Arc Tutorial

Paul Graham’s tutorial on Arc, re-printed here for convenience:

[This is a brief tutorial on Arc.  It's intended for readers with
little programming experience and no Lisp experience.  It is thus
also an introduction to Lisp.]

Arc programs consist of expressions.  The simplest expressions are
things like numbers and strings, which evaluate to themselves.

arc> 25
arc> "foo"

Several expressions enclosed within parentheses are also an expression.
These are called lists.  When a list is evaluated, the elements are
evaluated from left to right, and the value of the first (presumably
a function) is passed the values of the rest.  Whatever it returns
is returned as the value of the expression.

arc> (+ 1 2)

Here's what just happened.  First +, 1, and 2 were evaluated,
returning the plus function, 1, and 2 respectively.  1 and 2 were
then passed to the plus function, which returned 3, which was
returned as the value of the whole expression.

(Macros introduce a twist, because they transform lists before
they're evaluated.  We'll get to them later.)

Since expression and evaluation are both defined recursively,
programs can be as complex as you want:

arc> (+ (+ 1 2) (+ 3 (+ 4 5)))

Putting the + before the numbers looks odd when you're used to
writing "1 + 2," but it has the advantage that + can now take any
number of arguments, not just two:

arc> (+)
arc> (+ 1)
arc> (+ 1 2)
arc> (+ 1 2 3)

This turns out to be a convenient property, especially when generating
code, which is a common thing to do in Lisp.

Lisp dialects like Arc have a data type most languages don't:
symbols.  We've already seen one: + is a symbol.  Symbols don't
evaluate to themselves the way numbers and strings do.  They return
whatever value they've been assigned.

If we give foo the value 13, it will return 13 when evaluated:

arc> (= foo 13)
arc> foo

You can turn off evaluation by putting a single quote character
before an expression.  So 'foo returns the symbol foo.

arc> 'foo

Particularly observant readers may be wondering how we got away
with using foo as the first argument to =.  If the arguments are
evaluated left to right, why didn't this cause an error when foo
was evaluated?  There are some operators that violate the usual
evaluation rule, and = is one of them.  Its first argument isn't

If you quote a list, you get back the list itself.  

arc> (+ 1 2)
arc> '(+ 1 2)
(+ 1 2)

The first expression returns the number 3.  The second, because it
was quoted, returns a list consisting of the symbol + and the numbers
1 and 2.

You can build up lists with cons, which returns a list with a new
element on the front:

arc> (cons 'f '(a b))
(f a b)

It doesn't change the original list:

arc> (= x '(a b))
(a b)
arc> (cons 'f x)
(f a b)
arc> x
(a b)

The empty list is represented by the symbol nil, which is defined
to evaluate to itself.  So to make a list of one element you say:

arc> (cons 'a nil)

You can take lists apart with car and cdr, which return the first
element and everything but the first element respectively:

arc> (car '(a b c))
arc> (cdr '(a b c))
(b c)

To create a list with many elements use list, which does a series
of conses:

arc> (list 'a 1 "foo" '(b))
(a 1 "foo" (b))
arc> (cons 'a (cons 1 (cons "foo" (cons '(b) nil))))
(a 1 "foo" (b))

Notice that lists can contain elements of any type.

There are 4 parentheses at the end of that call to cons.  How do
Lisp programmers deal with this?  They don't.  You could add or
subtract a right paren from that expression and most wouldn't notice.
Lisp programmers don't count parens.  They read code by indentation,
not parens, and when writing code they let the editor match parens
(use :set sm in vi, M-x lisp-mode in Emacs).

Like Common Lisp assignment, Arc's = is not just for variables, but
can reach inside structures.  So you can use it to modify lists:

arc> x
(a b)
arc> (= (car x) 'z)
arc> x
(z b)

Lists are useful in exploratory programming because they're so
flexible.  You don't have to commit in advance to exactly what a
list represents.  For example, you can use a list of two numbers
to represent a point on a plane.  Some would think it more proper
to define a point object with two fields, x and y.  But if you use
lists to represent points, then when you expand your program to
deal with n dimensions, all you have to do is make the new code
default to zero for missing coordinates, and any remaining planar
code will continue to work.

Or if you decide to expand in another direction and allow partially
evaluated points, you can start using symbols representing variables
as components of points, and once again, all the existing code will
continue to work.

