A Lisp program consists of expressions or forms (see section Kinds of Forms). We control the order of execution of the forms by enclosing them in control structures. Control structures are special forms which control when, whether, or how many times to execute the forms they contain.
The simplest order of execution is sequential execution: first form a, then form b, and so on. This is what happens when you write several forms in succession in the body of a function, or at top level in a file of Lisp code--the forms are executed in the order written. We call this textual order. For example, if a function body consists of two forms a and b, evaluation of the function evaluates first a and then b, and the function's value is the value of b.
Explicit control structures make possible an order of execution other than sequential.
Emacs Lisp provides several kinds of control structure, including other varieties of sequencing, conditionals, iteration, and (controlled) jumps--all discussed below. The built-in control structures are special forms since their subforms are not necessarily evaluated or not evaluated sequentially. You can use macros to define your own control structure constructs (see section Macros).
if, cond.
and, or, not.
while loops.
Evaluating forms in the order they appear is the most common way
control passes from one form to another. In some contexts, such as in a
function body, this happens automatically. Elsewhere you must use a
control structure construct to do this: progn, the simplest
control construct of Lisp.
A progn special form looks like this:
(progn a b c ...)
and it says to execute the forms a, b, c and so on, in
that order. These forms are called the body of the progn form.
The value of the last form in the body becomes the value of the entire
progn.
In the early days of Lisp, progn was the only way to execute
two or more forms in succession and use the value of the last of them.
But programmers found they often needed to use a progn in the
body of a function, where (at that time) only one form was allowed. So
the body of a function was made into an "implicit progn":
several forms are allowed just as in the body of an actual progn.
Many other control structures likewise contain an implicit progn.
As a result, progn is not used as often as it used to be. It is
needed now most often inside an unwind-protect, and,
or, or in the then-part of an if.
(progn (print "The first form")
(print "The second form")
(print "The third form"))
-| "The first form"
-| "The second form"
-| "The third form"
=> "The third form"
Two other control constructs likewise evaluate a series of forms but return a different value:
(prog1 (print "The first form")
(print "The second form")
(print "The third form"))
-| "The first form"
-| "The second form"
-| "The third form"
=> "The first form"
Here is a way to remove the first element from a list in the variable
x, then return the value of that former element:
(prog1 (car x) (setq x (cdr x)))
(prog2 (print "The first form")
(print "The second form")
(print "The third form"))
-| "The first form"
-| "The second form"
-| "The third form"
=> "The second form"
Conditional control structures choose among alternatives. Emacs Lisp
has two conditional forms: if, which is much the same as in other
languages, and cond, which is a generalized case statement.
if chooses between the then-form and the else-forms
based on the value of condition. If the evaluated condition is
non-nil, then-form is evaluated and the result returned.
Otherwise, the else-forms are evaluated in textual order, and the
value of the last one is returned. (The else part of if is
an example of an implicit progn. See section Sequencing.)
If condition has the value nil, and no else-forms are
given, if returns nil.
if is a special form because the branch that is not selected is
never evaluated--it is ignored. Thus, in the example below,
true is not printed because print is never called.
(if nil
(print 'true)
'very-false)
=> very-false
cond chooses among an arbitrary number of alternatives. Each
clause in the cond must be a list. The CAR of this
list is the condition; the remaining elements, if any, the
body-forms. Thus, a clause looks like this:
(condition body-forms...)
cond tries the clauses in textual order, by evaluating the
condition of each clause. If the value of condition is
non-nil, the clause "succeeds"; then cond evaluates its
body-forms, and the value of the last of body-forms becomes
the value of the cond. The remaining clauses are ignored.
If the value of condition is nil, the clause "fails", so
the cond moves on to the following clause, trying its
condition.
If every condition evaluates to nil, so that every clause
fails, cond returns nil.
A clause may also look like this:
(condition)
Then, if condition is non-nil when tested, the value of
condition becomes the value of the cond form.
The following example has four clauses, which test for the cases where
the value of x is a number, string, buffer and symbol,
respectively:
(cond ((numberp x) x)
((stringp x) x)
((bufferp x)
(setq temporary-hack x) ; multiple body-forms
(buffer-name x)) ; in one clause
((symbolp x) (symbol-value x)))
Often we want to execute the last clause whenever none of the previous
clauses was successful. To do this, we use t as the
condition of the last clause, like this: (t
body-forms). The form t evaluates to t, which is
never nil, so this clause never fails, provided the cond
gets to it at all.
