edit: API for the interactive editor

Table of content

The edit: module is the interface to the Elvish editor.

Function usages are given in the same format as in the reference for the builtin module.

This document is incomplete.


Modes and Submodules

The Elvish editor has different modes, and exactly one mode is active at the same time. Each mode has its own UI and keybindings. For instance, the default insert mode lets you modify the current command. The completion mode (triggered by Tab by default) shows you all candidates for completion, and you can use arrow keys to navigate those candidates.

~/elvish> vim 1.0-release.md                elf@host
 COMPLETING argument 
1.0-release.md   README.md    syntaxes/
Dockerfile       cmd/         vscode/  
LICENSE          go.mod       website/ 
Makefile         go.sum
PACKAGING.md     pkg/       

Each mode has its own submodule under edit:. For instance, builtin functions and configuration variables for the completion mode can be found in the edit:completion: module.

The primary modes supported now are insert, command, completion, navigation, history, histlist, location, and lastcmd. The last 4 are “listing modes”, and their particularity is documented below.


Elvish has two prompts: the (normal) left-hand prompt and the right-side prompt (rprompt). Most of this section only documents the left-hand prompt, but API for rprompt is the same other than the variable name: just replace prompt with rprompt.

To customize the prompt, assign a function to edit:prompt. The function may write value outputs or byte outputs:

  • Value outputs may be either strings or styled values; they are joiend with no spaces in between.

  • Byte outputs are output as-is, including any newlines. Any SGR escape sequences included in the byte outputs will be parsed, but any other escape sequences or control character will be removed.

If you mix value and byte outputs, the order in which they appear is non-deterministic.

Prefer using styled to output styled text; the support for SGR escape sequences is mostly for compatibility with external cross-shell prompts.

The default prompt and rprompt are equivalent to:

set edit:prompt = { tilde-abbr $pwd; put '> ' }
set edit:rprompt = (constantly (styled (whoami)@(hostname) inverse))

Some more examples:

~> set edit:prompt = { tilde-abbr $pwd; styled '> ' green }
~> # ">" is now green
~> set edit:prompt = { echo '$' }
# Cursor will be on the next line as `echo` outputs a trailing newline

Stale Prompt

Elvish never waits for the prompt function to finish. Instead, the prompt function is always executed on a separate thread, and Elvish updates the screen when the function finishes.

However, this can be misleading when the function is slow: this means that the prompt on the screen may not contain the latest information. To deal with this, if the prompt function does not finish within a certain threshold - by default 0.2 seconds, Elvish marks the prompt as stale: it still shows the old stale prompt content, but transforms it using a stale transformer. The default stale transformer applies reverse-video to the whole prompt.

The threshold is customizable with $edit:prompt-stale-threshold; it specifies the threshold in seconds.

The transformer is customizable with $edit:prompt-stale-transform. It is a function; the function is called with one argument, a styled text, and the output is interpreted in the same way as prompt functions. Some examples are:

# The following effectively disables marking of stale prompt.
set edit:prompt-stale-transform = {|x| put $x }
# Show stale prompts in inverse; equivalent to the default.
set edit:prompt-stale-transform = {|x| styled $x inverse }
# Gray out stale prompts.
set edit:prompt-stale-transform = {|x| styled $x bright-black }

To see the transformer in action, try the following example (assuming default $edit:prompt-stale-transform):

var n = 0
set edit:prompt = { sleep 2; put $n; set n = (+ $n 1); put ': ' }
set edit:-prompt-eagerness = 10 # update prompt on each keystroke
set edit:prompt-stale-threshold = 0.5

And then start typing. Type one character; the prompt becomes inverse after 0.5 second: this is when Elvish starts to consider the prompt as stale. The prompt will return normal after 2 seconds, and the counter in the prompt is updated: this is when the prompt function finishes.

Another thing you will notice is that, if you type a few characters quickly (in less than 2 seconds, to be precise), the prompt is only updated twice. This is because Elvish never does two prompt updates in parallel: prompt updates are serialized. If a prompt update is required when the prompt function is still running, Elvish simply queues another update. If an update is already queued, Elvish does not queue another update. The reason why exactly two updates happen in this case, and how this algorithm ensures freshness of the prompt is left as an exercise to the reader.

Prompt Eagerness

The occasions when the prompt should get updated can be controlled with $edit:-prompt-eagerness:

  • The prompt is always updated when the editor becomes active – when Elvish starts, or a command finishes execution, or when the user presses Enter.

  • If $edit:-prompt-eagerness >= 5, it is updated when the working directory changes.

  • If $edit:-prompt-eagerness >= 10, it is updated on each keystroke.

The default value is 5.