In exploratory programming, it's as important to avoid premature
specification as premature optimization.

The most exciting thing lists can represent is code.  The lists you
build with cons are the same things programs are made out of.  This
means you can write programs that write programs.  The usual way
to do this is with something called a macro.  We'll get to those
later.  First, functions.

We've already seen some functions: +, cons, car, and cdr.  You can
define new ones with def, which takes a symbol to use as the name,
a list of symbols representing the parameters, and then zero or
more expressions called the body.  When the function is called,
those expressions will be evaluated in order with the symbols in
the body temporarily set ("bound") to the corresponding argument.
Whatever the last expression returns will be returned as the value
of the call.

Here's a function that takes two numbers and returns their average:

arc> (def average (x y)
       (/ (+ x y) 2))
#<procedure: average>
arc> (average 2 4)

The body of the function consists of one expression, (/ (+ x y) 2).
It's common for functions to consist of one expression; in purely
functional code (code with no side-effects) they always do.

Notice that def, like =, doesn't evaluate all its arguments.  It
is another of those operators with its own evaluation rule.

What's the strange-looking object returned as the value of the def
expression?  That's what a function looks like.  In Arc, as in most
Lisps, functions are a data type, just like numbers or strings.

As the literal representation of a string is a series of characters
surrounded by double quotes, the literal representation of a function
is a list consisting of the symbol fn, followed by its parameters,
followed by its body.  So you could represent a function to return
the average of two numbers as:

arc> (fn (x y) (/ (+ x y) 2))

There's nothing semantically special about named functions as there
is in some other languages.  All def does is basically this:

arc> (= average (fn (x y) (/ (+ x y) 2)))
#<procedure: average>

And of course you can use a literal function wherever you could use
a symbol whose value is one, e.g.

arc> ((fn (x y) (/ (+ x y) 2)) 2 4)

This expression has three elements, (fn (x y) (/ (+ x y) 2)), which
yields a function that returns averages, and the numbers 2 and 4.
So when you evaluate all three expressions and pass the values of
the second and third to the value of the first, you pass 2 and 4
to a function that returns averages, and the result is 3.

There's one thing you can't do with functions that you can do with
data types like symbols and strings: you can't print them out in a
way that could be read back in.  The reason is that the function
could be a closure; displaying closures is a tricky problem.

In Arc, data structures can be used wherever functions are, and
they behave as functions from indices to whatever's stored there.
So to get the first element of a string you say:

arc> ("foo" 0)

That return value is what a literal character looks like, incidentally.

Expressions with data structures in functional position also work
as the first argument to =.

arc> (= s "foo")
arc> (= (s 0) #m)
arc> s

There are two commonly used operators for establishing temporary
variables, let and with.  The first is for just one variable.

arc> (let x 1
       (+ x (* x 2)))

To bind multiple variables, use with.

arc> (with (x 3 y 4)
       (sqrt (+ (expt x 2) (expt y 2))))

So far we've only had things printed out implicity as a result of
evaluating them.  The standard way to print things out in the middle
of evaluation is with pr or prn.  They take multiple arguments and
print them in order; prn also prints a newline at the end.  Here's
a variant of average that tells us what its arguments were:

arc> (def average (x y)
       (prn "my arguments were: " (list x y))
       (/ (+ x y) 2))
*** redefining average
#<procedure: average>
arc> (average 100 200)
my arguments were: (100 200)

The standard conditional operator is if.  Like = and def, it doesn't
evaluate all its arguments.  When given three arguments, it evaluates
the first, and if that returns true, it returns the value of the
second, otherwise the value of the third:

arc> (if (odd 1) 'a 'b)
arc> (if (odd 2) 'a 'b)

Returning true means returning anything except nil.  Nil is
conventionally used to represent falsity as well as the empty list.
The symbol t (which like nil evaluates to itself) is often used to
represent truth, but any value other than nil would serve just as

arc> (odd 1)
arc> (odd 2)

It sometimes causes confusion to use the same thing for falsity and
the empty list, but many years of Lisp programming have convinced
me it's a net win, because the empty list is set-theoretic false,
and many Lisp programs think in sets.

If the third argument is missing it defaults to nil.

arc> (if (odd 2) 'a)

An if with more than three arguments is equivalent to a nested if.  

