home | help
PATH(1) 			   fish-shell				 PATH(1)

NAME
     path - manipulate and check paths

SYNOPSIS
     path basename GENERAL_OPTIONS [(-E | --no-extension)] [PATH ...]
     path dirname GENERAL_OPTIONS  [PATH ...]
     path extension GENERAL_OPTIONS [PATH ...]
     path filter GENERAL_OPTIONS [-v | --invert]
	 [-d] [-f] [-l] [-r] [-w] [-x]
	 [(-t | --type) TYPE] [(-p | --perm) PERMISSION] [--all] [PATH ...]
     path is GENERAL_OPTIONS [(-v | --invert)] [(-t | --type) TYPE]
	 [-d] [-f] [-l] [-r] [-w] [-x]
	 [(-p | --perm) PERMISSION] [PATH ...]
     path mtime GENERAL_OPTIONS [(-R | --relative)] [PATH ...]
     path normalize GENERAL_OPTIONS [PATH ...]
     path resolve GENERAL_OPTIONS [PATH ...]
     path change-extension GENERAL_OPTIONS EXTENSION [PATH ...]
     path sort GENERAL_OPTIONS [-r | --reverse]
	 [-u | --unique] [--key=(basename | dirname | path)] [PATH ...]

     GENERAL_OPTIONS
	 [-z | --null-in] [-Z | --null-out] [-q | --quiet]

DESCRIPTION
     path performs operations on paths.

     PATH  arguments  are  taken  from the command line unless standard input is
     connected to a pipe or a file, in which case they are  read  from	standard
     input,  one  PATH per line. It is an error to supply PATH arguments on both
     the command line and on standard input.

     Arguments starting with - are normally interpreted as switches;  --  causes
     the  following  arguments	not to be treated as switches even if they begin
     with -. Switches and required arguments are recognized only on the  command
     line.

     When  a  path starts with -, path filter and path normalize will prepend ./
     on output to avoid it being interpreted as an  option  otherwise,	so  it's
     safe  to  pass  path's  output  to  other commands that can handle relative
     paths.

     All subcommands accept a -q or --quiet switch, which suppresses  the  usual
     output  but  exits  with the documented status. In this case these commands
     will quit early, without reading all of the available input.

     All subcommands also accept a -Z or --null-out  switch,  which  makes  them
     print  output  separated  with NUL instead of newlines. This is for further
     processing, e.g. passing to another path, or xargs -0. This is  not  recom-
     mended when the output goes to the terminal or a command substitution.

     All  subcommands also accept a -z or --null-in switch, which makes them ac-
     cept arguments from stdin separated with NULL-bytes. Since Unix paths can't
     contain NULL, that makes it possible to handle all possible paths and  read
     input  from  e.g.	find  -print0. If arguments are given on the commandline
     this has no effect. This should mostly be unnecessary since path  automati-
     cally  starts splitting on NULL if one appears in the first PATH_MAX bytes,
     PATH_MAX being the operating system's maximum length for a path plus a NULL
     byte.

     Some subcommands operate on the paths as strings and so work on nonexistent
     paths, while others need to access the paths themselves and so  filter  out
     nonexistent paths.

     The following subcommands are available.

BASENAME SUBCOMMAND
	path basename [-E | --no-extension] [-z | --null-in] [-Z | --null-out] [-q | --quiet] [PATH ...]

     path  basename returns the last path component of the given path, by remov-
     ing the directory prefix and removing trailing slashes. In other words,  it
     is the part that is not the dirname. For files you might call it the "file-
     name".

     If  the  -E or ---no-extension option is used and the base name contained a
     period, the path is returned with the extension (or the last extension) re-
     moved, i.e. the "filename" without  an  extension	(akin  to  calling  path
     change-extension "" (path basename $path)).

     It returns 0 if there was a basename, i.e. if the path wasn't empty or just
     slashes.

   Examples
	>_ path basename ./foo.mp4
	foo.mp4

	>_ path basename ../banana
	banana

	>_ path basename /usr/bin/
	bin

	>_ path basename /usr/bin/*
	# This prints all files in /usr/bin/
	# A selection:
	cp
	fish
	grep
	rm

DIRNAME SUBCOMMAND
	path dirname [-z | --null-in] [-Z | --null-out] [-q | --quiet] [PATH ...]

     path  dirname  returns the dirname for the given path. This is the part be-
     fore the last "/", discounting trailing slashes. In other words, it is  the
     part that is not the basename (discounting superfluous slashes).

