Why is there a /bin/echo and why would I want to use it?
I noticed that there is a binary executable /bin/echo
on my Ubuntu MATE 17.04 system.
I thought, that's odd, because
$ type echo
echo is a shell builtin
Cursory testing suggests that /bin/echo
does the same sort of thing as the Bash builtin echo
:
$ /bin/echo foo
foo
$ /bin/echo $USER
zanna
So, why is there another version of echo
separate from the Bash program, and why or when would I want to use it?
Solution 1:
If you open up a bash
prompt and type in an echo
command, that uses a shell builtin rather than running /bin/echo
. The reasons it is still important for /bin/echo
to exist are:
- You're not always using a shell. Under a variety of circumstances, you run an executable directly and not through a shell.
- At least in theory, some shells don't have an
echo
builtin. This is not actually required.
To expand on #1, suppose you wanted to move all the regular files whose names started with abc
anywhere in src
to dest
. There are several ways to do that but one of them is:
find src -name 'abc*' -type f -exec mv -nv {} dest/ \;
But suppose, rather than just running that, you want to see every command that will be run first. Well, then you can prepend echo
to the command, just as you might in other contexts:
find src -name 'abc*' -type f -exec echo mv -nv {} dest/ \;
But find
doesn't use a shell. That runs /bin/echo
.
Besides find
with -exec
or -execdir
, the /bin/echo
executable will be called by other programs that themselves run programs but not through a shell. This happens with the xargs
command (which is related to find
), as well as in a number of other contexts, such as the Exec=
line of a .desktop
file. Another example is when you run sudo echo
, which can be handy for testing if sudo
is working.
Similarly, some shells have a printf
builtin but /usr/bin/printf
also exists.
A less common possible reason you might deliberately use /bin/echo
is if you were relying on the differences between it and the echo
command provided by your shell. man echo
documents /bin/echo
; help echo
in bash
documents the bash
builtin. echo
is not very portable, because different implementations--both across operating systems and across shells on the same operating system--support different options (e.g., -e
) and differ in their treatment of backslashes. Of course, it's better to avoid relying on such details, and use printf
instead, which is far more portable.
In bash
, you can make the type
builtin show /bin/echo
as well--assuming /bin
is in your $PATH
as it always should be--by passing it the -a
flag:
$ type -a echo
echo is a shell builtin
echo is /bin/echo
Solution 2:
Eliah's done a great job of answering this, but I want to comment about the "why is there another version of echo
separate from the Bash program" part. That's the wrong question.
The right question is: why is this a builtin in the first place, when it could have been (and is) a perfectly fine external command?
For simplicity, take a look at the builtins in dash, a measly 38 (bash has 61, for comparison, going by the output of compgen -b
):
. continue getopts readonly type
: echo hash return ulimit
[ eval jobs set umask
alias exec kill shift unalias
bg exit local test unset
break export printf times wait
cd false pwd trap
command fg read true
How many of these need to be builtins? [
, echo
, false
, printf
, pwd
, test
, and true
don't need to be builtins: They don't do anything that only a builtin can do (affect or obtain shell state that's not available to external commands). Bash's printf
at least takes advantage of being a builtin: printf -v var
saves the output to the variable var
. time
in bash is also special: by being a keyword, you can time arbitrary command lists in bash (dash doesn't have a time
equivalent). pwd
doesn't need to be a builtin either - any external command is going to inherit the current working directory (and it's an external command too). :
is an exception - you need a NOP, and :
is it. The rest do actions that an external command can easily do.
So, a fifth of these builtins don't need to be builtins. Why, then? The dash
manpage* actually explains in passing why these are builtins (emphasis mine):
Builtins This section lists the builtin commands which are builtin because they need to perform some operation that can't be performed by a separate process. In addition to these, there are several other commands that may be builtin for efficiency (e.g. printf(1), echo(1), test(1), etc).
That's pretty much it: these builtins are there because they're used so often, interactively and in scripts, and their functionality is simple enough, that the shell can do the job. And so it happens: some (most?) shells took on the job.** Go back to the sh
from 2.9 BSD, and you won't find an echo
builtin.
So, it's entirely possible a minimal shell can skip implementing such commands as builtins (I don't think any current shell does, though). The GNU coreutils project doesn't assume that you're going to run them in a particular shell, and POSIX requires these commands. So, coreutils provides these anyway, and skips those which don't have any meaning outside of the shell.
* This is almost identical to the corresponding manpage text for the Almquist shell, which is what dash, the Debian Almquist shell, is based on.
** zsh
takes this idea to the extreme: the commands you get by loading various modules, like zmv
, are things you wouldn't think a shell need even get into. At that point, the real question is: why would you use bash instead of zsh, which has all these builtins?