2.2 Simple Definitions and Expressions
A program module is written as
#lang ‹langname› ‹topform›*
where a ‹topform› is either a ‹definition› or an ‹expr›. The REPL also evaluates ‹topform›s.
In syntax specifications, text with a gray background, such as #lang, represents literal text. Whitespace must appear between such literals and nonterminals like ‹id›, except that whitespace is not required before or after (, ), [, or ]. A comment, which starts with ; and runs until the end of the line, is treated the same as whitespace.
Following the usual conventions, * in a grammar means zero or more repetitions of the preceding element, + means one or more repetitions of the preceding element, and {} groups a sequence as an element for repetition.
2.2.1 Definitions
A definition of the form
Definitions: define (later in this guide) explains more about definitions.
( define ‹id› ‹expr› )
binds ‹id› to the result of ‹expr›, while
( define ( ‹id› ‹id›* ) ‹expr›+ )
binds the first ‹id› to a function (also called a procedure) that takes arguments as named by the remaining ‹id›s. In the function case, the ‹expr›s are the body of the function. When the function is called, it returns the result of the last ‹expr›.
Examples: | |||
| |||
| |||
> five | |||
5 | |||
> (piece "hello world") | |||
"hello" |
Under the hood, a function definition is really the same as a non-function definition, and a function name does not have to be used in a function call. A function is just another kind of value, though the printed form is necessarily less complete than the printed form of a number or string.
Examples: |
> piece |
#<procedure:piece> |
#<procedure:substring> |
A function definition can include multiple expressions for the function’s body. In that case, only the value of the last expression is returned when the function is called. The other expressions are evaluated only for some side-effect, such as printing.
Examples: | ||||
| ||||
> (greet "universe") | ||||
returning a greeting for universe... | ||||
"hello universe" |
Scheme programmers prefer to avoid assignment statements. It’s important, though, to understand that multiple expressions are allowed in a definition body, because it explains why the following nogreet function simply returns its argument:
(define (nogreet name) |
string-append "hello " name) |
> (nogreet "world") |
"world" |
Within nogreet, there are no parentheses around string-append "hello " name, so they are three separate expressions instead of one function-call expression. The expressions string-append and "hello " are evaluated, but the results are never used. Instead, the result of the function is just the result of the expression name.
2.2.2 An Aside on Indenting Code
Line breaks and indentation are not significant for parsing Scheme programs, but most Scheme programmers use a standard set of conventions to make code more readable. For example, the body of a definition is typically indented under the first line of the definition. Identifiers are written immediately after an open parenthesis with no extra space, and closing parentheses never go on their own line.
DrScheme automatically indents according to the standard style when you type Enter in a program or REPL expression. For example, if you hit Enter after typing (define (greet name), then DrScheme automatically inserts two spaces for the next line. If you change a region of code, you can select it in DrScheme and hit Tab, and DrScheme will re-indent the code (without inserting any line breaks). Editors like Emacs offer a Scheme mode with similar indentation support.
Re-indenting not only makes the code easier to read, it gives you extra feedback that your parentheses are matched in the way that you intended. For example, if you leave out a closing parenthesis after the last argument to a function, automatic indentation starts the next line under the first argument, instead of under the define keyword:
(define (nogreet name |
(string-append "hello " name))) |
Furthermore, when an open parenthesis has no matching close parenthesis in a program, both mzscheme and DrScheme use the source’s indentation to suggest where it might be missing.
2.2.3 Identifiers
Scheme’s syntax for identifiers is especially liberal. Excluding the special characters
Identifiers and Binding (later in this guide) explains more about identifiers.
( ) [ ] { } " , ' ` ; # | \
and except for the sequences of characters that make number constants, almost any sequence of non-whitespace characters forms an ‹id›. For example substring is an identifier. Also, string-append and a+b are identifiers, as opposed to arithmetic expressions. Here are several more examples:
+ |
Apple |
integer? |
call/cc |
call-with-composable-continuation |
x-1+3i |
2.2.4 Function Calls (Procedure Applications)
We have already seen many function calls, which are called procedure applications in more traditional Scheme terminology. The syntax of a function call is
Function Calls (later in this guide) explains more about function calls.
( ‹id› ‹expr›* )
where the number of ‹expr›s determines the number of arguments supplied to the function named by ‹id›.
The scheme language pre-defines many function identifiers, such as substring and string-append. More examples are below.
