Scheme Programming/Numbers and Expressions
In this section, we'll continue to work with numbers, since they're familiar and easy to reason about. We'll also introduce the Scheme type system and consider more carefully how Scheme expressions are evaluated.
Types[edit  edit source]
First, though, what do we mean when we say a Scheme object is a "number" or a "boolean"? These names denote the type of an object. Very generally, knowing an object's type gives us information about what we can do with it. If we know that and both have type boolean, we know that the AND operation of logic is defined on and ; that is, the value of ( AND ) will be meaningful. We say that such a meaningful expression is welltyped. If, on the other hand, has type number, we have no way to evaluate ( AND ), because we don't know how to make sense of AND unless both of its arguments are booleans. In this case, the expression is illtyped and its value is undefined.
Scheme is a stronglytyped language, which means that illtyped expressions are forbidden; evaluating one will cause Scheme to report an error.^{[1]} For example, we can expect Scheme to reject the following expression:
> (+ 10 #t)
;; Error: (+) bad argument type  not a number: #t
From the error message, we can infer that the Scheme procedure
+
expects its arguments to be numbers. Since #t
is a boolean, the expression (+ 10 #t)
is illtyped;
no meaningful value can be obtained, so the execution of the program
halts. In this way, Scheme prevents further surprises which we
might encounter were execution to continue with a bogus value.
How do we get type information about Scheme objects? Scheme types are
defined by type predicates. These are procedures
which, given any Scheme object as an argument, return #t
if the object has a specific type, and #f
otherwise. By
applying a type predicate foo?
for a type foo, then,
we can divide the set of all Scheme objects into objects of type foo
and the rest.
So far, we've worked with numbers and booleans, which have the type
predicates number?
and, as you might have guessed,
boolean?
.
> (number? 10)
#t
> (number? #t)
#f
> (boolean? 3)
#f
> (boolean? #f)
#t
Scheme guarantees^{[2]} that the basic
types, including numbers and booleans, are disjoint. This
means, for example, that no Scheme object is a number and a
boolean; it might be one or the other or neither, but it cannot be
both. In typepredicate terms, there's no Scheme
object x
such that (number? x)
and
(boolean? x)
are both #t
.
We won't use basic type predicates like number?
too often
in this book, but it's important to understand how they define the
Scheme type system. They are used constantly in core Scheme and
library procedures, for example, to ensure that arguments are welltyped.
Numbers[edit  edit source]
We looked at several examples of simple numerical programs in the previous section. So far, though, the programs we've seen have only used integers. Scheme has a rich system of number types, called the numerical tower, which lets us compute with rational, real, and complex numbers.
> (+ 9/10 4/5)
17/10
> (* 2.423 5.39245)
13.06590635
> 2.762e8
276200000.0 ; 2.762 * 10^8
> ( 4+2i 1+7i)
35i
> (* 3+5i (+ 1.3 (/ 3 2)))
8.4+14.0i
As these examples show, Scheme also gives us many ways to write
numbers. Rational numbers are written in the form
a/b
, reals can be written in exponential
("scientific") notation as men
(which gives the
value ), and complex numbers are written in
the rectangular form a+bi
(or
abi
), where a is the real part and b the
imaginary part.^{[3]}
As we see in the final, nested example, Scheme allows us to mix different kinds of numbers without having to manually convert them to a common form.
Scheme also provides a wealth of numerical procedures. Here are a few examples:
> (gcd 36 60) ; greatest common divisor
12
> (min 3 4.7 2.1) ; min gives the minimum of its arguments
2.1
> (log 18) ; natural logarithm of 18
2.89037175789616
> (floorremainder 15 8) ; integer division remainder (modulo)
7
See R7RS § 6.2.6 for a comprehensive list of numerical procedures.
Expressions[edit  edit source]
By this point, we've seen Scheme evaluate a number of expressions
(and, hopefully, you've tried them out in your Scheme interpreter).
However, we've been vague about the meaning of Scheme expressions.
It's easy to guess that (+ 2 3)
is evaluated by adding
the numbers 2 and 3, but guessing isn't enough when we're faced with
complex nesting or expressions containing things less familiar than,
say, natural numbers or addition. We need a model of evaluation for
the Scheme expressions we've been seeing. What follows is a
simplified version, but one that is correct for the programs we'll
look at in this chapter.
The expression (+ 2 3)
may seem a bit trivial to
evaluate, but let's step through it. This expression is an
application of the operator +
to the operands
2
and 3
. To evaluate an application,
 First, evaluate the operator and operands.
 Then, apply the procedure which is the value of the operator to the values of the operands.
We saw in the very first example in "A taste of Scheme" that asking Scheme to evaluate a number gives us the number back as the value. This observation gives us the general rule of evaluation for numbers:
 Evaluating numbers: Numbers are selfevaluating.
With this rule in hand, it's trivial to evaluate the operands. Now
we need the value of +
, the operator. This value is a
procedure object. This value is opaque; that is, it's a "black
box" for performing addition that Scheme gives us, and we can't see
its inner workings. We can then apply this value to the values of
the operands to obtain 5, the value of the whole expression.
A more interesting example is the expression
(* (+ 2 3) ( 7 5))
. Here, the operator is
*
and the operands are (+ 2 3)
and
( 7 5)
. To find the value of this application,
we must evaluate these subexpressions, which are themselves
applications! Here is one possible sequence of evaluation
steps:^{[4]}
(* (+ 2 3) ( 7 5)) = (* 5 ( 7 5)) = (* 5 2) = 10
More complicated expressions are evaluated by exactly the same process; see exercise 1 below. So long as we are dealing with expressions built entirely out of applications, then, we know how to evaluate them; we recursively apply our evaluation rules until the expression is fully evaluated.
Exercises (Solutions) 


Notes[edit  edit source]
 ↑ Some popular languages,
including JavaScript and PHP, are weaklytyped. Roughly, these
languages permit combinations which would be illtyped in languages
with strong typing. This is often accomplished by automatically
converting values of one type to those of another; for example, the
JavaScript expression
4 + "five"
(which combines a number and a character string) evaluates to the string"4five"
. Here, the numeric operand4
has been implicitly converted to the string"4"
. Scheme, like other stronglytyped languages, performs no such magic; values must be converted explicitly.  ↑ R7RS § 3.2
 ↑ Scheme also provides
polar notation for complex
numbers. We can write
r@t
to denote the complex number with modulus r and argument t (in radians).  ↑ We don't actually know the order in which the operands are evaluated. At the moment, this makes no difference, but it is important to understand that Scheme will evaluate them in some (unspecified) order.