Harry R. Schwartz

Software engineer, nominal scientist, gentleman of the internet.

Lives in Brooklyn with one girlfriend and nine six :( houseplants.

Organizes the New York Emacs Meetup and Ruby Project Night NYC. Writes code for fun and thoughtbot.


An Introduction to Emacs Lisp

08 Apr 2014

This post is derived from a talk I gave at the New York Emacs Meetup. A video of that talk is also available.


What we’ll be covering

We’ll go over the basic features of Emacs Lisp, including an introduction to higher-order functions, but not including macros.

What we won’t be covering

Most of the Emacs standard library. This tutorial covers the features of the language, but it doesn’t address how we manipulate buffers, regions, and so on.

We’ll also be making passing references to macros, but we won’t be exploring them in any detail.

Introduction

Emacs can be thought of as a giant REPL. Like bash, python, irb, and other REPLs, it Reads in an expression, Evaluates it, Prints the results, and Loops to read the next expression.

In Emacs’ case, the REPL is started when we launch the Emacs application. Every time we interact with the editor, we’re just executing some code in the context of that REPL. Every keystroke, every mouse click, and every menu action correspond to evaluating an expression in this REPL.

Similarly, your .emacs file is just the first code that’s executed by that REPL after it starts.

Don’t worry if this doesn’t make too much sense now; you’ll get it soon. =)

To execute this code as we go along, you’ll first want to open up a new buffer with C-x b and name it whatever you’d like. Run M-x emacs-lisp-mode to put it into the correct mode.

Paste code snippets into this buffer, place your cursor (called “point” in Emacs lingo) right after the expression, and hit C-x C-e. That’ll evaluate the expression in the Emacs REPL. The results of the expression will appear in the echo area in the bottom-left corner of your window.

Atoms

Atoms are the simplest objects in Emacs Lisp. They evaluate to themselves, so evaluating these objects will just return themselves.

We have integers:

42

Floats:

3.0

And strings:

"foo!"

Functions

Lisp expressions are either atoms or applications of functions (or macro expressions, technically, but that’s outside the scope of this article).

To call a function, we wrap it all in parentheses, like so:

(+ 1 2 3)

That calls the function + on the arguments 1, 2, and 3, evaluating to 6.

When evaluating a function, Lisp first evaluates the arguments to the function, the applies the function to those arguments. This means that the arguments to a function can themselves be function calls:

(+ (* 2 3)
   (/ 8 4))

In this example, the interpreter first executes (* 2 3) and (/ 8 4), which evaluate to 6 and 2 respectively. These values become the arguments to +, so the entire expression evaluates to 8.

Lists

Lisp is really interested in lists; it’s the LISt Processing language, after all.

Let’s try to evaluate a list:

(1 2 3)

When we execute that, we get the error: Invalid function: 1. That’s reasonable: the interpreter is trying to execute the function 1 on the arguments 2 and 3, but of course 1 isn’t really a function.

If we want to refer to a list without trying to evaluate it, we can wrap it in a call to the quote function. In the same way that quoting a word in English refers to the word and not the thing it represents, quoting an expression in Lisp refers to the list itself. To say that another way, quoting an expression treats it as data, not as code.

(quote (1 2 3))

And boom! We get (1 2 3), just like we wanted.

It turns out that quoting is something we do pretty often in Lisp, so there’s a bit of handy syntax to make it easier.

'(1 2 3)

'(1 2 3) evaluates to (1 2 3), too, and we can generally use ' and quote interchangeably. We’ll be using ' from here on out, because it’s terser.

We can quote nested lists, too:

'(1 2 (3 4 5) (6 (7)))
; => (1 2 (3 4 5) (6 (7)))

We can also quote a list that looks like a function call:

'(+ 1 2 3)
; => (+ 1 2 3)

This just evaluates to the four-element list (+ 1 2 3). Notice that nothing is evaluated inside the quoted expression.

We can also create a list with the list function:

(list 1 2 3)
; => (1 2 3)

Now that we know how to refer to whole lists, we can start manipulating them. The simplest function is car, which just returns the first element of the list.

(car '(1 2 3))
; => 1

Conversely, cdr returns the list minus the first element.

(cdr '(1 2 3))
; => (2 3)

Why are these functions called car and cdr? It’s just an historical accident, but you can read the whole story here. They’re basically vestigial. We’d probably name these functions first and rest if we were rewriting Emacs from scratch today, but because they’re already used so widely, it’s just something you’ll need to memorize. Sorry. =(

Let’s try evaluating the empty list.

'()

Bizarrely, this evaluates to nil! In Lisp, nil and the empty list are equivalent. They’re literally the same thing. You can append objects to nil to create lists, and you can test for the empty list in conditionals (where it’s falsy).

The empty list evaluates to itself:

()

As does nil:

nil

When we take the cdr of a one-element list, we get the empty list, nil:

(cdr '(42))

The function null determines if a value is nil (or ()). This is especially useful in recursive functions, where we often use the empty list as a base case.

(null nil)
(null '())

We’ve seen how to decompose lists with car and cdr. The reverse operation, building up a list, is accomplished with cons.

cons (think “construct”) takes an expression and a list. It returns a new list whose car is the result of the expression and whose cdr is the list.

(cons 1 '(2 3))

By consing 1 onto '(2 3) we created the list '(1 2 3).

consing a value onto an empty list creates a list containing that value:

(cons "foo" '())
; => ("foo")

We can chain these cons calls together to build up a list from scratch:

(cons 1 (cons 2 (cons 3 '())))

Under the hood, the list function is just a series of cons calls.

This technique is very common when designing recursive functions. We often recurse through a list using car and cdr and/or build up a result with cons.

There’s also a handy append function for joining two lists to create a new one:

(append '(1 2) '(3 4))

This just creates a new list, (1 2 3 4).

A fun exercise (after you’ve read the next couple sections) might be implementing your own version of append using car, cdr, and cons.

Variables

Like virtually every other language, Emacs Lisp has variables. Emacs has many variables that are already defined and in use; for example, path-separator contains a string that separates file paths in the system’s $PATH variable.

path-separator

In my system, evaluating path-separator returns ":".

Trying to evaluate an undefined variable raises an error:

some-list

I got the error Symbol's value as variable is void: some-list. That just means that the symbol some-list doesn’t point to a variable. Let’s fix that.

(set 'some-list '(1 2 3))

We assign variables with set. set takes the name of a variable (quoted, so it’s not evaluated) and a value, and sets the variable to that value. So, if we try to evaluate some-list again:

some-list
; => '(1 2 3)

Unsurprisingly, we set variables fairly often. A long time ago, Lisp programmers decided that having to quote the name of the variable was a bit of a hassle, so they added setq as an alias. setq just implicitly quotes the first variable, so it’s a bit more convenient. In practice, you’ll rarely see set; most folks favor setq, so I’ll be using that exclusively from here on.

(setq my-list '(foo bar baz))
my-list
; => (foo bar baz)

“But Harry,” you might justly complain, “I thought you said that the arguments to the function would be evaluated before the function was executed! How does setq automatically quote that variable name? It’s just a function, right?”

The answer is that setq actually isn’t just a function. setq is an example of a macro. Macros are a bit like functions, but they also make it possible to manipulate their arguments before they get executed—in this case, the setq macro implicitly wraps a quote around the first argument. Macros are a big topic that’s way outside the scope of this tutorial, but I think they’re just awesome, so I’d highly recommend checking them out when you get a chance. They’re one of the things that make Lisp really fun to work with.

Now, back to variables.

setq defines a variable globally. Defining all of our variables globally is shoddy practice, so we also have locally-scoped variables. These are defined with a let expression. let also does some delayed-evaluation trickery. Its first argument is a list of two-element pairs. Each pair contains a variable name and its initial value.

Every subsequent argument to let is evaluated after those variables have been defined.

(let ((a 1)
      (b 5))
  (+ a b))

Evaluating the expression above should yield 6. a was bound to 1, b was bound to 5, and we summed them. The return value of a let expression is the return value of the last expression in it (in this case, (+ a b)).

To prove to ourselves that a and b were only defined in the scope of the let expression, let’s try evaluating a:

a

Phew, Symbol's value as a variable is void: a. So that’s good; a isn’t defined any more.

Sometimes we want to define a few variables in a let expression, and have the value of one depend on the value of another. let doesn’t allow this (you could think of the values in a let expression as being bound in parallel), but a variant, let*, does.

(let* ((a 3)
       (b (+ a 5)))
  (+ a b))

This should return 11, since a was bound to 3 and b was bound to 8.

Why wouldn’t we always just use let*? Heck, given that we have let*, why have let at all?

There are some purity arguments – the implementation of let is quite simple, but let*’s is significantly more complex – but that’s not really a compelling reason unless you’re a language designer.

When possible, you should use let because it’s intention-revealing. Other programmers will be able to read through your code and immediately know that there are no dependencies between the variable bindings. That makes reading code a bit easier, and other programmers reading your code (including yourself, in a few months) will appreciate it.

Defining functions

Okay, so. We can evaluate atoms, call functions, build and decompose lists, and bind variables globally or locally. We’re making some good progress.

Here’s the simplest function I can think of.

(defun say-hello ()
  "hello!")

In this expression, defun defines a function called say-hello. The empty list indicates that it takes no arguments. Every subsequent argument to defun will be executed sequentially when the function is called, and the return value of the last expression will be the return value of the function. In this case, the body of the function is just the string "hello!".

Sound good? Let’s call this thing.

(say-hello)

Sweet, it said "hello!" Why hello, function! :D

Next, let’s try a function that takes an argument. How about squareing a number?

(defun square (x)
  (* x x))

(square 2)

And behold, 4. Lookin’ fine.

We might wonder, “What happens if we send something non-numeric to our function?” Well,

(square "uh-oh")

We get the error Wrong type argument: number-or-marker-p, "uh-oh". This is Lisp’s way of saying that some function (*, in this case) can only operate on numbers.

Lisp is a dynamic, interpreted language. It doesn’t have a strong static typing system like Haskell (or even Java!), so there’s no way for it to detect problems like this in advance. This actually gives us a lot of power, and it’s often terribly convenient, but it does mean that we can shoot ourselves in the foot sometimes.

(Languages like Ruby and Python often overcome these pitfalls with solid suites of unit tests. Lisp is just as testable as those languages, but it doesn’t have a strong testing culture. I’ll save that rant for another day.)

Let’s look at a more complex example. In 2D space, to determine the distance between two points we use the Euclidean distance metric (usually formulated as “the square root of the sum of the squares”). Suppose we have two points, (x1, y1) and (x2, y2). Let’s implement a function to compute the distance between them:

(defun distance (x1 y1 x2 y2)
  (sqrt (+ (square (- x2 x1))
           (square (- y2 y1)))))

Here we’ve defined a function that takes four arguments and returns the square root of the sum of the squares. Note that we used our previously-defined square function as well as the built-in sqrt.

(distance 3 0 0 4)

Lo and behold, the distance between (3, 0) and (0, 4) is 5.0, as we’d expect.

Conditionals

We’ll be looking at three types of conditionals, from the simplest to the most general. First, though, we should talk about Boolean expressions.

In Emacs Lisp, every value is truthy except nil and the empty list (). Notably, 0 and "" and also both truthy. If we want to speak directly about truth, we use t, the equivalent of true or True in other languages. There’s no equivalent false value in Emacs Lisp; we generally use nil instead.

In summary: use t (or anything else, really) for truth, and nil for falsity.

The functions for and, or, and not are, reasonably enough, and, or, and not. No big shocks here. and and or can both take any number of arguments.

(and t "" 0 7)
; => 7

(or nil "foo" '() "bar")
; => "foo"

(not nil)
; => t

Note in the above examples that and and or return the first argument that satisfies them. or will return the first truthy value (or nil, if nothing’s truthy), and and will return the last argument, if they’re all truthy (and nil otherwise). You may be tempted to use these return values to write very terse code, but I’d advise you to resist that temptation. Overly clever code will only be harder for you to understand later.

Some functions – = or null, for example – just return t or nil. Lisp programmers refer to such functions as predicates. Predicates are usually (though not always) distinguished by being suffixed by “-p”. For example, when we tried to square a string in the last section, the interpreter claimed that our string didn’t satisfy number-or-marker-p, which we can now deduce was a predicate that tested the input of the * function.

Now that we can write our own predicates, we can get to conditionals.

If we want to take an action only if a predicate is true, and otherwise do nothing, we can use a when expression.