(if a b c d e)

is equivalent to

(if a
    (if c

If you're used to languages with elseif, this pattern will be
familiar.  [1]

Each argument to if is a single expression, so if you want to do
multiple things depending on the result of a test, combine them
into one expression with do.

arc> (do (prn "hello")
         (+ 2 3))

If you just want several expressions to be evaluated when some
condition is true, you could say

(if a
    (do b

but this situation is so common there's a separate operator for it.

(when a

The and and or operators are like conditionals because they don't
evaluate more arguments than they have to.

arc> (and nil
          (pr "you'll never see this"))

The negation operator is called no, a name that also works when
talking about nil as the empty list.  Here's a function to return
the length of a list:

arc> (def mylen (xs)
       (if (no xs)
           (+ 1 (mylen (cdr xs)))))
#<procedure: mylen>

If the list is nil the function will immediately return 0.  Otherwise
it returns 1 more than the length of the cdr of the list.

arc> (mylen nil)
arc> (mylen '(a b))

I called it mylen because there's already a function called len for
this.   You're welcome to redefine Arc functions, but redefining
len this way might break code that depended on it, because len works
on more than lists.

The standard comparison operator is is, which returns true if its
arguments are identical or, if strings, have the same characters. 

arc> (is 'a 'a)
arc> (is "foo" "foo")
arc> (let x (list 'a)
       (is x x))
arc> (is (list 'a) (list 'a))

Note that is returns false for two lists with the same elements.
There's another operator for that, iso (from isomorphic).

arc> (iso (list 'a) (list 'a))

If you want to test whether something is one of several alternatives,
you could say (or (is x y) (is x z) ...), but this situation is
common enough that there's an operator for it.

arc> (let x 'a
       (in x 'a 'b 'c))

The case operator takes alternating keys and expressions and returns
the value of the expression after the key that matches.  You can
supply a final expression as the default.

arc> (def translate (sym)
       (case sym
         apple 'mela
         onion 'cipolla
#<procedure: translate>
arc> (translate 'apple)
arc> (translate 'syzygy)

Arc has a variety of iteration operators.  For a range of numbers,
use for.

arc> (for i 1 10
       (pr i " "))
1 2 3 4 5 6 7 8 9 10 nil

To iterate through the elements of a list or string, use each.

arc> (each x '(a b c d e)
       (pr x " "))
a b c d e nil

Those nils you see at the end each time are not printed out by the
code in the loop.  They're the return values of the iteration

To continue iterating while some condition is true, use while.

arc> (let x 10
       (while (> x 5)
         (= x (- x 1))
         (pr x)))

There's also a more general loop operator that's similar to the C
for operator and tends to be rarely used in practice, and a simple
repeat operator for doing something n times:

arc> (repeat 5 (pr "la "))
la la la la la nil

The map function takes a function and a list and returns the result
of applying the function to successive elements.

arc> (map (fn (x) (+ x 10)) '(1 2 3))
(11 12 13)

Actually it can take any number of sequences, and keeps going till
the shortest runs out:

arc> (map + '(1 2 3 4) '(100 200 300))
(101 202 303)

Since functions of one argument are so often used in Lisp programs,
Arc has a special notation for them.  [... _ ...]  is an abbreviation
for (fn (_) (... _ ...)).  So our first map example could have been

arc> (map [+ _ 10] '(1 2 3))
(11 12 13)

Removing variables is a particularly good way to make programs
shorter.  An unnecessary variable increases the conceptual load of
a program by more than just what it adds to the length.

You can compose functions by putting a colon between the names.
I.e. (foo:bar x y) is equivalent to (foo (bar x y)).  Composed
functions are convenient as arguments.

arc> (map odd:car '((1 2) (4 5) (7 9)))
(t nil t)

You can also negate a function by putting a tilde (~) before the

arc> (map ~odd '(1 2 3 4 5))
(nil t nil t nil)

There are a number of functions like map that apply functions to
successive elements of a sequence.  The most commonly used is keep,
which returns the elements satisfying some test.

arc> (keep odd '(1 2 3 4 5 6 7))
(1 3 5 7)

