Great Command-line Combinations

Tools with graphical user interfaces (GUIs) can be easy to learn. All of their commands and options are typically laid out on menus and dialog boxes, making it easy to discover what the tool can do. If “easy” and “intuitive” are your main criteria for programs, then a GUI tool may always be the right choice.

But the simplicity of GUI interfaces can also make them weak and inflexible. For instance, imagine that you’re a programmer and you’re cleaning out your Linux filesystem. You’d like to find object files that haven’t been accessed in six months or more, all through your filesystem, and run make clean in those directories. Can a graphical filesystem browser do that in one step? Probably not. However, the power of the many non-GUI Linux utility programs, joining forces through the shell, would let you make quick work of that job.

Let’s look at some powerful things you can do on a command-line. Even if you don’t want to do exactly these things, you’re likely to get some ideas for related uses. This “related-ness” ability of the shell and its command-line — letting you combine tools to do just what you need to do — is the very spirit of “Power Tools.”

Faster than Emacs and vi

If you make a mistake on the command line, how can you fix it? Most modern shells let you recall and change previous command lines by pressing the up-arrow key (or CONTROL-p or ESCAPE-k) and editing the command-line using Emacs or vi commands. Before shells did this, though, the C shell had another way to edit: history substitution, a set of shell operators starting with the ! (often called “bang”) character. Some of these operators are still faster and easier to use than command-line editing. Here are three:

1. !$ (“bang dollar”) expands to the last argument on the previous command line. This is usually a filename.

Let’s see an example. You just listed the file /a/b/foo.c. Now you’d like to edit it with vi. There’s no need to move to the previous line and then use editing commands to change ls -l to vi. You can get the previous filename with just two keystrokes: !$. The shell shows the expanded command line before running it:

$ ls -l /a/b/foo.c
-rw-r--r-- ... Apr 20 11:31 /a/b/foo.c
$ vi !$
vi /a/b/foo.c
... vi starts ...

2. !* expands to all arguments on the previous command line. This lets you repeat a list of (say) filenames with a new command or fix a command name you mistyped.

For example, there’s no such command emsca [sic], but the typo is easily remedied with !*:

% emsca /b/c/foo.d /e/f/bar.g
emsca: Command not found.
% emacs !*
emacs /b/c/foo.d /e/f/bar.g
...emacs starts...

3. The sequence ^old^new changes the string old on the previous command line to new.

So, in the previous example, we could have typed ^sca^acs. A shorter version of this operator — a single ^ — is great for removing extra characters (with no replacement). For example, if you type two gs instead of one, repeat the command and remove one of them with ^g:

$ rm /some/longg/pathname
rm: /some/longg/pathname: Not found
$ ^g
rm /some/long/pathname

Merging Output with a Subshell

How can you collect the output of a lot of commands and use it somewhere else? In a GUI, you can copy from each command’s window and paste to another — assuming that all of those windows are “copy-able.”

In a shell, you can collect command outputs by using a subshell. The operator ( ) runs one or more commands in another shell process. You can also redirect the output of that subshell as you would any other process. Subshells let you run multiple commands in a subshell and collect all of their outputs at once.

For instance, you’d like to email your project leader a copy of the make output from three different project directories. Here’s one way:

% (cd adir; make; cd ../bdir; make; \
?  cd ../cdir; make) | mail

The subshell operators run a series of commands separated by semicolons (;), the shell’s command separator. Typing a backslash (\) at the end of a line tells the shell to keep reading another line, so the shell prints its secondary prompt (here, a ?). After the closing parenthesis, we pipe the output of the subshell, which is the output of all its commands. The mail program reads the text from its standard input, via the pipe, and sends it to

What if you redirect the subshell’s output to a file? Which directory is the output written into? (In other words, do the cd commands in the subshell have any effect?) Changes in the subshell don’t affect the parent shell. The parent shell stays in whatever current directory it had.

For instance, in the command below, the make_outputs file is created in the current working directory before the subshell even starts running:

% (cd adir; make; cd ../bdir; make; \
? cd ../cdir; make) > make_outputs

This kind of technique works anywhere, not just with cd commands. It’s also a nice lead-in to the next handy command-line technique: redirecting the output of a loop.