RPrompt Persistency

By default, the rprompt is only shown while the editor is active: as soon as you press Enter, it is erased. If you want to keep it, simply set $edit:rprompt-persistent to $true:

set edit:rprompt-persistent = $true


Each mode has its own keybinding, accessible as the binding variable in its module. For instance, the binding table for insert mode is $edit:insert:binding. To see current bindings, simply print the binding table: pprint $edit:insert:binding (replace insert with any other mode).

The global key binding table, $edit:global-binding is consulted when a key is not handled by the active mode.

A binding tables is simply a map that maps keys to functions. For instance, to bind Alt-x in insert mode to exit Elvish, simply do:

set edit:insert:binding[Alt-x] = { exit }

Outputs from a bound function always appear above the Elvish prompt. You can see this by doing the following:

set edit:insert:binding[Alt-x] = { echo 'output from a bound function!' }

and press Alt-x in insert mode. It allows you to put debugging outputs in bound functions without messing up the terminal.

Internally, this is implemented by connecting their output to a pipe. This does the correct thing in most cases, but if you are sure you want to do something to the terminal, redirect the output to /dev/tty. Since this will break Elvish’s internal tracking of the terminal state, you should also do a full redraw with edit:redraw &full=$true. For instance, the following binds Ctrl-L to clearing the terminal:

set edit:insert:binding[Ctrl-L] = { clear > /dev/tty; edit:redraw &full=$true }

(The same functionality is already available as a builtin, edit:clear.)

Bound functions have their inputs redirected to /dev/null.

Format of Keys

A key consists of a key name, preceded by zero or more key modifiers. Both are case-sensitive.

The key name may be either:

  • A simple character, such as x.

    Most of the time you should use the lower-case form of letters, except for Ctrl- bindings, which usually require the upper-case form (so Alt-x but Ctrl-X). This quirk may go away in future.

  • A function key from these symbols:

    F1  F2  F3  F4  F5  F6  F7  F8  F9  F10  F11  F12
    Up  Down  Right  Left
    Home  Insert  Delete  End  PageUp  PageDown
    Tab  Enter  Backspace

    Among these, Tab and Enter are equivalent to their ASCII counterparts "\t" and "\n".

Key modifiers are one of the following:

A  Alt
C  Ctrl
M  Meta
S  Shift

Modifiers end with either a - or +, and can be stacked. Examples:

  • S-F1

  • Ctrl-X or

  • Alt+Enter

  • C+A-X.

Note: The Shift modifier is only applicable to function keys such as F1. You cannot write Shift-m as a synonym for M.

You may not actually be able to use the full range of possible keys for several reasons:

  • Some Ctrl-[letter] keys have special functions, like Ctrl-C, Ctrl-Z, Ctrl-S and Ctrl-Z.

  • Some Ctrl-[letter] keys are equivalent to single keypresses, like Ctrl-J (equivalent to Enter) and Ctrl-I (equivalent to Tab).

  • Keys involving multiple modifiers may not be supported by the terminal emulator, especially when the base key is a function key.

Listing Modes

The modes histlist, location and lastcmd are all listing modes: They all show a list, and you can filter items and accept items.

Because they are very similar, you may want to change their bindings at the same time. This is made possible by the $edit:listing:binding binding table (listing is not a “real” mode but an “abstract” mode). These modes still have their own binding tables like $edit:histlist:binding, and bindings there have higher precedence over those in the shared $edit:listing:binding table.

Moreover, there are a lot of builtin functions in the edit:listing module like edit:listing:down (for moving down selection). They always apply to whichever listing mode is active.

Caveat: Bindings to Start Modes

Note that keybindings to start modes live in the binding table of the insert mode, not the target mode. For instance, if you want to be able to use Alt-l to start location mode, you should modify $edit:insert:binding[Alt-l]:

set edit:insert:binding[Alt-l] = { edit:location:start }

One tricky case is the history mode. You can press ▲︎ to start searching for history, and continue pressing it to search further. However, when the first press happens, the editor is in insert mode, while with subsequent presses, the editor is in history mode. Hence this binding actually relies on two entries, $edit:insert:binding[Up] and $edit:history:binding[Up].

So for instance if you want to be able to use Ctrl-P for this, you need to modify both bindings:

set edit:insert:binding[Ctrl-P] =  { edit:history:start }
set edit:history:binding[Ctrl-P] = { edit:history:up }

Filter DSL

The completion, history listing, location and navigation modes all support filtering the items to show using a filter DSL. It uses a small subset of Elvish’s expression syntax, and can be any of the following:

  • A literal string (barewords and single-quoted or double-quoted strings all work) matches items containing the string. If the string is all lower case, the match is done case-insensitively; otherwise the match is case-sensitive.