(when (= (+ 2 2) 4)
  "passed sanity check!")

Pfew, 2 + 2 = 4 and the world is sane.

Personally, I don’t use when all that much, but it shows up occasionally in my .emacs configuration. For example, there are certain things I only want to do if Emacs is running as a stand-alone application (not in a terminal), and I find that when is a nice fit for that.

More usefully, if evaluates a predicate, then evaluates one expression if it was true, and the other if it was false. Let’s wrap an if in a function to try it out.

(defun evens-or-odds (n)
  (if (= 0 (% n 2))
      "even!"
    "odd!"))

(evens-or-odds 4)
(evens-or-odds 3)

We’re probably unsurprised to learn that 4 is "even!" and 3 is "odd!".

if is great, but it’d be nice if we had an extended if/elseif/…/elseif/else construct like we do in other languages. And, of course, we do! cond is a generalization of if that can match an arbitrary number of cases. It takes a collection of lists, each of which starts with a Boolean expression. It runs through each such expression searching for one that matches, then evaluates the remaining elements in that list and returns the result. This makes more sense as an example:

(defun pick-a-word (n)
  (cond
   ((= n 1) "bulbous")
   ((= n 2) "bouffant")
   ((= n 3) "beluga")
   (t "gazebo!")))

(pick-a-word 2)
; => "bouffant"

(pick-a-word -72)
; => "gazebo!"

Notice that our last condition is just t. Since t is always truthy, we can use it to indicate an else or otherwise case. If none of the other expressions are truthy, t definitely will be.

Recursion

Recursion is the process of calling a function that calls itself. It’s pretty trippy, but it’s extraordinarily useful and not too difficult to use once you get the hang of it.

A recursive function generally contains a conditional. One branch of the conditional involves somehow simplifying the problem – perhaps taking an element off a list or decrementing a number – then calling the function itself with that simplified quantity and using the results. The other branch is the base case, which kicks in when the thing being investigated is at its simplest – perhaps the list is empty, or the number had been reduced to 0.

I’m not going to go into any great depth with recursion, because it’s a technique with some complex implications. However, I think you should see a simple recursive function in Lisp. An implementation of the factorial function is the canonical example; I don’t dare buck tradition by demonstrating anything else.

(defun factorial (n)
  (if (< n 1)
      1
    (* n (factorial (- n 1)))))

(factorial 5)
; => 120

In this case, factorial calls itself repeatedly until the base case kicks in when n hits 0. The stack of function calls unwinds, eventually returning the product (* 5 (* 4 (* 3 (* 2 (* 1 1))))), which evaluates to 120.

Anonymous functions

"foo" is a string literal. 42 is an integer literal. In the same way, anonymous functions are function literals.

Variables point to literal values. When we (setq foo "bar"), we’re pointing the symbol foo to the literal value "bar". Likewise, when we (defun baz () "quuz"), we’re pointing the symbol baz to a function. Without referencing baz, how could we refer to the function that baz points to?

Lisp uses a macro called lambda to create function literals. The name is a reference to the lambda calculus, which you emphatically do not have to understand to wield Lisp perfectly effectively.

A note on vocabulary: most programmers use the terms function literal, anonymous function, and lambda or lambda function more or less interchangeably. Unless you’re a researcher working in programming language theory they mean essentially the same thing.

Okay, let’s define us a lambda:

(lambda (x) (* x x x))

Painless! This is a function that takes an argument and cubes it. Simple enough!

However, we notice that it didn’t really do anything. To execute a lambda, just use it like you’d use a function:

((lambda (x) (* x x x)) 5)

Boom, 125. We created a lambda which was immediately evaluated and applied to 5.

We can also call a lambda with funcall:

(funcall (lambda (x) (* x x x)) 3)
; => 27

funcall takes a function and (optionally) some arguments, and calls the function on those arguments. The function doesn’t even have to be anonymous.

(funcall '+ 1 2)
; => 3

This is handy when the function’s stored in a variable; we’ll see that show up a little later.

So here’s a cool trick. Since lambdas are just function literals, and since we can bind variable names to literals, we can replicate the functionality of defun on our own!

(fset 'cube (lambda (x) (* x x x)))

(cube 4)

64! Pretty awesome.

