119

Typical Unix/Linux programs accept the command line inputs as an argument count (int argc) and an argument vector (char *argv[]). The first element of argv is the program name - followed by the actual arguments.

Why is the program name passed to the executable as an argument? Are there any examples of programs using their own name (maybe some kind of exec situation)?

  • 6
    like mv and cp ? – Archemar Oct 12 '16 at 05:26
  • 9
    On Debian sh is symlink to dash. They behave different, when called as sh or as dash – Motte001 Oct 12 '16 at 15:48
  • @Archemar I don't think that mv and cp are symlinks, at least on CentOS 6.4 they aren't. – Alexej Magura Oct 12 '16 at 15:52
  • 21
    @AlexejMagura If you use something like busybox (common on rescue-discs and such), then pretty much everything (cp, mv, rm, ls, ...) is a symbolic link to busybox. – Baard Kopperud Oct 12 '16 at 16:20
  • 11
    I'm finding this really hard to ignore, so I'll say it: you probably mean "GNU" programs (gcc, bash, gunzip, most of the rest of the OS...), as Linux is just the kernel. – wizzwizz4 Oct 12 '16 at 17:50
  • 1
    You need to ask Dennis Ritchie, the original designer of Unix and C (Unix was designed by others as well but the C language was designed by Dennis Ritchie). In the beginning, the design of C was not rigidly determined and sometimes is the result of Dennis Ritchie's personal preferences. – Sam Hobbs Oct 12 '16 at 22:18
  • 11
    @wizzwizz4 What's wrong with "Typical Unix/Linux programs"? I read it like "Typical programs running on Unix/Linux". That's much better than your restriction to certain GNU programs. Dennis Ritchie was certainly not using any GNU programs. BTW the Hurd kernel is an example of a GNU program which does not have a main function... – rudimeier Oct 13 '16 at 11:49
  • 2
    That's incredibly pedantic @wizzwizz4 - people say "Windows programs" or "macOS apps" all the time; nobody assumes they're referring only to first-party bundlewares. – OJFord Oct 15 '16 at 18:22
  • 1
    @rudimeier Linux is one program. GNU is the collective name of the majority of the rest of the programs. If Linux hadn't come along before Hurd was finished, you probably wouldn't even have heard of Linux. Android is based on Linux, but you don't see that on this website because it's not Unix-like. GNU is Unix-like, but Linux can't even be compared to it. I'll stop this argument now, else it will go on for months. If you want to carry on we should take it to chat, but I recommend we don't. – wizzwizz4 Oct 16 '16 at 11:22
  • @OllieFord But people don't say "XNU apps", do they? People also don't attribute the creation of the majority of Windows to the creators of ntoskrnl, and don't assume that they could replace explorer, cmd, dwm, dasHost , lsass, wininit etc. and still have Windows, so long as they kept ntoskrnl. – wizzwizz4 Oct 16 '16 at 11:25
  • 1
    Why does argv include the program name Why shouldn't it? – Braden Best Oct 16 '16 at 19:13
  • 1
    @wizzwizz4 Right, Linux is a kernel, not an OS. Though most people mean GNU/Linux when they say "Linux" in the context of OS. It's a colloquialism, and we're all aware of Stallman's meme – Braden Best Oct 16 '16 at 19:25
  • 1
    @BradenBest Unfortunately, we are not all aware that "Linux" is often an abbreviation of GNU/Linux, especially in the context of OS. I personally think that "*nix" and "Unix-like" are better phrases than "Unix/Linux", not only because Linux isn't anything like Unix but also because it entirely discounts FreeBSD and other Unix-like operating systems. But each to their own, I suppose. – wizzwizz4 Oct 16 '16 at 20:36
  • @wizzwizz4 I agree, you should inform people when it's necessary, as a user ought to have a good understanding of the software they're using. But you shouldn't force them to use terminology they don't want to use. Nobody calls Windows 7 "NT6.1/ntoskrnl", or OS X Yosemite "OSX10.10/XNU", just as not many people call Linux "GNU/Linux". Together, GNU and Linux make up a nifty OS that is grater than the sum of its parts. We just happen to call this combination "Linux". In effect, there's "Linux" the kernel, and "Linux" the family of OS. As long as that's understood, there shouldn't be problems. – Braden Best Oct 16 '16 at 21:50
  • Basically, and this is in counter to the point you made about "XNU apps" and "ntoskrnl explorer", "Linux" isn't meant as the name of the kernel (instead we say "Linux kernel"); it's meant as the name of the OS that is formed when you combine Linux and GNU. Just as Windows is the name of the OS that is combined when you combine the Windows user-land software with ntoskrnl.exe, and OS X is the OS you get when you combine the Mac OS user-land with XNU. They could have called OS X "Apple XNU", and Microsoft could have called Windows 7 "Microsoft NT 6.1"... but they didn't. – Braden Best Oct 16 '16 at 21:55
  • Is the intent to wonder why argv[0] doesn't simply pass the first parameter? Or are you fundamentally wondering what the good is in a program knowing how it was called? Or perhaps why argv[0] exists, i.e., why have a "zeroeth" arg? – user2338816 Oct 17 '16 at 00:35
  • @wizzwizz4 "Don't flood the comments", I agree. Some final words: I agree that *nix is a good name for all unix-like systems and I even use it myself sometimes. I also wish Linux (the OS family) had a different name, as I recognize how strange it is to name the OS after one or two of its parts (both Linux and GNU/Linux are idiosyncratic). Something like, I dunno, Penguix, Tuxix, GNuxix, PenGNUx, PenGNUix. "My favorite system is Arch Pengnuix". Eh, that's really hard to pronounce, I keep slurring it as "Pain-wicks"...huh – Braden Best Oct 17 '16 at 17:31
  • @BradenBest They were thinking of LiGNUx, but both parties agreed that it sounded rather rubbish. Deleting earlier comment to make room for this one. – wizzwizz4 Oct 17 '16 at 17:44
  • related: https://stackoverflow.com/questions/2050961/is-argv0-name-of-executable-an-accepted-standard-or-just-a-common-conventi/42291142#42291142 – Ciro Santilli OurBigBook.com Oct 11 '18 at 08:30

11 Answers11

133

To begin with, note that argv[0] is not necessarily the program name. It is what the caller puts into argv[0] of the execve system call (e.g. see this question on Stack Overflow). (All other variants of exec are not system calls but interfaces to execve.)

Suppose, for instance, the following (using execl):

execl("/var/tmp/mybackdoor", "top", NULL);

/var/tmp/mybackdoor is what is executed but argv[0] is set to top, and this is what ps or (the real) top would display. See this answer on U&L SE for more on this.

Setting all of this aside: Before the advent of fancy filesystems like /proc, argv[0] was the only way for a process to learn about its own name. What would that be good for?

  • Several programs customize their behavior depending on the name by which they were called (usually by symbolic or hard links, for example BusyBox's utilities; several more examples are provided in other answers to this question).
  • Moreover, services, daemons and other programs that log through syslog often prepend their name to the log entries; without this, event tracking would become next to infeasible.
countermode
  • 7,533
  • 5
  • 31
  • 58
  • 18
    Examples of such programs are bunzip2, bzcat and bzip2, for which first two are symlinks to the third one. – Ruslan Oct 12 '16 at 07:37
  • 5
    @Ruslan Interestingly zcat is not a symlink. They seem to avoid the downsides of this technique using a shell script instead. But they fail to print a complete --help output because somebody who added options to gzip forgot to maintain zcat too. – rudimeier Oct 12 '16 at 09:17
  • 3
    For as long as I can remember, the GNU coding standards have discouraged the use of argv[0] to change program behavior (section "Standards for Interfaces Generally" in the current version). gunzip is a historical exception. –  Oct 12 '16 at 11:58
  • 19
    busybox is another excellent example. It can be called by 308 different names to invoke different commands: https://busybox.net/downloads/BusyBox.html#commands – Pepijn Schmitz Oct 12 '16 at 13:24
  • 2
    Many, many more programs also inject their argv[0] in their usage/help output instead of hard-coding their name. Some in full, some just the basename. – spectras Oct 15 '16 at 14:13
  • more importantly, programs can behave differently depending upon the name they were invoked under. wdel and wput. Busybox is a good example, and it has been a past (possibly unwise ) usage to provide security backdoors with a high argc use, and a particular name. – mckenzm Oct 17 '16 at 23:32
67

Plenty:

  • Bash runs in POSIX mode when argv[0] is sh. It runs as a login shell when argv[0] begins with -.
  • Vim behaves differently when run as vi, view, evim, eview, ex, vimdiff, etc.
  • Busybox, as already mentioned.
  • In systems with systemd as init, shutdown, reboot, etc. are symlinks to systemctl.
  • and so on.
muru
  • 72,889
  • 7
    Another one is sendmail and mail. Every single unix MTA comes with a symlink for those two commands, and is designed to emulate the original's behaviour when called as such, meaning that any unix program that needs to send mail knows exactly how they can do so. – Shadur-don't-feed-the-AI Oct 12 '16 at 09:51
  • 4
    an other common case: test and [: when you call the former, it handles an error if the last argument is ]. (on actual Debian stable these commands are two different programs, but previous versions and MacOs still uses the same program). And tex, latex and so on: the binary is the same, but looking how it was called, it choose the proper configuration file. init is similar. – Giacomo Catenazzi Oct 12 '16 at 19:25
  • 4
    Related, [ considers it an error if the last argument is not ]. – chepner Oct 13 '16 at 15:49
  • 1
    I guess this answers the second question, but not the first. I very much doubt some OS designer sat down and said »Hey, it would be cool if I had the same program doing different things just based on its executable name. I guess I'll include the name in its argument array, then.« – Joey Oct 18 '16 at 05:52
  • @Joey Yes, the wording is intended to convey that (Q: "Are there any ...?" A: "Plenty: ...") – muru Oct 18 '16 at 06:34
37

Historically, argv is just an array of pointers to the "words" of the commandline, so it makes sense to start with the first "word", which happens to be the name of the program.

And there's quite a few programs that behave differently according to which name is used to call them, so you can just create different links to them and get different "commands". The most extreme example I can think of is busybox, which acts like several dozen different "commands" depending on how it is called.

Edit: References for Unix 1st edition, as requested

One can see e.g. from the main function of cc that argc and argv were already used. The shell copies arguments to the parbuf inside the newarg part of the loop, while treating the command itself in the same way as the arguments. (Of course, later on it executes only the first argument, which is the name of the command). It looks like execv and relatives didn't exist then.

Anko
  • 4,526
dirkt
  • 32,309
  • 1
    please add references that back this up. – Lesmana Oct 12 '16 at 10:24
  • From a quick skimming, exec takes the name of the command to execute and a zero-terminated array of char pointers (best seen at http://minnie.tuhs.org/cgi-bin/utree.pl?file=V1/u0.s, where exec takes references to label 2 and label 1, and at label 2: appears etc/init\0, and at label 1: appears a reference to label 2, and a terminating zero), which is basically what execve does today minus envp. – ninjalj Oct 13 '16 at 19:49
  • 1
    execv and execl have existed "forever" (i.e., since the early to mid 1970s) — execv was a system call and execl was a library function that called it.   execve didn't exist then because the environment didn't exist then.   The other members of the family were added later. – G-Man Says 'Reinstate Monica' Oct 14 '16 at 05:41
  • @G-Man Can you point me to execv in the v1 source I linked? Just curious. – dirkt Oct 14 '16 at 06:56
23

Use cases:

You can use the program name to change the program behavior.

For example you could create some symlinks to the actual binary.

One famous example where this technique is used is the busybox project which installs only one single binary and many symlinks to it. (ls, cp, mv, etc). They are doing it to save storage space because their targets are small embedded devices.

This is also used in setarch from util-linux:

$ ls -l /usr/bin/ | grep setarch
lrwxrwxrwx 1 root root           7 2015-11-05 02:15 i386 -> setarch
lrwxrwxrwx 1 root root           7 2015-11-05 02:15 linux32 -> setarch
lrwxrwxrwx 1 root root           7 2015-11-05 02:15 linux64 -> setarch
-rwxr-xr-x 1 root root       14680 2015-10-22 16:54 setarch
lrwxrwxrwx 1 root root           7 2015-11-05 02:15 x86_64 -> setarch

Here they are using this technique basically to avoid many duplicate source files or just to keep the sources more readable.

