The Common Lisp Cookbook – Error and exception handling

Table of Contents

The Common Lisp Cookbook – Error and exception handling

📢 New videos: web dev demo part 1, dynamic page with HTMX, Weblocks demo

📕 Get the EPUB and PDF

Common Lisp has mechanisms for error and condition handling as found in other languages, and can do more.

What is a condition ?

Just like in languages that support exception handling (Java, C++, Python, etc.), a condition represents, for the most part, an “exceptional” situation. However, even more so than those languages, a condition in Common Lisp can represent a general situation where some branching in program logic needs to take place, not necessarily due to some error condition. Due to the highly interactive nature of Lisp development (the Lisp image in conjunction with the REPL), this makes perfect sense in a language like Lisp rather than say, a language like Java or even Python, which has a very primitive REPL. In most cases, however, we may not need (or even allow) the interactivity that this system offers us. Thankfully, the same system works just as well even in non-interactive mode.


Let’s dive into it step by step. More resources are given afterwards.

Ignoring all errors, returning nil

Sometimes you know that a function can fail and you just want to ignore it: use ignore-errors:

  (/ 3 0))
; in: IGNORE-ERRORS (/ 3 0)
;     (/ 3 0)
;   Lisp error during constant folding:
;   arithmetic error DIVISION-BY-ZERO signalled
;   Operation was (/ 3 0).
; compilation unit finished
;   caught 1 STYLE-WARNING condition

We get a welcome division-by-zero warning but the code runs well and it returns two things: nil and the condition that was signaled. We could not choose what to return.

Remember that we can inspect the condition with a right click in Slime.

Catching any condition (handler-case)

ignore-errors is built from handler-case. We can write the previous example by catching the general error but now we can return whatever we want:

(handler-case (/ 3 0)
  (error (c)
    (format t "We caught a condition.~&")
    (values 0 c)))
; in: HANDLER-CASE (/ 3 0)
;     (/ 3 0)
;   Lisp error during constant folding:
;   Condition DIVISION-BY-ZERO was signalled.
; compilation unit finished
;   caught 1 STYLE-WARNING condition
We caught a condition.
#<DIVISION-BY-ZERO {1004846AE3}>

We also returned two values, 0 and the signaled condition.

The general form of handler-case is

(handler-case (code that errors out)
   (condition-type (the-condition) ;; <-- optional argument
   (another-condition (the-condition)

Catching a specific condition

We can specify what condition to handle:

(handler-case (/ 3 0)
  (division-by-zero (c)
    (format t "Caught division by zero: ~a~%" c)))
;; …
;; Caught division by zero: arithmetic error DIVISION-BY-ZERO signalled
;; Operation was (/ 3 0).
;; NIL

This workflow is similar to a try/catch as found in other languages, but we can do more.

handler-case VS handler-bind

handler-case is similar to the try/catch forms that we find in other languages.

handler-bind (see the next examples), is what to use when we need absolute control over what happens when a signal is raised. It allows us to use the debugger and restarts, either interactively or programmatically.

If some library doesn’t catch all conditions and lets some bubble out to us, we can see the restarts (established by restart-case) anywhere deep in the stack, including restarts established by other libraries that this library called. And we can see the stack trace, with every frame that was called and, in some lisps, even see local variables and such. Once we handler-case, we “forget” about this, everything is unwound. handler-bind does not rewind the stack.

Before we properly see handler-bind, let’s study conditions and restarts.

Defining and making conditions

We define conditions with define-condition and we make (initialize) them with make-condition.

(define-condition my-division-by-zero (error)

(make-condition 'my-division-by-zero)
;; #<MY-DIVISION-BY-ZERO {1005A5FE43}>

It’s better if we give more information to it when we create a condition, so let’s use slots:

(define-condition my-division-by-zero (error)
  ((dividend :initarg :dividend
             :initform nil
             :reader dividend)) ;; <-- we'll get the dividend with (dividend condition). See the CLOS tutorial if needed.
  (:documentation "Custom error when we encounter a division by zero.")) ;; good practice ;)

Now when we’ll “signal” or “throw” the condition in our code we’ll be able to populate it with information to be consumed later:

(make-condition 'my-division-by-zero :dividend 3)
;; #<MY-DIVISION-BY-ZERO {1005C18653}>

Note: here's a quick reminder on classes, if you are not fully operational on the Common Lisp Object System.

(make-condition 'my-division-by-zero :dividend 3)
;;                                   ^^ this is the ":initarg"

and :reader dividend created a generic function that is a “getter” for the dividend of a my-division-by-zero object:

(make-condition 'my-division-by-zero :dividend 3)
;; #<MY-DIVISION-BY-ZERO {1005C18653}>
(dividend *)
;; 3

an “:accessor” would be both a getter and a setter.