Others include rem, which does the opposite of keep; all, which
returns true if the function is true of every element; some, which
returns true if the function is true of any element; pos, which
returns the position of the first element for which the function
returns true; and trues, which returns a list of all the non-nil
return values:

arc> (rem odd '(1 2 3 4 5 6))
(2 4 6)
arc> (all odd '(1 3 5 7))
arc> (some even '(1 3 5 7))
arc> (pos even '(1 2 3 4 5))
arc> (trues [if (odd _) (+ _ 10)]
            '(1 2 3 4 5))
(11 13 15)

If functions like this are given a first argument that isn't a
function, it's treated like a function that tests for equality to

arc> (rem 'a '(a b a c u s))
(b c u s)

and they all work on strings as well as lists.

arc> (rem #a "abacus")

Lists can be used to represent a wide variety of data structures,
but if you want to store key/value pairs efficiently, Arc also has
hash tables.

arc> (= airports (table))
arc> (= (airports "Boston") 'bos)

If you want to create a hash table filled with values, you can use
listtab, which takes a list of key/value pairs and returns the
corresponding hash table.

arc> (let h (listtab '((x 1) (y 2)))
       (h 'y))

There's also an abbreviated form where you don't need to group the
arguments or quote the keys.

arc> (let h (obj x 1 y 2)
       (h 'y))

Like lists and strings, hash tables can be used wherever functions are.

arc> (= codes (obj "Boston" 'bos "San Francisco" 'sfo "Paris" 'cdg))
#hash(("Boston" . bos) ("Paris" . cdg) ("San Francisco" . sfo))
arc> (map codes '("Paris" "Boston" "San Francisco"))
(cdg bos sfo)

The function keys returns the keys in a hash table, and vals returns
the values.  

arc> (keys codes)
("Boston" "Paris" "San Francisco")

There is a function called maptable for hash tables that is like
map for lists, except that it returns the original table instead
of a new one.

arc> (maptable (fn (k v) (prn v " " k))
sfo San Francisco
cdg Paris
bos Boston
#hash(("Boston" . bos) ("Paris" . cdg) ("San Francisco" . sfo))

[Note: Like functions, hash tables can't be printed out in a way
that can be read back in.  We hope to fix that though.]

There is a tradition in Lisp going back to McCarthy's 1960 paper [2]
of using lists to represent key/value pairs:

arc> (= codes '(("Boston" bos) ("Paris" cdg) ("San Francisco" sfo)))
(("Boston" bos) ("Paris" cdg) ("San Francisco" sfo))

This is called an association list, or alist for short.  I once
thought alists were just a hack, but there are many things you can
do with them that you can't do with hash tables, including sort
them, build them up incrementally in recursive functions, have
several that share the same tail, and preserve old values.

The function alref returns the first value corresponding to a key
in an alist:

arc> (alref codes "Boston")

There are a couple operators for building strings.  The most general
is string, which takes any number of arguments and mushes them into
a string:

arc> (string 99 " bottles of " 'bee #r)
"99 bottles of beer"

Every argument will appear as it would look if printed out by pr,
except nil, which is ignored.

There's also tostring, which is like do except any output generated
during the evaluation of its body is sent to a string, which is
returned as the value of the whole expression.

arc> (tostring
       (prn "domesday")
       (prn "book"))

You can find the types of things using type, and convert them to
new types using coerce.

arc> (map type (list 'foo 23 23.5 '(a) nil car "foo" #a))
(sym int num cons sym fn string char)
arc> (coerce #A 'int)
arc> (coerce "foo" 'cons)
(#f #o #o)
arc> (coerce "99" 'int)
arc> (coerce "99" 'int 16)

The push and pop operators treat list as stacks, pushing a new
element on the front and popping one off respectively.

arc> (= x '(c a b))
(c a b)
arc> (pop x)
arc> x
(a b)
arc> (push 'f x)
(f a b)
arc> x
(f a b)

Like =, they work within structures, not just on variables.

arc> (push 'l (cdr x))
(l a b)
arc> x
(f l a b)

To increment or decrement use ++ or --:

arc> (let x '(1 2 3)
       (++ (car x))
(2 2 3)

There's also a more general operator called zap that changes something
to the result any function returns when applied to it.  I.e. (++ x)
is equivalent to (zap [+ _ 1] x).

The sort function returns a copy of a sequence sorted according to
the function given as the first argument.

arc> (sort < '(2 9 3 7 5 1))
(1 2 3 5 7 9)

It doesn't change the original, so if you want to sort the value
of a particular variable (or place within a structure), use zap:

arc> (= x '(2 9 3 7 5 1))
(2 9 3 7 5 1)
arc> (zap [sort < _] x)
(1 2 3 5 7 9)
arc> x
(1 2 3 5 7 9)

If you want to modify a sorted list by inserting a new element
at the right place, use insort:

arc> (insort < 4 x)
(1 2 3 4 5 7 9)
arc> x
(1 2 3 4 5 7 9)

In practice the things one needs to sort are rarely just lists of
numbers.  More often you'll need to sort things according to some
property other than their value, e.g.

arc> (sort (fn (x y) (< (len x) (len y)))
           '("orange" "pea" "apricot" "apple"))
("pea" "apple" "orange" "apricot")

Arc's sort is stable, meaning the relative positions of elements
judged equal by the comparison function won't change:

arc> (sort (fn (x y) (< (len x) (len y)))
           '("aa" "bb" "cc"))
("aa" "bb" "cc")

Since comparison functions other than > or < are so often needed,
Arc has a compare function to build them:

arc> (sort (compare < len)
           '("orange" "pea" "apricot" "apple"))
("pea" "apple" "orange" "apricot")

We've seen several functions so far that take optional arguments
or varying numbers of arguments.  To make a parameter optional,
just say (o x) instead of x. Optional parameters default to nil.

arc> (def greet (name (o punc))
       (string "hello " name punc))
#<procedure: greet>
arc> (greet 'joe)
"hello joe"
arc> (greet 'joe #!)
"hello joe!"

Functions can have as many optional parameters as you want, but
they have to come at the end of the parameter list.  

If you put an expression after the name of an optional parameter,
it will be evaluated if necessary to produce a default value.  The
expression can refer to preceding parameters.

arc> (def greet (name (o punc (case name who #? #!)))
       (string "hello " name punc))
*** redefining greet
#<procedure: greet>
arc> (greet 'who)
"hello who?"

To make a function that takes any number of arguments, put a period
and a space before the last parameter, and it will get bound to a
list of the values of all the remaining arguments:

arc> (def foo (x y . z)
       (list x y z))
#<procedure: foo>
arc> (foo (+ 1 2) (+ 3 4) (+ 5 6) (+ 7 8))
(3 7 (11 15))

This type of parameter is called a "rest parameter" because it gets
the rest of the arguments.  If you want all the arguments to a
function to be collected in one parameter, just use it in place of
the whole parameter list.

(These conventions are not as random as they seem.  The parameter
list mirrors the form of the arguments, and a list terminated by
something other than nil is represented as e.g. (a b . c).)

To supply a list of arguments to a function, use apply:

arc> (apply + '(1 2 3))

Now that we have rest parameters and apply, we can write a version
of average that takes any number of arguments.

arc> (def average args
       (/ (apply + args) (len args)))
#<procedure: average>
arc> (average 1 2 3)

We know enough now to start writing macros.  Macros are basically
functions that generate code.  Of course, generating code is easy;
just call list.

arc> (list '+ 1 2)
(+ 1 2)

What macros offer is a way of getting code generated this way into
your programs.  Here's a (rather stupid) macro definition:

arc> (mac foo ()
       (list '+ 1 2))
*** redefining foo
#3(tagged mac #<procedure>)

Notice that a macro definition looks exactly like a function
definition, but with def replaced by mac.  

What this macro says is that whenever the expression (foo) occurs
in your code, it shouldn't be evaluated in the normal way like a
function call.  Instead it should be replaced by the result of
evaluating the body of the macro definition, (list '+ 1 2).
This is called the "expansion" of the macro call.

In other words, if you've defined foo as above, putting (foo)
anywhere in your code is equivalent to putting (+ 1 2) there.

arc> (+ 10 (foo))

This is a rather useless macro, because it doesn't take any arguments.
Here's a more useful one:

arc> (mac when (test . body)
       (list 'if test (cons 'do body)))
*** redefining when
#3(tagged mac #<procedure>)

We've just redefined the built-in when operator.  That would
ordinarily be an alarming idea, but fortunately the definition we
supplied is the same as the one it already had.

arc> (when 1
       (pr "hello ")
hello 2

What the definition above says is that when you have to evaluate
an expression whose first element is when, replace it by the result
of applying

(fn (test . body)
  (list 'if test (cons 'do body)))

to the arguments.  Let's try it by hand and see what we get.

arc> (apply (fn (test . body)
              (list 'if test (cons 'do body)))
            '(1 (pr "hello ") 2))
(if 1 (do (pr "hello ") 2))

So when Arc has to evaluate 

(when 1
  (pr "hello ")

the macro we defined transforms that into 

(if 1
    (do (pr "hello ")

first, and when that in turn is evaluated, it produces the behavior
we saw above.

