Builtin Functions and Variables


The builtin module contains facilities that are potentially useful to all users.

Using builtin: explicitly

The builtin module is consulted implicitly when resolving unqualified names, and Elvish’s namespacing mechanism makes it impossible for other modules to redefine builtin symbols. It’s almost always sufficient (and safe) to use builtin functions and variables with their unqualified names.

Nonetheless, the builtin module is also available as a pre-defined module. It can be imported with use builtin, which makes all the builtin symbols available under the builtin: namespace. This can be useful in several cases:

  • To refer to a builtin function when it is shadowed locally. This is especially useful when the function that shadows the builtin one is a wrapper:

    use builtin
    fn cd {|@args|
        echo running my cd function
        builtin:cd $@args

    Note that the shadowing of cd is only in effect in the local lexical scope.

  • To introspect the builtin module, for example keys $builtin:.

Usage Notation

The usage of a builtin command is described by giving an example usage, using variables as arguments. For instance, the repeat command takes two arguments and is described as:

repeat $n $v

Optional arguments are represented with a trailing ?, while variadic arguments with a trailing .... For instance, the count command takes an optional list:

count $inputs?

While the put command takes an arbitrary number of arguments:

put $values...

Options are given along with their default values. For instance, the echo command takes a sep option and arbitrary arguments:

echo &sep=' ' $value...

(When you call functions, options are always optional.)

Commands taking value inputs

Most commands that take value inputs (e.g. count, each) can take the inputs in one of two ways:

  1. From the pipeline:

    ~> put lorem ipsum | count # count number of inputs
    ~> put 10 100 | each {|x| + 1 $x } # apply function to each input
    ▶ (num 11)
    ▶ (num 101)

    If the previous command outputs bytes, one line becomes one string input, as if there is an implicit from-lines (this behavior is subject to change):

    ~> print "a\nb\nc\n" | count # count number of lines
    ▶ 3
    ~> use str
    ~> print "a\nb\nc\n" | each $str:to-upper~ # apply to each line
    ▶ A
    ▶ B
    ▶ C
  2. From an argument – an iterable value:

    ~> count [lorem ipsum] # count number of elements in argument
    ~> each {|x| + 1 $x } [10 100] # apply to each element in argument
    ▶ 11
    ▶ 101

    Strings, and in future, other sequence types are also supported:

    ~> count lorem
    ▶ 5

When documenting such commands, the optional argument is always written as $inputs?.

Note: You should prefer the first form, unless using it requires explicit put commands. Avoid count [(some-command)] or each $some-func [(some-command)]; they are equivalent to some-command | count or some-command | each $some-func.

Rationale: An alternative way to design this is to make (say) count take an arbitrary number of arguments, and count its arguments; when there is 0 argument, count inputs. However, this leads to problems in code like count *; the intention is clearly to count the number of files in the current directory, but when the current directory is empty, count will wait for inputs. Hence it is required to put the input in a list: count [*] unambiguously supplies input in the argument, even if there is no file.

Numeric commands

Wherever a command expects a number argument, that argument can be supplied either with a typed number or a string that can be converted to a number. This includes numeric comparison commands like ==.

When a command outputs numbers, it always outputs a typed number.


~> + 2 10
▶ (num 12)
~> == 2 (num 2)
▶ $true

Exactness-preserving commands

Some numeric commands are designated exactness-preserving. When such commands are called with only exact numbers (i.e. integers or rationals), they will always output an exact number. Examples:

~> + 10 1/10
▶ (num 101/10)
~> * 12 5/17
▶ (num 60/17)

If the condition above is not satisfied – i.e. when a numeric command is not designated exactness-preserving, or when at least one of the arguments is inexact (i.e. a floating-point number), the result is an inexact number, unless otherwise documented. Examples:

~> + 10 0.1
▶ (num 10.1)
~> + 10 1e1
▶ (num 20.0)
~> use math
~> math:sin 1
▶ (num 0.8414709848078965)

There are some cases where the result is exact despite the use of inexact arguments or non-exactness-preserving commands. Such cases are always documented in their respective commands.

Unstable features

The name of some variables and functions have a leading -. This is a convention to say that it is subject to change and should not be depended upon. They are either only useful for debug purposes, or have known issues in the interface or implementation, and in the worst case will make Elvish crash. (Before 1.0, all features are subject to change, but those ones are sure to be changed.)



A blackhole variable.

Values assigned to it will be discarded. Referencing it always results in $nil.


A list of functions to run after changing directory. These functions are always called with directory to change it, which might be a relative path. The following example also shows $before-chdir:

~> set before-chdir = [{|dir| echo "Going to change to "$dir", pwd is "$pwd }]
~> set after-chdir = [{|dir| echo "Changed to "$dir", pwd is "$pwd }]
~> cd /usr
Going to change to /usr, pwd is /Users/xiaq
Changed to /usr, pwd is /usr
/usr> cd local
Going to change to local, pwd is /usr
Changed to local, pwd is /usr/local

Note: The use of echo above is for illustrative purposes. When Elvish is used interactively, the working directory may be changed in location mode or navigation mode, and outputs from echo can garble the terminal. If you are writing a plugin that works with the interactive mode, it’s better to use edit:notify.

See also $before-chdir.


A list containing command-line arguments. Analogous to argv in some other languages. Examples:

~> echo 'put $args' > args.elv
~> elvish args.elv foo -bar
▶ [foo -bar]
~> elvish -c 'put $args' foo -bar
▶ [foo -bar]

As demonstrated above, this variable does not contain the name of the script used to invoke it. For that information, use the src command.

See also src.


A list of functions to run before changing directory. These functions are always called with the new working directory.

See also $after-chdir.


A map that exposes information about the Elvish binary. Running put $buildinfo | to-json will produce the same output as elvish -buildinfo -json.

See also $version.


The boolean false value.


A special value useful for representing the lack of values.


Whether to notify success of background jobs, defaulting to $true.

Failures of background jobs are always notified.


Number of background jobs.


The special value used by ?() to signal absence of exceptions.


A list of search paths, kept in sync with $E:PATH. It is easier to use than $E:PATH.


The process ID of the current Elvish process.


The present working directory. Setting this variable has the same effect as cd. This variable is most useful in a temporary assignment.


## Updates all git repositories
use path
for x [*/] {
  tmp pwd = $x
  if (path:is-dir .git) {
    git pull

Etymology: the pwd command.

See also cd.


The boolean true value.


A string put before value outputs (such as those of put). Defaults to '▶ '. Example:

~> put lorem ipsum
▶ lorem
▶ ipsum
~> set value-out-indicator = 'val> '
~> put lorem ipsum
val> lorem
val> ipsum

Note that you almost always want some trailing whitespace for readability.


The full version of the Elvish binary as a string. This is the same information reported by elvish -version and the value of $buildinfo[version].

Note: In general it is better to perform functionality tests rather than testing $version. For example, do something like

has-key $builtin: new-var

to test if variable new-var is available rather than comparing against $version to see if the elvish version is equal to or newer than the version that introduced new-var.

See also $buildinfo.



+ $num...

Outputs the sum of all arguments, or 0 when there are no arguments.

This command is exactness-preserving.


~> + 5 2 7
▶ (num 14)
~> + 1/2 1/3 1/4
▶ (num 13/12)
~> + 1/2 0.5
▶ (num 1.0)


- $x-num $y-num...