Another use case would be a program which needs to load some modules or data at runtime. Having the program path makes you able to load modules from a path relative to the program location.

Moreover many programs print error messages including the program name.

Why:

  1. Because it's POSIX convention (man 3p execve):

argv is an array of argument strings passed to the new program. By convention, the first of these strings should contain the filename associated with the file being executed.

  1. It's C standard (at least C99 and C11):

If the value of argc is greater than zero, the string pointed to by argv[0] represents the program name; argv[0][0] shall be the null character if the program name is not available from the host environment.

Note the C Standard says "program name" not "filename".

rudimeier
  • 10,315
  • 3
    Doesn't this break if you reach the symlink from another symlink? – user541686 Oct 12 '16 at 06:55
  • 3
    @Mehrdad, Yes that's the downside and can be confusing for the user. – rudimeier Oct 12 '16 at 07:30
  • @rudimeier: Your 'Why' items are not really reasons, they're just a "homunculus", i.e. it just begs the question of why do the standard require this to be the case. – einpoklum Oct 15 '16 at 11:10
  • @einpoklum OP's question was: Why is the program name passed to the executable? I answered: Because POSIX and C standard tells us to do so. How you think that's not really a reason? If the docs I've quoted would not exist then probably many programs would not pass the program name. – rudimeier Oct 15 '16 at 23:35
  • The OP is effectively asking "WHY do the POSIX and C standards say to do this?" Granted the wording was at an abstracted level, but it seems clear. Realistically, the only way to know is to ask the originators. – user2338816 Oct 17 '16 at 00:30
23

In addition to programs altering their behaviour depending on how they were called, I find argv[0] useful in printing the usage of a program, like so:

printf("Usage: %s [arguments]\n", argv[0]);

This causes the usage message to always use the name through which it was called. If the program is renamed, its usage message changes with it. It even includes the path name it was called with:

# cat foo.c 
#include <stdio.h>
int main(int argc, char **argv) { printf("Usage: %s [arguments]\n", argv[0]); }
# gcc -Wall -o foo foo.c
# mv foo /usr/bin 
# cd /usr/bin 
# ln -s foo bar
# foo
Usage: foo [arguments]
# bar
Usage: bar [arguments]
# ./foo
Usage: ./foo [arguments]
# /usr/bin/foo
Usage: /usr/bin/foo [arguments]

It's a nice touch, especially for small special-purpose tools/scripts that might live all over the place.

This seems common practice in GNU tools as well, see ls for example:

% ls --qq
ls: unrecognized option '--qq'
Try 'ls --help' for more information.
% /bin/ls --qq
/bin/ls: unrecognized option '--qq'
Try '/bin/ls --help' for more information.
marcelm
  • 2,485
  • 3
    +1. I was going to suggest the same. Strange that so many people focus on changing behaviour and fail to mention probably the most obvious and much more widespread usage. – The Vee Oct 13 '16 at 11:51
6

One executes the program typing: program_name0 arg1 arg2 arg3 ....

So the shell should already divide the token, and the first token is already the program name. And BTW so there are the same indices on program side and on shell.

I think this was just a convenience trick (on very very beginning), and, as you see in other answers, it was also very handy, so this tradition was continued and set as API.

5

Basically, argv includes the program name so that you can write error messages like prgm: file: No such file or directory, which would be implemented with something like this:

    fprintf( stderr, "%s: %s: No such file or directory\n", argv[0], argv[1] );
user628544
  • 1,565
3

Another example of an application of this is this program, which replaces itself with... itself, until you type something that isn't y.

#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>

int main (int argc, char** argv) {

  (void) argc;

  printf("arg: %s\n", argv[1]);
  int count = atoi(argv[1]);

  if ( getchar() == 'y' ) {

    ++count;

    char buf[20];
    sprintf(buf, "%d", count);

    char* newargv[3];
    newargv[0] = argv[0];
    newargv[1] = buf;
    newargv[2] = NULL;

    execve(argv[0], newargv, NULL);
  }

  return count;
}

Obviously, kind of a contrived if interesting example, but I think this may have real uses -- for example, a self-updating binary, which rewrites its own memory space with a new version of itself that it downloaded or changed.

