Make links to mac/make-package and

[emacs.git] / etc / NEWS.21

GNU Emacs NEWS -- history of user-visible changes.  2006-05-31
Copyright (C) 2000, 2001, 2002, 2003, 2006
          Free Software Foundation, Inc.
See the end for copying conditions.

This file is about changes in emacs version 21.


\f
* Emacs 21.4 is a bug-fix release with no user-visible changes.


\f
* Installation changes in Emacs 21.3

** Support for GNU/Linux on little-endian MIPS and on IBM S390 has
been added.

\f
* Changes in Emacs 21.3

** The obsolete C mode (c-mode.el) has been removed to avoid problems
with Custom.

** UTF-16 coding systems are available, encoding the same characters
as mule-utf-8.

** There is a new language environment for UTF-8 (set up automatically
in UTF-8 locales).

** Translation tables are available between equivalent characters in
different Emacs charsets -- for instance `e with acute' coming from the
Latin-1 and Latin-2 charsets.  User options `unify-8859-on-encoding-mode'
and `unify-8859-on-decoding-mode' respectively turn on translation
between ISO 8859 character sets (`unification') on encoding
(e.g. writing a file) and decoding (e.g. reading a file).  Note that
`unify-8859-on-encoding-mode' is useful and safe, but
`unify-8859-on-decoding-mode' can cause text to change when you read
it and write it out again without edits, so it is not generally advisable.
By default `unify-8859-on-encoding-mode' is turned on.

** In Emacs running on the X window system, the default value of
`selection-coding-system' is now `compound-text-with-extensions'.

If you want the old behavior, set selection-coding-system to
compound-text, which may be significantly more efficient.  Using
compound-text-with-extensions seems to be necessary only for decoding
text from applications under XFree86 4.2, whose behavior is actually
contrary to the compound text specification.


\f
* Installation changes in Emacs 21.2

** Support for BSD/OS 5.0 has been added.

** Support for AIX 5.1 was added.

\f
* Changes in Emacs 21.2

** Emacs now supports compound-text extended segments in X selections.

X applications can use `extended segments' to encode characters in
compound text that belong to character sets which are not part of the
list of approved standard encodings for X, e.g. Big5.  To paste
selections with such characters into Emacs, use the new coding system
compound-text-with-extensions as the value of selection-coding-system.

** The default values of `tooltip-delay' and `tooltip-hide-delay'
were changed.

** On terminals whose erase-char is ^H (Backspace), Emacs
now uses normal-erase-is-backspace-mode.

** When the *scratch* buffer is recreated, its mode is set from
initial-major-mode, which normally is lisp-interaction-mode,
instead of using default-major-mode.

** The new option `Info-scroll-prefer-subnodes' causes Info to behave
like the stand-alone Info reader (from the GNU Texinfo package) as far
as motion between nodes and their subnodes is concerned.  If it is t
(the default), Emacs behaves as before when you type SPC in a menu: it
visits the subnode pointed to by the first menu entry.  If this option
is nil, SPC scrolls to the end of the current node, and only then goes
to the first menu item, like the stand-alone reader does.

This change was already in Emacs 21.1, but wasn't advertised in the
NEWS.

\f
* Lisp Changes in Emacs 21.2

** The meanings of scroll-up-aggressively and scroll-down-aggressively
have been interchanged, so that the former now controls scrolling up,
and the latter now controls scrolling down.

** The variable `compilation-parse-errors-filename-function' can
be used to transform filenames found in compilation output.


\f
* Installation Changes in Emacs 21.1

See the INSTALL file for information on installing extra libraries and
fonts to take advantage of the new graphical features and extra
charsets in this release.

** Support for GNU/Linux on IA64 machines has been added.

** Support for LynxOS has been added.

** There are new configure options associated with the support for
images and toolkit scrollbars.  Use the --help option in `configure'
to list them.

** You can build a 64-bit Emacs for SPARC/Solaris systems which
support 64-bit executables and also on Irix 6.5.  This increases the
maximum buffer size.  See etc/MACHINES for instructions.  Changes to
build on other 64-bit systems should be straightforward modulo any
necessary changes to unexec.

** There is a new configure option `--disable-largefile' to omit
Unix-98-style support for large files if that is available.

** There is a new configure option `--without-xim' that instructs
Emacs to not use X Input Methods (XIM), if these are available.

** `movemail' defaults to supporting POP.  You can turn this off using
the --without-pop configure option, should that be necessary.

** This version can be built for the Macintosh, but does not implement
all of the new display features described below.  The port currently
lacks unexec, asynchronous processes, and networking support.  See the
"Emacs and the Mac OS" appendix in the Emacs manual, for the
description of aspects specific to the Mac.

** Note that the MS-Windows port does not yet implement various of the
new display features described below.

\f
* Changes in Emacs 21.1

** Emacs has a new redisplay engine.

The new redisplay handles characters of variable width and height.
Italic text can be used without redisplay problems.  Fonts containing
oversized characters, i.e. characters larger than the logical height
of a font can be used.  Images of various formats can be displayed in
the text.

** Emacs has a new face implementation.

The new faces no longer fundamentally use X font names to specify the
font.  Instead, each face has several independent attributes--family,
height, width, weight and slant--that it may or may not specify.
These attributes can be merged from various faces, and then together
specify a font.

Faces are supported on terminals that can display color or fonts.
These terminal capabilities are auto-detected.  Details can be found
under Lisp changes, below.

** Emacs can display faces on TTY frames.

Emacs automatically detects terminals that are able to display colors.
Faces with a weight greater than normal are displayed extra-bright, if
the terminal supports it.  Faces with a weight less than normal and
italic faces are displayed dimmed, if the terminal supports it.
Underlined faces are displayed underlined if possible.  Other face
attributes such as `overline', `strike-through', and `box' are ignored
on terminals.

The command-line options `-fg COLOR', `-bg COLOR', and `-rv' are now
supported on character terminals.

Emacs automatically remaps all X-style color specifications to one of
the colors supported by the terminal.  This means you could have the
same color customizations that work both on a windowed display and on
a TTY or when Emacs is invoked with the -nw option.

** New default font is Courier 12pt under X.

** Sound support

Emacs supports playing sound files on GNU/Linux and FreeBSD (Voxware
driver and native BSD driver, a.k.a. Luigi's driver).  Currently
supported file formats are RIFF-WAVE (*.wav) and Sun Audio (*.au).
You must configure Emacs with the option `--with-sound=yes' to enable
sound support.

** Emacs now resizes mini-windows if appropriate.

If a message is longer than one line, or minibuffer contents are
longer than one line, Emacs can resize the minibuffer window unless it
is on a frame of its own.  You can control resizing and the maximum
minibuffer window size by setting the following variables:

- User option: max-mini-window-height

Maximum height for resizing mini-windows.  If a float, it specifies a
fraction of the mini-window frame's height.  If an integer, it
specifies a number of lines.

Default is 0.25.

- User option: resize-mini-windows

How to resize mini-windows.  If nil, don't resize.  If t, always
resize to fit the size of the text.  If `grow-only', let mini-windows
grow only, until they become empty, at which point they are shrunk
again.

Default is `grow-only'.

** LessTif support.

Emacs now runs with the LessTif toolkit (see
<http://www.lesstif.org>).  You will need version 0.92.26, or later.

** LessTif/Motif file selection dialog.

When Emacs is configured to use LessTif or Motif, reading a file name
from a menu will pop up a file selection dialog if `use-dialog-box' is
non-nil.

** File selection dialog on MS-Windows is supported.

When a file is visited by clicking File->Open, the MS-Windows version
now pops up a standard file selection dialog where you can select a
file to visit.  File->Save As also pops up that dialog.

** Toolkit scroll bars.

Emacs now uses toolkit scroll bars if available.  When configured for
LessTif/Motif, it will use that toolkit's scroll bar.  Otherwise, when
configured for Lucid and Athena widgets, it will use the Xaw3d scroll
bar if Xaw3d is available.  You can turn off the use of toolkit scroll
bars by specifying `--with-toolkit-scroll-bars=no' when configuring
Emacs.

When you encounter problems with the Xaw3d scroll bar, watch out how
Xaw3d is compiled on your system.  If the Makefile generated from
Xaw3d's Imakefile contains a `-DNARROWPROTO' compiler option, and your
Emacs system configuration file `s/your-system.h' does not contain a
define for NARROWPROTO, you might consider adding it.  Take
`s/freebsd.h' as an example.

Alternatively, if you don't have access to the Xaw3d source code, take
a look at your system's imake configuration file, for example in the
directory `/usr/X11R6/lib/X11/config' (paths are different on
different systems).  You will find files `*.cf' there.  If your
system's cf-file contains a line like `#define NeedWidePrototypes NO',
add a `#define NARROWPROTO' to your Emacs system configuration file.

The reason for this is that one Xaw3d function uses `double' or
`float' function parameters depending on the setting of NARROWPROTO.
This is not a problem when Imakefiles are used because each system's
imake configuration file contains the necessary information.  Since
Emacs doesn't use imake, this has do be done manually.

** Tool bar support.

Emacs supports a tool bar at the top of a frame under X.  For details
of how to define a tool bar, see the page describing Lisp-level
changes.  Tool-bar global minor mode controls whether or not it is
displayed and is on by default.  The appearance of the bar is improved
if Emacs has been built with XPM image support.  Otherwise monochrome
icons will be used.

To make the tool bar more useful, we need contributions of extra icons
for specific modes (with copyright assignments).

** Tooltips.

Tooltips are small X windows displaying a help string at the current
mouse position.  The Lisp package `tooltip' implements them.  You can
turn them off via the user option `tooltip-mode'.

Tooltips also provides support for GUD debugging.  If activated,
variable values can be displayed in tooltips by pointing at them with
the mouse in source buffers.  You can customize various aspects of the
tooltip display in the group `tooltip'.

** Automatic Hscrolling

Horizontal scrolling now happens automatically if
`automatic-hscrolling' is set (the default).  This setting can be
customized.

If a window is scrolled horizontally with set-window-hscroll, or
scroll-left/scroll-right (C-x <, C-x >), this serves as a lower bound
for automatic horizontal scrolling.  Automatic scrolling will scroll
the text more to the left if necessary, but won't scroll the text more
to the right than the column set with set-window-hscroll etc.

** When using a windowing terminal, each Emacs window now has a cursor
of its own.  By default, when a window is selected, the cursor is
solid; otherwise, it is hollow.  The user-option
`cursor-in-non-selected-windows' controls how to display the
cursor in non-selected windows.  If nil, no cursor is shown, if
non-nil a hollow box cursor is shown.

** Fringes to the left and right of windows are used to display
truncation marks, continuation marks, overlay arrows and alike.  The
foreground, background, and stipple of these areas can be changed by
customizing face `fringe'.

** The mode line under X is now drawn with shadows by default.
You can change its appearance by modifying the face `mode-line'.
In particular, setting the `:box' attribute to nil turns off the 3D
appearance of the mode line.  (The 3D appearance makes the mode line
occupy more space, and thus might cause the first or the last line of
the window to be partially obscured.)

The variable `mode-line-inverse-video', which was used in older
versions of emacs to make the mode-line stand out, is now deprecated.
However, setting it to nil will cause the `mode-line' face to be
ignored, and mode-lines to be drawn using the default text face.

** Mouse-sensitive mode line.

Different parts of the mode line have been made mouse-sensitive on all
systems which support the mouse.  Moving the mouse to a
mouse-sensitive part in the mode line changes the appearance of the
mouse pointer to an arrow, and help about available mouse actions is
displayed either in the echo area, or in the tooltip window if you
have enabled one.

Currently, the following actions have been defined:

- Mouse-1 on the buffer name in the mode line goes to the next buffer.

- Mouse-3 on the buffer-name goes to the previous buffer.

- Mouse-2 on the read-only or modified status in the mode line (`%' or
`*') toggles the status.

- Mouse-3 on the major mode name displays a major mode menu.

- Mouse-3 on the mode name displays a minor-mode menu.

** Hourglass pointer

Emacs can optionally display an hourglass pointer under X.  You can
turn the display on or off by customizing group `cursor'.

** Blinking cursor

M-x blink-cursor-mode toggles a blinking cursor under X and on
terminals having terminal capabilities `vi', `vs', and `ve'.  Blinking
and related parameters like frequency and delay can be customized in
the group `cursor'.

** New font-lock support mode `jit-lock-mode'.

This support mode is roughly equivalent to `lazy-lock' but is
generally faster.  It supports stealth and deferred fontification.
See the documentation of the function `jit-lock-mode' for more
details.

Font-lock uses jit-lock-mode as default support mode, so you don't
have to do anything to activate it.

** The default binding of the Delete key has changed.

The new user-option `normal-erase-is-backspace' can be set to
determine the effect of the Delete and Backspace function keys.

On window systems, the default value of this option is chosen
according to the keyboard used.  If the keyboard has both a Backspace
key and a Delete key, and both are mapped to their usual meanings, the
option's default value is set to t, so that Backspace can be used to
delete backward, and Delete can be used to delete forward.  On
keyboards which either have only one key (usually labeled DEL), or two
keys DEL and BS which produce the same effect, the option's value is
set to nil, and these keys delete backward.

If not running under a window system, setting this option accomplishes
a similar effect by mapping C-h, which is usually generated by the
Backspace key, to DEL, and by mapping DEL to C-d via
`keyboard-translate'.  The former functionality of C-h is available on
the F1 key.  You should probably not use this setting on a text-only
terminal if you don't have both Backspace, Delete and F1 keys.

Programmatically, you can call function normal-erase-is-backspace-mode
to toggle the behavior of the Delete and Backspace keys.

** The default for user-option `next-line-add-newlines' has been
changed to nil, i.e. C-n will no longer add newlines at the end of a
buffer by default.

** The <home> and <end> keys now move to the beginning or end of the
current line, respectively.  C-<home> and C-<end> move to the
beginning and end of the buffer.

** Emacs now checks for recursive loads of Lisp files.  If the
recursion depth exceeds `recursive-load-depth-limit', an error is
signaled.

** When an error is signaled during the loading of the user's init
file, Emacs now pops up the *Messages* buffer.

** Emacs now refuses to load compiled Lisp files which weren't
compiled with Emacs.  Set `load-dangerous-libraries' to t to change
this behavior.

The reason for this change is an incompatible change in XEmacs's byte
compiler.  Files compiled with XEmacs can contain byte codes that let
Emacs dump core.

** Toggle buttons and radio buttons in menus.

When compiled with LessTif (or Motif) support, Emacs uses toolkit
widgets for radio and toggle buttons in menus.  When configured for
Lucid, Emacs draws radio buttons and toggle buttons similar to Motif.

** The menu bar configuration has changed.  The new configuration is
more CUA-compliant.  The most significant change is that Options is
now a separate menu-bar item, with Mule and Customize as its submenus.

** Item Save Options on the Options menu allows saving options set
using that menu.

** Highlighting of trailing whitespace.

When `show-trailing-whitespace' is non-nil, Emacs displays trailing
whitespace in the face `trailing-whitespace'.  Trailing whitespace is
defined as spaces or tabs at the end of a line.  To avoid busy
highlighting when entering new text, trailing whitespace is not
displayed if point is at the end of the line containing the
whitespace.

** C-x 5 1 runs the new command delete-other-frames which deletes
all frames except the selected one.

** The new user-option `confirm-kill-emacs' can be customized to
let Emacs ask for confirmation before exiting.

** The header line in an Info buffer is now displayed as an emacs
header-line (which is like a mode-line, but at the top of the window),
so that it remains visible even when the buffer has been scrolled.
This behavior may be disabled by customizing the option
`Info-use-header-line'.

** Polish, Czech, German, and French translations of Emacs' reference card
have been added.  They are named `pl-refcard.tex', `cs-refcard.tex',
`de-refcard.tex' and `fr-refcard.tex'.  Postscript files are included.

** An `Emacs Survival Guide', etc/survival.tex, is available.

** A reference card for Dired has been added.  Its name is
`dired-ref.tex'.  A French translation is available in
`fr-drdref.tex'.

** C-down-mouse-3 is bound differently.  Now if the menu bar is not
displayed it pops up a menu containing the items which would be on the
menu bar.  If the menu bar is displayed, it pops up the major mode
menu or the Edit menu if there is no major mode menu.

** Variable `load-path' is no longer customizable through Customize.

You can no longer use `M-x customize-variable' to customize `load-path'
because it now contains a version-dependent component.  You can still
use `add-to-list' and `setq' to customize this variable in your
`~/.emacs' init file or to modify it from any Lisp program in general.

** C-u C-x = provides detailed information about the character at
point in a pop-up window.

** Emacs can now support 'wheeled' mice (such as the MS IntelliMouse)
under XFree86.  To enable this, use the `mouse-wheel-mode' command, or
customize the variable `mouse-wheel-mode'.

The variables `mouse-wheel-follow-mouse' and `mouse-wheel-scroll-amount'
determine where and by how much buffers are scrolled.

** Emacs' auto-save list files are now by default stored in a
sub-directory `.emacs.d/auto-save-list/' of the user's home directory.
(On MS-DOS, this subdirectory's name is `_emacs.d/auto-save.list/'.)
You can customize `auto-save-list-file-prefix' to change this location.

** The function `getenv' is now callable interactively.

** The new user-option `even-window-heights' can be set to nil
to prevent `display-buffer' from evening out window heights.

** The new command M-x delete-trailing-whitespace RET will delete the
trailing whitespace within the current restriction.  You can also add
this function to `write-file-hooks' or `local-write-file-hooks'.

** When visiting a file with M-x find-file-literally, no newlines will
be added to the end of the buffer even if `require-final-newline' is
non-nil.

** The new user-option `find-file-suppress-same-file-warnings' can be
set to suppress warnings ``X and Y are the same file'' when visiting a
file that is already visited under a different name.

** The new user-option `electric-help-shrink-window' can be set to
nil to prevent adjusting the help window size to the buffer size.

** New command M-x describe-character-set reads a character set name
and displays information about that.

** The new variable `auto-mode-interpreter-regexp' contains a regular
expression matching interpreters, for file mode determination.

This regular expression is matched against the first line of a file to
determine the file's mode in `set-auto-mode' when Emacs can't deduce a
mode from the file's name.  If it matches, the file is assumed to be
interpreted by the interpreter matched by the second group of the
regular expression.  The mode is then determined as the mode
associated with that interpreter in `interpreter-mode-alist'.

** New function executable-make-buffer-file-executable-if-script-p is
suitable as an after-save-hook as an alternative to `executable-chmod'.

** The most preferred coding-system is now used to save a buffer if
buffer-file-coding-system is `undecided' and it is safe for the buffer
contents.  (The most preferred is set by set-language-environment or
by M-x prefer-coding-system.)  Thus if you visit an ASCII file and
insert a non-ASCII character from your current language environment,
the file will be saved silently with the appropriate coding.
Previously you would be prompted for a safe coding system.

** The many obsolete language `setup-...-environment' commands have
been removed -- use `set-language-environment'.

** The new Custom option `keyboard-coding-system' specifies a coding
system for keyboard input.

** New variable `inhibit-iso-escape-detection' determines if Emacs'
coding system detection algorithm should pay attention to ISO2022's
escape sequences.  If this variable is non-nil, the algorithm ignores
such escape sequences.  The default value is nil, and it is
recommended not to change it except for the special case that you
always want to read any escape code verbatim.  If you just want to
read a specific file without decoding escape codes, use C-x RET c
(`universal-coding-system-argument').  For instance, C-x RET c latin-1
RET C-x C-f filename RET.

** Variable `default-korean-keyboard' is initialized properly from the
environment variable `HANGUL_KEYBOARD_TYPE'.

** New command M-x list-charset-chars reads a character set name and
displays all characters in that character set.

** M-x set-terminal-coding-system (C-x RET t) now allows CCL-based
coding systems such as cpXXX and cyrillic-koi8.

** Emacs now attempts to determine the initial language environment
and preferred and locale coding systems systematically from the
LC_ALL, LC_CTYPE, and LANG environment variables during startup.

** New language environments `Polish', `Latin-8' and `Latin-9'.
Latin-8 and Latin-9 correspond respectively to the ISO character sets
8859-14 (Celtic) and 8859-15 (updated Latin-1, with the Euro sign).
GNU Intlfonts doesn't support these yet but recent X releases have
8859-15.  See etc/INSTALL for information on obtaining extra fonts.
There are new Leim input methods for Latin-8 and Latin-9 prefix (only)
and Polish `slash'.

** New language environments `Dutch' and `Spanish'.
These new environments mainly select appropriate translations
of the tutorial.

** In Ethiopic language environment, special key bindings for
function keys are changed as follows.  This is to conform to "Emacs
Lisp Coding Convention".

    new  command                            old-binding
    ---  -------                            -----------
    f3   ethio-fidel-to-sera-buffer         f5
    S-f3 ethio-fidel-to-sera-region         f5
    C-f3 ethio-fidel-to-sera-mail-or-marker f5

    f4   ethio-sera-to-fidel-buffer         unchanged
    S-f4 ethio-sera-to-fidel-region         unchanged
    C-f4 ethio-sera-to-fidel-mail-or-marker unchanged

    S-f5 ethio-toggle-punctuation           f3
    S-f6 ethio-modify-vowel                 f6
    S-f7 ethio-replace-space                f7
    S-f8 ethio-input-special-character      f8
    S-f9 ethio-replace-space                unchanged
    C-f9 ethio-toggle-space                 f2

** There are new Leim input methods.
New input methods "turkish-postfix", "turkish-alt-postfix",
"greek-mizuochi", "TeX", and "greek-babel" are now part of the Leim
package.

** The rule of input method "slovak" is slightly changed.  Now the
rules for translating "q" and "Q" to "`" (backquote) are deleted, thus
typing them inserts "q" and "Q" respectively.  Rules for translating
"=q", "+q", "=Q", and "+Q" to "`" are also deleted.  Now, to input
"`", you must type "=q".

** When your terminal can't display characters from some of the ISO
8859 character sets but can display Latin-1, you can display
more-or-less mnemonic sequences of ASCII/Latin-1 characters instead of
empty boxes (under a window system) or question marks (not under a
window system).  Customize the option `latin1-display' to turn this
on.

** M-; now calls comment-dwim which tries to do something clever based
on the context.  M-x kill-comment is now an alias to comment-kill,
defined in newcomment.el.  You can choose different styles of region
commenting with the variable `comment-style'.

** New user options `display-time-mail-face' and
`display-time-use-mail-icon' control the appearance of mode-line mail
indicator used by the display-time package.  On a suitable display the
indicator can be an icon and is mouse-sensitive.

** On window-systems, additional space can be put between text lines
on the display using several methods

- By setting frame parameter `line-spacing' to PIXELS.  PIXELS must be
a positive integer, and specifies that PIXELS number of pixels should
be put below text lines on the affected frame or frames.

- By setting X resource `lineSpacing', class `LineSpacing'.  This is
equivalent to specifying the frame parameter.

- By specifying `--line-spacing=N' or `-lsp N' on the command line.

- By setting buffer-local variable `line-spacing'.  The meaning is
the same, but applies to the a particular buffer only.

** The new command `clone-indirect-buffer' can be used to create
an indirect buffer that is a twin copy of the current buffer.  The
command `clone-indirect-buffer-other-window', bound to C-x 4 c,
does the same but displays the indirect buffer in another window.

** New user options `backup-directory-alist' and
`make-backup-file-name-function' control the placement of backups,
typically in a single directory or in an invisible sub-directory.

** New commands iso-iso2sgml and iso-sgml2iso convert between Latin-1
characters and the corresponding SGML (HTML) entities.

** New X resources recognized

*** The X resource `synchronous', class `Synchronous', specifies
whether Emacs should run in synchronous mode.  Synchronous mode
is useful for debugging X problems.

Example:

  emacs.synchronous: true

*** The X resource `visualClass, class `VisualClass', specifies the
visual Emacs should use.  The resource's value should be a string of
the form `CLASS-DEPTH', where CLASS is the name of the visual class,
and DEPTH is the requested color depth as a decimal number.  Valid
visual class names are

  TrueColor
  PseudoColor
  DirectColor
  StaticColor
  GrayScale
  StaticGray

Visual class names specified as X resource are case-insensitive, i.e.
`pseudocolor', `Pseudocolor' and `PseudoColor' all have the same
meaning.

The program `xdpyinfo' can be used to list the visual classes
supported on your display, and which depths they have.  If
`visualClass' is not specified, Emacs uses the display's default
visual.

Example:

  emacs.visualClass: TrueColor-8

*** The X resource `privateColormap', class `PrivateColormap',
specifies that Emacs should use a private colormap if it is using the
default visual, and that visual is of class PseudoColor.  Recognized
resource values are `true' or `on'.

Example:

  emacs.privateColormap: true

** Faces and frame parameters.

There are four new faces `scroll-bar', `border', `cursor' and `mouse'.
Setting the frame parameters `scroll-bar-foreground' and
`scroll-bar-background' sets foreground and background color of face
`scroll-bar' and vice versa.  Setting frame parameter `border-color'
sets the background color of face `border' and vice versa.  Likewise
for frame parameters `cursor-color' and face `cursor', and frame
parameter `mouse-color' and face `mouse'.

Changing frame parameter `font' sets font-related attributes of the
`default' face and vice versa.  Setting frame parameters
`foreground-color' or `background-color' sets the colors of the
`default' face and vice versa.

** New face `menu'.

The face `menu' can be used to change colors and font of Emacs' menus.

** New frame parameter `screen-gamma' for gamma correction.

The new frame parameter `screen-gamma' specifies gamma-correction for
colors.  Its value may be nil, the default, in which case no gamma
correction occurs, or a number > 0, usually a float, that specifies
the screen gamma of a frame's display.

PC monitors usually have a screen gamma of 2.2.  smaller values result
in darker colors.  You might want to try a screen gamma of 1.5 for LCD
color displays.  The viewing gamma Emacs uses is 0.4545. (1/2.2).

The X resource name of this parameter is `screenGamma', class
`ScreenGamma'.

** Tabs and variable-width text.

Tabs are now displayed with stretch properties; the width of a tab is
defined as a multiple of the normal character width of a frame, and is
independent of the fonts used in the text where the tab appears.
Thus, tabs can be used to line up text in different fonts.

** Enhancements of the Lucid menu bar

*** The Lucid menu bar now supports the resource "margin".

        emacs.pane.menubar.margin: 5

The default margin is 4 which makes the menu bar appear like the
LessTif/Motif one.

*** Arrows that indicate sub-menus are now drawn with shadows, as in
LessTif and Motif.

** A block cursor can be drawn as wide as the glyph under it under X.

As an example: if a block cursor is over a tab character, it will be
drawn as wide as that tab on the display.  To do this, set
`x-stretch-cursor' to a non-nil value.

** Empty display lines at the end of a buffer may be marked with a
bitmap (this is similar to the tilde displayed by vi and Less).

This behavior is activated by setting the buffer-local variable
`indicate-empty-lines' to a non-nil value.  The default value of this
variable is found in `default-indicate-empty-lines'.

** There is a new "aggressive" scrolling method.

When scrolling up because point is above the window start, if the
value of the buffer-local variable `scroll-up-aggressively' is a
number, Emacs chooses a new window start so that point ends up that
fraction of the window's height from the top of the window.

When scrolling down because point is below the window end, if the
value of the buffer-local variable `scroll-down-aggressively' is a
number, Emacs chooses a new window start so that point ends up that
fraction of the window's height from the bottom of the window.

** You can now easily create new *Info* buffers using either
M-x clone-buffer, C-u m <entry> RET or C-u g <entry> RET.
M-x clone-buffer can also be used on *Help* and several other special
buffers.

** The command `Info-search' now uses a search history.

** Listing buffers with M-x list-buffers (C-x C-b) now shows
abbreviated file names.  Abbreviations can be customized by changing
`directory-abbrev-alist'.

** A new variable, backup-by-copying-when-privileged-mismatch, gives
the highest file uid for which backup-by-copying-when-mismatch will be
forced on.  The assumption is that uids less than or equal to this
value are special uids (root, bin, daemon, etc.--not real system
users) and that files owned by these users should not change ownership,
even if your system policy allows users other than root to edit them.

The default is 200; set the variable to nil to disable the feature.

** The rectangle commands now avoid inserting undesirable spaces,
notably at the end of lines.

All these functions have been rewritten to avoid inserting unwanted
spaces, and an optional prefix now allows them to behave the old way.

** The function `replace-rectangle' is an alias for `string-rectangle'.

** The new command M-x string-insert-rectangle is like `string-rectangle',
but inserts text instead of replacing it.

** The new command M-x query-replace-regexp-eval acts like
query-replace-regexp, but takes a Lisp expression which is evaluated
after each match to get the replacement text.

** M-x query-replace recognizes a new command `e' (or `E') that lets
you edit the replacement string.

** The new command mail-abbrev-complete-alias, bound to `M-TAB'
(if you load the library `mailabbrev'), lets you complete mail aliases
in the text, analogous to lisp-complete-symbol.

** The variable `echo-keystrokes' may now have a floating point value.

** If your init file is compiled (.emacs.elc), `user-init-file' is set
to the source name (.emacs.el), if that exists, after loading it.

** The help string specified for a menu-item whose definition contains
the property `:help HELP' is now displayed under X, on MS-Windows, and
MS-DOS, either in the echo area or with tooltips.  Many standard menus
displayed by Emacs now have help strings.

--
** New user option `read-mail-command' specifies a command to use to
read mail from the menu etc.

** The environment variable `EMACSLOCKDIR' is no longer used on MS-Windows.
This environment variable was used when creating lock files.  Emacs on
MS-Windows does not use this variable anymore.  This change was made
before Emacs 21.1, but wasn't documented until now.

** Highlighting of mouse-sensitive regions is now supported in the
MS-DOS version of Emacs.

** The new command `msdos-set-mouse-buttons' forces the MS-DOS version
of Emacs to behave as if the mouse had a specified number of buttons.
This comes handy with mice that don't report their number of buttons
correctly.  One example is the wheeled mice, which report 3 buttons,
but clicks on the middle button are not passed to the MS-DOS version
of Emacs.

** Customize changes

*** Customize now supports comments about customized items.  Use the
`State' menu to add comments, or give a prefix argument to
M-x customize-set-variable or M-x customize-set-value.  Note that
customization comments will cause the customizations to fail in
earlier versions of Emacs.

*** The new option `custom-buffer-done-function' says whether to kill
Custom buffers when you've done with them or just bury them (the
default).

*** If Emacs was invoked with the `-q' or `--no-init-file' options, it
does not allow you to save customizations in your `~/.emacs' init
file.  This is because saving customizations from such a session would
wipe out all the other customizationss you might have on your init
file.

** If Emacs was invoked with the `-q' or `--no-init-file' options, it
does not save disabled and enabled commands for future sessions, to
avoid overwriting existing customizations of this kind that are
already in your init file.

** New features in evaluation commands

*** The commands to evaluate Lisp expressions, such as C-M-x in Lisp
modes, C-j in Lisp Interaction mode, and M-:, now bind the variables
print-level, print-length, and debug-on-error based on the new
customizable variables eval-expression-print-level,
eval-expression-print-length, and eval-expression-debug-on-error.

The default values for the first two of these variables are 12 and 4
respectively, which means that `eval-expression' now prints at most
the first 12 members of a list and at most 4 nesting levels deep (if
the list is longer or deeper than that, an ellipsis `...'  is
printed).

<RET> or <mouse-2> on the printed text toggles between an abbreviated
printed representation and an unabbreviated one.

The default value of eval-expression-debug-on-error is t, so any error
during evaluation produces a backtrace.

*** The function `eval-defun' (C-M-x) now loads Edebug and instruments
code when called with a prefix argument.

** CC mode changes.

Note: This release contains changes that might not be compatible with
current user setups (although it's believed that these
incompatibilities will only show in very uncommon circumstances).
However, since the impact is uncertain, these changes may be rolled
back depending on user feedback.  Therefore there's no forward
compatibility guarantee wrt the new features introduced in this
release.

*** The hardcoded switch to "java" style in Java mode is gone.
CC Mode used to automatically set the style to "java" when Java mode
is entered.  This has now been removed since it caused too much
confusion.

However, to keep backward compatibility to a certain extent, the
default value for c-default-style now specifies the "java" style for
java-mode, but "gnu" for all other modes (as before).  So you won't
notice the change if you haven't touched that variable.

*** New cleanups, space-before-funcall and compact-empty-funcall.
Two new cleanups have been added to c-cleanup-list:

space-before-funcall causes a space to be inserted before the opening
parenthesis of a function call, which gives the style "foo (bar)".

compact-empty-funcall causes any space before a function call opening
parenthesis to be removed if there are no arguments to the function.
It's typically useful together with space-before-funcall to get the
style "foo (bar)" and "foo()".

*** Some keywords now automatically trigger reindentation.
Keywords like "else", "while", "catch" and "finally" have been made
"electric" to make them reindent automatically when they continue an
earlier statement.  An example:

for (i = 0; i < 17; i++)
  if (a[i])
    res += a[i]->offset;
else

Here, the "else" should be indented like the preceding "if", since it
continues that statement. CC Mode will automatically reindent it after
the "else" has been typed in full, since it's not until then it's
possible to decide whether it's a new statement or a continuation of
the preceding "if".

CC Mode uses Abbrev mode to achieve this, which is therefore turned on
by default.

*** M-a and M-e now moves by sentence in multiline strings.
Previously these two keys only moved by sentence in comments, which
meant that sentence movement didn't work in strings containing
documentation or other natural language text.

The reason it's only activated in multiline strings (i.e. strings that
contain a newline, even when escaped by a '\') is to avoid stopping in
the short strings that often reside inside statements.  Multiline
strings almost always contain text in a natural language, as opposed
to other strings that typically contain format specifications,
commands, etc.  Also, it's not that bothersome that M-a and M-e misses
sentences in single line strings, since they're short anyway.

*** Support for autodoc comments in Pike mode.
Autodoc comments for Pike are used to extract documentation from the
source, like Javadoc in Java.  Pike mode now recognize this markup in
comment prefixes and paragraph starts.

*** The comment prefix regexps on c-comment-prefix may be mode specific.
When c-comment-prefix is an association list, it specifies the comment
line prefix on a per-mode basis, like c-default-style does.  This
change came about to support the special autodoc comment prefix in
Pike mode only.

*** Better handling of syntactic errors.
The recovery after unbalanced parens earlier in the buffer has been
improved; CC Mode now reports them by dinging and giving a message
stating the offending line, but still recovers and indent the
following lines in a sane way (most of the time).  An "else" with no
matching "if" is handled similarly.  If an error is discovered while
indenting a region, the whole region is still indented and the error
is reported afterwards.

*** Lineup functions may now return absolute columns.
A lineup function can give an absolute column to indent the line to by
returning a vector with the desired column as the first element.

*** More robust and warning-free byte compilation.
Although this is strictly not a user visible change (well, depending
on the view of a user), it's still worth mentioning that CC Mode now
can be compiled in the standard ways without causing trouble.  Some
code have also been moved between the subpackages to enhance the
modularity somewhat.  Thanks to Martin Buchholz for doing the
groundwork.

*** c-style-variables-are-local-p now defaults to t.
This is an incompatible change that has been made to make the behavior
of the style system wrt global variable settings less confusing for
non-advanced users.  If you know what this variable does you might
want to set it to nil in your .emacs, otherwise you probably don't
have to bother.

Defaulting c-style-variables-are-local-p to t avoids the confusing
situation that occurs when a user sets some style variables globally
and edits both a Java and a non-Java file in the same Emacs session.
If the style variables aren't buffer local in this case, loading of
the second file will cause the default style (either "gnu" or "java"
by default) to override the global settings made by the user.

*** New initialization procedure for the style system.
When the initial style for a buffer is determined by CC Mode (from the
variable c-default-style), the global values of style variables now
take precedence over the values specified by the chosen style.  This
is different than the old behavior: previously, the style-specific
settings would override the global settings.  This change makes it
possible to do simple configuration in the intuitive way with
Customize or with setq lines in one's .emacs file.

By default, the global value of every style variable is the new
special symbol set-from-style, which causes the value to be taken from
the style system.  This means that in effect, only an explicit setting
of a style variable will cause the "overriding" behavior described
above.

Also note that global settings override style-specific settings *only*
when the initial style of a buffer is chosen by a CC Mode major mode
function.  When a style is chosen in other ways --- for example, by a
call like (c-set-style "gnu") in a hook, or via M-x c-set-style ---
then the style-specific values take precedence over any global style
values.  In Lisp terms, global values override style-specific values
only when the new second argument to c-set-style is non-nil; see the
function documentation for more info.

The purpose of these changes is to make it easier for users,
especially novice users, to do simple customizations with Customize or
with setq in their .emacs files.  On the other hand, the new system is
intended to be compatible with advanced users' customizations as well,
such as those that choose styles in hooks or whatnot.  This new system
is believed to be almost entirely compatible with current
configurations, in spite of the changed precedence between style and
global variable settings when a buffer's default style is set.

(Thanks to Eric Eide for clarifying this explanation a bit.)

**** c-offsets-alist is now a customizable variable.
This became possible as a result of the new initialization behavior.

This variable is treated slightly differently from the other style
variables; instead of using the symbol set-from-style, it will be
completed with the syntactic symbols it doesn't already contain when
the style is first initialized.  This means it now defaults to the
empty list to make all syntactic elements get their values from the
style system.

**** Compatibility variable to restore the old behavior.
In case your configuration doesn't work with this change, you can set
c-old-style-variable-behavior to non-nil to get the old behavior back
as far as possible.

*** Improvements to line breaking and text filling.
CC Mode now handles this more intelligently and seamlessly wrt the
surrounding code, especially inside comments.  For details see the new
chapter about this in the manual.

**** New variable to recognize comment line prefix decorations.
The variable c-comment-prefix-regexp has been added to properly
recognize the line prefix in both block and line comments.  It's
primarily used to initialize the various paragraph recognition and
adaptive filling variables that the text handling functions uses.

**** New variable c-block-comment-prefix.
This is a generalization of the now obsolete variable
c-comment-continuation-stars to handle arbitrary strings.

**** CC Mode now uses adaptive fill mode.
This to make it adapt better to the paragraph style inside comments.

It's also possible to use other adaptive filling packages inside CC
Mode, notably Kyle E. Jones' Filladapt mode (http://wonderworks.com/).
A new convenience function c-setup-filladapt sets up Filladapt for use
inside CC Mode.

Note though that the 2.12 version of Filladapt lacks a feature that
causes it to work suboptimally when c-comment-prefix-regexp can match
the empty string (which it commonly does).  A patch for that is
available from the CC Mode web site (http://www.python.org/emacs/
cc-mode/).

**** The variables `c-hanging-comment-starter-p' and
`c-hanging-comment-ender-p', which controlled how comment starters and
enders were filled, are not used anymore.  The new version of the
function `c-fill-paragraph' keeps the comment starters and enders as
they were before the filling.

**** It's now possible to selectively turn off auto filling.
The variable c-ignore-auto-fill is used to ignore auto fill mode in
specific contexts, e.g. in preprocessor directives and in string
literals.

**** New context sensitive line break function c-context-line-break.
It works like newline-and-indent in normal code, and adapts the line
prefix according to the comment style when used inside comments.  If
you're normally using newline-and-indent, you might want to switch to
this function.

*** Fixes to IDL mode.
It now does a better job in recognizing only the constructs relevant
to IDL.  E.g. it no longer matches "class" as the beginning of a
struct block, but it does match the CORBA 2.3 "valuetype" keyword.
Thanks to Eric Eide.

*** Improvements to the Whitesmith style.
It now keeps the style consistently on all levels and both when
opening braces hangs and when they don't.

**** New lineup function c-lineup-whitesmith-in-block.

*** New lineup functions c-lineup-template-args and c-indent-multi-line-block.
See their docstrings for details.  c-lineup-template-args does a
better job of tracking the brackets used as parens in C++ templates,
and is used by default to line up continued template arguments.

*** c-lineup-comment now preserves alignment with a comment on the
previous line.  It used to instead preserve comments that started in
the column specified by comment-column.

*** c-lineup-C-comments handles "free form" text comments.
In comments with a long delimiter line at the start, the indentation
is kept unchanged for lines that start with an empty comment line
prefix.  This is intended for the type of large block comments that
contain documentation with its own formatting.  In these you normally
don't want CC Mode to change the indentation.

*** The `c' syntactic symbol is now relative to the comment start
instead of the previous line, to make integers usable as lineup
arguments.

*** All lineup functions have gotten docstrings.

*** More preprocessor directive movement functions.
c-down-conditional does the reverse of c-up-conditional.
c-up-conditional-with-else and c-down-conditional-with-else are
variants of these that also stops at "#else" lines (suggested by Don
Provan).

*** Minor improvements to many movement functions in tricky situations.

** Dired changes

*** New variable `dired-recursive-deletes' determines if the delete
command will delete non-empty directories recursively.  The default
is, delete only empty directories.

*** New variable `dired-recursive-copies' determines if the copy
command will copy directories recursively.  The default is, do not
copy directories recursively.

*** In command `dired-do-shell-command' (usually bound to `!') a `?'
in the shell command has a special meaning similar to `*', but with
the difference that the command will be run on each file individually.

*** The new command `dired-find-alternate-file' (usually bound to `a')
replaces the Dired buffer with the buffer for an alternate file or
directory.

*** The new command `dired-show-file-type' (usually bound to `y') shows
a message in the echo area describing what type of file the point is on.
This command invokes the external program `file' do its work, and so
will only work on systems with that program, and will be only as
accurate or inaccurate as it is.

*** Dired now properly handles undo changes of adding/removing `-R'
from ls switches.

*** Dired commands that prompt for a destination file now allow the use
of the `M-n' command in the minibuffer to insert the source filename,
which the user can then edit.  This only works if there is a single
source file, not when operating on multiple marked files.

** Gnus changes.

The Gnus NEWS entries are short, but they reflect sweeping changes in
four areas: Article display treatment, MIME treatment,
internationalization and mail-fetching.

*** The mail-fetching functions have changed.  See the manual for the
many details.  In particular, all procmail fetching variables are gone.

If you used procmail like in

(setq nnmail-use-procmail t)
(setq nnmail-spool-file 'procmail)
(setq nnmail-procmail-directory "~/mail/incoming/")
(setq nnmail-procmail-suffix "\\.in")

this now has changed to

(setq mail-sources
      '((directory :path "~/mail/incoming/"
                   :suffix ".in")))

More information is available in the info doc at Select Methods ->
Getting Mail -> Mail Sources

*** Gnus is now a MIME-capable reader.  This affects many parts of
Gnus, and adds a slew of new commands.  See the manual for details.
Separate MIME packages like RMIME, mime-compose etc., will probably no
longer work; remove them and use the native facilities.

The FLIM/SEMI package still works with Emacs 21, but if you want to
use the native facilities, you must remove any mailcap.el[c] that was
installed by FLIM/SEMI version 1.13 or earlier.

*** Gnus has also been multilingualized.  This also affects too many
parts of Gnus to summarize here, and adds many new variables.  There
are built-in facilities equivalent to those of gnus-mule.el, which is
now just a compatibility layer.

*** gnus-mule.el is now just a compatibility layer over the built-in
Gnus facilities.

*** gnus-auto-select-first can now be a function to be
called to position point.

*** The user can now decide which extra headers should be included in
summary buffers and NOV files.

*** `gnus-article-display-hook' has been removed.  Instead, a number
of variables starting with `gnus-treat-' have been added.

*** The Gnus posting styles have been redone again and now work in a
subtly different manner.

*** New web-based backends have been added: nnslashdot, nnwarchive
and nnultimate.  nnweb has been revamped, again, to keep up with
ever-changing layouts.

*** Gnus can now read IMAP mail via nnimap.

*** There is image support of various kinds and some sound support.

** Changes in Texinfo mode.

*** A couple of new key bindings have been added for inserting Texinfo
macros

  Key binding   Macro
  -------------------------
  C-c C-c C-s   @strong
  C-c C-c C-e   @emph
  C-c C-c u     @uref
  C-c C-c q     @quotation
  C-c C-c m     @email
  C-c C-o       @<block> ... @end <block>
  M-RET         @item

*** The " key now inserts either " or `` or '' depending on context.

** Changes in Outline mode.

There is now support for Imenu to index headings.  A new command
`outline-headers-as-kill' copies the visible headings in the region to
the kill ring, e.g. to produce a table of contents.

** Changes to Emacs Server

*** The new option `server-kill-new-buffers' specifies what to do
with buffers when done with them.  If non-nil, the default, buffers
are killed, unless they were already present before visiting them with
Emacs Server.  If nil, `server-temp-file-regexp' specifies which
buffers to kill, as before.

Please note that only buffers are killed that still have a client,
i.e. buffers visited with `emacsclient --no-wait' are never killed in
this way.

** Both emacsclient and Emacs itself now accept command line options
of the form +LINE:COLUMN in addition to +LINE.

** Changes to Show Paren mode.

*** Overlays used by Show Paren mode now use a priority property.
The new user option show-paren-priority specifies the priority to
use.  Default is 1000.

** New command M-x check-parens can be used to find unbalanced paren
groups and strings in buffers in Lisp mode (or other modes).

** Changes to hideshow.el

*** Generalized block selection and traversal

A block is now recognized by its start and end regexps (both strings),
and an integer specifying which sub-expression in the start regexp
serves as the place where a `forward-sexp'-like function can operate.
See the documentation of variable `hs-special-modes-alist'.

*** During incremental search, if Hideshow minor mode is active,
hidden blocks are temporarily shown.  The variable `hs-headline' can
be used in the mode line format to show the line at the beginning of
the open block.

*** User option `hs-hide-all-non-comment-function' specifies a
function to be called at each top-level block beginning, instead of
the normal block-hiding function.

*** The command `hs-show-region' has been removed.

*** The key bindings have changed to fit the Emacs conventions,
roughly imitating those of Outline minor mode.  Notably, the prefix
for all bindings is now `C-c @'.  For details, see the documentation
for `hs-minor-mode'.

*** The variable `hs-show-hidden-short-form' has been removed, and
hideshow.el now always behaves as if this variable were set to t.

** Changes to Change Log mode and Add-Log functions

*** If you invoke `add-change-log-entry' from a backup file, it makes
an entry appropriate for the file's parent.  This is useful for making
log entries by comparing a version with deleted functions.

**** New command M-x change-log-merge merges another log into the
current buffer.

*** New command M-x change-log-redate fixes any old-style date entries
in a log file.

*** Change Log mode now adds a file's version number to change log
entries if user-option `change-log-version-info-enabled' is non-nil.
Unless the file is under version control the search for a file's
version number is performed based on regular expressions from
`change-log-version-number-regexp-list' which can be customized.
Version numbers are only found in the first 10 percent of a file.

*** Change Log mode now defines its own faces for font-lock highlighting.

** Changes to cmuscheme

*** The user-option `scheme-program-name' has been renamed
`cmuscheme-program-name' due to conflicts with xscheme.el.

** Changes in Font Lock

*** The new function `font-lock-remove-keywords' can be used to remove
font-lock keywords from the current buffer or from a specific major mode.

*** Multi-line patterns are now supported.  Modes using this, should
set font-lock-multiline to t in their font-lock-defaults.

*** `font-lock-syntactic-face-function' allows major-modes to choose
the face used for each string/comment.

*** A new standard face `font-lock-doc-face'.
Meant for Lisp docstrings, Javadoc comments and other "documentation in code".

** Changes to Shell mode

*** The `shell' command now accepts an optional argument to specify the buffer
to use, which defaults to "*shell*".  When used interactively, a
non-default buffer may be specified by giving the `shell' command a
prefix argument (causing it to prompt for the buffer name).

** Comint (subshell) changes

These changes generally affect all modes derived from comint mode, which
include shell-mode, gdb-mode, scheme-interaction-mode, etc.

*** Comint now by default interprets some carriage-control characters.
Comint now removes CRs from CR LF sequences, and treats single CRs and
BSs in the output in a way similar to a terminal (by deleting to the
beginning of the line, or deleting the previous character,
respectively).  This is achieved by adding `comint-carriage-motion' to
the `comint-output-filter-functions' hook by default.

*** By default, comint no longer uses the variable `comint-prompt-regexp'
to distinguish prompts from user-input.  Instead, it notices which
parts of the text were output by the process, and which entered by the
user, and attaches `field' properties to allow emacs commands to use
this information.  Common movement commands, notably beginning-of-line,
respect field boundaries in a fairly natural manner.  To disable this
feature, and use the old behavior, customize the user option
`comint-use-prompt-regexp-instead-of-fields'.

*** Comint now includes new features to send commands to running processes
and redirect the output to a designated buffer or buffers.

*** The command M-x comint-redirect-send-command reads a command and
buffer name from the mini-buffer.  The command is sent to the current
buffer's process, and its output is inserted into the specified buffer.

The command M-x comint-redirect-send-command-to-process acts like
M-x comint-redirect-send-command but additionally reads the name of
the buffer whose process should be used from the mini-buffer.

*** Packages based on comint now highlight user input and program prompts,
and support choosing previous input with mouse-2.  To control these features,
see the user-options `comint-highlight-input' and `comint-highlight-prompt'.

*** The new command `comint-write-output' (usually bound to `C-c C-s')
saves the output from the most recent command to a file.  With a prefix
argument, it appends to the file.

*** The command `comint-kill-output' has been renamed `comint-delete-output'
(usually bound to `C-c C-o'); the old name is aliased to it for
compatibility.

*** The new function `comint-add-to-input-history' adds commands to the input
ring (history).

*** The new variable `comint-input-history-ignore' is a regexp for
identifying history lines that should be ignored, like tcsh time-stamp
strings, starting with a `#'.  The default value of this variable is "^#".

** Changes to Rmail mode

*** The new user-option rmail-user-mail-address-regexp can be
set to fine tune the identification of the correspondent when
receiving new mail.  If it matches the address of the sender, the
recipient is taken as correspondent of a mail.  If nil, the default,
`user-login-name' and `user-mail-address' are used to exclude yourself
as correspondent.

Usually you don't have to set this variable, except if you collect
mails sent by you under different user names.  Then it should be a
regexp matching your mail addresses.

*** The new user-option rmail-confirm-expunge controls whether and how
to ask for confirmation before expunging deleted messages from an
Rmail file.  You can choose between no confirmation, confirmation
with y-or-n-p, or confirmation with yes-or-no-p.  Default is to ask
for confirmation with yes-or-no-p.

*** RET is now bound in the Rmail summary to rmail-summary-goto-msg,
like `j'.

*** There is a new user option `rmail-digest-end-regexps' that
specifies the regular expressions to detect the line that ends a
digest message.

*** The new user option `rmail-automatic-folder-directives' specifies
in which folder to put messages automatically.

*** The new function `rmail-redecode-body' allows to fix a message
with non-ASCII characters if Emacs happens to decode it incorrectly
due to missing or malformed "charset=" header.

** The new user-option `mail-envelope-from' can be used to specify
an envelope-from address different from user-mail-address.

** The variable mail-specify-envelope-from controls whether to
use the -f option when sending mail.

** The Rmail command `o' (`rmail-output-to-rmail-file') now writes the
current message in the internal `emacs-mule' encoding, rather than in
the encoding taken from the variable `buffer-file-coding-system'.
This allows to save messages whose characters cannot be safely encoded
by the buffer's coding system, and makes sure the message will be
displayed correctly when you later visit the target Rmail file.

If you want your Rmail files be encoded in a specific coding system
other than `emacs-mule', you can customize the variable
`rmail-file-coding-system' to set its value to that coding system.

** Changes to TeX mode

*** The default mode has been changed from `plain-tex-mode' to
`latex-mode'.

*** latex-mode now has a simple indentation algorithm.

*** M-f and M-p jump around \begin...\end pairs.

*** Added support for outline-minor-mode.

** Changes to RefTeX mode

*** RefTeX has new support for index generation.  Index entries can be
    created with `C-c <', with completion available on index keys.
    Pressing `C-c /' indexes the word at the cursor with a default
    macro.  `C-c >' compiles all index entries into an alphabetically
    sorted *Index* buffer which looks like the final index.  Entries
    can be edited from that buffer.

*** Label and citation key selection now allow to select several
    items and reference them together (use `m' to mark items, `a' or
    `A' to use all marked entries).

*** reftex.el has been split into a number of smaller files to reduce
    memory use when only a part of RefTeX is being used.

*** a new command `reftex-view-crossref-from-bibtex' (bound to `C-c &'
    in BibTeX-mode) can be called in a BibTeX database buffer in order
    to show locations in LaTeX documents where a particular entry has
    been cited.

** Emacs Lisp mode now allows multiple levels of outline headings.
The level of a heading is determined from the number of leading
semicolons in a heading line.  Toplevel forms starting with a `('
in column 1 are always made leaves.

** The M-x time-stamp command (most commonly used on write-file-hooks)
has the following new features:

*** The patterns for finding the time stamp and for updating a pattern
may match text spanning multiple lines.  For example, some people like
to have the filename and date on separate lines.  The new variable
time-stamp-inserts-lines controls the matching for multi-line patterns.

*** More than one time stamp can be updated in the same file.  This
feature is useful if you need separate time stamps in a program source
file to both include in formatted documentation and insert in the
compiled binary.  The same time-stamp will be written at each matching
pattern.  The variable time-stamp-count enables this new feature; it
defaults to 1.

** Partial Completion mode now completes environment variables in
file names.

** Ispell changes

*** The command `ispell' now spell-checks a region if
transient-mark-mode is on, and the mark is active.  Otherwise it
spell-checks the current buffer.

*** Support for synchronous subprocesses - DOS/Windoze - has been
added.

*** An "alignment error" bug was fixed when a manual spelling
correction is made and re-checked.

*** Italian, Portuguese, and Slovak dictionary definitions have been added.

*** Region skipping performance has been vastly improved in some
cases.

*** Spell checking HTML buffers has been improved and isn't so strict
on syntax errors.

*** The buffer-local words are now always placed on a new line at the
end of the buffer.

*** Spell checking now works in the MS-DOS version of Emacs.

*** The variable `ispell-format-word' has been renamed to
`ispell-format-word-function'.  The old name is still available as
alias.

** Makefile mode changes

*** The mode now uses the abbrev table `makefile-mode-abbrev-table'.

*** Conditionals and include statements are now highlighted when
Fontlock mode is active.

** Isearch changes

*** Isearch now puts a call to `isearch-resume' in the command history,
so that searches can be resumed.

*** In Isearch mode, C-M-s and C-M-r are now bound like C-s and C-r,
respectively, i.e. you can repeat a regexp isearch with the same keys
that started the search.

*** In Isearch mode, mouse-2 in the echo area now yanks the current
selection into the search string rather than giving an error.

*** There is a new lazy highlighting feature in incremental search.

Lazy highlighting is switched on/off by customizing variable
`isearch-lazy-highlight'.  When active, all matches for the current
search string are highlighted.  The current match is highlighted as
before using face `isearch' or `region'.  All other matches are
highlighted using face `isearch-lazy-highlight-face' which defaults to
`secondary-selection'.

The extra highlighting makes it easier to anticipate where the cursor
will end up each time you press C-s or C-r to repeat a pending search.
Highlighting of these additional matches happens in a deferred fashion
using "idle timers," so the cycles needed do not rob isearch of its
usual snappy response.

If `isearch-lazy-highlight-cleanup' is set to t, highlights for
matches are automatically cleared when you end the search.  If it is
set to nil, you can remove the highlights manually with `M-x
isearch-lazy-highlight-cleanup'.

** VC Changes

VC has been overhauled internally.  It is now modular, making it
easier to plug-in arbitrary version control backends.  (See Lisp
Changes for details on the new structure.)  As a result, the mechanism
to enable and disable support for particular version systems has
changed: everything is now controlled by the new variable
`vc-handled-backends'.  Its value is a list of symbols that identify
version systems; the default is '(RCS CVS SCCS).  When finding a file,
each of the backends in that list is tried in order to see whether the
file is registered in that backend.

When registering a new file, VC first tries each of the listed
backends to see if any of them considers itself "responsible" for the
directory of the file (e.g. because a corresponding subdirectory for
master files exists).  If none of the backends is responsible, then
the first backend in the list that could register the file is chosen.
As a consequence, the variable `vc-default-back-end' is now obsolete.

The old variable `vc-master-templates' is also obsolete, although VC
still supports it for backward compatibility.  To define templates for
RCS or SCCS, you should rather use the new variables
vc-{rcs,sccs}-master-templates.  (There is no such feature under CVS
where it doesn't make sense.)

The variables `vc-ignore-vc-files' and `vc-handle-cvs' are also
obsolete now, you must set `vc-handled-backends' to nil or exclude
`CVS' from the list, respectively, to achieve their effect now.

*** General Changes

The variable `vc-checkout-carefully' is obsolete: the corresponding
checks are always done now.

VC Dired buffers are now kept up-to-date during all version control
operations.

`vc-diff' output is now displayed in `diff-mode'.
`vc-print-log' uses `log-view-mode'.
`vc-log-mode' (used for *VC-Log*) has been replaced by `log-edit-mode'.

The command C-x v m (vc-merge) now accepts an empty argument as the
first revision number.  This means that any recent changes on the
current branch should be picked up from the repository and merged into
the working file (``merge news'').

The commands C-x v s (vc-create-snapshot) and C-x v r
(vc-retrieve-snapshot) now ask for a directory name from which to work
downwards.

*** Multiple Backends

VC now lets you register files in more than one backend.  This is
useful, for example, if you are working with a slow remote CVS
repository.  You can then use RCS for local editing, and occasionally
commit your changes back to CVS, or pick up changes from CVS into your
local RCS archives.

To make this work, the ``more local'' backend (RCS in our example)
should come first in `vc-handled-backends', and the ``more remote''
backend (CVS) should come later.  (The default value of
`vc-handled-backends' already has it that way.)

You can then commit changes to another backend (say, RCS), by typing
C-u C-x v v RCS RET (i.e. vc-next-action now accepts a backend name as
a revision number).  VC registers the file in the more local backend
if that hasn't already happened, and commits to a branch based on the
current revision number from the more remote backend.

If a file is registered in multiple backends, you can switch to
another one using C-x v b (vc-switch-backend).  This does not change
any files, it only changes VC's perspective on the file.  Use this to
pick up changes from CVS while working under RCS locally.

After you are done with your local RCS editing, you can commit your
changes back to CVS using C-u C-x v v CVS RET.  In this case, the
local RCS archive is removed after the commit, and the log entry
buffer is initialized to contain the entire RCS change log of the file.

*** Changes for CVS

There is a new user option, `vc-cvs-stay-local'.  If it is `t' (the
default), then VC avoids network queries for files registered in
remote repositories.  The state of such files is then only determined
by heuristics and past information.  `vc-cvs-stay-local' can also be a
regexp to match against repository hostnames; only files from hosts
that match it are treated locally.  If the variable is nil, then VC
queries the repository just as often as it does for local files.

If `vc-cvs-stay-local' is on, then VC also makes local backups of
repository versions.  This means that ordinary diffs (C-x v =) and
revert operations (C-x v u) can be done completely locally, without
any repository interactions at all.  The name of a local version
backup of FILE is FILE.~REV.~, where REV is the repository version
number.  This format is similar to that used by C-x v ~
(vc-version-other-window), except for the trailing dot.  As a matter
of fact, the two features can each use the files created by the other,
the only difference being that files with a trailing `.' are deleted
automatically after commit.  (This feature doesn't work on MS-DOS,
since DOS disallows more than a single dot in the trunk of a file
name.)

If `vc-cvs-stay-local' is on, and there have been changes in the
repository, VC notifies you about it when you actually try to commit.
If you want to check for updates from the repository without trying to
commit, you can either use C-x v m RET to perform an update on the
current file, or you can use C-x v r RET to get an update for an
entire directory tree.

The new user option `vc-cvs-use-edit' indicates whether VC should call
"cvs edit" to make files writeable; it defaults to `t'.  (This option
is only meaningful if the CVSREAD variable is set, or if files are
"watched" by other developers.)

The commands C-x v s (vc-create-snapshot) and C-x v r
(vc-retrieve-snapshot) are now also implemented for CVS.  If you give
an empty snapshot name to the latter, that performs a `cvs update',
starting at the given directory.

*** Lisp Changes in VC

VC has been restructured internally to make it modular.  You can now
add support for arbitrary version control backends by writing a
library that provides a certain set of backend-specific functions, and
then telling VC to use that library.  For example, to add support for
a version system named SYS, you write a library named vc-sys.el, which
provides a number of functions vc-sys-... (see commentary at the top
of vc.el for a detailed list of them).  To make VC use that library,
you need to put it somewhere into Emacs' load path and add the symbol
`SYS' to the list `vc-handled-backends'.

** The customizable EDT emulation package now supports the EDT
SUBS command and EDT scroll margins.  It also works with more
terminal/keyboard configurations and it now works under XEmacs.
See etc/edt-user.doc for more information.

** New modes and packages

*** The new global minor mode `minibuffer-electric-default-mode'
automatically hides the `(default ...)' part of minibuffer prompts when
the default is not applicable.

*** Artist is an Emacs lisp package that allows you to draw lines,
rectangles and ellipses by using your mouse and/or keyboard.  The
shapes are made up with the ascii characters |, -, / and \.

Features are:

- Intersecting: When a `|' intersects with a `-', a `+' is
  drawn, like this:   |         \ /
                    --+--        X
                      |         / \

- Rubber-banding: When drawing lines you can interactively see the
  result while holding the mouse button down and moving the mouse.  If
  your machine is not fast enough (a 386 is a bit too slow, but a
  pentium is well enough), you can turn this feature off.  You will
  then see 1's and 2's which mark the 1st and 2nd endpoint of the line
  you are drawing.

- Arrows: After having drawn a (straight) line or a (straight)
  poly-line, you can set arrows on the line-ends by typing < or >.

- Flood-filling: You can fill any area with a certain character by
  flood-filling.

- Cut copy and paste: You can cut, copy and paste rectangular
  regions.  Artist also interfaces with the rect package (this can be
  turned off if it causes you any trouble) so anything you cut in
  artist can be yanked with C-x r y and vice versa.

- Drawing with keys: Everything you can do with the mouse, you can
  also do without the mouse.

- Aspect-ratio: You can set the variable artist-aspect-ratio to
  reflect the height-width ratio for the font you are using. Squares
  and circles are then drawn square/round.  Note, that once your
  ascii-file is shown with font with a different height-width ratio,
  the squares won't be square and the circles won't be round.

- Drawing operations: The following drawing operations are implemented:

    lines               straight-lines
    rectangles          squares
    poly-lines          straight poly-lines
    ellipses            circles
    text (see-thru)     text (overwrite)
    spray-can           setting size for spraying
    vaporize line       vaporize lines
    erase characters    erase rectangles

  Straight lines are lines that go horizontally, vertically or
  diagonally.  Plain lines go in any direction.  The operations in
  the right column are accessed by holding down the shift key while
  drawing.

  It is possible to vaporize (erase) entire lines and connected lines
  (rectangles for example) as long as the lines being vaporized are
  straight and connected at their endpoints.  Vaporizing is inspired
  by the drawrect package by Jari Aalto <jari.aalto@poboxes.com>.

- Picture mode compatibility: Artist is picture mode compatible (this
  can be turned off).

*** The new package Eshell is an operating system command shell
implemented entirely in Emacs Lisp.  Use `M-x eshell' to invoke it.
It functions similarly to bash and zsh, and allows running of Lisp
functions and external commands using the same syntax.  It supports
history lists, aliases, extended globbing, smart scrolling, etc.  It
will work on any platform Emacs has been ported to.  And since most of
the basic commands -- ls, rm, mv, cp, ln, du, cat, etc. -- have been
rewritten in Lisp, it offers an operating-system independent shell,
all within the scope of your Emacs process.

*** The new package timeclock.el is a mode is for keeping track of time
intervals.  You can use it for whatever purpose you like, but the
typical scenario is to keep track of how much time you spend working
on certain projects.

*** The new package hi-lock.el provides commands to highlight matches
of interactively entered regexps.  For example,

  M-x highlight-regexp RET clearly RET RET

will highlight all occurrences of `clearly' using a yellow background
face.  New occurrences of `clearly' will be highlighted as they are
typed.  `M-x unhighlight-regexp RET' will remove the highlighting.
Any existing face can be used for highlighting and a set of
appropriate faces is provided.  The regexps can be written into the
current buffer in a form that will be recognized the next time the
corresponding file is read.  There are commands to highlight matches
to phrases and to highlight entire lines containing a match.

*** The new package zone.el plays games with Emacs' display when
Emacs is idle.

*** The new package tildify.el allows to add hard spaces or other text
fragments in accordance with the current major mode.

*** The new package xml.el provides a simple but generic XML
parser. It doesn't parse the DTDs however.

*** The comment operations are now provided by the newcomment.el
package which allows different styles of comment-region and should
be more robust while offering the same functionality.
`comment-region' now doesn't always comment a-line-at-a-time, but only
comments the region, breaking the line at point if necessary.

*** The Ebrowse package implements a C++ class browser and tags
facilities tailored for use with C++.  It is documented in a
separate Texinfo file.

*** The PCL-CVS package available by either running M-x cvs-examine or
by visiting a CVS administrative directory (with a prefix argument)
provides an alternative interface to VC-dired for CVS.  It comes with
`log-view-mode' to view RCS and SCCS logs and `log-edit-mode' used to
enter check-in log messages.

*** The new package called `woman' allows to browse Unix man pages
without invoking external programs.

The command `M-x woman' formats manual pages entirely in Emacs Lisp
and then displays them, like `M-x manual-entry' does.  Unlike
`manual-entry', `woman' does not invoke any external programs, so it
is useful on systems such as MS-DOS/MS-Windows where the `man' and
Groff or `troff' commands are not readily available.

The command `M-x woman-find-file' asks for the file name of a man
page, then formats and displays it like `M-x woman' does.

*** The new command M-x re-builder offers a convenient interface for
authoring regular expressions with immediate visual feedback.

The buffer from which the command was called becomes the target for
the regexp editor popping up in a separate window.  Matching text in
the target buffer is immediately color marked during the editing.
Each sub-expression of the regexp will show up in a different face so
even complex regexps can be edited and verified on target data in a
single step.

On displays not supporting faces the matches instead blink like
matching parens to make them stand out.  On such a setup you will
probably also want to use the sub-expression mode when the regexp
contains such to get feedback about their respective limits.

*** glasses-mode is a minor mode that makes
unreadableIdentifiersLikeThis readable.  It works as glasses, without
actually modifying content of a buffer.

*** The package ebnf2ps translates an EBNF to a syntactic chart in
PostScript.

Currently accepts ad-hoc EBNF, ISO EBNF and Bison/Yacc.

The ad-hoc default EBNF syntax has the following elements:

    ;           comment (until end of line)
    A           non-terminal
    "C"         terminal
    ?C?         special
    $A          default non-terminal
    $"C"        default terminal
    $?C?        default special
    A = B.      production (A is the header and B the body)
    C D         sequence (C occurs before D)
    C | D       alternative (C or D occurs)
    A - B       exception (A excluding B, B without any non-terminal)
    n * A       repetition (A repeats n (integer) times)
    (C)         group (expression C is grouped together)
    [C]         optional (C may or not occurs)
    C+          one or more occurrences of C
    {C}+        one or more occurrences of C
    {C}*        zero or more occurrences of C
    {C}         zero or more occurrences of C
    C / D       equivalent to: C {D C}*
    {C || D}+   equivalent to: C {D C}*
    {C || D}*   equivalent to: [C {D C}*]
    {C || D}    equivalent to: [C {D C}*]

Please, see ebnf2ps documentation for EBNF syntax and how to use it.

*** The package align.el will align columns within a region, using M-x
align.  Its mode-specific rules, based on regular expressions,
determine where the columns should be split.  In C and C++, for
example, it will align variable names in declaration lists, or the
equal signs of assignments.

*** `paragraph-indent-minor-mode' is a new minor mode supporting
paragraphs in the same style as `paragraph-indent-text-mode'.

*** bs.el is a new package for buffer selection similar to
list-buffers or electric-buffer-list.  Use M-x bs-show to display a
buffer menu with this package.  See the Custom group `bs'.

*** find-lisp.el is a package emulating the Unix find command in Lisp.

*** calculator.el is a small calculator package that is intended to
replace desktop calculators such as xcalc and calc.exe.  Actually, it
is not too small - it has more features than most desktop calculators,
and can be customized easily to get many more functions.  It should
not be confused with "calc" which is a much bigger mathematical tool
which answers different needs.

*** The minor modes cwarn-mode and global-cwarn-mode highlights
suspicious C and C++ constructions.  Currently, assignments inside
expressions, semicolon following `if', `for' and `while' (except, of
course, after a `do .. while' statement), and C++ functions with
reference parameters are recognized.  The modes require font-lock mode
to be enabled.

*** smerge-mode.el provides `smerge-mode', a simple minor-mode for files
containing diff3-style conflict markers, such as generated by RCS.

*** 5x5.el is a simple puzzle game.

*** hl-line.el provides `hl-line-mode', a minor mode to highlight the
current line in the current buffer.  It also provides
`global-hl-line-mode' to provide the same behavior in all buffers.

*** ansi-color.el translates ANSI terminal escapes into text-properties.

Please note: if `ansi-color-for-comint-mode' and
`global-font-lock-mode' are non-nil, loading ansi-color.el will
disable font-lock and add `ansi-color-apply' to
`comint-preoutput-filter-functions' for all shell-mode buffers.  This
displays the output of "ls --color=yes" using the correct foreground
and background colors.

*** delphi.el provides a major mode for editing the Delphi (Object
Pascal) language.

*** quickurl.el provides a simple method of inserting a URL based on
the text at point.

*** sql.el provides an interface to SQL data bases.

*** fortune.el uses the fortune program to create mail/news signatures.

*** whitespace.el is a package for warning about and cleaning bogus
whitespace in a file.

*** PostScript mode (ps-mode) is a new major mode for editing PostScript
files. It offers: interaction with a PostScript interpreter, including
(very basic) error handling; fontification, easily customizable for
interpreter messages; auto-indentation; insertion of EPSF templates and
often used code snippets; viewing of BoundingBox; commenting out /
uncommenting regions; conversion of 8bit characters to PostScript octal
codes. All functionality is accessible through a menu.

*** delim-col helps to prettify columns in a text region or rectangle.

Here is an example of columns:

horse   apple   bus
dog     pineapple       car     EXTRA
porcupine       strawberry      airplane

Doing the following settings:

   (setq delimit-columns-str-before "[ ")
   (setq delimit-columns-str-after " ]")
   (setq delimit-columns-str-separator ", ")
   (setq delimit-columns-separator "\t")


Selecting the lines above and typing:

   M-x delimit-columns-region

It results:

[ horse    , apple     , bus     ,       ]
[ dog      , pineapple , car     , EXTRA ]
[ porcupine, strawberry, airplane,       ]

delim-col has the following options:

   delimit-columns-str-before           Specify a string to be inserted
                                        before all columns.

   delimit-columns-str-separator        Specify a string to be inserted
                                        between each column.

   delimit-columns-str-after            Specify a string to be inserted
                                        after all columns.

   delimit-columns-separator            Specify a regexp which separates
                                        each column.

delim-col has the following commands:

   delimit-columns-region       Prettify all columns in a text region.
   delimit-columns-rectangle    Prettify all columns in a text rectangle.

*** Recentf mode maintains a menu for visiting files that were
operated on recently.  User option recentf-menu-filter specifies a
menu filter function to change the menu appearance. For example, the
recent file list can be displayed:

- organized by major modes, directories or user defined rules.
- sorted by file paths, file names, ascending or descending.
- showing paths relative to the current default-directory

The `recentf-filter-changer' menu filter function allows to
dynamically change the menu appearance.

*** elide-head.el provides a mechanism for eliding boilerplate header
text.

*** footnote.el provides `footnote-mode', a minor mode supporting use
of footnotes.  It is intended for use with Message mode, but isn't
specific to Message mode.

*** diff-mode.el provides `diff-mode', a major mode for
viewing/editing context diffs (patches).  It is selected for files
with extension `.diff', `.diffs', `.patch' and `.rej'.

*** EUDC, the Emacs Unified Directory Client, provides a common user
interface to access directory servers using different directory
protocols.  It has a separate manual.

*** autoconf.el provides a major mode for editing configure.in files
for Autoconf, selected automatically.

*** windmove.el provides moving between windows.

*** crm.el provides a facility to read multiple strings from the
minibuffer with completion.

*** todo-mode.el provides management of TODO lists and integration
with the diary features.

*** autoarg.el provides a feature reported from Twenex Emacs whereby
numeric keys supply prefix args rather than self inserting.

*** The function `turn-off-auto-fill' unconditionally turns off Auto
Fill mode.

*** pcomplete.el is a library that provides programmable completion
facilities for Emacs, similar to what zsh and tcsh offer.  The main
difference is that completion functions are written in Lisp, meaning
they can be profiled, debugged, etc.

*** antlr-mode is a new major mode for editing ANTLR grammar files.
It is automatically turned on for files whose names have the extension
`.g'.

** Changes in sort.el

The function sort-numeric-fields interprets numbers starting with `0'
as octal and numbers starting with `0x' or `0X' as hexadecimal.  The
new user-option sort-numeric-base can be used to specify a default
numeric base.

** Changes to Ange-ftp

*** Ange-ftp allows you to specify of a port number in remote file
names cleanly.  It is appended to the host name, separated by a hash
sign, e.g. `/foo@bar.org#666:mumble'.  (This syntax comes from EFS.)

*** If the new user-option `ange-ftp-try-passive-mode' is set, passive
ftp mode will be used if the ftp client supports that.

*** Ange-ftp handles the output of the w32-style clients which
output ^M at the end of lines.

** The recommended way of using Iswitchb is via the new global minor
mode `iswitchb-mode'.

** Just loading the msb package doesn't switch on Msb mode anymore.
If you have `(require 'msb)' in your .emacs, please replace it with
`(msb-mode 1)'.

** Changes in Flyspell mode

*** Flyspell mode has various new options.  See the `flyspell' Custom
group.

*** The variable `flyspell-generic-check-word-p' has been renamed
to `flyspell-generic-check-word-predicate'.  The old name is still
available as alias.

** The user option `backward-delete-char-untabify-method' controls the
behavior of `backward-delete-char-untabify'.  The following values
are recognized:

`untabify' -- turn a tab to many spaces, then delete one space;
`hungry'   -- delete all whitespace, both tabs and spaces;
`all'      -- delete all whitespace, including tabs, spaces and newlines;
nil        -- just delete one character.

Default value is `untabify'.

[This change was made in Emacs 20.3 but not mentioned then.]

** In Cperl mode `cperl-invalid-face' should now be a normal face
symbol, not double-quoted.

** Some packages are declared obsolete, to be removed in a future
version.  They are:  auto-show, c-mode, hilit19, hscroll, ooutline,
profile, rnews, rnewspost, and sc.  Their implementations have been
moved to lisp/obsolete.

** auto-compression mode is no longer enabled just by loading jka-compr.el.
To control it, set `auto-compression-mode' via Custom or use the
`auto-compression-mode' command.

** `browse-url-gnome-moz' is a new option for
`browse-url-browser-function', invoking Mozilla in GNOME, and
`browse-url-kde' can be chosen for invoking the KDE browser.

** The user-option `browse-url-new-window-p' has been renamed to
`browse-url-new-window-flag'.

** The functions `keep-lines', `flush-lines' and `how-many' now
operate on the active region in Transient Mark mode.

** `gnus-user-agent' is a new possibility for `mail-user-agent'.  It
is like `message-user-agent', but with all the Gnus paraphernalia.

** The Strokes package has been updated.  If your Emacs has XPM
support, you can use it for pictographic editing.  In Strokes mode,
use C-mouse-2 to compose a complex stoke and insert it into the
buffer.  You can encode or decode a strokes buffer with new commands
M-x strokes-encode-buffer and M-x strokes-decode-buffer.  There is a
new command M-x strokes-list-strokes.

** Hexl contains a new command `hexl-insert-hex-string' which inserts
a string of hexadecimal numbers read from the mini-buffer.

** Hexl mode allows to insert non-ASCII characters.

The non-ASCII characters are encoded using the same encoding as the
file you are visiting in Hexl mode.

** Shell script mode changes.

Shell script mode (sh-script) can now indent scripts for shells
derived from sh and rc.  The indentation style is customizable, and
sh-script can attempt to "learn" the current buffer's style.

** Etags changes.

*** In DOS, etags looks for file.cgz if it cannot find file.c.

*** New option --ignore-case-regex is an alternative to --regex.  It is now
possible to bind a regexp to a language, by prepending the regexp with
{lang}, where lang is one of the languages that `etags --help' prints out.
This feature is useful especially for regex files, where each line contains
a regular expression.  The manual contains details.

*** In C and derived languages, etags creates tags for function
declarations when given the --declarations option.

*** In C++, tags are created for "operator".  The tags have the form
"operator+", without spaces between the keyword and the operator.

*** You shouldn't generally need any more the -C or -c++ option: etags
automatically switches to C++ parsing when it meets the `class' or
`template' keywords.

*** Etags now is able to delve at arbitrary deeps into nested structures in
C-like languages.  Previously, it was limited to one or two brace levels.

*** New language Ada: tags are functions, procedures, packages, tasks, and
types.

*** In Fortran, `procedure' is not tagged.

*** In Java, tags are created for "interface".

*** In Lisp, "(defstruct (foo", "(defun (operator" and similar constructs
are now tagged.

*** In makefiles, tags the targets.

*** In Perl, the --globals option tags global variables.  my and local
variables are tagged.

*** New language Python: def and class at the beginning of a line are tags.

*** .ss files are Scheme files, .pdb is Postscript with C syntax, .psw is
for PSWrap.

** Changes in etags.el

*** The new user-option tags-case-fold-search can be used to make
tags operations case-sensitive or case-insensitive.  The default
is to use the same setting as case-fold-search.

*** You can display additional output with M-x tags-apropos by setting
the new variable tags-apropos-additional-actions.

If non-nil, the variable's value should be a list of triples (TITLE
FUNCTION TO-SEARCH).  For each triple, M-x tags-apropos processes
TO-SEARCH and lists tags from it.  TO-SEARCH should be an alist,
obarray, or symbol.  If it is a symbol, the symbol's value is used.

TITLE is a string to use to label the list of tags from TO-SEARCH.

FUNCTION is a function to call when an entry is selected in the Tags
List buffer.  It is called with one argument, the selected symbol.

A useful example value for this variable might be something like:

  '(("Emacs Lisp" Info-goto-emacs-command-node obarray)
    ("Common Lisp" common-lisp-hyperspec common-lisp-hyperspec-obarray)
    ("SCWM" scwm-documentation scwm-obarray))

*** The face tags-tag-face can be used to customize the appearance
of tags in the output of M-x tags-apropos.

*** Setting tags-apropos-verbose to a non-nil value displays the
names of tags files in the *Tags List* buffer.

*** You can now search for tags that are part of the filename itself.
If you have tagged the files topfile.c subdir/subfile.c
/tmp/tempfile.c, you can now search for tags "topfile.c", "subfile.c",
"dir/sub", "tempfile", "tempfile.c".  If the tag matches the file name,
point will go to the beginning of the file.

*** Compressed files are now transparently supported if
auto-compression-mode is active.  You can tag (with Etags) and search
(with find-tag) both compressed and uncompressed files.

*** Tags commands like M-x tags-search no longer change point
in buffers where no match is found.  In buffers where a match is
found, the original value of point is pushed on the marker ring.

** Fortran mode has a new command `fortran-strip-sequence-nos' to
remove text past column 72.  The syntax class of `\' in Fortran is now
appropriate for C-style escape sequences in strings.

** SGML mode's default `sgml-validate-command' is now `nsgmls'.

** A new command `view-emacs-problems' (C-h P) displays the PROBLEMS file.

** The Dabbrev package has a new user-option `dabbrev-ignored-regexps'
containing a list of regular expressions.  Buffers matching a regular
expression from that list, are not checked.

** Emacs can now figure out modification times of remote files.
When you do C-x C-f /user@host:/path/file RET and edit the file,
and someone else modifies the file, you will be prompted to revert
the buffer, just like for the local files.

** The buffer menu (C-x C-b) no longer lists the *Buffer List* buffer.

** When invoked with a prefix argument, the command `list-abbrevs' now
displays local abbrevs, only.

** Refill minor mode provides preliminary support for keeping
paragraphs filled as you modify them.

** The variable `double-click-fuzz' specifies how much the mouse
may be moved between clicks that are recognized as a pair.  Its value
is measured in pixels.

** The new global minor mode `auto-image-file-mode' allows image files
to be visited as images.

** Two new user-options `grep-command' and `grep-find-command'
were added to compile.el.

** Withdrawn packages

*** mldrag.el has been removed.  mouse.el provides the same
functionality with aliases for the mldrag functions.

*** eval-reg.el has been obsoleted by changes to edebug.el and removed.

*** ph.el has been obsoleted by EUDC and removed.

\f
* Incompatible Lisp changes in 21.1

There are a few Lisp changes which are not backwards-compatible and
may require changes to existing code. Here is a list for reference.
See the sections below for details.

** Since `format' preserves text properties, the idiom
`(format "%s" foo)' no longer works to copy and remove properties.
Use `copy-sequence' to copy the string, then use `set-text-properties'
to remove the properties of the copy.

** Since the `keymap' text property now has significance, some code
which uses both `local-map' and `keymap' properties (for portability)
may, for instance, give rise to duplicate menus when the keymaps from
these properties are active.

** The change in the treatment of non-ASCII characters in search
ranges may affect some code.

** A non-nil value for the LOCAL arg of add-hook makes the hook
buffer-local even if `make-local-hook' hasn't been called, which might
make a difference to some code.

** The new treatment of the minibuffer prompt might affect code which
operates on the minibuffer.

** The new character sets `eight-bit-control' and `eight-bit-graphic'
cause `no-conversion' and `emacs-mule-unix' coding systems to produce
different results when reading files with non-ASCII characters
(previously, both coding systems would produce the same results).
Specifically, `no-conversion' interprets each 8-bit byte as a separate
character.  This makes `no-conversion' inappropriate for reading
multibyte text, e.g. buffers written to disk in their internal MULE
encoding (auto-saving does that, for example).  If a Lisp program
reads such files with `no-conversion', each byte of the multibyte
sequence, including the MULE leading codes such as \201, is treated as
a separate character, which prevents them from being interpreted in
the buffer as multibyte characters.

Therefore, Lisp programs that read files which contain the internal
MULE encoding should use `emacs-mule-unix'.  `no-conversion' is only
appropriate for reading truly binary files.

** Code that relies on the obsolete `before-change-function' and
`after-change-function' to detect buffer changes will now fail.  Use
`before-change-functions' and `after-change-functions' instead.

** Code that uses `concat' with integer args now gets an error, as
long promised.  So does any code that uses derivatives of `concat',
such as `mapconcat'.

** The function base64-decode-string now always returns a unibyte
string.

** Not a Lisp incompatibility as such but, with the introduction of
extra private charsets, there is now only one slot free for a new
dimension-2 private charset.  User code which tries to add more than
one extra will fail unless you rebuild Emacs with some standard
charset(s) removed; that is probably inadvisable because it changes
the emacs-mule encoding.  Also, files stored in the emacs-mule
encoding using Emacs 20 with additional private charsets defined will
probably not be read correctly by Emacs 21.

** The variable `directory-sep-char' is slated for removal.
Not really a change (yet), but a projected one that you should be
aware of: The variable `directory-sep-char' is deprecated, and should
not be used.  It was always ignored on GNU/Linux and Unix systems and
on MS-DOS, but the MS-Windows port tried to support it by adapting the
behavior of certain primitives to the value of this variable.  It
turned out that such support cannot be reliable, so it was decided to
remove this variable in the near future.  Lisp programs are well
advised not to set it to anything but '/', because any different value
will not have any effect when support for this variable is removed.

\f
* Lisp changes made after edition 2.6 of the Emacs Lisp Manual,
(Display-related features are described in a page of their own below.)

** Function assq-delete-all replaces function assoc-delete-all.

** The new function animate-string, from lisp/play/animate.el
allows the animated display of strings.

** The new function `interactive-form' can be used to obtain the
interactive form of a function.

** The keyword :set-after in defcustom allows to specify dependencies
between custom options.  Example:

  (defcustom default-input-method nil
    "*Default input method for multilingual text (a string).
  This is the input method activated automatically by the command
  `toggle-input-method' (\\[toggle-input-method])."
    :group 'mule
    :type '(choice (const nil) string)
    :set-after '(current-language-environment))

This specifies that default-input-method should be set after
current-language-environment even if default-input-method appears
first in a custom-set-variables statement.

** The new hook `kbd-macro-termination-hook' is run at the end of
function execute-kbd-macro.  Functions on this hook are called with no
args.  The hook is run independent of how the macro was terminated
(signal or normal termination).

** Functions `butlast' and `nbutlast' for removing trailing elements
from a list are now available without requiring the CL package.

** The new user-option `even-window-heights' can be set to nil
to prevent `display-buffer' from evening out window heights.

** The user-option `face-font-registry-alternatives' specifies
alternative font registry names to try when looking for a font.

** Function `md5' calculates the MD5 "message digest"/"checksum".

** Function `delete-frame' runs `delete-frame-hook' before actually
deleting the frame.  The hook is called with one arg, the frame
being deleted.

** `add-hook' now makes the hook local if called with a non-nil LOCAL arg.

** The treatment of non-ASCII characters in search ranges has changed.
If a range in a regular expression or the arg of
skip-chars-forward/backward starts with a unibyte character C and ends
with a multibyte character C2, the range is divided into two: one is
C..?\377, the other is C1..C2, where C1 is the first character of C2's
charset.

** The new function `display-message-or-buffer' displays a message in
the echo area or pops up a buffer, depending on the length of the
message.

** The new macro `with-auto-compression-mode' allows evaluating an
expression with auto-compression-mode enabled.

** In image specifications, `:heuristic-mask' has been replaced
with the more general `:mask' property.

** Image specifications accept more `:conversion's.

** A `?' can be used in a symbol name without escaping it with a
backslash.

** Reading from the mini-buffer now reads from standard input if Emacs
is running in batch mode.  For example,

  (message "%s" (read t))

will read a Lisp expression from standard input and print the result
to standard output.

** The argument of `down-list', `backward-up-list', `up-list',
`kill-sexp', `backward-kill-sexp' and `mark-sexp' is now optional.

** If `display-buffer-reuse-frames' is set, function `display-buffer'
will raise frames displaying a buffer, instead of creating a new
frame or window.

** Two new functions for removing elements from lists/sequences
were added

- Function: remove ELT SEQ

Return a copy of SEQ with all occurrences of ELT removed.  SEQ must be
a list, vector, or string.  The comparison is done with `equal'.

- Function: remq ELT LIST

Return a copy of LIST with all occurrences of ELT removed.  The
comparison is done with `eq'.

** The function `delete' now also works with vectors and strings.

** The meaning of the `:weakness WEAK' argument of make-hash-table
has been changed: WEAK can now have new values `key-or-value' and
`key-and-value', in addition to `nil', `key', `value', and `t'.

** Function `aset' stores any multibyte character in any string
without signaling "Attempt to change char length of a string".  It may
convert a unibyte string to multibyte if necessary.

** The value of the `help-echo' text property is called as a function
or evaluated, if it is not a string already, to obtain a help string.

** Function `make-obsolete' now has an optional arg to say when the
function was declared obsolete.

** Function `plist-member' is renamed from `widget-plist-member' (which is
retained as an alias).

** Easy-menu's :filter now takes the unconverted form of the menu and
the result is automatically converted to Emacs' form.

** The new function `window-list' has been defined

- Function: window-list &optional FRAME WINDOW MINIBUF

Return a list of windows on FRAME, starting with WINDOW.  FRAME nil or
omitted means use the selected frame.  WINDOW nil or omitted means use
the selected window.  MINIBUF t means include the minibuffer window,
even if it isn't active.  MINIBUF nil or omitted means include the
minibuffer window only if it's active.  MINIBUF neither nil nor t
means never include the minibuffer window.

** There's a new function `get-window-with-predicate' defined as follows

- Function: get-window-with-predicate PREDICATE &optional MINIBUF ALL-FRAMES DEFAULT

Return a window satisfying PREDICATE.

This function cycles through all visible windows using `walk-windows',
calling PREDICATE on each one.  PREDICATE is called with a window as
argument.  The first window for which PREDICATE returns a non-nil
value is returned.  If no window satisfies PREDICATE, DEFAULT is
returned.

Optional second arg MINIBUF t means count the minibuffer window even
if not active.  MINIBUF nil or omitted means count the minibuffer iff
it is active.  MINIBUF neither t nor nil means not to count the
minibuffer even if it is active.

Several frames may share a single minibuffer; if the minibuffer
counts, all windows on all frames that share that minibuffer count
too.  Therefore, if you are using a separate minibuffer frame
and the minibuffer is active and MINIBUF says it counts,
`walk-windows' includes the windows in the frame from which you
entered the minibuffer, as well as the minibuffer window.

ALL-FRAMES is the optional third argument.
ALL-FRAMES nil or omitted means cycle within the frames as specified above.
ALL-FRAMES = `visible' means include windows on all visible frames.
ALL-FRAMES = 0 means include windows on all visible and iconified frames.
ALL-FRAMES = t means include windows on all frames including invisible frames.
If ALL-FRAMES is a frame, it means include windows on that frame.
Anything else means restrict to the selected frame.

** The function `single-key-description' now encloses function key and
event names in angle brackets.  When called with a second optional
argument non-nil, angle brackets won't be printed.

** If the variable `message-truncate-lines' is bound to t around a
call to `message', the echo area will not be resized to display that
message; it will be truncated instead, as it was done in 20.x.
Default value is nil.

** The user option `line-number-display-limit' can now be set to nil,
meaning no limit.

** The new user option `line-number-display-limit-width' controls
the maximum width of lines in a buffer for which Emacs displays line
numbers in the mode line.  The default is 200.

** `select-safe-coding-system' now also checks the most preferred
coding-system if buffer-file-coding-system is `undecided' and
DEFAULT-CODING-SYSTEM is not specified,

** The function `subr-arity' provides information about the argument
list of a primitive.

** `where-is-internal' now also accepts a list of keymaps.

** The text property `keymap' specifies a key map which overrides the
buffer's local map and the map specified by the `local-map' property.
This is probably what most current uses of `local-map' want, rather
than replacing the local map.

** The obsolete variables `before-change-function' and
`after-change-function' are no longer acted upon and have been
removed.  Use `before-change-functions' and `after-change-functions'
instead.

** The function `apropos-mode' runs the hook `apropos-mode-hook'.

** `concat' no longer accepts individual integer arguments,
as promised long ago.

** The new function `float-time' returns the current time as a float.

** The new variable auto-coding-regexp-alist specifies coding systems
for reading specific files, analogous to auto-coding-alist, but
patterns are checked against file contents instead of file names.

\f
* Lisp changes in Emacs 21.1 (see following page for display-related features)

** The new package rx.el provides an alternative sexp notation for
regular expressions.

- Function: rx-to-string SEXP

Translate SEXP into a regular expression in string notation.

- Macro: rx SEXP

Translate SEXP into a regular expression in string notation.

The following are valid subforms of regular expressions in sexp
notation.

STRING
     matches string STRING literally.

CHAR
     matches character CHAR literally.

`not-newline'
     matches any character except a newline.
                        .
`anything'
     matches any character

`(any SET)'
     matches any character in SET.  SET may be a character or string.
     Ranges of characters can be specified as `A-Z' in strings.

'(in SET)'
     like `any'.

`(not (any SET))'
     matches any character not in SET

`line-start'
     matches the empty string, but only at the beginning of a line
     in the text being matched

`line-end'
     is similar to `line-start' but matches only at the end of a line

`string-start'
     matches the empty string, but only at the beginning of the
     string being matched against.

`string-end'
     matches the empty string, but only at the end of the
     string being matched against.

`buffer-start'
     matches the empty string, but only at the beginning of the
     buffer being matched against.

`buffer-end'
     matches the empty string, but only at the end of the
     buffer being matched against.

`point'
     matches the empty string, but only at point.

`word-start'
     matches the empty string, but only at the beginning or end of a
     word.

`word-end'
     matches the empty string, but only at the end of a word.

`word-boundary'
     matches the empty string, but only at the beginning or end of a
     word.

`(not word-boundary)'
     matches the empty string, but not at the beginning or end of a
     word.

`digit'
     matches 0 through 9.

`control'
     matches ASCII control characters.

`hex-digit'
     matches 0 through 9, a through f and A through F.

`blank'
     matches space and tab only.

`graphic'
     matches graphic characters--everything except ASCII control chars,
     space, and DEL.

`printing'
     matches printing characters--everything except ASCII control chars
     and DEL.

`alphanumeric'
     matches letters and digits.  (But at present, for multibyte characters,
     it matches anything that has word syntax.)

`letter'
     matches letters.  (But at present, for multibyte characters,
     it matches anything that has word syntax.)

`ascii'
     matches ASCII (unibyte) characters.

`nonascii'
     matches non-ASCII (multibyte) characters.

`lower'
     matches anything lower-case.

`upper'
     matches anything upper-case.

`punctuation'
     matches punctuation.  (But at present, for multibyte characters,
     it matches anything that has non-word syntax.)

`space'
     matches anything that has whitespace syntax.

`word'
     matches anything that has word syntax.

`(syntax SYNTAX)'
     matches a character with syntax SYNTAX.  SYNTAX must be one
     of the following symbols.

     `whitespace'               (\\s- in string notation)
     `punctuation'              (\\s.)
     `word'                     (\\sw)
     `symbol'                   (\\s_)
     `open-parenthesis'         (\\s()
     `close-parenthesis'        (\\s))
     `expression-prefix'        (\\s')
     `string-quote'             (\\s\")
     `paired-delimiter'         (\\s$)
     `escape'                   (\\s\\)
     `character-quote'          (\\s/)
     `comment-start'            (\\s<)
     `comment-end'              (\\s>)

`(not (syntax SYNTAX))'
     matches a character that has not syntax SYNTAX.

`(category CATEGORY)'
     matches a character with category CATEGORY.  CATEGORY must be
     either a character to use for C, or one of the following symbols.

     `consonant'                        (\\c0 in string notation)
     `base-vowel'                       (\\c1)
     `upper-diacritical-mark'           (\\c2)
     `lower-diacritical-mark'           (\\c3)
     `tone-mark'                        (\\c4)
     `symbol'                           (\\c5)
     `digit'                            (\\c6)
     `vowel-modifying-diacritical-mark' (\\c7)
     `vowel-sign'                       (\\c8)
     `semivowel-lower'                  (\\c9)
     `not-at-end-of-line'               (\\c<)
     `not-at-beginning-of-line'         (\\c>)
     `alpha-numeric-two-byte'           (\\cA)
     `chinse-two-byte'                  (\\cC)
     `greek-two-byte'                   (\\cG)
     `japanese-hiragana-two-byte'       (\\cH)
     `indian-two-byte'                  (\\cI)
     `japanese-katakana-two-byte'       (\\cK)
     `korean-hangul-two-byte'           (\\cN)
     `cyrillic-two-byte'                (\\cY)
     `ascii'                            (\\ca)
     `arabic'                           (\\cb)
     `chinese'                          (\\cc)
     `ethiopic'                         (\\ce)
     `greek'                            (\\cg)
     `korean'                           (\\ch)
     `indian'                           (\\ci)
     `japanese'                         (\\cj)
     `japanese-katakana'                (\\ck)
     `latin'                            (\\cl)
     `lao'                              (\\co)
     `tibetan'                          (\\cq)
     `japanese-roman'                   (\\cr)
     `thai'                             (\\ct)
     `vietnamese'                       (\\cv)
     `hebrew'                           (\\cw)
     `cyrillic'                         (\\cy)
     `can-break'                        (\\c|)

`(not (category CATEGORY))'
     matches a character that has not category CATEGORY.

`(and SEXP1 SEXP2 ...)'
     matches what SEXP1 matches, followed by what SEXP2 matches, etc.

`(submatch SEXP1 SEXP2 ...)'
     like `and', but makes the match accessible with `match-end',
     `match-beginning', and `match-string'.

`(group SEXP1 SEXP2 ...)'
     another name for `submatch'.

`(or SEXP1 SEXP2 ...)'
     matches anything that matches SEXP1 or SEXP2, etc.  If all
     args are strings, use `regexp-opt' to optimize the resulting
     regular expression.

`(minimal-match SEXP)'
     produce a non-greedy regexp for SEXP.  Normally, regexps matching
     zero or more occurrences of something are \"greedy\" in that they
     match as much as they can, as long as the overall regexp can
     still match.  A non-greedy regexp matches as little as possible.

`(maximal-match SEXP)'
     produce a greedy regexp for SEXP.  This is the default.

`(zero-or-more SEXP)'
     matches zero or more occurrences of what SEXP matches.

`(0+ SEXP)'
     like `zero-or-more'.

`(* SEXP)'
     like `zero-or-more', but always produces a greedy regexp.

`(*? SEXP)'
     like `zero-or-more', but always produces a non-greedy regexp.

`(one-or-more SEXP)'
     matches one or more occurrences of A.

`(1+ SEXP)'
     like `one-or-more'.

`(+ SEXP)'
     like `one-or-more', but always produces a greedy regexp.

`(+? SEXP)'
     like `one-or-more', but always produces a non-greedy regexp.

`(zero-or-one SEXP)'
     matches zero or one occurrences of A.

`(optional SEXP)'
     like `zero-or-one'.

`(? SEXP)'
     like `zero-or-one', but always produces a greedy regexp.

`(?? SEXP)'
     like `zero-or-one', but always produces a non-greedy regexp.

`(repeat N SEXP)'
     matches N occurrences of what SEXP matches.

`(repeat N M SEXP)'
     matches N to M occurrences of what SEXP matches.

`(eval FORM)'
      evaluate FORM and insert result.  If result is a string,
      `regexp-quote' it.

`(regexp REGEXP)'
      include REGEXP in string notation in the result.

*** The features `md5' and `overlay' are now provided by default.

*** The special form `save-restriction' now works correctly even if the
buffer is widened inside the save-restriction and changes made outside
the original restriction.  Previously, doing this would cause the saved
restriction to be restored incorrectly.

*** The functions `find-charset-region' and `find-charset-string' include
`eight-bit-control' and/or `eight-bit-graphic' in the returned list
when they find 8-bit characters.  Previously, they included `ascii' in a
multibyte buffer and `unknown' in a unibyte buffer.

*** The functions `set-buffer-multibyte', `string-as-multibyte' and
`string-as-unibyte' change the byte sequence of a buffer or a string
if it contains a character from the `eight-bit-control' character set.

*** The handling of multibyte sequences in a multibyte buffer is
changed.  Previously, a byte sequence matching the pattern
[\200-\237][\240-\377]+ was interpreted as a single character
regardless of the length of the trailing bytes [\240-\377]+.  Thus, if
the sequence was longer than what the leading byte indicated, the
extra trailing bytes were ignored by Lisp functions.  Now such extra
bytes are independent 8-bit characters belonging to the charset
eight-bit-graphic.

** Fontsets are now implemented using char-tables.

A fontset can now be specified for each independent character, for
a group of characters or for a character set rather than just for a
character set as previously.

*** The arguments of the function `set-fontset-font' are changed.
They are NAME, CHARACTER, FONTNAME, and optional FRAME.  The function
modifies fontset NAME to use FONTNAME for CHARACTER.

CHARACTER may be a cons (FROM . TO), where FROM and TO are non-generic
characters.  In that case FONTNAME is used for all characters in the
range FROM and TO (inclusive).  CHARACTER may be a charset.  In that
case FONTNAME is used for all character in the charset.

FONTNAME may be a cons (FAMILY . REGISTRY), where FAMILY is the family
name of a font and REGISTRY is a registry name of a font.

*** Variable x-charset-registry has been deleted.  The default charset
registries of character sets are set in the default fontset
"fontset-default".

*** The function `create-fontset-from-fontset-spec' ignores the second
argument STYLE-VARIANT.  It never creates style-variant fontsets.

** The method of composing characters is changed.  Now character
composition is done by a special text property `composition' in
buffers and strings.

*** Charset composition is deleted.  Emacs never creates a `composite
character' which is an independent character with a unique character
code.  Thus the following functions handling `composite characters'
have been deleted: composite-char-component,
composite-char-component-count, composite-char-composition-rule,
composite-char-composition-rule and decompose-composite-char delete.
The variables leading-code-composition and min-composite-char have
also been deleted.

*** Three more glyph reference points are added.  They can be used to
specify a composition rule.  See the documentation of the variable
`reference-point-alist' for more detail.

*** The function `compose-region' takes new arguments COMPONENTS and
MODIFICATION-FUNC.  With COMPONENTS, you can specify not only a
composition rule but also characters to be composed.  Such characters
may differ between buffer and string text.

*** The function `compose-string' takes new arguments START, END,
COMPONENTS, and MODIFICATION-FUNC.

*** The function `compose-string' puts text property `composition'
directly on the argument STRING instead of returning a new string.
Likewise, the function `decompose-string' just removes text property
`composition' from STRING.

*** The new function `find-composition' returns information about
a composition at a specified position in a buffer or a string.

*** The function `decompose-composite-char' is now labeled as
obsolete.

** The new coding system `mac-roman' is primarily intended for use on
the Macintosh but may be used generally for Macintosh-encoded text.

** The new character sets `mule-unicode-0100-24ff',
`mule-unicode-2500-33ff', and `mule-unicode-e000-ffff' have been
introduced for Unicode characters in the range U+0100..U+24FF,
U+2500..U+33FF, U+E000..U+FFFF respectively.

Note that the character sets are not yet unified in Emacs, so
characters which belong to charsets such as Latin-2, Greek, Hebrew,
etc. and the same characters in the `mule-unicode-*' charsets are
different characters, as far as Emacs is concerned.  For example, text
which includes Unicode characters from the Latin-2 locale cannot be
encoded by Emacs with ISO 8859-2 coding system.

** The new coding system `mule-utf-8' has been added.
It provides limited support for decoding/encoding UTF-8 text.  For
details, please see the documentation string of this coding system.

** The new character sets `japanese-jisx0213-1' and
`japanese-jisx0213-2' have been introduced for the new Japanese
standard JIS X 0213 Plane 1 and Plane 2.

** The new character sets `latin-iso8859-14' and `latin-iso8859-15'
have been introduced.

** The new character sets `eight-bit-control' and `eight-bit-graphic'
have been introduced for 8-bit characters in the ranges 0x80..0x9F and
0xA0..0xFF respectively.  Note that the multibyte representation of
eight-bit-control is never exposed; this leads to an exception in the
emacs-mule coding system, which encodes everything else to the
buffer/string internal representation.  Note that to search for
eight-bit-graphic characters in a multibyte buffer, the search string
must be multibyte, otherwise such characters will be converted to
their multibyte equivalent.

** If the APPEND argument of `write-region' is an integer, it seeks to
that offset in the file before writing.

** The function `add-minor-mode' has been added for convenience and
compatibility with XEmacs (and is used internally by define-minor-mode).

** The function `shell-command' now sets the default directory of the
`*Shell Command Output*' buffer to the default directory of the buffer
from which the command was issued.

** The functions `query-replace', `query-replace-regexp',
`query-replace-regexp-eval' `map-query-replace-regexp',
`replace-string', `replace-regexp', and `perform-replace' take two
additional optional arguments START and END that specify the region to
operate on.

** The new function `count-screen-lines' is a more flexible alternative
to `window-buffer-height'.

- Function: count-screen-lines &optional BEG END COUNT-FINAL-NEWLINE WINDOW

Return the number of screen lines in the region between BEG and END.
The number of screen lines may be different from the number of actual
lines, due to line breaking, display table, etc.

Optional arguments BEG and END default to `point-min' and `point-max'
respectively.

If region ends with a newline, ignore it unless optional third argument
COUNT-FINAL-NEWLINE is non-nil.

The optional fourth argument WINDOW specifies the window used for
obtaining parameters such as width, horizontal scrolling, and so
on. The default is to use the selected window's parameters.

Like `vertical-motion', `count-screen-lines' always uses the current
buffer, regardless of which buffer is displayed in WINDOW. This makes
possible to use `count-screen-lines' in any buffer, whether or not it
is currently displayed in some window.

** The new function `mapc' is like `mapcar' but doesn't collect the
argument function's results.

** The functions base64-decode-region and base64-decode-string now
signal an error instead of returning nil if decoding fails.  Also,
`base64-decode-string' now always returns a unibyte string (in Emacs
20, it returned a multibyte string when the result was a valid multibyte
sequence).

** The function sendmail-user-agent-compose now recognizes a `body'
header in the list of headers passed to it.

** The new function member-ignore-case works like `member', but
ignores differences in case and text representation.

** The buffer-local variable cursor-type can be used to specify the
cursor to use in windows displaying a buffer.  Values are interpreted
as follows:

  t             use the cursor specified for the frame (default)
  nil           don't display a cursor
  `bar'         display a bar cursor with default width
  (bar . WIDTH) display a bar cursor with width WIDTH
  others        display a box cursor.

** The variable open-paren-in-column-0-is-defun-start controls whether
an open parenthesis in column 0 is considered to be the start of a
defun.  If set, the default, it is considered a defun start.  If not
set, an open parenthesis in column 0 has no special meaning.

** The new function `string-to-syntax' can be used to translate syntax
specifications in string form as accepted by `modify-syntax-entry' to
the cons-cell form that is used for the values of the `syntax-table'
text property, and in `font-lock-syntactic-keywords'.

Example:

  (string-to-syntax "()")
    => (4 . 41)

** Emacs' reader supports CL read syntax for integers in bases
other than 10.

*** `#BINTEGER' or `#bINTEGER' reads INTEGER in binary (radix 2).
INTEGER optionally contains a sign.

  #b1111
    => 15
  #b-1111
    => -15

*** `#OINTEGER' or `#oINTEGER' reads INTEGER in octal (radix 8).

  #o666
    => 438

*** `#XINTEGER' or `#xINTEGER' reads INTEGER in hexadecimal (radix 16).

  #xbeef
    => 48815

*** `#RADIXrINTEGER' reads INTEGER in radix RADIX, 2 <= RADIX <= 36.

  #2R-111
    => -7
  #25rah
    => 267

** The function `documentation-property' now evaluates the value of
the given property to obtain a string if it doesn't refer to etc/DOC
and isn't a string.

** If called for a symbol, the function `documentation' now looks for
a `function-documentation' property of that symbol.  If it has a non-nil
value, the documentation is taken from that value.  If the value is
not a string, it is evaluated to obtain a string.

** The last argument of `define-key-after' defaults to t for convenience.

** The new function `replace-regexp-in-string' replaces all matches
for a regexp in a string.

** `mouse-position' now runs the abnormal hook
`mouse-position-function'.

** The function string-to-number now returns a float for numbers
that don't fit into a Lisp integer.

** The variable keyword-symbols-constants-flag has been removed.
Keywords are now always considered constants.

** The new function `delete-and-extract-region' deletes text and
returns it.

** The function `clear-this-command-keys' now also clears the vector
returned by function `recent-keys'.

** Variables `beginning-of-defun-function' and `end-of-defun-function'
can be used to define handlers for the functions that find defuns.
Major modes can define these locally instead of rebinding C-M-a
etc. if the normal conventions for defuns are not appropriate for the
mode.

** easy-mmode-define-minor-mode now takes an additional BODY argument
and is renamed `define-minor-mode'.

** If an abbrev has a hook function which is a symbol, and that symbol
has a non-nil `no-self-insert' property, the return value of the hook
function specifies whether an expansion has been done or not.  If it
returns nil, abbrev-expand also returns nil, meaning "no expansion has
been performed."

When abbrev expansion is done by typing a self-inserting character,
and the abbrev has a hook with the `no-self-insert' property, and the
hook function returns non-nil meaning expansion has been done,
then the self-inserting character is not inserted.

** The function `intern-soft' now accepts a symbol as first argument.
In this case, that exact symbol is looked up in the specified obarray,
and the function's value is nil if it is not found.

** The new macro `with-syntax-table' can be used to evaluate forms
with the syntax table of the current buffer temporarily set to a
specified table.

  (with-syntax-table TABLE &rest BODY)

Evaluate BODY with syntax table of current buffer set to a copy of
TABLE.  The current syntax table is saved, BODY is evaluated, and the
saved table is restored, even in case of an abnormal exit.  Value is
what BODY returns.

** Regular expressions now support intervals \{n,m\} as well as
Perl's shy-groups \(?:...\) and non-greedy *? +? and ?? operators.
Also back-references like \2 are now considered as an error if the
corresponding subgroup does not exist (or is not closed yet).
Previously it would have been silently turned into `2' (ignoring the `\').

** The optional argument BUFFER of function file-local-copy has been
removed since it wasn't used by anything.

** The file name argument of function `file-locked-p' is now required
instead of being optional.

** The new built-in error `text-read-only' is signaled when trying to
modify read-only text.

** New functions and variables for locales.

The new variable `locale-coding-system' specifies how to encode and
decode strings passed to low-level message functions like strerror and
time functions like strftime.  The new variables
`system-messages-locale' and `system-time-locale' give the system
locales to be used when invoking these two types of functions.

The new function `set-locale-environment' sets the language
environment, preferred coding system, and locale coding system from
the system locale as specified by the LC_ALL, LC_CTYPE, and LANG
environment variables.  Normally, it is invoked during startup and need
not be invoked thereafter.  It uses the new variables
`locale-language-names', `locale-charset-language-names', and
`locale-preferred-coding-systems' to make its decisions.

** syntax tables now understand nested comments.
To declare a comment syntax as allowing nesting, just add an `n'
modifier to either of the characters of the comment end and the comment
start sequences.

** The function `pixmap-spec-p' has been renamed `bitmap-spec-p'
because `bitmap' is more in line with the usual X terminology.

** New function `propertize'

The new function `propertize' can be used to conveniently construct
strings with text properties.

- Function: propertize STRING &rest PROPERTIES

Value is a copy of STRING with text properties assigned as specified
by PROPERTIES.  PROPERTIES is a sequence of pairs PROPERTY VALUE, with
PROPERTY being the name of a text property and VALUE being the
specified value of that property.  Example:

  (propertize "foo" 'face 'bold 'read-only t)

** push and pop macros.

Simple versions of the push and pop macros of Common Lisp
are now defined in Emacs Lisp.  These macros allow only symbols
as the place that holds the list to be changed.

(push NEWELT LISTNAME)  add NEWELT to the front of LISTNAME's value.
(pop LISTNAME)          return first elt of LISTNAME, and remove it
                        (thus altering the value of LISTNAME).

** New dolist and dotimes macros.

Simple versions of the dolist and dotimes macros of Common Lisp
are now defined in Emacs Lisp.

(dolist (VAR LIST [RESULT]) BODY...)
      Execute body once for each element of LIST,
      using the variable VAR to hold the current element.
      Then return the value of RESULT, or nil if RESULT is omitted.

(dotimes (VAR COUNT [RESULT]) BODY...)
      Execute BODY with VAR bound to successive integers running from 0,
      inclusive, to COUNT, exclusive.
      Then return the value of RESULT, or nil if RESULT is omitted.

** Regular expressions now support Posix character classes such as
[:alpha:], [:space:] and so on.  These must be used within a character
class--for instance, [-[:digit:].+] matches digits or a period
or a sign.

[:digit:]  matches 0 through 9
[:cntrl:]  matches ASCII control characters
[:xdigit:]  matches 0 through 9, a through f and A through F.
[:blank:]  matches space and tab only
[:graph:]  matches graphic characters--everything except ASCII control chars,
           space, and DEL.
[:print:]  matches printing characters--everything except ASCII control chars
           and DEL.
[:alnum:]  matches letters and digits.
           (But at present, for multibyte characters,
            it matches anything that has word syntax.)
[:alpha:]  matches letters.
           (But at present, for multibyte characters,
            it matches anything that has word syntax.)
[:ascii:]  matches ASCII (unibyte) characters.
[:nonascii:]  matches non-ASCII (multibyte) characters.
[:lower:]  matches anything lower-case.
[:punct:]  matches punctuation.
           (But at present, for multibyte characters,
            it matches anything that has non-word syntax.)
[:space:]  matches anything that has whitespace syntax.
[:upper:]  matches anything upper-case.
[:word:]   matches anything that has word syntax.

** Emacs now has built-in hash tables.

The following functions are defined for hash tables:

- Function: make-hash-table ARGS

The argument list ARGS consists of keyword/argument pairs.  All arguments
are optional.  The following arguments are defined:

:test TEST

TEST must be a symbol specifying how to compare keys.  Default is `eql'.
Predefined are `eq', `eql' and `equal'.  If TEST is not predefined,
it must have been defined with `define-hash-table-test'.

:size SIZE

SIZE must be an integer > 0 giving a hint to the implementation how
many elements will be put in the hash table.  Default size is 65.

:rehash-size REHASH-SIZE

REHASH-SIZE specifies by how much to grow a hash table once it becomes
full.  If REHASH-SIZE is an integer, add that to the hash table's old
size to get the new size.  Otherwise, REHASH-SIZE must be a float >
1.0, and the new size is computed by multiplying REHASH-SIZE with the
old size.  Default rehash size is 1.5.

:rehash-threshold THRESHOLD

THRESHOLD must be a float > 0 and <= 1.0 specifying when to resize the
hash table.  It is resized when the ratio of (number of entries) /
(size of hash table) is >= THRESHOLD.  Default threshold is 0.8.

:weakness WEAK

WEAK must be either nil, one of the symbols `key, `value',
`key-or-value', `key-and-value', or t, meaning the same as
`key-and-value'.  Entries are removed from weak tables during garbage
collection if their key and/or value are not referenced elsewhere
outside of the hash table.  Default are non-weak hash tables.

- Function: makehash &optional TEST

Similar to make-hash-table, but only TEST can be specified.

- Function: hash-table-p TABLE

Returns non-nil if TABLE is a hash table object.

- Function: copy-hash-table TABLE

Returns a copy of TABLE.  Only the table itself is copied, keys and
values are shared.

- Function: hash-table-count TABLE

Returns the number of entries in TABLE.

- Function: hash-table-rehash-size TABLE

Returns the rehash size of TABLE.

- Function: hash-table-rehash-threshold TABLE

Returns the rehash threshold of TABLE.

- Function: hash-table-rehash-size TABLE

Returns the size of TABLE.

- Function: hash-table-test TABLE

Returns the test TABLE uses to compare keys.

- Function: hash-table-weakness TABLE

Returns the weakness specified for TABLE.

- Function: clrhash TABLE

Clear TABLE.

- Function: gethash KEY TABLE &optional DEFAULT

Look up KEY in TABLE and return its associated VALUE or DEFAULT if
not found.

- Function: puthash KEY VALUE TABLE

Associate KEY with VALUE in TABLE.  If KEY is already associated with
another value, replace the old value with VALUE.

- Function: remhash KEY TABLE

Remove KEY from TABLE if it is there.

- Function: maphash FUNCTION TABLE

Call FUNCTION for all elements in TABLE.  FUNCTION must take two
arguments KEY and VALUE.

- Function: sxhash OBJ

Return a hash code for Lisp object OBJ.

- Function: define-hash-table-test NAME TEST-FN HASH-FN

Define a new hash table test named NAME.  If NAME is specified as
a test in `make-hash-table', the table created will use TEST-FN for
comparing keys, and HASH-FN to compute hash codes for keys.  Test
and hash function are stored as symbol property `hash-table-test'
of NAME with a value of (TEST-FN HASH-FN).

TEST-FN must take two arguments and return non-nil if they are the same.

HASH-FN must take one argument and return an integer that is the hash
code of the argument.  The function should use the whole range of
integer values for hash code computation, including negative integers.

Example: The following creates a hash table whose keys are supposed to
be strings that are compared case-insensitively.

  (defun case-fold-string= (a b)
    (compare-strings a nil nil b nil nil t))

  (defun case-fold-string-hash (a)
    (sxhash (upcase a)))

  (define-hash-table-test 'case-fold 'case-fold-string=
                          'case-fold-string-hash))

  (make-hash-table :test 'case-fold)

** The Lisp reader handles circular structure.

It now works to use the #N= and #N# constructs to represent
circular structures.  For example, #1=(a . #1#) represents
a cons cell which is its own cdr.

** The Lisp printer handles circular structure.

If you bind print-circle to a non-nil value, the Lisp printer outputs
#N= and #N# constructs to represent circular and shared structure.

** If the second argument to `move-to-column' is anything but nil or
t, that means replace a tab with spaces if necessary to reach the
specified column, but do not add spaces at the end of the line if it
is too short to reach that column.

** perform-replace has a new feature:  the REPLACEMENTS argument may
now be a cons cell (FUNCTION . DATA).  This means to call FUNCTION
after each match to get the replacement text.  FUNCTION is called with
two arguments: DATA, and the number of replacements already made.

If the FROM-STRING contains any upper-case letters,
perform-replace also turns off `case-fold-search' temporarily
and inserts the replacement text without altering case in it.

** The function buffer-size now accepts an optional argument
to specify which buffer to return the size of.

** The calendar motion commands now run the normal hook
calendar-move-hook after moving point.

** The new variable small-temporary-file-directory specifies a
directory to use for creating temporary files that are likely to be
small.  (Certain Emacs features use this directory.)  If
small-temporary-file-directory is nil, they use
temporary-file-directory instead.

** The variable `inhibit-modification-hooks', if non-nil, inhibits all
the hooks that track changes in the buffer.  This affects
`before-change-functions' and `after-change-functions', as well as
hooks attached to text properties and overlay properties.

** assq-delete-all is a new function that deletes all the
elements of an alist which have a car `eq' to a particular value.

** make-temp-file provides a more reliable way to create a temporary file.

make-temp-file is used like make-temp-name, except that it actually
creates the file before it returns.  This prevents a timing error,
ensuring that no other job can use the same name for a temporary file.

** New exclusive-open feature in `write-region'

The optional seventh arg is now called MUSTBENEW.  If non-nil, it insists
on a check for an existing file with the same name.  If MUSTBENEW
is `excl', that means to get an error if the file already exists;
never overwrite. If MUSTBENEW is neither nil nor `excl', that means
ask for confirmation before overwriting, but do go ahead and
overwrite the file if the user gives confirmation.

If the MUSTBENEW argument in `write-region' is `excl',
that means to use a special feature in the `open' system call
to get an error if the file exists at that time.
The error reported is `file-already-exists'.

** Function `format' now handles text properties.

Text properties of the format string are applied to the result string.
If the result string is longer than the format string, text properties
ending at the end of the format string are extended to the end of the
result string.

Text properties from string arguments are applied to the result
string where arguments appear in the result string.

Example:

  (let ((s1 "hello, %s")
        (s2 "world"))
     (put-text-property 0 (length s1) 'face 'bold s1)
     (put-text-property 0 (length s2) 'face 'italic s2)
     (format s1 s2))

results in a bold-face string with an italic `world' at the end.

** Messages can now be displayed with text properties.

Text properties are handled as described above for function `format'.
The following example displays a bold-face message with an italic
argument in it.

  (let ((msg "hello, %s!")
        (arg "world"))
     (put-text-property 0 (length msg) 'face 'bold msg)
     (put-text-property 0 (length arg) 'face 'italic arg)
     (message msg arg))

** Sound support

Emacs supports playing sound files on GNU/Linux and the free BSDs
(Voxware driver and native BSD driver, aka as Luigi's driver).

Currently supported file formats are RIFF-WAVE (*.wav) and Sun Audio
(*.au).  You must configure Emacs with the option `--with-sound=yes'
to enable sound support.

Sound files can be played by calling (play-sound SOUND).  SOUND is a
list of the form `(sound PROPERTY...)'.  The function is only defined
when sound support is present for the system on which Emacs runs.  The
functions runs `play-sound-functions' with one argument which is the
sound to play, before playing the sound.

The following sound properties are supported:

- `:file FILE'

FILE is a file name.  If FILE isn't an absolute name, it will be
searched relative to `data-directory'.

- `:data DATA'

DATA is a string containing sound data.  Either :file or :data
may be present, but not both.

- `:volume VOLUME'

VOLUME must be an integer in the range 0..100 or a float in the range
0..1.  This property is optional.

- `:device DEVICE'

DEVICE is a string specifying the system device on which to play the
sound.  The default device is system-dependent.

Other properties are ignored.

An alternative interface is called as
(play-sound-file FILE &optional VOLUME DEVICE).

** `multimedia' is a new Finder keyword and Custom group.

** keywordp is a new predicate to test efficiently for an object being
a keyword symbol.

** Changes to garbage collection

*** The function garbage-collect now additionally returns the number
of live and free strings.

*** There is a new variable `strings-consed' holding the number of
strings that have been consed so far.

\f
* Lisp-level Display features added after release 2.6 of the Emacs
Lisp Manual

** The user-option `resize-mini-windows' controls how Emacs resizes
mini-windows.

** The function `pos-visible-in-window-p' now has a third optional
argument, PARTIALLY.  If a character is only partially visible, nil is
returned, unless PARTIALLY is non-nil.

** On window systems, `glyph-table' is no longer used.

** Help strings in menu items are now used to provide `help-echo' text.

** The function `image-size' can be used to determine the size of an
image.

- Function: image-size SPEC &optional PIXELS FRAME

Return the size of an image as a pair (WIDTH . HEIGHT).

SPEC is an image specification.  PIXELS non-nil means return sizes
measured in pixels, otherwise return sizes measured in canonical
character units (fractions of the width/height of the frame's default
font).  FRAME is the frame on which the image will be displayed.
FRAME nil or omitted means use the selected frame.

** The function `image-mask-p' can be used to determine if an image
has a mask bitmap.

- Function: image-mask-p SPEC &optional FRAME

Return t if image SPEC has a mask bitmap.
FRAME is the frame on which the image will be displayed.  FRAME nil
or omitted means use the selected frame.

** The function `find-image' can be used to find a usable image
satisfying one of a list of specifications.

** The STRING argument of `put-image' and `insert-image' is now
optional.

** Image specifications may contain the property `:ascent center' (see
below).

\f
* New Lisp-level Display features in Emacs 21.1

** The function tty-suppress-bold-inverse-default-colors can be used
to make Emacs avoid displaying text with bold black foreground on TTYs.

Some terminals, notably PC consoles, emulate bold text by displaying
text in brighter colors.  On such a console, a bold black foreground
is displayed in a gray color.  If this turns out to be hard to read on
your monitor---the problem occurred with the mode line on
laptops---you can instruct Emacs to ignore the text's boldness, and to
just display it black instead.

This situation can't be detected automatically.  You will have to put
a line like

  (tty-suppress-bold-inverse-default-colors t)

in your `.emacs'.

** New face implementation.

Emacs faces have been reimplemented from scratch.  They don't use XLFD
font names anymore and face merging now works as expected.

*** New faces.

Each face can specify the following display attributes:

   1. Font family or fontset alias name.

   2. Relative proportionate width, aka character set width or set
   width (swidth), e.g. `semi-compressed'.

   3. Font height in 1/10pt

   4. Font weight, e.g. `bold'.

   5. Font slant, e.g. `italic'.

   6. Foreground color.

   7. Background color.

   8. Whether or not characters should be underlined, and in what color.

   9. Whether or not characters should be displayed in inverse video.

   10. A background stipple, a bitmap.

   11. Whether or not characters should be overlined, and in what color.

   12. Whether or not characters should be strike-through, and in what
   color.

   13. Whether or not a box should be drawn around characters, its
   color, the width of the box lines, and 3D appearance.

Faces are frame-local by nature because Emacs allows to define the
same named face (face names are symbols) differently for different
frames.  Each frame has an alist of face definitions for all named
faces.  The value of a named face in such an alist is a Lisp vector
with the symbol `face' in slot 0, and a slot for each of the face
attributes mentioned above.

There is also a global face alist `face-new-frame-defaults'.  Face
definitions from this list are used to initialize faces of newly
created frames.

A face doesn't have to specify all attributes.  Those not specified
have a nil value.  Faces specifying all attributes are called
`fully-specified'.

*** Face merging.

The display style of a given character in the text is determined by
combining several faces.  This process is called `face merging'.  Any
aspect of the display style that isn't specified by overlays or text
properties is taken from the `default' face.  Since it is made sure
that the default face is always fully-specified, face merging always
results in a fully-specified face.

*** Face realization.

After all face attributes for a character have been determined by
merging faces of that character, that face is `realized'.  The
realization process maps face attributes to what is physically
available on the system where Emacs runs.  The result is a `realized
face' in form of an internal structure which is stored in the face
cache of the frame on which it was realized.

Face realization is done in the context of the charset of the
character to display because different fonts and encodings are used
for different charsets.  In other words, for characters of different
charsets, different realized faces are needed to display them.

Except for composite characters, faces are always realized for a
specific character set and contain a specific font, even if the face
being realized specifies a fontset.  The reason is that the result of
the new font selection stage is better than what can be done with
statically defined font name patterns in fontsets.

In unibyte text, Emacs' charsets aren't applicable; function
`char-charset' reports ASCII for all characters, including those >
0x7f.  The X registry and encoding of fonts to use is determined from
the variable `face-default-registry' in this case.  The variable is
initialized at Emacs startup time from the font the user specified for
Emacs.

Currently all unibyte text, i.e. all buffers with
`enable-multibyte-characters' nil are displayed with fonts of the same
registry and encoding `face-default-registry'.  This is consistent
with the fact that languages can also be set globally, only.

**** Clearing face caches.

The Lisp function `clear-face-cache' can be called to clear face caches
on all frames.  If called with a non-nil argument, it will also unload
unused fonts.

*** Font selection.

Font selection tries to find the best available matching font for a
given (charset, face) combination.  This is done slightly differently
for faces specifying a fontset, or a font family name.

If the face specifies a fontset name, that fontset determines a
pattern for fonts of the given charset.  If the face specifies a font
family, a font pattern is constructed.  Charset symbols have a
property `x-charset-registry' for that purpose that maps a charset to
an XLFD registry and encoding in the font pattern constructed.

Available fonts on the system on which Emacs runs are then matched
against the font pattern.  The result of font selection is the best
match for the given face attributes in this font list.

Font selection can be influenced by the user.

The user can specify the relative importance he gives the face
attributes width, height, weight, and slant by setting
face-font-selection-order (faces.el) to a list of face attribute
names.  The default is (:width :height :weight :slant), and means
that font selection first tries to find a good match for the font
width specified by a face, then---within fonts with that width---tries
to find a best match for the specified font height, etc.

Setting `face-font-family-alternatives' allows the user to specify
alternative font families to try if a family specified by a face
doesn't exist.

Setting `face-font-registry-alternatives' allows the user to specify
all alternative font registry names to try for a face specifying a
registry.

Please note that the interpretations of the above two variables are
slightly different.

Setting face-ignored-fonts allows the user to ignore specific fonts.


**** Scalable fonts

Emacs can make use of scalable fonts but doesn't do so by default,
since the use of too many or too big scalable fonts may crash XFree86
servers.

To enable scalable font use, set the variable
`scalable-fonts-allowed'.  A value of nil, the default, means never use
scalable fonts.  A value of t means any scalable font may be used.
Otherwise, the value must be a list of regular expressions.  A
scalable font may then be used if it matches a regular expression from
that list.  Example:

  (setq scalable-fonts-allowed '("muleindian-2$"))

allows the use of scalable fonts with registry `muleindian-2'.

*** Functions and variables related to font selection.

- Function: x-family-fonts &optional FAMILY FRAME

Return a list of available fonts of family FAMILY on FRAME.  If FAMILY
is omitted or nil, list all families.  Otherwise, FAMILY must be a
string, possibly containing wildcards `?' and `*'.

If FRAME is omitted or nil, use the selected frame.  Each element of
the result is a vector [FAMILY WIDTH POINT-SIZE WEIGHT SLANT FIXED-P
FULL REGISTRY-AND-ENCODING].  FAMILY is the font family name.
POINT-SIZE is the size of the font in 1/10 pt.  WIDTH, WEIGHT, and
SLANT are symbols describing the width, weight and slant of the font.
These symbols are the same as for face attributes.  FIXED-P is non-nil
if the font is fixed-pitch.  FULL is the full name of the font, and
REGISTRY-AND-ENCODING is a string giving the registry and encoding of
the font.  The result list is sorted according to the current setting
of the face font sort order.

- Function: x-font-family-list

Return a list of available font families on FRAME.  If FRAME is
omitted or nil, use the selected frame.  Value is a list of conses
(FAMILY . FIXED-P) where FAMILY is a font family, and FIXED-P is
non-nil if fonts of that family are fixed-pitch.

- Variable: font-list-limit

Limit for font matching.  If an integer > 0, font matching functions
won't load more than that number of fonts when searching for a
matching font.  The default is currently 100.

*** Setting face attributes.

For the most part, the new face implementation is interface-compatible
with the old one.  Old face attribute related functions are now
implemented in terms of the new functions `set-face-attribute' and
`face-attribute'.

Face attributes are identified by their names which are keyword
symbols.  All attributes can be set to `unspecified'.

The following attributes are recognized:

`:family'

VALUE must be a string specifying the font family, e.g. ``courier'',
or a fontset alias name.  If a font family is specified, wild-cards `*'
and `?' are allowed.

`:width'

VALUE specifies the relative proportionate width of the font to use.
It must be one of the symbols `ultra-condensed', `extra-condensed',
`condensed', `semi-condensed', `normal', `semi-expanded', `expanded',
`extra-expanded', or `ultra-expanded'.

`:height'

VALUE must be either an integer specifying the height of the font to use
in 1/10 pt, a floating point number specifying the amount by which to
scale any underlying face, or a function, which is called with the old
height (from the underlying face), and should return the new height.

`:weight'

VALUE specifies the weight of the font to use.  It must be one of the
symbols `ultra-bold', `extra-bold', `bold', `semi-bold', `normal',
`semi-light', `light', `extra-light', `ultra-light'.

`:slant'

VALUE specifies the slant of the font to use.  It must be one of the
symbols `italic', `oblique', `normal', `reverse-italic', or
`reverse-oblique'.

`:foreground', `:background'

VALUE must be a color name, a string.

`:underline'

VALUE specifies whether characters in FACE should be underlined.  If
VALUE is t, underline with foreground color of the face.  If VALUE is
a string, underline with that color.  If VALUE is nil, explicitly
don't underline.

`:overline'

VALUE specifies whether characters in FACE should be overlined.  If
VALUE is t, overline with foreground color of the face.  If VALUE is a
string, overline with that color.  If VALUE is nil, explicitly don't
overline.

`:strike-through'

VALUE specifies whether characters in FACE should be drawn with a line
striking through them.  If VALUE is t, use the foreground color of the
face.  If VALUE is a string, strike-through with that color.  If VALUE
is nil, explicitly don't strike through.

`:box'

VALUE specifies whether characters in FACE should have a box drawn
around them.  If VALUE is nil, explicitly don't draw boxes.  If
VALUE is t, draw a box with lines of width 1 in the foreground color
of the face.  If VALUE is a string, the string must be a color name,
and the box is drawn in that color with a line width of 1.  Otherwise,
VALUE must be a property list of the form `(:line-width WIDTH
:color COLOR :style STYLE)'.  If a keyword/value pair is missing from
the property list, a default value will be used for the value, as
specified below.  WIDTH specifies the width of the lines to draw; it
defaults to 1.  COLOR is the name of the color to draw in, default is
the foreground color of the face for simple boxes, and the background
color of the face for 3D boxes.  STYLE specifies whether a 3D box
should be draw.  If STYLE is `released-button', draw a box looking
like a released 3D button.  If STYLE is `pressed-button' draw a box
that appears like a pressed button.  If STYLE is nil, the default if
the property list doesn't contain a style specification, draw a 2D
box.

`:inverse-video'

VALUE specifies whether characters in FACE should be displayed in
inverse video. VALUE must be one of t or nil.

`:stipple'

If VALUE is a string, it must be the name of a file of pixmap data.
The directories listed in the `x-bitmap-file-path' variable are
searched.  Alternatively, VALUE may be a list of the form (WIDTH
HEIGHT DATA) where WIDTH and HEIGHT are the size in pixels, and DATA
is a string containing the raw bits of the bitmap.  VALUE nil means
explicitly don't use a stipple pattern.

For convenience, attributes `:family', `:width', `:height', `:weight',
and `:slant' may also be set in one step from an X font name:

`:font'

Set font-related face attributes from VALUE.  VALUE must be a valid
XLFD font name.  If it is a font name pattern, the first matching font
is used--this is for compatibility with the behavior of previous
versions of Emacs.

For compatibility with Emacs 20, keywords `:bold' and `:italic' can
be used to specify that a bold or italic font should be used.  VALUE
must be t or nil in that case.  A value of `unspecified' is not allowed."

Please see also the documentation of `set-face-attribute' and
`defface'.

`:inherit'

VALUE is the name of a face from which to inherit attributes, or a list
of face names.  Attributes from inherited faces are merged into the face
like an underlying face would be, with higher priority than underlying faces.

*** Face attributes and X resources

The following X resource names can be used to set face attributes
from X resources:

  Face attribute        X resource              class
-----------------------------------------------------------------------
  :family               attributeFamily .       Face.AttributeFamily
  :width                attributeWidth          Face.AttributeWidth
  :height               attributeHeight         Face.AttributeHeight
  :weight               attributeWeight         Face.AttributeWeight
  :slant                attributeSlant          Face.AttributeSlant
   foreground           attributeForeground     Face.AttributeForeground
  :background           attributeBackground .   Face.AttributeBackground
  :overline             attributeOverline       Face.AttributeOverline
  :strike-through       attributeStrikeThrough  Face.AttributeStrikeThrough
  :box                  attributeBox            Face.AttributeBox
  :underline            attributeUnderline      Face.AttributeUnderline
  :inverse-video        attributeInverse        Face.AttributeInverse
  :stipple              attributeStipple        Face.AttributeStipple
        or              attributeBackgroundPixmap
                                                Face.AttributeBackgroundPixmap
  :font                 attributeFont           Face.AttributeFont
  :bold                 attributeBold           Face.AttributeBold
  :italic               attributeItalic .       Face.AttributeItalic
  :font                 attributeFont           Face.AttributeFont

*** Text property `face'.

The value of the `face' text property can now be a single face
specification or a list of such specifications.  Each face
specification can be

1. A symbol or string naming a Lisp face.

2. A property list of the form (KEYWORD VALUE ...) where each
   KEYWORD is a face attribute name, and VALUE is an appropriate value
   for that attribute.  Please see the doc string of `set-face-attribute'
   for face attribute names.

3. Conses of the form (FOREGROUND-COLOR . COLOR) or
   (BACKGROUND-COLOR . COLOR) where COLOR is a color name.  This is
   for compatibility with previous Emacs versions.

** Support functions for colors on text-only terminals.

The function `tty-color-define' can be used to define colors for use
on TTY and MSDOS frames.  It maps a color name to a color number on
the terminal.  Emacs defines a couple of common color mappings by
default.  You can get defined colors with a call to
`defined-colors'.  The function `tty-color-clear' can be
used to clear the mapping table.

** Unified support for colors independent of frame type.

The new functions `defined-colors', `color-defined-p', `color-values',
and `display-color-p' work for any type of frame.  On frames whose
type is neither x nor w32, these functions transparently map X-style
color specifications to the closest colors supported by the frame
display.  Lisp programs should use these new functions instead of the
old `x-defined-colors', `x-color-defined-p', `x-color-values', and
`x-display-color-p'.  (The old function names are still available for
compatibility; they are now aliases of the new names.)  Lisp programs
should no more look at the value of the variable window-system to
modify their color-related behavior.

The primitives `color-gray-p' and `color-supported-p' also work for
any frame type.

** Platform-independent functions to describe display capabilities.

The new functions `display-mouse-p', `display-popup-menus-p',
`display-graphic-p', `display-selections-p', `display-screens',
`display-pixel-width', `display-pixel-height', `display-mm-width',
`display-mm-height', `display-backing-store', `display-save-under',
`display-planes', `display-color-cells', `display-visual-class', and
`display-grayscale-p' describe the basic capabilities of a particular
display.  Lisp programs should call these functions instead of testing
the value of the variables `window-system' or `system-type', or calling
platform-specific functions such as `x-display-pixel-width'.

The new function `display-images-p' returns non-nil if a particular
display can display image files.

** The minibuffer prompt is now actually inserted in the minibuffer.

This makes it possible to scroll through the prompt, if you want to.
To disallow this completely (like previous versions of emacs), customize
the variable `minibuffer-prompt-properties', and turn on the
`Inviolable' option.

The function `minibuffer-prompt-end' returns the current position of the
end of the minibuffer prompt, if the minibuffer is current.
Otherwise, it returns `(point-min)'.

** New `field' abstraction in buffers.

There is now code to support an abstraction called `fields' in emacs
buffers.  A field is a contiguous region of text with the same `field'
property (which can be a text property or an overlay).

Many emacs functions, such as forward-word, forward-sentence,
forward-paragraph, beginning-of-line, etc., stop moving when they come
to the boundary between fields; beginning-of-line and end-of-line will
not let the point move past the field boundary, but other movement
commands continue into the next field if repeated.  Stopping at field
boundaries can be suppressed programmatically by binding
`inhibit-field-text-motion' to a non-nil value around calls to these
functions.

Now that the minibuffer prompt is inserted into the minibuffer, it is in
a separate field from the user-input part of the buffer, so that common
editing commands treat the user's text separately from the prompt.

The following functions are defined for operating on fields:

- Function: constrain-to-field NEW-POS OLD-POS &optional ESCAPE-FROM-EDGE ONLY-IN-LINE INHIBIT-CAPTURE-PROPERTY

Return the position closest to NEW-POS that is in the same field as OLD-POS.

A field is a region of text with the same `field' property.
If NEW-POS is nil, then the current point is used instead, and set to the
constrained position if that is different.

If OLD-POS is at the boundary of two fields, then the allowable
positions for NEW-POS depends on the value of the optional argument
ESCAPE-FROM-EDGE: If ESCAPE-FROM-EDGE is nil, then NEW-POS is
constrained to the field that has the same `field' char-property
as any new characters inserted at OLD-POS, whereas if ESCAPE-FROM-EDGE
is non-nil, NEW-POS is constrained to the union of the two adjacent
fields.  Additionally, if two fields are separated by another field with
the special value `boundary', then any point within this special field is
also considered to be `on the boundary'.

If the optional argument ONLY-IN-LINE is non-nil and constraining
NEW-POS would move it to a different line, NEW-POS is returned
unconstrained.  This useful for commands that move by line, like
C-n or C-a, which should generally respect field boundaries
only in the case where they can still move to the right line.

If the optional argument INHIBIT-CAPTURE-PROPERTY is non-nil, and OLD-POS has
a non-nil property of that name, then any field boundaries are ignored.

Field boundaries are not noticed if `inhibit-field-text-motion' is non-nil.

- Function: delete-field &optional POS

Delete the field surrounding POS.
A field is a region of text with the same `field' property.
If POS is nil, the value of point is used for POS.

- Function: field-beginning &optional POS ESCAPE-FROM-EDGE

Return the beginning of the field surrounding POS.
A field is a region of text with the same `field' property.
If POS is nil, the value of point is used for POS.
If ESCAPE-FROM-EDGE is non-nil and POS is at the beginning of its
field, then the beginning of the *previous* field is returned.

- Function: field-end &optional POS ESCAPE-FROM-EDGE

Return the end of the field surrounding POS.
A field is a region of text with the same `field' property.
If POS is nil, the value of point is used for POS.
If ESCAPE-FROM-EDGE is non-nil and POS is at the end of its field,
then the end of the *following* field is returned.

- Function: field-string &optional POS

Return the contents of the field surrounding POS as a string.
A field is a region of text with the same `field' property.
If POS is nil, the value of point is used for POS.

- Function: field-string-no-properties &optional POS

Return the contents of the field around POS, without text-properties.
A field is a region of text with the same `field' property.
If POS is nil, the value of point is used for POS.

** Image support.

Emacs can now display images.  Images are inserted into text by giving
strings or buffer text a `display' text property containing one of
(AREA IMAGE) or IMAGE.  The display of the `display' property value
replaces the display of the characters having that property.

If the property value has the form (AREA IMAGE), AREA must be one of
`(margin left-margin)', `(margin right-margin)' or `(margin nil)'.  If
AREA is `(margin nil)', IMAGE will be displayed in the text area of a
window, otherwise it will be displayed in the left or right marginal
area.

IMAGE is an image specification.

*** Image specifications

Image specifications are lists of the form `(image PROPS)' where PROPS
is a property list whose keys are keyword symbols.  Each
specifications must contain a property `:type TYPE' with TYPE being a
symbol specifying the image type, e.g. `xbm'.  Properties not
described below are ignored.

The following is a list of properties all image types share.

`:ascent ASCENT'

ASCENT must be a number in the range 0..100, or the symbol `center'.
If it is a number, it specifies the percentage of the image's height
to use for its ascent.

If not specified, ASCENT defaults to the value 50 which means that the
image will be centered with the base line of the row it appears in.

If ASCENT is `center' the image is vertically centered around a
centerline which is the vertical center of text drawn at the position
of the image, in the manner specified by the text properties and
overlays that apply to the image.

`:margin MARGIN'

MARGIN must be either a number >= 0 specifying how many pixels to put
as margin around the image, or a pair (X . Y) with X specifying the
horizontal margin and Y specifying the vertical margin.  Default is 0.

`:relief RELIEF'

RELIEF is analogous to the `:relief' attribute of faces.  Puts a relief
around an image.

`:conversion ALGO'

Apply an image algorithm to the image before displaying it.

ALGO `laplace' or `emboss' means apply a Laplace or ``emboss''
edge-detection algorithm to the image.

ALGO `(edge-detection :matrix MATRIX :color-adjust ADJUST)' means
apply a general edge-detection algorithm.  MATRIX must be either a
nine-element list or a nine-element vector of numbers.  A pixel at
position x/y in the transformed image is computed from original pixels
around that position.  MATRIX specifies, for each pixel in the
neighborhood of x/y, a factor with which that pixel will influence the
transformed pixel; element 0 specifies the factor for the pixel at
x-1/y-1, element 1 the factor for the pixel at x/y-1 etc. as shown
below.

  (x-1/y-1  x/y-1  x+1/y-1
   x-1/y    x/y    x+1/y
   x-1/y+1  x/y+1  x+1/y+1)

The resulting pixel is computed from the color intensity of the color
resulting from summing up the RGB values of surrounding pixels,
multiplied by the specified factors, and dividing that sum by the sum
of the factors' absolute values.

Laplace edge-detection currently uses a matrix of

  (1  0  0
   0  0  0
   9  9 -1)

Emboss edge-detection uses a matrix of

  ( 2 -1  0
   -1  0  1
    0  1 -2)

ALGO `disabled' means transform the image so that it looks
``disabled''.

`:mask MASK'

If MASK is `heuristic' or `(heuristic BG)', build a clipping mask for
the image, so that the background of a frame is visible behind the
image.  If BG is not specified, or if BG is t, determine the
background color of the image by looking at the 4 corners of the
image, assuming the most frequently occurring color from the corners is
the background color of the image.  Otherwise, BG must be a list `(RED
GREEN BLUE)' specifying the color to assume for the background of the
image.

If MASK is nil, remove a mask from the image, if it has one.  Images
in some formats include a mask which can be removed by specifying
`:mask nil'.

`:file FILE'

Load image from FILE.  If FILE is not absolute after expanding it,
search for the image in `data-directory'.  Some image types support
building images from data.  When this is done, no `:file' property
may be present in the image specification.

`:data DATA'

Get image data from DATA.  (As of this writing, this is not yet
supported for image type `postscript').  Either :file or :data may be
present in an image specification, but not both.  All image types
support strings as DATA, some types allow additional types of DATA.

*** Supported image types

**** XBM, image type `xbm'.

XBM images don't require an external library.  Additional image
properties supported are:

`:foreground FG'

FG must be a string specifying the image foreground color, or nil
meaning to use the default.  Default is the frame's foreground color.

`:background BG'

BG must be a string specifying the image background color, or nil
meaning to use the default.  Default is the frame's background color.

XBM images can be constructed from data instead of file.  In this
case, the image specification must contain the following properties
instead of a `:file' property.

`:width WIDTH'

WIDTH specifies the width of the image in pixels.

`:height HEIGHT'

HEIGHT specifies the height of the image in pixels.

`:data DATA'

DATA must be either

   1. a string large enough to hold the bitmap data, i.e. it must
   have a size >= (WIDTH + 7) / 8 * HEIGHT

   2. a bool-vector of size >= WIDTH * HEIGHT

   3. a vector of strings or bool-vectors, one for each line of the
   bitmap.

   4. a string that's an in-memory XBM file.  Neither width nor
   height may be specified in this case because these are defined
   in the file.

**** XPM, image type `xpm'

XPM images require the external library `libXpm', package
`xpm-3.4k.tar.gz', version 3.4k or later.  Make sure the library is
found when Emacs is configured by supplying appropriate paths via
`--x-includes' and `--x-libraries'.

Additional image properties supported are:

`:color-symbols SYMBOLS'

SYMBOLS must be a list of pairs (NAME . COLOR), with NAME being the
name of color as it appears in an XPM file, and COLOR being an X color
name.

XPM images can be built from memory instead of files.  In that case,
add a `:data' property instead of a `:file' property.

The XPM library uses libz in its implementation so that it is able
to display compressed images.

**** PBM, image type `pbm'

PBM images don't require an external library.  Color, gray-scale and
mono images are supported.  Additional image properties supported for
mono images are:

`:foreground FG'

FG must be a string specifying the image foreground color, or nil
meaning to use the default.  Default is the frame's foreground color.

`:background FG'

BG must be a string specifying the image background color, or nil
meaning to use the default.  Default is the frame's background color.

**** JPEG, image type `jpeg'

Support for JPEG images requires the external library `libjpeg',
package `jpegsrc.v6a.tar.gz', or later.  There are no additional image
properties defined.

**** TIFF, image type `tiff'

Support for TIFF images requires the external library `libtiff',
package `tiff-v3.4-tar.gz', or later.  There are no additional image
properties defined.

**** GIF, image type `gif'

Support for GIF images requires the external library `libungif', package
`libungif-4.1.0', or later.

Additional image properties supported are:

`:index INDEX'

INDEX must be an integer >= 0.  Load image number INDEX from a
multi-image GIF file.  If INDEX is too large, the image displays
as a hollow box.

This could be used to implement limited support for animated GIFs.
For example, the following function displays a multi-image GIF file
at point-min in the current buffer, switching between sub-images
every 0.1 seconds.

(defun show-anim (file max)
  "Display multi-image GIF file FILE which contains MAX subimages."
  (display-anim (current-buffer) file 0 max t))

(defun display-anim (buffer file idx max first-time)
  (when (= idx max)
    (setq idx 0))
  (let ((img (create-image file nil nil :index idx)))
    (save-excursion
      (set-buffer buffer)
      (goto-char (point-min))
      (unless first-time (delete-char 1))
      (insert-image img "x"))
    (run-with-timer 0.1 nil 'display-anim buffer file (1+ idx) max nil)))

**** PNG, image type `png'

Support for PNG images requires the external library `libpng',
package `libpng-1.0.2.tar.gz', or later.  There are no additional image
properties defined.

**** Ghostscript, image type `postscript'.

Additional image properties supported are:

`:pt-width WIDTH'

WIDTH is width of the image in pt (1/72 inch).  WIDTH must be an
integer.  This is a required property.

`:pt-height HEIGHT'

HEIGHT specifies the height of the image in pt (1/72 inch).  HEIGHT
must be a integer.  This is an required property.

`:bounding-box BOX'

BOX must be a list or vector of 4 integers giving the bounding box of
the PS image, analogous to the `BoundingBox' comment found in PS
files.  This is an required property.

Part of the Ghostscript interface is implemented in Lisp.  See
lisp/gs.el.

*** Lisp interface.

The variable `image-types' contains a list of those image types
which are supported in the current configuration.

Images are stored in an image cache and removed from the cache when
they haven't been displayed for `image-cache-eviction-delay seconds.
The function `clear-image-cache' can be used to clear the image cache
manually.  Images in the cache are compared with `equal', i.e. all
images with `equal' specifications share the same image.

*** Simplified image API, image.el

The new Lisp package image.el contains functions that simplify image
creation and putting images into text.  The function `create-image'
can be used to create images.  The macro `defimage' can be used to
define an image based on available image types.  The functions
`put-image' and `insert-image' can be used to insert an image into a
buffer.

** Display margins.

Windows can now have margins which are used for special text
and images.

To give a window margins, either set the buffer-local variables
`left-margin-width' and `right-margin-width', or call
`set-window-margins'.  The function `window-margins' can be used to
obtain the current settings.  To make `left-margin-width' and
`right-margin-width' take effect, you must set them before displaying
the buffer in a window, or use `set-window-buffer' to force an update
of the display margins.

You can put text in margins by giving it a `display' text property
containing a pair of the form `(LOCATION . VALUE)', where LOCATION is
one of `left-margin' or `right-margin' or nil.  VALUE can be either a
string, an image specification or a stretch specification (see later
in this file).

** Help display

Emacs displays short help messages in the echo area, when the mouse
moves over a tool-bar item or a piece of text that has a text property
`help-echo'.  This feature also applies to strings in the mode line
that have a `help-echo' property.

If the value of the `help-echo' property is a function, that function
is called with three arguments WINDOW, OBJECT and POSITION.  WINDOW is
the window in which the help was found.

If OBJECT is a buffer, POS is the position in the buffer where the
`help-echo' text property was found.

If OBJECT is an overlay, that overlay has a `help-echo' property, and
POS is the position in the overlay's buffer under the mouse.

If OBJECT is a string (an overlay string or a string displayed with
the `display' property), POS is the position in that string under the
mouse.

If the value of the `help-echo' property is neither a function nor a
string, it is evaluated to obtain a help string.

For tool-bar and menu-bar items, their key definition is used to
determine the help to display.  If their definition contains a
property `:help FORM', FORM is evaluated to determine the help string.
For tool-bar items without a help form, the caption of the item is
used as help string.

The hook `show-help-function' can be set to a function that displays
the help string differently.  For example, enabling a tooltip window
causes the help display to appear there instead of in the echo area.

** Vertical fractional scrolling.

The display of text in windows can be scrolled smoothly in pixels.
This is useful, for example, for making parts of large images visible.

The function `window-vscroll' returns the current value of vertical
scrolling, a non-negative fraction of the canonical character height.
The function `set-window-vscroll' can be used to set the vertical
scrolling value.  Here is an example of how these function might be
used.

  (global-set-key [A-down]
    #'(lambda ()
        (interactive)
        (set-window-vscroll (selected-window)
                            (+ 0.5 (window-vscroll)))))
  (global-set-key [A-up]
    #'(lambda ()
        (interactive)
        (set-window-vscroll (selected-window)
                            (- (window-vscroll) 0.5)))))

** New hook `fontification-functions'.

Functions from `fontification-functions' are called from redisplay
when it encounters a region of text that is not yet fontified.  This
variable automatically becomes buffer-local when set.  Each function
is called with one argument, POS.

At least one of the hook functions should fontify one or more
characters starting at POS in the current buffer.  It should mark them
as fontified by giving them a non-nil value of the `fontified' text
property.  It may be reasonable for these functions to check for the
`fontified' property and not put it back on, but they do not have to.

** Tool bar support.

Emacs supports a tool bar at the top of a frame under X.  The frame
parameter `tool-bar-lines' (X resource "toolBar", class "ToolBar")
controls how may lines to reserve for the tool bar.  A zero value
suppresses the tool bar.  If the value is non-zero and
`auto-resize-tool-bars' is non-nil the tool bar's size will be changed
automatically so that all tool bar items are visible.

*** Tool bar item definitions

Tool bar items are defined using `define-key' with a prefix-key
`tool-bar'.  For example `(define-key global-map [tool-bar item1] ITEM)'
where ITEM is a list `(menu-item CAPTION BINDING PROPS...)'.

CAPTION is the caption of the item, If it's not a string, it is
evaluated to get a string.  The caption is currently not displayed in
the tool bar, but it is displayed if the item doesn't have a `:help'
property (see below).

BINDING is the tool bar item's binding.  Tool bar items with keymaps as
binding are currently ignored.

The following properties are recognized:

`:enable FORM'.

FORM is evaluated and specifies whether the tool bar item is enabled
or disabled.

`:visible FORM'

FORM is evaluated and specifies whether the tool bar item is displayed.

`:filter FUNCTION'

FUNCTION is called with one parameter, the same list BINDING in which
FUNCTION is specified as the filter.  The value FUNCTION returns is
used instead of BINDING to display this item.

`:button (TYPE SELECTED)'

TYPE must be one of `:radio' or `:toggle'.  SELECTED is evaluated
and specifies whether the button is selected (pressed) or not.

`:image IMAGES'

IMAGES is either a single image specification or a vector of four
image specifications.  If it is a vector, this table lists the
meaning of each of the four elements:

   Index        Use when item is
   ----------------------------------------
     0          enabled and selected
     1          enabled and deselected
     2          disabled and selected
     3          disabled and deselected

If IMAGE is a single image specification, a Laplace edge-detection
algorithm is used on that image to draw the image in disabled state.

`:help HELP-STRING'.

Gives a help string to display for the tool bar item.  This help
is displayed when the mouse is moved over the item.

The function `toolbar-add-item' is a convenience function for adding
toolbar items generally, and `tool-bar-add-item-from-menu' can be used
to define a toolbar item with a binding copied from an item on the
menu bar.

The default bindings use a menu-item :filter to derive the tool-bar
dynamically from variable `tool-bar-map' which may be set
buffer-locally to override the global map.

*** Tool-bar-related variables.

If `auto-resize-tool-bar' is non-nil, the tool bar will automatically
resize to show all defined tool bar items.  It will never grow larger
than 1/4 of the frame's size.

If `auto-raise-tool-bar-buttons' is non-nil, tool bar buttons will be
raised when the mouse moves over them.

You can add extra space between tool bar items by setting
`tool-bar-button-margin' to a positive integer specifying a number of
pixels, or a pair of integers (X . Y) specifying horizontal and
vertical margins .  Default is 1.

You can change the shadow thickness of tool bar buttons by setting
`tool-bar-button-relief' to an integer.  Default is 3.

*** Tool-bar clicks with modifiers.

You can bind commands to clicks with control, shift, meta etc. on
a tool bar item.  If

  (define-key global-map [tool-bar shell]
    '(menu-item "Shell" shell
                :image (image :type xpm :file "shell.xpm")))

is the original tool bar item definition, then

  (define-key global-map [tool-bar S-shell] 'some-command)

makes a binding to run `some-command' for a shifted click on the same
item.

** Mode line changes.

*** Mouse-sensitive mode line.

The mode line can be made mouse-sensitive by displaying strings there
that have a `local-map' text property.  There are three ways to display
a string with a `local-map' property in the mode line.

1. The mode line spec contains a variable whose string value has
a `local-map' text property.

2. The mode line spec contains a format specifier (e.g. `%12b'), and
that format specifier has a `local-map' property.

3. The mode line spec contains a list containing `:eval FORM'.  FORM
is evaluated.  If the result is a string, and that string has a
`local-map' property.

The same mechanism is used to determine the `face' and `help-echo'
properties of strings in the mode line.  See `bindings.el' for an
example.

*** If a mode line element has the form `(:eval FORM)', FORM is
evaluated and the result is used as mode line element.

*** You can suppress mode-line display by setting the buffer-local
variable mode-line-format to nil.

*** A headerline can now be displayed at the top of a window.

This mode line's contents are controlled by the new variable
`header-line-format' and `default-header-line-format' which are
completely analogous to `mode-line-format' and
`default-mode-line-format'.  A value of nil means don't display a top
line.

The appearance of top mode lines is controlled by the face
`header-line'.

The function `coordinates-in-window-p' returns `header-line' for a
position in the header-line.

** Text property `display'

The `display' text property is used to insert images into text,
replace text with other text, display text in marginal area, and it is
also used to control other aspects of how text displays.  The value of
the `display' property should be a display specification, as described
below, or a list or vector containing display specifications.

*** Replacing text, displaying text in marginal areas

To replace the text having the `display' property with some other
text, use a display specification of the form `(LOCATION STRING)'.

If LOCATION is `(margin left-margin)', STRING is displayed in the left
marginal area, if it is `(margin right-margin)', it is displayed in
the right marginal area, and if LOCATION is `(margin nil)' STRING
is displayed in the text.  In the latter case you can also use the
simpler form STRING as property value.

*** Variable width and height spaces

To display a space of fractional width or height, use a display
specification of the form `(LOCATION STRECH)'.  If LOCATION is
`(margin left-margin)', the space is displayed in the left marginal
area, if it is `(margin right-margin)', it is displayed in the right
marginal area, and if LOCATION is `(margin nil)' the space is
displayed in the text.  In the latter case you can also use the
simpler form STRETCH as property value.

The stretch specification STRETCH itself is a list of the form `(space
PROPS)', where PROPS is a property list which can contain the
properties described below.

The display of the fractional space replaces the display of the
characters having the `display' property.

- :width WIDTH

Specifies that the space width should be WIDTH times the normal
character width.  WIDTH can be an integer or floating point number.

- :relative-width FACTOR

Specifies that the width of the stretch should be computed from the
first character in a group of consecutive characters that have the
same `display' property.  The computation is done by multiplying the
width of that character by FACTOR.

- :align-to HPOS

Specifies that the space should be wide enough to reach HPOS.  The
value HPOS is measured in units of the normal character width.

Exactly one of the above properties should be used.

- :height HEIGHT

Specifies the height of the space, as HEIGHT, measured in terms of the
normal line height.

- :relative-height FACTOR

The height of the space is computed as the product of the height
of the text having the `display' property and FACTOR.

- :ascent ASCENT

Specifies that ASCENT percent of the height of the stretch should be
used for the ascent of the stretch, i.e. for the part above the
baseline.  The value of ASCENT must be a non-negative number less or
equal to 100.

You should not use both `:height' and `:relative-height' together.

*** Images

A display specification for an image has the form `(LOCATION
. IMAGE)', where IMAGE is an image specification.  The image replaces,
in the display, the characters having this display specification in
their `display' text property.  If LOCATION is `(margin left-margin)',
the image will be displayed in the left marginal area, if it is
`(margin right-margin)' it will be displayed in the right marginal
area, and if LOCATION is `(margin nil)' the image will be displayed in
the text.  In the latter case you can also use the simpler form IMAGE
as display specification.

*** Other display properties

- (space-width FACTOR)

Specifies that space characters in the text having that property
should be displayed FACTOR times as wide as normal; FACTOR must be an
integer or float.

- (height HEIGHT)

Display text having this property in a font that is smaller or larger.

If HEIGHT is a list of the form `(+ N)', where N is an integer, that
means to use a font that is N steps larger.  If HEIGHT is a list of
the form `(- N)', that means to use a font that is N steps smaller.  A
``step'' is defined by the set of available fonts; each size for which
a font is available counts as a step.

If HEIGHT is a number, that means to use a font that is HEIGHT times
as tall as the frame's default font.

If HEIGHT is a symbol, it is called as a function with the current
height as argument.  The function should return the new height to use.

Otherwise, HEIGHT is evaluated to get the new height, with the symbol
`height' bound to the current specified font height.

- (raise FACTOR)

FACTOR must be a number, specifying a multiple of the current
font's height.  If it is positive, that means to display the characters
raised.  If it is negative, that means to display them lower down.  The
amount of raising or lowering is computed without taking account of the
`height' subproperty.

*** Conditional display properties

All display specifications can be conditionalized.  If a specification
has the form `(when CONDITION . SPEC)', the specification SPEC applies
only when CONDITION yields a non-nil value when evaluated.  During the
evaluation, `object' is bound to the string or buffer having the
conditional display property; `position' and `buffer-position' are
bound to the position within `object' and the buffer position where
the display property was found, respectively.  Both positions can be
different when object is a string.

The normal specification consisting of SPEC only is equivalent to
`(when t . SPEC)'.

** New menu separator types.

Emacs now supports more than one menu separator type.  Menu items with
item names consisting of dashes only (including zero dashes) are
treated like before.  In addition, the following item names are used
to specify other menu separator types.

- `--no-line' or `--space', or `--:space', or `--:noLine'

No separator lines are drawn, but a small space is inserted where the
separator occurs.

- `--single-line' or `--:singleLine'

A single line in the menu's foreground color.

- `--double-line' or `--:doubleLine'

A double line in the menu's foreground color.

- `--single-dashed-line' or `--:singleDashedLine'

A single dashed line in the menu's foreground color.

- `--double-dashed-line' or `--:doubleDashedLine'

A double dashed line in the menu's foreground color.

- `--shadow-etched-in' or `--:shadowEtchedIn'

A single line with 3D sunken appearance.  This is the form
displayed for item names consisting of dashes only.

- `--shadow-etched-out' or `--:shadowEtchedOut'

A single line with 3D raised appearance.

- `--shadow-etched-in-dash' or `--:shadowEtchedInDash'

A single dashed line with 3D sunken appearance.

- `--shadow-etched-out-dash' or `--:shadowEtchedOutDash'

A single dashed line with 3D raise appearance.

- `--shadow-double-etched-in' or `--:shadowDoubleEtchedIn'

Two lines with 3D sunken appearance.

- `--shadow-double-etched-out' or `--:shadowDoubleEtchedOut'

Two lines with 3D raised appearance.

- `--shadow-double-etched-in-dash' or `--:shadowDoubleEtchedInDash'

Two dashed lines with 3D sunken appearance.

- `--shadow-double-etched-out-dash' or `--:shadowDoubleEtchedOutDash'

Two dashed lines with 3D raised appearance.

Under LessTif/Motif, the last four separator types are displayed like
the corresponding single-line separators.

** New frame parameters for scroll bar colors.

The new frame parameters `scroll-bar-foreground' and
`scroll-bar-background' can be used to change scroll bar colors.
Their value must be either a color name, a string, or nil to specify
that scroll bars should use a default color.  For toolkit scroll bars,
default colors are toolkit specific.  For non-toolkit scroll bars, the
default background is the background color of the frame, and the
default foreground is black.

The X resource name of these parameters are `scrollBarForeground'
(class ScrollBarForeground) and `scrollBarBackground' (class
`ScrollBarBackground').

Setting these parameters overrides toolkit specific X resource
settings for scroll bar colors.

** You can set `redisplay-dont-pause' to a non-nil value to prevent
display updates from being interrupted when input is pending.

** Changing a window's width may now change its window start if it
starts on a continuation line.  The new window start is computed based
on the window's new width, starting from the start of the continued
line as the start of the screen line with the minimum distance from
the original window start.

** The variable `hscroll-step' and the functions
`hscroll-point-visible' and `hscroll-window-column' have been removed
now that proper horizontal scrolling is implemented.

** Windows can now be made fixed-width and/or fixed-height.

A window is fixed-size if its buffer has a buffer-local variable
`window-size-fixed' whose value is not nil.  A value of `height' makes
windows fixed-height, a value of `width' makes them fixed-width, any
other non-nil value makes them both fixed-width and fixed-height.

The following code makes all windows displaying the current buffer
fixed-width and fixed-height.

  (set (make-local-variable 'window-size-fixed) t)

A call to enlarge-window on a window gives an error if that window is
fixed-width and it is tried to change the window's width, or if the
window is fixed-height, and it is tried to change its height.  To
change the size of a fixed-size window, bind `window-size-fixed'
temporarily to nil, for example

  (let ((window-size-fixed nil))
     (enlarge-window 10))

Likewise, an attempt to split a fixed-height window vertically,
or a fixed-width window horizontally results in a error.

** The cursor-type frame parameter is now supported on MS-DOS
terminals.  When Emacs starts, it by default changes the cursor shape
to a solid box, as it does on Unix.  The `cursor-type' frame parameter
overrides this as it does on Unix, except that the bar cursor is
horizontal rather than vertical (since the MS-DOS display doesn't
support a vertical-bar cursor).


\f
----------------------------------------------------------------------
Copyright information:

Copyright (C) 2000, 2001, 2002, 2003, 2006
          Free Software Foundation, Inc.

   Permission is granted to anyone to make or distribute verbatim copies
   of this document as received, in any medium, provided that the
   copyright notice and this permission notice are preserved,
   thus giving the recipient permission to redistribute in turn.

   Permission is granted to distribute modified versions
   of this document, or of portions of it,
   under the above conditions, provided also that they
   carry prominent notices stating who last changed them.
\f
Local variables:
mode: outline
paragraph-separate: "[  \f]*$"
end: