show-addition 3 4)
	 (show-addition 6 8)
	 'done)
3+4=7
6+8=14
DONE

We use display, which doesn't move to the next line after printing its argument:

(define (show-addition x y)
  (display x)
  (display '+)
  (display y)
  (display '=)
  (show (+ x y)))

(The last one is a show because we do want to start a new line after it.)

What if you just want to print a blank line? You use newline:

(define (verse n)
  (show (cons n '(bottles of beer on the wall)))
  (show (cons n '(bottles of beer)))
  (show '(if one of those bottles should happen to fall))
  (show (cons (- n 1) '(bottles of beer on the wall)))
  (newline))                                 ; replaces (show '())

In fact, show isn't an official Scheme primitive; we wrote it in terms of display and newline.

Strings

Throughout the book we've occasionally used strings, that is, words enclosed in double-quote marks so that Scheme will permit the use of punctuation or other unusual characters. Strings also preserve the case of letters, so they can be used to beautify our song even more. Since any character can be in a string, including spaces, the easiest thing to do in this case is to treat all the letters, spaces, and punctuation characters of each line of the song as one long word. (If we wanted to be able to compute functions of the individual words in each line, that wouldn't be such a good idea.)

(define (verse n)
  (display n)
  (show " bottles of beer on the wall,")
  (display n)
  (show " bottles of beer.")
  (show "If one of those bottles should happen to fall,")
  (display (- n 1))
  (show " bottles of beer on the wall.")
  (newline))

> (verse 6)
6 bottles of beer on the wall,
6 bottles of beer.
If one of those bottles should happen to fall,
5 bottles of beer on the wall.

#F                               ; or whatever is returned by (newline)

It's strange to think of " bottles of beer on the wall," as a single word. But the rule is that anything inside double quotes counts as a single word. It doesn't have to be an English word.

A Higher-Order Procedure for Sequencing

Sometimes we want to print each element of a list separately:

(define (show-list lst)
  (if (null? lst)
      'done
      (begin (show (car lst))
	     (show-list (cdr lst)))))

> (show-list '((dig a pony) (doctor robert) (for you blue)))
(DIG A PONY)
(DOCTOR ROBERT)
(FOR YOU BLUE)
DONE

Like other patterns of computation involving lists, this one can be abstracted into a higher-order procedure. (We can't call it a "higher-order function" because this one is for computations with side effects.) The procedure for-each is part of standard Scheme:

> (for-each show '((mean mr mustard) (no reply) (tell me why)))
(MEAN MR MUSTARD)
(NO REPLY)
(TELL ME WHY)

The value returned by for-each is unspecified.

Why couldn't we just use map for this purpose? There are two reasons. One is just an efficiency issue: Map constructs a list containing the values returned by each of its sub-computations; in this example, it would be a list of three instances of the unspecified value returned by show. But we aren't going to use that list for anything, so there's no point in constructing it. The second reason is more serious. In functional programming, the order of evaluation of subexpressions is unspecified. For example, when we evaluate the expression

(- (+ 4 5) (* 6 7))

we don't know whether the addition or the multiplication happens first. Similarly, the order in which map computes the results for each element is unspecified. That's okay as long as the ultimately returned list of results is in the right order. But when we are using side effects, we do care about the order of evaluation. In this case, we want to make sure that the elements of the argument list are printed from left to right. For-each guarantees this ordering.

Tic-Tac-Toe Revisited

We're working up toward playing a game of tic-tac-toe against the computer. But as a first step, let's have the computer play against itself. What we already have is ttt, a strategy function: one that takes a board position as argument (and also a letter x or o) and returns the chosen next move. In order to play a game of tic-tac-toe, we need two players; to make it more interesting, each should have its own strategy. So we'll write another one, quickly, that just moves in the first empty square it sees:

(define (stupid-ttt position letter)
  (location '_ position))

(define (location letter word)
  (if (equal? letter (first word))
      1
      (+ 1 (location letter (bf word)))))

Now we can write a program that takes two strategies as arguments and actually plays a game between them.