Outputs the result of subtracting from $x-num all the $y-nums, working from left to right. When no $y-num is given, outputs the negation of $x-num instead (in other words, - $x-num is equivalent to - 0 $x-num).

This command is exactness-preserving.


~> - 5
▶ (num -5)
~> - 5 2
▶ (num 3)
~> - 5 2 7
▶ (num -4)
~> - 1/2 1/3
▶ (num 1/6)
~> - 1/2 0.3
▶ (num 0.2)
~> - 10
▶ (num -10)


* $num...

Outputs the product of all arguments, or 1 when there are no arguments.

This command is exactness-preserving. Additionally, when any argument is exact 0 and no other argument is a floating-point infinity, the result is exact 0.


~> * 2 5 7
▶ (num 70)
~> * 1/2 0.5
▶ (num 0.25)
~> * 0 0.5
▶ (num 0)


/ $x-num $y-num...

Outputs the result of dividing $x-num with all the $y-nums, working from left to right. When no $y-num is given, outputs the reciprocal of $x-num instead (in other words, / $y-num is equivalent to / 1 $y-num).

Dividing by exact 0 raises an exception. Dividing by inexact 0 results with either infinity or NaN according to floating-point semantics.

This command is exactness-preserving. Additionally, when $x-num is exact 0 and no $y-num is exact 0, the result is exact 0.


~> / 2
▶ (num 1/2)
~> / 2.0
▶ (num 0.5)
~> / 10 5
▶ (num 2)
~> / 2 5
▶ (num 2/5)
~> / 2 5 7
▶ (num 2/35)
~> / 0 1.0
▶ (num 0)
~> / 2 0
Exception: bad value: divisor must be number other than exact 0, but is exact 0
[tty 6], line 1: / 2 0
~> / 2 0.0
▶ (num +Inf)

When given no argument, this command is equivalent to cd /, due to the implicit cd feature. (The implicit cd feature will probably change to avoid this oddity).


% $x $y

Outputs the remainder after dividing $x by $y. The result has the same sign as $x.


~> % 10 3
▶ (num 1)
~> % -10 3
▶ (num -1)
~> % 10 -3
▶ (num 1)

Note that % requires both arguments to be within the range of signed integers the size of a machine word, and throws an exception otherwise:

~> % (math:pow 2 63) 3
Exception: wrong type for arg #0: must be integer

This limit may be lifted in the future.

< <= == != > >=

<  $number... # less
<= $number... # less or equal
== $number... # equal
!= $number... # not equal
>  $number... # greater
>= $number... # greater or equal

Number comparisons. All of them accept an arbitrary number of arguments:

  1. When given fewer than two arguments, all output $true.

  2. When given two arguments, output whether the two arguments satisfy the named relationship.

  3. When given more than two arguments, output whether every adjacent pair of numbers satisfy the named relationship.


~> == 3 3.0
▶ $true
~> < 3 4
▶ $true
~> < 3 4 10
▶ $true
~> < 6 9 1
▶ $false

As a consequence of rule 3, the != command outputs $true as long as any adjacent pair of numbers are not equal, even if some numbers that are not adjacent are equal:

~> != 5 5 4
▶ $false
~> != 5 6 5
▶ $true

<s <=s ==s !=s >s >=s

<s  $string... # less
<=s $string... # less or equal
==s $string... # equal
!=s $string... # not equal
>s  $string... # greater
>=s $string... # greater or equal

String comparisons. They behave similarly to their number counterparts when given multiple arguments. Examples:

~> >s lorem ipsum
▶ $true
~> ==s 1 1.0
▶ $false
~> >s 8 12
▶ $true


all $inputs?

Takes value inputs, and outputs those values unchanged.

This is an identity function for the value channel; in other words, a | all is equivalent to just a if a only has value output.

This command can be used inside output capture (i.e. (all)) to turn value inputs into arguments. For example:

~> echo '["foo","bar"] ["lorem","ipsum"]' | from-json
▶ [foo bar]
▶ [lorem ipsum]
~> echo '["foo","bar"] ["lorem","ipsum"]' | from-json | put (all)[0]
▶ foo
▶ lorem

The latter pipeline is equivalent to the following:

~> put (echo '["foo","bar"] ["lorem","ipsum"]' | from-json)[0]
▶ foo
▶ lorem

In general, when (all) appears in the last command of a pipeline, it is equivalent to just moving the previous commands of the pipeline into (). The choice is a stylistic one; the (all) variant is longer overall, but can be more readable since it allows you to avoid putting an excessively long pipeline inside an output capture, and keeps the data flow within the pipeline.

Putting the value capture inside [] (i.e. [(all)]) is useful for storing all value inputs in a list for further processing:

~> fn f { var inputs = [(all)]; put $inputs[1] }
~> put foo bar baz | f
▶ bar

The all command can also take “inputs” from an iterable argument. This can be used to flatten lists or strings (although not recursively):

~> all [foo [lorem ipsum]]
▶ foo
▶ [lorem ipsum]
~> all foo
▶ f
▶ o
▶ o

This can be used together with (one) to turn a single iterable value in the pipeline into its elements:

~> echo '["foo","bar","lorem"]' | from-json
▶ [foo bar lorem]
~> echo '["foo","bar","lorem"]' | from-json | all (one)
▶ foo
▶ bar
▶ lorem

When given byte inputs, the all command currently functions like from-lines, although this behavior is subject to change:

~> print "foo\nbar\n" | all
▶ foo
▶ bar

See also one.


assoc $container $k $v

Output a slightly modified version of $container, such that its value at $k is $v. Applies to both lists and to maps.

When $container is a list, $k may be a negative index. However, slice is not yet supported.

~> assoc [foo bar quux] 0 lorem
▶ [lorem bar quux]
~> assoc [foo bar quux] -1 ipsum
▶ [foo bar ipsum]
~> assoc [&k=v] k v2
▶ [&k=v2]
~> assoc [&k=v] k2 v2
▶ [&k2=v2 &k=v]

Etymology: Clojure.

See also dissoc.


base $base $number...

Outputs a string for each $number written in $base. The $base must be between 2 and 36, inclusive. Examples:

~> base 2 1 3 4 16 255
▶ 1
▶ 11
▶ 100
▶ 10000
▶ 11111111
~> base 16 1 3 4 16 255
▶ 1
▶ 3
▶ 4
▶ 10
▶ ff


benchmark &min-runs=5 &min-time=1s &on-end=$nil &on-run-end=$nil $callable

Runs $callable repeatedly, and reports statistics about how long each run takes.

If the &on-end callback is not given, benchmark prints the average, standard deviation, minimum and maximum of the time it took to run $callback, and the number of runs. If the &on-end callback is given, benchmark instead calls it with a map containing these metrics, keyed by avg, stddev, min, max and runs. Each duration value (i.e. all except runs) is given as the number of seconds.

The number of runs is controlled by &min-runs and &min-time. The $callable is run at least &min-runs times, and when the total duration is at least &min-time.

The &min-runs option must be a non-negative integer within the range of the machine word.

The &min-time option must be a string representing a non-negative duration, specified as a sequence of decimal numbers with a unit suffix (the numbers may have fractional parts), such as “300ms”, “1.5h” and “1h45m7s”. Valid time units are “ns”, “us” (or “µs”), “ms”, “s”, “m”, “h”.

If &on-run-end is given, it is called after each call to $callable, with the time that call took, given as the number of seconds.

If $callable throws an exception, benchmark terminates and propagates the exception after the &on-end callback (or the default printing behavior) finishes. The duration of the call that throws an exception is not passed to &on-run-end, nor is it included when calculating the statistics for &on-end. If the first call to $callable throws an exception and &on-end is $nil, nothing is printed and any &on-end callback is not called.

If &on-run-end is given and throws an exception, benchmark terminates and propagates the exception after the &on-end callback (or the default printing behavior) finishes, unless $callable has already thrown an exception

If &on-end throws an exception, the exception is propagated, unless $callable or &on-run-end has already thrown an exception.


~> benchmark { }
98ns ± 382ns (min 0s, max 210.417µs, 10119226 runs)
~> benchmark &on-end={|m| put $m[avg]} { }
▶ (num 9.8e-08)
~> benchmark &on-run-end={|d| echo $d} { sleep 0.3 }
301.088324ms ± 234.298µs (min 300.629166ms, max 301.260333ms, 5 runs)

See also time.


bool $value

Convert a value to boolean. In Elvish, only $false and errors are booleanly false. Everything else, including 0, empty strings and empty lists, is booleanly true:

~> bool $true
▶ $true
~> bool $false
▶ $false
~> bool $ok
▶ $true
~> bool ?(fail haha)
▶ $false
~> bool ''
▶ $true
~> bool []
▶ $true
~> bool abc
▶ $true

See also not.



Raises the special “break” exception. When raised inside a loop it is captured and causes the loop to terminate.

Because break raises an exception it can be caught by a try block. If not caught, either implicitly by a loop or explicitly, it causes a failure like any other uncaught exception.

See the discussion about flow commands and exceptions

Note: You can create a break function and it will shadow the builtin command. If you do so you should explicitly invoke the builtin. For example:

~> use builtin
~> fn break { put 'break'; builtin:break; put 'should not appear' }
~> for x [a b c] { put $x; break; put 'unexpected' }
▶ a
▶ break


call $fn $args $opts

Calls $fn with $args as the arguments, and $opts as the option. Useful for calling a function with dynamic option keys.


~> var f = {|a &k1=v1 &k2=v2| put $a $k1 $k2 }
~> call $f [foo] [&k1=bar]
▶ foo
▶ bar
▶ v2


cd $dirname

Changes directory.

This affects the entire process, including parallel tasks that are started implicitly (such as prompt functions) or explicitly (such as one started by peach).

Note that Elvish’s cd does not support cd -.

See also $pwd.


compact $inputs?

Replaces consecutive runs of equal values with a single copy. Similar to the uniq command on Unix.


~> put a a b b c | compact
▶ a
▶ b
▶ c
~> compact [a a b b c]
▶ a
▶ b
▶ c
~> put a b a | compact
▶ a
▶ b
▶ a


compare &total=$false $a $b

Outputs the number -1 if $a is smaller than $b, 0 if $a is equal to $b, and 1 if $a is greater than $b.

The following algorithm is used:

  1. If $a and $b have the same type and that type is listed below, they are compared accordingly:

    • Booleans: $false is smaller than $true.

    • Typed numbers: Compared numerically, consistent with the number comparison commands, except that NaN values are considered equal to each other and smaller than all other numbers.

    • Strings: Compared lexicographically by bytes, consistent with the string comparison commands. For UTF-8 encoded strings, this is equivalent to comparing by codepoints.

      Beware that strings that look like numbers are compared as strings, not numbers.

    • Lists: Compared lexicographically by elements, with elements compared recursively.

  2. If eq $a $b is true, compare $a $b outputs the number 0.

  3. Otherwise the behavior depends on the &total option:

    • If it is $false (the default), compare throws an exception complaning that the two values can’t be compared.

    • If it is $true, compare compares the types of $a and $b: if they have the same type, it outputs 0; if they have different types, it outputs -1 and 1 depending on which type comes first in an internal ordering of all types.

      The internal ordering of all types is unspecified, but it is guaranteed to be consistent during the same Elvish session. For example, if compare &total $a $b outputs -1 when $a is a number and $b is a string, it will always output -1 for such pairs.

      This creates an artificial total order, which is mainly useful for sorting values of mixed types.


~> compare a b
▶ (num 1)
~> compare b a
▶ (num -1)
~> compare x x
▶ (num 0)
~> compare (num 10) (num 1)
▶ (num 1)
~> compare a (num 10)
Exception: bad value: inputs to "compare" or "order" must be comparable values, but is uncomparable values
[tty 3]:1:1: compare a (num 10)
~> compare &total a (num 10)
▶ (num 1)
~> compare &total (num 10) a
▶ (num -1)

See also order.


conj $list $more...

Outputs a list created from adding values in $more to the end of $list.

The output is the same as [$@list $more...], but the time complexity is guaranteed to be O(m), where m is the number of values in $more.


~> conj [] a
▶ [a]
~> conj [a b] c d
▶ [a b c d]

Etymology: Clojure.


constantly $value...

Output a function that takes no arguments and outputs $values when called. Examples:

~> var f = (constantly lorem ipsum)
~> $f
▶ lorem
▶ ipsum

The above example is equivalent to simply var f = { put lorem ipsum }; it is most useful when the argument is not a literal value, e.g.

~> var f = (constantly (uname))
~> $f
▶ Darwin
~> $f
▶ Darwin

The above code only calls uname once when defining $f. In contrast, if $f is defined as var f = { put (uname) }, every time you invoke $f, uname will be called.

Etymology: Clojure.



Raises the special “continue” exception. When raised inside a loop it is captured and causes the loop to begin its next iteration.

Because continue raises an exception it can be caught by a try block. If not caught, either implicitly by a loop or explicitly, it causes a failure like any other uncaught exception.

See the discussion about flow commands and exceptions

Note: You can create a continue function and it will shadow the builtin command. If you do so you should explicitly invoke the builtin. For example:

~> use builtin
~> fn continue { put 'continue'; builtin:continue; put 'should not appear' }
~> for x [a b c] { put $x; continue; put 'unexpected' }
▶ a
▶ continue
▶ b
▶ continue
▶ c
▶ continue


count $input-list?

Count the number of inputs.


~> count lorem # count bytes in a string
▶ 5
~> count [lorem ipsum]
▶ 2
~> range 100 | count
▶ 100
~> seq 100 | count
▶ 100


defer $fn

Schedules a function to be called when execution reaches the end of the current closure. The function is called with no arguments or options, and any exception it throws gets propagated.


~> { defer { put foo }; put bar }
▶ bar
▶ foo
~> defer { put foo }
Exception: defer must be called from within a closure
[tty 2], line 1: defer { put foo }


deprecate $msg

Shows the given deprecation message to stderr. If called from a function or module, also shows the call site of the function or import site of the module. Does nothing if the combination of the call site and the message has been shown before.

~> deprecate msg
deprecation: msg
~> fn f { deprecate msg }
~> f
deprecation: msg
[tty 3], line 1: f
~> f # a different call site; shows deprecate message
deprecation: msg
[tty 4], line 1: f
~> fn g { f }
~> g
deprecation: msg
[tty 5], line 1: fn g { f }
~> g # same call site, no more deprecation message


dissoc $map $k

Output a slightly modified version of $map, with the key $k removed. If $map does not contain $k as a key, the same map is returned.