For example,
(cond ((eq a 'hack) 'foo)
(t "default"))
=> "default"
This expression is a cond which returns foo if the value
of a is 1, and returns the string "default" otherwise.
Any conditional construct can be expressed with cond or with
if. Therefore, the choice between them is a matter of style.
For example:
(if a b c) == (cond (a b) (t c))
This section describes three constructs that are often used together
with if and cond to express complicated conditions. The
constructs and and or can also be used individually as
kinds of multiple conditional constructs.
t if condition is nil, and nil otherwise.
The function not is identical to null, and we recommend
using the name null if you are testing for an empty list.
and special form tests whether all the conditions are
true. It works by evaluating the conditions one by one in the
order written.
If any of the conditions evaluates to nil, then the result
of the and must be nil regardless of the remaining
conditions; so and returns right away, ignoring the
remaining conditions.
If all the conditions turn out non-nil, then the value of
the last of them becomes the value of the and form.
Here is an example. The first condition returns the integer 1, which is
not nil. Similarly, the second condition returns the integer 2,
which is not nil. The third condition is nil, so the
remaining condition is never evaluated.
(and (print 1) (print 2) nil (print 3))
-| 1
-| 2
=> nil
Here is a more realistic example of using and:
(if (and (consp foo) (eq (car foo) 'x))
(message "foo is a list starting with x"))
Note that (car foo) is not executed if (consp foo) returns
nil, thus avoiding an error.
and can be expressed in terms of either if or cond.
For example:
(and arg1 arg2 arg3) == (if arg1 (if arg2 arg3)) == (cond (arg1 (cond (arg2 arg3))))
or special form tests whether at least one of the
conditions is true. It works by evaluating all the
conditions one by one in the order written.
If any of the conditions evaluates to a non-nil value, then
the result of the or must be non-nil; so or returns
right away, ignoring the remaining conditions. The value it
returns is the non-nil value of the condition just evaluated.
If all the conditions turn out nil, then the or
expression returns nil.
For example, this expression tests whether x is either 0 or
nil:
(or (eq x nil) (eq x 0))
Like the and construct, or can be written in terms of
cond. For example:
(or arg1 arg2 arg3)
==
(cond (arg1)
(arg2)
(arg3))
You could almost write or in terms of if, but not quite:
(if arg1 arg1
(if arg2 arg2
arg3))
This is not completely equivalent because it can evaluate arg1 or
arg2 twice. By contrast, (or arg1 arg2
arg3) never evaluates any argument more than once.
Iteration means executing part of a program repetitively. For
example, you might want to repeat some computation once for each element
of a list, or once for each integer from 0 to n. You can do this
in Emacs Lisp with the special form while:
while first evaluates condition. If the result is
non-nil, it evaluates forms in textual order. Then it
reevaluates condition, and if the result is non-nil, it
evaluates forms again. This process repeats until condition
evaluates to nil.
There is no limit on the number of iterations that may occur. The loop
will continue until either condition evaluates to nil or
until an error or throw jumps out of it (see section Nonlocal Exits).
The value of a while form is always nil.
(setq num 0)
=> 0
(while (< num 4)
(princ (format "Iteration %d." num))
(setq num (1+ num)))
-| Iteration 0.
-| Iteration 1.
-| Iteration 2.
-| Iteration 3.
=> nil
If you would like to execute something on each iteration before the
end-test, put it together with the end-test in a progn as the
first argument of while, as shown here:
(while (progn
(forward-line 1)
(not (looking-at "^$"))))
This moves forward one line and continues moving by lines until it
reaches an empty. It is unusual in that the while has no body,
just the end test (which also does the real work of moving point).
A nonlocal exit is a transfer of control from one point in a program to another remote point. Nonlocal exits can occur in Emacs Lisp as a result of errors; you can also use them under explicit control. Nonlocal exits unbind all variable bindings made by the constructs being exited.
catch and throw
Most control constructs affect only the flow of control within the
construct itself. The function throw is the exception to this
rule of normal program execution: it performs a nonlocal exit on
request. (There are other exceptions, but they are for error handling
only.) throw is used inside a catch, and jumps back to
that catch. For example:
(catch 'foo
(progn
...
