Hacker News Reader

Comment by guerrilla | Hacker News Reader

It's pretty much useless for the same purpose:

    $ which ls
    /usr/bin/ls
    $ type ls
    ls is aliased to `ls --color=auto'
    $ command -v ls
    alias ls='ls --color=auto'

Suddenly I find myself irritated at Debian despite not even using it. Why would they not just include the GNU one?

View on HN

That depends on whether your shell has a built-in which or not. Mine says

    % which ls
    ls: aliased to /bin/ls --color=auto

Which makes way more sense. Your "which" is not telling you what will actually be executed when you run the `ls` command there. `command`, on the other hand, is guaranteed to be a built-in, has consistent behavior, and has defined, consistent output, unlike `which`. What `which` outputs will be different depending on shell and what implementation of `which` you actually have installed.

That depends on which which you mean. Which on GNU has always been this:

    NAME
       which - shows the full path of (shell) commands.
    DESCRIPTION
       Which takes one or more arguments. For each of its arguments it prints to stdout the full path of the
       executables that would have been executed when this argument had been entered at the shell prompt. It
       does this by searching for an executable or script in the directories listed in the environment vari‐
       able PATH using the same algorithm as bash(1).

Unless you're using a shell that subverts that by providing its own built-in, like Zsh, which is allowed because there is no standard for `which`, and depending on its behavior can be problematic and inherently non-portable.

That description is also doesn't correctly describe the behavior of the command if the shell has any aliases or built-ins of that name. If you have an alias that points to a different command, then `which` is distinctly not printing the executable that would have been executed.

> That description is also doesn't correctly describe the behavior of the command if the shell has any aliases or built-ins of that name. If you have an alias that points to a different command, then `which` is distinctly not printing the executable that would have been executed.

I have some bad news.

    % command -v which
    Unknown option: v
    _______
    < which >
    -------
            \   ^__^
             \  (oo)\_______
                (__)\       )\/\
                    ||----w |
                    ||     ||
    % alias
    command=cowsay
    run-help=man
    which-command=whence
    %

Interestingly, at least one shell will not let you do this, but it's not entirely POSIX compliant anyway:

    [I] ⋊> ~ echo $FISH_VERSION
    3.3.1
    [I] ⋊> ~ alias command cowsay
    - (line 1): function: The name 'command' is reserved, and cannot be used as a function name
    function command --wraps cowsay --description 'alias command cowsay';  cowsay $argv; end
    ^
    from sourcing file -
            called on line 70 of file /nix/store/gwc21f4ra55h0x0b8xbwnpjlc6223z3q-fish-3.3.1/share/fish/functions/alias.fish
    in function 'alias' with arguments 'command cowsay'

It's also probably worth noting at this point that portability isn't the same kind of issue for interactive shells as it is for scripts, and you should probably not expect to be using or encountering aliases in scripts at all, if you can avoid it.

As I said, it depends on which which you mean. In this case, Debian is changing it's behavior away from how it's been for 28 years or whatever and breaking heaps of code and people's habits in the process.

Yes, you're right. I think it's fine to depend on the existing which behavior for the current use cases. I just disagree that the behavior is really more useful than `command -v` for any sort of build or scripting purposes. It's definitely more useful as a user-facing utility to have a very recognizable name, like `which`. I'm one of the people who had never heard of `command -v` before now, and I'd used `which` for scripting, because I assumed it was standardized. I just don't see much use case for a shell command that finds a command in the path while specifically ignoring all aliases, functions, and built-ins over something like `command -v`.

> In this case, Debian is changing it's behavior away from how it's been for 28 years

In this case, Debian has voted to keep it the same: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=994275

> I just don't see much use case for a shell command that finds a command in the path while specifically ignoring all aliases, functions, and built-ins over something like `command -v`.

It's useful if you want to track down an executable! For example, if you're performing an Arch Linux install for the first time in years, and you notice that a program is missing on your install but present on the installation media, you can use `which` in combination with `realpath` and `pacman -Qf` to find out what you need to install to get it. (This happened to me a couple days ago, when I decided to really revisit Arch for the first time in over a decade.)

Also, `command -v` reporting builtins is kinda against the spirit of the `command` command in the first place. From, for example, the bash help:

    $ help command
    command: command [-pVv] command [arg ...]
        Execute a simple command or display information about commands.

        Runs COMMAND with ARGS suppressing  shell function lookup, or display
        information about the specified COMMANDs.  Can be used to invoke commands
        on disk when a function with the same name exists.

        Options:
          -p    use a default value for PATH that is guaranteed to find all of
                the standard utilities
          -v    print a description of COMMAND similar to the `type' builtin
          -V    print a more verbose description of each COMMAND

        Exit Status:
        Returns exit status of COMMAND, or failure if COMMAND is not found.

> Runs COMMAND with ARGS suppressing shell function lookup

Or in the ksh manual:

       command [ -pvxV ] name [ arg ... ]
              Without  the  -v  or  -V options, command executes name with the arguments given by arg.  The -p option causes a default path to be searched rather than the one defined by the value of PATH.  Functions will not be searched for when finding name.  In addition, if name refers to a special built-in, none of the special properties associated with the leading daggers will be honored.  (For example, the predefined alias redirect=′command exec′ prevents a  script  from  terminating when  an invalid redirection is given.)  With the -x option, if command execution would result in a failure because there are too many arguments, errno E2BIG, the shell will invoke command name multiple times with a subset of the arguments on each invocation.  Arguments that occur prior to the first word that expands to multiple arguments and after the last word that expands to multiple arguments will be passed on each invocation.  The exit status will be the maximum invocation exit status.  With the -v option, command is equivalent to the built-in whence command described below.  The -V option causes command to act like whence -v.

> Functions will not be searched for when finding name. In addition, if name refers to a special built-in, none of the special properties associated with the leading daggers will be honored.

and so on.

You can use the `command` command to determine which `which` you use ;)

    % which which
    which: shell built-in command
    % command which which
    /run/current-system/sw/bin/which

I like to use `which` together with `realpath` to see what exact version of a program I'm using, e.g. (in Fish),

    [I] ⋊> ~ realpath (command which which)
    /nix/store/3w3rvxhlv5dcmdih72da6m613qyav9kw-which-2.21/bin/which

Idk if it's also POSIX, but `command` also typically has an analogue called `builtin` that you can use ensure that you are not looking at an external command, e.g.:

    [I] ⋊> ~ builtin which which   # Fish doesn't have a builtin called `which`
    fish: Unknown builtin “which”
    [I] ⋊> ~ command command command # and I don't have an external command called `command`
    command: command not found

It can also be useful for figuring out which package owns an executable you're running, e.g., on Debian-based systems (also with Fish):

    > apt show (dpkg -S (realpath (command which php)) | cut -d':' -f1)
    Package: php7.4-cli
    Version: 7.4.25-1+ubuntu18.04.1+deb.sury.org+1
    Priority: optional
    Section: php
    Source: php7.4
    Maintainer: Debian PHP Maintainers <team+pkg-php@tracker.debian.org>
    Installed-Size: 4,711 kB
    Provides: php-cli, phpapi-20190902
    Depends: libedit2 (>= 2.11-20080614-4), libmagic1, mime-support, php7.4-common (= 7.4.25-1+ubuntu18.04.1+deb.sury.org+1), php7.4-json, php7.4-opcache, php7.4-readline, tzdata, ucf, libargon2-1 (>= 0~20171227), libc6 (>= 2.27), libpcre2-8-0 (>= 10.32), libsodium23 (>= 1.0.14), libssl1.1 (>= 1.1.0), libxml2 (>= 2.8.0), zlib1g (>= 1:1.1.4)
    Suggests: php-pear
    Download-Size: 1,398 kB
    APT-Sources: http://ppa.launchpad.net/ondrej/php/ubuntu bionic/main amd64 Packages
    Description: command-line interpreter for the PHP scripting language
    This package provides the /usr/bin/php7.4 command interpreter, useful for
    testing PHP scripts from a shell or performing general shell scripting tasks.
    .
    The following extensions are built in: Core date filter hash libxml openssl
    pcntl pcre Reflection session sodium SPL standard zlib.
    .
    PHP (recursive acronym for PHP: Hypertext Preprocessor) is a widely-used
    open source general-purpose scripting language that is especially suited
    for web development and can be embedded into HTML.

    N: There is 1 additional record. Please use the '-a' switch to see it

> What `which` outputs will be different depending on shell

This is also true of `command -v`, whose behavior with respect to builtins varies per shell, and is not implemented in some shells.

Fish:

    [I] ⋊> ~ fish --version                                                                                                  10:55:42
    fish, version 3.3.1
    [I] ⋊> ~ command -v command
    [I] ⋊> ~ command -v which
    /run/current-system/sw/bin/which

tcsh:

    > tcsh --version
    tcsh 6.22.04 (Astron) 2021-04-26 (x86_64-unknown-linux) options wide,nls,dl,al,kan,sm,rh,color,filec
    > command -v command
    command: Command not found.
    > command -v which
    command: Command not found.
    > builtins | grep command
    >

ksh:

    $ ksh --version
      version         sh (AT&T Research) 2020.0.0
    $ command -v command
    'command '
    $ command -v which
    /run/current-system/sw/bin/which
    $

mksh:

    $ echo $KSH_VERSION
    @(#)MIRBSD KSH R59 2020/10/31
    $ command -v command
    command
    $ command -v which
    /run/current-system/sw/bin/which
    $

pwsh:

    PS> $PSVersionTable

    Name                           Value
    ----                           -----
    PSVersion                      7.1.4
    PSEdition                      Core
    GitCommitId                    7.1.4
    OS                             Linux 5.14.12 #1-NixOS SMP Wed Oct 13 07:42:04 UTC 2021
    Platform                       Unix
    PSCompatibleVersions           {1.0, 2.0, 3.0, 4.0…}
    PSRemotingProtocolVersion      2.3
    SerializationVersion           1.1.0.1
    WSManStackVersion              3.0

    PS> command -v command
    PS> command -v which
    PS> command which

    CommandType     Name                                               Version    Source
    -----------     ----                                               -------    ------
    Application     which                                              0.0.0.0    /run/current-system/sw/bin/which

Granted, some of those shells (Fish, tcsh, PowerShell) don't aim for full POSIX compliance, and the most popular shells (bash, dash, zsh) all behave the same way as mksh in the example above. But you can also see that the output given for builtins varies among POSIX shell implementations by comparing the output of the AT&T Korn shell to the MirBSD Korn shell.

tcsh on my system:

    $ tcsh --version
    tcsh 6.21.00 (Astron) 2019-05-08 (x86_64-apple-darwin) options wide,nls,dl,bye,al,kan,sm,rh,color,filec
    $ tcsh
    % command -v command
    command
    % command -V command
    command is a shell builtin
    % builtins | grep command
    %

Yeah, this is probably because macOS has done something very, very weird, and added an executable `/usr/bin/command` to the system. These are the contents on the old MacBook I have for work right here:

    #!/bin/sh
    # $FreeBSD: src/usr.bin/alias/generic.sh,v 1.2 2005/10/24 22:32:19 cperciva Exp $
    # This file is in the public domain.
    builtin `echo %{0##*/} | tr \[:upper:] \[:lower]` ${1+"$@"}

If you have SIP disabled, try renaming `/usr/bin/command` to `/usr/bin/command.wtf` and see if `tcsh` still acts like there's a `command` command.

I do have SIP disabled for DYLD_LIBRARY_PATH and a handful of other things, but macOS still says /usr/bin/command is on a read-only file system.

[edit]

Not hard to remount root as writable

    $ sudo mount -uw /
    Password:
    $ sudo mv /usr/bin/command /usr/bin/command.save
    $ tcsh
    % command -v command
    command: Command not found.
    % command -V command
    command: Command not found.
    %

Now reversing the changes

   % exit
   exit
   $ sudo mv /usr/bin/command.save /usr/bin/command
   $ sudo mount -ur /
   mount_apfs: volume could not be mounted: Invalid argument
   mount: / failed with 66

Looks like I'll have to reboot to get the read-only state back.

Someone should make this into a blog post at this point :P

This is a very unlikely output when run from a script using the /bin/sh interpreter on Debian, though.

If that is the output you've gone out of your way to create an alias in a script, in which case it's reasonable output. It is what will happen when the script runs that command, after all.

I'm certain there's masses of code that depends on `which` responding the way it does and scripts with aliases in them regardless of whether that was the right way to do it or not, so your point is probably irrelevant in the grand scheme of things. Think about all that enterprise install and setup crap. People still depend on that spaghetti trash working.

I think aliases are only used in interactive shells:

    $ sh -c 'command -v ls'
    /bin/ls

Reply on news.ycombinator.com