The result of ls * , ls ** and ls ***

Question

I know using the command ls will list all the directories. But what does the ls * command do ? I used it and it just lists the directories. Does the star in front of ls mean how deep it will list the directories?

Answer 1

ls lists the files and content of directories it is being passed as arguments, and if no argument is given, it lists the current directory. It can also be passed a number of options that affect its behaviour (see man ls for details).

If ls is being passed an argument called *, it will look for a file or directory called * in the current directory and list it just like any other. ls doesn't treat the * character in any other way than any other one.

However if ls * is a shell command line, that is code in the language of a Unix shell, then the shell will expand that * according to its globbing (also referred to as Filename Generation or Filename/Pathname Expansion) rules.

While different shells support different globbing operators, most of them agree on the simplest one *. * as a pattern means any number of characters, so * as a glob will expand to the list of files in the current directories that match that pattern. There's an exception however that a leading dot (.) character in a file name has to be matched explicitly, so * actually expands to the list of files and directories not starting with . (in lexical order).

For instance, if the current directory contains the files called ., .., .foo, -l and foo bar, * will be expanded by the shell to two arguments to pass to ls: -l and foo bar, so it will be as if you had typed:

ls -l "foo bar"

or

'ls' "-l" foo\ bar

Which are three ways to run exactly the same command. In all 3 cases, the ls command (which will probably be executed from /bin/ls from a lookup of directories mentioned in $PATH) will be passed those 3 arguments: "ls", "-l" and "foo bar".

Incidentally, in this case, ls will treat the first (strictly speaking second) one as an option.

Now, as I said, different shells have different globbing operators. A few decades ago, zsh introduced the **/ operator¹ which means to match any level of subdirectories, short for (*/)# and ***/ which is the same except that it follows symlinks while descending the directories.

A few years ago (July 2003, ksh93o+), ksh93 decided to copy that behaviour but decided to make it optional, and only covered the ** case (not ***). Also, while ** alone was not special in zsh² (just meant the same as * like in other traditional shells since ** means any number of character followed by any number of characters), in ksh93, ** meant the same as **/* (so any file or directory below the current one (excluding hidden files)³.

bash copied ksh93 a few years later (February 2009, bash 4.0), with the same syntax but an unfortunate difference: bash's ** was like zsh's ***, that is it was following symlinks when recursing into sub-directories which is generally not what you want it do and can have nasty side effects. It was partly fixed in bash-4.3 in that symlinks were still followed, but recursion stopped there. It was fully fixed in 5.0.

yash added ** in version 2.0 in 2008, enabled with the extended-glob option. Its implementation is closer to zsh's in that ** alone is not special. In version 2.15 (2009), it added *** like in zsh and two of its own extensions: .** and .*** to include hidden dirs when recursing (in zsh, the D glob qualifier (as in **/*(D)) will consider hidden files and directories, but if you only want to traverse hidden dirs but not expand hidden files, you need ((*|.*)/)#* or **/[^.]*(D)).

fish also supports **. Like earlier version of bash, it follows symlinks when descending the directory tree. In that shell however **/* is not the same as **. ** is more an extension of * that can span several directories. In fish, **/*.c will match a/b/c.c but not a.c, while a**.c will match a.c and ab/c/d.c and zsh's **/.* for instance has to be written {,**/}.*. There, *** is understood as ** followed by * so the same as **.

tcsh also added a globstar option in V6.17.01 (May 2010) and supports both ** and *** à la zsh.

So in tcsh, bash and ksh93, (when the corresponding option is enabled (globstar)) or fish, ** expands all the files and directories below the current one, and *** is the same as ** for fish, a symlink traversing ** for tcsh with globstar, and the same as * in bash and ksh93 (though it's not impossible that future versions of those shells will also traverse symlinks).

Above, you'll have noticed the need to make sure none of the expansions is interpreted as an options. For that, you'd do:

ls -- *

Or:

ls ./*

There are some commands (it doesn't matter for ls) where the second is preferable since even with the -- some filenames may be treated specially. It's the case of - for most text utilities, cd and pushd and filenames that contain the = character for awk for instance. Prepending ./ to all the arguments removes their special meaning (at least for the cases mentioned above).