(throw 'foo t)
...))
The throw transfers control straight back to the corresponding
catch, which returns immediately. The code following the
throw is not executed. The second argument of throw is used
as the return value of the catch.
The throw and the catch are matched through the first
argument: throw searches for a catch whose first argument
is eq to the one specified. Thus, in the above example, the
throw specifies foo, and the catch specifies the
same symbol, so that catch is applicable. If there is more than
one applicable catch, the innermost one takes precedence.
Executing throw exits all Lisp constructs up to the matching
catch, including function calls. When binding constructs such as
let or function calls are exited in this way, the bindings are
unbound, just as they are when these constructs exit normally
(see section Local Variables). Likewise, throw restores the buffer
and position saved by save-excursion (see section Excursions), and
the narrowing status saved by save-restriction and the window
selection saved by save-window-excursion (see section Window Configurations). It also runs any cleanups established with the
unwind-protect special form when it exits that form
(see section Cleaning Up from Nonlocal Exits).
The throw need not appear lexically within the catch
that it jumps to. It can equally well be called from another function
called within the catch. As long as the throw takes place
chronologically after entry to the catch, and chronologically
before exit from it, it has access to that catch. This is why
throw can be used in commands such as exit-recursive-edit
that throw back to the editor command loop (see section Recursive Editing).
Common Lisp note: Most other versions of Lisp, including Common Lisp, have several ways of transferring control nonsequentially:
return,return-from, andgo, for example. Emacs Lisp has onlythrow.
catch establishes a return point for the throw function. The
return point is distinguished from other such return points by tag,
which may be any Lisp object. The argument tag is evaluated normally
before the return point is established.
With the return point in effect, catch evaluates the forms of the
body in textual order. If the forms execute normally, without
error or nonlocal exit, the value of the last body form is returned from
the catch.
If a throw is done within body specifying the same value
tag, the catch exits immediately; the value it returns is
whatever was specified as the second argument of throw.
throw is to return from a return point previously
established with catch. The argument tag is used to choose
among the various existing return points; it must be eq to the value
specified in the catch. If multiple return points match tag,
the innermost one is used.
The argument value is used as the value to return from that
catch.
If no return point is in effect with tag tag, then a no-catch
error is signaled with data (tag value).
catch and throw
One way to use catch and throw is to exit from a doubly
nested loop. (In most languages, this would be done with a "go to".)
Here we compute (foo i j) for i and j
varying from 0 to 9:
(defun search-foo ()
(catch 'loop
(let ((i 0))
(while (< i 10)
(let ((j 0))
(while (< j 10)
(if (foo i j)
(throw 'loop (list i j)))
(setq j (1+ j))))
(setq i (1+ i))))))
If foo ever returns non-nil, we stop immediately and return a
list of i and j. If foo always returns nil, the
catch returns normally, and the value is nil, since that
is the result of the while.
Here are two tricky examples, slightly different, showing two
return points at once. First, two return points with the same tag,
hack:
(defun catch2 (tag)
(catch tag
(throw 'hack 'yes)))
=> catch2
(catch 'hack
(print (catch2 'hack))
'no)
-| yes
=> no
Since both return points have tags that match the throw, it goes to
the inner one, the one established in catch2. Therefore,
catch2 returns normally with value yes, and this value is
printed. Finally the second body form in the outer catch, which is
'no, is evaluated and returned from the outer catch.
Now let's change the argument given to catch2:
(defun catch2 (tag)
(catch tag
(throw 'hack 'yes)))
=> catch2
(catch 'hack
(print (catch2 'quux))
'no)
=> yes
We still have two return points, but this time only the outer one has
the tag hack; the inner one has the tag quux instead.
Therefore, throw makes the outer catch return the value
yes. The function print is never called, and the
body-form 'no is never evaluated.
When Emacs Lisp attempts to evaluate a form that, for some reason, cannot be evaluated, it signals an error.
When an error is signaled, Emacs's default reaction is to print an error message and terminate execution of the current command. This is the right thing to do in most cases, such as if you type C-f at the end of the buffer.
In complicated programs, simple termination may not be what you want.