Building up expressions using calls to list and cons can get unwieldy,
so most Lisp dialects have an abbreviation called backquote that
makes generating lists easier.

If you put a single open-quote character (`) before an expression,
it turns off evaluation just like the ordinary quote (') does,

arc> `(a b c)
(a b c)

except that if you put a comma before an expression within the list,
evaluation gets turned back on for that expression.

arc> (let x 2
       `(a ,x c))
(a 2 c)

A backquoted expression is like a quoted expression with holes in it.

You can also put a comma-at (,@) in front of anything within a
backquoted expression, and in that case its value (which must be a
list) will get spliced into whatever list you're currently in.

arc> (let x '(1 2)
       `(a ,@x c))
(a 1 2 c)

With backquote we can make the definition of when more readable.

(mac when (test . body)
  `(if ,test (do ,@body)))

In fact, this is the definition of when in the Arc source.

One of the keys to understanding macros is to remember that macro
calls aren't function calls.  Macro calls look like function calls.
Macro definitions even look a lot like function definitions.  But
something fundamentally different is happening.  You're transforming
code, not evaluating it.  Macros live in the land of the names, not
the land of the things they refer to.

For example, consider this definition of repeat:

arc> (mac repeat (n . body)
       `(for x 1 ,n ,@body))
#3(tagged mac #<procedure>)

Looks like it works, right?

arc> (repeat 3 (pr "blub "))
blub blub blub nil

But if you use it in certain contexts, strange things happen.

arc> (let x "blub "
       (repeat 3 (pr x)))

We can see what's going wrong if we look at the expansion.  The
code above is equivalent to

(let x "blub "
  (for x 1 3 (pr x)))

Now the bug is obvious.  The macro uses the variable x to hold the
count while iterating, and that gets in the way of the x we're
trying to print.

Some people worry unduly about this kind of bug.  It caused the
Scheme committee to adopt a plan for "hygienic" macros that was
probably a mistake.  It seems to me that the solution is not to
encourage the noob illusion that macro calls are function calls.
People writing macros need to remember that macros live in the land
of names.  Naturally in the land of names you have to worry about
using the wrong names, just as in the land of values you have to
remember not to use the wrong values-- for example, not to use zero
as a divisor.

The way to fix repeat is to use a symbol that couldn't occur in
source code instead of x.  In Arc you can get one by calling the
function uniq.  So the correct definition of repeat (and in fact
the one in the Arc source) is

(mac repeat (n . body)
  `(for ,(uniq) 1 ,n ,@body))

If you need one or more uniqs for use in a macro, you can use w/uniq,
which takes either a variable or list of variables you want bound to
uniqs.  Here's the definition of a variant of do called do1 that's
like do but returns the value of its first argument instead of the
last (useful if you want to print a message after something happens,
but return the something, not the message):

(mac do1 args
  (w/uniq g
    `(let ,g ,(car args)
       ,@(cdr args)

Sometimes you actually want to "capture" variables, as it's called,
in macro definitions.  The following variant of when, which binds the
variable it to the value of the test, turns out to be very useful:

(mac awhen (expr . body)
  `(let it ,expr (if it (do ,@body))))

In a sense, you now know all about macros-- in the same sense that,
if you know the axioms in Euclid, you know all the theorems.  A lot
follows from these simple ideas, and it can take years to explore
the territory they define.  At least, it took me years.  But it's
a path worth following.  Because macro calls can expand into further
macro calls, you can generate massively complex expressions with
them-- code you would have had to write by hand otherwise.  And yet
programs built up out of layers of macros turn out to be very
manageable.  I wouldn't be surprised if some parts of my code go
through 10 or 20 levels of macroexpansion before the compiler sees
them, but I don't know, because I've never had to look.