~> dissoc [&foo=bar &lorem=ipsum] foo
▶ [&lorem=ipsum]
~> dissoc [&foo=bar &lorem=ipsum] k
▶ [&lorem=ipsum &foo=bar]

See also assoc.


drop $n $inputs?

Ignores the first $n value inputs and outputs the rest. If $n is larger than the number of value inputs, outputs nothing.


~> range 10 | drop 8
▶ (num 8)
▶ (num 9)
~> range 2 | drop 10
~> drop 2 [a b c d e]
▶ c
▶ d
▶ e
~> use str
~> str:split ' ' 'how are you?' | drop 1
▶ are
▶ 'you?'

Etymology: Haskell.

See also take.


each $f $inputs?

Calls $f on each value input.

An exception raised from break is caught by each, and will cause it to terminate early.

An exception raised from continue is swallowed and can be used to terminate a single iteration early.


~> range 5 8 | each {|x| * $x $x }
▶ 25
▶ 36
▶ 49
~> each {|x| put $x[:3] } [lorem ipsum]
▶ lor
▶ ips

See also peach.

Etymology: Various languages, as for each. Happens to have the same name as the iteration construct of Factor.


eawk $f $inputs?

For each value input, calls $f with the input followed by all its fields. A break command will cause eawk to stop processing inputs. A continue command will exit $f, but is ignored by eawk.

It should behave the same as the following functions:

fn eawk {|f @rest|
  each {|line|
    var @fields = (re:split '[ \t]+' (str:trim $line " \t"))
    $f $line $@fields
  } $@rest

This command allows you to write code very similar to awk scripts using anonymous functions. Example:

~> echo " lorem ipsum\n1 2" | awk '{ print $1 }'
~> echo " lorem ipsum\n1 2" | eawk {|line a b| put $a }
▶ lorem
▶ 1

Note: Since Elvish allows variable names consisting solely of digits, you can also do the following:

~> echo " lorem ipsum\n1 2" | eawk {|0 1 2| put $1 }
▶ lorem
▶ 1


echo &sep=' ' $value...

Print all arguments, joined by the sep option, and followed by a newline.


~> echo Hello   elvish
Hello elvish
~> echo "Hello   elvish"
Hello   elvish
~> echo &sep=, lorem ipsum

Notes: The echo builtin does not treat -e or -n specially. For instance, echo -n just prints -n. Use double-quoted strings to print special characters, and print to suppress the trailing newline.

See also print.

Etymology: Bourne sh.


eq $values...

Determines whether all $values are equal. Writes $true when given no or one argument.

Two values are equal when they have the same type and value.

For complex data structures like lists and maps, comparison is done recursively. A pseudo-map is equal to another pseudo-map with the same internal type (which is not exposed to Elvish code now) and value.

~> eq a a
▶ $true
~> eq [a] [a]
▶ $true
~> eq [&k=v] [&k=v]
▶ $true
~> eq a [b]
▶ $false

See also is and not-eq.

Etymology: Perl.


eval $code &ns=$nil &on-end=$nil

Evaluates $code, which should be a string. The evaluation happens in a new, restricted namespace, whose initial set of variables can be specified by the &ns option. After evaluation completes, the new namespace is passed to the callback specified by &on-end if it is not nil.

The namespace specified by &ns is never modified; it will not be affected by the creation or deletion of variables by $code. However, the values of the variables may be mutated by $code.

If the &ns option is $nil (the default), a temporary namespace built by amalgamating the local and upvalue scopes of the caller is used.

If $code fails to parse or compile, the parse error or compilation error is raised as an exception.

Basic examples that do not modify the namespace or any variable:

~> eval 'put x'
▶ x
~> var x = foo
~> eval 'put $x'
▶ foo
~> var ns = (ns [&x=bar])
~> eval &ns=$ns 'put $x'
▶ bar

Examples that modify existing variables:

~> var y = foo
~> eval 'set y = bar'
~> put $y
▶ bar

Examples that creates new variables and uses the callback to access it:

~> eval 'var z = lorem'
~> put $z
compilation error: variable $z not found
[ttz 2], line 1: put $z
~> var saved-ns = $nil
~> eval &on-end={|ns| set saved-ns = $ns } 'var z = lorem'
~> put $saved-ns[z]
▶ lorem

Note that when using variables from an outer scope, only those that have been referenced are captured as upvalues (see closure semantics) and thus accessible to eval:

~> var a b
~> fn f {|code| nop $a; eval $code }
~> f 'echo $a'
~> f 'echo $b'
Exception: compilation error: variable $b not found
[eval 2], line 1: echo $b
Traceback: [... omitted ...]


exact-num $string-or-number

Coerces the argument to an exact number. If the argument is infinity or NaN, an exception is thrown.

If the argument is a string, it is converted to a typed number first. If the argument is already an exact number, it is returned as is.


~> exact-num (num 0.125)
▶ (num 1/8)
~> exact-num 0.125
▶ (num 1/8)
~> exact-num (num 1)
▶ (num 1)

Beware that seemingly simple fractions that can’t be represented precisely in binary can result in the denominator being a very large power of 2:

~> exact-num 0.1
▶ (num 3602879701896397/36028797018963968)

See also num and inexact-num.


exec $command? $args...

Replace the Elvish process with an external $command, defaulting to elvish, passing the given arguments. This decrements $E:SHLVL before starting the new process.

This command always raises an exception on Windows with the message “not supported on Windows”.


exit $status?

Exit the Elvish process with $status (defaulting to 0).


external $program

Construct a callable value for the external program $program. Example:

~> var x = (external man)
~> $x ls # opens the manpage for ls

See also has-external and search-external.


fail $v

Throws an exception; $v may be any type. If $v is already an exception, fail rethrows it.

~> fail bad
Exception: bad
[tty 9], line 1: fail bad
~> put ?(fail bad)
▶ [&reason=[&content=bad &type=fail]]
~> fn f { fail bad }
~> fail ?(f)
Exception: bad
  [tty 7], line 1:
    fn f { fail bad }
  [tty 8], line 1:
    fail ?(f)


float64 $string-or-number

Constructs a floating-point number.

This command is deprecated; use num to construct a typed number, or inexact-num to construct an inexact number.



Takes bytes stdin, parses it as JSON and puts the result on structured stdout. The input can contain multiple JSONs, and whitespace between them are ignored.

Numbers in JSON are parsed as follows:

  • Numbers without fractional parts are parsed as exact integers, and arbitrary precision is supported.

  • Numbers with fractional parts (even if it’s .0) are parsed as inexact floating-point numbers, and the parsing may fail if the number can’t be represented.


~> echo '"a"' | from-json
▶ a
~> echo '["lorem", "ipsum"]' | from-json
▶ [lorem ipsum]
~> echo '{"lorem": "ipsum"}' | from-json
▶ [&lorem=ipsum]
~> # multiple JSONs running together
echo '"a""b"["x"]' | from-json
▶ a
▶ b
▶ [x]
~> # multiple JSONs separated by newlines
echo '"a"
{"k": "v"}' | from-json
▶ a
▶ [&k=v]
~> echo '[42, 100000000000000000000, 42.0, 42.2]' | from-json
▶ [(num 42) (num 100000000000000000000) (num 42.0) (num 42.2)]

See also to-json.



Splits byte input into lines, and writes them to the value output. Value input is ignored.