For example, the program may have made temporary changes in data
structures, or created temporary buffers that should be deleted before
the program is finished. In such cases, you would use
unwind-protect to establish cleanup expressions to be
evaluated in case of error. (See section Cleaning Up from Nonlocal Exits.) Occasionally, you may
wish the program to continue execution despite an error in a subroutine.
In these cases, you would use condition-case to establish
error handlers to recover control in case of error.
Resist the temptation to use error handling to transfer control from
one part of the program to another; use catch and throw
instead. See section Explicit Nonlocal Exits: catch and throw.
Most errors are signaled "automatically" within Lisp primitives
which you call for other purposes, such as if you try to take the
CAR of an integer or move forward a character at the end of the
buffer; you can also signal errors explicitly with the functions
error and signal.
Quitting, which happens when the user types C-g, is not considered an error, but it is handled almost like an error. See section Quitting.
format (see section Conversion of Characters and Strings) to
format-string and args.
These examples show typical uses of error:
(error "You have committed an error.
Try something else.")
error--> You have committed an error.
Try something else.
(error "You have committed %d errors." 10)
error--> You have committed 10 errors.
error works by calling signal with two arguments: the
error symbol error, and a list containing the string returned by
format.
If you want to use your own string as an error message verbatim, don't
just write (error string). If string contains
`%', it will be interpreted as a format specifier, with undesirable
results. Instead, use (error "%s" string).
The argument error-symbol must be an error symbol---a symbol
bearing a property error-conditions whose value is a list of
condition names. This is how Emacs Lisp classifies different sorts of
errors.
The number and significance of the objects in data depends on
error-symbol. For example, with a wrong-type-arg error,
there are two objects in the list: a predicate that describes the type
that was expected, and the object that failed to fit that type.
See section Error Symbols and Condition Names, for a description of error symbols.
Both error-symbol and data are available to any error
handlers that handle the error: condition-case binds a local
variable to a list of the form (error-symbol .
data) (see section Writing Code to Handle Errors). If the error is not handled,
these two values are used in printing the error message.
The function signal never returns (though in older Emacs versions
it could sometimes return).
(signal 'wrong-number-of-arguments '(x y))
error--> Wrong number of arguments: x, y
(signal 'no-such-error '("My unknown error condition."))
error--> peculiar error: "My unknown error condition."
Common Lisp note: Emacs Lisp has nothing like the Common Lisp concept of continuable errors.
When an error is signaled, signal searches for an active
handler for the error. A handler is a sequence of Lisp
expressions designated to be executed if an error happens in part of the
Lisp program. If the error has an applicable handler, the handler is
executed, and control resumes following the handler. The handler
executes in the environment of the condition-case that
established it; all functions called within that condition-case
have already been exited, and the handler cannot return to them.
If there is no applicable handler for the error, the current command is terminated and control returns to the editor command loop, because the command loop has an implicit handler for all kinds of errors. The command loop's handler uses the error symbol and associated data to print an error message.
An error that has no explicit handler may call the Lisp debugger. The
debugger is enabled if the variable debug-on-error (see section Entering the Debugger on an Error) is non-nil. Unlike error handlers, the debugger runs
in the environment of the error, so that you can examine values of
variables precisely as they were at the time of the error.
The usual effect of signaling an error is to terminate the command
that is running and return immediately to the Emacs editor command loop.
You can arrange to trap errors occurring in a part of your program by
establishing an error handler, with the special form
condition-case. A simple example looks like this:
(condition-case nil
(delete-file filename)
(error nil))
This deletes the file named filename, catching any error and
returning nil if an error occurs.
The second argument of condition-case is called the
protected form. (In the example above, the protected form is a
call to delete-file.) The error handlers go into effect when
this form begins execution and are deactivated when this form returns.
They remain in effect for all the intervening time. In particular, they
are in effect during the execution of functions called by this form, in
their subroutines, and so on. This is a good thing, since, strictly
speaking, errors can be signaled only by Lisp primitives (including
signal and error) called by the protected form, not by the
protected form itself.
The arguments after the protected form are handlers. Each handler
lists one or more condition names (which are symbols) to specify
which errors it will handle. The error symbol specified when an error
is signaled also defines a list of condition names. A handler applies
to an error if they have any condition names in common. In the example
above, there is one handler, and it specifies one condition name,
error, which covers all errors.
