16. TIPS

This section of the manual explains some of the more complex, obscure, and useful features of Elvis (or larger subjects). The following subjects are discussed:
  • 16.1 Information via the Web
  • 16.2 Using Elvis as a Web browser
  • 16.3 How to debug macros
  • 16.4 Running your compiler from within Elvis
  • 16.5 Internationalization
  • 16.6 Aliases
  • 16.6.1 Some example aliases
  • 16.6.2 Using aliases to add network protocols
  • 16.7 The spell-checker
  • 16.8 Auto commands
  • 16.8.3 Events
  • 16.8.4 Patterns
  • 16.9 How to make Elvis run faster
  • 16.1 Information via the Web

    Here are some URLs (World Wide Web links to other documents) which are relevant to vi. Each of these, in turn, has links to other sites. Directly or indirectly, these links will lead you to a huge amount of information about vi.

    I've tried to limit this list to advertised sites; there are others that I run across occasionally, but their URLs tend to vary over time, so it isn't a good idea to place them in a static document such as this one. It makes for a short list, though.

    ftp://ftp.cs.pdx.edu/pub/elvis/README.html
    The "home" site of Elvis. This is where you can find the latest official release.
    ftp://ftp.cs.pdx.edu/pub/elvis/unreleased/README.html
    Prerelease versions of Elvis can often be found here. Prerelease versions are identified by a letter appended to the version number they they're expected to be released as. For example, 2.2a is a prerelease version of what will eventually be released as version 2.2. Also, I append "-alpha" in the early stages when new features are still being added, or "-beta" if no new features are expected before the release. Typically, new versions are uploaded once a month or so; watch for announcements on the comp.editors newsgroup.
    http://elvis.vi-editor.org/index.html
    This where Herbert (a.k.a. Martin Dietze, who ported Elvis to OS/2) is putting an Elvis home page. It has links to all relevant web pages, archives, and people's email addresses. The manual will also be available here. He is also considering adding a searchable online database of aliases and how-to articles. This may also become the preferred way to submit bug reports, suggestions, or any other contribution to the project.
    http://georg.f-451.net/elvis/
    This is Georg Neis' web page for Elvis. He's the maintainer of the Debian Elvis package.
    http://groups.yahoo.com/group/elvis-editor/
    This is a Yahoo mailing group. It is used mostly by Elvis' developers, but user questions are welcome too.
    http://www.phys.columbia.edu/~flame/vi.htm
    This is a Web page for Elvis, maintained by JaeSub Hong. It has some screen shots, macros, and syntax-coloring definitions. It also has the online manual.
    http://www.thomer.com/vi/vi.html
    "The VI lover's home page," by Thomer Gil. This contains links to a wide variety of vi documentation, and practically all vi clones. A very good resource! This is a new location; previously it was located at http://www.cs.vu.nl/~tmgil/vi.html.
    vi-editor.org
    Sven Guckes' generic vi page. He and Thomer Gil seem to have a friendly competition going on to see who can make the best vi page, and we are all beneficiaries.
    http://www.bsyse.wsu.edu/~rnelson/editors/editors.htm
    A compendium of text editors. It describes all kinds of text editors, not just vi clones. It lists the features and supported platforms for each editor.
    http://bluneon.gq.nu/puter/ed00.html
    Another compendium of text editors.' This one is oriented toward writing natural language text, instead of computer source code. The maintainer specifically mentions the alt.tv.x-files.creative newsgroup.
    http://darren.hiebert.com/ctags/
    Home page for Darren Hiebert's exuberant ctags program. It supports all features of Elvis' ctags, except for the "ln" attribute, and the "-r" flag for generating a "refs" file. Exuberant ctags has a smarter parsing algorithm, which causes it to generate fewer bogus tags (i.e., tags for things that really shouldn't have tags). It also adds the ability to generate tags for "enum" values and a few other useful things.
    http://www.fleiner.com/jtags/
    Home page of Claudio Fleiner's "jtags" program -- a tag file generator for Java source.
    http://www.tamacom.com/global/
    Home page for Shigio Yamaguchi's global function reference utility. You can make Elvis use it instead of the normal built-in tag searcher by giving Elvis the following command:
    :se tagprg="global -t $1"

    Once you have run global's gtags utility, you can do some powerful things such as search for each place where a function (e.g., m_front()) is called, like this:

    :ta -r m_front
    http://www.nicemice.net/par/
    Home page for Adam Costello's par program, which is a much more sophisticated text formatter than the fmt program distributed with Elvis. It even does a very good job of reformatting comments! Sadly, it is one of those rare programs that doesn't handle tab characters correctly.
    http://www.cygwin.com/
    This is the home page for the CygWin tools -- ports of GNU utilities to Microsoft Windows, by Cygnus and Red Hat. If you're looking for more Unix utilities to go with Elvis, this is the place.
    ftp://alf.uib.no/pub/vi
    An archive site containing many macro packages and other information about vi. Nearly all of it should apply equally well to Elvis.
    http://developer.berlios.de/projects/ex-vi/
    This is the source code to the real vi!
    http://www.vim.org/
    http://invisible-island.net/vile/vile.html
    http://www.bostic.com/vi/
    http://www.winvi.de/
    http://www.fwiarda.com/software.htm#pvic
    http://jvi.sourceforge.net/
    These are the home pages for some other vi clones: vim, vile, nvi, WinVi, pvic, and javi respectively. The last one is interesting because it is written in Java!
    http://www.vim.org/util.html
    The Vim utilities page. Most of these also work with Elvis.
    ftp://ftp.funet.fi/pub/doc/posix/p1003.2a/d8/5.10
    ftp://ftp.funet.fi/pub/doc/posix/p1003.2a/d8/5.35
    These are old drafts of the POSIX standards for ex and vi, respectively. These URLs might not be valid very long.
    http://docs.freebsd.org/44doc/usd/12.vi/paper.html
    This is the official BSD documentation for vi.
    http://roxanne.roxanne.org/~eric/vi_editor/
    An HTML version of E. Larry Lidz' "vi editor FAQ". A plain-text version of this FAQ is occasionally posted in the comp.editors newsgroup. (I think the official URL of the HTML version is http://www.macom.co.il/vi/index.html, but that site seems to be unreachable now.)
    http://alumni.caltech.edu/~dank/nansi/
    This is a home page for the NANSI.SYS and NNANSI.SYS drivers, which accelerate the screen updates under MS-DOS.
    ftp://ftp.leo.org:/pub/comp/os/os2/leo/gnu/emx+gcc/
    ftp://ftp-os2.cdrom.com:/pub/os2/emx09d/
    ftp://ftp-os2.nmsu.edu:/pub/os2/dev/emx/v0.9d/
    These are all places where you can find the EMX.DLL library for OS/2. The first one is the "home" site, and the others are popular OS/2 archives. You need EMX only if you want to run the "termcap" (elvisemx.exe) or "x11" (elvisx11.exe) versions of Elvis under OS/2.
    http://set.gmd.de/~veit/os2/xf86os2.html
    This is the home page for the OS/2 version of XFree86. You need this (in addition to EMX) if you want to run the graphical X11 version of Elvis.
    ftp://ftp.ox.ac.uk/pub/wordlists/
    A collection of word lists, for a variety of languages. This may be useful in conjunction with Elvis' spell checker. The home page for these lists is http://www.hyphenologist.co.uk/.
    http://www.mso.anu.edu.au/~ralph/OPTED/
    OPTED is a public domain english-language dictionary, distributed as a set of HTML files. It is part of Project Gutenburg, and is derived from an old Webster's Unabridged Dictionary. This is more than just a word list -- it includes definitions for the words.
    http://www.dcs.shef.ac.uk/research/ilash/Moby/
    The Moby project -- a collection of word lists and similar resources.
    http://www.htmlcodetutorial.com/quicklist.html
    This is an online HTML authoring site. Despite the name, it works better as a reference than as a tutorial. The above link is actually just a page within the larger site, because the main page (http://www.htmlcodetutorial.com/) has a big nasty pop-up advertisement. Sadly, their pages look ugly when viewed with Elvis because they depend on JavaScript and forms.