So, the general form of define-condition looks and feels like a regular class definition, but despite the similarities, conditions are not standard objects.

A difference is that we can’t use slot-value on slots.

Signaling (throwing) conditions: error, warn, signal

We can use error in two ways:

With our own condition we can do:

(error 'my-division-by-zero :dividend 3)
;; which is a shortcut for
(error (make-condition 'my-division-by-zero :dividend 3))

Throwing these conditions will enter the interactive debugger, where the user may select a restart.

warn will not enter the debugger (create warning conditions by subclassing simple-warning).

Use signal if you do not want to enter the debugger, but you still want to signal to the upper levels that something exceptional happened.

And that can be anything. For example, it can be used to track progress during an operation. You would create a condition with a percent slot, signal one when progress is made, and the higher level code would handle it and display it to the user. See the resources below for more.

Conditions hierarchy

The class precedence list of simple-error is simple-error, simple-condition, error, serious-condition, condition, t.

The class precedence list of simple-warning is simple-warning, simple-condition, warning, condition, t.

Custom error messages (:report)

So far, when throwing our error, we saw this default text in the debugger:

Condition COMMON-LISP-USER::MY-DIVISION-BY-ZERO was signalled.
   [Condition of type MY-DIVISION-BY-ZERO]

We can do better by giving a :report function in our condition declaration:

(define-condition my-division-by-zero (error)
  ((dividend :initarg :dividend
             :initform nil
             :accessor dividend))
  ;; the :report is the message into the debugger:
  (:report (lambda (condition stream)
     (format stream
             "You were going to divide ~a by zero.~&"
             (dividend condition)))))


(error 'my-division-by-zero :dividend 3)
;; Debugger:
;; You were going to divide 3 by zero.
;;    [Condition of type MY-DIVISION-BY-ZERO]

Inspecting the stacktrace

That’s another quick reminder, not a Slime tutorial. In the debugger, you can inspect the stacktrace, the arguments to the function calls, go to the erroneous source line (with v in Slime), execute code in the context (e), etc.

Often, you can edit a buggy function, compile it (with the C-c C-c shortcut in Slime), choose the “RETRY” restart and see your code pass.

All this depends on compiler options, wether it is optimized for debugging, speed or security.

See our debugging section.

Restarts, interactive choices in the debugger

Restarts are the choices we get in the debugger, which always has the RETRY and ABORT ones.

By handling restarts we can start over the operation as if the error didn’t occur (as seen in the stack).

Using assert’s optional restart

In its simple form assert does what we know:

(assert (realp 3))
;; NIL = passed

When the assertion fails, we are prompted into the debugger:

(defun divide (x y)
  (assert (not (zerop y)))
  (/ x y))

(divide 3 0)
;; The assertion (NOT #1=(ZEROP Y)) failed with #1# = T.
;;    [Condition of type SIMPLE-ERROR]
;; Restarts:
;;  0: [CONTINUE] Retry assertion.
;;  1: [RETRY] Retry SLIME REPL evaluation request.
;;  …

It also accepts an optional parameter to offer to change values:

(defun divide (x y)
  (assert (not (zerop y))
          (y)   ;; list of values that we can change.
          "Y can not be zero. Please change it") ;; custom error message.
  (/ x y))

Now we get a new restart that offers to change the value of Y:

(divide 3 0)
;; Y can not be zero. Please change it
;;    [Condition of type SIMPLE-ERROR]
;; Restarts:
;;  0: [CONTINUE] Retry assertion with new value for Y.  <--- new restart
;;  1: [RETRY] Retry SLIME REPL evaluation request.
;;  …

and when we choose it, we are prompted for a new value in the REPL:

The old value of Y is 0.
Do you want to supply a new value?  (y or n) y

Type a form to be evaluated:
3/2  ;; and our result.

Defining restarts (restart-case)

All this is good but we might want more custom choices. We can add restarts on the top of the list by wrapping our function call inside restart-case.

(defun divide-with-restarts (x y)
  (restart-case (/ x y)
    (return-zero ()  ;; <-- creates a new restart called "RETURN-ZERO"
    (divide-by-one ()
      (/ x 1))))
(divide-with-restarts 3 0)

In case of any error (we’ll improve on that with handler-bind), we’ll get those two new choices at the top of the debugger:

That’s allright but let’s just write more human-friendy “reports”:

(defun divide-with-restarts (x y)
  (restart-case (/ x y)
    (return-zero ()
      :report "Return 0"  ;; <-- added
    (divide-by-one ()
      :report "Divide by 1"
      (/ x 1))))
(divide-with-restarts 3 0)
;; Nicer restarts:
;;  0: [RETURN-ZERO] Return 0
;;  1: [DIVIDE-BY-ONE] Divide by 1

That’s better, but we lack the ability to change an operand, as we did with the assert example above.