The search for an applicable handler checks all the established handlers
starting with the most recently established one. Thus, if two nested
condition-case forms offer to handle the same error, the inner of
the two will actually handle it.
When an error is handled, control returns to the handler. Before this
happens, Emacs unbinds all variable bindings made by binding constructs
that are being exited and executes the cleanups of all
unwind-protect forms that are exited. Once control arrives at
the handler, the body of the handler is executed.
After execution of the handler body, execution continues by returning
from the condition-case form. Because the protected form is
exited completely before execution of the handler, the handler cannot
resume execution at the point of the error, nor can it examine variable
bindings that were made within the protected form. All it can do is
clean up and proceed.
condition-case is often used to trap errors that are
predictable, such as failure to open a file in a call to
insert-file-contents. It is also used to trap errors that are
totally unpredictable, such as when the program evaluates an expression
read from the user.
Error signaling and handling have some resemblance to throw and
catch, but they are entirely separate facilities. An error
cannot be caught by a catch, and a throw cannot be handled
by an error handler (though using throw when there is no suitable
catch signals an error that can be handled).
condition-case form; in this case, the condition-case has
no effect. The condition-case form makes a difference when an
error occurs during protected-form.
Each of the handlers is a list of the form (conditions
body...). Here conditions is an error condition name
to be handled, or a list of condition names; body is one or more
Lisp expressions to be executed when this handler handles an error.
Here are examples of handlers:
(error nil) (arith-error (message "Division by zero")) ((arith-error file-error) (message "Either division by zero or failure to open a file"))
Each error that occurs has an error symbol that describes what
kind of error it is. The error-conditions property of this
symbol is a list of condition names (see section Error Symbols and Condition Names). Emacs
searches all the active condition-case forms for a handler that
specifies one or more of these condition names; the innermost matching
condition-case handles the error. Within this
condition-case, the first applicable handler handles the error.
After executing the body of the handler, the condition-case
returns normally, using the value of the last form in the handler body
as the overall value.
The argument var is a variable. condition-case does not
bind this variable when executing the protected-form, only when it
handles an error. At that time, it binds var locally to a list of
the form (error-symbol . data), giving the
particulars of the error. The handler can refer to this list to decide
what to do. For example, if the error is for failure opening a file,
the file name is the second element of data---the third element of
var.
If var is nil, that means no variable is bound. Then the
error symbol and associated data are not available to the handler.
Here is an example of using condition-case to handle the error
that results from dividing by zero. The handler prints out a warning
message and returns a very large number.
(defun safe-divide (dividend divisor)
(condition-case err
;; Protected form.
(/ dividend divisor)
;; The handler.
(arith-error ; Condition.
(princ (format "Arithmetic error: %s" err))
1000000)))
=> safe-divide
(safe-divide 5 0)
-| Arithmetic error: (arith-error)
=> 1000000
The handler specifies condition name arith-error so that it will handle only division-by-zero errors. Other kinds of errors will not be handled, at least not by this condition-case. Thus,
(safe-divide nil 3)
error--> Wrong type argument: integer-or-marker-p, nil
Here is a condition-case that catches all kinds of errors,
including those signaled with error:
(setq baz 34)
=> 34
(condition-case err
(if (eq baz 35)
t
;; This is a call to the function error.
(error "Rats! The variable %s was %s, not 35." 'baz baz))
;; This is the handler; it is not a form.
(error (princ (format "The error was: %s" err))
2))
-| The error was: (error "Rats! The variable baz was 34, not 35.")
=> 2
When you signal an error, you specify an error symbol to specify the kind of error you have in mind. Each error has one and only one error symbol to categorize it. This is the finest classification of errors defined by the Emacs Lisp language.
These narrow classifications are grouped into a hierarchy of wider
classes called error conditions, identified by condition
names. The narrowest such classes belong to the error symbols
themselves: each error symbol is also a condition name. There are also
condition names for more extensive classes, up to the condition name
error which takes in all kinds of errors. Thus, each error has
one or more condition names: error, the error symbol if that
is distinct from error, and perhaps some intermediate
classifications.
In order for a symbol to be an error symbol, it must have an
error-conditions property which gives a list of condition names.
This list defines the conditions that this kind of error belongs to.
(The error symbol itself, and the symbol error, should always be
members of this list.) Thus, the hierarchy of condition names is
defined by the error-conditions properties of the error symbols.