    16.2 Using Elvis as a Web browser

    NOTE: The following information doesn't apply to the MS-DOS version of Elvis, because that version doesn't support the ftp and http protocols. But for Win32, Unix, and OS/2...

    You can use Elvis as a light-weight Web browser. Surfing with Elvis isn't as much fun as surfing with a multimedia-capable browser such as Netscape or MSIE, but Elvis does have some advantages: it starts up much faster, it feels like vi, and you can edit whatever you download.

    There's no special trick to loading a Web page. Just give a URL where Elvis expects a filename, and Elvis will read the Web page via the Internet. You can follow links in Web pages just as you do in Elvis' online manual. HTML pages are displayed in the "html" display mode, and anything else uses the "hex" or "normal" display mode by default. All data is fetched in binary so data files aren't mangled; however, this also means that newlines aren't converted, which may make non-HTML text files look ugly.

    Elvis has built-in support for the HTTP and FTP protocols. Other protocols may be indirectly supported, via an HTTP proxy as indicated by the elvis.net configuration file.

    By default, FTP access is anonymous. However, if you give a file name which starts with "/~/" then Elvis will attempt to login to the FTP server using you own account, as described in the elvis.ftp or ~/.netrc configuration file. You can also use "/~username/" to use some other user account listed in .netrc. For example, "ftp://localhost/directory/file" uses anonymous FTP, but "ftp://localhost/~/directory/file" uses your own account.

    Elvis can write via FTP as well as read; see the Internet chapter.

    Elvis also doesn't support inline graphic images, but that isn't as big of a problem as you might think. If you download an image, Elvis will simply load it into a buffer and then display that buffer in the "hex" display mode. You can then write that buffer's contents to a file, or in Unix you can send it directly to an image viewer via a command such as ":w !xv -".

    The easiest way to save an image (or any other buffer) to a local file is via the command ":w (dirfile(filename))". In fact, you might want to add the following lines to your ~/.exrc file to make the (F2) key save the current buffer to a file, and the (F3) key send it to the xv image viewer:

    	map #2 :w (dirfile(filename))^M
    	map #3 :w !xv -^M

    To make images easier to fetch, any <IMG SRC=url> tag which isn't already part of a hypertext link will be interpreted as a link to the SRC url. This allows you to download an image by moving the cursor onto it and hitting the (Enter) key.

    Elvis doesn't support frames either, so a similar trick was used for <FRAME SRC=url> tags. Elvis displays the name of each frame; those names serve as links to the contents of the frame.

    Because Elvis is primarily an editor, not a Web browser, I deliberately made the "html" display mode rather picky, so that any questionable entities in your own HTML documents will call attention to themselves. When you're using Elvis to browse other peoples' documents, though, this can be annoying, so I modified it slightly to be more forgiving when you're viewing read-only documents. (All Web pages are read-only.)

    And you already know that Elvis' support for the <TABLE> macros is very poor, right? If you encounter a Web page which looks really ugly in Elvis, you can bet it uses tables.

    Elvis doesn't support forms, or secure connections. Well, Elvis can display mock-ups of forms; they just don't work. They probably never will. You have to draw the line somewhere.

    Elvis doesn't always choose the best display mode for HTML pages. It uses "html" if the file name ends with ".html" or ".htm", or if the document's text begins with "<!", "<H", or "<h". For all other documents, it uses the "normal" or "hex" display mode by default. If Elvis chooses the wrong display mode, you can use the :display command to switch to a different display mode.

    The command ":e foo" will always load the local file "foo" from your current directory. This is true of all commands which normally act on files -- unless you give a complete URL, Elvis assumes it should work with local files. However, while in the "html" display mode, the command ":ta foo" will use the same protocol, site, and directory as the page you're already viewing, because that's how the "html" display mode interprets tags.

    16.3 How to debug macros

    There are two ways to create a macro in Elvis: You can either assign a series of commands to a keystroke (or series of keystrokes) via the :map command, or you can store a series of commands in a cut buffer and execute them via the visual @x command. You will often use a combination of techniques, in which :map macro constructs a customized @x macro and runs it.

    (Aliases are a separate issue, discussed later in this chapter. The information in this section does not apply to aliases. Elvis does not yet offer any special tools for debugging aliases.)

    Elvis has several features that make debugging macros much easier. For example, you can create a window which continuously displays the contents of a given cut buffer, such as "m, via the command:

    	:(Elvis cut buffer m)split
    
    or, more concisely:
    	:("m)sp
    

    Note: The cut buffer must exist before you can display it. Cut buffers are created the first time anything is yanked into them.

    The maptrace option allows you to trace the execution of macros. You can either allow it to run through the macro, or wait for a keypress before each mapped command character. You can also use the :break and :unbreak commands to set or clear a breakpoint on a given :map macro. Breakpoints cause the maptrace option to switch from "run" to "step" when that macro is expanded.

    The maplog option can be used to log the trace information to a buffer named "Elvis map log". The idea here is that you will give the command...

    	:se mt=r mlog=r
    
    ... (or its full-length form, :set maptrace=run maplog=reset) before starting the macro, and then when the macro fails you can give the command...
    	:("Eml)sp
    
    ... (or its full-length form, :(Elvis map log)split) to see what it was doing shortly before the failure. Note that the maplog option has no effect if maptrace is "off".

    Warning: Elvis has a single keystroke queue which is shared by all windows. Because of this, while Elvis is running a macro in one window you can't switch to another window and type in commands. Even if the GUI allows you to switch windows without using the keyboard, doing so will simply force the macro to continue execution in the new window. So don't switch windows while a macro is running!

    Here's a debugging methodology that works for me:

    1. Begin by loading the macro package and a test file.
    2. Give the command ":se mt=r mlog=r", and run the macro.
    3. If the macro fails, give the command ":("Eml)sp" to find out what commands executed immediately before the failure. In particular look for a :map macro that was expanded shortly before the failure.
    4. Set a breakpoint on that macro with ":bre macrokey".
    5. Turn off logging, via ":se mlog=o".
    6. Reload the test file.
    7. Execute the macro again. When the macro with the breakpoint is encountered, Elvis will switch to single-step mode. Step slowly through the next few instructions, looking for one which does something unexpected.

    If your macro reveals a bug in Elvis, please let me know! My email address is kirkenda@cs.pdx.edu. Please tell me which version of Elvis you're using, as reported by the :version command.

    16.4 Running your compiler from within Elvis

    Elvis can parse most compilers' error messages. When it parses an error message, Elvis loads the faulty file, moves the cursor to the line where the error was reported, and shows the descriptive portion of the error message on the bottom row of the window. You can step through all reported errors very quickly, making changes along the way.

    Usually, you will invoke your compiler or "make" utility via the :cc or :make commands. The only difference between these commands is that :cc invokes the program named by the ccprg option, and :make uses the makeprg option.

    Both of those options' values are evaluated using the simpler expression syntax, with $1 set to any extra command-line parameters, and $2 set to the current file name.

    You can also read error messages from some other program with the command ":errlist !program", or read them from a file with the command ":errlist filename".

    I often invoke Elvis via the command "elvis +make" so Elvis will attempt to compile the program, and move the cursor to the first error (if there are any errors).

    All of the compiler's output text is collected into a buffer named "Elvis error list". If you wish, you can view this list in a separate window via this command:

    :(Elvis error list)split

    Here's how Elvis parses each line of compiler output: Starting from the left, it divides the line into "words", which are defined as a series of letters, digits, and/or certain punctuation characters.

    If the word is the name of an existing directory, then Elvis remembers that directory name. In later lines, Elvis will allow file names to be given relative to that directory, in addition to the current directory. This particular feature is intended to work with the directory lines generated by the GNU version of the "make" program.

    If the word looks like a number, and no line number has been seen yet, then the word is taken to be a line number. If the word is the name of an existing, writable file (or any existing file if the anyerror option is set) in either the current directory or the directory remembered from a previous line as described above, then the word is taken to be a file name. Other words are ignored.

    When Elvis has found both a file name and a line number, then it skips over any whitespace or punctuation immediately following them, and uses the remainder of the line as the error's description.

    If Elvis fails to find a file name/line number pair, then it skips that whole line of compiler output.

    Immediately after collecting compiler output, Elvis moves the cursor to the source of the first error. After that, you can use :errlist (with no arguments) or the visual * command to step through each following error.

    Each time Elvis collects a new set of error messages, it remembers how many lines are in each buffer. Later, when you insert or delete lines to correct an error, Elvis can compare the current number of lines to original number of lines, and adjust the reported line numbers accordingly.

    Here's something that may be useful for PERL programmers. PERL's error messages follow two distinct formats:

    description in file file at line line
    description at file line line
    Neither of these looks like any recognizable compiler error message format; consequently, Elvis can't parse PERL's error messages directly. But here's a way around that. The following PERL program is a filter that reformats PERL's error messages to look like normal compiler error messages.
    	#!/usr/bin/perl
    	$\ = "\n";
    	while (<>) {
    	    chop;
    	    s/(.*) in file ([^ ]*) at line (\d*).*/$2($3): $1/;
    	    s/(.*) at ([^ ]*) line (\d*)\.$/$2($3): $1/;
    	    print;
    	}

    To use this script, store it in a file named "perlerr" and turn on the file's "execute" permissions, and then set Elvis' ccprg option as follows:
    For CSH: :set ccprg="perl -c ($1?$1:$2) |& perlerr"
    Other shells: :set ccprg="perl -c ($1?$1:$2) 2>&1 | perlerr"

    NOTE: You can't simply cut&paste the above Perl script into a file, because it contains some HTML code which Perl wouldn't understand. (The diamond is written as "&lt;&gt;".) A better way is to visually select those lines via the shift-V command, and then use the :wascii alias to save the formatted text to a file as plain ASCII text. You'll still need to edit the text file to remove leading whitespace and possibly some blank lines, but that's pretty easy.

    16.5 Internationalization

    Elvis can be configured to translate its messages into different languages, and to use different symbol sets. These things are accomplished via the elvis.msg file and :digraph command, respectively.

    Elvis locates the elvis.msg file during initialization. Ordinarily it searches through each directory named in the ELVISPATH environment variable. However, if there is an environment variable named LC_ALL, LC_MESSAGES, or LANG (listed in order or precedence) which is set to a non-null value, then Elvis will look for elvis.msg first in a subdirectory whose name matches the environment variable's value. For example, if LC_ALL is unset, LC_MESSAGES=german, and ELVISPATH=~/.elvis:/usr/local/lib/elvis, then Elvis would try to load its messages from...

    1. ~/.elvis/german/elvis.msg
    2. ~/.elvis/elvis.msg
    3. /usr/local/lib/elvis/german/elvis.msg
    4. /usr/local/lib/elvis/elvis.msg

    The digraph table tells Elvis which pairs of ASCII characters can be combined to form a single non-ASCII character. This table is configured via the :digraph command. To enter a digraph, type <Ctrl-K> and then the two ASCII characters. Elvis will store the corresponding non-ASCII character instead of the two ASCII characters. See the Input Mode chapter for more information.

    The digraph table affects more than just keyboard input. It also affects "html" mode, and character type classifications.

    Digraphs are used by the "html" display mode to translate character entities into characters. For example, when Elvis encounters &ntilde; in an HTML document, it tries to find a digraph which combines 'n' with '~'. If there is such a digraph, Elvis will use it to display an 'ñ'; if not, then Elvis will display a plain 'n' character.

    The digraph table affects the character classes, too. This, in turn, affects the definition of a "word", as used by the visual w command, among others. A non-ascii character is treated as an uppercase letter if, according to the digraph table, it is the result of combining an ASCII uppercase letter with either a punctuation character or a second uppercase letter. A similar rule holds for lowercase letters.

    Also, Elvis tries to find uppercase/lowercase pairs through the digraph table. This is used for case conversions, as performed by the visual ~ command, or the \U metacharacter in the :s/old/new command.

    There is no way to specify a sorting order. This means, in particular, that the regular expression /[a-z]/ will only match the ASCII lowercase letters, not the non-ASCII ones. However, the regular expression /[[:lower:]]/ will match all lowercase letters including the non-ASCII ones.

    The default elvis.ini file tries to load digraphs by executing either elvis.pc8 for MS-DOS, OS/2, or text-mode Win32, or elvis.lat for any other operating system.

    The "win32" version of the "termcap" user interface has a codepage option which determines which symbol set is used for console output. If you change codepage, you should also adjust your digraph table.

    16.6 Aliases

    Aliases provide a simple way to add a new name for an existing ex command, or series of ex commands.

    The syntax of Elvis' :alias command is intended to resemble that of the csh Unix shell. The simplest example of an alias is...

    :alias save w
    
    ... which would allow you to write your file out by running ":save", as an alternative to the standard ":w". If you pass any arguments to ":save" then they'll be appended to the ":w" command. For example, ":save foo" would be interpreted as ":w foo".

    Here's another example. On Unix systems, this will make ":ls" display a directory listing.

    :alias ls !!ls -C

    Note that the above example requires two exclamation marks. This is because the "!" character is special in aliases -- Elvis' aliases allow you to use special symbols to indicate where arguments belong in the command text, and all of those symbols begin with a "!" character. When you invoke the alias, all of the symbols are replaced by argument values before the command text is executed. Here is a complete list of the replacement symbols:

         .--------.-------------------------------------------------.
         | SYMBOL | REPLACED BY                                     |
         |--------|-------------------------------------------------|
         |   !<   | first address line, if any addresses given      |
         |   !>   | last address line, if any addresses given       |
         |   !%   | address range, if any addresses given           |
         |   !?   | "!" if the alias is invoked with a "!" suffix   |
         |   !*   | the entire argument string except for "!" suffix|
         |   !^   | the first word from the argument string         |
         |   !$   | the last word from the argument string          |
         |   !$*  | all but the last word from the argument string  |
         |   !n   | the nth word (where n is a single digit)        |
         |   !n*  | everything from the nth word to the end         |
         |   !!   | a single, literal "!" character                 |
         | !name= | value from a name=value argument                |
         | !name& | URL-encoded version of !name=                   |
         ^--------^-------------------------------------------------^

    Using any of the !*, !^, !$, !name=, !name&, or !1 through !9 symbols in the command string has the side-effect of disabling the normal behavior of appending the arguments to the command. Or to phrase that another way: If the command text doesn't explicitly say what to do with arguments, then Elvis will assume it should simply append them.

    The other symbols, such as !% and !?, have no such default behavior. If your macro is going to use addresses or a "!" suffix, then you must explicitly include !% or !? (respectively) in the command string.

    Here's a simple alias for playing around with these:

    :alias show echo !!<=!< !!%=!% !!?=!? !!*=!* !!^=!^ !!2=!2 !!$=!$

    The !name= and !name& symbols are intended mostly for use in user defined protocols, where they offer an easy way to examine the values of URL form fields. In any other type of alias, they search through the alias's arguments for one of the form name=value, and return the value portion of that arguments. Note that "!=" (without a name) is not substituted, so you don't need to do anything special to use the != operator in expressions within aliases.

    Here's a more sophisticated version of the ":save" alias. This version allows you to use ":save!" as an alias for ":w!".

    :alias save w!?

    Here's a macro that converts a range of lines to uppercase. If invoked without any addresses, it will change only the current line, because that's the default for the :s command.

    :alias upper !%s/.*/\U&/

    You can define multi-line aliases by enclosing the lines in curly braces. The following example uses this technique to make a slightly smarter version of the "save" alias:

    :alias save {
     "Write a file, but only if it has been modified
     if modified
     then w!? !*
    }

    Note that the first line of the alias's body is a comment. (Comments start with a " character.) This is a good idea because when the :alias command is invoked with no arguments, it lists the names and first lines of all aliases. Putting a descriptive comment in the first line allows you to see what each alias does simply by examining that list.

    If a multi-line alias is going to use arguments, then it must include !*, !^, !$, or !n symbols. Elvis does not, by default, append arguments to the end of a multi-line alias; it only does that for single-line aliases.

    An alias can have the same name as a built-in command, but aliases can't be recursive. Together, these two rules mean that you can use an alias to change the behavior of a built-in command. For example, the following alias makes the :w command perform an RCS checkout operation if you don't already have write permission for a file. The "w" command inside the command text refers to the normal :write command since it isn't allowed to be a recursive call to the "w" alias.

    :alias w {
     "Write a file, checking it out first if necessary
     if readonly && "!%!?!*" == ""
     then !!co -l %
     then w!!
     else !%w!? !*
    }
    
    You can optionally insert a ':' character between the '!' and the second character of any of these symbols. This has no effect; it is allowed simply to remain a little closer to CSH's alias syntax.

    You can also optionally insert a '\' character between the '!' and the second character. This does have an effect: It causes a backslash to be inserted before any characters which would otherwise receive special treatment if they appeared in a regular expression. Specifically, it will always insert a backslash before '\', '/', '^', '$', '.', '[', or '*'. Note that this is not sensitive to the magic or magicchar options; in effect, it assumes that magic and magicchar are set to their default values. Also, it never inserts a backslash before a '?' character even if it is used in a regular expression which is delimited by '?' characters. The following "find" alias will search for literal text:

    :alias find /!\*
    
    In addition, you can optionally specify a default value for an argument, by placing the value in parentheses between the '!' and the second character. Here's an example which acts like echo, except that if you don't tell it what to echo then it will echo "Howdy!":
    :alias greet echo !(Howdy!)*
    
    If necessary, you can insert both a backslash and a parenthesized default value. The backslash quoting will be applied to the given argument value, but not the parenthesized default value. Here's a variation of the "find" alias which searches for literal text, or if you don't specify any text to find then it searches for a { character at the front of a line.
    :alias find /!(^{)\*
    
    Some vi commands are implemented via ex commands. If you create an alias with the same name as a built-in ex command, then the corresponding visual command will be affected. For example, the ZZ visual command runs :x, so the following alias would break the ZZ command...
    :alias x echo Winners never quit, and quitters never win
    

    16.6.1 Some example aliases

    The distribution comes with some handy aliases in a file named lib/elvis.ali. I suggest you look at them. The simple ones should give you some ideas of how to structure your own aliases, and the complex examples will give you a feel of what can be accomplished.

    Remember that the names of aliases must be spelled out in full; you can't abbreviate alias names the way can for built-ins. Also, you can display the definition of any alias by running ":alias aliasname".

    These examples are intended to be useful as well as instructive. They are loaded automatically when Elvis starts. The aliases in that file include:

    :lf [directory]
    List the contents of the current directory, or of a named directory. On Unix systems this works by invoking "ls -CF" on the arguments; on other systems it invokes "dir/w".
    :pwd
    Display the name of the current directory.
    :howto[!] word [word2]
    Load the "How To" appendix in a separate window, and search for a topic containing the given word or words. The words should be typed in lowercase. If you want to search all lines (not just topic lines) then run :howto! (with a ! suffix).
    :kwic word
    Build a table showing all occurrences of word in the online manual. This macro depends on the "grep" program to do the actual searching; if your system lacks "grep" then this won't work correctly. (Windows users should consider using the CygWin tools.)
    :man [section] topic
    Unix only. Read a man-page and display it in a new window.
    :save [filename]
    Write this file, but only if it has been modified.
    :[range]ww[!] [filename]
    Like the normal :w command, except that if you try to write a whole file back over itself, and that file is readonly, then the :ww alias will attempt to perform an RCS "checkout" operation on that file by running "!co -l filename".
    :courier [size] ["bold"]
    :luxi [size] ["bold"]
    :andale [size]
    X11 only. Select the Courier or Luxi mono fonts of the given size. If no size is specified, then it uses the default size, which is 14-point.
    :fork program
    X11 only. Start running program in parallel with Elvis, with its stdio disconnected from Elvis. This is necessary because if you just ran ":!program &" Elvis would wait for the program to exit, so it could display the program's output in the Elvis window.
    :copying
    Display the license.
    :features
    Report the configuration of your copy of Elvis. I.e., list the features which were enabled when Elvis was compiled.
    :customize filename
    Create a personal copy of one of Elvis' configuration files (unless you already have a personal copy of it), and start editing it. A typical example would be ":customize elvis.syn" to edit the syntax coloring rules.
    :[range]left
    :[range]right
    :[range]center
    For each line in the given range (or only the current line if no addresses are given), adjust the indentation so that its text is moved to the left, right, or center of the line.
    :[range]wascii[!] file
    Write WYSIWYG formatted text out to a file as plain ASCII text. This is intended to make it easy to save ex scripts (or other source code) that are embedded within HTML pages such as this manual. It works by temporarily setting lptype to "dumb", and then writing the text via :lp.
    :[range]lpd[!] file
    Print text, formatting it in the current display mode, regardless of the buffer's usual display mode. It works by temporarily setting the buffer's bufdisplay option to the value of the window's display option.
    :[range]cfmt
    Adjust the line breaks in a C or C++ comment block.
    :wrapset [label] optionname
    Display the value of an option, inserting spaces after each comma and wrapping it to fit the width of the screen. The optional label may contain multiple words, and causes the option's value to be neatly indented.
    :text words
    :btext words
    Search through all the *.c or *.h files in the current directory for the given words, and treat the found locations as a series of matching tags. The :text alias moves the cursor to the first location; you can use an argumentless :tag command to move' the each successive location. The :btext alias browses the matches, similar to the :browse command.
    :load [name]
    List the installed scripts, or run the script with a given name. The scripts reside in a "scripts/" subdirectory inside any of the directories listed in elvispath. Elvis comes with a variety of scripts to configure Elvis for various purposes. In most cases, if you run :mkexrc after running a script, then the effects of the script become permanent.
    :theme [name]
    List all installed themes, or load the theme with a given name. All of the themes use background images, and only the "x11" and "windows" interfaces currently support images, so this alias is only defined if you're using one of those.

    Themes are basically the same as :load scripts. They just reside in a different subdirectory, and are intended to configure colors instead of aliases and options.

    :mktheme name
    For "x11" and "windows" only. Create a theme by bundling up the current color configuration and the images it uses. You only need to do this if you want to share the theme; the :mkexrc command is perfectly capable of storing your color settings for your own use.

    You should give this command in an empty edit buffer, with Elvis configured the way you want it to look, and any background images located in the current directory. After you run this alias, the edit buffer will be filled with an ex script and the images. You'll want to edit it to add a description, add your name to it, and maybe tweak the settings. Then save it to a file, and send file to me at kirkenda@cs.pdx.edu.

    The theme should set the foreground & background for "normal", "idle", and the widgets and cursor. It may also set the background only of highlight faces such as "selection", "hlsearch", and "spell". Other faces, such as "comment", should not be set in the theme; instead, users are responsible for setting up their own favorite colors using the :color command's "or colorname" clauses to supply a variety of colors to choose from.

    readTHEME
    For "x11" and "windows" only. This alias isn't intended to be run by the user directly. Instead, it implements the "theme:" network protocol, using the techniques described in the next section. You can run ":e theme:/" to see a web page of downloadable themes.

    16.6.2 Using aliases to add network protocols

    Elvis has built-in support for URLs that use the "ftp:" and "http:" protocols, as well as the "file:" and "buffer:" pseudo-protocols that access local files or buffers.

    Most Elvis versions also allow you to add support for new protocols by defining aliases that handle reading and writing via that protocol. (Expressions can use the feature("proto") function to test for this feature.) The aliases must have the names "readXXX" and "writeXXX", where "XXX" is the uppercase name of the protocol.

    The following example supports the "mailto:" protocol, on Unix/Linux systems:

    	alias readMAILTO {
    	  "initialize a mailto: message
    	  se noro reol=text
    	  if exists($HOME/".signature")
    	    read ~/.signature
    	    1 i --
    	  }
    	}
    	alias writeMAILTO {
    	  "send a mailto: message
    	  w !!mail -s"!(no subject)subject=" !2 >/dev/null 2>&1
    	  se nomod
    	}

    After loading these aliases, the command ":e mailto:user@system.com" will create an empty buffer named "mailto:user@system.com", and then run the readMAILTO alias with the cursor located in that buffer.

    The first thing that readMAILTO does is turn off the readonly option, and set the readeol option to "text". Elvis normally sets these to readonly and reol=binary because that's safest when downloading files via FTP or HTTP. Since the whole point of the "mailto:" protocol is to write text, though, those settings are inappropriate, so readMAILTO changes them.

    Other than that, readMAILTO merely needs to insert a copy of the ~/.signature file, if one exists.

    You can then edit the message however you like, interactively. When you want to send the message, give the :w command. Because the buffer's name (actually the filename option) starts with "mailto:", Elvis knows it should use the writeMAILTO alias to write it.

    The writeMAILTO alias pipes the message out to the standard Unix "mail" program. This is a little trickier than it should be, because "mail" can fork off a spooler process which can run for several minutes. This could make Elvis' x11 user interface hang, because it waits for stdout/stderr to close before resuming its normal activity, and the spooler process inherits the stdout/stderr from the mail process. The solution is to redirect the mail process' stdout/stderr to /dev/null.

    After that, it turns off the buffer's modified option, so that other commands will know it is safe to delete that buffer.

    Both the readXXX and writeXXX aliases are invoked with arguments derived from the URL. The arguments contain pieces of the URL, as follows:

    .--------.-------------------------------------------------------.
    | SYMBOL | REPLACEMENT TEXT                                      |
    |--------|-------------------------------------------------------|
    |   !*   | Whole URL - protocol://sitename/resource#anchor?args  |
    |   !1   | Site name, if the URL contained one.                  |
    |   !2   | Resource name                                         |
    |   !3   | The anchor name, if the URL contained one.            |
    | !4 - !9| Form fields, if any (name=value, URL-decoded)         |
    | !name= | Value of the form field named name, URL-decoded       |
    | !name& | Value of the form field named name, URL-encoded       |
    ^--------^-------------------------------------------------------^
    
    Example:                                                  !term&
                                                             .------.
      http://www.techweb.com/encyclopedia/defineterm.yb?term=firewall
             ^------!1-----^^----------!2-------------^ ^-----!4----^

    In the "mailto:" example above, writeMAILTO used !2 as the email address of the person being mailed to. It also used the subject field value, if one was supplied, but instead of trying to locate it by searching through !4 - !9, it searched by name using the !name= notation. Specifically, it used "!(no subject)subject=" to find the value of the subject field, or use "no subject" by default.

    The difference between the !name= and !name& notations is that !name& is replaced with URL-encoding intact, while !name= is replaced with the plain-text form of the parameter. So !name& works better for plugging a parameter into a newly constructed URL, while !name= works better for most other purposes.

    The userprotocol option is turned on for buffers loaded via a readPROTOCOL alias. The default elvis.arf script is sensitive to this; it skips certain setup steps, since the default setup steps might not be appropriate for a user-defined protocol. If you don't like this behavior, then simply add a "set nouserprotocol" command to your readPROTOCOL alias.

    16.7 The spell-checker

    Elvis can be compiled with a built-in spell checker. To find out if your version of Elvis includes the spell-check feature, run the following:
    	:calc feature("spell")

    If it says True then you're all set. You should be able to enable spell-checking via the command:

    	:set spell

    After that, any suspected misspellings will be highlighted. You can change the highlighting style via ":color spell ..."

    16.7.1 Dictionaries

    Elvis has two spelling dictionaries: One loaded from "tags" files, which is useful mostly for detecting misspelled function names while you're editing source code; and one for natural-language words, which can be loaded from a large sorted file, or configured via the :words command.

    Elvis is intended primarily for editing the source code of programs. In source code, some words should be checked only against the "tags" dictionary (function names and data types), some should be checked against both dictionaries (comments, mostly), and some should be unchecked (variables, since they often are not listed in the "tags" file). Elvis' spell checker can do that!

    The spell-checking rules are oriented around faces. Faces are also used to control the font & color attributes of text, as configured via the :color command. Each face has a name, such as "comment" or "function". The spell checker doesn't care about the appearance of a face, but it cares a great deal about its name.

    For each face, Elvis allows you to specify one of three levels of spell-checking:

              .------.-----------------------------------------.
              | FLAG | TYPE OF SPELL-CHECKING PERFORMED        |
              |------|-----------------------------------------|
              |  *   | Check against both dictionaries         |
              |  +   | Check against only the tags dictionary  |
              |  -   | Don't perform any spell checking        |
              ^------^-----------------------------------------^

    By default, most faces check against both dictionaries. The elvis.spe initialization file configures some faces to check against only the tags dictionary, or skip checking entirely.

    If the defaults don't meet your needs, you can use the :check command to change the rules for any face. For example, the "variable" face isn't normally subjected to any type of spell checking. It would be nice if this could be checked against the tags dictionary, but most tags generators don't generate tags for all variables, so many valid variable names would be marked as misspelled. However, if you have a tags generator that does a thorough job on variables then you might want to enable spell-checking of variables against the tags dictionary. You probably wouldn't want to check variable names against the natural-language dictionary, though. The following command configures variable checking this way:

    	:check variable

    Similarly, the "string" face isn't normally checked because strings tend to contain non-human text such as "\n" or "%d". If you want to enable checking against both dictionaries, you can use the following command:

    	:check *string

    The "*" causes Elvis to use both dictionaries. In the earlier "variable" example, we could also have used a "+" flag to select just the tags dictionary, but that's the default so we omitted it. The "-" flag disables spell-checking:

    	:check -variable string

    By the way, the :check command supports name completion, so you can use the <Tab> key to see what faces are available.

    16.7.2 Configuring the tags dictionary

    If any buffer has its spell option turned on, then Elvis will automatically load the tags dictionary from any "tags" files that it finds. (For a discussion of tags, see the Tags chapter, and the ctags man-page.)

    When loading tags, Elvis searches through all of the file names in the tags and elvispath options. If any of those names is a directory, then Elvis looks for a file named "tags" in that directory.

    Elvis loads every tag from every file that it finds this way. All of these tag names are combined in RAM to form a single tags directory. Other tag information is ignored; Elvis only stores the names.

    As long as you generate tags files in the usual way, Elvis should be able to find them without any trouble.

    To make it even easier, if the "tags" file in the current directory is altered then Elvis will reload the tags dictionary automatically. (It doesn't check the "tags" files in other directories, though.)

    All words in the tags dictionary are considered to be case-sensitive.

    16.7.3 Configuring the natural-language dictionary

    There are three ways to load words into the natural-language dictionary: the spelldict option, and the :words and :wordfile commands.

    The way Elvis uses this dictionary is a bit tricky. The first thing you must realize is that the dictionary is maintained in RAM. It can pull words from the spelldict file as needed, but that file isn't the dictionary -- the in-memory data structure is.

    The next big quirk is that the dictionary contains both correct and misspelled words. Elvis must remember misspelled words so that it can recognize them again quickly, without having to look them up in the spelldict file every time. This is very important because Elvis spell-checks every word on the screen, after every keystroke. Elvis can check any word against its in-memory dictionary in a fraction of a microsecond, but file searches are far slower.

    You can use the :words command to add words to the natural-language dictionary. If you precede the words with a "-" then they'll be flagged as being incorrect. Any time Elvis displays that word on the screen, it'll be shown highlighted as an error. Words added without a "-" flag will be added as correct words.

    If Elvis looks for a word in its dictionary, and doesn't find it, then it will try to find the word in the spelldict file. If Elvis finds it, great! Elvis will add it to its dictionary as a correct word. Otherwise, Elvis will will try suffix substitution, as specified via the spellsuffix option, and check the resulting word against the spelldict file again. If Elvis finds that word, then it will add both the found word and the original word, both flagged as correct words. Otherwise, Elvis will add the original word, flagged as incorrect.

    Words pulled from the spelldict file are always considered to be case-insensitive, even if they're capitalized in that file. Words added via :wordfile are also case-insensitive. However, words added via the :words command are considered to be case-insensitive only when they're given in lowercase; if the a word contains any uppercase letters then it will be case-sensitive.

    Words added via :word or :wordfile are flagged as being personal, unless you append a "!" suffix to the command name. The distinction between personal words and other natural-language words is important, because :mkexrc will store any personal words in your .exrc file.

    16.7.4 Correcting errors

    Elvis has no built-in command for finding and correcting all of the spelling errors in a file. For that, I recommend an external program such as ispell.

    However, Elvis does have an spell option for highlighting misspelled words, and the show option supports a "spell" keyword which displays suggested corrections at the bottom of the screen. It also has a gs command for moving to the next misspelled word. If the cursor is already on a misspelled word before gs, and the show option is presenting a list of suggestions, then you can invoke gs with a "count" argument to fix the current word before moving to the next word. If none of the suggested alternative spellings are correct, you can use c gS (with an uppercase "S") to replace the word manually.

    Elvis' built in calculator has functions named spell() and spelltag(), which offer a way to access the spelling suggestions in your own aliases or scripts.

    The spell() function only checks the in-memory dictionary; it doesn't scan the spelldict file. Because of this, you'll probably want to load all words from that file into memory. There are two ways to do that: either invoke the :wordfile command explicitly, or set the spellautoload option to have it be loaded automatically the first time you need it.

    16.8 Auto commands

    16.8.1 Overview of Auto Commands

    Auto commands give you a way to cause commands to be executed automatically whenever a certain event occurs. This is sort of like hardware interrupts. Many of the events resemble GUI events.

    Elvis maintains a list of auto commands. Each auto command contains:

    The whole auto command feature is omitted from the MS-DOS version of Elvis, for the sake of compactness. Under MS-DOS, Elvis will continue to depend on elvis.arf and the related files.

    16.8.2 Groups

    For ease of editing, auto commands may be divided into groups. Each group has a name. The default group is named "END", but other groups have no preset names; you can choose your own names.

    To select a group for editing (and to create the group if it didn't exist before), use the :augroup command. Typically, you'll begin by using ":aug groupname" to select a group to edit, then wipe out any existing auto commands in that group and define new ones, and then use ":aug END" to mark the end of the group.

    	:aug buflist
    	:au!
    	:au BufNewFile,BufReadPost * morebuffers
    	:au BufDelete * lessbuffers
    	:aug END

    Note that the :augroup command only affects the behavior of later :autocmd commands. Afterward, when events occur, they will trigger all appropriate auto commands regardless of the group name. I.e., groups are useful for editing the auto command list, but have no effect on its actions.

    Well, actually there is one exception: The :doau command can optionally be told a specific group name to run. If you don't specify the group name, then it will trigger all appropriate events in all groups, just like a real (non-:doau) event.

    Elvis' version of :aug has a minor extension beyond vim's capabilities: when invoked with ! suffix (":aug! groupname") it marks that group as being a "system" group. Elvis' standard configuration scripts may use this notation to mark its own groups. The significance of this is, "system" groups aren't saved by the :mkexrc command -- they don't need to be stored in your custom initialization file since they're created by a standard script.

    16.8.3 Events

    Event names are case-insensitive, so "BufCreate" can be written as "bufcreate", or even "BuFcReAtE".

    The :doau command only accepts a single event name -- the one that it is supposed to simulate. However, the :au command accepts a comma-delimited list of event names (without any whitespace between them), or * to represent all events.

    When deleting/replacing events, if an auto command has been created with multiple events and you only delete/replace some of them, then the remaining ones will still be in effect. In other words, Elvis will remove the given events from each auto command's set of matching events, and will only completely delete an auto command if the remaining set of events is empty. (When deleting/replacing, it only effects auto commands where at least one of the new patterns is identical to one of the existing patterns.)

    Here's a list of built-in events, and what they mean in Elvis. Note that even "Not implemented" events can be triggered explicitly by :doau, so they aren't totally useless. You can also create new events with the :auevent command.

    AliasEnter (not in vim)
    At the start of an alias. The pattern is compared to the name of the alias, not a file, but you can still use the "*" wildcard. One common use for this is to set local options. For example, if you like to use Perl-ish regular expressions for interactive commands but still want to use traditional regular expression syntax inside aliases, you could use the following to temporarily reset the magicchar and magicperl options while running aliases:
    	:au AliasEnter * local magicchar=^$.*[ nomagicperl
    AliasLeave (not in vim)
    At the end of an alias. The opposite of AliasEnter, and not nearly as useful. Maybe good for tracing execution?
    BgChanged (not in vim)
    The normal background color is changed. The :color command is used to change the normal background color. The pattern is compared to "light" or "dark", depending on the chosen color. You can use this to adjust other backgrounds:
    	:au BgChanged light color x11.spell on pink
    	:au BgChanged dark  color x11.spell on red
    BufCreate
    A user buffer is created. A user buffer is one in which the internal option is off. All buffers that contain your text files are user buffers.

    Vim also sends this event after a buffer is renamed, but Elvis doesn't normally. If you want to make Elvis perform BufCreate when a buffer is renamed, then make the BufFilePost event run
    ":doau BufCreate", like this:

    	:au BufFilePost * doau BufCreate
    BufDelete
    A user buffer is destroyed. Note that Elvis has no explicit command for freeing buffers. It chooses when to free them automatically, if they aren't being used and haven't been modified. You can prevent Elvis from destroying a buffer by settings its retain option.

    You could use this event with BufCreate to maintain a list of open buffers. The data/scripts/xbuf.ex file, distributed with Elvis, contains an example of how this might be done in Elvis' x11 user interface.

    BufEnter
    A buffer becomes the "current" buffer. Specifically, this event is triggered when the buffer-specific options are swapped in, to be accessible via commands such as :set, or in expressions. See also the BufLeave event.

    Warning: Elvis can switch buffers extremely often. For example, if you're using the "x11" user interface with multiple windows and toolbars, then Elvis will make each window's buffer be the "current" buffer while it is evaluating the states of each toolbar button, after each keystroke.

    BufFilePost
    After changing a buffer's name. A buffer's name can be changed two ways: explicitly via the :file command, or implicitly by saving an anonymous buffer out to a file.
    BufFilePre
    Before changing a buffer's name. See also the BufFilePost event.
    BufHidden
    Not implemented.
    BufLeave
    A buffer will no longer be the "current" buffer. This is the other half of BufEnter event. When Elvis is switching buffers, it triggers a BufLeave event (with the old buffer) before switching, and a BufEnter event (with the new buffer) afterward.
    BufNewFile
    A buffer is created for a file that doesn't exist yet. This is similar to BufReadPost, except that BufNewFile is triggered only when there is no file to be read. You can use this to create a skeleton version of a file.
    BufReadPost
    After loading a file's contents. See also BufReadPre for before it has loaded, FileReadPost for inserting text from another file into an existing file's buffer, and FilterReadPost for inserting from a filter.
    BufReadPre
    Before loading a file's contents. This is useful for setting the readeol option. See also BufReadPost for after it has loaded, FileReadPre for inserting text from another file into an existing file's buffer, and FilterReadPre for inserting from a filter.
    BufRead
    Synonym for BufReadPost
    BufUnload
    Immediately before BufDelete events, for Elvis. Vim has an extra level of decrepitude for buffers, where the text has been discarded but the name and options have not yet been forgotten; vim performs BufUnload when a buffer enters this state, and BufDelete when it discards the rest of the buffer. Elvis doesn't have that intermediate step, so BufUnload and BufDelete are effectively the same thing.
    BufWritePost
    After a buffer has been saved to its file.
    BufWritePre
    Before a buffer will be saved to its file. If any auto command triggered by this event causes an error and the eventerrors option is set, then the text will not be written.
    BufWrite
    Synonym for BufWritePre.
    CursorHold
    Not implemented.
    DisplayEnter (not in vim)
    A window's display mode has changed. This is useful for changing options in a display-mode dependent way. The pattern is compared to the name of the new display mode, not the file name. See DisplayLeave for a way to change them back afterward, or DispMapEnter for a way to adjust the map table.
    DisplayLeave (not in vim)
    A window's display mode is about to change. This is useful for reverting options that were changed via a DisplayEnter event. The pattern is compared to the name of the new display mode, not the file name.
    DispMapEnter (not in vim)
    Keystrokes are received in a different display mode. This is useful for installing maps that are only intended to be used in specific display modes. The pattern is compared against the name of the new display mode, not the file name. Note that this event can be triggered simply by switching windows, if the two windows are using different display modes.

    Generally, Elvis will trigger a DispMapLeave event on the old mode before triggering the DispMapEnter event on the new mode. Both of these events will be triggered before the newly received keystrokes are subjected to maps.

    DispMapLeave (not in vim)
    Keystrokes are received after an old display mode is abandoned. This is useful for removing display-sensitive maps that were installed via DispMapEnter. The pattern is compared against the name of the old display mode, not the file name. Note that this event can be triggered simply by switching windows, if the two windows are using different display modes.
    Edit (Not in vim)
    Text is changed. While the event's auto commands are running, the '[ and '] marks indicate the changed region.

    The Edit event is triggered every time the buffer is changed. If you want to trigger an event for only the first change to a buffer, then you can use the OptSet event to watch for changes to the modified option. (However, OptSet doesn't set the '[ and '] marks.)

    FileAppendPost
    After appending all or part of the buffer onto a file.
    FileAppendPre
    Before appending all or part of the buffer onto a file. While the event's auto commands are running, the '[ and '] marks indicate the region to be written, and the aufilename option indicates the name of the file that it'll be written to. If any auto command triggered by this event causes an error and the eventerrors option is set, then the text will not be written.
    FileChangedShell
    Not implemented.
    FileEncoding
    Not implemented.
    FileReadPost
    After inserting from a file. This is different from BufReadPost, which loads a file into a buffer. FileReadPost is triggered by the :r command, but not the :e command. The auto command can use '[ and '] marks to determine which lines are new. See also FileReadPre for before the text has been inserted, and FilterReadPost for reading from a filter program.
    FileReadPre
    Before inserting from a file. This is different from BufReadPre, which loads a file into a buffer. FileReadPre is triggered by the :r command, but not the :e command. See also FileReadPost for after the text has been inserted, and FilterReadPre for reading from a filter program.
    FileType
    Not implemented. Elvis doesn't have a "filetype" option, though I may add one later. The mapmode and bufdisplay options serve a similar purpose.
    FileWritePost
    After writing all or part of the buffer out to a different file. While the event's auto commands are running, the aufilename option indicates the name of the file that was written.
    FileWritePre
    Before writing all or part of the buffer out to a different file. While the event's auto commands are running, the '[ and '] marks indicate the region to be written, and the aufilename option indicates the name of the file that it'll be written to. If any auto command triggered by this event causes an error and the eventerrors option is set, then the text will not be written.
    FilterReadPost
    After reading from a filter program. The auto command can use '[ and '] marks to determine which lines are new. See also FilterReadPre for before the text has been inserted, and FileReadPost for reading from a file.
    FilterReadPre
    Before reading from a filter program. See also FilterReadPost for after the text has been inserted, and FileReadPre for reading from a file.
    FilterWritePost
    After writing all or part of the buffer out to a filter program. While the event's auto commands are running, the aufilename option sores a copy of the name stores a copy of the command that it is writing to.
    FilterWritePre
    Before writing all or part of the buffer out to a filter program. While the event's auto commands are running, the '[ and '] marks indicate the region to be written, and the aufilename option stores a copy of the command that it is writing to. If any auto command triggered by this event causes an error and the eventerrors option is set, then the text will not be written.
    FocusGained
    Not implemented.
    FocusLost
    Not implemented.
    GUIEnter
    Not implemented.
    OptChanged (not in vim)
    A particular option's value is changed. The pattern for this event is a list of option names, not file name wildcards. While the auto command is running, the aufilename option contains the name of the actual option that was changed.

    NOTE: Options changed via the :set, :let, and :local commands always trigger this event for any changed options, but some options are changed internally by Elvis as a normal consequence of operation, and those changes don't always trigger the OptChanged event. For example, the bufchars option changes any time you insert or delete text, but this doesn't trigger OptChanged.

    OptSet (not in vim)
    A particular Boolean option is set or reset. The pattern for this event is a list of option names, not file name wildcards. To detect when a Boolean option is set (becomes true), use the name without any prefix; to detect when it is reset (becomes false), use the name with a "no" prefix. While the auto command is running, the aufilename option contains the name of the actual option that was changed, with a "no" prefix if it just became false. The following example causes the novice option to change a few other options:
    	:au OptSet novice set nomagic report=1 showmode

    NOTE: As with OptChanged, Elvis changes some options internally, and those changes don't always trigger an OptSet event.

    ScriptEnter (not in vim)
    At the start of a script. The pattern is compared to the name of the script file. One common use for this event is for setting local options. For example, if you like to use Perl-ish regular expressions for interactive commands but still want to use traditional regular expression syntax inside scripts, you could use the following to temporarily reset the magicchar and magicperl options while running scripts:
    	:au ScriptEnter * local magicchar=^$.*[ nomagicperl

    Note that ScriptEnter and ScriptLeave are executed both for scripts that are explicitly run via the :so command, and the configuration scripts that are run automatically: elvis.brf, elvis.arf, elvis.bwf, and elvis.awf.

    ScriptLeave (not in vim)
    At the end of a script. The opposite of ScriptEnter, and not nearly as useful. Maybe good for tracing execution?
    StdinReadPost
    Not implemented.
    StdinReadPre
    Not implemented.
    Syntax
    Not implemented. Elvis has no "syntax" option. The bufdisplay option serves a similar purpose, but isn't quite similar enough to justify triggering a Syntax event. You could have Elvis simulate it via the OptChanged event:
    	:au OptChanged bufdisplay doau Syntax
    TermChanged
    Not implemented. Elvis doesn't allow the term option to be changed after initialization is complete.
    User
    Never. The User event can only be triggered explicitly via the :doau command.
    VimEnter
    Not implemented.
    VimLeavePre
    Not implemented.
    VimLeave
    Not implemented.
    WinEnter
    A window becomes the "current" window. Specifically, this is triggered when the window-specific options for a given window are swapped in so they're accessible via commands such as :set, and in expressions. See also BufEnter and WinLeave.

    Warning: Elvis can switch windows extremely often. For example, if you're using the "x11" user interface with multiple windows and toolbars, then Elvis will make each window be the "current" window while it is evaluating the states of each toolbar's buttons, after each keystroke.

    WinLeave
    A window will no longer be the "current" window. When switching from one window to another, Elvis will first trigger a WinLeave event (with the old window), then switch windows, and then trigger a WinEnter event (with the new window). See also BufLeave.

    16.8.4 Patterns

    Patterns are used mostly to select file names. They use the same "wildcard" characters that are used for matching filenames on the command line. Different operating systems have slightly different rules. Specifically...

    The contents of [] wildcards is always interpreted in a case-sensitive manner, even in operating systems such as Windows which are normally case-insensitive with respect to filenames. If you want a [] expression to be case-insensitive, then you must explicitly give both the uppercase and lowercase versions of the letter.

    The chars list may be individual characters, or ranges of characters with a '-' between them. For example, [_a-zA-Z] matches an underscore or a letter.

    Although the syntax for [] wildcards is similar to the syntax for the [] metacharacter in regular expressions, the two are not the same. For example, you cannot use [[:alpha:]] to match any locale-dependent letter in a filename wildcard, though you can in regular expressions.

    16.9 How to make Elvis run faster

    This section describes some ways you can "tune" Elvis to run faster. None of these suggestions require recompiling Elvis.

    For example, my computer (AMD K6-200) can run 10 generations of the "life" macros in 41 seconds with the default configuration. If I invoke Elvis with a reduced block size (-b1024 on the command line) and an increased cache (:set blkcache=200 blkhash=300), it can run 10 generations in just 24 seconds.

    The blksize option

    Elvis uses fixed-size blocks to store buffers. The block size is chosen when the session file is created. The default is 2048 bytes (or 1024 bytes for MS-DOS), but you can override that by invoking Elvis with a -bblksize flag. The size must be a power of two, between 512 and 8192.

    The blksize option indicates the current block size. This is a read-only option; once Elvis has started, (and hence has already created the session file) it is too late to request a different block size.

    Generally, smaller blocks are better if your CPU is slow or you're only editing small files. Larger blocks are better if your disk is slow (e.g., the temporary file is stored on a remote disk, accessed via a network) or you're editing large files.

    The block cache

    In the interest of speed, Elvis has its own cache of blocks from the session file. The blkcache option tells Elvis how many blocks to store in the cache. You can change this value at any time. If Elvis requires more cached blocks for a given editing operation than the blkcache allows, then Elvis ignores blkcache and loads the required blocks into the expanded cache; you can't make blkcache too small. The default blkcache is 20 (except for MS-DOS, where it is 10), and the upper bound is 200 blocks. In MS-DOS, setting blkcache too high can cause Elvis to crash.

    The blkhit and blkmiss options count the number of cache hits and misses, so you can compute the efficiency of the cache as follows:

    	:calc (bh*100/(bh+bm))"%"

    You'll probably find that 98% or more of the block requests are being satisfied from the cache. However, since each miss takes thousands of times longer to complete than a hit, each fraction of a percent means a lot.

    In addition to the blocks themselves, the cache contains a hash table which allows Elvis to quickly determine whether a block is in the cache or not. If you increase the size of the cache, then you'll probably want to increase the size of the table as well. The table size is controlled by the blkhash option. Ideally, it should be set to a prime number somewhat larger than blkcache... or simply the largest possible value, since hash table entries are small. The default is 61, and the upper bound is 300.

    Syncing

    Elvis has a sync option which, if set, causes Elvis to force all dirty blocks out to disk occasionally. This is a very slow operation, so the sync option is normally turned off. If Elvis seems to be running exceptionally slowly, then you might want to verify that sync is off. You can check it by giving the command ":set sync?".

    Temporary files

    The session file, and any other temporary files, should be stored on a local hard disk. Storing them on a network drive will slow Elvis down a lot.

    The session option indicates where the current session file is stored, and the sessionpath option indicates which directories Elvis looked through when deciding where to put the session file. These are read-only options, since it is too late to choose a new location for the session file after the session file is already created. If you need to force Elvis to store its session files in a different directory, you should set the SESSIONPATH environment variable to a list of acceptable directories. Elvis will use the first directory from that list which exists and is writable.

    The directory option tells Elvis where to store other temporary files, such as those used for piping text through external programs. You can change its value at any time. (Note: the real vi also has a directory option, but only allows you to change it during initialization.)

    Screen updates

    Options which add information to the bottom row of the window, such as ruler and showmode, can slow down screen updates. If speed is critical, you should turn those options off.

    The optimize option attempts to eliminate superfluous screen updates while a macro is executing. It is normally on, but you may want to verify that. Some animation macros force it off.

    Elvis also has an animation option which, if optimize is on, causes Elvis to bypass some of the animation steps. The default value is 3. If animations look choppy then try reducing it to 1. Or set it to 10 or so for faster animation.

    The exrefresh option controls whether Elvis should refresh the screen after each line output by an ex command, or wait until the command completes and then update the screen once. It is normally off, so the screen is only updated once.

    Interestingly, the syntax-coloring display mode seems to run about as fast as the normal display mode. One possible exception would be if you're running Elvis over a slow modem connection then the extra escape sequences required for sending color information may slow down screen updates.

    If you have long lines, then Elvis may run somewhat faster when the wrap option is set. This is because Elvis always formats entire lines, even if only part of the line is visible on the screen, and the "nowrap" display style shows more long lines (one per row) than the "wrap" display style.

    Input

    The pollfrequency option indicates how often Elvis checks for an abort request from the user. (Abort requests are usually given by typing ^C or by clicking on a window while a macro or other time-consuming command is running). Smaller numbers give a quicker response to an abort request, but will cause the command itself to run slower.

    The keytime option indicates how long Elvis should wait after receiving an Esc character to distinguish between the <Esc> key, and some other function key which begins with an Esc. Longer times are more reliable, especially when you're running over a network. But shorter times allow a quicker response to the <Esc> key.

    MS-DOS-specific tips

    The single biggest improvement you can make is to switch from the 16-bit MS-DOS version to the 32-bit text-mode Win32 version. It only runs under Windows95 or WindowsNT, not MS-DOS, but you can make it run in full-screen mode which feels like MS-DOS. And it is *much* faster, because I really had to mangle the MS-DOS version of Elvis to make it fit in the lower 640k.

    The fancier ANSI drivers such as NANSI.SYS also help. They allow Elvis to redraw the screen very quickly. The URLs section of this chapter tells you where you can find NANSI.SYS on the Internet.

    Installing smartdrv.exe can be a big help. Storing temporary files on a RAM disk (in extended/expanded memory please!) can also help.

    Elvis also has a compile-time option, controlled by the FEATURE_RAM declaration in config.h, which allows Elvis to store its buffers in EMS/XMS memory instead of a file. This makes Elvis run much faster, but it has some disadvantages. If Elvis crashes, there will be no way to recover the contents of the edit buffers. Also the Microsoft functions for accessing EMS/XMS memory are very bulky, and also require a fairly large buffer; so if you enable FEATURE_RAM, then you must disable most of the other FEATURE_XXXX features in config.h, and even then Elvis may run out of memory occasionally. If you enable FEATURE_RAM, then to actually use that feature, you must invoke Elvis with "-f ram" on the command line.

    If you often use "-b 2048" or "-f ram", then you might consider setting up a batch file which runs them. For example, you could create a elvis.bat file containing...

    	elvis.exe -b 2048 -f ram %1 %2 %3 %4 %5 %6 %7 %8 %9