Oils Reference — Chapter Builtin Commands

This chapter in the Oils Reference describes builtin commands for OSH and YSH.

(in progress)

In This Chapter

Memory

cmd/append

Append word arguments to a list:

var mylist = :| hello |

append *.py (mylist)  # append all Python files

var myflags = []
append -- -c 'echo hi' (myflags)  # -- to avoid ambiguity

It's a shortcut for:

call myflags->append('-c')
call myflags->append('echo hi')

Similar names: append

pp

The pp builtin pretty prints values and interpreter state.

Pretty printing expressions is the most common:

$ var x = 42
$ pp (x + 5)
myfile.ysh:1: (Int)   47   # print value with code location

You can pass an unevaluated expression:

$ pp [x + 5]
myfile.ysh:1: (Int)   47   # evaluate first

The value command is a synonym for the interactive = operator:

$ pp value (x)
(Int)   42

$ = x 
(Int)   42

Print proc names and doc comments:

$ pp proc  # subject to change

You can also print low-level interpreter state. The trailing _ indicates that the exact format may change:

Examples:

$ var x = :| one two |

$ pp asdl_ (x)  # dump the ASDL "guts"

$ pp test_ (x)  # single-line stable format, for spec tests

# dump the ASDL representation of a "Cell", which is a location for a value
# (not the value itself)
$ pp cell_ x

Handle Errors

error

The error builtin interrupts shell execution.

If there's a surrounding try block, the _error register is set, and execution proceeds after the block.

Otherwise, the shell exits with a non-zero status.

Examples:

error 'Missing /tmp'            # program fails with status 10

try {
   error 'Another problem'
}
echo $[error.code] # => 10

Override the default error code of 10 with a named argument:

error 'Missing /tmp' (code=99)  # program fails with status 99

Named arguments add arbitrary properties to the resulting _error register:

error 'Oops' (path='foo.json')

See YSH Error Handling for more examples.

failed

A shortcut for (_error.code !== 0):

try {
  ls /tmp
}
if failed {
  echo 'ls failed'
}

It saves you 7 punctuation characters: ( _ . !== )

See YSH Error Handling for more examples.

try

Run a block of code, stopping at the first error. (This is implemented with shopt --set errexit)

try sets the _error register to a dict, and always returns 0.

try {
  ls /nonexistent
}
if (_error.code !== 0) {
  echo 'ls failed'
}

Handle expression errors:

try {
  var x = 42 / 0
}

And errors from compound commands:

try {
  ls | wc -l
  diff <(sort left.txt) <(sort right.txt)
}

The case statement can be useful:

try {
  grep PATTERN FILE.txt
}
case (_error.code) {
  (0)    { echo 'found' }
  (1)    { echo 'not found' }
  (else) { echo "grep returned status $[_error.code]" }
}

See YSH Error Handling for more examples.

boolstatus

Runs a command, and requires the exit code to be 0 or 1.

if boolstatus egrep '[0-9]+' myfile {  # e.g. aborts on status 2
  echo 'found'               # status 0 means found
} else {
  echo 'not found'           # status 1 means not found
}

It's meant for external commands that "return" more than 2 values, like true / false / fail, rather than pass / fail.

assert

Evaluates and expression, and fails if it is not truthy.

assert (false)   # fails
assert [false]   # also fails (the expression is evaluated)

It's common to pass an unevaluated expression with ===:

func f() { return (42) }

assert [43 === f()]

In this special case, you get a nicer error message:

Expected: 43 Got: 42

That is, the left-hand side should be the expected value, and the right-hand side should be the actual value.

Shell State

ysh-cd

It takes a block:

cd / {
  echo $PWD
}

ysh-shopt

It takes a block:

shopt --unset errexit {
  false
  echo 'ok'
}

shvar

Execute a block with a global variable set.

shvar IFS=/ {
  echo "ifs is $IFS"
}
echo "ifs restored to $IFS"

ctx

Execute a block with a shared "context" that can be updated using the ctx built-in.

var mydict = {}
ctx push (mydict) {
   # = mydict => {}
   ctx set (mykey='myval')
}
# = mydict => { mykey: 'myval' }

The context can be modified with ctx set (key=val), which updates or inserts the value at the given key.

