Common Lisp Extensions


Common Lisp Extensions

This document describes a set of Emacs Lisp facilities borrowed from Common Lisp. All the facilities are described here in detail. While this document does not assume any prior knowledge of Common Lisp, it does assume a basic familiarity with Emacs Lisp.

Overview

Common Lisp is a huge language, and Common Lisp systems tend to be massive and extremely complex. Emacs Lisp, by contrast, is rather minimalist in the choice of Lisp features it offers the programmer. As Emacs Lisp programmers have grown in number, and the applications they write have grown more ambitious, it has become clear that Emacs Lisp could benefit from many of the conveniences of Common Lisp.

The CL package adds a number of Common Lisp functions and control structures to Emacs Lisp. While not a 100% complete implementation of Common Lisp, CL adds enough functionality to make Emacs Lisp programming significantly more convenient.

Some Common Lisp features have been omitted from this package for various reasons:

The package described here was written by Dave Gillespie, `daveg@synaptics.com'. It is a total rewrite of the original 1986 `cl.el' package by Cesar Quiroz. Most features of the the Quiroz package have been retained; any incompatibilities are noted in the descriptions below. Care has been taken in this version to ensure that each function is defined efficiently, concisely, and with minimal impact on the rest of the Emacs environment.

Usage

Lisp code that uses features from the CL package should include at the beginning:

(require 'cl)

If you want to ensure that the new (Gillespie) version of CL is the one that is present, add an additional (require 'cl-19) call:

(require 'cl)
(require 'cl-19)

The second call will fail (with "`cl-19.el' not found") if the old `cl.el' package was in use.

It is safe to arrange to load CL at all times, e.g., in your `.emacs' file. But it's a good idea, for portability, to (require 'cl) in your code even if you do this.

Organization

The Common Lisp package is organized into four files:

`cl.el'
This is the "main" file, which contains basic functions and information about the package. This file is relatively compact--about 700 lines.
`cl-extra.el'
This file contains the larger, more complex or unusual functions. It is kept separate so that packages which only want to use Common Lisp fundamentals like the cadr function won't need to pay the overhead of loading the more advanced functions.
`cl-seq.el'
This file contains most of the advanced functions for operating on sequences or lists, such as delete-if and assoc*.
`cl-macs.el'
This file contains the features of the packages which are macros instead of functions. Macros expand when the caller is compiled, not when it is run, so the macros generally only need to be present when the byte-compiler is running (or when the macros are used in uncompiled code such as a `.emacs' file). Most of the macros of this package are isolated in `cl-macs.el' so that they won't take up memory unless you are compiling.

The file `cl.el' includes all necessary autoload commands for the functions and macros in the other three files. All you have to do is (require 'cl), and `cl.el' will take care of pulling in the other files when they are needed.

There is another file, `cl-compat.el', which defines some routines from the older `cl.el' package that are no longer present in the new package. This includes internal routines like setelt and zip-lists, deprecated features like defkeyword, and an emulation of the old-style multiple-values feature. See section Old CL Compatibility.

Installation

Installation of the CL package is simple: Just put the byte-compiled files `cl.elc', `cl-extra.elc', `cl-seq.elc', `cl-macs.elc', and `cl-compat.elc' into a directory on your load-path.

There are no special requirements to compile this package: The files do not have to be loaded before they are compiled, nor do they need to be compiled in any particular order.

You may choose to put the files into your main `lisp/' directory, replacing the original `cl.el' file there. Or, you could put them into a directory that comes before `lisp/' on your load-path so that the old `cl.el' is effectively hidden.

Also, format the `cl.texinfo' file and put the resulting Info files in the `info/' directory or another suitable place.

You may instead wish to leave this package's components all in their own directory, and then add this directory to your load-path and (Emacs 19 only) Info-directory-list. Add the directory to the front of the list so the old CL package and its documentation are hidden.

Naming Conventions

Except where noted, all functions defined by this package have the same names and calling conventions as their Common Lisp counterparts.

Following is a complete list of functions whose names were changed from Common Lisp, usually to avoid conflicts with Emacs. In each case, a `*' has been appended to the Common Lisp name to obtain the Emacs name:

defun*        defsubst*     defmacro*     function*
member*       assoc*        rassoc*       get*
remove*       delete*       mapcar*       sort*
floor*        ceiling*      truncate*     round*
mod*          rem*          random*       last*

Internal function and variable names in the package are prefixed by cl-. Here is a complete list of functions not prefixed by cl- which were not taken from Common Lisp:

member        delete        remove        remq
rassoc        floatp-safe   lexical-let   lexical-let*
callf         callf2        letf          letf*
defsubst*     defalias      add-hook      eval-when-compile

(Most of these are Emacs 19 features provided to Emacs 18 users, or introduced, like remq, for reasons of symmetry with similar features.)

The following simple functions and macros are defined in `cl.el'; they do not cause other components like `cl-extra' to be loaded.

eql           floatp-safe   abs           endp
evenp         oddp          plusp         minusp
butlast       nbutlast                    caar .. cddddr
list*         ldiff         rest          first .. tenth
member [1]    copy-list     subst         mapcar* [2]
adjoin [3]    acons         pairlis       when
unless        pop [4]       push [4]      pushnew [3,4]
incf [4]      decf [4]      proclaim      declaim
add-hook

[1] This is the Emacs 19-compatible function, not member*.

[2] Only for one sequence argument or two list arguments.

[3] Only if :test is eq, equal, or unspecified, and :key is not used.

[4] Only when place is a plain variable name.

@chapno=4

Program Structure

This section describes features of the CL package which have to do with programs as a whole: advanced argument lists for functions, and the eval-when construct.

@secno=1

Argument Lists

Emacs Lisp's notation for argument lists of functions is a subset of the Common Lisp notation. As well as the familiar &optional and &rest markers, Common Lisp allows you to specify default values for optional arguments, and it provides the additional markers &key and &aux.

Since argument parsing is built-in to Emacs, there is no way for this package to implement Common Lisp argument lists seamlessly. Instead, this package defines alternates for several Lisp forms which you must use if you need Common Lisp argument lists.

Special Form: defun* name arglist body...
This form is identical to the regular defun form, except that arglist is allowed to be a full Common Lisp argument list. Also, the function body is enclosed in an implicit block called name; see section Blocks and Exits.

Special Form: defsubst* name arglist body...
This is just like defun*, except that the function that is defined is automatically proclaimed inline, i.e., calls to it may be expanded into in-line code by the byte compiler. This is analogous to the defsubst form in Emacs 19; defsubst* uses a different method (compiler macros) which works in all version of Emacs, and also generates somewhat more efficient inline expansions. In particular, defsubst* arranges for the processing of keyword arguments, default values, etc., to be done at compile-time whenever possible.

Special Form: defmacro* name arglist body...
This is identical to the regular defmacro form, except that arglist is allowed to be a full Common Lisp argument list. The &environment keyword is supported as described in Steele. The &whole keyword is supported only within destructured lists (see below); top-level &whole cannot be implemented with the current Emacs Lisp interpreter. The macro expander body is enclosed in an implicit block called name.

Special Form: function* symbol-or-lambda
This is identical to the regular function form, except that if the argument is a lambda form then that form may use a full Common Lisp argument list.

Also, all forms (such as defsetf and flet) defined in this package that include arglists in their syntax allow full Common Lisp argument lists.

Note that it is not necessary to use defun* in order to have access to most CL features in your function. These features are always present; defun*'s only difference from defun is its more flexible argument lists and its implicit block.

The full form of a Common Lisp argument list is

(var...
 &optional (var initform svar)...
 &rest var
 &key ((keyword var) initform svar)...
 &aux (var initform)...)

Each of the five argument list sections is optional. The svar, initform, and keyword parts are optional; if they are omitted, then `(var)' may be written simply `var'.

The first section consists of zero or more required arguments. These arguments must always be specified in a call to the function; there is no difference between Emacs Lisp and Common Lisp as far as required arguments are concerned.

The second section consists of optional arguments. These arguments may be specified in the function call; if they are not, initform specifies the default value used for the argument. (No initform means to use nil as the default.) The initform is evaluated with the bindings for the preceding arguments already established; (a &optional (b (1+ a))) matches one or two arguments, with the second argument defaulting to one plus the first argument. If the svar is specified, it is an auxiliary variable which is bound to t if the optional argument was specified, or to nil if the argument was omitted. If you don't use an svar, then there will be no way for your function to tell whether it was called with no argument, or with the default value passed explicitly as an argument.

The third section consists of a single rest argument. If more arguments were passed to the function than are accounted for by the required and optional arguments, those extra arguments are collected into a list and bound to the "rest" argument variable. Common Lisp's &rest is equivalent to that of Emacs Lisp. Common Lisp accepts &body as a synonym for &rest in macro contexts; this package accepts it all the time.

The fourth section consists of keyword arguments. These are optional arguments which are specified by name rather than positionally in the argument list. For example,

(defun* foo (a &optional b &key c d (e 17)))

defines a function which may be called with one, two, or more arguments. The first two arguments are bound to a and b in the usual way. The remaining arguments must be pairs of the form :c, :d, or :e followed by the value to be bound to the corresponding argument variable. (Symbols whose names begin with a colon are called keywords, and they are self-quoting in the same way as nil and t.)

For example, the call (foo 1 2 :d 3 :c 4) sets the five arguments to 1, 2, 4, 3, and 17, respectively. If the same keyword appears more than once in the function call, the first occurrence takes precedence over the later ones. Note that it is not possible to specify keyword arguments without specifying the optional argument b as well, since (foo 1 :c 2) would bind b to the keyword :c, then signal an error because 2 is not a valid keyword.

If a keyword symbol is explicitly specified in the argument list as shown in the above diagram, then that keyword will be used instead of just the variable name prefixed with a colon. You can specify a keyword symbol which does not begin with a colon at all, but such symbols will not be self-quoting; you will have to quote them explicitly with an apostrophe in the function call.

Ordinarily it is an error to pass an unrecognized keyword to a function, e.g., (foo 1 2 :c 3 :goober 4). You can ask Lisp to ignore unrecognized keywords, either by adding the marker &allow-other-keys after the keyword section of the argument list, or by specifying an :allow-other-keys argument in the call whose value is non-nil. If the function uses both &rest and &key at the same time, the "rest" argument is bound to the keyword list as it appears in the call. For example:

(defun* find-thing (thing &rest rest &key need &allow-other-keys)
  (or (apply 'member* thing thing-list :allow-other-keys t rest)
      (if need (error "Thing not found"))))

This function takes a :need keyword argument, but also accepts other keyword arguments which are passed on to the member* function. allow-other-keys is used to keep both find-thing and member* from complaining about each others' keywords in the arguments.

As a (significant) performance optimization, this package implements the scan for keyword arguments by calling memq to search for keywords in a "rest" argument. Technically speaking, this is incorrect, since memq looks at the odd-numbered values as well as the even-numbered keywords. The net effect is that if you happen to pass a keyword symbol as the value of another keyword argument, where that keyword symbol happens to equal the name of a valid keyword argument of the same function, then the keyword parser will become confused. This minor bug can only affect you if you use keyword symbols as general-purpose data in your program; this practice is strongly discouraged in Emacs Lisp.

The fifth section of the argument list consists of auxiliary variables. These are not really arguments at all, but simply variables which are bound to nil or to the specified initforms during execution of the function. There is no difference between the following two functions, except for a matter of stylistic taste:

(defun* foo (a b &aux (c (+ a b)) d)
  body)

(defun* foo (a b)
  (let ((c (+ a b)) d)
    body))

Argument lists support destructuring. In Common Lisp, destructuring is only allowed with defmacro; this package allows it with defun* and other argument lists as well. In destructuring, any argument variable (var in the above diagram) can be replaced by a list of variables, or more generally, a recursive argument list. The corresponding argument value must be a list whose elements match this recursive argument list. For example:

(defmacro* dolist ((var listform &optional resultform)
                   &rest body)
  ...)

This says that the first argument of dolist must be a list of two or three items; if there are other arguments as well as this list, they are stored in body. All features allowed in regular argument lists are allowed in these recursive argument lists. In addition, the clause `&whole var' is allowed at the front of a recursive argument list. It binds var to the whole list being matched; thus (&whole all a b) matches a list of two things, with a bound to the first thing, b bound to the second thing, and all bound to the list itself. (Common Lisp allows &whole in top-level defmacro argument lists as well, but Emacs Lisp does not support this usage.)

One last feature of destructuring is that the argument list may be dotted, so that the argument list (a b . c) is functionally equivalent to (a b &rest c).

If the optimization quality safety is set to 0 (see section Declarations), error checking for wrong number of arguments and invalid keyword arguments is disabled. By default, argument lists are rigorously checked.

Time of Evaluation

Normally, the byte-compiler does not actually execute the forms in a file it compiles. For example, if a file contains (setq foo t), the act of compiling it will not actually set foo to t. This is true even if the setq was a top-level form (i.e., not enclosed in a defun or other form). Sometimes, though, you would like to have certain top-level forms evaluated at compile-time. For example, the compiler effectively evaluates defmacro forms at compile-time so that later parts of the file can refer to the macros that are defined.

Special Form: eval-when (situations...) forms...
This form controls when the body forms are evaluated. The situations list may contain any set of the symbols compile, load, and eval (or their long-winded ANSI equivalents, :compile-toplevel, :load-toplevel, and :execute).

The eval-when form is handled differently depending on whether or not it is being compiled as a top-level form. Specifically, it gets special treatment if it is being compiled by a command such as byte-compile-file which compiles files or buffers of code, and it appears either literally at the top level of the file or inside a top-level progn.

For compiled top-level eval-whens, the body forms are executed at compile-time if compile is in the situations list, and the forms are written out to the file (to be executed at load-time) if load is in the situations list.

For non-compiled-top-level forms, only the eval situation is relevant. (This includes forms executed by the interpreter, forms compiled with byte-compile rather than byte-compile-file, and non-top-level forms.) The eval-when acts like a progn if eval is specified, and like nil (ignoring the body forms) if not.

The rules become more subtle when eval-whens are nested; consult Steele (second edition) for the gruesome details (and some gruesome examples).

Some simple examples:

;; Top-level forms in foo.el:
(eval-when (compile)           (setq foo1 'bar))
(eval-when (load)              (setq foo2 'bar))
(eval-when (compile load)      (setq foo3 'bar))
(eval-when (eval)              (setq foo4 'bar))
(eval-when (eval compile)      (setq foo5 'bar))
(eval-when (eval load)         (setq foo6 'bar))
(eval-when (eval compile load) (setq foo7 'bar))

When `foo.el' is compiled, these variables will be set during the compilation itself:

foo1  foo3  foo5  foo7      ; `compile'

When `foo.elc' is loaded, these variables will be set:

foo2  foo3  foo6  foo7      ; `load'

And if `foo.el' is loaded uncompiled, these variables will be set:

foo4  foo5  foo6  foo7      ; `eval'

If these seven eval-whens had been, say, inside a defun, then the first three would have been equivalent to nil and the last four would have been equivalent to the corresponding setqs.

Note that (eval-when (load eval) ...) is equivalent to (progn ...) in all contexts. The compiler treats certain top-level forms, like defmacro (sort-of) and require, as if they were wrapped in (eval-when (compile load eval) ...).

Emacs 19 includes two special forms related to eval-when. One of these, eval-when-compile, is not quite equivalent to any eval-when construct and is described below. This package defines a version of eval-when-compile for the benefit of Emacs 18 users.

The other form, (eval-and-compile ...), is exactly equivalent to `(eval-when (compile load eval) ...)' and so is not itself defined by this package.

Special Form: eval-when-compile forms...
The forms are evaluated at compile-time; at execution time, this form acts like a quoted constant of the resulting value. Used at top-level, eval-when-compile is just like `eval-when (compile eval)'. In other contexts, eval-when-compile allows code to be evaluated once at compile-time for efficiency or other reasons.

This form is similar to the `#.' syntax of true Common Lisp.

Special Form: load-time-value form
The form is evaluated at load-time; at execution time, this form acts like a quoted constant of the resulting value.

Early Common Lisp had a `#,' syntax that was similar to this, but ANSI Common Lisp replaced it with load-time-value and gave it more well-defined semantics.

In a compiled file, load-time-value arranges for form to be evaluated when the `.elc' file is loaded and then used as if it were a quoted constant. In code compiled by byte-compile rather than byte-compile-file, the effect is identical to eval-when-compile. In uncompiled code, both eval-when-compile and load-time-value act exactly like progn.

(defun report ()
  (insert "This function was executed on: "
          (current-time-string)
          ", compiled on: "
          (eval-when-compile (current-time-string))
          ;; or '#.(current-time-string) in real Common Lisp
          ", and loaded on: "
          (load-time-value (current-time-string))))

Byte-compiled, the above defun will result in the following code (or its compiled equivalent, of course) in the `.elc' file:

(setq --temp-- (current-time-string))
(defun report ()
  (insert "This function was executed on: "
          (current-time-string)
          ", compiled on: "
          '"Wed Jun 23 18:33:43 1993"
          ", and loaded on: "
          --temp--))

Function Aliases

This section describes a feature from GNU Emacs 19 which this package makes available in other versions of Emacs.

Function: defalias symbol function
This function sets symbol's function cell to function. It is equivalent to fset, except that in GNU Emacs 19 it also records the setting in load-history so that it can be undone by a later unload-feature.

In other versions of Emacs, defalias is a synonym for fset.

Predicates

This section describes functions for testing whether various facts are true or false.

Type Predicates

The CL package defines a version of the Common Lisp typep predicate.

Function: typep object type
Check if object is of type type, where type is a (quoted) type name of the sort used by Common Lisp. For example, (typep foo 'integer) is equivalent to (integerp foo).

The type argument to the above function is either a symbol or a list beginning with a symbol.

The following function and macro (not technically predicates) are related to typep.

Function: coerce object type
This function attempts to convert object to the specified type. If object is already of that type as determined by typep, it is simply returned. Otherwise, certain types of conversions will be made: If type is any sequence type (string, list, etc.) then object will be converted to that type if possible. If type is character, then strings of length one and symbols with one-character names can be coerced. If type is float, then integers can be coerced in versions of Emacs that support floats. In all other circumstances, coerce signals an error.

Special Form: deftype name arglist forms...
This macro defines a new type called name. It is similar to defmacro in many ways; when name is encountered as a type name, the body forms are evaluated and should return a type specifier that is equivalent to the type. The arglist is a Common Lisp argument list of the sort accepted by defmacro*. The type specifier `(name args...)' is expanded by calling the expander with those arguments; the type symbol `name' is expanded by calling the expander with no arguments. The arglist is processed the same as for defmacro* except that optional arguments without explicit defaults use * instead of nil as the "default" default. Some examples:
(deftype null () '(satisfies null))    ; predefined
(deftype list () '(or null cons))      ; predefined
(deftype unsigned-byte (&optional bits)
  (list 'integer 0 (if (eq bits '*) bits (1- (lsh 1 bits)))))
(unsigned-byte 8)  ==  (integer 0 255)
(unsigned-byte)  ==  (integer 0 *)
unsigned-byte  ==  (integer 0 *)

The last example shows how the Common Lisp unsigned-byte type specifier could be implemented if desired; this package does not implement unsigned-byte by default.

The typecase and check-type macros also use type names. See section Conditionals. See section Assertions and Errors. The map, concatenate, and merge functions take type-name arguments to specify the type of sequence to return. See section Sequences.

Equality Predicates

This package defines two Common Lisp predicates, eql and equalp.

Function: eql a b
This function is almost the same as eq, except that if a and b are numbers of the same type, it compares them for numeric equality (as if by equal instead of eq). This makes a difference only for versions of Emacs that are compiled with floating-point support, such as Emacs 19. Emacs floats are allocated objects just like cons cells, which means that (eq 3.0 3.0) will not necessarily be true--if the two 3.0s were allocated separately, the pointers will be different even though the numbers are the same. But (eql 3.0 3.0) will always be true.

The types of the arguments must match, so (eql 3 3.0) is still false.

Note that Emacs integers are "direct" rather than allocated, which basically means (eq 3 3) will always be true. Thus eq and eql behave differently only if floating-point numbers are involved, and are indistinguishable on Emacs versions that don't support floats.

There is a slight inconsistency with Common Lisp in the treatment of positive and negative zeros. Some machines, notably those with IEEE standard arithmetic, represent +0 and -0 as distinct values. Normally this doesn't matter because the standard specifies that (= 0.0 -0.0) should always be true, and this is indeed what Emacs Lisp and Common Lisp do. But the Common Lisp standard states that (eql 0.0 -0.0) and (equal 0.0 -0.0) should be false on IEEE-like machines; Emacs Lisp does not do this, and in fact the only known way to distinguish between the two zeros in Emacs Lisp is to format them and check for a minus sign.

Function: equalp a b
This function is a more flexible version of equal. In particular, it compares strings case-insensitively, and it compares numbers without regard to type (so that (equalp 3 3.0) is true). Vectors and conses are compared recursively. All other objects are compared as if by equal.

This function differs from Common Lisp equalp in several respects. First, Common Lisp's equalp also compares characters case-insensitively, which would be impractical in this package since Emacs does not distinguish between integers and characters. In keeping with the idea that strings are less vector-like in Emacs Lisp, this package's equalp also will not compare strings against vectors of integers. Finally, Common Lisp's equalp compares hash tables without regard to ordering, whereas this package simply compares hash tables in terms of their underlying structure (which means vectors for Lucid Emacs 19 hash tables, or lists for other hash tables).

Also note that the Common Lisp functions member and assoc use eql to compare elements, whereas Emacs Lisp follows the MacLisp tradition and uses equal for these two functions. In Emacs, use member* and assoc* to get functions which use eql for comparisons.

Control Structure

The features described in the following sections implement various advanced control structures, including the powerful setf facility and a number of looping and conditional constructs.

Assignment

The psetq form is just like setq, except that multiple assignments are done in parallel rather than sequentially.

Special Form: psetq [symbol form]...
This special form (actually a macro) is used to assign to several variables simultaneously. Given only one symbol and form, it has the same effect as setq. Given several symbol and form pairs, it evaluates all the forms in advance and then stores the corresponding variables afterwards.
(setq x 2 y 3)
(setq x (+ x y)  y (* x y))
x
     => 5
y                     ; y was computed after x was set.
     => 15
(setq x 2 y 3)
(psetq x (+ x y)  y (* x y))
x
     => 5
y                     ; y was computed before x was set.
     => 6

The simplest use of psetq is (psetq x y y x), which exchanges the values of two variables. (The rotatef form provides an even more convenient way to swap two variables; see section Modify Macros.)

psetq always returns nil.

Generalized Variables

A "generalized variable" or "place form" is one of the many places in Lisp memory where values can be stored. The simplest place form is a regular Lisp variable. But the cars and cdrs of lists, elements of arrays, properties of symbols, and many other locations are also places where Lisp values are stored.

The setf form is like setq, except that it accepts arbitrary place forms on the left side rather than just symbols. For example, (setf (car a) b) sets the car of a to b, doing the same operation as (setcar a b) but without having to remember two separate functions for setting and accessing every type of place.

Generalized variables are analogous to "lvalues" in the C language, where `x = a[i]' gets an element from an array and `a[i] = x' stores an element using the same notation. Just as certain forms like a[i] can be lvalues in C, there is a set of forms that can be generalized variables in Lisp.

Basic Setf

The setf macro is the most basic way to operate on generalized variables.

Special Form: setf [place form]...
This macro evaluates form and stores it in place, which must be a valid generalized variable form. If there are several place and form pairs, the assignments are done sequentially just as with setq. setf returns the value of the last form.

The following Lisp forms will work as generalized variables, and so may legally appear in the place argument of setf:

Using any forms other than these in the place argument to setf will signal an error.

The setf macro takes care to evaluate all subforms in the proper left-to-right order; for example,

(setf (aref vec (incf i)) i)

looks like it will evaluate (incf i) exactly once, before the following access to i; the setf expander will insert temporary variables as necessary to ensure that it does in fact work this way no matter what setf-method is defined for aref. (In this case, aset would be used and no such steps would be necessary since aset takes its arguments in a convenient order.)

However, if the place form is a macro which explicitly evaluates its arguments in an unusual order, this unusual order will be preserved. Adapting an example from Steele, given

(defmacro wrong-order (x y) (list 'aref y x))

the form (setf (wrong-order a b) 17) will evaluate b first, then a, just as in an actual call to wrong-order.

Modify Macros

This package defines a number of other macros besides setf that operate on generalized variables. Many are interesting and useful even when the place is just a variable name.

Special Form: psetf [place form]...
This macro is to setf what psetq is to setq: When several places and forms are involved, the assignments take place in parallel rather than sequentially. Specifically, all subforms are evaluated from left to right, then all the assignments are done (in an undefined order).

Special Form: incf place &optional x
This macro increments the number stored in place by one, or by x if specified. The incremented value is returned. For example, (incf i) is equivalent to (setq i (1+ i)), and (incf (car x) 2) is equivalent to (setcar x (+ (car x) 2)).

Once again, care is taken to preserve the "apparent" order of evaluation. For example,

(incf (aref vec (incf i)))

appears to increment i once, then increment the element of vec addressed by i; this is indeed exactly what it does, which means the above form is not equivalent to the "obvious" expansion,

(setf (aref vec (incf i)) (1+ (aref vec (incf i))))   ; Wrong!

but rather to something more like

(let ((temp (incf i)))
  (setf (aref vec temp) (1+ (aref vec temp))))

Again, all of this is taken care of automatically by incf and the other generalized-variable macros.

As a more Emacs-specific example of incf, the expression (incf (point) n) is essentially equivalent to (forward-char n).

Special Form: decf place &optional x
This macro decrements the number stored in place by one, or by x if specified.

Special Form: pop place
This macro removes and returns the first element of the list stored in place. It is analogous to (prog1 (car place) (setf place (cdr place))), except that it takes care to evaluate all subforms only once.

Special Form: push x place
This macro inserts x at the front of the list stored in place. It is analogous to (setf place (cons x place)), except for evaluation of the subforms.

Special Form: pushnew x place &key :test :test-not :key
This macro inserts x at the front of the list stored in place, but only if x was not eql to any existing element of the list. The optional keyword arguments are interpreted in the same way as for adjoin. See section Lists as Sets.

Special Form: shiftf place... newvalue
This macro shifts the places left by one, shifting in the value of newvalue (which may be any Lisp expression, not just a generalized variable), and returning the value shifted out of the first place. Thus, (shiftf a b c d) is equivalent to
(prog1
    a
  (psetf a b
         b c
         c d))

except that the subforms of a, b, and c are actually evaluated only once each and in the apparent order.

Special Form: rotatef place...
This macro rotates the places left by one in circular fashion. Thus, (rotatef a b c d) is equivalent to
(psetf a b
       b c
       c d
       d a)

except for the evaluation of subforms. rotatef always returns nil. Note that (rotatef a b) conveniently exchanges a and b.

The following macros were invented for this package; they have no analogues in Common Lisp.

Special Form: letf (bindings...) forms...
This macro is analogous to let, but for generalized variables rather than just symbols. Each binding should be of the form (place value); the original contents of the places are saved, the values are stored in them, and then the body forms are executed. Afterwards, the places are set back to their original saved contents. This cleanup happens even if the forms exit irregularly due to a throw or an error.

For example,

(letf (((point) (point-min))
       (a 17))
  ...)

moves "point" in the current buffer to the beginning of the buffer, and also binds a to 17 (as if by a normal let, since a is just a regular variable). After the body exits, a is set back to its original value and point is moved back to its original position.

Note that letf on (point) is not quite like a save-excursion, as the latter effectively saves a marker which tracks insertions and deletions in the buffer. Actually, a letf of (point-marker) is much closer to this behavior. (point and point-marker are equivalent as setf places; each will accept either an integer or a marker as the stored value.)

Since generalized variables look like lists, let's shorthand of using `foo' for `(foo nil)' as a binding would be ambiguous in letf and is not allowed.

However, a binding specifier may be a one-element list `(place)', which is similar to `(place place)'. In other words, the place is not disturbed on entry to the body, and the only effect of the letf is to restore the original value of place afterwards. (The redundant access-and-store suggested by the (place place) example does not actually occur.)

In most cases, the place must have a well-defined value on entry to the letf form. The only exceptions are plain variables and calls to symbol-value and symbol-function. If the symbol is not bound on entry, it is simply made unbound by makunbound or fmakunbound on exit.

Special Form: letf* (bindings...) forms...
This macro is to letf what let* is to let: It does the bindings in sequential rather than parallel order.

Special Form: callf function place args...
This is the "generic" modify macro. It calls function, which should be an unquoted function name, macro name, or lambda. It passes place and args as arguments, and assigns the result back to place. For example, (incf place n) is the same as (callf + place n). Some more examples:
(callf abs my-number)
(callf concat (buffer-name) "<" (int-to-string n) ">")
(callf union happy-people (list joe bob) :test 'same-person)

See section Customizing Setf, for define-modify-macro, a way to create even more concise notations for modify macros. Note again that callf is an extension to standard Common Lisp.

Special Form: callf2 function arg1 place args...
This macro is like callf, except that place is the second argument of function rather than the first. For example, (push x place) is equivalent to (callf2 cons x place).

The callf and callf2 macros serve as building blocks for other macros like incf, pushnew, and define-modify-macro. The letf and letf* macros are used in the processing of symbol macros; see section Macro Bindings.

Customizing Setf

Common Lisp defines three macros, define-modify-macro, defsetf, and define-setf-method, that allow the user to extend generalized variables in various ways.

Special Form: define-modify-macro name arglist function [doc-string]
This macro defines a "read-modify-write" macro similar to incf and decf. The macro name is defined to take a place argument followed by additional arguments described by arglist. The call
(name place args...)

will be expanded to

(callf func place args...)

which in turn is roughly equivalent to

(setf place (func place args...))

For example:

(define-modify-macro incf (&optional (n 1)) +)
(define-modify-macro concatf (&rest args) concat)

Note that &key is not allowed in arglist, but &rest is sufficient to pass keywords on to the function.

Most of the modify macros defined by Common Lisp do not exactly follow the pattern of define-modify-macro. For example, push takes its arguments in the wrong order, and pop is completely irregular. You can define these macros "by hand" using get-setf-method, or consult the source file `cl-macs.el' to see how to use the internal setf building blocks.

Special Form: defsetf access-fn update-fn
This is the simpler of two defsetf forms. Where access-fn is the name of a function which accesses a place, this declares update-fn to be the corresponding store function. From now on,
(setf (access-fn arg1 arg2 arg3) value)

will be expanded to

(update-fn arg1 arg2 arg3 value)

The update-fn is required to be either a true function, or a macro which evaluates its arguments in a function-like way. Also, the update-fn is expected to return value as its result. Otherwise, the above expansion would not obey the rules for the way setf is supposed to behave.

As a special (non-Common-Lisp) extension, a third argument of t to defsetf says that the update-fn's return value is not suitable, so that the above setf should be expanded to something more like

(let ((temp value))
  (update-fn arg1 arg2 arg3 temp)
  temp)

Some examples of the use of defsetf, drawn from the standard suite of setf methods, are:

(defsetf car setcar)
(defsetf symbol-value set)
(defsetf buffer-name rename-buffer t)

Special Form: defsetf access-fn arglist (store-var) forms...
This is the second, more complex, form of defsetf. It is rather like defmacro except for the additional store-var argument. The forms should return a Lisp form which stores the value of store-var into the generalized variable formed by a call to access-fn with arguments described by arglist. The forms may begin with a string which documents the setf method (analogous to the doc string that appears at the front of a function).

For example, the simple form of defsetf is shorthand for

(defsetf access-fn (&rest args) (store)
  (append '(update-fn) args (list store)))

The Lisp form that is returned can access the arguments from arglist and store-var in an unrestricted fashion; macros like setf and incf which invoke this setf-method will insert temporary variables as needed to make sure the apparent order of evaluation is preserved.

Another example drawn from the standard package:

(defsetf nth (n x) (store)
  (list 'setcar (list 'nthcdr n x) store))

Special Form: define-setf-method access-fn arglist forms...
This is the most general way to create new place forms. When a setf to access-fn with arguments described by arglist is expanded, the forms are evaluated and must return a list of five items:
  1. A list of temporary variables.
  2. A list of value forms corresponding to the temporary variables above. The temporary variables will be bound to these value forms as the first step of any operation on the generalized variable.
  3. A list of exactly one store variable (generally obtained from a call to gensym).
  4. A Lisp form which stores the contents of the store variable into the generalized variable, assuming the temporaries have been bound as described above.
  5. A Lisp form which accesses the contents of the generalized variable, assuming the temporaries have been bound.

This is exactly like the Common Lisp macro of the same name, except that the method returns a list of five values rather than the five values themselves, since Emacs Lisp does not support Common Lisp's notion of multiple return values.

Once again, the forms may begin with a documentation string.

A setf-method should be maximally conservative with regard to temporary variables. In the setf-methods generated by defsetf, the second return value is simply the list of arguments in the place form, and the first return value is a list of a corresponding number of temporary variables generated by gensym. Macros like setf and incf which use this setf-method will optimize away most temporaries that turn out to be unnecessary, so there is little reason for the setf-method itself to optimize.

Function: get-setf-method place &optional env
This function returns the setf-method for place, by invoking the definition previously recorded by defsetf or define-setf-method. The result is a list of five values as described above. You can use this function to build your own incf-like modify macros. (Actually, it is better to use the internal functions cl-setf-do-modify and cl-setf-do-store, which are a bit easier to use and which also do a number of optimizations; consult the source code for the incf function for a simple example.)

The argument env specifies the "environment" to be passed on to macroexpand if get-setf-method should need to expand a macro in place. It should come from an &environment argument to the macro or setf-method that called get-setf-method.

See also the source code for the setf-methods for apply and substring, each of which works by calling get-setf-method on a simpler case, then massaging the result in various ways.

Modern Common Lisp defines a second, independent way to specify the setf behavior of a function, namely "setf functions" whose names are lists (setf name) rather than symbols. For example, (defun (setf foo) ...) defines the function that is used when setf is applied to foo. This package does not currently support setf functions. In particular, it is a compile-time error to use setf on a form which has not already been defsetf'd or otherwise declared; in newer Common Lisps, this would not be an error since the function (setf func) might be defined later.

@secno=4

Variable Bindings

These Lisp forms make bindings to variables and function names, analogous to Lisp's built-in let form.

See section Modify Macros, for the letf and letf* forms which are also related to variable bindings.

Dynamic Bindings

The standard let form binds variables whose names are known at compile-time. The progv form provides an easy way to bind variables whose names are computed at run-time.

Special Form: progv symbols values forms...
This form establishes let-style variable bindings on a set of variables computed at run-time. The expressions symbols and values are evaluated, and must return lists of symbols and values, respectively. The symbols are bound to the corresponding values for the duration of the body forms. If values is shorter than symbols, the last few symbols are made unbound (as if by makunbound) inside the body. If symbols is shorter than values, the excess values are ignored.

Lexical Bindings

The CL package defines the following macro which more closely follows the Common Lisp let form:

Special Form: lexical-let (bindings...) forms...
This form is exactly like let except that the bindings it establishes are purely lexical. Lexical bindings are similar to local variables in a language like C: Only the code physically within the body of the lexical-let (after macro expansion) may refer to the bound variables.
(setq a 5)
(defun foo (b) (+ a b))
(let ((a 2)) (foo a))
     => 4
(lexical-let ((a 2)) (foo a))
     => 7

In this example, a regular let binding of a actually makes a temporary change to the global variable a, so foo is able to see the binding of a to 2. But lexical-let actually creates a distinct local variable a for use within its body, without any effect on the global variable of the same name.

The most important use of lexical bindings is to create closures. A closure is a function object that refers to an outside lexical variable. For example:

(defun make-adder (n)
  (lexical-let ((n n))
    (function (lambda (m) (+ n m)))))
(setq add17 (make-adder 17))
(funcall add17 4)
     => 21

The call (make-adder 17) returns a function object which adds 17 to its argument. If let had been used instead of lexical-let, the function object would have referred to the global n, which would have been bound to 17 only during the call to make-adder itself.

(defun make-counter ()
  (lexical-let ((n 0))
    (function* (lambda (&optional (m 1)) (incf n m)))))
(setq count-1 (make-counter))
(funcall count-1 3)
     => 3
(funcall count-1 14)
     => 17
(setq count-2 (make-counter))
(funcall count-2 5)
     => 5
(funcall count-1 2)
     => 19
(funcall count-2)
     => 6

Here we see that each call to make-counter creates a distinct local variable n, which serves as a private counter for the function object that is returned.

Closed-over lexical variables persist until the last reference to them goes away, just like all other Lisp objects. For example, count-2 refers to a function object which refers to an instance of the variable n; this is the only reference to that variable, so after (setq count-2 nil) the garbage collector would be able to delete this instance of n. Of course, if a lexical-let does not actually create any closures, then the lexical variables are free as soon as the lexical-let returns.

Many closures are used only during the extent of the bindings they refer to; these are known as "downward funargs" in Lisp parlance. When a closure is used in this way, regular Emacs Lisp dynamic bindings suffice and will be more efficient than lexical-let closures:

(defun add-to-list (x list)
  (mapcar (function (lambda (y) (+ x y))) list))
(add-to-list 7 '(1 2 5))
     => (8 9 12)

Since this lambda is only used while x is still bound, it is not necessary to make a true closure out of it.

You can use defun or flet inside a lexical-let to create a named closure. If several closures are created in the body of a single lexical-let, they all close over the same instance of the lexical variable.

The lexical-let form is an extension to Common Lisp. In true Common Lisp, all bindings are lexical unless declared otherwise.

Special Form: lexical-let* (bindings...) forms...
This form is just like lexical-let, except that the bindings are made sequentially in the manner of let*.

Function Bindings

These forms make let-like bindings to functions instead of variables.

Special Form: flet (bindings...) forms...
This form establishes let-style bindings on the function cells of symbols rather than on the value cells. Each binding must be a list of the form `(name arglist forms...)', which defines a function exactly as if it were a defun* form. The function name is defined accordingly for the duration of the body of the flet; then the old function definition, or lack thereof, is restored.

While flet in Common Lisp establishes a lexical binding of name, Emacs Lisp flet makes a dynamic binding. The result is that flet affects indirect calls to a function as well as calls directly inside the flet form itself.

You can use flet to disable or modify the behavior of a function in a temporary fashion. This will even work on Emacs primitives, although note that some calls to primitive functions internal to Emacs are made without going through the symbol's function cell, and so will not be affected by flet. For example,

(flet ((message (&rest args) (push args saved-msgs)))
  (do-something))

This code attempts to replace the built-in function message with a function that simply saves the messages in a list rather than displaying them. The original definition of message will be restored after do-something exits. This code will work fine on messages generated by other Lisp code, but messages generated directly inside Emacs will not be caught since they make direct C-language calls to the message routines rather than going through the Lisp message function.

Functions defined by flet may use the full Common Lisp argument notation supported by defun*; also, the function body is enclosed in an implicit block as if by defun*. See section Program Structure.

Special Form: labels (bindings...) forms...
The labels form is like flet, except that it makes lexical bindings of the function names rather than dynamic bindings. (In true Common Lisp, both flet and labels make lexical bindings of slightly different sorts; since Emacs Lisp is dynamically bound by default, it seemed more appropriate for flet also to use dynamic binding. The labels form, with its lexical binding, is fully compatible with Common Lisp.)

Lexical scoping means that all references to the named functions must appear physically within the body of the labels form. References may appear both in the body forms of labels itself, and in the bodies of the functions themselves. Thus, labels can define local recursive functions, or mutually-recursive sets of functions.

A "reference" to a function name is either a call to that function, or a use of its name quoted by quote or function to be passed on to, say, mapcar.

Macro Bindings

These forms create local macros and "symbol macros."

Special Form: macrolet (bindings...) forms...
This form is analogous to flet, but for macros instead of functions. Each binding is a list of the same form as the arguments to defmacro* (i.e., a macro name, argument list, and macro-expander forms). The macro is defined accordingly for use within the body of the macrolet.

Because of the nature of macros, macrolet is lexically scoped even in Emacs Lisp: The macrolet binding will affect only calls that appear physically within the body forms, possibly after expansion of other macros in the body.

Special Form: symbol-macrolet (bindings...) forms...
This form creates symbol macros, which are macros that look like variable references rather than function calls. Each binding is a list `(var expansion)'; any reference to var within the body forms is replaced by expansion.
(setq bar '(5 . 9))
(symbol-macrolet ((foo (car bar)))
  (incf foo))
bar
     => (6 . 9)

A setq of a symbol macro is treated the same as a setf. I.e., (setq foo 4) in the above would be equivalent to (setf foo 4), which in turn expands to (setf (car bar) 4).

Likewise, a let or let* binding a symbol macro is treated like a letf or letf*. This differs from true Common Lisp, where the rules of lexical scoping cause a let binding to shadow a symbol-macrolet binding. In this package, only lexical-let and lexical-let* will shadow a symbol macro.

There is no analogue of defmacro for symbol macros; all symbol macros are local. A typical use of symbol-macrolet is in the expansion of another macro:

(defmacro* my-dolist ((x list) &rest body)
  (let ((var (gensym)))
    (list 'loop 'for var 'on list 'do
          (list* 'symbol-macrolet (list (list x (list 'car var)))
                 body))))

(setq mylist '(1 2 3 4))
(my-dolist (x mylist) (incf x))
mylist
     => (2 3 4 5)

In this example, the my-dolist macro is similar to dolist (see section Iteration) except that the variable x becomes a true reference onto the elements of the list. The my-dolist call shown here expands to

(loop for G1234 on mylist do
      (symbol-macrolet ((x (car G1234)))
        (incf x)))

which in turn expands to

(loop for G1234 on mylist do (incf (car G1234)))

See section Loop Facility, for a description of the loop macro. This package defines a nonstandard in-ref loop clause that works much like my-dolist.

Conditionals

These conditional forms augment Emacs Lisp's simple if, and, or, and cond forms.

Special Form: when test forms...
This is a variant of if where there are no "else" forms, and possibly several "then" forms. In particular,
(when test a b c)

is entirely equivalent to

(if test (progn a b c) nil)

Special Form: unless test forms...
This is a variant of if where there are no "then" forms, and possibly several "else" forms:
(unless test a b c)

is entirely equivalent to

(when (not test) a b c)

Special Form: case keyform clause...
This macro evaluates keyform, then compares it with the key values listed in the various clauses. Whichever clause matches the key is executed; comparison is done by eql. If no clause matches, the case form returns nil. The clauses are of the form
(keylist body-forms...)

where keylist is a list of key values. If there is exactly one value, and it is not a cons cell or the symbol nil or t, then it can be used by itself as a keylist without being enclosed in a list. All key values in the case form must be distinct. The final clauses may use t in place of a keylist to indicate a default clause that should be taken if none of the other clauses match. (The symbol otherwise is also recognized in place of t. To make a clause that matches the actual symbol t, nil, or otherwise, enclose the symbol in a list.)

For example, this expression reads a keystroke, then does one of four things depending on whether it is an `a', a `b', a RET or C-j, or anything else.

(case (read-char)
  (?a (do-a-thing))
  (?b (do-b-thing))
  ((?\r ?\n) (do-ret-thing))
  (t (do-other-thing)))

Special Form: ecase keyform clause...
This macro is just like case, except that if the key does not match any of the clauses, an error is signaled rather than simply returning nil.

Special Form: typecase keyform clause...
This macro is a version of case that checks for types rather than values. Each clause is of the form `(type body...)'. See section Type Predicates, for a description of type specifiers. For example,
(typecase x
  (integer (munch-integer x))
  (float (munch-float x))
  (string (munch-integer (string-to-int x)))
  (t (munch-anything x)))

The type specifier t matches any type of object; the word otherwise is also allowed. To make one clause match any of several types, use an (or ...) type specifier.

Special Form: etypecase keyform clause...
This macro is just like typecase, except that if the key does not match any of the clauses, an error is signaled rather than simply returning nil.

Blocks and Exits

Common Lisp blocks provide a non-local exit mechanism very similar to catch and throw, but lexically rather than dynamically scoped. This package actually implements block in terms of catch; however, the lexical scoping allows the optimizing byte-compiler to omit the costly catch step if the body of the block does not actually return-from the block.

Special Form: block name forms...
The forms are evaluated as if by a progn. However, if any of the forms execute (return-from name), they will jump out and return directly from the block form. The block returns the result of the last form unless a return-from occurs.

The block/return-from mechanism is quite similar to the catch/throw mechanism. The main differences are that block names are unevaluated symbols, rather than forms (such as quoted symbols) which evaluate to a tag at run-time; and also that blocks are lexically scoped whereas catch/throw are dynamically scoped. This means that functions called from the body of a catch can also throw to the catch, but the return-from referring to a block name must appear physically within the forms that make up the body of the block. They may not appear within other called functions, although they may appear within macro expansions or lambdas in the body. Block names and catch names form independent name-spaces.

In true Common Lisp, defun and defmacro surround the function or expander bodies with implicit blocks with the same name as the function or macro. This does not occur in Emacs Lisp, but this package provides defun* and defmacro* forms which do create the implicit block.

The Common Lisp looping constructs defined by this package, such as loop and dolist, also create implicit blocks just as in Common Lisp.

Because they are implemented in terms of Emacs Lisp catch and throw, blocks have the same overhead as actual catch constructs (roughly two function calls). However, Zawinski and Furuseth's optimizing byte compiler (standard in Emacs 19) will optimize away the catch if the block does not in fact contain any return or return-from calls that jump to it. This means that do loops and defun* functions which don't use return don't pay the overhead to support it.

Special Form: return-from name [result]
This macro returns from the block named name, which must be an (unevaluated) symbol. If a result form is specified, it is evaluated to produce the result returned from the block. Otherwise, nil is returned.

Special Form: return [result]
This macro is exactly like (return-from nil result). Common Lisp loops like do and dolist implicitly enclose themselves in nil blocks.

Iteration

The macros described here provide more sophisticated, high-level looping constructs to complement Emacs Lisp's basic while loop.

Special Form: loop forms...
The CL package supports both the simple, old-style meaning of loop and the extremely powerful and flexible feature known as the Loop Facility or Loop Macro. This more advanced facility is discussed in the following section; see section Loop Facility. The simple form of loop is described here.

If loop is followed by zero or more Lisp expressions, then (loop exprs...) simply creates an infinite loop executing the expressions over and over. The loop is enclosed in an implicit nil block. Thus,

(loop (foo)  (if (no-more) (return 72))  (bar))

is exactly equivalent to

(block nil (while t (foo)  (if (no-more) (return 72))  (bar)))

If any of the expressions are plain symbols, the loop is instead interpreted as a Loop Macro specification as described later. (This is not a restriction in practice, since a plain symbol in the above notation would simply access and throw away the value of a variable.)

Special Form: do (spec...) (end-test [result...]) forms...
This macro creates a general iterative loop. Each spec is of the form
(var [init [step]])

The loop works as follows: First, each var is bound to the associated init value as if by a let form. Then, in each iteration of the loop, the end-test is evaluated; if true, the loop is finished. Otherwise, the body forms are evaluated, then each var is set to the associated step expression (as if by a psetq form) and the next iteration begins. Once the end-test becomes true, the result forms are evaluated (with the vars still bound to their values) to produce the result returned by do.

The entire do loop is enclosed in an implicit nil block, so that you can use (return) to break out of the loop at any time.

If there are no result forms, the loop returns nil. If a given var has no step form, it is bound to its init value but not otherwise modified during the do loop (unless the code explicitly modifies it); this case is just a shorthand for putting a (let ((var init)) ...) around the loop. If init is also omitted it defaults to nil, and in this case a plain `var' can be used in place of `(var)', again following the analogy with let.

This example (from Steele) illustrates a loop which applies the function f to successive pairs of values from the lists foo and bar; it is equivalent to the call (mapcar* 'f foo bar). Note that this loop has no body forms at all, performing all its work as side effects of the rest of the loop.

(do ((x foo (cdr x))
     (y bar (cdr y))
     (z nil (cons (f (car x) (car y)) z)))
  ((or (null x) (null y))
   (nreverse z)))

Special Form: do* (spec...) (end-test [result...]) forms...
This is to do what let* is to let. In particular, the initial values are bound as if by let* rather than let, and the steps are assigned as if by setq rather than psetq.

Here is another way to write the above loop:

(do* ((xp foo (cdr xp))
      (yp bar (cdr yp))
      (x (car xp) (car xp))
      (y (car yp) (car yp))
      z)
  ((or (null xp) (null yp))
   (nreverse z))
  (push (f x y) z))

Special Form: dolist (var list [result]) forms...
This is a more specialized loop which iterates across the elements of a list. list should evaluate to a list; the body forms are executed with var bound to each element of the list in turn. Finally, the result form (or nil) is evaluated with var bound to nil to produce the result returned by the loop. The loop is surrounded by an implicit nil block.

Special Form: dotimes (var count [result]) forms...
This is a more specialized loop which iterates a specified number of times. The body is executed with var bound to the integers from zero (inclusive) to count (exclusive), in turn. Then the result form is evaluated with var bound to the total number of iterations that were done (i.e., (max 0 count)) to get the return value for the loop form. The loop is surrounded by an implicit nil block.

Special Form: do-symbols (var [obarray [result]]) forms...
This loop iterates over all interned symbols. If obarray is specified and is not nil, it loops over all symbols in that obarray. For each symbol, the body forms are evaluated with var bound to that symbol. The symbols are visited in an unspecified order. Afterward the result form, if any, is evaluated (with var bound to nil) to get the return value. The loop is surrounded by an implicit nil block.

Special Form: do-all-symbols (var [result]) forms...
This is identical to do-symbols except that the obarray argument is omitted; it always iterates over the default obarray.

See section Mapping over Sequences, for some more functions for iterating over vectors or lists.

Loop Facility

A common complaint with Lisp's traditional looping constructs is that they are either too simple and limited, such as Common Lisp's dotimes or Emacs Lisp's while, or too unreadable and obscure, like Common Lisp's do loop.

To remedy this, recent versions of Common Lisp have added a new construct called the "Loop Facility" or "loop macro," with an easy-to-use but very powerful and expressive syntax.

Loop Basics

The loop macro essentially creates a mini-language within Lisp that is specially tailored for describing loops. While this language is a little strange-looking by the standards of regular Lisp, it turns out to be very easy to learn and well-suited to its purpose.

Since loop is a macro, all parsing of the loop language takes place at byte-compile time; compiled loops are just as efficient as the equivalent while loops written longhand.

Special Form: loop clauses...
A loop construct consists of a series of clauses, each introduced by a symbol like for or do. Clauses are simply strung together in the argument list of loop, with minimal extra parentheses. The various types of clauses specify initializations, such as the binding of temporary variables, actions to be taken in the loop, stepping actions, and final cleanup.

Common Lisp specifies a certain general order of clauses in a loop:

(loop name-clause
      var-clauses...
      action-clauses...)

The name-clause optionally gives a name to the implicit block that surrounds the loop. By default, the implicit block is named nil. The var-clauses specify what variables should be bound during the loop, and how they should be modified or iterated throughout the course of the loop. The action-clauses are things to be done during the loop, such as computing, collecting, and returning values.

The Emacs version of the loop macro is less restrictive about the order of clauses, but things will behave most predictably if you put the variable-binding clauses with, for, and repeat before the action clauses. As in Common Lisp, initially and finally clauses can go anywhere.

Loops generally return nil by default, but you can cause them to return a value by using an accumulation clause like collect, an end-test clause like always, or an explicit return clause to jump out of the implicit block. (Because the loop body is enclosed in an implicit block, you can also use regular Lisp return or return-from to break out of the loop.)

The following sections give some examples of the Loop Macro in action, and describe the particular loop clauses in great detail. Consult the second edition of Steele's Common Lisp, the Language, for additional discussion and examples of the loop macro.

Loop Examples

Before listing the full set of clauses that are allowed, let's look at a few example loops just to get a feel for the loop language.

(loop for buf in (buffer-list)
      collect (buffer-file-name buf))

This loop iterates over all Emacs buffers, using the list returned by buffer-list. For each buffer buf, it calls buffer-file-name and collects the results into a list, which is then returned from the loop construct. The result is a list of the file names of all the buffers in Emacs' memory. The words for, in, and collect are reserved words in the loop language.

(loop repeat 20 do (insert "Yowsa\n"))

This loop inserts the phrase "Yowsa" twenty times in the current buffer.

(loop until (eobp) do (munch-line) (forward-line 1))

This loop calls munch-line on every line until the end of the buffer. If point is already at the end of the buffer, the loop exits immediately.

(loop do (munch-line) until (eobp) do (forward-line 1))

This loop is similar to the above one, except that munch-line is always called at least once.

(loop for x from 1 to 100
      for y = (* x x)
      until (>= y 729)
      finally return (list x (= y 729)))

This more complicated loop searches for a number x whose square is 729. For safety's sake it only examines x values up to 100; dropping the phrase `to 100' would cause the loop to count upwards with no limit. The second for clause defines y to be the square of x within the loop; the expression after the = sign is reevaluated each time through the loop. The until clause gives a condition for terminating the loop, and the finally clause says what to do when the loop finishes. (This particular example was written less concisely than it could have been, just for the sake of illustration.)

Note that even though this loop contains three clauses (two fors and an until) that would have been enough to define loops all by themselves, it still creates a single loop rather than some sort of triple-nested loop. You must explicitly nest your loop constructs if you want nested loops.

For Clauses

Most loops are governed by one or more for clauses. A for clause simultaneously describes variables to be bound, how those variables are to be stepped during the loop, and usually an end condition based on those variables.

The word as is a synonym for the word for. This word is followed by a variable name, then a word like from or across that describes the kind of iteration desired. In Common Lisp, the phrase being the sometimes precedes the type of iteration; in this package both being and the are optional. The word each is a synonym for the, and the word that follows it may be singular or plural: `for x being the elements of y' or `for x being each element of y'. Which form you use is purely a matter of style.

The variable is bound around the loop as if by let:

(setq i 'happy)
(loop for i from 1 to 10 do (do-something-with i))
i
     => happy
for var from expr1 to expr2 by expr3
This type of for clause creates a counting loop. Each of the three sub-terms is optional, though there must be at least one term so that the clause is marked as a counting clause. The three expressions are the starting value, the ending value, and the step value, respectively, of the variable. The loop counts upwards by default (expr3 must be positive), from expr1 to expr2 inclusively. If you omit the from term, the loop counts from zero; if you omit the to term, the loop counts forever without stopping (unless stopped by some other loop clause, of course); if you omit the by term, the loop counts in steps of one. You can replace the word from with upfrom or downfrom to indicate the direction of the loop. Likewise, you can replace to with upto or downto. For example, `for x from 5 downto 1' executes five times with x taking on the integers from 5 down to 1 in turn. Also, you can replace to with below or above, which are like upto and downto respectively except that they are exclusive rather than inclusive limits:
(loop for x to 10 collect x)
     => (0 1 2 3 4 5 6 7 8 9 10)
(loop for x below 10 collect x)
     => (0 1 2 3 4 5 6 7 8 9)
The by value is always positive, even for downward-counting loops. Some sort of from value is required for downward loops; `for x downto 5' is not a legal loop clause all by itself.
for var in list by function
This clause iterates var over all the elements of list, in turn. If you specify the by term, then function is used to traverse the list instead of cdr; it must be a function taking one argument. For example:
(loop for x in '(1 2 3 4 5 6) collect (* x x))
     => (1 4 9 16 25 36)
(loop for x in '(1 2 3 4 5 6) by 'cddr collect (* x x))
     => (1 9 25)
for var on list by function
This clause iterates var over all the cons cells of list.
(loop for x on '(1 2 3 4) collect x)
     => ((1 2 3 4) (2 3 4) (3 4) (4))
With by, there is no real reason that the on expression must be a list. For example:
(loop for x on first-animal by 'next-animal collect x)
where (next-animal x) takes an "animal" x and returns the next in the (assumed) sequence of animals, or nil if x was the last animal in the sequence.
for var in-ref list by function
This is like a regular in clause, but var becomes a setf-able "reference" onto the elements of the list rather than just a temporary variable. For example,
(loop for x in-ref my-list do (incf x))
increments every element of my-list in place. This clause is an extension to standard Common Lisp.
for var across array
This clause iterates var over all the elements of array, which may be a vector or a string.
(loop for x across "aeiou"
      do (use-vowel (char-to-string x)))
for var across-ref array
This clause iterates over an array, with var a setf-able reference onto the elements; see in-ref above.
for var being the elements of sequence
This clause iterates over the elements of sequence, which may be a list, vector, or string. Since the type must be determined at run-time, this is somewhat less efficient than in or across. The clause may be followed by the additional term `using (index var2)' to cause var2 to be bound to the successive indices (starting at 0) of the elements. This clause type is taken from older versions of the loop macro, and is not present in modern Common Lisp. The `using (sequence ...)' term of the older macros is not supported.
for var being the elements of-ref sequence
This clause iterates over a sequence, with var a setf-able reference onto the elements; see in-ref above.
for var being the symbols [of obarray]
This clause iterates over symbols, either over all interned symbols or over all symbols in obarray. The loop is executed with var bound to each symbol in turn. The symbols are visited in an unspecified order. As an example,
(loop for sym being the symbols
      when (fboundp sym)
      when (string-match "^map" (symbol-name sym))
      collect sym)
returns a list of all the functions whose names begin with `map'. The Common Lisp words external-symbols and present-symbols are also recognized but are equivalent to symbols in Emacs Lisp. Due to a minor implementation restriction, it will not work to have more than one for clause iterating over symbols, hash tables, keymaps, overlays, or intervals in a given loop. Fortunately, it would rarely if ever be useful to do so. It is legal to mix one of these types of clauses with other clauses like for ... to or while.
for var being the hash-keys of hash-table
This clause iterates over the entries in hash-table. For each hash table entry, var is bound to the entry's key. If you write `the hash-values' instead, var is bound to the values of the entries. The clause may be followed by the additional term `using (hash-values var2)' (where hash-values is the opposite word of the word following the) to cause var and var2 to be bound to the two parts of each hash table entry.
for var being the key-codes of keymap
This clause iterates over the entries in keymap. In GNU Emacs 18 and 19, keymaps are either alists or vectors, and key-codes are integers or symbols. In Lucid Emacs 19, keymaps are a special new data type, and key-codes are symbols or lists of symbols. The iteration does not enter nested keymaps or inherited (parent) keymaps. You can use `the key-bindings' to access the commands bound to the keys rather than the key codes, and you can add a using clause to access both the codes and the bindings together.
for var being the key-seqs of keymap
This clause iterates over all key sequences defined by keymap and its nested keymaps, where var takes on values which are strings in Emacs 18 or vectors in Emacs 19. The strings or vectors are reused for each iteration, so you must copy them if you wish to keep them permanently. You can add a `using (key-bindings ...)' clause to get the command bindings as well.
for var being the overlays [of buffer] ...
This clause iterates over the Emacs 19 "overlays" or Lucid Emacs "extents" of a buffer (the clause extents is synonymous with overlays). Under Emacs 18, this clause iterates zero times. If the of term is omitted, the current buffer is used. This clause also accepts optional `from pos' and `to pos' terms, limiting the clause to overlays which overlap the specified region.
for var being the intervals [of buffer] ...
This clause iterates over all intervals of a buffer with constant text properties. The variable var will be bound to conses of start and end positions, where one start position is always equal to the previous end position. The clause allows of, from, to, and property terms, where the latter term restricts the search to just the specified property. The of term may specify either a buffer or a string. This clause is useful only in GNU Emacs 19; in other versions, all buffers and strings consist of a single interval.
for var being the frames
This clause iterates over all frames, i.e., X window system windows open on Emacs files. This clause works only under Emacs 19. The clause screens is a synonym for frames. The frames are visited in next-frame order starting from selected-frame.
for var being the windows [of frame]
This clause iterates over the windows (in the Emacs sense) of the current frame, or of the specified frame. (In Emacs 18 there is only ever one frame, and the of term is not allowed there.)
for var being the buffers
This clause iterates over all buffers in Emacs. It is equivalent to `for var in (buffer-list)'.
for var = expr1 then expr2
This clause does a general iteration. The first time through the loop, var will be bound to expr1. On the second and successive iterations it will be set by evaluating expr2 (which may refer to the old value of var). For example, these two loops are effectively the same:
(loop for x on my-list by 'cddr do ...)
(loop for x = my-list then (cddr x) while x do ...)
Note that this type of for clause does not imply any sort of terminating condition; the above example combines it with a while clause to tell when to end the loop. If you omit the then term, expr1 is used both for the initial setting and for successive settings:
(loop for x = (random) when (> x 0) return x)
This loop keeps taking random numbers from the (random) function until it gets a positive one, which it then returns.

If you include several for clauses in a row, they are treated sequentially (as if by let* and setq). You can instead use the word and to link the clauses, in which case they are processed in parallel (as if by let and psetq).

(loop for x below 5 for y = nil then x collect (list x y))
     => ((0 nil) (1 1) (2 2) (3 3) (4 4))
(loop for x below 5 and y = nil then x collect (list x y))
     => ((0 nil) (1 0) (2 1) (3 2) (4 3))

In the first loop, y is set based on the value of x that was just set by the previous clause; in the second loop, x and y are set simultaneously so y is set based on the value of x left over from the previous time through the loop.

Another feature of the loop macro is destructuring, similar in concept to the destructuring provided by defmacro. The var part of any for clause can be given as a list of variables instead of a single variable. The values produced during loop execution must be lists; the values in the lists are stored in the corresponding variables.

(loop for (x y) in '((2 3) (4 5) (6 7)) collect (+ x y))
     => (5 9 13)

In loop destructuring, if there are more values than variables the trailing values are ignored, and if there are more variables than values the trailing variables get the value nil. If nil is used as a variable name, the corresponding values are ignored. Destructuring may be nested, and dotted lists of variables like (x . y) are allowed.

Iteration Clauses

Aside from for clauses, there are several other loop clauses that control the way the loop operates. They might be used by themselves, or in conjunction with one or more for clauses.

repeat integer
This clause simply counts up to the specified number using an internal temporary variable. The loops
(loop repeat n do ...)
(loop for temp to n do ...)
are identical except that the second one forces you to choose a name for a variable you aren't actually going to use.
while condition
This clause stops the loop when the specified condition (any Lisp expression) becomes nil. For example, the following two loops are equivalent, except for the implicit nil block that surrounds the second one:
(while cond forms...)
(loop while cond do forms...)
until condition
This clause stops the loop when the specified condition is true, i.e., non-nil.
always condition
This clause stops the loop when the specified condition is nil. Unlike while, it stops the loop using return nil so that the finally clauses are not executed. If all the conditions were non-nil, the loop returns t:
(if (loop for size in size-list always (> size 10))
    (some-big-sizes)
  (no-big-sizes))
never condition
This clause is like always, except that the loop returns t if any conditions were false, or nil otherwise.
thereis condition
This clause stops the loop when the specified form is non-nil; in this case, it returns that non-nil value. If all the values were nil, the loop returns nil.

Accumulation Clauses

These clauses cause the loop to accumulate information about the specified Lisp form. The accumulated result is returned from the loop unless overridden, say, by a return clause.

collect form
This clause collects the values of form into a list. Several examples of collect appear elsewhere in this manual. The word collecting is a synonym for collect, and likewise for the other accumulation clauses.
append form
This clause collects lists of values into a result list using append.
nconc form
This clause collects lists of values into a result list by destructively modifying the lists rather than copying them.
concat form
This clause concatenates the values of the specified form into a string. (It and the following clause are extensions to standard Common Lisp.)
vconcat form
This clause concatenates the values of the specified form into a vector.
count form
This clause counts the number of times the specified form evaluates to a non-nil value.
sum form
This clause accumulates the sum of the values of the specified form, which must evaluate to a number.
maximize form
This clause accumulates the maximum value of the specified form, which must evaluate to a number. The return value is undefined if maximize is executed zero times.
minimize form
This clause accumulates the minimum value of the specified form.

Accumulation clauses can be followed by `into var' to cause the data to be collected into variable var (which is automatically let-bound during the loop) rather than an unnamed temporary variable. Also, into accumulations do not automatically imply a return value. The loop must use some explicit mechanism, such as finally return, to return the accumulated result.

It is legal for several accumulation clauses of the same type to accumulate into the same place. From Steele:

(loop for name in '(fred sue alice joe june)
      for kids in '((bob ken) () () (kris sunshine) ())
      collect name
      append kids)
     => (fred bob ken sue alice joe kris sunshine june)

Other Clauses

This section describes the remaining loop clauses.

with var = value
This clause binds a variable to a value around the loop, but otherwise leaves the variable alone during the loop. The following loops are basically equivalent:
(loop with x = 17 do ...)
(let ((x 17)) (loop do ...))
(loop for x = 17 then x do ...)
Naturally, the variable var might be used for some purpose in the rest of the loop. For example:
(loop for x in my-list  with res = nil  do (push x res)
      finally return res)
This loop inserts the elements of my-list at the front of a new list being accumulated in res, then returns the list res at the end of the loop. The effect is similar to that of a collect clause, but the list gets reversed by virtue of the fact that elements are being pushed onto the front of res rather than the end. If you omit the = term, the variable is initialized to nil. (Thus the `= nil' in the above example is unnecessary.) Bindings made by with are sequential by default, as if by let*. Just like for clauses, with clauses can be linked with and to cause the bindings to be made by let instead.
if condition clause
This clause executes the following loop clause only if the specified condition is true. The following clause should be an accumulation, do, return, if, or unless clause. Several clauses may be linked by separating them with and. These clauses may be followed by else and a clause or clauses to execute if the condition was false. The whole construct may optionally be followed by the word end (which may be used to disambiguate an else or and in a nested if). The actual non-nil value of the condition form is available by the name it in the "then" part. For example:
(setq funny-numbers '(6 13 -1))
     => (6 13 -1)
(loop for x below 10
      if (oddp x)
        collect x into odds
        and if (memq x funny-numbers) return (cdr it) end
      else
        collect x into evens
      finally return (vector odds evens))
     => [(1 3 5 7 9) (0 2 4 6 8)]
(setq funny-numbers '(6 7 13 -1))
     => (6 7 13 -1)
(loop <same thing again>)
     => (13 -1)
Note the use of and to put two clauses into the "then" part, one of which is itself an if clause. Note also that end, while normally optional, was necessary here to make it clear that the else refers to the outermost if clause. In the first case, the loop returns a vector of lists of the odd and even values of x. In the second case, the odd number 7 is one of the funny-numbers so the loop returns early; the actual returned value is based on the result of the memq call.
when condition clause
This clause is just a synonym for if.
unless condition clause
The unless clause is just like if except that the sense of the condition is reversed.
named name
This clause gives a name other than nil to the implicit block surrounding the loop. The name is the symbol to be used as the block name.
initially [do] forms...
This keyword introduces one or more Lisp forms which will be executed before the loop itself begins (but after any variables requested by for or with have been bound to their initial values). initially clauses can appear anywhere; if there are several, they are executed in the order they appear in the loop. The keyword do is optional.
finally [do] forms...
This introduces Lisp forms which will be executed after the loop finishes (say, on request of a for or while). initially and finally clauses may appear anywhere in the loop construct, but they are executed (in the specified order) at the beginning or end, respectively, of the loop.
finally return form
This says that form should be executed after the loop is done to obtain a return value. (Without this, or some other clause like collect or return, the loop will simply return nil.) Variables bound by for, with, or into will still contain their final values when form is executed.
do forms...
The word do may be followed by any number of Lisp expressions which are executed as an implicit progn in the body of the loop. Many of the examples in this section illustrate the use of do.
return form
This clause causes the loop to return immediately. The following Lisp form is evaluated to give the return value of the loop form. The finally clauses, if any, are not executed. Of course, return is generally used inside an if or unless, as its use in a top-level loop clause would mean the loop would never get to "loop" more than once. The clause `return form' is equivalent to `do (return form)' (or return-from if the loop was named). The return clause is implemented a bit more efficiently, though.

While there is no high-level way to add user extensions to loop (comparable to defsetf for setf, say), this package does offer two properties called cl-loop-handler and cl-loop-for-handler which are functions to be called when a given symbol is encountered as a top-level loop clause or for clause, respectively. Consult the source code in file `cl-macs.el' for details.

This package's loop macro is compatible with that of Common Lisp, except that a few features are not implemented: loop-finish and data-type specifiers. Naturally, the for clauses which iterate over keymaps, overlays, intervals, frames, windows, and buffers are Emacs-specific extensions.

Multiple Values

Common Lisp functions can return zero or more results. Emacs Lisp functions, by contrast, always return exactly one result. This package makes no attempt to emulate Common Lisp multiple return values; Emacs versions of Common Lisp functions that return more than one value either return just the first value (as in compiler-macroexpand) or return a list of values (as in get-setf-method). This package does define placeholders for the Common Lisp functions that work with multiple values, but in Emacs Lisp these functions simply operate on lists instead. The values form, for example, is a synonym for list in Emacs.

Special Form: multiple-value-bind (var...) values-form forms...
This form evaluates values-form, which must return a list of values. It then binds the vars to these respective values, as if by let, and then executes the body forms. If there are more vars than values, the extra vars are bound to nil. If there are fewer vars than values, the excess values are ignored.

Special Form: multiple-value-setq (var...) form
This form evaluates form, which must return a list of values. It then sets the vars to these respective values, as if by setq. Extra vars or values are treated the same as in multiple-value-bind.

The older Quiroz package attempted a more faithful (but still imperfect) emulation of Common Lisp multiple values. The old method "usually" simulated true multiple values quite well, but under certain circumstances would leave spurious return values in memory where a later, unrelated multiple-value-bind form would see them.

Since a perfect emulation is not feasible in Emacs Lisp, this package opts to keep it as simple and predictable as possible.

Macros

This package implements the various Common Lisp features of defmacro, such as destructuring, &environment, and &body. Top-level &whole is not implemented for defmacro due to technical difficulties. See section Argument Lists.

Destructuring is made available to the user by way of the following macro:

Special Form: destructuring-bind arglist expr forms...
This macro expands to code which executes forms, with the variables in arglist bound to the list of values returned by expr. The arglist can include all the features allowed for defmacro argument lists, including destructuring. (The &environment keyword is not allowed.) The macro expansion will signal an error if expr returns a list of the wrong number of arguments or with incorrect keyword arguments.

This package also includes the Common Lisp define-compiler-macro facility, which allows you to define compile-time expansions and optimizations for your functions.

Special Form: define-compiler-macro name arglist forms...
This form is similar to defmacro, except that it only expands calls to name at compile-time; calls processed by the Lisp interpreter are not expanded, nor are they expanded by the macroexpand function.

The argument list may begin with a &whole keyword and a variable. This variable is bound to the macro-call form itself, i.e., to a list of the form `(name args...)'. If the macro expander returns this form unchanged, then the compiler treats it as a normal function call. This allows compiler macros to work as optimizers for special cases of a function, leaving complicated cases alone.

For example, here is a simplified version of a definition that appears as a standard part of this package:

(define-compiler-macro member* (&whole form a list &rest keys)
  (if (and (null keys)
           (eq (car-safe a) 'quote)
           (not (floatp-safe (cadr a))))
      (list 'memq a list)
    form))

This definition causes (member* a list) to change to a call to the faster memq in the common case where a is a non-floating-point constant; if a is anything else, or if there are any keyword arguments in the call, then the original member* call is left intact. (The actual compiler macro for member* optimizes a number of other cases, including common :test predicates.)

Function: compiler-macroexpand form
This function is analogous to macroexpand, except that it expands compiler macros rather than regular macros. It returns form unchanged if it is not a call to a function for which a compiler macro has been defined, or if that compiler macro decided to punt by returning its &whole argument. Like macroexpand, it expands repeatedly until it reaches a form for which no further expansion is possible.

See section Macro Bindings, for descriptions of the macrolet and symbol-macrolet forms for making "local" macro definitions.

Declarations

Common Lisp includes a complex and powerful "declaration" mechanism that allows you to give the compiler special hints about the types of data that will be stored in particular variables, and about the ways those variables and functions will be used. This package defines versions of all the Common Lisp declaration forms: declare, locally, proclaim, declaim, and the.

Most of the Common Lisp declarations are not currently useful in Emacs Lisp, as the byte-code system provides little opportunity to benefit from type information, and special declarations are redundant in a fully dynamically-scoped Lisp. A few declarations are meaningful when the optimizing Emacs 19 byte compiler is being used, however. Under the earlier non-optimizing compiler, these declarations will effectively be ignored.

Function: proclaim decl-spec
This function records a "global" declaration specified by decl-spec. Since proclaim is a function, decl-spec is evaluated and thus should normally be quoted.

Special Form: declaim decl-specs...
This macro is like proclaim, except that it takes any number of decl-spec arguments, and the arguments are unevaluated and unquoted. The declaim macro also puts an (eval-when (compile load eval) ...) around the declarations so that they will be registered at compile-time as well as at run-time. (This is vital, since normally the declarations are meant to influence the way the compiler treats the rest of the file that contains the declaim form.)

Special Form: declare decl-specs...
This macro is used to make declarations within functions and other code. Common Lisp allows declarations in various locations, generally at the beginning of any of the many "implicit progns" throughout Lisp syntax, such as function bodies, let bodies, etc. Currently the only declaration understood by declare is special.

Special Form: locally declarations... forms...
In this package, locally is no different from progn.

Special Form: the type form
Type information provided by the is ignored in this package; in other words, (the type form) is equivalent to form. Future versions of the optimizing byte-compiler may make use of this information.

For example, mapcar can map over both lists and arrays. It is hard for the compiler to expand mapcar into an in-line loop unless it knows whether the sequence will be a list or an array ahead of time. With (mapcar 'car (the vector foo)), a future compiler would have enough information to expand the loop in-line. For now, Emacs Lisp will treat the above code as exactly equivalent to (mapcar 'car foo).

Each decl-spec in a proclaim, declaim, or declare should be a list beginning with a symbol that says what kind of declaration it is. This package currently understands special, inline, notinline, optimize, and warn declarations. (The warn declaration is an extension of standard Common Lisp.) Other Common Lisp declarations, such as type and ftype, are silently ignored.

special
Since all variables in Emacs Lisp are "special" (in the Common Lisp sense), special declarations are only advisory. They simply tell the optimizing byte compiler that the specified variables are intentionally being referred to without being bound in the body of the function. The compiler normally emits warnings for such references, since they could be typographical errors for references to local variables. The declaration (declare (special var1 var2)) is equivalent to (defvar var1) (defvar var2) in the optimizing compiler, or to nothing at all in older compilers (which do not warn for non-local references). In top-level contexts, it is generally better to write (defvar var) than (declaim (special var)), since defvar makes your intentions clearer. But the older byte compilers can not handle defvars appearing inside of functions, while (declare (special var)) takes care to work correctly with all compilers.
inline
The inline decl-spec lists one or more functions whose bodies should be expanded "in-line" into calling functions whenever the compiler is able to arrange for it. For example, the Common Lisp function cadr is declared inline by this package so that the form (cadr x) will expand directly into (car (cdr x)) when it is called in user functions, for a savings of one (relatively expensive) function call. The following declarations are all equivalent. Note that the defsubst form is a convenient way to define a function and declare it inline all at once, but it is available only in Emacs 19.
(declaim (inline foo bar))
(eval-when (compile load eval) (proclaim '(inline foo bar)))
(proclaim-inline foo bar)      ; Lucid Emacs only
(defsubst foo (...) ...)       ; instead of defun; Emacs 19 only
Note: This declaration remains in effect after the containing source file is done. It is correct to use it to request that a function you have defined should be inlined, but it is impolite to use it to request inlining of an external function. In Common Lisp, it is possible to use (declare (inline ...)) before a particular call to a function to cause just that call to be inlined; the current byte compilers provide no way to implement this, so (declare (inline ...)) is currently ignored by this package.
notinline
The notinline declaration lists functions which should not be inlined after all; it cancels a previous inline declaration.
optimize
This declaration controls how much optimization is performed by the compiler. Naturally, it is ignored by the earlier non-optimizing compilers. The word optimize is followed by any number of lists like (speed 3) or (safety 2). Common Lisp defines several optimization "qualities"; this package ignores all but speed and safety. The value of a quality should be an integer from 0 to 3, with 0 meaning "unimportant" and 3 meaning "very important." The default level for both qualities is 1. In this package, with the Emacs 19 optimizing compiler, the speed quality is tied to the byte-compile-optimize flag, which is set to nil for (speed 0) and to t for higher settings; and the safety quality is tied to the byte-compile-delete-errors flag, which is set to t for (safety 3) and to nil for all lower settings. (The latter flag controls whether the compiler is allowed to optimize out code whose only side-effect could be to signal an error, e.g., rewriting (progn foo bar) to bar when it is not known whether foo will be bound at run-time.) Note that even compiling with (safety 0), the Emacs byte-code system provides sufficient checking to prevent real harm from being done. For example, barring serious bugs in Emacs itself, Emacs will not crash with a segmentation fault just because of an error in a fully-optimized Lisp program. The optimize declaration is normally used in a top-level proclaim or declaim in a file; Common Lisp allows it to be used with declare to set the level of optimization locally for a given form, but this will not work correctly with the current version of the optimizing compiler. (The declare will set the new optimization level, but that level will not automatically be unset after the enclosing form is done.)
warn
This declaration controls what sorts of warnings are generated by the byte compiler. Again, only the optimizing compiler generates warnings. The word warn is followed by any number of "warning qualities," similar in form to optimization qualities. The currently supported warning types are redefine, callargs, unresolved, and free-vars; in the current system, a value of 0 will disable these warnings and any higher value will enable them. See the documentation for the optimizing byte compiler for details.

Symbols

This package defines several symbol-related features that were missing from Emacs Lisp.

Property Lists

These functions augment the standard Emacs Lisp functions get and put for operating on properties attached to symbols. There are also functions for working with property lists as first-class data structures not attached to particular symbols.

Function: get* symbol property &optional default
This function is like get, except that if the property is not found, the default argument provides the return value. (The Emacs Lisp get function always uses nil as the default; this package's get* is equivalent to Common Lisp's get.)

The get* function is setf-able; when used in this fashion, the default argument is allowed but ignored.

Function: remprop symbol property
This function removes the entry for property from the property list of symbol. It returns a true value if the property was indeed found and removed, or nil if there was no such property. (This function was probably omitted from Emacs originally because, since get did not allow a default, it was very difficult to distinguish between a missing property and a property whose value was nil; thus, setting a property to nil was close enough to remprop for most purposes.)

Function: getf place property &optional default
This function scans the list place as if it were a property list, i.e., a list of alternating property names and values. If an even-numbered element of place is found which is eq to property, the following odd-numbered element is returned. Otherwise, default is returned (or nil if no default is given).

In particular,

(get sym prop)  ==  (getf (symbol-plist sym) prop)

It is legal to use getf as a setf place, in which case its place argument must itself be a legal setf place. The default argument, if any, is ignored in this context. The effect is to change (via setcar) the value cell in the list that corresponds to property, or to cons a new property-value pair onto the list if the property is not yet present.

(put sym prop val)  ==  (setf (getf (symbol-plist sym) prop) val)

The get and get* functions are also setf-able. The fact that default is ignored can sometimes be useful:

(incf (get* 'foo 'usage-count 0))

Here, symbol foo's usage-count property is incremented if it exists, or set to 1 (an incremented 0) otherwise.

When not used as a setf form, getf is just a regular function and its place argument can actually be any Lisp expression.

Special Form: remf place property
This macro removes the property-value pair for property from the property list stored at place, which is any setf-able place expression. It returns true if the property was found. Note that if property happens to be first on the list, this will effectively do a (setf place (cddr place)), whereas if it occurs later, this simply uses setcdr to splice out the property and value cells.

@secno=2

Creating Symbols

These functions create unique symbols, typically for use as temporary variables.

Function: gensym &optional x
This function creates a new, uninterned symbol (using make-symbol) with a unique name. (The name of an uninterned symbol is relevant only if the symbol is printed.) By default, the name is generated from an increasing sequence of numbers, `G1000', `G1001', `G1002', etc. If the optional argument x is a string, that string is used as a prefix instead of `G'. Uninterned symbols are used in macro expansions for temporary variables, to ensure that their names will not conflict with "real" variables in the user's code.

Variable: *gensym-counter*
This variable holds the counter used to generate gensym names. It is incremented after each use by gensym. In Common Lisp this is initialized with 0, but this package initializes it with a random (time-dependent) value to avoid trouble when two files that each used gensym in their compilation are loaded together. (Uninterned symbols become interned when the compiler writes them out to a file and the Emacs loader loads them, so their names have to be treated a bit more carefully than in Common Lisp where uninterned symbols remain uninterned after loading.)

Function: gentemp &optional x
This function is like gensym, except that it produces a new interned symbol. If the symbol that is generated already exists, the function keeps incrementing the counter and trying again until a new symbol is generated.

The Quiroz `cl.el' package also defined a defkeyword form for creating self-quoting keyword symbols. This package automatically creates all keywords that are called for by &key argument specifiers, and discourages the use of keywords as data unrelated to keyword arguments, so the defkeyword form has been discontinued.

@chapno=11

Numbers

This section defines a few simple Common Lisp operations on numbers which were left out of Emacs Lisp.

@secno=1

Predicates on Numbers

These functions return t if the specified condition is true of the numerical argument, or nil otherwise.

Function: plusp number
This predicate tests whether number is positive. It is an error if the argument is not a number.

Function: minusp number
This predicate tests whether number is negative. It is an error if the argument is not a number.

Function: oddp integer
This predicate tests whether integer is odd. It is an error if the argument is not an integer.

Function: evenp integer
This predicate tests whether integer is even. It is an error if the argument is not an integer.

Function: floatp-safe object
This predicate tests whether object is a floating-point number. On systems that support floating-point, this is equivalent to floatp. On other systems, this always returns nil.

@secno=3

Numerical Functions

These functions perform various arithmetic operations on numbers.

Function: abs number
This function returns the absolute value of number. (Newer versions of Emacs provide this as a built-in function; this package defines abs only for Emacs 18 versions which don't provide it as a primitive.)

Function: expt base power
This function returns base raised to the power of number. (Newer versions of Emacs provide this as a built-in function; this package defines expt only for Emacs 18 versions which don't provide it as a primitive.)

Function: gcd &rest integers
This function returns the Greatest Common Divisor of the arguments. For one argument, it returns the absolute value of that argument. For zero arguments, it returns zero.

Function: lcm &rest integers
This function returns the Least Common Multiple of the arguments. For one argument, it returns the absolute value of that argument. For zero arguments, it returns one.

Function: isqrt integer
This function computes the "integer square root" of its integer argument, i.e., the greatest integer less than or equal to the true square root of the argument.

Function: floor* number &optional divisor
This function implements the Common Lisp floor function. It is called floor* to avoid name conflicts with the simpler floor function built-in to Emacs 19.

With one argument, floor* returns a list of two numbers: The argument rounded down (toward minus infinity) to an integer, and the "remainder" which would have to be added back to the first return value to yield the argument again. If the argument is an integer x, the result is always the list (x 0). If the argument is an Emacs 19 floating-point number, the first result is a Lisp integer and the second is a Lisp float between 0 (inclusive) and 1 (exclusive).

With two arguments, floor* divides number by divisor, and returns the floor of the quotient and the corresponding remainder as a list of two numbers. If (floor* x y) returns (q r), then q*y + r = x, with r between 0 (inclusive) and r (exclusive). Also, note that (floor* x) is exactly equivalent to (floor* x 1).

This function is entirely compatible with Common Lisp's floor function, except that it returns the two results in a list since Emacs Lisp does not support multiple-valued functions.

Function: ceiling* number &optional divisor
This function implements the Common Lisp ceiling function, which is analogous to floor except that it rounds the argument or quotient of the arguments up toward plus infinity. The remainder will be between 0 and minus r.

Function: truncate* number &optional divisor
This function implements the Common Lisp truncate function, which is analogous to floor except that it rounds the argument or quotient of the arguments toward zero. Thus it is equivalent to floor* if the argument or quotient is positive, or to ceiling* otherwise. The remainder has the same sign as number.

Function: round* number &optional divisor
This function implements the Common Lisp round function, which is analogous to floor except that it rounds the argument or quotient of the arguments to the nearest integer. In the case of a tie (the argument or quotient is exactly halfway between two integers), it rounds to the even integer.

Function: mod* number divisor
This function returns the same value as the second return value of floor.

Function: rem* number divisor
This function returns the same value as the second return value of truncate.

These definitions are compatible with those in the Quiroz `cl.el' package, except that this package appends `*' to certain function names to avoid conflicts with existing Emacs 19 functions, and that the mechanism for returning multiple values is different.

@secno=8

Random Numbers

This package also provides an implementation of the Common Lisp random number generator. It uses its own additive-congruential algorithm, which is much more likely to give statistically clean random numbers than the simple generators supplied by many operating systems.

Function: random* number &optional state
This function returns a random nonnegative number less than number, and of the same type (either integer or floating-point). The state argument should be a random-state object which holds the state of the random number generator. The function modifies this state object as a side effect. If state is omitted, it defaults to the variable *random-state*, which contains a pre-initialized random-state object.

Variable: *random-state*
This variable contains the system "default" random-state object, used for calls to random* that do not specify an alternative state object. Since any number of programs in the Emacs process may be accessing *random-state* in interleaved fashion, the sequence generated from this variable will be irreproducible for all intents and purposes.

Function: make-random-state &optional state
This function creates or copies a random-state object. If state is omitted or nil, it returns a new copy of *random-state*. This is a copy in the sense that future sequences of calls to (random* n) and (random* n s) (where s is the new random-state object) will return identical sequences of random numbers.

If state is a random-state object, this function returns a copy of that object. If state is t, this function returns a new random-state object seeded from the date and time. As an extension to Common Lisp, state may also be an integer in which case the new object is seeded from that integer; each different integer seed will result in a completely different sequence of random numbers.

It is legal to print a random-state object to a buffer or file and later read it back with read. If a program wishes to use a sequence of pseudo-random numbers which can be reproduced later for debugging, it can call (make-random-state t) to get a new sequence, then print this sequence to a file. When the program is later rerun, it can read the original run's random-state from the file.

Function: random-state-p object
This predicate returns t if object is a random-state object, or nil otherwise.

Implementation Parameters

This package defines several useful constants having to with numbers.

Variable: most-positive-fixnum
This constant equals the largest value a Lisp integer can hold. It is typically 2^23-1 or 2^25-1.

Variable: most-negative-fixnum
This constant equals the smallest (most negative) value a Lisp integer can hold.

The following parameters have to do with floating-point numbers. This package determines their values by exercising the computer's floating-point arithmetic in various ways. Because this operation might be slow, the code for initializing them is kept in a separate function that must be called before the parameters can be used.

Function: cl-float-limits
This function makes sure that the Common Lisp floating-point parameters like most-positive-float have been initialized. Until it is called, these parameters will be nil. If this version of Emacs does not support floats (e.g., most versions of Emacs 18), the parameters will remain nil. If the parameters have already been initialized, the function returns immediately.

The algorithm makes assumptions that will be valid for most modern machines, but will fail if the machine's arithmetic is extremely unusual, e.g., decimal.

Since true Common Lisp supports up to four different floating-point precisions, it has families of constants like most-positive-single-float, most-positive-double-float, most-positive-long-float, and so on. Emacs has only one floating-point precision, so this package omits the precision word from the constants' names.

Variable: most-positive-float
This constant equals the largest value a Lisp float can hold. For those systems whose arithmetic supports infinities, this is the largest finite value. For IEEE machines, the value is approximately 1.79e+308.

Variable: most-negative-float
This constant equals the most-negative value a Lisp float can hold. (It is assumed to be equal to (- most-positive-float).)

Variable: least-positive-float
This constant equals the smallest Lisp float value greater than zero. For IEEE machines, it is about 4.94e-324 if denormals are supported or 2.22e-308 if not.

Variable: least-positive-normalized-float
This constant equals the smallest normalized Lisp float greater than zero, i.e., the smallest value for which IEEE denormalization will not result in a loss of precision. For IEEE machines, this value is about 2.22e-308. For machines that do not support the concept of denormalization and gradual underflow, this constant will always equal least-positive-float.

Variable: least-negative-float
This constant is the negative counterpart of least-positive-float.

Variable: least-negative-normalized-float
This constant is the negative counterpart of least-positive-normalized-float.

Variable: float-epsilon
This constant is the smallest positive Lisp float that can be added to 1.0 to produce a distinct value. Adding a smaller number to 1.0 will yield 1.0 again due to roundoff. For IEEE machines, epsilon is about 2.22e-16.

Variable: float-negative-epsilon
This is the smallest positive value that can be subtracted from 1.0 to produce a distinct value. For IEEE machines, it is about 1.11e-16.

@chapno=13

Sequences

Common Lisp defines a number of functions that operate on sequences, which are either lists, strings, or vectors. Emacs Lisp includes a few of these, notably elt and length; this package defines most of the rest.

Sequence Basics

Many of the sequence functions take keyword arguments; see section Argument Lists. All keyword arguments are optional and, if specified, may appear in any order.

The :key argument should be passed either nil, or a function of one argument. This key function is used as a filter through which the elements of the sequence are seen; for example, (find x y :key 'car) is similar to (assoc* x y): It searches for an element of the list whose car equals x, rather than for an element which equals x itself. If :key is omitted or nil, the filter is effectively the identity function.

The :test and :test-not arguments should be either nil, or functions of two arguments. The test function is used to compare two sequence elements, or to compare a search value with sequence elements. (The two values are passed to the test function in the same order as the original sequence function arguments from which they are derived, or, if they both come from the same sequence, in the same order as they appear in that sequence.) The :test argument specifies a function which must return true (non-nil) to indicate a match; instead, you may use :test-not to give a function which returns false to indicate a match. The default test function is :test 'eql.

Many functions which take item and :test or :test-not arguments also come in -if and -if-not varieties, where a predicate function is passed instead of item, and sequence elements match if the predicate returns true on them (or false in the case of -if-not). For example:

(remove* 0 seq :test '=)  ==  (remove-if 'zerop seq)

to remove all zeros from sequence seq.

Some operations can work on a subsequence of the argument sequence; these function take :start and :end arguments which default to zero and the length of the sequence, respectively. Only elements between start (inclusive) and end (exclusive) are affected by the operation. The end argument may be passed nil to signify the length of the sequence; otherwise, both start and end must be integers, with 0 <= start <= end <= (length seq). If the function takes two sequence arguments, the limits are defined by keywords :start1 and :end1 for the first, and :start2 and :end2 for the second.

A few functions accept a :from-end argument, which, if non-nil, causes the operation to go from right-to-left through the sequence instead of left-to-right, and a :count argument, which specifies an integer maximum number of elements to be removed or otherwise processed.

The sequence functions make no guarantees about the order in which the :test, :test-not, and :key functions are called on various elements. Therefore, it is a bad idea to depend on side effects of these functions. For example, :from-end may cause the sequence to be scanned actually in reverse, or it may be scanned forwards but computing a result "as if" it were scanned backwards. (Some functions, like mapcar* and every, do specify exactly the order in which the function is called so side effects are perfectly acceptable in those cases.)

Strings in GNU Emacs 19 may contain "text properties" as well as character data. Except as noted, it is undefined whether or not text properties are preserved by sequence functions. For example, (remove* ?A str) may or may not preserve the properties of the characters copied from str into the result.

Mapping over Sequences

These functions "map" the function you specify over the elements of lists or arrays. They are all variations on the theme of the built-in function mapcar.

Function: mapcar* function seq &rest more-seqs
This function calls function on successive parallel sets of elements from its argument sequences. Given a single seq argument it is equivalent to mapcar; given n sequences, it calls the function with the first elements of each of the sequences as the n arguments to yield the first element of the result list, then with the second elements, and so on. The mapping stops as soon as the shortest sequence runs out. The argument sequences may be any mixture of lists, strings, and vectors; the return sequence is always a list.

Common Lisp's mapcar accepts multiple arguments but works only on lists; Emacs Lisp's mapcar accepts a single sequence argument. This package's mapcar* works as a compatible superset of both.

Function: map result-type function seq &rest more-seqs
This function maps function over the argument sequences, just like mapcar*, but it returns a sequence of type result-type rather than a list. result-type must be one of the following symbols: vector, string, list (in which case the effect is the same as for mapcar*), or nil (in which case the results are thrown away and map returns nil).

Function: maplist function list &rest more-lists
This function calls function on each of its argument lists, then on the cdrs of those lists, and so on, until the shortest list runs out. The results are returned in the form of a list. Thus, maplist is like mapcar* except that it passes in the list pointers themselves rather than the cars of the advancing pointers.

Function: mapc function seq &rest more-seqs
This function is like mapcar*, except that the values returned by function are ignored and thrown away rather than being collected into a list. The return value of mapc is seq, the first sequence.

Function: mapl function list &rest more-lists
This function is like maplist, except that it throws away the values returned by function.

Function: mapcan function seq &rest more-seqs
This function is like mapcar*, except that it concatenates the return values (which must be lists) using nconc, rather than simply collecting them into a list.

Function: mapcon function list &rest more-lists
This function is like maplist, except that it concatenates the return values using nconc.

Function: some predicate seq &rest more-seqs
This function calls predicate on each element of seq in turn; if predicate returns a non-nil value, some returns that value, otherwise it returns nil. Given several sequence arguments, it steps through the sequences in parallel until the shortest one runs out, just as in mapcar*. You can rely on the left-to-right order in which the elements are visited, and on the fact that mapping stops immediately as soon as predicate returns non-nil.

Function: every predicate seq &rest more-seqs
This function calls predicate on each element of the sequence(s) in turn; it returns nil as soon as predicate returns nil for any element, or t if the predicate was true for all elements.

Function: notany predicate seq &rest more-seqs
This function calls predicate on each element of the sequence(s) in turn; it returns nil as soon as predicate returns a non-nil value for any element, or t if the predicate was nil for all elements.

Function: notevery predicate seq &rest more-seqs
This function calls predicate on each element of the sequence(s) in turn; it returns a non-nil value as soon as predicate returns nil for any element, or t if the predicate was true for all elements.

Function: reduce function seq &key :from-end :start :end :initial-value :key
This function combines the elements of seq using an associative binary operation. Suppose function is * and seq is the list (2 3 4 5). The first two elements of the list are combined with (* 2 3) = 6; this is combined with the next element, (* 6 4) = 24, and that is combined with the final element: (* 24 5) = 120. Note that the * function happens to be self-reducing, so that (* 2 3 4 5) has the same effect as an explicit call to reduce.

If :from-end is true, the reduction is right-associative instead of left-associative:

(reduce '- '(1 2 3 4))
     == (- (- (- 1 2) 3) 4) => -8
(reduce '- '(1 2 3 4) :from-end t)
     == (- 1 (- 2 (- 3 4))) => -2

If :key is specified, it is a function of one argument which is called on each of the sequence elements in turn.

If :initial-value is specified, it is effectively added to the front (or rear in the case of :from-end) of the sequence. The :key function is not applied to the initial value.

If the sequence, including the initial value, has exactly one element then that element is returned without ever calling function. If the sequence is empty (and there is no initial value), then function is called with no arguments to obtain the return value.

All of these mapping operations can be expressed conveniently in terms of the loop macro. In compiled code, loop will be faster since it generates the loop as in-line code with no function calls.

Sequence Functions

This section describes a number of Common Lisp functions for operating on sequences.

Function: subseq sequence start &optional end
This function returns a given subsequence of the argument sequence, which may be a list, string, or vector. The indices start and end must be in range, and start must be no greater than end. If end is omitted, it defaults to the length of the sequence. The return value is always a copy; it does not share structure with sequence.

As an extension to Common Lisp, start and/or end may be negative, in which case they represent a distance back from the end of the sequence. This is for compatibility with Emacs' substring function. Note that subseq is the only sequence function that allows negative start and end.

You can use setf on a subseq form to replace a specified range of elements with elements from another sequence. The replacement is done as if by replace, described below.

Function: concatenate result-type &rest seqs
This function concatenates the argument sequences together to form a result sequence of type result-type, one of the symbols vector, string, or list. The arguments are always copied, even in cases such as (concatenate 'list '(1 2 3)) where the result is identical to an argument.

Function: fill seq item &key :start :end
This function fills the elements of the sequence (or the specified part of the sequence) with the value item.

Function: replace seq1 seq2 &key :start1 :end1 :start2 :end2
This function copies part of seq2 into part of seq1. The sequence seq1 is not stretched or resized; the amount of data copied is simply the shorter of the source and destination (sub)sequences. The function returns seq1.

If seq1 and seq2 are eq, then the replacement will work correctly even if the regions indicated by the start and end arguments overlap. However, if seq1 and seq2 are lists which share storage but are not eq, and the start and end arguments specify overlapping regions, the effect is undefined.

Function: remove* item seq &key :test :test-not :key :count :start :end :from-end
This returns a copy of seq with all elements matching item removed. The result may share storage with or be eq to seq in some circumstances, but the original seq will not be modified. The :test, :test-not, and :key arguments define the matching test that is used; by default, elements eql to item are removed. The :count argument specifies the maximum number of matching elements that can be removed (only the leftmost count matches are removed). The :start and :end arguments specify a region in seq in which elements will be removed; elements outside that region are not matched or removed. The :from-end argument, if true, says that elements should be deleted from the end of the sequence rather than the beginning (this matters only if count was also specified).

Function: delete* item seq &key :test :test-not :key :count :start :end :from-end
This deletes all elements of seq which match item. It is a destructive operation. Since Emacs Lisp does not support stretchable strings or vectors, this is the same as remove* for those sequence types. On lists, remove* will copy the list if necessary to preserve the original list, whereas delete* will splice out parts of the argument list. Compare append and nconc, which are analogous non-destructive and destructive list operations in Emacs Lisp.

The predicate-oriented functions remove-if, remove-if-not, delete-if, and delete-if-not are defined similarly.

Function: delete item list
This MacLisp-compatible function deletes from list all elements which are equal to item. The delete function is built-in to Emacs 19; this package defines it equivalently in Emacs 18.

Function: remove item list
This function removes from list all elements which are equal to item. This package defines it for symmetry with delete, even though remove is not built-in to Emacs 19.

Function: remq item list
This function removes from list all elements which are eq to item. This package defines it for symmetry with delq, even though remq is not built-in to Emacs 19.

Function: remove-duplicates seq &key :test :test-not :key :start :end :from-end
This function returns a copy of seq with duplicate elements removed. Specifically, if two elements from the sequence match according to the :test, :test-not, and :key arguments, only the rightmost one is retained. If :from-end is true, the leftmost one is retained instead. If :start or :end is specified, only elements within that subsequence are examined or removed.

Function: delete-duplicates seq &key :test :test-not :key :start :end :from-end
This function deletes duplicate elements from seq. It is a destructive version of remove-duplicates.

Function: substitute new old seq &key :test :test-not :key :count :start :end :from-end
This function returns a copy of seq, with all elements matching old replaced with new. The :count, :start, :end, and :from-end arguments may be used to limit the number of substitutions made.

Function: nsubstitute new old seq &key :test :test-not :key :count :start :end :from-end
This is a destructive version of substitute; it performs the substitution using setcar or aset rather than by returning a changed copy of the sequence.

The substitute-if, substitute-if-not, nsubstitute-if, and nsubstitute-if-not functions are defined similarly. For these, a predicate is given in place of the old argument.

Searching Sequences

These functions search for elements or subsequences in a sequence. (See also member* and assoc*; see section Lists.)

Function: find item seq &key :test :test-not :key :start :end :from-end
This function searches seq for an element matching item. If it finds a match, it returns the matching element. Otherwise, it returns nil. It returns the leftmost match, unless :from-end is true, in which case it returns the rightmost match. The :start and :end arguments may be used to limit the range of elements that are searched.

Function: position item seq &key :test :test-not :key :start :end :from-end
This function is like find, except that it returns the integer position in the sequence of the matching item rather than the item itself. The position is relative to the start of the sequence as a whole, even if :start is non-zero. The function returns nil if no matching element was found.

Function: count item seq &key :test :test-not :key :start :end
This function returns the number of elements of seq which match item. The result is always a nonnegative integer.

The find-if, find-if-not, position-if, position-if-not, count-if, and count-if-not functions are defined similarly.

Function: mismatch seq1 seq2 &key :test :test-not :key :start1 :end1 :start2 :end2 :from-end
This function compares the specified parts of seq1 and seq2. If they are the same length and the corresponding elements match (according to :test, :test-not, and :key), the function returns nil. If there is a mismatch, the function returns the index (relative to seq1) of the first mismatching element. This will be the leftmost pair of elements which do not match, or the position at which the shorter of the two otherwise-matching sequences runs out.

If :from-end is true, then the elements are compared from right to left starting at (1- end1) and (1- end2). If the sequences differ, then one plus the index of the rightmost difference (relative to seq1) is returned.

An interesting example is (mismatch str1 str2 :key 'upcase), which compares two strings case-insensitively.

Function: search seq1 seq2 &key :test :test-not :key :from-end :start1 :end1 :start2 :end2
This function searches seq2 for a subsequence that matches seq1 (or part of it specified by :start1 and :end1.) Only matches which fall entirely within the region defined by :start2 and :end2 will be considered. The return value is the index of the leftmost element of the leftmost match, relative to the start of seq2, or nil if no matches were found. If :from-end is true, the function finds the rightmost matching subsequence.

Sorting Sequences

Function: sort* seq predicate &key :key
This function sorts seq into increasing order as determined by using predicate to compare pairs of elements. predicate should return true (non-nil) if and only if its first argument is less than (not equal to) its second argument. For example, < and string-lessp are suitable predicate functions for sorting numbers and strings, respectively; > would sort numbers into decreasing rather than increasing order.

This function differs from Emacs' built-in sort in that it can operate on any type of sequence, not just lists. Also, it accepts a :key argument which is used to preprocess data fed to the predicate function. For example,

(setq data (sort data 'string-lessp :key 'downcase))

sorts data, a sequence of strings, into increasing alphabetical order without regard to case. A :key function of car would be useful for sorting association lists.

The sort* function is destructive; it sorts lists by actually rearranging the cdr pointers in suitable fashion.

Function: stable-sort seq predicate &key :key
This function sorts seq stably, meaning two elements which are equal in terms of predicate are guaranteed not to be rearranged out of their original order by the sort.

In practice, sort* and stable-sort are equivalent in Emacs Lisp because the underlying sort function is stable by default. However, this package reserves the right to use non-stable methods for sort* in the future.

Function: merge type seq1 seq2 predicate &key :key
This function merges two sequences seq1 and seq2 by interleaving their elements. The result sequence, of type type (in the sense of concatenate), has length equal to the sum of the lengths of the two input sequences. The sequences may be modified destructively. Order of elements within seq1 and seq2 is preserved in the interleaving; elements of the two sequences are compared by predicate (in the sense of sort) and the lesser element goes first in the result. When elements are equal, those from seq1 precede those from seq2 in the result. Thus, if seq1 and seq2 are both sorted according to predicate, then the result will be a merged sequence which is (stably) sorted according to predicate.

Lists

The functions described here operate on lists.

List Functions

This section describes a number of simple operations on lists, i.e., chains of cons cells.

Function: caddr x
This function is equivalent to (car (cdr (cdr x))). Likewise, this package defines all 28 cxxxr functions where xxx is up to four `a's and/or `d's. All of these functions are setf-able, and calls to them are expanded inline by the byte-compiler for maximum efficiency.

Function: first x
This function is a synonym for (car x). Likewise, the functions second, third, ..., through tenth return the given element of the list x.

Function: rest x
This function is a synonym for (cdr x).

Function: endp x
Common Lisp defines this function to act like null, but signaling an error if x is neither a nil nor a cons cell. This package simply defines endp as a synonym for null.

Function: list-length x
This function returns the length of list x, exactly like (length x), except that if x is a circular list (where the cdr-chain forms a loop rather than terminating with nil), this function returns nil. (The regular length function would get stuck if given a circular list.)

Function: last* x &optional n
This function returns the last cons, or the nth-to-last cons, of the list x. If n is omitted it defaults to 1. The "last cons" means the first cons cell of the list whose cdr is not another cons cell. (For normal lists, the cdr of the last cons will be nil.) This function returns nil if x is nil or shorter than n. Note that the last element of the list is (car (last x)).

The Emacs function last does the same thing except that it does not handle the optional argument n.

Function: butlast x &optional n
This function returns the list x with the last element, or the last n elements, removed. If n is greater than zero it makes a copy of the list so as not to damage the original list. In general, (append (butlast x n) (last x n)) will return a list equal to x.

Function: nbutlast x &optional n
This is a version of butlast that works by destructively modifying the cdr of the appropriate element, rather than making a copy of the list.

Function: list* arg &rest others
This function constructs a list of its arguments. The final argument becomes the cdr of the last cell constructed. Thus, (list* a b c) is equivalent to (cons a (cons b c)), and (list* a b nil) is equivalent to (list a b).

(Note that this function really is called list* in Common Lisp; it is not a name invented for this package like member* or defun*.)

Function: ldiff list sublist
If sublist is a sublist of list, i.e., is eq to one of the cons cells of list, then this function returns a copy of the part of list up to but not including sublist. For example, (ldiff x (cddr x)) returns the first two elements of the list x. The result is a copy; the original list is not modified. If sublist is not a sublist of list, a copy of the entire list is returned.

Function: copy-list list
This function returns a copy of the list list. It copies dotted lists like (1 2 . 3) correctly.

Function: copy-tree x &optional vecp
This function returns a copy of the tree of cons cells x. Unlike copy-sequence (and its alias copy-list), which copies only along the cdr direction, this function copies (recursively) along both the car and the cdr directions. If x is not a cons cell, the function simply returns x unchanged. If the optional vecp argument is true, this function copies vectors (recursively) as well as cons cells.

Function: tree-equal x y &key :test :test-not :key
This function compares two trees of cons cells. If x and y are both cons cells, their cars and cdrs are compared recursively. If neither x nor y is a cons cell, they are compared by eql, or according to the specified test. The :key function, if specified, is applied to the elements of both trees. See section Sequences.

@secno=3

Substitution of Expressions

These functions substitute elements throughout a tree of cons cells. (See section Sequence Functions, for the substitute function, which works on just the top-level elements of a list.)

Function: subst new old tree &key :test :test-not :key
This function substitutes occurrences of old with new in tree, a tree of cons cells. It returns a substituted tree, which will be a copy except that it may share storage with the argument tree in parts where no substitutions occurred. The original tree is not modified. This function recurses on, and compares against old, both cars and cdrs of the component cons cells. If old is itself a cons cell, then matching cells in the tree are substituted as usual without recursively substituting in that cell. Comparisons with old are done according to the specified test (eql by default). The :key function is applied to the elements of the tree but not to old.

Function: nsubst new old tree &key :test :test-not :key
This function is like subst, except that it works by destructive modification (by setcar or setcdr) rather than copying.

The subst-if, subst-if-not, nsubst-if, and nsubst-if-not functions are defined similarly.

Function: sublis alist tree &key :test :test-not :key
This function is like subst, except that it takes an association list alist of old-new pairs. Each element of the tree (after applying the :key function, if any), is compared with the cars of alist; if it matches, it is replaced by the corresponding cdr.

Function: nsublis alist tree &key :test :test-not :key
This is a destructive version of sublis.

Lists as Sets

These functions perform operations on lists which represent sets of elements.

Function: member item list
This MacLisp-compatible function searches list for an element which is equal to item. The member function is built-in to Emacs 19; this package defines it equivalently in Emacs 18. See the following function for a Common-Lisp compatible version.

Function: member* item list &key :test :test-not :key
This function searches list for an element matching item. If a match is found, it returns the cons cell whose car was the matching element. Otherwise, it returns nil. Elements are compared by eql by default; you can use the :test, :test-not, and :key arguments to modify this behavior. See section Sequences.

Note that this function's name is suffixed by `*' to avoid the incompatible member function defined in Emacs 19. (That function uses equal for comparisons; it is equivalent to (member* item list :test 'equal).)

The member-if and member-if-not functions analogously search for elements which satisfy a given predicate.

Function: tailp sublist list
This function returns t if sublist is a sublist of list, i.e., if sublist is eql to list or to any of its cdrs.

Function: adjoin item list &key :test :test-not :key
This function conses item onto the front of list, like (cons item list), but only if item is not already present on the list (as determined by member*). If a :key argument is specified, it is applied to item as well as to the elements of list during the search, on the reasoning that item is "about" to become part of the list.

Function: union list1 list2 &key :test :test-not :key
This function combines two lists which represent sets of items, returning a list that represents the union of those two sets. The result list will contain all items which appear in list1 or list2, and no others. If an item appears in both list1 and list2 it will be copied only once. If an item is duplicated in list1 or list2, it is undefined whether or not that duplication will survive in the result list. The order of elements in the result list is also undefined.

Function: nunion list1 list2 &key :test :test-not :key
This is a destructive version of union; rather than copying, it tries to reuse the storage of the argument lists if possible.

Function: intersection list1 list2 &key :test :test-not :key
This function computes the intersection of the sets represented by list1 and list2. It returns the list of items which appear in both list1 and list2.

Function: nintersection list1 list2 &key :test :test-not :key
This is a destructive version of intersection. It tries to reuse storage of list1 rather than copying. It does not reuse the storage of list2.

Function: set-difference list1 list2 &key :test :test-not :key
This function computes the "set difference" of list1 and list2, i.e., the set of elements that appear in list1 but not in list2.

Function: nset-difference list1 list2 &key :test :test-not :key
This is a destructive set-difference, which will try to reuse list1 if possible.

Function: set-exclusive-or list1 list2 &key :test :test-not :key
This function computes the "set exclusive or" of list1 and list2, i.e., the set of elements that appear in exactly one of list1 and list2.

Function: nset-exclusive-or list1 list2 &key :test :test-not :key
This is a destructive set-exclusive-or, which will try to reuse list1 and list2 if possible.

Function: subsetp list1 list2 &key :test :test-not :key
This function checks whether list1 represents a subset of list2, i.e., whether every element of list1 also appears in list2.

Association Lists

An association list is a list representing a mapping from one set of values to another; any list whose elements are cons cells is an association list.

Function: assoc* item a-list &key :test :test-not :key
This function searches the association list a-list for an element whose car matches (in the sense of :test, :test-not, and :key, or by comparison with eql) a given item. It returns the matching element, if any, otherwise nil. It ignores elements of a-list which are not cons cells. (This corresponds to the behavior of assq and assoc in Emacs Lisp; Common Lisp's assoc ignores nils but considers any other non-cons elements of a-list to be an error.)

Function: rassoc* item a-list &key :test :test-not :key
This function searches for an element whose cdr matches item. If a-list represents a mapping, this applies the inverse of the mapping to item.

Function: rassoc item a-list
This function searches like rassoc* with a :test argument of equal. It is analogous to Emacs Lisp's standard assoc function, which derives from the MacLisp rather than the Common Lisp tradition.

The assoc-if, assoc-if-not, rassoc-if, and rassoc-if-not functions are defined similarly.

Two simple functions for constructing association lists are:

Function: acons key value alist
This is equivalent to (cons (cons key value) alist).

Function: pairlis keys values &optional alist
This is equivalent to (nconc (mapcar* 'cons keys values) alist).

Hash Tables

A hash table is a data structure that maps "keys" onto "values." Keys and values can be arbitrary Lisp data objects. Hash tables have the property that the time to search for a given key is roughly constant; simpler data structures like association lists take time proportional to the number of entries in the list.

Function: make-hash-table &key :test :size
This function creates and returns a hash-table object whose function for comparing elements is :test (eql by default), and which is allocated to fit about :size elements. The :size argument is purely advisory; the table will stretch automatically if you store more elements in it. If :size is omitted, a reasonable default is used.

Common Lisp allows only eq, eql, equal, and equalp as legal values for the :test argument. In this package, any reasonable predicate function will work, though if you use something else you should check the details of the hashing function described below to make sure it is suitable for your predicate.

Some versions of Emacs (like Lucid Emacs 19) include a built-in hash table type; in these versions, make-hash-table with a test of eq will use these built-in hash tables. In all other cases, it will return a hash-table object which takes the form of a list with an identifying "tag" symbol at the front. All of the hash table functions in this package can operate on both types of hash table; normally you will never know which type is being used.

This function accepts the additional Common Lisp keywords :rehash-size and :rehash-threshold, but it ignores their values.

Function: gethash key table &optional default
This function looks up key in table. If key exists in the table, in the sense that it matches any of the existing keys according to the table's test function, then the associated value is returned. Otherwise, default (or nil) is returned.

To store new data in the hash table, use setf on a call to gethash. If key already exists in the table, the corresponding value is changed to the stored value. If key does not already exist, a new entry is added to the table and the table is reallocated to a larger size if necessary. The default argument is allowed but ignored in this case. The situation is exactly analogous to that of get*; see section Property Lists.

Function: remhash key table
This function removes the entry for key from table. If an entry was removed, it returns t. If key does not appear in the table, it does nothing and returns nil.

Function: clrhash table
This function removes all the entries from table, leaving an empty hash table.

Function: maphash function table
This function calls function for each entry in table. It passes two arguments to function, the key and the value of the given entry. The return value of function is ignored; maphash itself returns nil. See section Loop Facility, for an alternate way of iterating over hash tables.

Function: hash-table-count table
This function returns the number of entries in table. Warning: The current implementation of Lucid Emacs 19 hash-tables does not decrement the stored count when remhash removes an entry. Therefore, the return value of this function is not dependable if you have used remhash on the table and the table's test is eq. A slower, but reliable, way to count the entries is (loop for x being the hash-keys of table count t).

Function: hash-table-p object
This function returns t if object is a hash table, nil otherwise. It recognizes both types of hash tables (both Lucid Emacs built-in tables and tables implemented with special lists.)

Sometimes when dealing with hash tables it is useful to know the exact "hash function" that is used. This package implements hash tables using Emacs Lisp "obarrays," which are the same data structure that Emacs Lisp uses to keep track of symbols. Each hash table includes an embedded obarray. Key values given to gethash are converted by various means into strings, which are then looked up in the obarray using intern and intern-soft. The symbol, or "bucket," corresponding to a given key string includes as its symbol-value an association list of all key-value pairs which hash to that string. Depending on the test function, it is possible for many entries to hash to the same bucket. For example, if the test is eql, then the symbol foo and two separately built strings "foo" will create three entries in the same bucket. Search time is linear within buckets, so hash tables will be most effective if you arrange not to store too many things that hash the same.

The following algorithm is used to convert Lisp objects to hash strings:

Thus, for example, searching among many buffer objects in a hash table will devolve to a (still fairly fast) linear-time search through a single bucket, whereas searching for different symbols will be very fast since each symbol will, in general, hash into its own bucket.

The size of the obarray in a hash table is automatically adjusted as the number of elements increases.

As a special case, make-hash-table with a :size argument of 0 or 1 will create a hash-table object that uses a single association list rather than an obarray of many lists. For very small tables this structure will be more efficient since lookup does not require converting the key to a string or looking it up in an obarray. However, such tables are guaranteed to take time proportional to their size to do a search.

@chapno=18

Structures

The Common Lisp structure mechanism provides a general way to define data types similar to C's struct types. A structure is a Lisp object containing some number of slots, each of which can hold any Lisp data object. Functions are provided for accessing and setting the slots, creating or copying structure objects, and recognizing objects of a particular structure type.

In true Common Lisp, each structure type is a new type distinct from all existing Lisp types. Since the underlying Emacs Lisp system provides no way to create new distinct types, this package implements structures as vectors (or lists upon request) with a special "tag" symbol to identify them.

Special Form: defstruct name slots...
The defstruct form defines a new structure type called name, with the specified slots. (The slots may begin with a string which documents the structure type.) In the simplest case, name and each of the slots are symbols. For example,
(defstruct person name age sex)

defines a struct type called person which contains three slots. Given a person object p, you can access those slots by calling (person-name p), (person-age p), and (person-sex p). You can also change these slots by using setf on any of these place forms:

(incf (person-age birthday-boy))

You can create a new person by calling make-person, which takes keyword arguments :name, :age, and :sex to specify the initial values of these slots in the new object. (Omitting any of these arguments leaves the corresponding slot "undefined," according to the Common Lisp standard; in Emacs Lisp, such uninitialized slots are filled with nil.)

Given a person, (copy-person p) makes a new object of the same type whose slots are eq to those of p.

Given any Lisp object x, (person-p x) returns true if x looks like a person, false otherwise. (Again, in Common Lisp this predicate would be exact; in Emacs Lisp the best it can do is verify that x is a vector of the correct length which starts with the correct tag symbol.)

Accessors like person-name normally check their arguments (effectively using person-p) and signal an error if the argument is the wrong type. This check is affected by (optimize (safety ...)) declarations. Safety level 1, the default, uses a somewhat optimized check that will detect all incorrect arguments, but may use an uninformative error message (e.g., "expected a vector" instead of "expected a person"). Safety level 0 omits all checks except as provided by the underlying aref call; safety levels 2 and 3 do rigorous checking that will always print a descriptive error message for incorrect inputs. See section Declarations.

(setq dave (make-person :name "Dave" :sex 'male))
     => [cl-struct-person "Dave" nil male]
(setq other (copy-person dave))
     => [cl-struct-person "Dave" nil male]
(eq dave other)
     => nil
(eq (person-name dave) (person-name other))
     => t
(person-p dave)
     => t
(person-p [1 2 3 4])
     => nil
(person-p "Bogus")
     => nil
(person-p '[cl-struct-person counterfeit person object])
     => t

In general, name is either a name symbol or a list of a name symbol followed by any number of struct options; each slot is either a slot symbol or a list of the form `(slot-name default-value slot-options...)'. The default-value is a Lisp form which is evaluated any time an instance of the structure type is created without specifying that slot's value.

Common Lisp defines several slot options, but the only one implemented in this package is :read-only. A non-nil value for this option means the slot should not be setf-able; the slot's value is determined when the object is created and does not change afterward.

(defstruct person
  (name nil :read-only t)
  age
  (sex 'unknown))

Any slot options other than :read-only are ignored.

For obscure historical reasons, structure options take a different form than slot options. A structure option is either a keyword symbol, or a list beginning with a keyword symbol possibly followed by arguments. (By contrast, slot options are key-value pairs not enclosed in lists.)

(defstruct (person (:constructor create-person)
                   (:type list)
                   :named)
  name age sex)

The following structure options are recognized.

@itemmax=0 in @advance@leftskip-.5leindent
:conc-name
The argument is a symbol whose print name is used as the prefix for the names of slot accessor functions. The default is the name of the struct type followed by a hyphen. The option (:conc-name p-) would change this prefix to p-. Specifying nil as an argument means no prefix, so that the slot names themselves are used to name the accessor functions.
:constructor
In the simple case, this option takes one argument which is an alternate name to use for the constructor function. The default is make-name, e.g., make-person. The above example changes this to create-person. Specifying nil as an argument means that no standard constructor should be generated at all. In the full form of this option, the constructor name is followed by an arbitrary argument list. See section Program Structure, for a description of the format of Common Lisp argument lists. All options, such as &rest and &key, are supported. The argument names should match the slot names; each slot is initialized from the corresponding argument. Slots whose names do not appear in the argument list are initialized based on the default-value in their slot descriptor. Also, &optional and &key arguments which don't specify defaults take their defaults from the slot descriptor. It is legal to include arguments which don't correspond to slot names; these are useful if they are referred to in the defaults for optional, keyword, or &aux arguments which do correspond to slots. You can specify any number of full-format :constructor options on a structure. The default constructor is still generated as well unless you disable it with a simple-format :constructor option.
(defstruct
 (person
  (:constructor nil)   ; no default constructor
  (:constructor new-person (name sex &optional (age 0)))
  (:constructor new-hound (&key (name "Rover")
                                (dog-years 0)
                           &aux (age (* 7 dog-years))
                                (sex 'canine))))
 name age sex)
The first constructor here takes its arguments positionally rather than by keyword. (In official Common Lisp terminology, constructors that work By Order of Arguments instead of by keyword are called "BOA constructors." No, I'm not making this up.) For example, (new-person "Jane" 'female) generates a person whose slots are "Jane", 0, and female, respectively. The second constructor takes two keyword arguments, :name, which initializes the name slot and defaults to "Rover", and :dog-years, which does not itself correspond to a slot but which is used to initialize the age slot. The sex slot is forced to the symbol canine with no syntax for overriding it.
:copier
The argument is an alternate name for the copier function for this type. The default is copy-name. nil means not to generate a copier function. (In this implementation, all copier functions are simply synonyms for copy-sequence.)
:predicate
The argument is an alternate name for the predicate which recognizes objects of this type. The default is name-p. nil means not to generate a predicate function. (If the :type option is used without the :named option, no predicate is ever generated.) In true Common Lisp, typep is always able to recognize a structure object even if :predicate was used. In this package, typep simply looks for a function called typename-p, so it will work for structure types only if they used the default predicate name.
:include
This option implements a very limited form of C++-style inheritance. The argument is the name of another structure type previously created with defstruct. The effect is to cause the new structure type to inherit all of the included structure's slots (plus, of course, any new slots described by this struct's slot descriptors). The new structure is considered a "specialization" of the included one. In fact, the predicate and slot accessors for the included type will also accept objects of the new type. If there are extra arguments to the :include option after the included-structure name, these options are treated as replacement slot descriptors for slots in the included structure, possibly with modified default values. Borrowing an example from Steele:
(defstruct person name (age 0) sex)
     => person
(defstruct (astronaut (:include person (age 45)))
  helmet-size
  (favorite-beverage 'tang))
     => astronaut

(setq joe (make-person :name "Joe"))
     => [cl-struct-person "Joe" 0 nil]
(setq buzz (make-astronaut :name "Buzz"))
     => [cl-struct-astronaut "Buzz" 45 nil nil tang]

(list (person-p joe) (person-p buzz))
     => (t t)
(list (astronaut-p joe) (astronaut-p buzz))
     => (nil t)

(person-name buzz)
     => "Buzz"
(astronaut-name joe)
     => error: "astronaut-name accessing a non-astronaut"
Thus, if astronaut is a specialization of person, then every astronaut is also a person (but not the other way around). Every astronaut includes all the slots of a person, plus extra slots that are specific to astronauts. Operations that work on people (like person-name) work on astronauts just like other people.
:print-function
In full Common Lisp, this option allows you to specify a function which is called to print an instance of the structure type. The Emacs Lisp system offers no hooks into the Lisp printer which would allow for such a feature, so this package simply ignores :print-function.
:type
The argument should be one of the symbols vector or list. This tells which underlying Lisp data type should be used to implement the new structure type. Vectors are used by default, but (:type list) will cause structure objects to be stored as lists instead. The vector representation for structure objects has the advantage that all structure slots can be accessed quickly, although creating vectors is a bit slower in Emacs Lisp. Lists are easier to create, but take a relatively long time accessing the later slots.
:named
This option, which takes no arguments, causes a characteristic "tag" symbol to be stored at the front of the structure object. Using :type without also using :named will result in a structure type stored as plain vectors or lists with no identifying features. The default, if you don't specify :type explicitly, is to use named vectors. Therefore, :named is only useful in conjunction with :type.
(defstruct (person1) name age sex)
(defstruct (person2 (:type list) :named) name age sex)
(defstruct (person3 (:type list)) name age sex)

(setq p1 (make-person1))
     => [cl-struct-person1 nil nil nil]
(setq p2 (make-person2))
     => (person2 nil nil nil)
(setq p3 (make-person3))
     => (nil nil nil)

(person1-p p1)
     => t
(person2-p p2)
     => t
(person3-p p3)
     => error: function person3-p undefined
Since unnamed structures don't have tags, defstruct is not able to make a useful predicate for recognizing them. Also, accessors like person3-name will be generated but they will not be able to do any type checking. The person3-name function, for example, will simply be a synonym for car in this case. By contrast, person2-name is able to verify that its argument is indeed a person2 object before proceeding.
:initial-offset
The argument must be a nonnegative integer. It specifies a number of slots to be left "empty" at the front of the structure. If the structure is named, the tag appears at the specified position in the list or vector; otherwise, the first slot appears at that position. Earlier positions are filled with nil by the constructors and ignored otherwise. If the type :includes another type, then :initial-offset specifies a number of slots to be skipped between the last slot of the included type and the first new slot.

Except as noted, the defstruct facility of this package is entirely compatible with that of Common Lisp.

@chapno=23

Assertions and Errors

This section describes two macros that test assertions, i.e., conditions which must be true if the program is operating correctly. Assertions never add to the behavior of a Lisp program; they simply make "sanity checks" to make sure everything is as it should be.

If the optimization property speed has been set to 3, and safety is less than 3, then the byte-compiler will optimize away the following assertions. Because assertions might be optimized away, it is a bad idea for them to include side-effects.

Special Form: assert test-form [show-args string args...]
This form verifies that test-form is true (i.e., evaluates to a non-nil value). If so, it returns nil. If the test is not satisfied, assert signals an error.

A default error message will be supplied which includes test-form. You can specify a different error message by including a string argument plus optional extra arguments. Those arguments are simply passed to error to signal the error.

If the optional second argument show-args is t instead of nil, then the error message (with or without string) will also include all non-constant arguments of the top-level form. For example:

(assert (> x 10) t "x is too small: %d")

This usage of show-args is an extension to Common Lisp. In true Common Lisp, the second argument gives a list of places which can be setf'd by the user before continuing from the error. Since Emacs Lisp does not support continuable errors, it makes no sense to specify places.

Special Form: check-type form type [string]
This form verifies that form evaluates to a value of type type. If so, it returns nil. If not, check-type signals a wrong-type-argument error. The default error message lists the erroneous value along with type and form themselves. If string is specified, it is included in the error message in place of type. For example:
(check-type x (integer 1 *) "a positive integer")

See section Type Predicates, for a description of the type specifiers that may be used for type.

Note that in Common Lisp, the first argument to check-type must be a place suitable for use by setf, because check-type signals a continuable error that allows the user to modify place.

The following error-related macro is also defined:

Special Form: ignore-errors forms...
This executes forms exactly like a progn, except that errors are ignored during the forms. More precisely, if an error is signaled then ignore-errors immediately aborts execution of the forms and returns nil. If the forms complete successfully, ignore-errors returns the result of the last form.

Efficiency Concerns

Macros

Many of the advanced features of this package, such as defun*, loop, and setf, are implemented as Lisp macros. In byte-compiled code, these complex notations will be expanded into equivalent Lisp code which is simple and efficient. For example, the forms

(incf i n)
(push x (car p))

are expanded at compile-time to the Lisp forms

(setq i (+ i n))
(setcar p (cons x (car p)))

which are the most efficient ways of doing these respective operations in Lisp. Thus, there is no performance penalty for using the more readable incf and push forms in your compiled code.

Interpreted code, on the other hand, must expand these macros every time they are executed. For this reason it is strongly recommended that code making heavy use of macros be compiled. (The features labeled "Special Form" instead of "Function" in this manual are macros.) A loop using incf a hundred times will execute considerably faster if compiled, and will also garbage-collect less because the macro expansion will not have to be generated, used, and thrown away a hundred times.

You can find out how a macro expands by using the cl-prettyexpand function.

Function: cl-prettyexpand form &optional full
This function takes a single Lisp form as an argument and inserts a nicely formatted copy of it in the current buffer (which must be in Lisp mode so that indentation works properly). It also expands all Lisp macros which appear in the form. The easiest way to use this function is to go to the *scratch* buffer and type, say,
(cl-prettyexpand '(loop for x below 10 collect x))

and type C-x C-e immediately after the closing parenthesis; the expansion

(block nil
  (let* ((x 0)
         (G1004 nil))
    (while (< x 10)
      (setq G1004 (cons x G1004))
      (setq x (+ x 1)))
    (nreverse G1004)))

will be inserted into the buffer. (The block macro is expanded differently in the interpreter and compiler, so cl-prettyexpand just leaves it alone. The temporary variable G1004 was created by gensym.)

If the optional argument full is true, then all macros are expanded, including block, eval-when, and compiler macros. Expansion is done as if form were a top-level form in a file being compiled. For example,

(cl-prettyexpand '(pushnew 'x list))
     -| (setq list (adjoin 'x list))
(cl-prettyexpand '(pushnew 'x list) t)
     -| (setq list (if (memq 'x list) list (cons 'x list)))
(cl-prettyexpand '(caddr (member* 'a list)) t)
     -| (car (cdr (cdr (memq 'a list))))

Note that adjoin, caddr, and member* all have built-in compiler macros to optimize them in common cases.

Error Checking

Common Lisp compliance has in general not been sacrificed for the sake of efficiency. A few exceptions have been made for cases where substantial gains were possible at the expense of marginal incompatibility. One example is the use of memq (which is treated very efficiently by the byte-compiler) to scan for keyword arguments; this can become confused in rare cases when keyword symbols are used as both keywords and data values at once. This is extremely unlikely to occur in practical code, and the use of memq allows functions with keyword arguments to be nearly as fast as functions that use &optional arguments.

The Common Lisp standard (as embodied in Steele's book) uses the phrase "it is an error if" to indicate a situation which is not supposed to arise in complying programs; implementations are strongly encouraged but not required to signal an error in these situations. This package sometimes omits such error checking in the interest of compactness and efficiency. For example, do variable specifiers are supposed to be lists of one, two, or three forms; extra forms are ignored by this package rather than signaling a syntax error. The endp function is simply a synonym for null in this package. Functions taking keyword arguments will accept an odd number of arguments, treating the trailing keyword as if it were followed by the value nil.

Argument lists (as processed by defun* and friends) are checked rigorously except for the minor point just mentioned; in particular, keyword arguments are checked for validity, and &allow-other-keys and :allow-other-keys are fully implemented. Keyword validity checking is slightly time consuming (though not too bad in byte-compiled code); you can use &allow-other-keys to omit this check. Functions defined in this package such as find and member* do check their keyword arguments for validity.

Optimizing Compiler

The byte-compiler that comes with Emacs 18 normally fails to expand macros that appear in top-level positions in the file (i.e., outside of defuns or other enclosing forms). This would have disastrous consequences to programs that used such top-level macros as defun*, eval-when, and defstruct. To work around this problem, the CL package patches the Emacs 18 compiler to expand top-level macros. This patch will apply to your own macros, too, if they are used in a top-level context. The patch will not harm versions of the Emacs 18 compiler which have already had a similar patch applied, nor will it affect the optimizing Emacs 19 byte-compiler written by Jamie Zawinski and Hallvard Furuseth. The patch is applied to the byte compiler's code in Emacs' memory, not to the `bytecomp.elc' file stored on disk.

The Emacs 19 compiler (for Emacs 18) is available from various Emacs Lisp archive sites such as archive.cis.ohio-state.edu. Its use is highly recommended; many of the Common Lisp macros emit code which can be improved by optimization. In particular, blocks (whether explicit or implicit in constructs like defun* and loop) carry a fair run-time penalty; the optimizing compiler removes blocks which are not actually referenced by return or return-from inside the block.

Common Lisp Compatibility

Following is a list of all known incompatibilities between this package and Common Lisp as documented in Steele (2nd edition).

Certain function names, such as member, assoc, and floor, were already taken by (incompatible) Emacs Lisp functions; this package appends `*' to the names of its Common Lisp versions of these functions.

The word defun* is required instead of defun in order to use extended Common Lisp argument lists in a function. Likewise, defmacro* and function* are versions of those forms which understand full-featured argument lists. The &whole keyword does not work in defmacro argument lists (except inside recursive argument lists).

In order to allow an efficient implementation, keyword arguments use a slightly cheesy parser which may be confused if a keyword symbol is passed as the value of another keyword argument. (Specifically, (memq :keyword rest-of-arguments) is used to scan for :keyword among the supplied keyword arguments.)

The eql and equal predicates do not distinguish between IEEE floating-point plus and minus zero. The equalp predicate has several differences with Common Lisp; see section Predicates.

The setf mechanism is entirely compatible, except that setf-methods return a list of five values rather than five values directly. Also, the new "setf function" concept (typified by (defun (setf foo) ...)) is not implemented.

The do-all-symbols form is the same as do-symbols with no obarray argument. In Common Lisp, this form would iterate over all symbols in all packages. Since Emacs obarrays are not a first-class package mechanism, there is no way for do-all-symbols to locate any but the default obarray.

The loop macro is complete except that loop-finish and type specifiers are unimplemented.

The multiple-value return facility treats lists as multiple values, since Emacs Lisp cannot support multiple return values directly. The macros will be compatible with Common Lisp if values or values-list is always used to return to a multiple-value-bind or other multiple-value receiver; if values is used without multiple-value-... or vice-versa the effect will be different from Common Lisp.

Many Common Lisp declarations are ignored, and others match the Common Lisp standard in concept but not in detail. For example, local special declarations, which are purely advisory in Emacs Lisp, do not rigorously obey the scoping rules set down in Steele's book.

The variable *gensym-counter* starts out with a pseudo-random value rather than with zero. This is to cope with the fact that generated symbols become interned when they are written to and loaded back from a file.

The defstruct facility is compatible, except that structures are of type :type vector :named by default rather than some special, distinct type. Also, the :type slot option is ignored.

The second argument of check-type is treated differently.

Old CL Compatibility

Following is a list of all known incompatibilities between this package and the older Quiroz `cl.el' package.

This package's emulation of multiple return values in functions is incompatible with that of the older package. That package attempted to come as close as possible to true Common Lisp multiple return values; unfortunately, it could not be 100% reliable and so was prone to occasional surprises if used freely. This package uses a simpler method, namely replacing multiple values with lists of values, which is more predictable though more noticeably different from Common Lisp.

The defkeyword form and keywordp function are not implemented in this package.

The member, floor, ceiling, truncate, round, mod, and rem functions are suffixed by `*' in this package to avoid collision with existing functions in Emacs 18 or Emacs 19. The older package simply redefined these functions, overwriting the built-in meanings and causing serious portability problems with Emacs 19. (Some more recent versions of the Quiroz package changed the names to cl-member, etc.; this package defines the latter names as aliases for member*, etc.)

Certain functions in the old package which were buggy or inconsistent with the Common Lisp standard are incompatible with the conforming versions in this package. For example, eql and member were synonyms for eq and memq in that package, setf failed to preserve correct order of evaluation of its arguments, etc.

Finally, unlike the older package, this package is careful to prefix all of its internal names with cl-. Except for a few functions which are explicitly defined as additional features (such as floatp-safe and letf), this package does not export any non-`cl-' symbols which are not also part of Common Lisp.

The cl-compat package

The CL package includes emulations of some features of the old `cl.el', in the form of a compatibility package cl-compat. To use it, put (require 'cl-compat) in your program.

The old package defined a number of internal routines without cl- prefixes or other annotations. Call to these routines may have crept into existing Lisp code. cl-compat provides emulations of the following internal routines: pair-with-newsyms, zip-lists, unzip-lists, reassemble-arglists, duplicate-symbols-p, safe-idiv.

Some setf forms translated into calls to internal functions that user code might call directly. The functions setnth, setnthcdr, and setelt fall in this category; they are defined by cl-compat, but the best fix is to change to use setf properly.

The cl-compat file defines the keyword functions keywordp, keyword-of, and defkeyword, which are not defined by the new CL package because the use of keywords as data is discouraged.

The build-klist mechanism for parsing keyword arguments is emulated by cl-compat; the with-keyword-args macro is not, however, and in any case it's best to change to use the more natural keyword argument processing offered by defun*.

Multiple return values are treated differently by the two Common Lisp packages. The old package's method was more compatible with true Common Lisp, though it used heuristics that caused it to report spurious multiple return values in certain cases. The cl-compat package defines a set of multiple-value macros that are compatible with the old CL package; again, they are heuristic in nature, but they are guaranteed to work in any case where the old package's macros worked. To avoid name collision with the "official" multiple-value facilities, the ones in cl-compat have capitalized names: Values, Values-list, Multiple-value-bind, etc.

The functions cl-floor, cl-ceiling, cl-truncate, and cl-round are defined by cl-compat to use the old-style multiple-value mechanism, just as they did in the old package. The newer floor* and friends return their two results in a list rather than as multiple values. Note that older versions of the old package used the unadorned names floor, ceiling, etc.; cl-compat cannot use these names because they conflict with Emacs 19 built-ins.

Porting Common Lisp

This package is meant to be used as an extension to Emacs Lisp, not as an Emacs implementation of true Common Lisp. Some of the remaining differences between Emacs Lisp and Common Lisp make it difficult to port large Common Lisp applications to Emacs. For one, some of the features in this package are not fully compliant with ANSI or Steele; see section Common Lisp Compatibility. But there are also quite a few features that this package does not provide at all. Here are some major omissions that you will want watch out for when bringing Common Lisp code into Emacs.

Function Index

Jump to: a - b - c - d - e - f - g - h - i - l - m - n - o - p - r - s - t - u - w

a

  • abs
  • acons
  • adjoin
  • assert
  • assoc*
  • assoc-if
  • assoc-if-not
  • b

  • block
  • butlast
  • c

  • caddr
  • callf
  • callf2
  • case
  • ceiling*
  • check-type
  • cl-float-limits
  • cl-prettyexpand
  • clrhash
  • coerce
  • compiler-macroexpand
  • concatenate
  • copy-list
  • copy-tree
  • count
  • count-if
  • count-if-not
  • d

  • decf
  • declaim
  • declare
  • defalias
  • define-compiler-macro
  • define-modify-macro
  • define-setf-method
  • defmacro*
  • defsetf, defsetf
  • defstruct
  • defsubst*
  • deftype
  • defun*
  • delete
  • delete*
  • delete-duplicates
  • delete-if
  • delete-if-not
  • destructuring-bind
  • do
  • do*
  • do-all-symbols
  • do-symbols
  • dolist
  • dotimes
  • e

  • ecase
  • endp
  • eql
  • equalp
  • etypecase
  • eval-when
  • eval-when-compile
  • evenp
  • every
  • expt
  • f

  • fill
  • find
  • find-if
  • find-if-not
  • first
  • flet
  • floatp-safe
  • floor*
  • function*
  • g

  • gcd
  • gensym
  • gentemp
  • get*
  • get-setf-method
  • getf
  • gethash
  • h

  • hash-table-count
  • hash-table-p
  • i

  • ignore-errors
  • incf
  • intersection
  • isqrt
  • l

  • labels
  • last*
  • lcm
  • ldiff
  • letf
  • letf*
  • lexical-let
  • lexical-let*
  • list*
  • list-length
  • load-time-value
  • locally
  • loop, loop
  • m

  • macrolet
  • make-hash-table
  • make-random-state
  • map
  • mapc
  • mapcan
  • mapcar*
  • mapcon
  • maphash
  • mapl
  • maplist
  • member
  • member*
  • member-if
  • member-if-not
  • merge
  • minusp
  • mismatch
  • mod*
  • multiple-value-bind
  • multiple-value-setq
  • n

  • nbutlast
  • nintersection
  • notany
  • notevery
  • nset-difference
  • nset-exclusive-or
  • nsublis
  • nsubst
  • nsubst-if
  • nsubst-if-not
  • nsubstitute
  • nsubstitute-if
  • nsubstitute-if-not
  • nunion
  • o

  • oddp
  • p

  • pairlis
  • plusp
  • pop
  • position
  • position-if
  • position-if-not
  • proclaim
  • progv
  • psetf
  • psetq
  • push
  • pushnew
  • r

  • random*
  • random-state-p
  • rassoc
  • rassoc*
  • rassoc-if
  • rassoc-if-not
  • reduce
  • rem*
  • remf
  • remhash
  • remove
  • remove*
  • remove-duplicates
  • remove-if
  • remove-if-not
  • remprop
  • remq
  • replace
  • rest
  • return
  • return-from
  • rotatef
  • round*
  • s

  • search
  • set-difference
  • set-exclusive-or
  • setf
  • shiftf
  • some
  • sort*
  • stable-sort
  • sublis
  • subseq
  • subsetp
  • subst
  • subst-if
  • subst-if-not
  • substitute
  • substitute-if
  • substitute-if-not
  • symbol-macrolet
  • t

  • tailp
  • the
  • tree-equal
  • truncate*
  • typecase
  • typep
  • u

  • union
  • unless
  • w

  • when
  • Variable Index

    Jump to: * - f - l - m

    *

  • *gensym-counter*
  • *random-state*
  • f

  • float-epsilon
  • float-negative-epsilon
  • l

  • least-negative-float
  • least-negative-normalized-float
  • least-positive-float
  • least-positive-normalized-float
  • m

  • most-negative-fixnum
  • most-negative-float
  • most-positive-fixnum
  • most-positive-float

  • This document was generated on 6 November 2000 using texi2html 1.56k.