(define (play-ttt x-strat o-strat)
  (play-ttt-helper x-strat o-strat '_________ 'x))

(define (play-ttt-helper x-strat o-strat position whose-turn)
  (cond ((already-won? position (opponent whose-turn))
	 (list (opponent whose-turn) 'wins!))
	((tie-game? position) '(tie game))
	(else (let ((square (if (equal? whose-turn 'x)
				(x-strat position 'x)
				(o-strat position 'o))))
		(play-ttt-helper x-strat
				 o-strat
				 (add-move square whose-turn position)
				 (opponent whose-turn))))))
We use a helper procedure because we need to keep track of two pieces of information besides the strategy procedures: the current board position and whose turn it is (x or o). The helper procedure is invoked recursively for each move. First it checks whether the game is already over (won or tied).[8] If not, the helper procedure invokes the current player's strategy procedure, which returns the square number for the next move. For the recursive call, the arguments are the same two strategies, the new position after the move, and the letter for the other player.

We still need add-move, the procedure that takes a square and an old position as arguments and returns the new position.

(define (add-move square letter position)
  (if (= square 1)
      (word letter (bf position))
      (word (first position)
	    (add-move (- square 1) letter (bf position)))))

> (play-ttt ttt stupid-ttt)
(X WINS!)

> (play-ttt stupid-ttt ttt)
(O WINS!)

Accepting User Input

The work we did in the last section was purely functional. We didn't print anything (except the ultimate return value, as always) and we didn't have to read information from a human player, because there wasn't one.

You might expect that the structure of an interactive game program would be very different, with a top-level procedure full of sequential operations. But the fact is that we hardly have to change anything to turn this into an interactive game. All we need is a new "strategy" procedure that asks the user where to move, instead of computing a move based on built-in rules.

(define (ask-user position letter)
  (print-position position)
  (display letter)
  (display "'s move: ")
  (read))

(define (print-position position)            ;; first version
  (show position))

(Ultimately we're going to want a beautiful two-dimensional display of the current position, but we don't want to get distracted by that just now. That's why we've written a trivial temporary version.)

> (play-ttt ttt ask-user)
____X____
O'S MOVE: 1
O___XX___
O'S MOVE: 4
O__OXXX__
O'S MOVE: 3
OXOOXXX__
O'S MOVE: 8
(TIE GAME)

What the user typed is just the single digits shown in boldface at the ends of the lines.

What's new here is that we invoke the procedure read. It waits for you to type a Scheme expression, and returns that expression. Don't be confused: Read does not evaluate what you type. It returns exactly the same expression that you type:

(define (echo)
  (display "What? ")
  (let ((expr (read)))
    (if (equal? expr 'stop)
	'okay
	(begin
	 (show expr)
	 (echo)))))

> (echo)
What? hello
HELLO
What? (+ 2 3)
(+ 2 3)
What? (first (glass onion))
(FIRST (GLASS ONION))
What? stop
OKAY

Aesthetic Board Display

Here's our beautiful position printer:

(define (print-position position)
  (print-row (subword position 1 3))
  (show "-+-+-")
  (print-row (subword position 4 6))
  (show "-+-+-")
  (print-row (subword position 7 9))
  (newline))

(define (print-row row)
  (maybe-display (first row))
  (display "|")
  (maybe-display (first (bf row)))
  (display "|")
  (maybe-display (last row))
  (newline))

(define (maybe-display letter)
  (if (not (equal? letter '_))
      (display letter)
      (display " ")))

(define (subword wd start end)
  ((repeated bf (- start 1))
   ((repeated bl (- (count wd) end))
    wd)))[9]
Here's how it works:

> (print-position '_x_oo__xx)
 |X| 
-+-+-
O|O| 
-+-+-
 |X|X

Reading and Writing Normal Text

The read procedure works fine as long as what you type looks like a Lisp program. That is, it reads one expression at a time. In the tic-tac-toe program the user types a single number, which is a Scheme expression, so read works fine. But what if we want to read more than one word?