In example Scheme code throughout the documentation, uses of pre-defined names are hyperlinked to the reference manual. So, you can click on an identifier to get full details about its use.
> (string-append "hello" " " "scheme") ; append strings |
"hello scheme" |
> (substring "hello scheme" 6 12) ; extract a substring |
"scheme" |
> (string-length "scheme") ; get a string's length |
6 |
> (string? "hello scheme") ; recognize strings |
#t |
> (string? 1) |
#f |
> (sqrt 16) ; find a square root |
4 |
> (sqrt -16) |
0+4i |
> (+ 1 2) ; add numbers |
3 |
> (- 2 1) ; subtract numbers |
1 |
> (< 2 1) ; compare numbers |
#f |
> (>= 2 1) |
#t |
> (number? "hello scheme") ; recognize numbers |
#f |
> (number? 1) |
#t |
> (equal? 1 "hello") ; compare anything |
#f |
> (equal? 1 1) |
#t |
2.2.5 Conditionals with if, and, or, and cond
The next simplest kind of expression is an if conditional:
( if ‹expr› ‹expr› ‹expr› )
Conditionals (later in this guide) explains more about conditionals.
The first ‹expr› is always evaluated. If it produces a non-#f value, then the second ‹expr› is evaluated for the result of the whole if expression, otherwise the third ‹expr› is evaluated for the result.
Examples: | |||
| |||
"smaller" |
(define (reply s) |
"hi!" |
"huh?")) |
> (reply "hello scheme") |
"hi!" |
> (reply "λx:(μα.α→α).xx") |
"huh?" |
Complex conditionals can be formed by nesting if expressions. For example, you could make the reply function work when given non-strings:
(define (reply s) |
"hi!" |
"huh?")) |
"huh?")) |
Instead of duplicating the "huh?" case, this function is better written as
(define (reply s) |
#f)) |
"hi!" |
"huh?")) |
but these kinds of nested ifs are difficult to read. Scheme provides more readable shortcuts through the and and or forms, which work with any number of expressions:
Combining Tests: and and or (later in this guide) explains more about and and or.
( and ‹expr›* ) |
( or ‹expr›* ) |
The and form short-circuits: it stops and returns #f when an expression produces #f, otherwise it keeps going. The or form similarly short-circuits when it encounters a true result.
Examples: | |||||||
| |||||||
> (reply "hello scheme") | |||||||
"hi!" | |||||||
> (reply 17) | |||||||
"huh?" |
Another common pattern of nested ifs involves a sequence of tests, each with its own result:
(define (reply-more s) |
"hi!" |
"bye!" |
(if (equal? "?" (substring s (- (string-length s) 1))) |
"I don't know" |
"huh?")))) |
The shorthand for a sequence of tests is the cond form:
Chaining Tests: cond (later in this guide) explains more about cond.
( cond {[ ‹expr› ‹expr›* ]}* )
A cond form contains a sequence of clauses between square brackets. In each clause, the first ‹expr› is a test expression. If it produces true, then the clause’s remaining ‹expr›s are evaluated, and the last one in the clause provides the answer for the entire cond expression; the rest of the clauses are ignored. If the test ‹expr› produces #f, then the clause’s remaining ‹expr›s are ignored, and evaluation continues with the next clause. The last clause can use else as a synonym for a #t test expression.
Using cond, the reply-more function can be more clearly written as follows:
(define (reply-more s) |
(cond |
"hi!"] |
"bye!"] |
[(equal? "?" (substring s (- (string-length s) 1))) |
"I don't know"] |
[else "huh?"])) |
> (reply-more "hello scheme") |
"hi!" |
> (reply-more "goodbye cruel world") |
"bye!" |
> (reply-more "what is your favorite color?") |
"I don't know" |
> (reply-more "mine is lime green") |
"huh?" |
The use of square brackets for cond clauses is a convention. In Scheme, parentheses and square brackets are actually interchangable, as long as ( is matched with ) and [ is matched with ]. Using square brackets in a few key places makes Scheme code even more readable.
2.2.6 Function Calls, Again
In our earlier grammar of function calls, we oversimplified. The actual syntax of a function call allows an arbitrary expression for the function, instead of just an ‹id›:
Function Calls (later in this guide) explains more about function calls.
( ‹expr› ‹expr›* )
The first ‹expr› is often an ‹id›, such as string-append or +, but it can be anything that evaluates to a function. For example, it can be a conditional expression:
(define (double v) |
((if (string? v) string-append +) v v)) |
> (double "hello") |
"hellohello" |
> (double 5) |
10 |
Syntactically, the first expression in a function call could even be a number – but that leads to an error, since a number is not a function.
> (1 2 3 4) |
procedure application: expected procedure, given: 1; |
arguments were: 2 3 4 |
When you accidentally omit a function name or when you use parentheses around an expression, you’ll most often get an “expected a procedure” error like this one.
2.2.7 Anonymous Functions with lambda
Programming in Scheme would be tedious if you had to name all of your numbers. Instead of writing (+ 1 2), you’d have to write
Functions: lambda (later in this guide) explains more about lambda.
> (define a 1) |
> (define b 2) |
> (+ a b) |
3 |
It turns out that having to name all your functions can be tedious, too. For example, you might have a function twice that takes a function and an argument. Using twice is convenient if you already have a name for the function, such as sqrt:
(define (twice f v) |
(f (f v))) |
> (twice sqrt 16) |
2 |
If you want to call a function that is not yet defined, you could define it, and then pass it to twice:
(define (louder s) |
(string-append s "!")) |
> (twice louder "hello") |
"hello!!" |
But if the call to twice is the only place where louder is used, it’s a shame to have to write a whole definition. In Scheme, you can use a lambda expression to produce a function directly. The lambda form is followed by identifiers for the function’s arguments, and then the function’s body expressions:
( lambda ( ‹id›* ) ‹expr›+ )
Evaluating a lambda form by itself produces a function:
> (lambda (s) (string-append s "!")) |
#<procedure> |
Using lambda, the above call to twice can be re-written as
| ||
"hello!!" | ||
| ||
"hello?!?!" |
Another use of lambda is as a result for a function that generates functions:
(define (make-add-suffix s2) |
(lambda (s) (string-append s s2))) |
> (twice (make-add-suffix "!") "hello") |
"hello!!" |
> (twice (make-add-suffix "?!") "hello") |
"hello?!?!" |
> (twice (make-add-suffix "...") "hello") |
"hello......" |
Scheme is a lexically scoped language, which means that s2 in the function returned by make-add-suffix always refers to the argument for the call that created the function. In other words, the lambda-generated function “remembers” the right s2:
> (define louder (make-add-suffix "!")) |
> (define less-sure (make-add-suffix "?")) |
> (twice less-sure "really") |
"really??" |
> (twice louder "really") |
"really!!" |
We have so far referred to definitions of the form (define ‹id› ‹expr›) as “non-function definitions.” This characterization is misleading, because the ‹expr› could be a lambda form, in which case the definition is equivalent to using the “function” definition form. For example, the following two definitions of louder are equivalent:
(define (louder s) |
(string-append s "!")) |
|
(define louder |
(lambda (s) |
(string-append s "!"))) |
> louder |
#<procedure:louder> |
Note that the expression for louder in the second case is an “anonymous” function written with lambda, but, if possible, the compiler infers a name, anyway, to make printing and error reporting as informative as possible.
2.2.8 Local Binding with define, let, and let*
It’s time to retract another simplification in our grammar of Scheme. In the body of a function, definitions can appear before the body expressions:
Internal Definitions (later in this guide) explains more about local (internal) definitions.
( define ( ‹id› ‹id›* ) ‹definition›* ‹expr›+ ) |
( lambda ( ‹id›* ) ‹definition›* ‹expr›+ ) |
Definitions at the start of a function body are local to the function body.
Examples: | ||||||||||
| ||||||||||
> (converse "hello!") | ||||||||||
"hi!" | ||||||||||
> (converse "urp") | ||||||||||
"huh?" | ||||||||||
> starts? ; outside of converse, so... | ||||||||||
reference to undefined identifier: starts? |
Another way to create local bindings is the let form. An advantage of let is that it can be used in any expression position. Also, let binds many identifiers at once, instead of requiring a separate define for each identifier.
Internal Definitions (later in this guide) explains more about let and let*.
( let ( {[ ‹id› ‹expr› ]}* ) ‹expr›+ )
Each binding clause is an ‹id› and a ‹expr› surrounded by square brackets, and the expressions after the clauses are the body of the let. In each clause, the ‹id› is bound to the result of the ‹expr› for use in the body.
| |||
"adding 1 and 2 produces 3" |
The bindings of a let form are available only in the body of the let, so the binding clauses cannot refer to each other. The let* form, in contrast, allows later clauses to use earlier bindings:
| ||||
"adding 1 and 2 produces 3" |