The context can also be updated with ctx emit field (value).

ctx push (mydict) {
   # = mydict => {}
   ctx emit mylist (0)
   # = mydict => { mylist: [0] }
   ctx emit mylist (1)
}
# = mydict => { mylist: [0, 1] }

Contexts can be nested, resulting in a stack of contexts.

ctx push (mydict1) {
    ctx set (dict=1)
    ctx push (mydict2) {
        ctx set (dict=2)
    }
}
# = mydict1 => { dict: 1 }
# = mydict2 => { dict: 2 }

ctx is useful for creating DSLs, such as a mini-parseArgs.

proc parser (; place ; ; block_def) {
  var p = {}
  ctx push (p, block_def)
  call place->setValue(p)
}

proc flag (short_name, long_name; type; help) {
  ctx emit flag ({short_name, long_name, type, help})
}

proc arg (name) {
  ctx emit arg ({name})
}

parser (&spec) {
  flag -t --tsv (Bool, help='Output as TSV')
  flag -r --recursive (Bool, help='Recurse into the given directory')
  flag -N --count (Int, help='Process no more than N files')
  arg path
}

push-registers

Save global registers like $? on a stack. It's useful for preventing plugins from interfering with user code. Example:

status_42         # returns 42 and sets $?
push-registers {  # push a new frame
  status_43       # top of stack changed here
  echo done
}                 # stack popped
echo $?           # 42, read from new top-of-stack

Current list of registers:

Regex data underlying BASH_REMATCH, _group(), _start(), _end()
$?             
_error                # set by the try builtin
PIPESTATUS            # aka  _pipeline_status
_process_sub_status

Modules

runproc

Runs a named proc with the given arguments. It's often useful as the only top level statement in a "task file":

proc p {
  echo hi
}
runproc @ARGV

Like 'builtin' and 'command', it affects the lookup of the first word.

source-guard

Registers a name in the global "module" dict. Returns 0 if it doesn't exist, or 1 if it does.

Use it like this in executable files:

source-guard main || return 0   

And like this in libraries:

source-guard myfile.ysh || return 0   

is-main

The is-main builtin returns 1 (false) if the current file was executed with the source builtin.

In the "main" file, including -c or stdin input, it returns 0 (true).

Use it like this:

if is-main {
  runproc @ARGV
}

use

TODO

Reuse code from other files, respecting namespaces.

use lib/foo.ysh  # foo myproc, $[foo.attr]
                 # implicit $_this_dir aka relative import

Bind a specific name:

use lib/foo.ysh (&myvar)  # makes 'myvar' available

Bind multiple names:

use lib/foo.ysh (&myvar) {
  pick log die
}

Maybe:

use lib/foo.ysh (&myvar) {
  pick log (&mylog)
  pick die (&mydie)
}

Also a declaration

use --extern grep sed

I/O

ysh-read

YSH adds long flags to shell's read:

read --all               # whole file including trailing \n, fills $_reply
read --all (&x)          # fills $x

read --num-bytes 3       # read N bytes, fills _reply
read --num-bytes 3 (&x)  # fills $x

read --raw-line             # unbuffered read of line, omitting trailing \n
read --raw-line (&x)        # fills $x

read --raw-line --with-eol  # include the trailing \n

And a convenience:

read -0                 # read until NUL, synonym for read -r -d ''

You may want to use fromJson8() or fromJson() after reading a line.

ysh-echo

Print arguments to stdout, separated by a space.

ysh$ echo hi there
hi there

The simple_echo option means that flags aren't accepted, and -- is not accepted.

ysh$ echo -n
-n

See the YSH FAQ for details.

write

write fixes problems with shell's echo builtin.

The default separator is a newline, and the default terminator is a newline.

Examples:

write -- ale bean         # write two lines

write -n -- ale bean      # synonym for --end '', like echo -n
write --sep '' --end '' -- a b        # write 2 bytes
write --sep $'\t' --end $'\n' -- a b  # TSV line

You may want to use toJson8() or toJson() before writing:

write -- $[toJson8(mystr)]
write -- $[toJson(mystr)]

fork

Run a command, but don't wait for it to finish.

fork { sleep 1 }
wait -n

In YSH, use fork rather than shell's & (ampersand).

forkwait

The preferred alternative to shell's (). Prefer cd with a block if possible.

forkwait {
  not_mutated=zzz
}
echo $not_mutated

fopen

Runs a block passed to it. It's designed so redirects have a prefix syntax:

fopen >out.txt {
  echo 1
  echo 2
}

Rather than shell style:

{ echo 1
  echo 2
} >out.txt

When a block is long, the former is more readable.

Hay Config

hay

haynode

Data Formats

json

Write JSON:

var d = {name: 'bob', age: 42}
json write (d)           # default indentation of 2
json write (d, space=0)  # no indentation

Read JSON:

echo hi | json read  # fills $_reply by default

Or use an explicit place:

var x = ''
json read (&x) < myfile.txt

Related: err-json-encode and err-json-decode

json8

Like json, but on the encoding side:

On decoding side:

Related: err-json8-encode and err-json8-decode

Testing

TODO: describe

External Lang

TODO: when

I/O

These builtins take input and output. They're often used with redirects.

read

read FLAG* VAR*

Read a line from stdin, split it into tokens with the $IFS algorithm, and assign the tokens to the given variables. When no VARs are given, assign to $REPLY.

Note: When writing ySH, prefer the extensions documented in ysh-read. The read builtin is confusing because -r needs to be explicitly enabled.

Flags:

-a ARRAY  assign the tokens to elements of this array
-d CHAR   use DELIM as delimiter, instead of newline
-n NUM    read up to NUM characters, respecting delimiters
-p STR    print the string PROMPT before reading input
-r        raw mode: don't let backslashes escape characters
-s        silent: do not echo input coming from a terminal
-t NUM    time out and fail after TIME seconds
          -t 0 returns whether any input is available
-u FD     read from file descriptor FD instead of 0 (stdin)

echo

echo FLAG* ARG*

Prints ARGs to stdout, separated by a space, and terminated by a newline.

Flags:

-e  enable interpretation of backslash escapes
-n  omit the trailing newline

See char-escapes.

printf

printf FLAG* FMT ARG*

Formats values and prints them. The FMT string contain three types of objects:

  1. Literal Characters
  2. Character escapes like \t. See char-escapes.
  3. Percent codes like %s that specify how to format each each ARG.

If not enough ARGS are passed, the empty string is used. If too many are passed, the FMT string will be "recycled".

Flags:

-v VAR  Write output in variable VAR instead of standard output.

Format specifiers:

%%  Prints a single "%".
%b  Interprets backslash escapes while printing.
%q  Prints the argument escaping the characters needed to make it reusable
    as shell input.
%d  Print as signed decimal number.
%i  Same as %d.
%o  Print as unsigned octal number.
%u  Print as unsigned decimal number.
%x  Print as unsigned hexadecimal number with lower-case hex-digits (a-f).
%X  Same as %x, but with upper-case hex-digits (A-F).
%f  Print as floating point number.
%e  Print as a double number, in "±e" format (lower-case e).
%E  Same as %e, but with an upper-case E.
%g  Interprets the argument as double, but prints it like %f or %e.
%G  Same as %g, but print it like %E.
%c  Print as a single char, only the first character is printed.
%s  Print as string
%n  The number of characters printed so far is stored in the variable named
    in the argument.
%a  Interprets the argument as double, and prints it like a C99 hexadecimal
    floating-point literal.
%A  Same as %a, but print it like %E.
%(FORMAT)T  Prints date and time, according to FORMAT as a format string
            for strftime(3). The argument is the number of seconds since
            epoch. It can also be -1 (current time, also the default value
            if there is no argument) or -2 (shell startup time).

readarray

Alias for mapfile.

mapfile

mapfile FLAG* ARRAY?

Reads lines from stdin into the variable named ARRAY (default ${MAPFILE[@]}).

Flags:

-t       Remove the trailing newline from every line

Run Code

These builtins accept shell code and run it.

source

source SCRIPT ARG*

Execute SCRIPT with the given ARGs, in the context of the current shell. That is, existing variables will be modified.


Oils extension: If the SCRIPT starts with ///, we look for scripts embedded in the oils-for-unix binary. Example:

source ///osh/two.sh     # load embedded script

: ${LIB_OSH=fallback/dir}
source $LIB_OSH/two.sh   # same thing

The LIB_OSH form is useful for writing a script that works under both bash and OSH.

eval

eval ARG+

Creates a string by joining ARGs with a space, then runs it as a shell command.

Example:

 # Create the string echo "hello $name" and run it.
 a='echo'
 b='"hello $name"'
 eval $a $b

Tips:

YSH eval:

var myblock = ^(echo hi)
eval (myblock)  # => hi

trap

trap FLAG* CMD SIGNAL*

Registers the shell string CMD to be run after the SIGNALs are received. If the CMD is empty, then the signal is ignored.

Flags:

-l  Lists all signals and their signal number
-p  Prints a list of the installed signal handlers

Tip:

Prefer passing the name of a shell function to trap.

Set Options

The set and shopt builtins set global shell options. YSH code should use the more natural shopt.

set

set FLAG* ARG*

Sets global shell options. Short style:

set -e

Long style:

set -o errexit

Set the arguments array:

set -- 1 2 3

shopt

shopt FLAG* OPTION* BLOCK?

Sets global shell options.

Flags:

-s --set    Turn the named options on
-u --unset  Turn the named options off
-p          Print option values
-o          Use older set of options, normally controlled by 'set -o'
-q          Return 0 if the option is true, else 1

Examples:

shopt --set errexit

You can set or unset multiple options with the groups strict:all, ysh:upgrade, and ysh:all.

If a block is passed, then the mutated options are pushed onto a stack, the block is executed, and then options are restored to their original state.

Working Dir

These 5 builtins deal with the working directory of the shell.

cd

cd FLAG* DIR

Changes the working directory of the current shell process to DIR.

If DIR isn't specified, change to $HOME. If DIR is -, change to $OLDPWD (a variable that the sets to the previous working directory.)

Flags:

-L  Follow symbolic links, i.e. change to the TARGET of the symlink.
    (default).
-P  Don't follow symbolic links.

pwd

pwd FLAG*

Prints the current working directory.

Flags:

-L  Follow symbolic links if present (default)
-P  Don't follow symbolic links.  Print the link instead of the target.

pushd

pushd DIR

Add DIR to the directory stack, then change the working directory to DIR. Typically used with popd and dirs.

popd

popd

Removes a directory from the directory stack, and changes the working directory to it. Typically used with pushd and dirs.

dirs

dirs FLAG*

Shows the contents of the directory stack. Typically used with pushd and popd.

Flags:

-c  Clear the dir stack.
-l  Show the dir stack, but with the real path instead of ~.
-p  Show the dir stack, but formatted as one line per entry.
-v  Like -p, but numbering each line.

Completion

These builtins implement our bash-compatible autocompletion system.

complete

Registers completion policies for different commands.

compgen

Generates completion candidates inside a user-defined completion function.

It can also be used in scripts, i.e. outside a completion function.

compopt

Changes completion options inside a user-defined completion function.

compadjust

Adjusts COMP_ARGV according to specified delimiters, and optionally set variables cur, prev, words (an array), and cword. May also set 'split'.

This is an OSH extension that makes it easier to run the bash-completion project.

compexport

Complete an entire shell command string. For example,

compexport -c 'echo $H'

will complete variables like $HOME. And

compexport -c 'ha'

will complete builtins like hay, as well as external commands.

Shell Process

These builtins mutate the state of the shell process.

exec

exec BIN_PATH ARG*

Replaces the running shell with the binary specified, which is passed ARGs. BIN_PATH must exist on the file system; i.e. it can't be a shell builtin or function.

umask

umask MODE?

Sets the bit mask that determines the permissions for new files and directories. The mask is subtracted from 666 for files and 777 for directories.

Oils currently supports writing masks in octal.

If no MODE, show the current mask.

ulimit

ulimit --all
ulimit -a
ulimit FLAGS* -RESOURCE_FLAG VALUE?

ulimit FLAGS* VALUE?  # discouraged

Show and modify process resource limits.

Flags:

-S  for soft limit
-H  for hard limit

-c -d -f ...  # ulimit --all shows all resource flags

Show a table of resources:

ulimit --all
ulimit -a

For example, the table shows that -n is the flag that controls the number file descriptors, the soft and hard limit for -n, and the multiplication "factor" for the integer VALUE you pass.


Here are examples of using resource flags.

Get the soft limit for the number of file descriptors:

ulimit -S -n
ulimit -n     # same thing

Get the hard limit:

ulimit -H -n

Set the soft or hard limit:

ulimit -S -n 100
ulimit -H -n 100

Set both limits:

ulimit -n 100

A special case that's discouraged: with no resource flag, -f is assumed:

ulimit      # equivalent to ulimit -f
ulimit 100  # equivalent to ulimit -f 100

times

times

Shows the user and system time used by the shell and its child processes.

Child Process

jobs

jobs

Shows all jobs running in the shell and their status.

wait

wait FLAG* ARG

Wait for processes to exit.

If the ARG is a PID, wait only for that job, and return its status.

If there's no ARG, wait for all child processes.

Flags:

-n  Wait for the next process to exit, rather than a specific process.

Wait can be interrupted by a signal, in which case the exit code indicates the signal number.

fg

fg JOB?

Returns a job running in the background to the foreground. If no JOB is specified, use the latest job.

External

test

test OP ARG
test ARG OP ARG
[ OP ARG ]      # [ is an alias for test that requires closing ]
[ ARG OP ARG ]

Evaluates a conditional expression and returns 0 (true) or 1 (false).

Note that [ is the name of a builtin, not an operator in the language. Use 'test' to avoid this confusion.

String expressions:

-n STR           True if STR is not empty.
                 'test STR' is usually equivalent, but discouraged.
-z STR           True if STR is empty.
STR1 = STR2      True if the strings are equal.
STR1 != STR2     True if the strings are not equal.
STR1 < STR2      True if STR1 sorts before STR2 lexicographically.
STR1 > STR2      True if STR1 sorts after STR2 lexicographically.
                 Note: < and > should be quoted like \< and \>

File expressions:

-a FILE          Synonym for -e.
-b FILE          True if FILE is a block special file.
-c FILE          True if FILE is a character special file.
-d FILE          True if FILE is a directory.
-e FILE          True if FILE exists.
-f FILE          True if FILE is a regular file.
-g FILE          True if FILE has the sgid bit set.
-G FILE          True if current user's group is also FILE's group.
-h FILE          True if FILE is a symbolic link.
-L FILE          True if FILE is a symbolic link.
-k FILE          True if FILE has the sticky bit set.
-O FILE          True if current user is the file owner.
-p FILE          True if FILE is a named pipe (FIFO).
-r FILE          True if FILE is readable.
-s FILE          True if FILE has size bigger than 0.
-S FILE          True if FILE is a socket file.
-t FD            True if file descriptor FD is open and refers to a terminal.
-u FILE          True if FILE has suid bit set.
-w FILE          True if FILE is writable.
-x FILE          True if FILE is executable.
FILE1 -nt FILE2  True if FILE1 is newer than FILE2 (mtime).
FILE1 -ot FILE2  True if FILE1 is older than FILE2 (mtime).
FILE1 -ef FILE2  True if FILE1 is a hard link to FILE2.

Arithmetic expressions coerce arguments to integers, then compare:

INT1 -eq INT2    True if they're equal.
INT1 -ne INT2    True if they're not equal.
INT1 -lt INT2    True if INT1 is less than INT2.
INT1 -le INT2    True if INT1 is less or equal than INT2.
INT1 -gt INT2    True if INT1 is greater than INT2.
INT1 -ge INT2    True if INT1 is greater or equal than INT2.

Other expressions:

-o OPTION        True if the shell option OPTION is set.
-v VAR           True if the variable VAR is set.

The test builtin also supports POSIX conditionals like -a, -o, !, and ( ), but these are discouraged.

Oils supports these long flags:

--dir            same as -d
--exists         same as -e
--file           same as -f
--symlink        same as -L

getopts

getopts SPEC VAR ARG*

A single iteration of flag parsing. The SPEC is a sequence of flag characters, with a trailing : to indicate that the flag takes an argument:

ab    # accept  -a and -b
xy:z  # accept -x, -y arg, and -z

The input is "$@" by default, unless ARGs are passed.

On each iteration, the flag character is stored in VAR. If the flag has an argument, it's stored in $OPTARG. When an error occurs, VAR is set to ? and $OPTARG is unset.

Returns 0 if a flag is parsed, or 1 on end of input or another error.

Example:

while getopts "ab:" flag; do
    case $flag in
        a)   flag_a=1 ;;
        b)   flag_b=$OPTARG" ;;
        '?') echo 'Invalid Syntax'; break ;;
    esac
done

Notes:

kill

Unimplemented.

Introspection

help

Usage: help TOPIC?

Examples:

    help               # this help
    help echo          # help on the 'echo' builtin
    help command-sub   # help on command sub $(date)

    help oils-usage    # identical to oils-for-unix --help
    help osh-usage     #              osh --help
    help ysh-usage     #              ysh --help

hash

hash

Display information about remembered commands.

hash FLAG* CMD+

Determine the locations of commands using $PATH, and remember them.

Flag:

-r       Discard all remembered locations.

cmd/type

type FLAG* NAME+

Print the type of each NAME, if it were the first word of a command. Is it a shell keyword, builtin command, shell function, alias, or executable file on $PATH?

Flags:

-a  Show all possible candidates, not just the first one
-f  Don't search for shell functions
-P  Only search for executable files
-t  Print a single word: alias, builtin, file, function, or keyword

Similar names: type

Modeled after the bash type builtin.

Word Lookup

command

command FLAG* CMD ARG*

Look up CMD as a shell builtin or executable file, and execute it with the given ARGs. That is, the lookup ignores shell functions named CMD.

Flags:

-v  Instead of executing CMD, print a description of it.
    Similar to the 'type' builtin.

builtin

builtin CMD ARG*

Look up CMD as a shell builtin, and execute it with the given ARGs. That is, the lookup ignores shell functions and executables named CMD.

Interactive

alias

alias NAME=CODE

Make NAME a shortcut for executing CODE, e.g. alias hi='echo hello'.

alias NAME

Show the value of this alias.

alias

Show a list of all aliases.

Tips:

Prefer shell functions like:

ls() {
  command ls --color "$@"
}

to aliases like:

alias ls='ls --color'

Functions are less likely to cause parsing problems.

unalias

unalias NAME

Remove the alias NAME.

history

history FLAG*

Display and manipulate the shell's history entries.

history NUM

Show the last NUM history entries.

Flags:

-c      Clears the history.
-d POS  Deletes the history entry at position POS.

Unsupported

enable

Bash has this, but OSH won't implement it.

Generated on Wed, 07 Aug 2024 22:42:07 +0000