In addition to the error-conditions list, the error symbol
should have an error-message property whose value is a string to
be printed when that error is signaled but not handled. If the
error-message property exists, but is not a string, the error
message `peculiar error' is used.
Here is how we define a new error symbol, new-error:
(put 'new-error
'error-conditions
'(error my-own-errors new-error))
=> (error my-own-errors new-error)
(put 'new-error 'error-message "A new error")
=> "A new error"
This error has three condition names: new-error, the narrowest
classification; my-own-errors, which we imagine is a wider
classification; and error, which is the widest of all.
Naturally, Emacs will never signal new-error on its own; only
an explicit call to signal (see section How to Signal an Error) in your
code can do this:
(signal 'new-error '(x y))
error--> A new error: x, y
This error can be handled through any of the three condition names.
This example handles new-error and any other errors in the class
my-own-errors:
(condition-case foo
(bar nil t)
(my-own-errors nil))
The significant way that errors are classified is by their condition
names--the names used to match errors with handlers. An error symbol
serves only as a convenient way to specify the intended error message
and list of condition names. It would be cumbersome to give
signal a list of condition names rather than one error symbol.
By contrast, using only error symbols without condition names would
seriously decrease the power of condition-case. Condition names
make it possible to categorize errors at various levels of generality
when you write an error handler. Using error symbols alone would
eliminate all but the narrowest level of classification.
See section Standard Errors, for a list of all the standard error symbols and their conditions.
The unwind-protect construct is essential whenever you
temporarily put a data structure in an inconsistent state; it permits
you to ensure the data are consistent in the event of an error or throw.
unwind-protect executes the body with a guarantee that the
cleanup-forms will be evaluated if control leaves body, no
matter how that happens. The body may complete normally, or
execute a throw out of the unwind-protect, or cause an
error; in all cases, the cleanup-forms will be evaluated.
If the body forms finish normally, unwind-protect returns
the value of the last body form, after it evaluates the
cleanup-forms. If the body forms do not finish,
unwind-protect does not return any value in the normal sense.
Only the body is actually protected by the unwind-protect.
If any of the cleanup-forms themselves exits nonlocally (e.g., via
a throw or an error), unwind-protect is not
guaranteed to evaluate the rest of them. If the failure of one of the
cleanup-forms has the potential to cause trouble, then protect it
with another unwind-protect around that form.
The number of currently active unwind-protect forms counts,
together with the number of local variable bindings, against the limit
max-specpdl-size (see section Local Variables).
For example, here we make an invisible buffer for temporary use, and make sure to kill it before finishing:
(save-excursion
(let ((buffer (get-buffer-create " *temp*")))
(set-buffer buffer)
(unwind-protect
body
(kill-buffer buffer))))
You might think that we could just as well write (kill-buffer
(current-buffer)) and dispense with the variable buffer.
However, the way shown above is safer, if body happens to get an
error after switching to a different buffer! (Alternatively, you could
write another save-excursion around the body, to ensure that the
temporary buffer becomes current in time to kill it.)
Here is an actual example taken from the file `ftp.el'. It
creates a process (see section Processes) to try to establish a connection
to a remote machine. As the function ftp-login is highly
susceptible to numerous problems that the writer of the function cannot
anticipate, it is protected with a form that guarantees deletion of the
process in the event of failure. Otherwise, Emacs might fill up with
useless subprocesses.
(let ((win nil))
(unwind-protect
(progn
(setq process (ftp-setup-buffer host file))
(if (setq win (ftp-login process host user password))
(message "Logged in")
(error "Ftp login failed")))
(or win (and process (delete-process process)))))
This example actually has a small bug: if the user types C-g to
quit, and the quit happens immediately after the function
ftp-setup-buffer returns but before the variable process is
set, the process will not be killed. There is no easy way to fix this bug,
but at least it is very unlikely.
Here is another example which uses unwind-protect to make sure
to kill a temporary buffer. In this example, the value returned by
unwind-protect is used.
(defun shell-command-string (cmd)
"Return the output of the shell command CMD, as a string."
(save-excursion
(set-buffer (generate-new-buffer " OS*cmd"))
(shell-command cmd t)
(unwind-protect
(buffer-string)
(kill-buffer (current-buffer)))))