functions.texi 77.8 KB
Newer Older
Glenn Morris's avatar
Glenn Morris committed
1 2
@c -*-texinfo-*-
@c This is part of the GNU Emacs Lisp Reference Manual.
3
@c Copyright (C) 1990-1995, 1998-1999, 2001-2014 Free Software
4
@c Foundation, Inc.
Glenn Morris's avatar
Glenn Morris committed
5
@c See the file elisp.texi for copying conditions.
6
@node Functions
Glenn Morris's avatar
Glenn Morris committed
7 8 9 10 11 12 13
@chapter Functions

  A Lisp program is composed mainly of Lisp functions.  This chapter
explains what functions are, how they accept arguments, and how to
define them.

@menu
14 15 16 17 18 19 20 21
* What Is a Function::          Lisp functions vs. primitives; terminology.
* Lambda Expressions::          How functions are expressed as Lisp objects.
* Function Names::              A symbol can serve as the name of a function.
* Defining Functions::          Lisp expressions for defining functions.
* Calling Functions::           How to use an existing function.
* Mapping Functions::           Applying a function to each element of a list, etc.
* Anonymous Functions::         Lambda expressions are functions with no names.
* Function Cells::              Accessing or setting the function definition
Glenn Morris's avatar
Glenn Morris committed
22
                            of a symbol.
23 24 25 26 27 28 29 30
* Closures::                    Functions that enclose a lexical environment.
* Advising Functions::          Adding to the definition of a function.
* Obsolete Functions::          Declaring functions obsolete.
* Inline Functions::            Functions that the compiler will expand inline.
* Declare Form::                Adding additional information about a function.
* Declaring Functions::         Telling the compiler that a function is defined.
* Function Safety::             Determining whether a function is safe to call.
* Related Topics::              Cross-references to specific Lisp primitives
Glenn Morris's avatar
Glenn Morris committed
31 32 33 34 35 36
                            that have a special bearing on how functions work.
@end menu

@node What Is a Function
@section What Is a Function?

37 38 39 40 41 42 43 44 45 46 47
@cindex return value
@cindex value of function
@cindex argument
  In a general sense, a function is a rule for carrying out a
computation given input values called @dfn{arguments}.  The result of
the computation is called the @dfn{value} or @dfn{return value} of the
function.  The computation can also have side effects, such as lasting
changes in the values of variables or the contents of data structures.

  In most computer languages, every function has a name.  But in Lisp,
a function in the strictest sense has no name: it is an object which
48
can @emph{optionally} be associated with a symbol (e.g., @code{car})
49 50
that serves as the function name.  @xref{Function Names}.  When a
function has been given a name, we usually also refer to that symbol
51
as a ``function'' (e.g., we refer to ``the function @code{car}'').
52 53 54 55 56 57 58 59 60 61
In this manual, the distinction between a function name and the
function object itself is usually unimportant, but we will take note
wherever it is relevant.

  Certain function-like objects, called @dfn{special forms} and
@dfn{macros}, also accept arguments to carry out computations.
However, as explained below, these are not considered functions in
Emacs Lisp.

  Here are important terms for functions and function-like objects:
Glenn Morris's avatar
Glenn Morris committed
62 63

@table @dfn
64
@item lambda expression
65
A function (in the strict sense, i.e., a function object) which is
66 67 68 69
written in Lisp.  These are described in the following section.
@ifnottex
@xref{Lambda Expressions}.
@end ifnottex
Glenn Morris's avatar
Glenn Morris committed
70 71 72 73 74

@item primitive
@cindex primitive
@cindex subr
@cindex built-in function
75
A function which is callable from Lisp but is actually written in C@.
76 77 78 79 80 81
Primitives are also called @dfn{built-in functions}, or @dfn{subrs}.
Examples include functions like @code{car} and @code{append}.  In
addition, all special forms (see below) are also considered
primitives.

Usually, a function is implemented as a primitive because it is a
82
fundamental part of Lisp (e.g., @code{car}), or because it provides a
83 84 85 86
low-level interface to operating system services, or because it needs
to run fast.  Unlike functions defined in Lisp, primitives can be
modified or added only by changing the C sources and recompiling
Emacs.  See @ref{Writing Emacs Primitives}.
Glenn Morris's avatar
Glenn Morris committed
87 88

@item special form
89 90 91 92 93
A primitive that is like a function but does not evaluate all of its
arguments in the usual way.  It may evaluate only some of the
arguments, or may evaluate them in an unusual order, or several times.
Examples include @code{if}, @code{and}, and @code{while}.
@xref{Special Forms}.
Glenn Morris's avatar
Glenn Morris committed
94 95 96

@item macro
@cindex macro
97 98 99 100 101
A construct defined in Lisp, which differs from a function in that it
translates a Lisp expression into another expression which is to be
evaluated instead of the original expression.  Macros enable Lisp
programmers to do the sorts of things that special forms can do.
@xref{Macros}.
Glenn Morris's avatar
Glenn Morris committed
102 103 104

@item command
@cindex command
105 106 107 108 109 110 111
An object which can be invoked via the @code{command-execute}
primitive, usually due to the user typing in a key sequence
@dfn{bound} to that command.  @xref{Interactive Call}.  A command is
usually a function; if the function is written in Lisp, it is made
into a command by an @code{interactive} form in the function
definition (@pxref{Defining Commands}).  Commands that are functions
can also be called from Lisp expressions, just like other functions.
Glenn Morris's avatar
Glenn Morris committed
112 113

Keyboard macros (strings and vectors) are commands also, even though
114 115 116 117 118 119 120 121 122
they are not functions.  @xref{Keyboard Macros}.  We say that a symbol
is a command if its function cell contains a command (@pxref{Symbol
Components}); such a @dfn{named command} can be invoked with
@kbd{M-x}.

@item closure
A function object that is much like a lambda expression, except that
it also encloses an ``environment'' of lexical variable bindings.
@xref{Closures}.
Glenn Morris's avatar
Glenn Morris committed
123 124

@item byte-code function
125 126
A function that has been compiled by the byte compiler.
@xref{Byte-Code Type}.
127 128 129

@item autoload object
@cindex autoload object
130 131 132
A place-holder for a real function.  If the autoload object is called,
Emacs loads the file containing the definition of the real function,
and then calls the real function.  @xref{Autoload}.
Glenn Morris's avatar
Glenn Morris committed
133 134
@end table

135 136 137
  You can use the function @code{functionp} to test if an object is a
function:

Glenn Morris's avatar
Glenn Morris committed
138 139
@defun functionp object
This function returns @code{t} if @var{object} is any kind of
140
function, i.e., can be passed to @code{funcall}.  Note that
141 142
@code{functionp} returns @code{t} for symbols that are function names,
and returns @code{nil} for special forms.
Glenn Morris's avatar
Glenn Morris committed
143 144
@end defun

145 146 147
@noindent
Unlike @code{functionp}, the next three functions do @emph{not} treat
a symbol as its function definition.
Glenn Morris's avatar
Glenn Morris committed
148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189

@defun subrp object
This function returns @code{t} if @var{object} is a built-in function
(i.e., a Lisp primitive).

@example
@group
(subrp 'message)            ; @r{@code{message} is a symbol,}
     @result{} nil                 ;   @r{not a subr object.}
@end group
@group
(subrp (symbol-function 'message))
     @result{} t
@end group
@end example
@end defun

@defun byte-code-function-p object
This function returns @code{t} if @var{object} is a byte-code
function.  For example:

@example
@group
(byte-code-function-p (symbol-function 'next-line))
     @result{} t
@end group
@end example
@end defun

@defun subr-arity subr
This function provides information about the argument list of a
primitive, @var{subr}.  The returned value is a pair
@code{(@var{min} . @var{max})}.  @var{min} is the minimum number of
args.  @var{max} is the maximum number or the symbol @code{many}, for a
function with @code{&rest} arguments, or the symbol @code{unevalled} if
@var{subr} is a special form.
@end defun

@node Lambda Expressions
@section Lambda Expressions
@cindex lambda expression

190 191
  A lambda expression is a function object written in Lisp.  Here is
an example:
Glenn Morris's avatar
Glenn Morris committed
192 193

@example
194 195 196
(lambda (x)
  "Return the hyperbolic cosine of X."
  (* 0.5 (+ (exp x) (exp (- x)))))
Glenn Morris's avatar
Glenn Morris committed
197 198 199
@end example

@noindent
200 201
In Emacs Lisp, such a list is a valid expression which evaluates to
a function object.
202 203 204 205 206 207 208

  A lambda expression, by itself, has no name; it is an @dfn{anonymous
function}.  Although lambda expressions can be used this way
(@pxref{Anonymous Functions}), they are more commonly associated with
symbols to make @dfn{named functions} (@pxref{Function Names}).
Before going into these details, the following subsections describe
the components of a lambda expression and what they do.
Glenn Morris's avatar
Glenn Morris committed
209 210

@menu
211 212 213 214
* Lambda Components::           The parts of a lambda expression.
* Simple Lambda::               A simple example.
* Argument List::               Details and special features of argument lists.
* Function Documentation::      How to put documentation in a function.
Glenn Morris's avatar
Glenn Morris committed
215 216 217 218 219
@end menu

@node Lambda Components
@subsection Components of a Lambda Expression

220
  A lambda expression is a list that looks like this:
Glenn Morris's avatar
Glenn Morris committed
221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260

@example
(lambda (@var{arg-variables}@dots{})
  [@var{documentation-string}]
  [@var{interactive-declaration}]
  @var{body-forms}@dots{})
@end example

@cindex lambda list
  The first element of a lambda expression is always the symbol
@code{lambda}.  This indicates that the list represents a function.  The
reason functions are defined to start with @code{lambda} is so that
other lists, intended for other uses, will not accidentally be valid as
functions.

  The second element is a list of symbols---the argument variable names.
This is called the @dfn{lambda list}.  When a Lisp function is called,
the argument values are matched up against the variables in the lambda
list, which are given local bindings with the values provided.
@xref{Local Variables}.

  The documentation string is a Lisp string object placed within the
function definition to describe the function for the Emacs help
facilities.  @xref{Function Documentation}.

  The interactive declaration is a list of the form @code{(interactive
@var{code-string})}.  This declares how to provide arguments if the
function is used interactively.  Functions with this declaration are called
@dfn{commands}; they can be called using @kbd{M-x} or bound to a key.
Functions not intended to be called in this way should not have interactive
declarations.  @xref{Defining Commands}, for how to write an interactive
declaration.

@cindex body of function
  The rest of the elements are the @dfn{body} of the function: the Lisp
code to do the work of the function (or, as a Lisp programmer would say,
``a list of Lisp forms to evaluate'').  The value returned by the
function is the value returned by the last element of the body.

@node Simple Lambda
261
@subsection A Simple Lambda Expression Example
Glenn Morris's avatar
Glenn Morris committed
262

263
  Consider the following example:
Glenn Morris's avatar
Glenn Morris committed
264 265 266 267 268 269

@example
(lambda (a b c) (+ a b c))
@end example

@noindent
270
We can call this function by passing it to @code{funcall}, like this:
Glenn Morris's avatar
Glenn Morris committed
271 272 273

@example
@group
274 275
(funcall (lambda (a b c) (+ a b c))
         1 2 3)
Glenn Morris's avatar
Glenn Morris committed
276 277 278 279 280 281 282 283 284 285 286 287 288 289
@end group
@end example

@noindent
This call evaluates the body of the lambda expression  with the variable
@code{a} bound to 1, @code{b} bound to 2, and @code{c} bound to 3.
Evaluation of the body adds these three numbers, producing the result 6;
therefore, this call to the function returns the value 6.

  Note that the arguments can be the results of other function calls, as in
this example:

@example
@group
290 291
(funcall (lambda (a b c) (+ a b c))
         1 (* 2 3) (- 5 4))
Glenn Morris's avatar
Glenn Morris committed
292 293 294 295 296 297 298 299
@end group
@end example

@noindent
This evaluates the arguments @code{1}, @code{(* 2 3)}, and @code{(- 5
4)} from left to right.  Then it applies the lambda expression to the
argument values 1, 6 and 1 to produce the value 8.

300 301 302 303 304 305 306 307 308
  As these examples show, you can use a form with a lambda expression
as its @sc{car} to make local variables and give them values.  In the
old days of Lisp, this technique was the only way to bind and
initialize local variables.  But nowadays, it is clearer to use the
special form @code{let} for this purpose (@pxref{Local Variables}).
Lambda expressions are mainly used as anonymous functions for passing
as arguments to other functions (@pxref{Anonymous Functions}), or
stored as symbol function definitions to produce named functions
(@pxref{Function Names}).
Glenn Morris's avatar
Glenn Morris committed
309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400

@node Argument List
@subsection Other Features of Argument Lists
@kindex wrong-number-of-arguments
@cindex argument binding
@cindex binding arguments
@cindex argument lists, features

  Our simple sample function, @code{(lambda (a b c) (+ a b c))},
specifies three argument variables, so it must be called with three
arguments: if you try to call it with only two arguments or four
arguments, you get a @code{wrong-number-of-arguments} error.

  It is often convenient to write a function that allows certain
arguments to be omitted.  For example, the function @code{substring}
accepts three arguments---a string, the start index and the end
index---but the third argument defaults to the @var{length} of the
string if you omit it.  It is also convenient for certain functions to
accept an indefinite number of arguments, as the functions @code{list}
and @code{+} do.

@cindex optional arguments
@cindex rest arguments
@kindex &optional
@kindex &rest
  To specify optional arguments that may be omitted when a function
is called, simply include the keyword @code{&optional} before the optional
arguments.  To specify a list of zero or more extra arguments, include the
keyword @code{&rest} before one final argument.

  Thus, the complete syntax for an argument list is as follows:

@example
@group
(@var{required-vars}@dots{}
 @r{[}&optional @var{optional-vars}@dots{}@r{]}
 @r{[}&rest @var{rest-var}@r{]})
@end group
@end example

@noindent
The square brackets indicate that the @code{&optional} and @code{&rest}
clauses, and the variables that follow them, are optional.

  A call to the function requires one actual argument for each of the
@var{required-vars}.  There may be actual arguments for zero or more of
the @var{optional-vars}, and there cannot be any actual arguments beyond
that unless the lambda list uses @code{&rest}.  In that case, there may
be any number of extra actual arguments.

  If actual arguments for the optional and rest variables are omitted,
then they always default to @code{nil}.  There is no way for the
function to distinguish between an explicit argument of @code{nil} and
an omitted argument.  However, the body of the function is free to
consider @code{nil} an abbreviation for some other meaningful value.
This is what @code{substring} does; @code{nil} as the third argument to
@code{substring} means to use the length of the string supplied.

@cindex CL note---default optional arg
@quotation
@b{Common Lisp note:} Common Lisp allows the function to specify what
default value to use when an optional argument is omitted; Emacs Lisp
always uses @code{nil}.  Emacs Lisp does not support ``supplied-p''
variables that tell you whether an argument was explicitly passed.
@end quotation

  For example, an argument list that looks like this:

@example
(a b &optional c d &rest e)
@end example

@noindent
binds @code{a} and @code{b} to the first two actual arguments, which are
required.  If one or two more arguments are provided, @code{c} and
@code{d} are bound to them respectively; any arguments after the first
four are collected into a list and @code{e} is bound to that list.  If
there are only two arguments, @code{c} is @code{nil}; if two or three
arguments, @code{d} is @code{nil}; if four arguments or fewer, @code{e}
is @code{nil}.

  There is no way to have required arguments following optional
ones---it would not make sense.  To see why this must be so, suppose
that @code{c} in the example were optional and @code{d} were required.
Suppose three actual arguments are given; which variable would the
third argument be for?  Would it be used for the @var{c}, or for
@var{d}?  One can argue for both possibilities.  Similarly, it makes
no sense to have any more arguments (either required or optional)
after a @code{&rest} argument.

  Here are some examples of argument lists and proper calls:

401
@example
402 403
(funcall (lambda (n) (1+ n))        ; @r{One required:}
         1)                         ; @r{requires exactly one argument.}
Glenn Morris's avatar
Glenn Morris committed
404
     @result{} 2
405 406 407
(funcall (lambda (n &optional n1)   ; @r{One required and one optional:}
           (if n1 (+ n n1) (1+ n))) ; @r{1 or 2 arguments.}
         1 2)
Glenn Morris's avatar
Glenn Morris committed
408
     @result{} 3
409 410 411
(funcall (lambda (n &rest ns)       ; @r{One required and one rest:}
           (+ n (apply '+ ns)))     ; @r{1 or more arguments.}
         1 2 3 4 5)
Glenn Morris's avatar
Glenn Morris committed
412
     @result{} 15
413
@end example
Glenn Morris's avatar
Glenn Morris committed
414 415 416 417 418

@node Function Documentation
@subsection Documentation Strings of Functions
@cindex documentation of function

419 420 421 422 423 424
  A lambda expression may optionally have a @dfn{documentation string}
just after the lambda list.  This string does not affect execution of
the function; it is a kind of comment, but a systematized comment
which actually appears inside the Lisp world and can be used by the
Emacs help facilities.  @xref{Documentation}, for how the
documentation string is accessed.
Glenn Morris's avatar
Glenn Morris committed
425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476

  It is a good idea to provide documentation strings for all the
functions in your program, even those that are called only from within
your program.  Documentation strings are like comments, except that they
are easier to access.

  The first line of the documentation string should stand on its own,
because @code{apropos} displays just this first line.  It should consist
of one or two complete sentences that summarize the function's purpose.

  The start of the documentation string is usually indented in the
source file, but since these spaces come before the starting
double-quote, they are not part of the string.  Some people make a
practice of indenting any additional lines of the string so that the
text lines up in the program source.  @emph{That is a mistake.}  The
indentation of the following lines is inside the string; what looks
nice in the source code will look ugly when displayed by the help
commands.

  You may wonder how the documentation string could be optional, since
there are required components of the function that follow it (the body).
Since evaluation of a string returns that string, without any side effects,
it has no effect if it is not the last form in the body.  Thus, in
practice, there is no confusion between the first form of the body and the
documentation string; if the only body form is a string then it serves both
as the return value and as the documentation.

  The last line of the documentation string can specify calling
conventions different from the actual function arguments.  Write
text like this:

@example
\(fn @var{arglist})
@end example

@noindent
following a blank line, at the beginning of the line, with no newline
following it inside the documentation string.  (The @samp{\} is used
to avoid confusing the Emacs motion commands.)  The calling convention
specified in this way appears in help messages in place of the one
derived from the actual arguments of the function.

  This feature is particularly useful for macro definitions, since the
arguments written in a macro definition often do not correspond to the
way users think of the parts of the macro call.

@node Function Names
@section Naming a Function
@cindex function definition
@cindex named function
@cindex function name

477 478
  A symbol can serve as the name of a function.  This happens when the
symbol's @dfn{function cell} (@pxref{Symbol Components}) contains a
479
function object (e.g., a lambda expression).  Then the symbol itself
480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497
becomes a valid, callable function, equivalent to the function object
in its function cell.

  The contents of the function cell are also called the symbol's
@dfn{function definition}.  The procedure of using a symbol's function
definition in place of the symbol is called @dfn{symbol function
indirection}; see @ref{Function Indirection}.  If you have not given a
symbol a function definition, its function cell is said to be
@dfn{void}, and it cannot be used as a function.

  In practice, nearly all functions have names, and are referred to by
their names.  You can create a named Lisp function by defining a
lambda expression and putting it in a function cell (@pxref{Function
Cells}).  However, it is more common to use the @code{defun} special
form, described in the next section.
@ifnottex
@xref{Defining Functions}.
@end ifnottex
Glenn Morris's avatar
Glenn Morris committed
498 499

  We give functions names because it is convenient to refer to them by
500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515
their names in Lisp expressions.  Also, a named Lisp function can
easily refer to itself---it can be recursive.  Furthermore, primitives
can only be referred to textually by their names, since primitive
function objects (@pxref{Primitive Function Type}) have no read
syntax.

  A function need not have a unique name.  A given function object
@emph{usually} appears in the function cell of only one symbol, but
this is just a convention.  It is easy to store it in several symbols
using @code{fset}; then each of the symbols is a valid name for the
same function.

  Note that a symbol used as a function name may also be used as a
variable; these two uses of a symbol are independent and do not
conflict.  (This is not the case in some dialects of Lisp, like
Scheme.)
Glenn Morris's avatar
Glenn Morris committed
516 517 518 519 520 521 522

@node Defining Functions
@section Defining Functions
@cindex defining a function

  We usually give a name to a function when it is first created.  This
is called @dfn{defining a function}, and it is done with the
523
@code{defun} macro.
Glenn Morris's avatar
Glenn Morris committed
524

525
@defmac defun name args [doc] [declare] [interactive] body@dots{}
Glenn Morris's avatar
Glenn Morris committed
526
@code{defun} is the usual way to define new Lisp functions.  It
527 528 529
defines the symbol @var{name} as a function with argument list
@var{args} and body forms given by @var{body}.  Neither @var{name} nor
@var{args} should be quoted.
Glenn Morris's avatar
Glenn Morris committed
530

531 532 533 534 535 536
@var{doc}, if present, should be a string specifying the function's
documentation string (@pxref{Function Documentation}).  @var{declare},
if present, should be a @code{declare} form specifying function
metadata (@pxref{Declare Form}).  @var{interactive}, if present,
should be an @code{interactive} form specifying how the function is to
be called interactively (@pxref{Interactive Call}).
Glenn Morris's avatar
Glenn Morris committed
537

538
The return value of @code{defun} is undefined.
Glenn Morris's avatar
Glenn Morris committed
539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565

Here are some examples:

@example
@group
(defun foo () 5)
(foo)
     @result{} 5
@end group

@group
(defun bar (a &optional b &rest c)
    (list a b c))
(bar 1 2 3 4 5)
     @result{} (1 2 (3 4 5))
@end group
@group
(bar 1)
     @result{} (1 nil nil)
@end group
@group
(bar)
@error{} Wrong number of arguments.
@end group

@group
(defun capitalize-backwards ()
566
  "Upcase the last letter of the word at point."
Glenn Morris's avatar
Glenn Morris committed
567 568 569 570 571 572 573 574 575 576
  (interactive)
  (backward-word 1)
  (forward-word 1)
  (backward-char 1)
  (capitalize-word 1))
@end group
@end example

Be careful not to redefine existing functions unintentionally.
@code{defun} redefines even primitive functions such as @code{car}
577 578 579 580
without any hesitation or notification.  Emacs does not prevent you
from doing this, because redefining a function is sometimes done
deliberately, and there is no way to distinguish deliberate
redefinition from unintentional redefinition.
581
@end defmac
Glenn Morris's avatar
Glenn Morris committed
582 583

@cindex function aliases
Xue Fuqiao's avatar
Xue Fuqiao committed
584
@cindex alias, for functions
585
@defun defalias name definition &optional doc
Glenn Morris's avatar
Glenn Morris committed
586
@anchor{Definition of defalias}
587
This function defines the symbol @var{name} as a function, with
Glenn Morris's avatar
Glenn Morris committed
588
definition @var{definition} (which can be any valid Lisp function).
589
Its return value is @emph{undefined}.
Glenn Morris's avatar
Glenn Morris committed
590

591 592
If @var{doc} is non-@code{nil}, it becomes the function documentation
of @var{name}.  Otherwise, any documentation provided by
Glenn Morris's avatar
Glenn Morris committed
593 594
@var{definition} is used.

595 596 597 598 599
@cindex defalias-fset-function property
Internally, @code{defalias} normally uses @code{fset} to set the definition.
If @var{name} has a @code{defalias-fset-function} property, however,
the associated value is used as a function to call in place of @code{fset}.

Glenn Morris's avatar
Glenn Morris committed
600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622
The proper place to use @code{defalias} is where a specific function
name is being defined---especially where that name appears explicitly in
the source file being loaded.  This is because @code{defalias} records
which file defined the function, just like @code{defun}
(@pxref{Unloading}).

By contrast, in programs that manipulate function definitions for other
purposes, it is better to use @code{fset}, which does not keep such
records.  @xref{Function Cells}.
@end defun

  You cannot create a new primitive function with @code{defun} or
@code{defalias}, but you can use them to change the function definition of
any symbol, even one such as @code{car} or @code{x-popup-menu} whose
normal definition is a primitive.  However, this is risky: for
instance, it is next to impossible to redefine @code{car} without
breaking Lisp completely.  Redefining an obscure function such as
@code{x-popup-menu} is less dangerous, but it still may not work as
you expect.  If there are calls to the primitive from C code, they
call the primitive's C definition directly, so changing the symbol's
definition will have no effect on them.

  See also @code{defsubst}, which defines a function like @code{defun}
623 624
and tells the Lisp compiler to perform inline expansion on it.
@xref{Inline Functions}.
Glenn Morris's avatar
Glenn Morris committed
625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729

@node Calling Functions
@section Calling Functions
@cindex function invocation
@cindex calling a function

  Defining functions is only half the battle.  Functions don't do
anything until you @dfn{call} them, i.e., tell them to run.  Calling a
function is also known as @dfn{invocation}.

  The most common way of invoking a function is by evaluating a list.
For example, evaluating the list @code{(concat "a" "b")} calls the
function @code{concat} with arguments @code{"a"} and @code{"b"}.
@xref{Evaluation}, for a description of evaluation.

  When you write a list as an expression in your program, you specify
which function to call, and how many arguments to give it, in the text
of the program.  Usually that's just what you want.  Occasionally you
need to compute at run time which function to call.  To do that, use
the function @code{funcall}.  When you also need to determine at run
time how many arguments to pass, use @code{apply}.

@defun funcall function &rest arguments
@code{funcall} calls @var{function} with @var{arguments}, and returns
whatever @var{function} returns.

Since @code{funcall} is a function, all of its arguments, including
@var{function}, are evaluated before @code{funcall} is called.  This
means that you can use any expression to obtain the function to be
called.  It also means that @code{funcall} does not see the
expressions you write for the @var{arguments}, only their values.
These values are @emph{not} evaluated a second time in the act of
calling @var{function}; the operation of @code{funcall} is like the
normal procedure for calling a function, once its arguments have
already been evaluated.

The argument @var{function} must be either a Lisp function or a
primitive function.  Special forms and macros are not allowed, because
they make sense only when given the ``unevaluated'' argument
expressions.  @code{funcall} cannot provide these because, as we saw
above, it never knows them in the first place.

@example
@group
(setq f 'list)
     @result{} list
@end group
@group
(funcall f 'x 'y 'z)
     @result{} (x y z)
@end group
@group
(funcall f 'x 'y '(z))
     @result{} (x y (z))
@end group
@group
(funcall 'and t nil)
@error{} Invalid function: #<subr and>
@end group
@end example

Compare these examples with the examples of @code{apply}.
@end defun

@defun apply function &rest arguments
@code{apply} calls @var{function} with @var{arguments}, just like
@code{funcall} but with one difference: the last of @var{arguments} is a
list of objects, which are passed to @var{function} as separate
arguments, rather than a single list.  We say that @code{apply}
@dfn{spreads} this list so that each individual element becomes an
argument.

@code{apply} returns the result of calling @var{function}.  As with
@code{funcall}, @var{function} must either be a Lisp function or a
primitive function; special forms and macros do not make sense in
@code{apply}.

@example
@group
(setq f 'list)
     @result{} list
@end group
@group
(apply f 'x 'y 'z)
@error{} Wrong type argument: listp, z
@end group
@group
(apply '+ 1 2 '(3 4))
     @result{} 10
@end group
@group
(apply '+ '(1 2 3 4))
     @result{} 10
@end group

@group
(apply 'append '((a b c) nil (x y z) nil))
     @result{} (a b c x y z)
@end group
@end example

For an interesting example of using @code{apply}, see @ref{Definition
of mapcar}.
@end defun

730 731
@cindex partial application of functions
@cindex currying
732
  Sometimes it is useful to fix some of the function's arguments at
733 734 735 736 737 738 739 740 741
certain values, and leave the rest of arguments for when the function
is actually called.  The act of fixing some of the function's
arguments is called @dfn{partial application} of the function@footnote{
This is related to, but different from @dfn{currying}, which
transforms a function that takes multiple arguments in such a way that
it can be called as a chain of functions, each one with a single
argument.}.
The result is a new function that accepts the rest of
arguments and calls the original function with all the arguments
742 743 744
combined.

  Here's how to do partial application in Emacs Lisp:
745 746 747 748 749 750 751 752 753

@defun apply-partially func &rest args
This function returns a new function which, when called, will call
@var{func} with the list of arguments composed from @var{args} and
additional arguments specified at the time of the call.  If @var{func}
accepts @var{n} arguments, then a call to @code{apply-partially} with
@w{@code{@var{m} < @var{n}}} arguments will produce a new function of
@w{@code{@var{n} - @var{m}}} arguments.

754 755 756
Here's how we could define the built-in function @code{1+}, if it
didn't exist, using @code{apply-partially} and @code{+}, another
built-in function:
757 758 759

@example
@group
760 761 762 763 764
(defalias '1+ (apply-partially '+ 1)
  "Increment argument by one.")
@end group
@group
(1+ 10)
765 766 767 768 769
     @result{} 11
@end group
@end example
@end defun

Glenn Morris's avatar
Glenn Morris committed
770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787
@cindex functionals
  It is common for Lisp functions to accept functions as arguments or
find them in data structures (especially in hook variables and property
lists) and call them using @code{funcall} or @code{apply}.  Functions
that accept function arguments are often called @dfn{functionals}.

  Sometimes, when you call a functional, it is useful to supply a no-op
function as the argument.  Here are two different kinds of no-op
function:

@defun identity arg
This function returns @var{arg} and has no side effects.
@end defun

@defun ignore &rest args
This function ignores any arguments and returns @code{nil}.
@end defun

788 789 790 791
  Some functions are user-visible @dfn{commands}, which can be called
interactively (usually by a key sequence).  It is possible to invoke
such a command exactly as though it was called interactively, by using
the @code{call-interactively} function.  @xref{Interactive Call}.
792

Glenn Morris's avatar
Glenn Morris committed
793 794 795 796 797 798
@node Mapping Functions
@section Mapping Functions
@cindex mapping functions

  A @dfn{mapping function} applies a given function (@emph{not} a
special form or macro) to each element of a list or other collection.
799 800 801 802 803 804
Emacs Lisp has several such functions; this section describes
@code{mapcar}, @code{mapc}, and @code{mapconcat}, which map over a
list.  @xref{Definition of mapatoms}, for the function @code{mapatoms}
which maps over the symbols in an obarray.  @xref{Definition of
maphash}, for the function @code{maphash} which maps over key/value
associations in a hash table.
Glenn Morris's avatar
Glenn Morris committed
805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820

  These mapping functions do not allow char-tables because a char-table
is a sparse array whose nominal range of indices is very large.  To map
over a char-table in a way that deals properly with its sparse nature,
use the function @code{map-char-table} (@pxref{Char-Tables}).

@defun mapcar function sequence
@anchor{Definition of mapcar}
@code{mapcar} applies @var{function} to each element of @var{sequence}
in turn, and returns a list of the results.

The argument @var{sequence} can be any kind of sequence except a
char-table; that is, a list, a vector, a bool-vector, or a string.  The
result is always a list.  The length of the result is the same as the
length of @var{sequence}.  For example:

821
@example
Glenn Morris's avatar
Glenn Morris committed
822 823 824 825 826
@group
(mapcar 'car '((a b) (c d) (e f)))
     @result{} (a c e)
(mapcar '1+ [1 2 3])
     @result{} (2 3 4)
827
(mapcar 'string "abc")
Glenn Morris's avatar
Glenn Morris committed
828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852
     @result{} ("a" "b" "c")
@end group

@group
;; @r{Call each function in @code{my-hooks}.}
(mapcar 'funcall my-hooks)
@end group

@group
(defun mapcar* (function &rest args)
  "Apply FUNCTION to successive cars of all ARGS.
Return the list of results."
  ;; @r{If no list is exhausted,}
  (if (not (memq nil args))
      ;; @r{apply function to @sc{car}s.}
      (cons (apply function (mapcar 'car args))
            (apply 'mapcar* function
                   ;; @r{Recurse for rest of elements.}
                   (mapcar 'cdr args)))))
@end group

@group
(mapcar* 'cons '(a b c) '(1 2 3 4))
     @result{} ((a . 1) (b . 2) (c . 3))
@end group
853
@end example
Glenn Morris's avatar
Glenn Morris committed
854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873
@end defun

@defun mapc function sequence
@code{mapc} is like @code{mapcar} except that @var{function} is used for
side-effects only---the values it returns are ignored, not collected
into a list.  @code{mapc} always returns @var{sequence}.
@end defun

@defun mapconcat function sequence separator
@code{mapconcat} applies @var{function} to each element of
@var{sequence}: the results, which must be strings, are concatenated.
Between each pair of result strings, @code{mapconcat} inserts the string
@var{separator}.  Usually @var{separator} contains a space or comma or
other suitable punctuation.

The argument @var{function} must be a function that can take one
argument and return a string.  The argument @var{sequence} can be any
kind of sequence except a char-table; that is, a list, a vector, a
bool-vector, or a string.

874
@example
Glenn Morris's avatar
Glenn Morris committed
875 876 877 878 879 880 881 882 883 884 885 886 887
@group
(mapconcat 'symbol-name
           '(The cat in the hat)
           " ")
     @result{} "The cat in the hat"
@end group

@group
(mapconcat (function (lambda (x) (format "%c" (1+ x))))
           "HAL-8000"
           "")
     @result{} "IBM.9111"
@end group
888
@end example
Glenn Morris's avatar
Glenn Morris committed
889 890 891 892 893 894
@end defun

@node Anonymous Functions
@section Anonymous Functions
@cindex anonymous function

895 896 897 898 899 900 901 902 903 904 905 906 907 908 909
  Although functions are usually defined with @code{defun} and given
names at the same time, it is sometimes convenient to use an explicit
lambda expression---an @dfn{anonymous function}.  Anonymous functions
are valid wherever function names are.  They are often assigned as
variable values, or as arguments to functions; for instance, you might
pass one as the @var{function} argument to @code{mapcar}, which
applies that function to each element of a list (@pxref{Mapping
Functions}).  @xref{describe-symbols example}, for a realistic example
of this.

  When defining a lambda expression that is to be used as an anonymous
function, you can in principle use any method to construct the list.
But typically you should use the @code{lambda} macro, or the
@code{function} special form, or the @code{#'} read syntax:

910 911 912 913 914 915 916 917
@defmac lambda args [doc] [interactive] body@dots{}
This macro returns an anonymous function with argument list
@var{args}, documentation string @var{doc} (if any), interactive spec
@var{interactive} (if any), and body forms given by @var{body}.

In effect, this macro makes @code{lambda} forms ``self-quoting'':
evaluating a form whose @sc{car} is @code{lambda} yields the form
itself:
Glenn Morris's avatar
Glenn Morris committed
918

919 920 921 922
@example
(lambda (x) (* x x))
     @result{} (lambda (x) (* x x))
@end example
Glenn Morris's avatar
Glenn Morris committed
923

924 925 926 927
The @code{lambda} form has one other effect: it tells the Emacs
evaluator and byte-compiler that its argument is a function, by using
@code{function} as a subroutine (see below).
@end defmac
Glenn Morris's avatar
Glenn Morris committed
928

929 930 931 932 933 934 935 936
@defspec function function-object
@cindex function quoting
This special form returns @var{function-object} without evaluating it.
In this, it is similar to @code{quote} (@pxref{Quoting}).  But unlike
@code{quote}, it also serves as a note to the Emacs evaluator and
byte-compiler that @var{function-object} is intended to be used as a
function.  Assuming @var{function-object} is a valid lambda
expression, this has two effects:
Glenn Morris's avatar
Glenn Morris committed
937

938 939 940 941
@itemize
@item
When the code is byte-compiled, @var{function-object} is compiled into
a byte-code function object (@pxref{Byte Compilation}).
Glenn Morris's avatar
Glenn Morris committed
942

943 944 945 946 947
@item
When lexical binding is enabled, @var{function-object} is converted
into a closure.  @xref{Closures}.
@end itemize
@end defspec
Glenn Morris's avatar
Glenn Morris committed
948

949 950 951
@cindex @samp{#'} syntax
The read syntax @code{#'} is a short-hand for using @code{function}.
The following forms are all equivalent:
Glenn Morris's avatar
Glenn Morris committed
952

953 954 955 956 957
@example
(lambda (x) (* x x))
(function (lambda (x) (* x x)))
#'(lambda (x) (* x x))
@end example
Glenn Morris's avatar
Glenn Morris committed
958

959 960 961 962
  In the following example, we define a @code{change-property}
function that takes a function as its third argument, followed by a
@code{double-property} function that makes use of
@code{change-property} by passing it an anonymous function:
Glenn Morris's avatar
Glenn Morris committed
963 964 965 966 967 968 969 970 971 972

@example
@group
(defun change-property (symbol prop function)
  (let ((value (get symbol prop)))
    (put symbol prop (funcall function value))))
@end group

@group
(defun double-property (symbol prop)
973
  (change-property symbol prop (lambda (x) (* 2 x))))
Glenn Morris's avatar
Glenn Morris committed
974 975 976 977
@end group
@end example

@noindent
978
Note that we do not quote the @code{lambda} form.
Glenn Morris's avatar
Glenn Morris committed
979

980 981 982
  If you compile the above code, the anonymous function is also
compiled.  This would not happen if, say, you had constructed the
anonymous function by quoting it as a list:
Glenn Morris's avatar
Glenn Morris committed
983

984
@c Do not unquote this lambda!
Glenn Morris's avatar
Glenn Morris committed
985 986 987
@example
@group
(defun double-property (symbol prop)
988
  (change-property symbol prop '(lambda (x) (* 2 x))))
Glenn Morris's avatar
Glenn Morris committed
989 990 991 992
@end group
@end example

@noindent
993 994 995 996
In that case, the anonymous function is kept as a lambda expression in
the compiled code.  The byte-compiler cannot assume this list is a
function, even though it looks like one, since it does not know that
@code{change-property} intends to use it as a function.
Glenn Morris's avatar
Glenn Morris committed
997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009

@node Function Cells
@section Accessing Function Cell Contents

  The @dfn{function definition} of a symbol is the object stored in the
function cell of the symbol.  The functions described here access, test,
and set the function cell of symbols.

  See also the function @code{indirect-function}.  @xref{Definition of
indirect-function}.

@defun symbol-function symbol
@kindex void-function
Chong Yidong's avatar
Chong Yidong committed
1010 1011
This returns the object in the function cell of @var{symbol}.  It does
not check that the returned object is a legitimate function.
Glenn Morris's avatar
Glenn Morris committed
1012

Chong Yidong's avatar
Chong Yidong committed
1013 1014 1015
If the function cell is void, the return value is @code{nil}.  To
distinguish between a function cell that is void and one set to
@code{nil}, use @code{fboundp} (see below).
Glenn Morris's avatar
Glenn Morris committed
1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034

@example
@group
(defun bar (n) (+ n 2))
(symbol-function 'bar)
     @result{} (lambda (n) (+ n 2))
@end group
@group
(fset 'baz 'bar)
     @result{} bar
@end group
@group
(symbol-function 'baz)
     @result{} bar
@end group
@end example
@end defun

@cindex void function cell
Chong Yidong's avatar
Chong Yidong committed
1035 1036 1037 1038
  If you have never given a symbol any function definition, we say
that that symbol's function cell is @dfn{void}.  In other words, the
function cell does not have any Lisp object in it.  If you try to call
the symbol as a function, Emacs signals a @code{void-function} error.
Glenn Morris's avatar
Glenn Morris committed
1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085

  Note that void is not the same as @code{nil} or the symbol
@code{void}.  The symbols @code{nil} and @code{void} are Lisp objects,
and can be stored into a function cell just as any other object can be
(and they can be valid functions if you define them in turn with
@code{defun}).  A void function cell contains no object whatsoever.

  You can test the voidness of a symbol's function definition with
@code{fboundp}.  After you have given a symbol a function definition, you
can make it void once more using @code{fmakunbound}.

@defun fboundp symbol
This function returns @code{t} if the symbol has an object in its
function cell, @code{nil} otherwise.  It does not check that the object
is a legitimate function.
@end defun

@defun fmakunbound symbol
This function makes @var{symbol}'s function cell void, so that a
subsequent attempt to access this cell will cause a
@code{void-function} error.  It returns @var{symbol}.  (See also
@code{makunbound}, in @ref{Void Variables}.)

@example
@group
(defun foo (x) x)
(foo 1)
     @result{}1
@end group
@group
(fmakunbound 'foo)
     @result{} foo
@end group
@group
(foo 1)
@error{} Symbol's function definition is void: foo
@end group
@end example
@end defun

@defun fset symbol definition
This function stores @var{definition} in the function cell of
@var{symbol}.  The result is @var{definition}.  Normally
@var{definition} should be a function or the name of a function, but
this is not checked.  The argument @var{symbol} is an ordinary evaluated
argument.

1086 1087 1088 1089
The primary use of this function is as a subroutine by constructs that define
or alter functions, like @code{defun} or @code{advice-add} (@pxref{Advising
Functions}).  You can also use it to give a symbol a function definition that
is not a function, e.g., a keyboard macro (@pxref{Keyboard Macros}):
Glenn Morris's avatar
Glenn Morris committed
1090

1091 1092 1093 1094 1095
@example
;; @r{Define a named keyboard macro.}
(fset 'kill-two-lines "\^u2\^k")
     @result{} "\^u2\^k"
@end example
Glenn Morris's avatar
Glenn Morris committed
1096

1097 1098 1099 1100
It you wish to use @code{fset} to make an alternate name for a
function, consider using @code{defalias} instead.  @xref{Definition of
defalias}.
@end defun
Glenn Morris's avatar
Glenn Morris committed
1101

1102 1103
@node Closures
@section Closures
Glenn Morris's avatar
Glenn Morris committed
1104

1105 1106
  As explained in @ref{Variable Scoping}, Emacs can optionally enable
lexical binding of variables.  When lexical binding is enabled, any
1107
named function that you create (e.g., with @code{defun}), as well as
1108 1109 1110
any anonymous function that you create using the @code{lambda} macro
or the @code{function} special form or the @code{#'} syntax
(@pxref{Anonymous Functions}), is automatically converted into a
1111
@dfn{closure}.
Glenn Morris's avatar
Glenn Morris committed
1112

1113
@cindex closure
1114 1115 1116 1117 1118 1119
  A closure is a function that also carries a record of the lexical
environment that existed when the function was defined.  When it is
invoked, any lexical variable references within its definition use the
retained lexical environment.  In all other respects, closures behave
much like ordinary functions; in particular, they can be called in the
same way as ordinary functions.
Glenn Morris's avatar
Glenn Morris committed
1120

1121
  @xref{Lexical Binding}, for an example of using a closure.
Glenn Morris's avatar
Glenn Morris committed
1122

1123 1124 1125 1126
  Currently, an Emacs Lisp closure object is represented by a list
with the symbol @code{closure} as the first element, a list
representing the lexical environment as the second element, and the
argument list and body forms as the remaining elements:
Glenn Morris's avatar
Glenn Morris committed
1127

1128 1129 1130 1131
@example
;; @r{lexical binding is enabled.}
(lambda (x) (* x x))
     @result{} (closure (t) (x) (* x x))
Glenn Morris's avatar
Glenn Morris committed
1132 1133
@end example

1134 1135 1136 1137 1138
@noindent
However, the fact that the internal structure of a closure is
``exposed'' to the rest of the Lisp world is considered an internal
implementation detail.  For this reason, we recommend against directly
examining or altering the structure of closure objects.
Glenn Morris's avatar
Glenn Morris committed
1139

1140 1141 1142 1143 1144
@node Advising Functions
@section Advising Emacs Lisp Functions
@cindex advising functions
@cindex piece of advice

1145 1146 1147 1148 1149 1150
When you need to modify a function defined in another library, or when you need
to modify a hook like @code{@var{foo}-function}, a process filter, or basically
any variable or object field which holds a function value, you can use the
appropriate setter function, such as @code{fset} or @code{defun} for named
functions, @code{setq} for hook variables, or @code{set-process-filter} for
process filters, but those are often too blunt, completely throwing away the
1151 1152
previous value.

1153 1154 1155
  The @dfn{advice} feature lets you add to the existing definition of
a function, by @dfn{advising the function}.  This is a cleaner method
than redefining the whole function.
1156

1157 1158 1159 1160 1161 1162 1163 1164
Emacs's advice system provides two sets of primitives for that: the core set,
for function values held in variables and object fields (with the corresponding
primitives being @code{add-function} and @code{remove-function}) and another
set layered on top of it for named functions (with the main primitives being
@code{advice-add} and @code{advice-remove}).

For example, in order to trace the calls to the process filter of a process
@var{proc}, you could use:
1165 1166

@example
1167 1168 1169 1170
(defun my-tracing-function (proc string)
  (message "Proc %S received %S" proc string))

(add-function :before (process-filter @var{proc}) #'my-tracing-function)
1171 1172
@end example

1173 1174 1175 1176
This will cause the process's output to be passed to @code{my-tracing-function}
before being passed to the original process filter.  @code{my-tracing-function}
receives the same arguments as the original function.  When you're done with
it, you can revert to the untraced behavior with:
1177 1178

@example
1179
(remove-function (process-filter @var{proc}) #'my-tracing-function)
1180 1181
@end example

1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193
Similarly, if you want to trace the execution of the function named
@code{display-buffer}, you could use:

@example
(defun his-tracing-function (orig-fun &rest args)
  (message "display-buffer called with args %S" args)
  (let ((res (apply orig-fun args)))
    (message "display-buffer returned %S" res)
    res))

(advice-add 'display-buffer :around #'his-tracing-function)
@end example
1194

1195 1196 1197 1198
Here, @code{his-tracing-function} is called instead of the original function
and receives the original function (additionally to that function's arguments)
as argument, so it can call it if and when it needs to.
When you're tired of seeing this output, you can revert to the untraced
1199 1200 1201 1202 1203 1204
behavior with:

@example
(advice-remove 'display-buffer #'his-tracing-function)
@end example

1205
The arguments @code{:before} and @code{:around} used in the above examples
1206 1207
specify how the two functions are composed, since there are many different
ways to do it.  The added function is also called an @emph{advice}.
1208 1209

@menu
1210 1211 1212 1213
* Core Advising Primitives::    Primitives to manipulate advice.
* Advising Named Functions::    Advising named functions.
* Advice combinators::          Ways to compose advices.
* Porting old advices::         Adapting code using the old defadvice.
1214 1215
@end menu

1216 1217
@node Core Advising Primitives
@subsection Primitives to manipulate advices
1218
@cindex advice, add and remove
1219 1220 1221 1222 1223 1224

@defmac add-function where place function &optional props
This macro is the handy way to add the advice @var{function} to the function
stored in @var{place} (@pxref{Generalized Variables}).

@var{where} determines how @var{function} is composed with the
1225
existing function, e.g. whether @var{function} should be called before, or
1226
after the original function.  @xref{Advice combinators}, for the list of
1227
available ways to compose the two functions.
1228 1229 1230 1231 1232 1233 1234

When modifying a variable (whose name will usually end with @code{-function}),
you can choose whether @var{function} is used globally or only in the current
buffer: if @var{place} is just a symbol, then @var{function} is added to the
global value of @var{place}.  Whereas if @var{place} is of the form
@code{(local @var{symbol})}, where @var{symbol} is an expression which returns
the variable name, then @var{function} will only be added in the
1235
current buffer.  Finally, if you want to modify a lexical variable, you will
Glenn Morris's avatar
Glenn Morris committed
1236
have to use @code{(var @var{variable})}.
1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248

Every function added with @code{add-function} can be accompanied by an
association list of properties @var{props}.  Currently only two of those
properties have a special meaning:

@table @code
@item name
This gives a name to the advice, which @code{remove-function} can use to
identify which function to remove.  Typically used when @var{function} is an
anonymous function.

@item depth
1249
This specifies how to order the advices, in case several advices are present.
1250 1251 1252 1253
By default, the depth is 0.  A depth of 100 indicates that this advice should
be kept as deep as possible, whereas a depth of -100 indicates that it
should stay as the outermost advice.  When two advices specify the same depth,
the most recently added advice will be outermost.
1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264

For a @code{:before} advice, being outermost means that this advice will be run
first, before any other advice, whereas being innermost means that it will run
right before the original function, with no other advice run between itself and
the original function.  Similarly, for an @code{:after} advice innermost means
that it will run right after the original function, with no other advice run in
between, whereas outermost means that it will be run very last after all
other advices.  An innermost @code{:override} advice will only override the
original function and other advices will apply to it, whereas an outermost
@code{:override} advice will override not only the original function but all
other advices applied to it as well.
1265
@end table
1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280

If @var{function} is not interactive, then the combined function will inherit
the interactive spec, if any, of the original function.  Else, the combined
function will be interactive and will use the interactive spec of
@var{function}.  One exception: if the interactive spec of @var{function}
is a function (rather than an expression or a string), then the interactive
spec of the combined function will be a call to that function with as sole
argument the interactive spec of the original function.  To interpret the spec
received as argument, use @code{advice-eval-interactive-spec}.

Note: The interactive spec of @var{function} will apply to the combined
function and should hence obey the calling convention of the combined function
rather than that of @var{function}.  In many cases, it makes no difference
since they are identical, but it does matter for @code{:around},
@code{:filter-args}, and @code{filter-return}, where @var{function}.
1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306
@end defmac

@defmac remove-function place function
This macro removes @var{function} from the function stored in
@var{place}.  This only works if @var{function} was added to @var{place}
using @code{add-function}.

@var{function} is compared with functions added to @var{place} using
@code{equal}, to try and make it work also with lambda expressions.  It is
additionally compared also with the @code{name} property of the functions added
to @var{place}, which can be more reliable than comparing lambda expressions
using @code{equal}.
@end defmac

@defun advice-function-member-p advice function-def
Return non-@code{nil} if @var{advice} is already in @var{function-def}.
Like for @code{remove-function} above, instead of @var{advice} being the actual
function, it can also be the @code{name} of the piece of advice.
@end defun

@defun advice-function-mapc f function-def
Call the function @var{f} for every advice that was added to
@var{function-def}.  @var{f} is called with two arguments: the advice function
and its properties.
@end defun

1307 1308 1309 1310 1311 1312 1313 1314
@defun advice-eval-interactive-spec spec
Evaluate the interactive @var{spec} just like an interactive call to a function
with such a spec would, and then return the corresponding list of arguments
that was built.  E.g. @code{(advice-eval-interactive-spec "r\nP")} will
return a list of three elements, containing the boundaries of the region and
the current prefix argument.
@end defun

1315 1316
@node Advising Named Functions
@subsection Advising Named Functions
1317
@cindex advising named functions
1318 1319

A common use of advice is for named functions and macros.
1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340
You could just use @code{add-function} as in:

@example
(add-function :around (symbol-function '@var{fun}) #'his-tracing-function)
@end example

  But you should use @code{advice-add} and @code{advice-remove} for that
instead.  This separate set of functions to manipulate pieces of advice applied
to named functions, offers the following extra features compared to
@code{add-function}: they know how to deal with macros and autoloaded
functions, they let @code{describe-function} preserve the original docstring as
well as document the added advice, and they let you add and remove advices
before a function is even defined.

  @code{advice-add} can be useful for altering the behavior of existing calls
to an existing function without having to redefine the whole function.
However, it can be a source of bugs, since existing callers to the function may
assume the old behavior, and work incorrectly when the behavior is changed by
advice.  Advice can also cause confusion in debugging, if the person doing the
debugging does not notice or remember that the function has been modified
by advice.
1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351

  For these reasons, advice should be reserved for the cases where you
cannot modify a function's behavior in any other way.  If it is
possible to do the same thing via a hook, that is preferable
(@pxref{Hooks}).  If you simply want to change what a particular key
does, it may be better to write a new command, and remap the old
command's key bindings to the new one (@pxref{Remapping Commands}).
In particular, Emacs's own source files should not put advice on
functions in Emacs.  (There are currently a few exceptions to this
convention, but we aim to correct them.)

1352 1353 1354 1355
  Special forms (@pxref{Special Forms}) cannot be advised, however macros can
be advised, in much the same way as functions.  Of course, this will not affect
code that has already been macro-expanded, so you need to make sure the advice
is installed before the macro is expanded.
1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367

  It is possible to advise a primitive (@pxref{What Is a Function}),
but one should typically @emph{not} do so, for two reasons.  Firstly,
some primitives are used by the advice mechanism, and advising them
could cause an infinite recursion.  Secondly, many primitives are
called directly from C, and such calls ignore advice; hence, one ends
up in a confusing situation where some calls (occurring from Lisp
code) obey the advice and other calls (from C code) do not.

@defun advice-add symbol where function &optional props
Add the advice @var{function} to the named function @var{symbol}.
@var{where} and @var{props} have the same meaning as for @code{add-function}
1368
(@pxref{Core Advising Primitives}).
1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387
@end defun

@defun advice-remove symbol function
Remove the advice @var{function} from the named function @var{symbol}.
@var{function} can also be the @code{name} of an advice.
@end defun

@defun advice-member-p function symbol
Return non-@code{nil} if the advice @var{function} is already in the named
function @var{symbol}.  @var{function} can also be the @code{name} of
an advice.
@end defun

@defun advice-mapc function symbol
Call @var{function} for every advice that was added to the named function
@var{symbol}.  @var{function} is called with two arguments: the advice function
and its properties.
@end defun

1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500
@node Advice combinators
@subsection Ways to compose advices

Here are the different possible values for the @var{where} argument of
@code{add-function} and @code{advice-add}, specifying how the advice
@var{function} and the original function should be composed.

@table @code
@item :before
Call @var{function} before the old function.  Both functions receive the
same arguments, and the return value of the composition is the return value of
the old function.  More specifically, the composition of the two functions
behaves like:
@example
(lambda (&rest r) (apply @var{function} r) (apply @var{oldfun} r))
@end example
@code{(add-function :before @var{funvar} @var{function})} is comparable for
single-function hooks to @code{(add-hook '@var{hookvar} @var{function})} for
normal hooks.

@item :after
Call @var{function} after the old function.  Both functions receive the
same arguments, and the return value of the composition is the return value of
the old function.  More specifically, the composition of the two functions
behaves like:
@example
(lambda (&rest r) (prog1 (apply @var{oldfun} r) (apply @var{function} r)))
@end example
@code{(add-function :after @var{funvar} @var{function})} is comparable for
single-function hooks to @code{(add-hook '@var{hookvar} @var{function}
'append)} for normal hooks.

@item :override
This completely replaces the old function with the new one.  The old function
can of course be recovered if you later call @code{remove-function}.

@item :around
Call @var{function} instead of the old function, but provide the old function
as an extra argument to @var{function}.  This is the most flexible composition.
For example, it lets you call the old function with different arguments, or
many times, or within a let-binding, or you can sometimes delegate the work to
the old function and sometimes override it completely.  More specifically, the
composition of the two functions behaves like:
@example
(lambda (&rest r) (apply @var{function} @var{oldfun} r))
@end example

@item :before-while
Call @var{function} before the old function and don't call the old
function if @var{function} returns @code{nil}.  Both functions receive the
same arguments, and the return value of the composition is the return value of
the old function.  More specifically, the composition of the two functions
behaves like:
@example
(lambda (&rest r) (and (apply @var{function} r) (apply @var{oldfun} r)))
@end example
@code{(add-function :before-while @var{funvar} @var{function})} is comparable
for single-function hooks to @code{(add-hook '@var{hookvar} @var{function})}
when @var{hookvar} is run via @code{run-hook-with-args-until-failure}.

@item :before-until
Call @var{function} before the old function and only call the old function if
@var{function} returns @code{nil}.  More specifically, the composition of the
two functions behaves like:
@example
(lambda (&rest r) (or (apply @var{function} r) (apply @var{oldfun} r)))
@end example
@code{(add-function :before-until @var{funvar} @var{function})} is comparable
for single-function hooks to @code{(add-hook '@var{hookvar} @var{function})}
when @var{hookvar} is run via @code{run-hook-with-args-until-success}.

@item :after-while
Call @var{function} after the old function and only if the old function
returned non-@code{nil}.  Both functions receive the same arguments, and the
return value of the composition is the return value of @var{function}.
More specifically, the composition of the two functions behaves like:
@example
(lambda (&rest r) (and (apply @var{oldfun} r) (apply @var{function} r)))
@end example
@code{(add-function :after-while @var{funvar} @var{function})} is comparable
for single-function hooks to @code{(add-hook '@var{hookvar} @var{function}
'append)} when @var{hookvar} is run via
@code{run-hook-with-args-until-failure}.

@item :after-until
Call @var{function} after the old function and only if the old function
returned @code{nil}.  More specifically, the composition of the two functions
behaves like:
@example
(lambda (&rest r) (or  (apply @var{oldfun} r) (apply @var{function} r)))
@end example
@code{(add-function :after-until @var{funvar} @var{function})} is comparable
for single-function hooks to @code{(add-hook '@var{hookvar} @var{function}
'append)} when @var{hookvar} is run via
@code{run-hook-with-args-until-success}.

@item :filter-args
Call @var{function} first and use the result (which should be a list) as the
new arguments to pass to the old function.  More specifically, the composition
of the two functions behaves like:
@example
(lambda (&rest r) (apply @var{oldfun} (funcall @var{function} r)))
@end example

@item :filter-return
Call the old function first and pass the result to @var{function}.
More specifically, the composition of the two functions behaves like:
@example
(lambda (&rest r) (funcall @var{function} (apply @var{oldfun} r)))
@end example
@end table


1501 1502
@node Porting old advices
@subsection Adapting code using the old defadvice
1503
@cindex old advices, porting
1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583

A lot of code uses the old @code{defadvice} mechanism, which is largely made
obsolete by the new @code{advice-add}, whose implementation and semantics is
significantly simpler.

An old advice such as:

@example
(defadvice previous-line (before next-line-at-end
                                 (&optional arg try-vscroll))
  "Insert an empty line when moving up from the top line."
  (if (and next-line-add-newlines (= arg 1)
           (save-excursion (beginning-of-line) (bobp)))
      (progn
        (beginning-of-line)
        (newline))))
@end example

could be translated in the new advice mechanism into a plain function:

@example
(defun previous-line--next-line-at-end (&optional arg try-vscroll)
  "Insert an empty line when moving up from the top line."
  (if (and next-line-add-newlines (= arg 1)
           (save-excursion (beginning-of-line) (bobp)))
      (progn
        (beginning-of-line)
        (newline))))
@end example

Obviously, this does not actually modify @code{previous-line}.  For that the
old advice needed:
@example
(ad-activate 'previous-line)
@end example
whereas the new advice mechanism needs:
@example
(advice-add 'previous-line :before #'previous-line--next-line-at-end)
@end example

Note that @code{ad-activate} had a global effect: it activated all pieces of
advice enabled for that specified function.  If you wanted to only activate or
deactivate a particular advice, you needed to @emph{enable} or @emph{disable}
that advice with @code{ad-enable-advice} and @code{ad-disable-advice}.
The new mechanism does away with this distinction.

An around advice such as:

@example
(defadvice foo (around foo-around)
  "Ignore case in `foo'."
  (let ((case-fold-search t))
    ad-do-it))
(ad-activate 'foo)
@end example

could translate into:

@example
(defun foo--foo-around (orig-fun &rest args)
  "Ignore case in `foo'."
  (let ((case-fold-search t))
    (apply orig-fun args)))
(advice-add 'foo :around #'foo--foo-around)
@end example

Regarding the advice's @emph{class}, note that the new @code{:before} is not
quite equivalent to the old @code{before}, because in the old advice you could
modify the function's arguments (e.g., with @code{ad-set-arg}), and that would
affect the argument values seen by the original function, whereas in the new
@code{:before}, modifying an argument via @code{setq} in the advice has no
effect on the arguments seen by the original function.
When porting a @code{before} advice which relied on this behavior, you'll need
to turn it into a new @code{:around} or @code{:filter-args} advice instead.

Similarly an old @code{after} advice could modify the returned value by
changing @code{ad-return-value}, whereas a new @code{:after} advice cannot, so
when porting such an old @code{after} advice, you'll need to turn it into a new
@code{:around} or @code{:filter-return} advice instead.

Glenn Morris's avatar
Glenn Morris committed
1584 1585
@node Obsolete Functions
@section Declaring Functions Obsolete
1586
@cindex obsolete functions
Glenn Morris's avatar
Glenn Morris committed
1587

1588 1589 1590 1591 1592 1593 1594