(define (music-critic)                       ;; first version
  (show "What's your favorite Beatles song?")
  (let ((song (read)))
    (show (se "I like" song "too."))))

> (music-critic)
What's your favorite Beatles song?
She Loves You
(I like SHE too.)

If the user had typed the song title in parentheses, then it would have been a single Scheme expression and read would have accepted it. But we don't want the users of our program to have to be typing parentheses all the time.

Scheme also lets you read one character at a time. This allows you to read any text, with no constraints on its format. The disadvantage is that you find yourself putting a lot of effort into minor details. We've provided a procedure read-line that reads one line of input and returns a sentence. The words in that sentence will contain any punctuation characters that appear on the line, including parentheses, which are not interpreted as sublist delimiters by read-line. Read-line also preserves the case of letters.

(define (music-critic)                       ;; second version
  (read-line)   ; See explanation below.
  (show "What's your favorite Beatles song?")
  (let ((song (read-line)))
    (show (se "I like" song "too."))))

> (music-critic)
What's your favorite Beatles song?
She Loves You
(I like She Loves You too.)

Why do we call read-line and ignore its result at the beginning of music-critic? It has to do with the interaction between read-line and read. Read treats what you type as a sequence of Scheme expressions; each invocation of read reads one of them. Read pays no attention to formatting details, such as several consecutive spaces or line breaks. If, for example, you type several expressions on the same line, it will take several invocations of read to read them all.

By contrast, read-line treats what you type as a sequence of lines, reading one line per invocation, so it does pay attention to line breaks.

Either of these ways to read input is sensible in itself, but if you mix the two, by invoking read sometimes and read-line sometimes in the same program, the results can be confusing. Suppose you type a line containing an expression and your program invokes read to read it. Since there might have been another expression on the line, read doesn't advance to the next line until you ask for the next expression. So if you now invoke read-line, thinking that it will read another line from the keyboard, it will instead return an empty list, because what it sees is an empty line—what's left after read uses up the expression you typed.

You may be thinking, "But music-critic doesn't call read!" That's true, but Scheme itself used read to read the expression that you used to invoke music-critic. So the first invocation of read-line is needed to skip over the spurious empty line.

Our solution works only if music-critic is invoked directly at a Scheme prompt. If music-critic were a subprocedure of some larger program that has already called read-line before calling music-critic, the extra read-line in music-critic would really read and ignore a useful line of text.

If you write a procedure using read-line that will sometimes be called directly and sometimes be used as a subprocedure, you can't include an extra read-line call in it. Instead, when you call your procedure directly from the Scheme prompt, you must say

> (begin (read-line) (my-procedure))

Another technical detail about read-line is that since it preserves the capitalization of words, its result may include strings, which will be shown in quotation marks if you return the value rather than showing it:

(define (music-critic-return)
  (read-line)
  (show "What's your favorite Beatles song?")
  (let ((song (read-line)))
    (se "I like" song "too.")))

> (music-critic-return)
What's your favorite Beatles song?
She Loves You
("I like" "She" "Loves" "You" "too.")

We have also provided show-line, which takes a sentence as argument. It prints the sentence without surrounding parentheses, followed by a newline. (Actually, it takes any list as argument; it prints all the parentheses except for the outer ones.)

(define (music-critic)
  (read-line)
  (show "What's your favorite Beatles song?")
  (let ((song (read-line)))
    (show-line (se "I like" song "too."))))

> (music-critic)
What's your favorite Beatles song?
She Loves You
I like She Loves You too.

The difference between show and show-line isn't crucial. It's just a matter of a pair of parentheses. The point is that read-line and show-line go together. Read-line reads a bunch of disconnected words and combines them into a sentence. Show-line takes a sentence and prints it as if it were a bunch of disconnected words. Later, when we read and write files in Chapter 22, this ability to print in the same form in which we read will be important.

Formatted Text

We've been concentrating on the use of sequential programming with explicit printing instructions for the sake of conversational programs. Another common application of sequential printing is to display tabular information, such as columns of numbers. The difficulty is to get the numbers to line up so that corresponding digits are in the same position, even when the numbers have very widely separated values. The align function can be used to convert a number to a printable word with a fixed number of positions before and after the decimal point:

(define (square-root-table nums)
  (if (null? nums)
      'done
      (begin (display (align (car nums) 7 1))
	     (show (align (sqrt (car nums)) 10 5))
	     (square-root-table (cdr nums)))))

> (square-root-table '(7 8 9 10 20 98 99 100 101 1234 56789))
    7.0   2.64575
    8.0   2.82843
    9.0   3.00000
   10.0   3.16228
   20.0   4.47214
   98.0   9.89949
   99.0   9.94987
  100.0  10.00000
  101.0  10.04988
 1234.0  35.12834
56789.0 238.30443
DONE

Align takes three arguments. The first is the value to be displayed. The second is the width of the column in which it will be displayed; the returned value will be a word with that many characters in it. The third argument is the number of digits that should be displayed to the right of the decimal point. (If this number is zero, then no decimal point will be displayed.) The width must be great enough to include all the digits, as well as the decimal point and minus sign, if any.

As the program example above indicates, align does not print anything. It's a function that returns a value suitable for printing with display or show.

What if the number is too big to fit in the available space?

> (align 12345679 4 0)
"123+"

Align returns a word containing the first few digits, as many as fit, ending with a plus sign to indicate that part of the value is missing.

Align can also be used to include non-numeric text in columns. If the first argument is not a number, then only two arguments are needed; the second is the column width. In this case align returns a word with extra spaces at the right, if necessary, so that the argument word will appear at the left in its column:

(define (name-table names)
  (if (null? names)
      'done
      (begin (display (align (cadar names) 11))
	     (show (caar names))
	     (name-table (cdr names)))))

> (name-table '((john lennon) (paul mccartney)
		(george harrison) (ringo starr)))
LENNON     JOHN
MCCARTNEY  PAUL
HARRISON   GEORGE
STARR      RINGO
DONE

As with numbers, if a non-numeric word won't fit in the allowed space, align returns a partial word ending with a plus sign.

This align function is not part of standard Scheme. Most programming languages, including some versions of Scheme, offer much more elaborate formatting capabilities with many alternate ways to represent both numbers and general text. Our version is a minimal capability to show the flavor and to meet the needs of projects in this book.

Sequential Programming and Order of Evaluation

Our expanded tic-tac-toe program includes both functional and sequential parts. The program computes its strategy functionally but uses sequences of commands to control the user interface by alternately printing information to the screen and reading information from the keyboard.

By adding sequential programming to our toolkit, we've increased our ability to write interactive programs. But there is a cost that goes along with this benefit: We now have to pay more attention to the order of events than we did in purely functional programs.

The obvious concern about order of events is that sequences of show expressions must come in the order in which we want them to appear, and read expressions must fit into the sequence properly so that the user is asked for the right information at the right time.

But there is another, less obvious issue about order of events. When the evaluation of expressions can have side effects in addition to returning values, the order of evaluation of argument subexpressions becomes important. Here's an example to show what we mean. Suppose we type the expression

(list (+ 3 4) (- 10 2))

The answer, of course, is (7 8). It doesn't matter whether Scheme computes the seven first (left to right) or the eight first (right to left). But here's a similar example in which it does matter:

(define (show-and-return x)
  (show x)
  x)

> (list (show-and-return (+ 3 4)) (show-and-return (- 10 2)))
8
7
(7 8)

The value that's ultimately returned, in this example, is the same as before. But the two numeric values that go into the list are also printed separately, so we can see which is computed first. (We've shown the case of right-to-left computation; your Scheme might be different.)

Suppose you want to make sure that the seven prints first, regardless of which order your Scheme uses. You could do this:

> (let ((left (show-and-return (+ 3 4))))
    (list left (show-and-return (- 10 2))))
7
8
(7 8)

The expression in the body of a let can't be evaluated until the let variables (such as left) have had their values computed.

It's hard to imagine a practical use for the artificial show-and-return procedure, but a similar situation arises whenever we use read. Suppose we want to write a procedure to ask a person for his or her full name, returning a two-element list containing the first and last name. A natural mistake to make would be to write this procedure:

(define (ask-for-name)                       ;; wrong
  (show "Please type your first name, then your last name:")
  (list (read) (read)))

> (ask-for-name)
Please type your first name, then your last name:
John
Lennon
(LENNON JOHN)

What went wrong? We happen to be using a version of Scheme that evaluates argument subexpressions from right to left. Therefore, the word John was read by the rightmost call to read, which provided the second argument to list. The best solution is to use let as we did above:

(define (ask-for-name)
  (show "Please type your first name, then your last name:")
  (let ((first-name (read)))
    (list first-name (read))))

Even this example looks artificially simple, because of the two invocations of read that are visibly right next to each other in the erroneous version. But look at play-ttt-helper. The word read doesn't appear in its body at all. But when we invoke it using ask-user as the strategy procedure for x, the expression

(x-strat position 'x)

hides an invocation of read. The structure of play-ttt-helper includes a let that controls the timing of that read. (As it turns out, in this particular case we could have gotten away with writing the program without let. The hidden invocation of read is the only subexpression with a side effect, so there aren't two effects that might get out of order. But we had to think carefully about the program to be sure of that.)

Pitfalls

It's easy to get confused about what is printed explicitly by your program and what is printed by Scheme's read-eval-print loop. Until now, all printing was of the second kind. Here's an example that doesn't do anything very interesting but will help make the point clear:

(define (name)
  (display "MATT ")
  'wright)

> (name)
MATT WRIGHT

At first glance it looks as if putting the word "Matt" inside a call to display is unnecessary. After all, the word wright is printed even without using display. But watch this:

> (bf (name))
MATT RIGHT

Every time you invoke name, whether or not as the entire expression used at a Scheme prompt, the word MATT is printed. But the word wright is returned, and may or may not be printed depending on the context in which name is invoked.

A sequence of expressions returns the value of the last expression. If that isn't what you want, you must remember the value you want to return using let:

(let ((result (compute-this-first)))
  (begin
   (compute-this-second)
   (compute-this-third)
   result))

Don't forget that the first call to read-line, or any call to read-line after a call to read, will probably read the empty line that read left behind.

Sometimes you want to use what the user typed more than once in your program. But don't forget that read has an effect as well as a return value. Don't try to read the same expression twice:

(define (ask-question question)              ;; wrong
  (show question)
  (cond ((equal? (read) 'yes) #t)
	((equal? (read) 'no) #f)
	(else (show "Please answer yes or no.")
	      (ask-question question))))

If the answer is yes, this procedure will work fine. But if not, the second invocation of read will read a second expression, not test the same expression again as intended. To avoid this problem, invoke read only once for each expression you want to read, and use let to remember the result:

(define (ask-question question)
  (show question)
  (let ((answer (read)))
    (cond ((equal? answer 'yes) #t)
	  ((equal? answer 'no) #f)
	  (else (show "Please answer yes or no.")
		(ask-question question)))))

Boring Exercises

20.1  What happens when we evaluate the following expression? What is printed, and what is the return value? Try to figure it out in your head before you try it on the computer.

(cond ((= 2 3) (show '(lady madonna)) '(i call your name))
      ((< 2 3) (show '(the night before)) '(hello little girl))
      (else '(p.s. i love you)))

20.2  What does newline return in your version of Scheme?

20.3  Define show in terms of newline and display.

Real Exercises

20.4  Write a program that carries on a conversation like the following example. What the user types is in boldface.

> (converse)
Hello, I'm the computer.  What's your name? Brian Harvey
Hi, Brian.  How are you? I'm fine.
Glad to hear it.

20.5  Our name-table procedure uses a fixed width for the column containing the last names of the people in the argument list. Suppose that instead of liking British-invasion music you are into late romantic Russian composers:

> (name-table '((piotr tchaikovsky) (nicolay rimsky-korsakov)
		(sergei rachmaninov) (modest musorgsky)))

Alternatively, perhaps you like jazz:

> (name-table '((bill evans) (paul motian) (scott lefaro)))

Modify name-table so that it figures out the longest last name in its argument list, adds two for spaces, and uses that number as the width of the first column.

20.6  The procedure ask-user isn't robust. What happens if you type something that isn't a number, or isn't between 1 and 9? Modify it to check that what the user types is a number between 1 and 9. If not, it should print a message and ask the user to try again.

20.7  Another problem with ask-user is that it allows a user to request a square that isn't free. If the user does this, what happens? Fix ask-user to ensure that this can't happen.

20.8  At the end of the game, if the computer wins or ties, you never find out which square it chose for its final move. Modify the program to correct this. (Notice that this exercise requires you to make play-ttt-helper non-functional.)

20.9  The way we invoke the game program isn't very user-friendly. Write a procedure game that asks you whether you wish to play x or o, then starts a game. (By definition, x plays first.) Then write a procedure games that allows you to keep playing repeatedly. It can ask "do you want to play again?" after each game. (Make sure that the outcome of each game is still reported, and that the user can choose whether to play x or o before each game.)


[1] The only exception is that we've used trace, which prints messages about the progress of a computation.

[2] We know that it's still not as beautiful as can be, because of the capital letters and parentheses, but we'll get to that later.

[3] Suppose show returns #f in your version of Scheme. Then you might see

> (show 7)
7
#F

But since the return value is unspecified, we try to write programs in such a way that we never use show's return value as the return value from our procedures. That's why we return values like burp.

[4] The term side effect is based on the idea that a procedure may have a useful return value as its main purpose and may also have an effect "on the side." It's a misnomer to talk about the side effect of show, since the effect is its main purpose. But nobody ever says "side return value"!

[5] In Chapter 4, we said that the body of a procedure was always one single expression. We lied. But as long as you don't use any procedures with side effects, it doesn't do you any good to evaluate more than one expression in a body.

[6] For example:

> (cond ((< 4 0)
	 (show '(how interesting))
	 (show '(4 is less than zero?))
	 #f)
	((> 4 0)
	 (show '(more reasonable))
	 (show '(4 really is more than zero))
	 'value)
	(else
	 (show '(you mean 4=0?))
	 #f))
(MORE REASONABLE)
(4 REALLY IS MORE THAN ZERO)
VALUE

[7] Sometimes people sloppily say that the procedure is a function. In fact, you may hear people be really sloppy and call a non-functional procedure a function!

[8] You wrote the procedures already-won? and tie-game? in Exercises 10.1 and 10.2:

(define (already-won? position who)
  (member? (word who who who)  (find-triples position)))

(define (tie-game? position)
  (not (member? '_ position)))

[9] Alternate version:

(define (subword wd start end)
  (cond ((> start 1) (subword (bf wd) (- start 1) (- end 1)))
	((< end (count wd)) (subword (bl wd) start end))
	(else wd)))

You can take your choice, depending on which you think is easier, recursion or higher-order functions.

(back to Table of Contents)

BACK chapter thread NEXT

Brian Harvey, bh@cs.berkeley.edu