One of the things you'll discover as you learn more about macros
is how much day-to-day coding in other languages consists of manually
generating macroexpansions.  Conversely, one of the most important
elements of learning to think like a Lisp programmer is to cultivate
a dissatisfaction with repetitive code.  When there are patterns
in source code, the response should not be to enshrine them in a
list of "best practices," or to find an IDE that can generate them.
Patterns in your code mean you're doing something wrong.  You should
write the macro that will generate them and call that instead.

Now that you've learned the basics of Arc programming, the best way
to learn more about the language is to try writing some programs
in it.  Here's how to write the hello-world of web apps:

arc> (defop hello req (pr "hello world"))
arc> (asv)
ready to serve port 8080

If you now go to http://localhost:8080/hello your new web app will
be waiting for you.  

Here are a couple slightly more complex hellos that hint at the
convenience of macros that store closures on the server:

(defop hello2 req
  (w/link (pr "there")
    (pr "here")))

(defop hello3 req
  (w/link (w/link (pr "end")
            (pr "middle"))
    (pr "start")))

(defop hello4 req
  (aform [w/link (pr "you said: " (arg _ "foo"))
           (pr "click here")]
    (input "foo")

See the sample application in blog.arc for ideas about how to make
web apps that do more.

We now know enough Arc to read the definitions of some of the
predefined functions.  Here are a few of the simpler ones.

(def cadr (xs)
  (car (cdr xs)))

(def no (x)
  (is x nil))

(def list args

(def isa (x y)
  (is (type x) y))

(def firstn (n xs)
  (if (and (> n 0) xs)
      (cons (car xs) (firstn (- n 1) (cdr xs)))

(def nthcdr (n xs)
  (if (> n 0)
      (nthcdr (- n 1) (cdr xs))

(def tuples (xs (o n 2))
  (if (no xs)
      (cons (firstn n xs)
            (tuples (nthcdr n xs) n))))

(def trues (f seq)
  (rem nil (map f seq)))

(mac unless (test . body)
  `(if (no ,test) (do ,@body)))

(mac n-of (n expr)
  (w/uniq ga
    `(let ,ga nil
       (repeat ,n (push ,expr ,ga))
       (rev ,ga))))

These definitions are taken from arc.arc.  As its name suggests,
reading that file is a good way to learn more about both Arc and
Arc programming techniques.  Nothing in it is used before it's
defined; it is an exercise in building the part of the language
written in Arc up from the "axioms" defined in ac.scm. I hoped this
would yield a simple language.  But since this is also the source
code of Arc, I've tried to balance simplicity with efficiency.  The
definitions aren't mathematically minimal if that would be insanely
inefficient; I tried that once, and they were.

The definitions in arc.arc are also an experiment in another way.
They are the language spec.  The spec for isa isn't prose, like
function specs in Common Lisp.  This is the spec for isa:

(def isa (x y)
  (is (type x) y))

It may sound rather dubious to say that the only spec for something
is its implementation.  It sounds like the sort of thing one might
say about C++, or the Common Lisp loop macro.  But that's also how
math works.  If the implementation is sufficiently abstract, it
starts to be a good idea to make specification and implementation

I agree with Abelson and Sussman that programs should be written
primarily for people to read rather than machines to execute.  The
Lisp defined as a model of computation in McCarthy's original paper
was.  It seems worth trying to preserve this as you grow Lisp into
a language for everyday use.


[1] Note to Lisp hackers: If you're used to the conventional Lisp
cond operator, this if amounts to the same thing, but with fewer
parentheses.  E.g.

(cond (a b)
      (c d)
      (t e))


(if a b
    c d

JMC's original cond didn't have implicit progn, so the parens around
each pair of clauses were unnecessary.  They became necessary soon
after, however, when cond started to have implicit progn in the
first Lisp implementations.  This probably prevented people from
realizing they hadn't originally been needed.  But most conds in
the wild seem to occur in purely functional code, and thus pay the
cost in parens of implicit progn without actually needing it.  My
experience so far suggests it's a net win to offer progn a la carte
instead of combining it with the default conditional operator.
Having to use explicit dos may even be an advantage, because it
calls attention to nonfunctional code.

[2] "Recursive Functions of Symbolic Expressions and Their Computation
by Machine, Part I," CACM, April 1960.