Changing a variable with restarts

The two restarts we defined didn’t ask for a new value. To do this, we add an :interactive lambda function to the restart, that asks for the user a new value with the input method of its choice. Here, we’ll use the regular read.

(defun divide-with-restarts (x y)
  (restart-case (/ x y)
    (return-zero ()
      :report "Return 0"
    (divide-by-one ()
      :report "Divide by 1"
      (/ x 1))
    (set-new-divisor (value)
      :report "Enter a new divisor"
      ;; Ask the user for a new value:
      :interactive (lambda () (prompt-new-value "Please enter a new divisor: "))
      ;; and call the divide function with the new value…
      ;; … possibly catching bad input again!
      (divide-with-restarts x value))))

(defun prompt-new-value (prompt)
  (format *query-io* prompt) ;; *query-io*: the special stream to make user queries.
  (force-output *query-io*)  ;; Ensure the user sees what he types.
  (list (read *query-io*)))  ;; We must return a list.

(divide-with-restarts 3 0)

When calling it, we are offered a new restart, we enter a new value, and we get our result:

(divide-with-restarts 3 0)
;; Debugger:
;; 2: [SET-NEW-DIVISOR] Enter a new divisor
;; Please enter a new divisor: 10
;; 3/10

Oh, you prefer a graphical user interface? We can use the zenity command line interface on GNU/Linux.

(defun prompt-new-value (prompt)
   (let ((input
          ;; We capture the program's output to a string.
          (with-output-to-string (s)
            (let* ((*standard-output* s))
              (uiop:run-program `("zenity"
                                  ,(format nil "--add-entry=~a" prompt))
                                :output s)))))
     ;; We get a string and we want a number.
     ;; We could also use parse-integer, the parse-number library, etc.
     (read-from-string input))))

Now try again and you should get a little window asking for a new number:

That’s fun, but that’s not all. Choosing restarts manually is not always (or often?) satisfactory. And by handling restarts we can start over the operation as if the error didn’t occur, as seen in the stack.

Calling restarts programmatically (handler-bind, invoke-restart)

We have a piece of code that we know can throw conditions. Here, divide-with-restarts can signal an error about a division by zero. What we want to do, is our higher-level code to automatically handle it and call the appropriate restart.

We can do this with handler-bind and invoke-restart:

(defun divide-and-handle-error (x y)
      ((division-by-zero (lambda (c)
                (format t "Got error: ~a~%" c) ;; error-message
                (format t "and will divide by 1~&")
                (invoke-restart 'divide-by-one))))
    (divide-with-restarts x y)))

(divide-and-handle-error 3 0)
;; Got error: arithmetic error DIVISION-BY-ZERO signalled
;; Operation was (/ 3 0).
;; and will divide by 1
;; 3

Using other restarts (find-restart)

Use find-restart.

find-restart 'name-of-restart will return the most recent bound restart with the given name, or nil.

Hiding and showing restarts

Restarts can be hidden. In restart-case, in addition to :report and :interactive, they also accept a :test key:

   (return-zero ()
     :test (lambda ()

Handling conditions (handler-bind)

We just saw a use for handler-bind.

Its general form is:

(handler-bind ((a-condition #'function-to-handle-it)
               (another-one #'another-function))
    (code that can...)
    (...error out))

If the handler returns normally (it declines to handle the condition), the condition continues to bubble up, searching for another handler, and it will find the interactive debugger (when it’s an error, not when it’s a simple condition).

We can study a real example with the unix-opts library, that parses command line arguments. It defined some conditions: unknown-option, missing-arg and arg-parser-failed, and it is up to us to write what to do in these cases.

(handler-bind ((opts:unknown-option #'unknown-option)
               (opts:missing-arg #'missing-arg)
               (opts:arg-parser-failed #'arg-parser-failed))

Our unknown-option function is simple and looks like this:

(defun unknown-option (condition)
  (format t "~s option is unknown.~%" (opts:option condition))
  (exit)) ;; <-- we return to the command line, no debugger.

it takes the condition as parameter, so we can read information from it if needed. Here we get the name of the erroneous option with the condition’s reader (opts:option condition).

Running some code, condition or not (“finally”) (unwind-protect)

The “finally” part of others try/catch/finally forms is done with unwind-protect.

It is the construct used in “with-“ macros, like with-open-file, which always closes the file after it.

With this example:

(unwind-protect (/ 3 0)
  (format t "This place is safe.~&"))

We do get the interactive debugger (we didn’t use handler-bind or anything), but our message is printed afterwards anyway.


You’re now more than ready to write some code and to dive into other resources!


See also

Page source: