vi reference

Table of Contents


Work in progress from my own HTML version.

This reference was originally developed by Maarten Litmaath [email protected], and has been contributed to by many others (see list at the end of document).

Warning: some vi versions don't support the more esoteric features described in this document. You can edit/redistribute this document freely, as long as you don't make false claims on original authorship.


Original Author: Maarten Limaath

Maarten Litmaath [email protected]
Rich Salz [email protected]
Eamonn McManus [email protected]
Diomidis Spinellis diomidis%[email protected]
Blair P. Houghton [email protected]
Rusty Haddock <{uunet,att,rutgers}!!fe2o3!rusty>
Panos Tsirigotis [email protected]
David J. MacKenzie [email protected]
Kevin Carothers [email protected]
Dan Mercer [email protected]
Miles Davis [email protected]
Patricia Bender
Ze-ev Shtadler
Paul Quare
Dave Deyerl
Lee Sailer
David Gast


default values1
<*>'*' must not be taken literally.
[*]'*' is optional.
<cr>carriage return
<ht>horizontal tab
<erase>your erase character.
<kill>your kill character.
<intr>your interrupt character.
<a-z>an element in the range [a...z].
Nnumber ('*' = allowed, '-' = not appropriate).
CHAR`char unequal to
WORD`word followed by

move commands

*$To the end of line <*> from the cursor.
-^To the first CHAR of the line.
*_To the first CHAR <*> - 1 lines lower.
*-To the first CHAR <*> lines higher.
-0To the first char of the line.
*f<char><*> <char>s to the right (find).
*t<char>Till before <*> <char>s to the right.
*F<char><*> <char>s to the left.
*T<char>Till after <*> <char>s to the left.
*;`Repeat latest ‘f’
*,Idem in opposite direction.
*w<*> words forward.
*W<*> WORDS forward.
*b<*> words backward.
*B<*> WORDS backward.
*eTo the end of word <*> forward.
*ETo the end of WORD <*> forward.
*GGo to line <*> (default EOF).
*HTo line <*> from top of the screen (home).
*LTo line <*> from bottom of the screen (last).
-MTo the middle line of the screen.
-)<*> sentences forward.
*(<*> sentences backward.
*}<*> paragraphs forward.
*{<*> paragraphs backward.
-]]To the next section (default EOF).
-[[To the previous section (default begin of file).
-```To the mark.
-'<a-z>To the first CHAR of the line with the mark.
-````To the cursor position before the latest absolute jump (of which are examples '/' and 'G').
-''To the first CHAR of the line on which the cursor was placed before the latest absolute jump.
-/<string>To the next occurrence of <string>.
-?<string>To the previous occurrence of <string>.
-n`Repeat latest ‘/’
-NIdem in opposite direction.
-%`Find the next bracket and go to its match (also with ‘{’


:ta <name> Search in the tags file[s] where <name> is defined (file, line) and go to it.
^]Use the name under the cursor in a ':ta' command.
^TPop the previous tag off the tagstack and return to its position.
:[x,y]g/<string>/<cmd>Search globally [from line x to y] for <string> and execute the 'ex' <cmd> on each occurrence.
:[x,y]v/<string>/<cmd>Execute <cmd> on the lines that don't match.

undoing changes

uUndo the latest change.
UUndo all changes on a line, while not having moved off it (unfortunately).
:q!Quit vi without writing.
:e!Re-edit a messed-up file.

appending text (end with <esc>)

*a<*> times after the cursor.
*A<*> times at the end of line.
*i<*> times before the cursor (insert).
*I<*> times before the first CHAR of the line.
*oOn a new line below the current (open).
*OOn a new line above the current.
*><move>Shift the lines described by <*><move> one shiftwidth to the right.
*>>Shift <*> lines one shiftwidth to the right.
*["<a-zA-Z1-9>]pPut the contents of the (default undo) buffer <*> times after the cursor. A buffer containing lines is put only once, below the current line.
*["<a-zA-Z1-9>]PPut the contents of the (default undo) buffer <*> times before the cursor. A buffer containing lines is put only once, above the current line.
*.Repeat previous command <*> times.  If the last command before a '.' command references a numbered buffer, the buffer number is incremented first (and the count is ignored):  