Example:

$ ./res 1
arg: 1
y
arg: 2
y
arg: 3
y
arg: 4
y
arg: 5
y
arg: 6
y
arg: 7
n

7 | $

Source, and some more info.

cat
  • 3,468
  • 4
  • 23
  • 51
1

The path to the program is argv[0], so that the program can retrieve configuration files etc. from its install directory.
This would be impossible without argv[0].

Thomas
  • 6,362
  • 2
    That's not a particularly good explanation - there's no reason we couldn't have standardised on something like (char *path_to_program, char **argv, int argc) for example – moopet Oct 17 '16 at 10:46
  • Afaik, most programs pull configuration from a standard location (~/.<program>, /etc/<program, $XDG_CONFIG_HOME) and either take a parameter to change it or have a compile-time option that bakes in a constant to the binary. – Xiong Chiamiov Oct 17 '16 at 18:32
1

ccache behaves this way in order to imitate different calls to compiler binaries. ccache is a compilation cache - the whole point is never to compile the same source code twice but instead return the object code from cache if possible.

From the ccache man page, "there are two ways to use ccache. You can either prefix your compilation commands with ccache or you can let ccache masquerade as the compiler by creating a symbolic link (named as the compiler) to ccache. The first method is most convenient if you just want to try out ccache or wish to use it for some specific projects. The second method is most useful for when you wish to use ccache for all your compilations."

The symlinks method involves running these commands:

cp ccache /usr/local/bin/
ln -s ccache /usr/local/bin/gcc
ln -s ccache /usr/local/bin/g++
ln -s ccache /usr/local/bin/cc
ln -s ccache /usr/local/bin/c++
... etc ...

... the effect of which is to allow ccache to snag any commands which would otherwise have gone to the compilers, thus allowing ccache to return a cached file or pass the command on to the actual compiler.

0

The following is a practical example to illustrate the usefulness of having the program name as argv[0] ($0 in bash):

To accomplish a certain task, I often need to read the manual pages. But this is often tiresome, and a lot of time is wasted when reading the man pages over and over again any time I need to use the same command or need to do the same thing in the future.

So I started making small, textual notes about each command I use the most, and use it in an outliner (emacs' outline-mode):

tcpdump help file

Using an outliner helps to collapse or show large portions of text, making navigation easier, and accessing desired information quicker. When used properly, it makes reading and maintaining documentation much more efficient.

To get help on a specific command, for example strace, instead of doing man strace, I did strace.help

In the beginning, I wrote aliases that looked like this:

alias strace.help='emacs ~/help/strace'
alias tcpdump.help='emacs ~/help/tcpdump'
alias ps.help='emacs ~/help/ps'
...

help aliases

but then I thought wait, this is really stupid. why should I have x aliases that all look the same? isn't this crying for refactoring? So I rewrote the aliases into a single aliase creating bash function, using eval inside a for loop, so the aliases were created dynamically, but that still seemed wrong...

Then I remembered the argv[0] trick that I saw while exploring the source code of some the most common unix/linux commands: write a single program that behaves differently depending on how it's called.

So I wrote a generic command.help script that looked like this:

#!/bin/bash
commandname=$(filename.path.basename "$0")
# make sure to stip the .help suffix
commandname=$(filename.ext.remove "$commandname")
if [[ -f ~/.bash_lib/help/$commandname ]]
then
    $EDITOR ~/.bash_lib/help/$commandname
else
    echo "no custom help available for $commandname"
fi

filename.path.basename and filename.ext.remove are short bash utils and showing their code has no relevance here.

command.help source code

I put that script in my $PATH, then I created multiple links command.help, with different names:

strace.help,
tcpdump.help,
ps.help
...

creating links

so now, anytime I add a new help note for a command, say zip, I only need to create a new link to command.help under the name zip.help, and it will magically open the help file for the zip command.

creating links

muru
  • 72,889
ychaouche
  • 998