~> { echo a; echo b } | from-lines
▶ a
▶ b
~> { echo a; put b } | from-lines
▶ a

See also from-terminated, read-upto, and to-lines.


from-terminated $terminator

Splits byte input into lines at each $terminator character, and writes them to the value output. If the byte input ends with $terminator, it is dropped. Value input is ignored.

The $terminator must be a single ASCII character such as "\x00" (NUL).

~> { echo a; echo b } | from-terminated "\x00"
▶ "a\nb\n"
~> print "a\x00b" | from-terminated "\x00"
▶ a
▶ b
~> print "a\x00b\x00" | from-terminated "\x00"
▶ a
▶ b

See also from-lines, read-upto, and to-terminated.



Force the Go garbage collector to run.

This is only useful for debug purposes.


get-env $name

Gets the value of an environment variable. Throws an exception if the environment variable does not exist.

Calling get-env VAR_NAME is similar to put $E:VAR_NAME, but allows the variable name to be dynamic, and throws an exception instead of producing an empty string for nonexistent environment variables.


~> get-env LANG
▶ zh_CN.UTF-8
~> get-env NO_SUCH_ENV
Exception: non-existent environment variable
[tty], line 1: get-env NO_SUCH_ENV

See also has-env, set-env, and unset-env.


has-env $name

Test whether an environment variable exists. This command has no equivalent operation using the E: namespace (but see https://b.elv.sh/1026).


~> has-env PATH
▶ $true
~> has-env NO_SUCH_ENV
▶ $false

See also get-env, set-env, and unset-env.


has-external $command

Test whether $command names a valid external command. Examples (your output might differ):

~> has-external cat
▶ $true
~> has-external lalala
▶ $false

See also external and search-external.


has-key $container $key

Determine whether $key is a key in $container. A key could be a map key or an index on a list or string. This includes a range of indexes.

Examples, maps:

~> has-key [&k1=v1 &k2=v2] k1
▶ $true
~> has-key [&k1=v1 &k2=v2] v1
▶ $false

Examples, lists:

~> has-key [v1 v2] 0
▶ $true
~> has-key [v1 v2] 1
▶ $true
~> has-key [v1 v2] 2
▶ $false
~> has-key [v1 v2] 0..2
▶ $true
~> has-key [v1 v2] 0..3
▶ $false

Examples, strings:

~> has-key ab 0
▶ $true
~> has-key ab 1
▶ $true
~> has-key ab 2
▶ $false
~> has-key ab 0:2
▶ $true
~> has-key ab 0:3
▶ $false


has-value $container $value

Determine whether $value is a value in $container.

Examples, maps:

~> has-value [&k1=v1 &k2=v2] v1
▶ $true
~> has-value [&k1=v1 &k2=v2] k1
▶ $false

Examples, lists:

~> has-value [v1 v2] v1
▶ $true
~> has-value [v1 v2] k1
▶ $false

Examples, strings:

~> has-value ab b
▶ $true
~> has-value ab c
▶ $false



Output all IP addresses of the current host.

This should be part of a networking module instead of the builtin module.


inexact-num $string-or-number

Coerces the argument to an inexact number.

If the argument is a string, it is converted to a typed number first. If the argument is already an inexact number, it is returned as is.


~> inexact-num (num 1)
▶ (num 1.0)
~> inexact-num (num 0.5)
▶ (num 0.5)
~> inexact-num (num 1/2)
▶ (num 0.5)
~> inexact-num 1/2
▶ (num 0.5)

Since the underlying representation for inexact numbers has limited range, numbers with very large magnitudes may be converted to an infinite value:

~> inexact-num 1000000000000000000
▶ (num 1e+18)
~> inexact-num 10000000000000000000
▶ (num +Inf)
~> inexact-num -10000000000000000000
▶ (num -Inf)

Likewise, numbers with very small magnitudes may be converted to 0:

~> use math
~> math:pow 10 -323
▶ (num 1/100000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000)
~> inexact-num (math:pow 10 -323)
▶ (num 1e-323)
~> math:pow 10 -324
▶ (num 1/1000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000)
~> inexact-num (math:pow 10 -324)
▶ (num 0.0)

See also num and exact-num.


is $values...

Determine whether all $values have the same identity. Writes $true when given no or one argument.

The definition of identity is subject to change. Do not rely on its behavior.

~> is a a
▶ $true
~> is a b
▶ $false
~> is [] []
▶ $true
~> is [a] [a]
▶ $false

See also eq.

Etymology: Python.


keys $map

Put all keys of $map on the structured stdout.


~> keys [&a=foo &b=bar &c=baz]
▶ a
▶ c
▶ b

Note that there is no guaranteed order for the keys of a map.


kind-of $value...

Output the kinds of $values. Example:

~> kind-of lorem [] [&]
▶ string
▶ list
▶ map

The terminology and definition of “kind” is subject to change.


-log $filename

Direct internal debug logs to the named file.

This is only useful for debug purposes.


make-map $input?

Outputs a map from the value inputs, each of which must be an iterable value with with two elements. The first element of each value is used as the key, and the second element is used as the value.

If the same key appears multiple times, the last value is used.


~> make-map [[k v]]
▶ [&k=v]
~> make-map [[k v1] [k v2]]
▶ [&k=v2]
~> put [k1 v1] [k2 v2] | make-map
▶ [&k1=v1 &k2=v2]
~> put aA bB | make-map
▶ [&a=A &b=B]


nop &any-opt= @value

Accepts arbitrary arguments and options and does exactly nothing.


~> nop
~> nop a b c
~> nop &k=v

Etymology: Various languages, in particular NOP in assembly languages.


not $value

Boolean negation. Examples:

~> not $true
▶ $false
~> not $false
▶ $true
~> not $ok
▶ $false
~> not ?(fail error)
▶ $true

Note: The related logical commands and and or are implemented as special commands instead, since they do not always evaluate all their arguments. The not command always evaluates its only argument, and is thus a normal command.

See also bool.


not-eq $values...

Determines whether every adjacent pair of $values are not equal. Note that this does not imply that $values are all distinct. Examples:

~> not-eq 1 2 3
▶ $true
~> not-eq 1 2 1
▶ $true
~> not-eq 1 1 2
▶ $false

See also eq.


ns $map

Constructs a namespace from $map, using the keys as variable names and the values as their values. Examples:

~> var n = (ns [&name=value])
~> put $n[name]
▶ value
~> var n: = (ns [&name=value])
~> put $n:name
▶ value


num $string-or-number

Constructs a typed number.

If the argument is a string, this command outputs the typed number the argument represents, or raises an exception if the argument is not a valid representation of a number. If the argument is already a typed number, this command outputs it as is.

This command is usually not needed for working with numbers; see the discussion of numeric commands.


~> num 10
▶ (num 10)
~> num 0x10
▶ (num 16)
~> num 1/12
▶ (num 1/12)
~> num 3.14
▶ (num 3.14)
~> num (num 10)
▶ (num 10)

See also exact-num and inexact-num.


one $inputs?

Takes exactly one value input and outputs it. If there are more than one value inputs, raises an exception.

This function can be used in a similar way to all, but is a better choice when you expect that there is exactly one output.

See also all.



Passes byte input to output, and discards value inputs.


~> { put value; echo bytes } | only-bytes



Passes value input to output, and discards byte inputs.