The first time I tried to do this I did something like (setq cube ...). That doesn’t work. Emacs Lisp keeps variables and functions in different namespaces (Emacs Lisp is a Lisp-2), so we have to use fset instead.

And no, there’s isn’t a fsetq :)

Higher-order functions

As we’ve seen, we can pretty much treat functions as just another kind of variable. There’s no reason that we can’t pass them in as arguments to other functions. Functions that take other functions as arguments are commonly called higher-order functions, but otherwise there’s nothing special about them. We can easily write our own.

(defun transform-unless-zero (fn n)
  (if (= n 0)
      0
    (funcall fn n)))

So, if n is 0, we just return 0, but otherwise we apply the function fn to n and return the result. Cool?

(transform-unless-zero (lambda (n) (+ 1 n)) 0)
; => 0

(transform-unless-zero (lambda (n) (+ 1 n)) 7)
; => 12

Higher-order functions aren’t terribly scary, and there’re really useful.

In fact, they’re so useful that they show up all over the place in Lisp. One of the most common is mapcar. It takes a function and a list, applies the function to each element in the list, and returns a new list of results.

(mapcar (lambda (n) (+ 1 n)) '(1 2 3 4))
; => (2 3 4 5)

Pretty cool! Note that mapcar doesn’t change the original list. If it’d been stored in a variable, that variable’s value wouldn’t have changed. So, for example:

(setq our-list '(1 2 3 4))
(mapcar (lambda (n) (+ 1 n)) '(1 2 3 4))
; => (2 3 4 5)

our-list
; => (1 2 3 4)

If the mapping had changed the variable, we’d have called it a destructive operation. We like to avoid destructive operations in Lisp (and many other languages), because they introduce side-effects that can make our code harder to reason about. (Incidentally, a function that operates deterministically with no side-effects is called referentially transparent. Impress your friends!)

We can also refer to functions by name, of course.

(mapcar 'upcase '("foo" "bar" "baz"))
; => ("FOO" "BAR" "BAZ")

It’s worth noting that applying mapcar to an empty list will always just return an empty list.

(mapcar 'upcase '())
; => ()

remove-if-not is another handy higher-order function. It takes a predicate and a list. It returns a new list that only contains those items that satisfy the predicate.

(remove-if-not 'oddp '(0 1 2 3 4 5 6 7 8 9))
; => (1 3 5 7 9)

It’s also not destructive, and it maintains the order of the items in the list.

Incidentally, if you find the name remove-if-not a bit confusing because it contains a double negative, you’re not the only one. I still have to consciously think it through sometimes. Other languages often refer to this function as keep-if, filter, or select, all of which seem a bit clearer to me.

An example

Okay, we’re in the home stretch. We’ve covered a lot of ground, so let’s pull it all together into a big, fancy example. We’re going to implement quicksort, a very clever recursive algorithm for efficiently sorting a list. Here’s how the algorithm works:

And here’s how we’d implement it in Emacs Lisp:

(defun qs (items)
  (if (null items)
      '()
    (let* ((pivot (car items))
           (rest (cdr items))
           (lesser (remove-if-not
                    (lambda (x) (<= x pivot)) rest))
           (greater (remove-if-not
                     (lambda (x) (> x pivot)) rest)))
      (append (qs lesser) (list pivot) (qs greater)))))

This might look a little intimidating, but there’s actually nothing new here. We’re defining a function called qs that takes the list items. If items is empty, it returns an empty list, which is the base case. Otherwise it defines a pivot, uses remove-if-not to create the lesser and greater lists, recursively calls qs on them, and joins the results together. Not really that bad, is it?

(qs '(3 5 7 8 4 2 5 7 0 8 4 6))
; => (0 2 3 4 4 5 5 6 7 7 8 8)

Will ya look at that! It even works!

In real life, of course, we’d probably use the built-in sort function. There’s no reason to build our own, and the built-in one is almost certainly more efficient.

Next steps

We just went from atoms to quicksort in under five thousand words. Not too shabby!

If you’d like to keep going with Emacs Lisp, there are a few resources I’d like to recommend:

I’ll also be writing a couple short blog posts in the next couple weeks about setting keybindings and using the built-in documentation. Keep an eye out! =)