  • A list [re $string] matches items matching the regular expression $string. The $string must be a literal string.

  • A list [and $expr...] matches items matching all of the $exprs.

  • A list [or $expr...] matches items matching any of the $exprs.

If the filter contains multiple expressions, they are ANDed, as if surrounded by an implicit [and ...].

Completion API

Argument Completer

There are two types of completions in Elvish: completion for internal data and completion for command arguments. The former includes completion for variable names (e.g. echo $Tab) and indices (e.g. echo $edit:insert:binding[Tab). These are the completions that Elvish can provide itself because they only depend on the internal state of Elvish.

The latter, in turn, is what happens when you type e.g. catTab. Elvish cannot provide completions for them without full knowledge of the command.

Command argument completions are programmable via the $edit:completion:arg-completer variable. When Elvish is completing an argument of command $x, it will call the value stored in $edit:completion:arg-completer[$x], with all the existing arguments, plus the command name in the front.

For example, if the user types man 1Tab, Elvish will call:

$edit:completion:arg-completer[man] man 1

If the user is starting a new argument when hitting Tab, Elvish will call the completer with a trailing empty string. For instance, if you do man 1SpaceTab, Elvish will call:

$edit:completion:arg-completer[man] man 1 ""

The output of this call becomes candidates. There are several ways of outputting candidates:

  • Writing byte output, e.g. “echo cand1; echo cand2”. Each line becomes a candidate. This has the drawback that you cannot put newlines in candidates. Only use this if you are sure that you candidates will not contain newlines – e.g. package names, usernames, but not file names, etc..

  • Write strings to value output, e.g. “put cand1 cand2”. Each string output becomes a candidate.

  • Use the edit:complex-candidate command, e.g.:

    edit:complex-candidate &code-suffix='' &display=$stem' ('$description')'  $stem

    See edit:complex-candidate for the full description of the arguments is accepts.

After receiving your candidates, Elvish will match your candidates against what the user has typed. Hence, normally you don’t need to (and shouldn’t) do any matching yourself.

That means that in many cases you can (and should) simply ignore the last argument to your completer. However, they can be useful for deciding what kind of things to complete. For instance, if you are to write a completer for ls, you want to see whether the last argument starts with - or not: if it does, complete an option; and if not, complete a filename.

Here is a very basic example of configuring a completer for the apt command. It only supports completing the install and remove command and package names after that:

var all-packages = [(apt-cache search '' | eawk {|0 1 @rest| put $1 })]

set edit:completion:arg-completer[apt] = {|@args|
    var n = (count $args)
    if (== $n 2) {
        # apt x<Tab> -- complete a subcommand name
        put install uninstall
    } elif (== $n 3) {
        put $@all-packages

Here is another slightly more complex example for the git command. It supports completing some common subcommands and then branch names after that:

fn all-git-branches {
    # Note: this assumes a recent version of git that supports the format
    # string used.
    git branch -a --format="%(refname:strip=2)" | eawk {|0 1 @rest| put $1 }

var common-git-commands = [
  add branch checkout clone commit diff init log merge
  pull push rebase reset revert show stash status

set edit:completion:arg-completer[git] = {|@args|
    var n = (count $args)
    if (== $n 2) {
        put $@common-git-commands
    } elif (>= $n 3) {


As stated above, after the completer outputs candidates, Elvish matches them with them with what the user has typed. For clarity, the part of the user input that is relevant to tab completion is called for the seed of the completion. For instance, in echo xTab, the seed is x.

Elvish first indexes the matcher table – $edit:completion:matcher – with the completion type to find a matcher. The completion type is currently one of variable, index, command, redir or argument. If the $edit:completion:matcher lacks the suitable key, $edit:completion:matcher[''] is used.

Elvish then calls the matcher with one argument – the seed, and feeds the text of all candidates to the input. The mather must output an identical number of booleans, indicating whether the candidate should be kept.

As an example, the following code configures a prefix matcher for all completion types:

set edit:completion:matcher[''] = {|seed| each {|cand| has-prefix $cand $seed } }

Elvish provides three builtin matchers, edit:match-prefix, edit:match-substr and edit:match-subseq. In addition to conforming to the matcher protocol, they accept two options &ignore-case and &smart-case. For example, if you want completion of arguments to use prefix matching and ignore case, use:

set edit:completion:matcher[argument] = {|seed| edit:match-prefix $seed &ignore-case=$true }

The default value of $edit:completion:matcher is [&''=$edit:match-prefix~], hence that candidates for all completion types are matched by prefix.


Hooks are functions that are executed at certain points in time. In Elvish this functionality is provided by variables that are a list of functions.

NOTE: Hook variables may be initialized with a non-empty list, and you may have modules that add their own hooks. In general you should append to a hook variable rather than assign a list of functions to it. That is, rather than doing set edit:some-hook = [ { put 'I ran' } ] you should do set edit:some-hook = [ $@hook-var { put 'I ran' } ].

These are the editor/REPL hooks:

  • $edit:before-readline: The functions are called before the editor runs. Each function is called with no arguments.

  • $edit:after-readline: The functions are called after the editor accepts a command for execution. Each function is called with a sole argument: the line just read.

  • $edit:after-command: The functions are called after the shell executes the command you entered (typically by pressing the Enter key). Each function is called with a sole argument: a map that provides information about the executed command. This hook is also called after your interactive RC file is executed and before the first prompt is output.

Example usage:

set edit:before-readline = [{ echo 'going to read' }]
set edit:after-readline = [{|line| echo 'just read '$line }]
set edit:after-command = [{|m| echo 'command took '$m[duration]' seconds' }]

Given the above hooks…

  1. Every time you accept a chunk of code (normally by pressing Enter) just read is printed.

  2. At the very beginning of an Elvish session, or after a chunk of code is handled, going to read is printed.

  3. After each non empty chunk of code is accepted and executed the string “command took … seconds` is output.

Word types

The editor supports operating on entire “words”. As intuitive as the concept of “word” is, there is actually no single definition for the concept. The editor supports the following three definitions of words:

  • A big word, or simply word, is a sequence of non-whitespace characters. This definition corresponds to the concept of “WORD” in vi.

  • A small word is a sequence of alphanumerical characters (”alnum small word”), or a sequence of non-alphanumerical, non-whitespace characters (”punctuation small word”). This definition corresponds to the concept of “word” in vi and zsh.

  • An alphanumerical word is a sequence of alphanumerical characters. This definition corresponds to the concept of “word” in bash.

Whitespace characters are those with the Unicode Whitespace property. Alphanumerical characters are those in the Unicode Letter or Number category.

A word boundary is an imaginary zero-length boundary around a word.

To see the difference between these definitions, consider the following string: abc++ /* xyz:

  • It contains three (big) words: abc++, /* and xyz.

  • It contains four small words, abc, ++, /* and xyz. Among them, abc and xyz are alnum small words, while ++ and /* are punctuation small words.

  • It contains two alnum words, abc and xyz.


The editor can identify autofix commands to fix some errors in the code.

For example, if you try to use a command from the str: module without importing it, the editor will offer use str as an autofix command:

~> str:                                     elf@host
Ctrl-A autofix: use str Tab Enter autofix first

As seen above, autofixes are also applied automatically by edit:completion:smart-start (the default binding for Tab) and edit:smart-enter (the default binding for Enter).



Contains the current position of the cursor, as a byte position within $edit:current-command.


Binding for the instant mode.


See Prompt Eagerness.


See Prompt Eagerness.


A map from simple abbreviations to their expansions.

An abbreviation is replaced by its expansion when it is typed in full and consecutively, without being interrupted by the use of other editing functionalities, such as cursor movements.

If more than one abbreviations would match, the longest one is used.


set edit:abbr['||'] = '| less'
set edit:abbr['>dn'] = '2>/dev/null'

With the definitions above, typing || anywhere expands to | less, and typing >dn anywhere expands to 2>/dev/null. However, typing a |, moving the cursor left, and typing another | does not expand to | less, since the abbreviation || was not typed consecutively.

See also $edit:command-abbr and $edit:small-word-abbr.


List of filters to run before adding a command to history.

A filter is a function that takes a command as argument and outputs a boolean value. If any of the filters outputs $false, the command is not saved to history, and the rest of the filters are not run. The default value of this list contains a filter which ignores command starts with space.


A list of functions to call after each interactive command completes. There is one pre-defined function used to populate the $edit:command-duration variable. Each function is called with a single map argument containing the following keys:

  • src: Information about the source that was executed, same as what src would output inside the code.

  • duration: A floating-point number representing the command execution duration in seconds.

  • error: An exception object if the command terminated with an exception, else $nil.

See also $edit:command-duration.


A list of functions to call after each readline cycle. Each function is called with a single string argument containing the code that has been read.


A list of functions to call before each readline cycle. Each function is called without any arguments.


A map from command abbreviations to their expansions.

A command abbreviation is replaced by its expansion when seen in the command position followed by a whitespace. This is similar to the Fish shell’s abbreviations, but does not trigger when executing a command with Enter – you must type a space first.


set edit:command-abbr['l'] = 'less'
set edit:command-abbr['gc'] = 'git commit'

See also $edit:abbr and $edit:small-word-abbr.