"1pu.u.u.u.u      - 'walk through' buffers 1 through 5 

"1P....           - restore them


deleting text

| Everything deleted can be stored into a buffer. This is achieved by putting a '"' and a letter <a-z> before the delete command. The deleted text will be in the buffer with the used letter. If <A-Z> is used as buffer name, the conjugate buffer <a-z> will be augmented instead of overwritten with the text. The undo buffer always contains the latest change. Buffers <1-9> contain the latest 9 LINE deletions ('"1' is most recent). |

*xDelete <*> chars under and after the cursor.
*X<*> chars before the cursor.
*d<move>From begin to endpoint of <*><move>.
*dd<*> lines.
-DThe rest of the line.
*<<move>Shift the lines described by <*><move> one shiftwidth to the left.
*<<Shift <*> lines one shiftwidth to the left.
*.Repeat latest command <*> times.

changing text (end with <esc>)

*r<char>Replace <*> chars by <char> - no <esc>.
*ROverwrite the rest of the line, appending change <*> - 1 times.
*sSubstitute <*> chars.
*SSubstitute <*> lines.
*c<move>Change from begin to endpoint of <*><move>.
*cc<*> lines.
*CThe rest of the line and <*> - 1 next lines.
*=<move>If the option 'lisp' is set, this command will realign the lines described by <*><move> as though they had been typed with the option 'ai' set too.
-~Switch lower and upper cases (should be an operator, like 'c').
*JJoin <*> lines (default 2).
*.Repeat latest command <*> times ('J' only once).
-&Repeat latest 'ex' substitute command, e.g. ':s/wrong/good'.
-:[x,y]s/<p>/<r>/<f>Substitute (on lines x through y) the pattern <p> (default the last pattern) with <r>.  Useful  flags <f> are 'g' for 'global' (i.e. change every non-overlapping occurrence of <p>) and 'c' for 'confirm' (type 'y' to confirm a particular substitution, else <cr>).  Instead of '/' any punctuation CHAR unequal to <lf> can be used as delimiter.

substitute replacement patterns

| The basic meta-characters for the replacement pattern are '&' and '~'; these are given as '\&' and '\~' when nomagic is set.  Each instance of '&' is replaced by the characters which the regular expression matched.  The meta-character '~' stands, in the replacement pattern, for the defining text of the previous replacement pattern.  Other meta-sequences possible in the replacement pattern are always introduced by the escaping character '\'.  The sequence '\n' (with 'n' in [1-9]) is replaced by the text matched by the n-th regular subexpression enclosed between '\(' and '\)'.  The sequences '\u' and '\l' cause the immediately following character in the replacement to be converted to upper- or lower-case respectively if this character is a letter.  The sequences '\U'  and '\L' turn such conversion on, either until '\E' or '\e' is encountered, or until the end of the replacement pattern. |

remembering text

| With yank commands you can put '"<a-zA-Z>' before the command, just as with delete commands.  Otherwise you only copy to the undo buffer. The use of buffers <a-z> is THE way of copying text to another file; see the ':e <file>' command. |

*y<move>Yank from begin to endpoint of <*><move>.
*yy<*> lines.
*YIdem (should be equivalent to 'y$' though).
-m<a-z>Mark the cursor position with a letter.

commands while in append|change mode

^@If typed as the first character of the insertion, it is replaced with the previous text inserted (max. 128 chars), after which the insertion is terminated.
^VDeprive the next char of its special meaning (e.g. <esc>).
^DOne shiftwidth to the left, but only if nothing else has been typed on the line.
0^D Remove all indentation on the current line (there must be no other chars on the line).
^^D Idem, but it is restored on the next line.
^T One shiftwidth to the right, but only if nothing else has been typed on the line.
^W One word back.
<kill>Back to the begin of the change on the current line.
<intr>Like <esc> (but you get a beep as well).