It should also be noted that most shells have a number of options that affect the globbing behaviour (like whether dot files are ignored or not, the sorting order, what to do if there's no match...), see also the $FIGNORE parameter in ksh

Also, in every shell but csh, tcsh, fish and zsh, if the globbing pattern doesn't match any file, the pattern is passed as an unexpanded argument which causes confusion and possibly bugs. For instance, if there's no non-hidden file in the current directory

ls *

Will actually call ls with the two arguments ls and *. And as there's no file at all, so none called * either, you'll see an error message from ls (not the shell) like: ls: cannot access *: No such file or directory, which has been known to make people think that it was ls that was actually expanding the globs.

The problem is even worse in cases like:

rm -- *.[ab]

If there's no *.a nor *.b file in the current directory, then you might end up deleting a file called *.[ab] by mistake (csh, tcsh, and zsh would report a no match error and wouldn't call rm (and fish doesn't support the [...] wildcards)).

If you do want to pass a literal * to ls, you have to quote that * character in some way as in ls \* or ls '*' or ls "*". In POSIX-like shells, globbing can be disabled altogether using set -o noglob or set -f (the latter not working in zsh unless in sh/ksh emulation).

---

^{¹ While (*/)# was always supported, it was first short-handed as ..../ in zsh-2.0 (and potentially before), then ****/ in 2.1 before getting its definitive form **/ in 2.2 (early 1992)}

^{² The globstarshort option, has since be added (in 2015) to allow ** and *** being used instead of **/* and ***/* respectively}

^{³ See also these few more oddities with the ksh93 globstar design, some of which were copied by bash.}.

Answer 2

The command ls defaults to ls .: List all entries in the current directory.

The command ls * means 'run ls on the expansion of the * shell pattern'

The * pattern is processed by the shell, and expands to all entries in the current directory, except those that start with a .. It will go one level deep.

The interpretation of double or triple * patterns depend on the actual shell used.

* is a wildcard that matches 0 or more characters. Some modern shells will recurse into subdirectories on seeing the ** pattern.

Answer 3

You can demystify the whole process by typing echo instead of ls first, to see what the command expands to:

$ echo *
Applications Downloads Documents tmp.html

So in this case, ls * expands to ls Applications Downloads Documents tmp.html

$ echo **
Applications Downloads Documents tmp.html

$ echo ***
Applications Downloads Documents tmp.html

So no change. This assumes you're using bash as your shell -- most people are, and different shells have different behavior. If you're using ash or csh or ksh or zsh, you may expect things to work differently. That's the point of having different shells.

So lets try something different (still with bash) so we get an idea of the the globbing (*) operator can do for us. For example, we can filter by part of the name:

$ echo D*
Downloads Documents

And interestingly, a trailing slash is an implicitly part of any directory name. So */ will yield only the directories (and symbolic links to directories):

$ echo */
Applications/ Downloads/ Documents/

And we can do some filtering at multiple levels by putting slashes in the middle:

$ echo D*/*/
Documents/Work/ /Documents/unfinished/

Since the Downloads directory doesn't contain any subdirectories, it does not end up in the output. This is very useful for just examining the files you want. I use commands like this all the time:

$ ls -l /home/*/public_html/wp-config.php

This lists, if there are any, all the wp-config.php files that exist at the base level of any user's public_html directory. Or perhaps to be more complete:

$ find /home/*/public_html/ -name wp-config.php

This will find any wp-config.php files in any user's public_html directories or any of their subdirectories, but it will operate more efficiently than just find /home/ -name wp-config.php because it won't examine anything but the public_html directories for each of the users.

Answer 4

In some shells, including bash 4.x with the globstar option enabled, ** will perform a recursive glob, descending matched directories. Additional asterisks don't further modify this operation.

Answer 5

If you want to "dive deep", use ls -R (recursive) option, or use find, like so:

find . -ls

"find" will dive down to the bottom of the directory tree (as will 'ls -R'), and has many more options, like listing directories (-type d), files only (-type f) or showing files having other characteristics (no user in /etc/passwd, specific permissions, and a whole lot more). "find" is also somewhat safer in scripting (due to inconsistent globbing rules between shells, as well as special escapes for files having dashes, etc).

shell wildcard globbing won't work with just an asterisk '*' on dotfiles. To list dotfiles only, use:

ls .??*