Duration, in seconds, of the most recent interactive command. This can be useful in your prompt to provide feedback on how long a command took to run. The initial value of this variable is the time to evaluate your rc.elv before printing the first prompt.

See also $edit:after-command.


Key bindings for command mode. This is currently a very small subset of Vi command mode bindings.

See also edit:command:start.


A map containing argument completers.


Keybinding for the completion mode.


A map mapping from context names to matcher functions. See the Matcher section.


Contains the content of the current input. Setting the variable will cause the cursor to move to the very end, as if edit-dot = (count $edit:current-command) has been invoked.

This API is subject to change.


A list of exceptions thrown from callbacks such as prompts. Useful for examining tracebacks and other metadata.


Global keybindings, consulted for keys not handled by mode-specific bindings.

See Keybindings.


Keybinding for the history listing mode.

Keys bound to edit:histlist:toggle-dedup (Ctrl-D by default) will be shown in the history listing UI.



Binding table for the history mode.


Binding map for the insert mode.

The key bound to edit:apply-autofix will be shown when an autofix is available.


A boolean used to control whether text pasted using bracketed paste in the terminal should be quoted as a string. Defaults to $false.


Keybinding for the last command mode.


Common binding table for listing modes.


Keybinding for the location mode.


A list of directories to hide in the location addon.


A list of directories to always show at the top of the list of the location addon.


A map mapping types of workspaces to their patterns.


Maximum height the editor is allowed to use, defaults to +Inf.

By default, the height of the editor is only restricted by the terminal height. Some modes like location mode can use a lot of lines; as a result, it can often occupy the entire terminal, and push up your scrollback buffer. Change this variable to a finite number to restrict the height of the editor.


Keybinding for the navigation mode.

Keys bound to edit:navigation:trigger-filter (Ctrl-F by default) and edit:navigation:trigger-shown-hidden (Ctrl-H by default) will be shown in the navigation mode UI.


Name of the currently selected file in navigation mode. $nil if not in navigation mode.


A list of 3 integers, used for specifying the width ratio of the 3 columns in navigation mode.


See Prompts.


See Stale Prompt.


See Stale Prompt.


See Prompts.


See RPrompt Persistency.


See Stale Prompt.


See Stale Prompt.


A map from small-word abbreviations to their expansions.

A small-word abbreviation is replaced by its expansion after it is typed in full and consecutively, and followed by another character (the trigger character). Furthermore, the expansion requires the following conditions to be satisfied:

  • The end of the abbreviation must be adjacent to a small-word boundary, i.e. the last character of the abbreviation and the trigger character must be from two different small-word categories.

  • The start of the abbreviation must also be adjacent to a small-word boundary, unless it appears at the beginning of the code buffer.

  • The cursor must be at the end of the buffer.

If more than one abbreviations would match, the longest one is used. See the description of small words for more information.

As an example, with the following configuration:

set edit:small-word-abbr['gcm'] = 'git checkout master'

In the following scenarios, the gcm abbreviation is expanded:

  • With an empty buffer, typing gcm and a space or semicolon;

  • When the buffer ends with a space, typing gcm and a space or semicolon.

The space or semicolon after gcm is preserved in both cases.

In the following scenarios, the gcm abbreviation is not expanded:

  • With an empty buffer, typing Xgcm and a space or semicolon (start of abbreviation is not adjacent to a small-word boundary);

  • When the buffer ends with X, typing gcm and a space or semicolon (end of abbreviation is not adjacent to a small-word boundary);

  • When the buffer is non-empty, move the cursor to the beginning, and typing gcm and a space (cursor not at the end of the buffer).

This example shows the case where the abbreviation consists of a single small word of alphanumerical characters, but that doesn’t have to be the case. For example, with the following configuration:

set edit:small-word-abbr['>dn'] = ' 2>/dev/null'

The abbreviation >dn starts with a punctuation character, and ends with an alphanumerical character. This means that it is expanded when it borders a whitespace or alphanumerical character to the left, and a whitespace or punctuation to the right; for example, typing ls>dn; will expand it.

Some extra examples of small-word abbreviations:

set edit:small-word-abbr['gcp'] = 'git cherry-pick -x'
set edit:small-word-abbr['ll'] = 'ls -ltr'

If both a simple abbreviation and a small-word abbreviation can be expanded, the simple abbreviation has priority.

See also $edit:abbr $edit:command-abbr.




Starts the instant mode. In instant mode, any text entered at the command line is evaluated immediately, with the output displayed.