     It  returns 0 if there was a dirname, i.e. if the path wasn't empty or just
     slashes.

   Examples
	>_ path dirname ./foo.mp4
	.

	>_ path dirname ../banana
	..

	>_ path dirname /usr/bin/
	/usr

EXTENSION SUBCOMMAND
	path extension [-z | --null-in] [-Z | --null-out] [-q | --quiet] [PATH ...]

     path extension returns the extension of the given path. This  is  the  part
     after  (and  including) the last ".", unless that "." followed a "/" or the
     basename is "." or "..", in which case there is no extension and  an  empty
     line is printed.

     If the filename ends in a ".", only a "." is printed.

     It returns 0 if there was an extension.

   Examples
	>_ path extension ./foo.mp4
	.mp4

	>_ path extension ../banana
	# an empty line, status 1

	>_ path extension ~/.config
	# an empty line, status 1

	>_ path extension ~/.config.d
	.d

	>_ path extension ~/.config.
	.

	>_ set -l path (path change-extension '' ./foo.mp4)
	>_ set -l extension (path extension ./foo.mp4)
	> echo $path$extension
	# reconstructs the original path again.
	./foo.mp4

FILTER SUBCOMMAND
	path filter [-z | --null-in] [-Z | --null-out] [-q | --quiet] \
	    [-d] [-f] [-l] [-r] [-w] [-x] \
	    [-v | --invert] [(-t | --type) TYPE] [(-p | --perm) PERMISSION] [--all] [PATH ...]

     path  filter returns all of the given paths that match the given checks. In
     all cases, the paths need to exist, nonexistent paths are always filtered.

     The available filters are:

     * -t or --type with the options: "dir", "file",  "link",  "block",  "char",
       "fifo"  and  "socket",  in  which  case the path needs to be a directory,
       file, link, block device, character device, named pipe or socket, respec-
       tively.

     * -d, -f and -l are short for --type=dir, --type=file and --type=link,  re-
       spectively. There are no shortcuts for the other types.

     * -p  or  --perm  with the options: "read", "write", and "exec", as well as
       "suid", "sgid", "user" (referring to the path owner) and "group"  (refer-
       ring  to  the  path's group), in which case the path needs to have all of
       the given permissions for the current user.

     * -r, -w and -x are short for --perm=read,  --perm=write  and  --perm=exec,
       respectively. There are no shortcuts for the other permissions.

     Note  that the path needs to be any of the given types, but have all of the
     given permissions. This is because having a path that is both writable  and
     executable  makes	sense,	but having a path that is both a directory and a
     file doesn't. Links will count as the type of the linked-to file, so  links
     to files count as files, links to directories count as directories.

     The  filter options can either be given as multiple options, or comma-sepa-
     rated - path filter -t dir,file or path filter --type dir --type  file  are
     equivalent.

     With  --invert,  the  meaning  of the filtering is inverted - any path that
     wouldn't pass (including by not existing) passes, and any path  that  would
     pass fails.

     When  a  path  starts with -, path filter will prepend ./ to avoid it being
     interpreted as an option otherwise.

     It returns 0 if at least one path passed the filter.

     With --all, return status 0 (true) if all paths pass the filter, and status
     1 (false) if any path fails. This is equivalent to not path filter  -v.  It
     produces no output, only a status.

     When  --all combined with --invert, it returns status 0 (true) if all paths
     fail the filter and status 1 (false) if any path passes.

     path is is shorthand for path filter -q, i.e. just checking without produc-
     ing output, see The is subcommand.

   Examples
	>_ path filter /usr/bin /usr/argagagji
	# The (hopefully) nonexistent argagagji is filtered implicitly:
	/usr/bin

	>_ path filter --type file /usr/bin /usr/bin/fish
	# Only fish is a file
	/usr/bin/fish

	>_ path filter --type file,dir --perm exec,write /usr/bin/fish /home/me
	# fish is a file, which passes, and executable, which passes,
	# but probably not writable, which fails.
	#
	# $HOME is a directory and both writable and executable, typically.
	# So it passes.
	/home/me

	>_ path filter -fdxw /usr/bin/fish /home/me
	# This is the same as above: "-f" is "--type=file", "-d" is "--type=dir",
	# "-x" is short for "--perm=exec" and "-w" short for "--perm=write"!
	/home/me

	>_ path filter -fx $PATH/*
	# Prints all possible commands - the first entry of each name is what fish would execute!

	>_ path filter --all /usr/bin /usr/argagagji
	# This returns 1 (false) because not all paths pass the filter.

IS SUBCOMMAND
	path is [-z | --null-in] [-Z | --null-out] [-q | --quiet] \
	    [-d] [-f] [-l] [-r] [-w] [-x] \
	    [-v | --invert] [(-t | --type) TYPE] [(-p | --perm) PERMISSION] [PATH ...]

     path is is short for path filter -q. It returns true if any  of  the  given
     files passes the filter, but does not produce any output.

     --quiet can still be passed for compatibility but is redundant. The options
     are the same as for path filter.

   Examples
	>_ path is /usr/bin /usr/argagagji
	# /usr/bin exists, so this returns a status of 0 (true). It prints nothing.
	>_ path is /usr/argagagji
	# /usr/argagagji does not, so this returns a status of 1 (false). It also prints nothing.
	>_ path is -fx /bin/sh
	# /bin/sh is usually an executable file, so this returns true.

MTIME SUBCOMMAND
	path mtime [-z | --null-in] [-Z | --null-out] [-q | --quiet] [-R | --relative] [PATH ...]

     path  mtime  returns the last modification time ("mtime" in unix jargon) of
     the given paths, in seconds since the unix epoch (the beginning of the  1st
     of January 1970).

     With  --relative (or -R), it prints the number of seconds since the modifi-
     cation time. It only reads the current time once at start, so in case  mul-
     tiple paths are given the times are all relative to the start of path mtime
     -R running.

     If you want to know if a file is newer or older than another file, consider
     using test -nt instead. See the test documentation.

     It returns 0 if reading mtime for any path succeeded.

   Examples
	>_ date +%s
	# This prints the current time as seconds since the epoch
	1657217847

	>_ path mtime /etc/
	1657213796

	>_ path mtime -R /etc/
	4078
	# So /etc/ on this system was last modified a little over an hour ago

	# This is the same as
	>_ math (date +%s) - (path mtime /etc/)

NORMALIZE SUBCOMMAND
	path normalize [-z | --null-in] [-Z | --null-out] [-q | --quiet] [PATH ...]

     path  normalize returns the normalized versions of all paths. That means it
     squashes duplicate "/", collapses "../" with earlier components and removes
     "." components.

     Unlike realpath or path resolve, it does not make the  paths  absolute.  It
     also  does not resolve any symlinks. As such it can operate on non-existent
     paths.

     Because it operates on paths as strings and doesn't  resolve  symlinks,  it
     works  sort  of like pwd -L and cd. E.g. path normalize link/.. will return
     ., just like cd link; cd .. would return to the current  directory.  For  a
     physical view of the filesystem, see path resolve.

     Leading "./" components are usually removed. But when a path starts with -,
     path normalize will add it instead to avoid confusion with options.

     It  returns  0 if any normalization was done, i.e. any given path wasn't in
     canonical form.

   Examples
	>_ path normalize /usr/bin//../../etc/fish
	# The "//" is squashed and the ".." components neutralize the components before
	/etc/fish

	>_ path normalize /bin//bash
	# The "//" is squashed, but /bin isn't resolved even if your system links it to /usr/bin.
	/bin/bash

	>_ path normalize ./my/subdirs/../sub2
	my/sub2

	>_ path normalize -- -/foo
	./-/foo

RESOLVE SUBCOMMAND
	path resolve [-z | --null-in] [-Z | --null-out] [-q | --quiet] [PATH ...]

     path resolve returns the normalized, physical and absolute versions of  all
     paths.  That  means it resolves symlinks and does what path normalize does:
     it squashes duplicate "/", collapses "../" with earlier components and  re-
     moves "." components. Then it turns that path into the absolute path start-
     ing from the filesystem root "/".

     It  is  similar to realpath, as it creates the "real", canonical version of
     the path. However, for paths that can't be resolved, e.g. if they don't ex-
     ist or form a symlink loop, it will resolve as far as it can and  normalize
     the rest.

     Because  it  resolves symlinks, it works sort of like pwd -P. E.g. path re-
     solve link/.. will return the parent directory of what the link points  to,
     just like cd link; cd (pwd -P)/.. would go to it. For a logical view of the
     filesystem, see path normalize.

     It  returns  0  if any normalization or resolution was done, i.e. any given
     path wasn't in canonical form.

   Examples
	>_ path resolve /bin//sh
	# The "//" is squashed, and /bin is resolved if your system links it to /usr/bin.
	# sh here is bash (this is common on linux systems)
	/usr/bin/bash

	>_ path resolve /bin/foo///bar/../baz
	# Assuming /bin exists and is a symlink to /usr/bin, but /bin/foo doesn't.
	# This resolves the /bin/ and normalizes the nonexistent rest:
	/usr/bin/foo/baz

CHANGE-EXTENSION SUBCOMMAND
	path change-extension [-z | --null-in] [-Z | --null-out] \
	    [-q | --quiet] EXTENSION [PATH ...]

     path change-extension returns the given paths, with their extension changed
     to the given new extension. The extension is the part after (and including)
     the last ".", unless that "." followed a "/" or  the  basename  is  "."  or
     "..",  in	which  case  there  is	no previous extension and the new one is
     added.

     If the extension is empty, any previous extension is stripped,  along  with
     the ".". This is, of course, the inverse of path extension.

     One  leading  dot	on  the  extension  is	ignored, so ".mp3" and "mp3" are
     treated the same.

     It returns 0 if it was given any paths.

   Examples
	>_ path change-extension mp4 ./foo.wmv
	./foo.mp4

	>_ path change-extension .mp4 ./foo.wmv
	./foo.mp4

	>_ path change-extension '' ../banana
	../banana

	>_ path change-extension '' ~/.config
	/home/alfa/.config

	>_ path change-extension '' ~/.config.d
	/home/alfa/.config

	>_ path change-extension '' ~/.config.
	/home/alfa/.config

SORT SUBCOMMAND
	path sort [-z | --null-in] [-Z | --null-out] \
	    [-q | --quiet] [-r | --reverse] \
	    [--key=basename|dirname|path] [PATH ...]

     path sort returns the given paths in sorted order. They are sorted  in  the
     same  order  as  globs  - alphabetically, but with runs of numerical digits
     compared numerically.

     With --reverse or -r the sort is reversed.

     With --key= only the given part of the path is compared, e.g. --key=dirname
     causes only the dirname to be compared, --key=basename  only  the	basename
     and --key=path causes the entire path to be compared (this is the default).

     With  --unique  or -u the sort is deduplicated, meaning only the first of a
     run that have the same key is kept. So if you are sorting by basename, then
     only the first of each basename is used.

     The sort used is stable, so sorting first by basename and then  by  dirname
     works and causes the files to be grouped according to directory.

     It currently returns 0 if it was given any paths.

   Examples
	>_ path sort 10-foo 2-bar
	2-bar
	10-foo

	>_ path sort --reverse 10-foo 2-bar
	10-foo
	2-bar

	>_ path sort --unique --key=basename $fish_function_path/*.fish
	# prints a list of all function files fish would use, sorted by name.

COMBINING PATH
     path is meant to be easy to combine with itself, other tools and fish.

     This is why

     * path's  output  is  automatically split by fish if it goes into a command
       substitution, so just doing (path ...) handles all paths, even those con-
       taining newlines, correctly

     * path has --null-in to handle null-delimited  input  (typically  automati-
       cally detected!), and --null-out to pass on null-delimited output

     Some examples of combining path:

	# Expand all paths in the current directory, leave only executable files, and print their resolved path
	path filter -zZ -xf -- * | path resolve -z

	# The same thing, but using find (note -maxdepth needs to come first or find will scream)
	# (this also depends on your particular version of find)
	# Note the `-z` is unnecessary for any sensible version of find - if `path` sees a NULL,
	# it will split on NULL automatically.
	find . -maxdepth 1 -type f -executable -print0 | path resolve -z

	set -l paths (path filter -p exec $PATH/fish -Z | path resolve)

COPYRIGHT
     fish-shell developers

4.6				  Aug 29, 2026				 PATH(1)

home | help