~> { put value; echo bytes } | only-values
▶ value


order &less-than=$nil &total=$false &key=$nil &reverse=$false $inputs?

Outputs the value inputs after sorting. The sorting process is stable.

By default, order sorts the values in ascending order, using the same comparator as compare, which only supports values of the same ordered type. Its options modify this behavior:

  • The &less-than option, if given, overrides the comparator. Its value should be a function that takes two arguments $a and $b and outputs a boolean indicating whether $a is less than $b. If the function throws an exception, order rethrows the exception without outputting any value.

    The default behavior of order is equivalent to order &less-than={|a b| == -1 (compare $a $b)}.

  • The &total option, if true, overrides the comparator to be same as compare &total=$true, which allows sorting values of mixed types and unordered types. The result groups values by their types. Groups of ordered types are sorted internally, and groups of unordered types retain their original relative order.

    Specifying &total=$true is equivalent to specifying &less-than={|a b| == -1 (compare &total=$true $a $b)}. It is an error to both specify &total=$true and a non-nil &less-than callback.

  • The &key option, if given, is a function that gets called with each input value. It must output a single value, which is used for comparison in place of the original value. The comparator used can be affected by $less-than or &total.

    If the function throws an exception, order rethrows the exception.

    Use of &key can usually be rewritten to use &less-than instead, but using &key can be faster. The &key callback is only called once for each element, whereas the &less-than callback is called O(n*lg(n)) times on average.

  • The &reverse option, if true, reverses the order of output.


~> put foo bar ipsum | order
▶ bar
▶ foo
▶ ipsum
~> order [(num 10) (num 1) (num 5)]
▶ (num 1)
▶ (num 5)
▶ (num 10)
~> order [[a b] [a] [b b] [a c]]
▶ [a]
▶ [a b]
▶ [a c]
▶ [b b]
~> order &reverse [a c b]
▶ c
▶ b
▶ a
~> order [a (num 2) c (num 0) b (num 1)]
Exception: bad value: inputs to "compare" or "order" must be comparable values, but is uncomparable values
[tty 3]:1:1: order [a (num 2) c (num 0) b (num 1)]
~> order &total [a (num 2) c (num 0) b (num 1)]
▶ (num 0)
▶ (num 1)
▶ (num 2)
▶ a
▶ b
▶ c
~> put [0 x] [1 a] [2 b] | order &key={|l| put $l[1]}
▶ [1 a]
▶ [2 b]
▶ [0 x]
~> order &less-than={|a b| eq $a x } [l x o r x e x m]
▶ x
▶ x
▶ x
▶ l
▶ o
▶ r
▶ e
▶ m

Beware that strings that look like numbers are treated as strings, not numbers. To sort strings as numbers, use an explicit &key or &less-than:

~> order [5 1 10]
▶ 1
▶ 10
▶ 5
~> order &key=$num~ [5 1 10]
▶ 1
▶ 5
▶ 10
~> order &less-than=$"<~" [5 1 10]
▶ 1
▶ 5
▶ 10

(The $"<~" syntax is a reference to the < function.)

See also compare.


peach &num-workers=(num +inf) $f $inputs?

Calls $f for each value input, possibly in parallel.

Like each, an exception raised from break will cause peach to terminate early. However due to the parallel nature of peach, the exact time of termination is non-deterministic, and termination is not guaranteed.

An exception raised from continue is swallowed and can be used to terminate a single iteration early.

The &num-workers option restricts the number of functions that may run in parallel, and must be either an exact positive or +inf. A value of +inf (the default) means no restriction. Note that peach &num-workers=1 is equivalent to each.

Example (your output will differ):

~> range 1 10 | peach {|x| + $x 10 }
▶ (num 12)
▶ (num 13)
▶ (num 11)
▶ (num 16)
▶ (num 18)
▶ (num 14)
▶ (num 17)
▶ (num 15)
▶ (num 19)
~> range 1 101 |
   peach {|x| if (== 50 $x) { break } else { put $x } } |
   + (all) # 1+...+49 = 1225; 1+...+100 = 5050
▶ (num 1328)

This command is intended for homogeneous processing of possibly unbound data. If you need to do a fixed number of heterogeneous things in parallel, use run-parallel.

See also each and run-parallel.


pprint $value...

Pretty-print representations of Elvish values. Examples:

~> pprint [foo bar]
~> pprint [&k1=v1 &k2=v2]

The output format is subject to change.

See also repr.


print &sep=' ' $value...

Like echo, just without the newline.

See also echo.

Etymology: Various languages, in particular Perl and zsh, whose prints do not print a trailing newline.


printf $template $value...

Prints values to the byte stream according to a template. If you need to inject the output into the value stream use this pattern: printf .... | slurp. That ensures that any newlines in the output of printf do not cause its output to be broken into multiple values, thus eliminating the newlines, which will occur if you do put (printf ....).

Like print, this command does not add an implicit newline; include an explicit "\n" in the formatting template instead. For example, printf "%.1f\n" (/ 10.0 3).

See Go’s fmt package for details about the formatting verbs and the various flags that modify the default behavior, such as padding and justification.

Unlike Go, each formatting verb has a single associated internal type, and accepts any argument that can reasonably be converted to that type:

  • The verbs %s, %q and %v convert the corresponding argument to a string in different ways:

    • %s uses to-string to convert a value to string.

    • %q uses repr to convert a value to string.

    • %v is equivalent to %s, and %#v is equivalent to %q.

  • The verb %t first convert the corresponding argument to a boolean using bool, and then uses its Go counterpart to format the boolean.

  • The verbs %b, %c, %d, %o, %O, %x, %X and %U first convert the corresponding argument to an integer using an internal algorithm, and use their Go counterparts to format the integer.

  • The verbs %e, %E, %f, %F, %g and %G first convert the corresponding argument to a floating-point number using float64, and then use their Go counterparts to format the number.

The special verb %% prints a literal % and consumes no argument.

Verbs not documented above are not supported.


~> printf "%10s %.2f\n" Pi $math:pi
        Pi 3.14
~> printf "%-10s %.2f %s\n" Pi $math:pi $math:pi
Pi         3.14 3.141592653589793
~> printf "%d\n" 0b11100111
~> printf "%08b\n" 231
~> printf "list is: %q\n" [foo bar 'foo bar']
list is: [foo bar 'foo bar']

Note: Compared to the POSIX printf command found in other shells, there are 3 key differences:

  • The behavior of the formatting verbs are based on Go’s fmt package instead of the POSIX specification.

  • The number of arguments after the formatting template must match the number of formatting verbs. The POSIX command will repeat the template string to consume excess values; this command does not have that behavior.

  • This command does not interpret escape sequences such as \n; just use double-quoted strings.

See also print, echo, pprint, and repr.


put $value...

Takes arbitrary arguments and write them to the structured stdout.


~> put a
▶ a
~> put lorem ipsum [a b] { ls }
▶ lorem
▶ ipsum
▶ [a b]
▶ <closure 0xc4202607e0>

Note: It is almost never necessary to use put (...) - just write the ... part. For example, put (eq a b) is the equivalent to just eq a b.

Etymology: Various languages, in particular C and Ruby as puts.



Output a pseudo-random number in the interval [0, 1). Example:

~> rand
▶ 0.17843564133528436


randint $low? $high

Output a pseudo-random integer N such that $low <= N < $high. If not given, $low defaults to 0. Examples:

~> # Emulate dice
randint 1 7
▶ 6


-randseed $seed

Sets the seed for the random number generator.


range &step $start=0 $end

Outputs numbers, starting from $start and ending before $end, using &step as the increment.

  • If $start <= $end, &step defaults to 1, and range outputs values as long as they are smaller than $end. An exception is thrown if &step is given a negative value.

  • If $start > $end, &step defaults to -1, and range outputs values as long as they are greater than $end. An exception is thrown if &step is given a positive value.

As a special case, if the outputs are floating point numbers, range also terminates if the values stop changing.

This command is exactness-preserving.


~> range 4
▶ (num 0)
▶ (num 1)
▶ (num 2)
▶ (num 3)
~> range 4 0
▶ (num 4)
▶ (num 3)
▶ (num 2)
▶ (num 1)
~> range -3 3 &step=2
▶ (num -3)
▶ (num -1)
▶ (num 1)
~> range 3 -3 &step=-2
▶ (num 3)
▶ (num 1)
▶ (num -1)
~> range (- (math:pow 2 53) 1) +inf
▶ (num 9007199254740991.0)
▶ (num 9007199254740992.0)

When using floating-point numbers, beware that numerical errors can result in an incorrect number of outputs:

~> range 0.9 &step=0.3
▶ (num 0.0)
▶ (num 0.3)
▶ (num 0.6)
▶ (num 0.8999999999999999)

Avoid this problem by using exact rationals:

~> range 9/10 &step=3/10
▶ (num 0)
▶ (num 3/10)
▶ (num 3/5)

One usage of this command is to execute something a fixed number of times by combining with each:

~> range 3 | each {|_| echo foo }

Etymology: Python.


read-bytes $n

Reads $n bytes, or until end-of-file, and outputs the bytes as a string value. The result may not be a valid UTF-8 string.


~> echo "a,b" | read-bytes 2
▶ 'a,'
~> echo "a,b" | read-bytes 10
▶ "a,b\n"



Reads a single line from byte input, and writes the line to the value output, stripping the line ending. A line can end with "\r\n", "\n", or end of file. Examples:

~> print line | read-line
▶ line
~> print "line\n" | read-line
▶ line
~> print "line\r\n" | read-line
▶ line
~> print "line-with-extra-cr\r\r\n" | read-line
▶ "line-with-extra-cr\r"


read-upto $terminator

Reads byte input until $terminator or end-of-file is encountered. It outputs the part of the input read as a string value. The output contains the trailing $terminator, unless read-upto terminated at end-of-file.

The $terminator must be a single ASCII character such as "\x00" (NUL).


~> echo "a,b,c" | read-upto ","
▶ 'a,'
~> echo "foo\nbar" | read-upto "\n"
▶ "foo\n"
~> echo "a.elv\x00b.elv" | read-upto "\x00"
▶ "a.elv\x00"
~> print "foobar" | read-upto "\n"
▶ foobar


repeat $n $value

Output $value for $n times. Example:

~> repeat 0 lorem
~> repeat 4 NAN

Etymology: Clojure.


repr $value...

Writes representation of $values, separated by space and followed by a newline. Example:

~> repr [foo 'lorem ipsum'] "aha\n"
[foo 'lorem ipsum'] "aha\n"

See also pprint.

Etymology: Python.


resolve $command

Output what $command resolves to in symbolic form. Command resolution is described in the language reference.


~> resolve echo
▶ <builtin echo>
~> fn f { }
~> resolve f
▶ <closure 0xc4201c24d0>
~> resolve cat
▶ <external cat>



Raises the special “return” exception. When raised inside a named function (defined by the fn keyword) it is captured by the function and causes the function to terminate. It is not captured by an ordinary anonymous function.

Because return raises an exception it can be caught by a try block. If not caught, either implicitly by a named function or explicitly, it causes a failure like any other uncaught exception.

See the discussion about flow commands and exceptions

Note: If you want to shadow the builtin return function with a local wrapper, do not define it with fn as fn swallows the special exception raised by return. Consider this example:

~> use builtin
~> fn return { put return; builtin:return }
~> fn test-return { put before; return; put after }
~> test-return
▶ before
▶ return
▶ after

Instead, shadow the function by directly assigning to return~:

~> use builtin
~> var return~ = { put return; builtin:return }
~> fn test-return { put before; return; put after }
~> test-return
▶ before
▶ return


run-parallel $callable...

Run several callables in parallel, and wait for all of them to finish.

If one or more callables throw exceptions, the other callables continue running, and a composite exception is thrown when all callables finish execution.

The behavior of run-parallel is consistent with the behavior of pipelines, except that it does not perform any redirections.

Here is an example that lets you pipe the stdout and stderr of a command to two different commands in order to independently capture the output of each byte stream:

~> fn capture {|f|
     var pout = (file:pipe)
     var perr = (file:pipe)
     var out err
     run-parallel {
       $f > $pout[w] 2> $perr[w]
       file:close $pout[w]
       file:close $perr[w]
     } {
       set out = (slurp < $pout[r])
       file:close $pout[r]
     } {
       set err = (slurp < $perr[r])
       file:close $perr[r]
     put $out $err
~> capture { echo stdout-test; echo stderr-test >&2 }
▶ "stdout-test\n"
▶ "stderr-test\n"

This command is intended for doing a fixed number of heterogeneous things in parallel. If you need homogeneous parallel processing of possibly unbound data, use peach instead.

See also peach.


search-external $command

Output the full path of the external $command. Throws an exception when not found. Example (your output might vary):

~> search-external cat
▶ /bin/cat

See also external and has-external.


set-env $name $value

Sets an environment variable to the given value. Calling set-env VAR_NAME value is similar to set E:VAR_NAME = value, but allows the variable name to be dynamic.


~> set-env X foobar
~> put $E:X
▶ foobar

See also get-env, has-env, and unset-env.


show $e

Shows the value to the output, which is assumed to be a VT-100-compatible terminal.

Currently, the only type of value that can be showed is exceptions, but this will likely expand in future.


~> var e = ?(fail lorem-ipsum)
~> show $e
Exception: lorem-ipsum
[tty 3], line 1: var e = ?(fail lorem-ipsum)


sleep $duration

Pauses for at least the specified duration. The actual pause duration depends on the system.

This only affects the current Elvish context. It does not affect any other contexts that might be executing in parallel as a consequence of a command such as peach.

A duration can be a simple number (with optional fractional value) without an explicit unit suffix, with an implicit unit of seconds.

A duration can also be a string written as a sequence of decimal numbers, each with optional fraction, plus a unit suffix. For example, “300ms”, “1.5h” or “1h45m7s”. Valid time units are “ns”, “us” (or “µs”), “ms”, “s”, “m”, “h”.

Passing a negative duration causes an exception; this is different from the typical BSD or GNU sleep command that silently exits with a success status without pausing when given a negative duration.

See the Go documentation for more information about how durations are parsed.


~> sleep 0.1    # sleeps 0.1 seconds
~> sleep 100ms  # sleeps 0.1 seconds
~> sleep 1.5m   # sleeps 1.5 minutes
~> sleep 1m30s  # sleeps 1.5 minutes
~> sleep -1
Exception: sleep duration must be >= zero
[tty 8], line 1: sleep -1



Reads bytes input into a single string, and put this string on structured stdout.


~> echo "a\nb" | slurp
▶ "a\nb\n"

Etymology: Perl, as File::Slurp.



Output a map describing the current source, which is the source file or interactive command that contains the call to src. The value contains the following fields:

  • name, a unique name of the current source. If the source originates from a file, it is the full path of the file.

  • code, the full body of the current source.

  • is-file, whether the source originates from a file.


~> src
▶ [&code=src &is-file=$false &name='[tty 1]']
~> elvish show-src.elv
▶ [&code="src\n" &is-file=$true &name=/home/elf/show-src.elv]
~> echo src > .config/elvish/lib/show-src.elv
~> use show-src
▶ [&code="src\n" &is-file=$true &name=/home/elf/.config/elvish/lib/show-src.elv]



Print a stack trace.

This is only useful for debug purposes.


styled $object $style-transformer...

Constructs a styled text by applying the supplied transformers to the supplied $object, which may be a string, a styled segment, or an existing styled text.

Each $style-transformer can be one of the following:

  • A boolean attribute name:

    • One of bold, dim, italic, underlined, blink and inverse for setting the corresponding attribute.

    • An attribute name prefixed by no- for unsetting the attribute.

    • An attribute name prefixed by toggle- for toggling the attribute between set and unset.

  • A color name for setting the text color, which may be one of the following:

    • One of the 8 basic ANSI colors: black, red, green, yellow, blue, magenta, cyan and white.

    • The bright variant of the 8 basic ANSI colors, with a bright- prefix.

    • Any color from the xterm 256-color palette, as colorX (such as color12).

    • A 24-bit RGB color written as #RRGGBB (such as '#778899').

      Note: You need to quote such values, since an unquoted # introduces a comment (e.g. use 'bg-#778899' instead of bg-#778899).

  • A color name prefixed by fg- to set the foreground color. This has the same effect as specifying the color name without the fg- prefix.

  • A color name prefixed by bg- to set the background color.

  • A function that receives a styled segment as the only argument and outputs a single styled segment, which will be applied to all the segments.

When a styled text is converted to a string the corresponding ANSI SGR code is built to render the style.

If the NO_COLOR environment variable is set and non-empty when Elvish starts, color output is suppressed. Modifications to NO_COLOR within Elvish (including from rc.elv) do not affect the current process, but will affect child Elvish processes.


echo (styled foo red bold) # prints red bold "foo"
echo (styled (styled foo red bold) green) # prints green bold "foo"

A styled text can contain multiple segments with different styles. Such styled texts can be constructed by concatenating multiple styled texts with the compounding syntax. Strings and styled segments are automatically “promoted” to styled texts when concatenating. Examples:

echo foo(styled bar red) # prints "foo" + red "bar"
echo (styled foo bold)(styled bar red) # prints bold "foo" + red "bar"

The individual segments in a styled text can be extracted by indexing:

var s = (styled abc red)(styled def green)
put $s[0] $s[1]

When printed to the terminal, a styled text is not affected by any existing SGR styles in effect, and it will always reset the SGR style afterwards. For example:

print "\e[1m"
echo (styled foo red)
echo bar
# "foo" will be printed as red, but not bold
# "bar" will be printed without any style


styled-segment $object &fg-color=default &bg-color=default &bold=$false &dim=$false &italic=$false &underlined=$false &blink=$false &inverse=$false

Constructs a styled segment, a building block for styled texts.

  • If $object is a string, constructs a styled segment with $object as the content, and the properties specified by the options.

  • If $object is a styled segment, constructs a styled segment that is a copy of $object, with the properties specified by the options overridden.

The properties of styled segments can be inspected by indexing into it. Valid keys are the same as the options to styled-segment, plus text for the string content:

~> var s = (styled-segment abc &bold)
~> put $s[text]
▶ abc
~> put $s[fg-color]
▶ default
~> put $s[bold]
▶ $true

Prefer the high-level styled command to build and transform styled texts. Styled segments are a low-level construct, and you only have to deal with it when building custom style transformers.

In the following example, a custom transformer sets the inverse property for every bold segment:

styled foo(styled bar bold) {|x| styled-segment $x &inverse=$x[bold] }
# transforms "foo" + bold "bar" into "foo" + bold and inverse "bar"


take $n $inputs?

Outputs the first $n value inputs. If $n is larger than the number of value inputs, outputs everything.


~> range 2 | take 10
▶ 0
▶ 1
~> take 3 [a b c d e]
▶ a
▶ b
▶ c
~> use str
~> str:split ' ' 'how are you?' | take 1
▶ how

Etymology: Haskell.

See also drop.


tilde-abbr $path

If $path represents a path under the home directory, replace the home directory with ~. Examples:

~> echo $E:HOME
~> tilde-abbr /Users/foo
▶ '~'
~> tilde-abbr /Users/foobar
▶ /Users/foobar
~> tilde-abbr /Users/foo/a/b
▶ '~/a/b'


time &on-end=$nil $callable

Runs the callable, and call $on-end with the duration it took, as a number in seconds. If $on-end is $nil (the default), prints the duration in human-readable form.

If $callable throws an exception, the exception is propagated after the on-end or default printing is done.

If $on-end throws an exception, it is propagated, unless $callable has already thrown an exception.


~> time { sleep 1 }
~> time { sleep 0.01 }
~> var t = ''
~> time &on-end={|x| set t = $x } { sleep 1 }
~> put $t
▶ (num 1.000925004)
~> time &on-end={|x| set t = $x } { sleep 0.01 }
~> put $t
▶ (num 0.011030208)

See also benchmark.



Takes structured stdin, convert it to JSON and puts the result on bytes stdout.

~> put a | to-json
~> put [lorem ipsum] | to-json
~> put [&lorem=ipsum] | to-json

See also from-json.


to-lines $inputs?

Writes each value input to a separate line in the byte output. Byte input is ignored.

~> put a b | to-lines
~> to-lines [a b]
~> { put a; echo b } | to-lines

See also from-lines and to-terminated.


to-string $value...

Convert arguments to string values.

~> to-string foo [a] [&k=v]
▶ foo
▶ '[a]'
▶ '[&k=v]'


to-terminated $terminator $inputs?

Writes each value input to the byte output with the specified terminator character. Byte input is ignored. This behavior is useful, for example, when feeding output into a program that accepts NUL terminated lines to avoid ambiguities if the values contains newline characters.

The $terminator must be a single ASCII character such as "\x00" (NUL).

~> put a b | to-terminated "\x00" | slurp
▶ "a\x00b\x00"
~> to-terminated "\x00" [a b] | slurp
▶ "a\x00b\x00"

See also from-terminated and to-lines.


unset-env $name

Unset an environment variable. Calling unset-env VAR_NAME is similar to del E:VAR_NAME, but allows the variable name to be dynamic.


~> set E:X = foo
~> unset-env X
~> has-env X
▶ $false
~> put $E:X
▶ ''

See also has-env, get-env, and set-env.


use-mod $use-spec

Imports a module, and outputs the namespace for the module.

Most code should use the use special command instead.


~> echo 'var x = value' > a.elv
~> put (use-mod ./a)[x]
▶ value


wcswidth $string

Output the width of $string when displayed on the terminal. Examples:

~> wcswidth a
▶ 1
~> wcswidth lorem
▶ 5
~> wcswidth 你好,世界
▶ 10