WARNING: Beware of unintended consequences when using destructive commands. For example, if you type sudo rm -rf /tmp/* in the instant mode, Elvish will attempt to evaluate sudo rm -rf / when you typed that far.


edit:add-var $name $init

Defines a new variable in the interactive REPL with an initial value. The new variable becomes available during the next REPL cycle.

Equivalent to running var $name = $init at a REPL prompt, but $name can be dynamic.

This is most useful for modules to modify the REPL namespace. Example:

~> cat .config/elvish/lib/a.elv
for i [(range 10)] {
  edit:add-var foo$i $i
~> use a
~> put $foo1 $foo2
▶ (num 1)
▶ (num 2)

Note that if you use a variable as the $init argument, edit:add-var doesn’t add the variable “itself” to the REPL namespace. The variable in the REPL namespace will have the initial value set to the variable’s value, but it is not an alias of the original variable:

~> cat .config/elvish/lib/b.elv
var foo = foo
edit:add-var foo $foo
~> use b
~> put $foo
▶ foo
~> set foo = bar
~> echo $b:foo

Importing definition from a module into the REPL

One common use of this command is to put the definitions of functions intended for REPL use in a module instead of your rc.elv. For example, if you want to define ll as ls -l, you can do so in your rc.elv directly:

fn ll {|@a| ls -l $@a }

But if you move the definition into a module (say util.elv in one of the module search directories, this function can only be used as util:ll (after use util). To make it usable directly as ll, you can add the following to util.elv:

edit:add-var ll~ $ll~

Conditionally importing a module

Another use case is to add a module or function to the REPL namespace conditionally. For example, to only import the unix module when actually running on Unix, a straightforward solution is to do the following in rc.elv:

use platform
if $platform:is-unix {
  use unix

This doesn’t work however, since what use does is introducing a variable named $unix:. Since all variables in Elvish are lexically scoped, the $unix: variable is only valid inside the if block.

This can be fixed by explicitly introducing the $unix: variable to the REPL namespace. The following works both from rc.elv and from a module:

use platform
if $platform:is-unix {
  use unix
  edit:add-var unix: $unix:


edit:add-vars $map

Takes a map from strings to arbitrary values. Equivalent to calling edit:add-var for each key-value pair in the map, but guarantees that all the names will be added at the same time.



Executes the currently suggested autofix.


edit:binding-table $map

Converts a normal map into a binding map.



Clears the screen.

This command should be used in place of the external clear command to clear the screen.



Closes the current active mode.


edit:command-history &cmd-only=$false &dedup=$false &newest-first

Outputs the command history.

By default, each entry is represented as a map, with an id key key for the sequence number of the command, and a cmd key for the text of the command. If &cmd-only is $true, only the text of each command is output.

All entries are output by default. If &dedup is $true, only the most recent instance of each command (when comparing just the cmd key) is output.

Commands are are output in oldest to newest order by default. If &newest-first is $true the output is in newest to oldest order instead.

As an example, either of the following extracts the text of the most recent command:

edit:command-history | put [(all)][-1][cmd]
edit:command-history &cmd-only &newest-first | take 1



Enter command mode. This mode is intended to emulate Vi’s command mode, but it is very incomplete right now.

See also $edit:command:binding.


edit:complete-filename $args...

Produces a list of filenames found in the directory of the last argument. All other arguments are ignored. If the last argument does not contain a path (either absolute or relative to the current directory), then the current directory is used. Relevant files are output as edit:complex-candidate objects.

This function is the default handler for any commands without explicit handlers in $edit:completion:arg-completer. See Argument Completer.


~> edit:complete-filename ''
▶ (edit:complex-candidate Applications &code-suffix=/ &style='01;34')
▶ (edit:complex-candidate Books &code-suffix=/ &style='01;34')
▶ (edit:complex-candidate Desktop &code-suffix=/ &style='01;34')
▶ (edit:complex-candidate Docsafe &code-suffix=/ &style='01;34')
▶ (edit:complex-candidate Documents &code-suffix=/ &style='01;34')
~> edit:complete-filename .elvish/
▶ (edit:complex-candidate .elvish/aliases &code-suffix=/ &style='01;34')
▶ (edit:complex-candidate .elvish/db &code-suffix=' ' &style='')
▶ (edit:complex-candidate .elvish/epm-installed &code-suffix=' ' &style='')
▶ (edit:complex-candidate .elvish/lib &code-suffix=/ &style='01;34')
▶ (edit:complex-candidate .elvish/rc.elv &code-suffix=' ' &style='')


edit:complete-getopt $args $opt-specs $arg-handlers

Produces completions according to a specification of accepted command-line options (both short and long options are handled), positional handler functions for each command position, and the current arguments in the command line. The arguments are as follows:

  • $args is an array containing the current arguments in the command line (without the command itself). These are the arguments as passed to the Argument Completer function.

  • $opt-specs is an array of maps, each one containing the definition of one possible command-line option. Matching options will be provided as completions when the last element of $args starts with a dash, but not otherwise. Each map can contain the following keys (at least one of short or long needs to be specified):

    • short contains the one-letter short option, if any, without the dash.

    • long contains the long option name, if any, without the initial two dashes.

    • arg-optional, if set to $true, specifies that the option receives an optional argument.

    • arg-required, if set to $true, specifies that the option receives a mandatory argument. Only one of arg-optional or arg-required can be set to $true.

    • desc can be set to a human-readable description of the option which will be displayed in the completion menu.

    • completer can be set to a function to generate possible completions for the option argument. The function receives as argument the element at that position and return zero or more candidates.

  • $arg-handlers is an array of functions, each one returning the possible completions for that position in the arguments. Each function receives as argument the last element of $args, and should return zero or more possible values for the completions at that point. The returned values can be plain strings or the output of edit:complex-candidate. If the last element of the list is the string ..., then the last handler is reused for all following arguments.


~> fn complete {|@args|
     opt-specs = [ [&short=a &long=all &desc="Show all"]
                   [&short=n &desc="Set name" &arg-required=$true
                    &completer= {|_| put name1 name2 }] ]
     arg-handlers = [ {|_| put first1 first2 }
                      {|_| put second1 second2 } ... ]
     edit:complete-getopt $args $opt-specs $arg-handlers
~> complete ''
▶ first1
▶ first2
~> complete '-'
▶ (edit:complex-candidate -a &display='-a (Show all)')
▶ (edit:complex-candidate --all &display='--all (Show all)')
▶ (edit:complex-candidate -n &display='-n (Set name)')
~> complete -n ''
▶ name1
▶ name2
~> complete -a ''
▶ first1
▶ first2
~> complete arg1 ''
▶ second1
▶ second2
~> complete arg1 arg2 ''
▶ second1
▶ second2

See also flag:parse-getopt.



Starts the completion mode after accepting any pending autofix.

If all the candidates share a non-empty prefix and that prefix starts with the seed, inserts the prefix instead.



Start the completion mode.


edit:complex-candidate $stem &display='' &code-suffix=''

Builds a complex candidate. This is mainly useful in argument completers.

The &display option controls how the candidate is shown in the UI. It can be a string or a styled text. If it is empty, $stem is used.

The &code-suffix option affects how the candidate is inserted into the code when it is accepted. By default, a quoted version of $stem is inserted. If $code-suffix is non-empty, it is added to that text, and the suffix is not quoted.


edit:del-var $name

Deletes a variable from the interactive REPL if it exists.

Equivalent to running del $name at a REPL prompt, but $name can be dynamic, and it is not an error to delete a non-existing variable.


edit:del-vars $list

Deletes variables from the interactive REPL.

Equivalent to calling edit:del-var for each element of the list, but guarantees that all the variables will be deleted at the same time.



Adds a notification saying “End of history”.



Starts the history listing mode.



Toggles deduplication in history listing mode.

When deduplication is on (the default), only the last occurrence of the same command is shown.



Replaces the content of the buffer with the current history mode entry, and closes history mode.



Walks to the next entry in history mode.



Walks to the next entry in history mode, or quit the history mode if already at the newest entry.



Import command history entries that happened after the current session started.



Starts the history mode.



Walks to the previous entry in history mode.


edit:insert-at-dot $text

Inserts the given text at the dot, moving the dot after the newly inserted text.



Inserts the last word of the last command.



Requests the next terminal input to be inserted uninterpreted.


edit:key $string

Parses a string into a key.



Deletes the last alnum word to the left of the dot.



Deletes the first alnum word to the right of the dot.



Deletes the text between the dot and the start of the current line.



Deletes the text between the dot and the end of the current line.



Kills one rune left of the dot. Does nothing if the dot is at the beginning of the buffer.



Kills one rune right of the dot. Does nothing if the dot is at the end of the buffer.



Deletes the last small word to the left of the dot.



Deletes the first small word to the right of the dot.



Deletes the last word to the left of the dot.



Deletes the first word to the right of the dot.



Starts the last command mode.



Accepts the current selected listing item.



Moves the cursor down in listing mode.



Moves the cursor down in listing mode, or to the first item if the last item is currently selected.



Moves the cursor down one page.



Moves the cursor up one page.


edit:listing:start-custom $items &binding=$nil &caption='' &keep-bottom=$false &accept=$nil &auto-accept=$false

Starts custom listing mode.

The $items argument can be as a list of maps, each map representing one item and having the following keys:

  • The value of the to-show key must be a string or a styled text. It is used in the listing UI.

  • The value of the to-filter key must be a string. It is used when filtering the item.

  • The value of the to-accept key must be a string. It is passed to the accept callback (see below).

Alternatively, the $items argument can be a function taking one argument. It will be called with the value of the filter (initially an empty string), and can output any number of maps containing the to-show and to-accept keys, with the same semantics as above. Any other key is ignored.

The &binding option, if specified, should be a binding map to use in the custom listing mode. Bindings from $edit:listing:binding are also used, after this map if it is specified.

The &caption option changes the caption of the mode. If empty, the caption defaults to ' LISTING '.

The &keep-bottom option, if true, makes the last item to get selected initially or when the filter changes.

The &accept option specifies a function to call when an item is accepted. It is passed the value of the to-accept key of the item.

The &auto-accept option, if true, accepts an item automatically when there is only one item being shown.



Moves the cursor up in listing mode.



Moves the cursor up in listing mode, or to the last item if the first item is currently selected.


edit:match-prefix $seed $inputs?

For each input, outputs whether the input has $seed as a prefix. Uses the result of to-string for non-string inputs.

Roughly equivalent to the following Elvish function, but more efficient:

use str
fn match-prefix {|seed @input|
  each {|x| str:has-prefix (to-string $x) $seed } $@input


edit:match-subseq $seed $inputs?

For each input, outputs whether the input has $seed as a subsequence. Uses the result of to-string for non-string inputs.


edit:match-substr $seed $inputs?

For each input, outputs whether the input has $seed as a substring. Uses the result of to-string for non-string inputs.

Roughly equivalent to the following Elvish function, but more efficient:

use str
fn match-substr {|seed @input|
  each {|x| str:has-contains (to-string $x) $seed } $@input



Moves the dot down one line, trying to preserve the visual horizontal position. Does nothing if dot is already on the last line of the buffer.



Moves the dot to the end of the current line.



Moves the dot left one rune. Does nothing if the dot is at the beginning of the buffer.



Moves the dot to the beginning of the last alnum word to the left of the dot.



Moves the dot to the beginning of the last small word to the left of the dot.



Moves the dot to the beginning of the last word to the left of the dot.



Moves the dot right one rune. Does nothing if the dot is at the end of the buffer.



Moves the dot to the beginning of the first alnum word to the right of the dot.



Moves the dot to the beginning of the first small word to the right of the dot.



Moves the dot to the beginning of the first word to the right of the dot.



Moves the dot to the start of the current line.



Moves the dot up one line, trying to preserve the visual horizontal position. Does nothing if dot is already on the first line of the buffer.



Inserts the selected filename.



Inserts the selected filename and closes the navigation addon.



Start the navigation mode.



Toggles the filtering status of the navigation addon.



Toggles whether the navigation addon should be showing hidden files.


edit:notify $message

Prints a notification message. The argument may be a string or a styled text.

If called while the editor is active, this will print the message above the editor, and redraw the editor.

If called while the editor is inactive, the message will be queued, and shown once the editor becomes active.


edit:redraw &full=$false

Triggers a redraw.

The &full option controls whether to do a full redraw. By default, all redraws performed by the line editor are incremental redraws, updating only the part of the screen that has changed from the last redraw. A full redraw updates the entire command line.


edit:replace-input $text

Equivalent to assigning $text to $edit:current-command.



Causes the Elvish REPL to terminate. If called from a key binding, takes effect after the key binding returns.



Causes the Elvish REPL to end the current read iteration and evaluate the code it just read. If called from a key binding, takes effect after the key binding returns.



If the current code is syntactically incomplete (like echo [), inserts a literal newline.

Otherwise, applies any pending autofixes and accepts the current line.



Toggles the value of [$edit:insert:quote-paste].



Swaps the alnum words to the left and right of the dot. If the dot is at the beginning of the buffer, it swaps the first two alnum words, and if the dot is at the end, it swaps the last two.



Swaps the runes to the left and right of the dot. If the dot is at the beginning of the buffer, swaps the first two runes, and if the dot is at the end, it swaps the last two.



Swaps the small words to the left and right of the dot. If the dot is at the beginning of the buffer, it swaps the first two small words, and if the dot is at the end, it swaps the last two.



Swaps the words to the left and right of the dot. If the dot is at the beginning of the buffer, swaps the first two words, and the dot is at the end, it swaps the last two.


edit:wordify $code

Breaks Elvish code into words.