writing, editing other files, and quitting vi

| In ':' 'ex' commands - if not the first CHAR on the line - '%' denotes the current file, '#' is a synonym for the alternate file (which normally is the previous file).  As first CHAR on the line '%' is a shorthand for '1,$'. Marks can be used for line numbers too: '<a-z>. In the ':w'|':f'|':cd'|':e'|':n' commands shell meta-characters can be used. |

:qQuit vi, unless the buffer has been changed.
:q!Quit vi without writing.
^ZSuspend vi.
:wWrite the file.
:w <name>Write to the file <name>.
:w >> <name>Append the buffer to the file <name>.
:w! <name>Overwrite the file <name>.
:x,y w <name>Write lines x through y to the file <name>.
:wqWrite the file and quit vi; some versions quit even if the write was unsuccessful! Use 'ZZ' instead.
ZZWrite if the buffer has been changed, and quit vi.  If you have invoked vi with the '-r' option, you'd better write the file explicitly ('w' or 'w!'), or quit the editor explicitly ('q!') if you don't want to overwrite the file - some versions of vi don't handle the 'recover' option very well.
:x [<file>] Idem [but write to <file>].
:x! [<file>]':w![<file>]' and ':q'.
:prePreserve the file - the buffer is saved as if the system had just crashed; for emergencies, when a ':w' command has failed and you don't know how to save your work (see 'vi -r').
:f <name>Set the working directory to <dir> (default home directory).
:cd! [<dir>] Idem, but don't save changes.
:e [+<cmd>] <file> Edit another file without quitting vi - the buffers are not changed (except the undo buffer), so text can be copied from one file to another this way.  [Execute the 'ex' command <cmd> (default '$') when the new file has been read into the buffer.]  <cmd> must contain no <sp> or <ht>.  See 'vi startup'.
:e! [+<cmd>] <file>Idem, without writing the current buffer.
^^Edit the alternate (normally the previous) file.
:rewRewind the argument list, edit the first file.
:rew!Idem, without writing the current buffer.
:n [+<cmd>] [<files>]Edit next file or specify a new argument list.
:n! [+<cmd>] [<files>]Idem, without writing the current buffer.
:argsGive the argument list, with the current file between '[' and ']'.

display Commands

^GGive file name, status, current line number and relative position.
^LRefresh the screen (sometimes '^P' or '^R').
^R Sometimes vi replaces a deleted line by a '@', to be deleted by '^R' (see option 'redraw').
[*]^EExpose <*> more lines at bottom, cursor stays put (if possible).
[*]^YExpose <*> more lines at top, cursor stays put (if possible).
[*]^DScroll <*> lines downward (default the number of the previous scroll; initialization: half a page).
[*]^UScroll <*> lines upward (default the number of the previous scroll; initialization: half a page).
[*]^F<*> pages forward.
[*]^B<*> pages backward (in older versions '^B' only works without count).

| If in the next commands the field <wi> is present, the windowsize will change to <wi>. The window will always be displayed at the bottom of the screen. |

[*]z[wi]<cr>Put line <*> at the top of the window (default the current line).
[*]z[wi]+Put line <*> at the top of the window (default the first line of the next page).
[*]z[wi]-Put line <*> at the bottom of the window (default the current line).
[*]z[wi]^ Put line <*> at the bottom of the window (default the last line of the previous page).
[*]z[wi]. Put line <*> in the centre of the window (default the current line).

Mapping and abbreviation

| When mapping take a look at the options 'to' and 'remap' (below).  |

:map <string> <seq><string> is interpreted as <seq>, e.g. ':map ^C :!cc %^V<cr>' to invoke 'cc' (the C compiler) from within the editor (vi replaces '%' with the current file name).
:mapShow all mappings.
:unmap <string>Deprive <string> of its mapping.  When vi complains about non-mapped macros (whereas no typos have been made), first do something like ':map <string> Z', followed by ':unmap <string>' ('Z' must not be a macro itself), or switch to 'ex' mode first with 'Q'.
:map! <string> <seq>Mapping in append mode, e.g.  

':map! \be begin^V<cr>end;^V<esc>O<ht>'

When in append mode <string> is preceded by '^V', no mapping is done. | | :map! | Show all append mode mappings. | | :unmap! <string>  | Deprive <string> of its mapping (see ':unmap'). | | :ab <string> <seq> | Whenever in append mode <string> is preceded and followed by a breakpoint (e.g. <sp> or ','), it is interpreted as <seq>, e.g. ':ab ^P procedure'.  A '^V' immediately following <string> inhibits expansion. | | :ab | Show all abbreviations. | | :unab <string> | Do not consider <string> an abbreviation anymore (see ':unmap'). | | @<a-z> | Consider the contents of the named register a command, e.g.:   



oopen a new line
0^Dremove indentation
:s/wrong/goodthis input text is an 'ex' substitute command
<esc>finish the input
"zdddelete the line just created into register 'z'
Now you can type '@z' to replace 'wrong' with 'good' on the current line.
@@Repeat last register command.

switch and shell commands

:An 'ex' command can be given.
:viSwitch from 'ex' to vi.
:shExecute a subshell, back to vi by '^D'.
:[x,y]!<cmd>Execute a shell <cmd> [on lines x through y; these lines will serve as input for <cmd> and will be replaced by its standard output].
:[x,y]!! [<args>]Repeat last shell command [and append <args>].
:[x,y]!<cmd> ! [<args>]Use the previous command (the second '!') in a new command.
[*]!<move><cmd>The shell executes <cmd>, with as standard input the lines described by <*><move>, next the standard output replaces those lines (think of 'cb', 'sort', 'nroff', etc.).
[*]!<move>!<args>Append <args> to the last <cmd> and execute it, using the lines described by the current <*><move>.
[*]!!<cmd>Give <*> lines as standard input to the shell <cmd>, next let the standard output replace those lines.
[*]!!! [<args>]Use the previous <cmd> [and append <args> to it].
:x,y w !<cmd>Let lines x to y be standard input for <cmd> (notice the <sp> between the 'w' and the '!').
:r!<cmd>Put the output of <cmd> onto a new line.
:r <name>Read the file <name> into the buffer.

vi startup

vi [<files>]Edit the files, start with the first page of the first file.

| The editor can be initialized by the shell variable 'EXINIT', which looks like: 

    <cmd>: set options
           map ...
           ab ...
    export EXINIT (in the Bourne shell) 

However, the list of initializations can also be put into a file. If this file is located in your home directory, and is named '.exrc' AND the variable 'EXINIT' is NOT set, the list will be executed automatically at startup time. However, vi will always execute the contents of a '.exrc' in the current directory, if you own the file. Else you have to give the execute ('source') command yourself: 

    :so file 

In a '.exrc' file a comment is introduced with a double quote character: the rest of the line is ignored.  Exception: if the last command on the line is a 'map[!]' or 'ab' command or a shell escape, a trailing comment is not recognized, but considered part of the command. 

On-line initializations can be given with 'vi +<cmd> file', e.g.:


| vi +x file | The cursor will immediately jump to line x (default last line). | | vi +/<string> file | Jump to the first occurrence of <string>. |

| You can start at a particular tag with:  |

| vi -t <tag> | Start in the right file in the right place. |

| Sometimes (e.g. if the system crashed while you were editing) it is possible to recover files lost in the editor by 'vi -r file'.  A plain 'vi -r' command shows the files you can recover.If you just want to view a file by using vi, and you want to avoid any change, instead of vi you can use the 'view' or 'vi -R' command: the option 'readonly' will be set automatically (with ':w!' you can override this option). |

the most important options

aiautoindent - In append mode after a <cr> the cursor will move directly below the first CHAR on the previous line.  However, if the option 'lisp' is set, the cursor will align at the first argument to the last open list.
awautowrite - Write at every shell escape (useful when compiling from within vi).
dir=<string>directory - The directory for vi to make temporary files (default '/tmp').
eberrorbells - Beeps when you goof (not on every terminal).
icignorecase - No distinction between upper and lower cases when searching.
lispRedefine the following commands:  

'(', ')'   - move backward (forward) over S-expressions 

'{', '}'   - idem, but don't stop at atoms 

'[[', ']]' - go to previous (next) line beginning with a '('

See option 'ai'. | | list | <lf> is shown as '$', <ht> as '^I'. | | magic | If this option is set (default), the chars '.', '[' and '*' have special meanings within search and 'ex' substitute commands.  To deprive such a char of its special function it must be preceded by a '\'.  If the option is turned off it's just the other way around.  Meta-chars:  

^<string>    - <string> must begin the line 

<string>$    - <string> must end the line 

.            - matches any char 

[a-z]        - matches any char in the range 

[^a-z]       - any char not in the range 

[<string>]   - matches any char in <string> 

[^<string>]  - any char not in <string> 

<char>*      - 0 or more <char>s 

\<<string>   - <string> must begin a word 

<string>\>   - <string> must end a word

| | modeline | When you read an existing file into the buffer, and this option is set, the first and last 5 lines are checked for editing commands in the following form:  

<sp>vi:set options|map ...|ab ...|!...:

Instead of <sp> a <ht> can be used, instead of 'vi' there can be 'ex'.  Warning: this option could have nasty results if you edit a file containing 'strange' modelines. | | nu | number - Numbers before the lines. | | para=<string> | paragraphs - Every pair of chars in <string> is considered a paragraph delimiter nroff macro(for '{' and '}').  A <sp> preceded by a '\' indicates the previous char is a single letter macro.  ':set para=P\ bp' introduces '.P' and '.bp' as paragraph delimiters.  Empty lines and section boundaries are paragraph boundaries too. | | redraw  | The screen remains up to date. | | remap | If on (default), macros are repeatedly expanded until they are unchanged. Example: if 'o' is mapped to 'A', and 'A' is mapped to 'I', then 'o' will map to 'I' if 'remap' is set, else it will map to 'A'. | | report=<*>  | Vi reports whenever e.g. a delete or yank command affects <*> or more lines. | | ro  | readonly - The file is not to be changed. However, ':w!' will override this option. | | sect=<string> | sections - Gives the section delimiters (for '[[' and ']]'); see option 'para'. A '{' beginning a line also starts a section (as in C functions). | | sh=<string> | shell - The program to be used for shell escapes (default '$SHELL' (default '/bin/sh')). | | sw=<*> | shiftwidth - Gives the shiftwidth (default 8 positions). | | sm | showmatch - Whenever you append a ')', vi shows its match if it's on the same page; also with '{' and '}'.  If there's no match at all, vi will beep. | | taglength=<*> | The number of significant characters in tags (0 = unlimited). | | tags=<string>  | The comma-separated list of tags files. | | terse | Short error messages. | | to | timeout - If this option is set, append mode mappings will be interpreted only if they're typed fast enough. | | ts=<*> | tabstop - The length of a <ht>; warning: this is only IN the editor, outside of it <ht>s have their normal length (default 8 positions). | | wa | writeany - No checks when writing (dangerous). | | warn | Warn you when you try to quit without writing. | | wi=<*> | window - The default number of lines vi shows. | | wm=<*> | wrapmargin - In append mode vi automaticallyputs a <lf> whenever there is a <sp> or <ht> within <wm> columns from the right margin (0 = don't put a <lf> in the file, yet put it on the screen). | | ws | wrapscan - When searching, the end is considered 'stuck' to the begin of the file. | | :set <option> | Turn <option> on. | | :set no<option>  | Turn <option> off. | | :set <option>=<value> | Set <option> to <value>. | | :set  | Show all non-default options and their values. | | :set <option>? | Show <option>'s value. | | :set all | Show all options and their values. |