Let a Loop Do the Work, Part I

Bourne-type shells like bash don’t only let you redirect the output (and input) of subshells. You can redirect the output of almost any shell construct, including if and case statements (tests) and loops. (This doesn’t work in csh or tcsh, though.)

You can use this technique to keep a log of what happens in a loop. For instance, start a loop by running echo to print a message or date to show the current time. Redirect the output of the loop to a file or pipe it to some command, and you gather the standard outputs of all commands within the loop, one after another.

Loops aren’t just for shell scripts! You can type them interactively at a prompt. Let’s re-run the previous example in bash with a for loop, adding an echo to output a label before each make output. (The bash secondary prompt is >.)

$ for d in [a-c]dir
> do
>   cd "$d"
>   echo "======= $d ======="
>   make
>   cd ..
> done | mail

The shell expands the wildcard [a-c]dir into adir bdir cdir and stores each of those names in the shell variable d before running the commands from do to done. The output of the loop is piped to the mail program. Joe gets a message like this:

======= adir =======
...make output from adir...
======= bdir =======
...make output from bdir...
======= cdir =======
...make output from cdir...

Of course, loops aren’t only useful when you redirect their outputs. They’re handy for any series of commands you need to repeat.

Let’s use the netpbm utilities (from to make PNG-format thumbnails, 100 pixels high, from a bunch of TIFF-format files. Listing One shows the loop typed at a shell prompt.

Listing One: Converting image files with netpbm

$ for file in *.tif
> do
>   echo "===== doing $file ===="
>   tifftopnm $file |
>   pnmscale -height 100 |
>   pnmtopng > ${file%.*}.png
> done
==== doing img0321.tif ====
tifftopnm: writing PPM file
==== doing img0343.tif ====
tifftopnm: writing PPM file
$ ls
img0321.png  img0343.png  img0369.png
img0321.tif  img0343.tif  img0369.tif

To get the same base filename with a .png extension (for example, to get img0321.png from img0321.tif, we’re using the parameter expansion operator ${file%.*}. It expands the filename from $file, removing everything from the dot to the end. That reads “expand $file, stripping off the smallest possible string from the end of the value that matches a dot followed by any number of other characters.” Then we add .png to the end of the stripped name. (In csh and tcsh, use ${file:r} instead.)

If you had a hundred TIFF files to convert, this could save you a lot of time!

Let a Loop Do the Work, Part II

Let’s get back to the example at the start of this article: searching a directory tree and removing old object files. There are several ways to tackle this problem. One is by writing a script in Perl or another language that’s good at handling Linux filesystems and file properties. But, if you have a shell waiting in your terminal window, why not type the commands you need right there?

If you’re new to shells — or the only “shell” you’ve used is COMMAND from Microsoft Windows — this may seem a bit risky: typing a command to clean your filesystem, not even writing a script. Well, you do need experience to understand what the command is doing. (But then, you need experience to write a script, too!)

A shell is actually a programming language interpreter. Its language is command-lines. Let’s use it. We’ll pair a loop with the Linux find utility (covered in the September 2002 column.)

We saw earlier that you can redirect the output of a Bourne-type shell loop. You can also redirect the input to a loop. Listing Two shows the technique (the lines have been numbered for reference). A while loop runs a command over and over until that command returns a nonzero exit status. The command to use here is read, which reads one line of input from its standard input and stores it in a shell variable.

Listing Two: Cleaning a directory tree

1: $ find /proj -type d -print |
2: > while read dir
3: > do
4: >   cd "$dir" || break
5: >   test "$(echo *.o)" = '*.o' && continue
6: >   test -z "$(find *.o -atime +180 -type f -print)" && continue
7: >   echo "Cleaning $dir"
8: >   make clean
9: > done 2>&1 | less

The output of find from line 1 is piped into a while loop, which starts on line 2. The standard input comes from find, via the pipe. When there’s no more output from find, read returns a nonzero exit status and the loop terminates.

Inside the loop, we start by changing the current directory at line 4. If that fails, cd returns a nonzero status and the || operator executes break to end the loop. (To make things simple, we gave find an absolute path, /proj — so it will output absolute pathnames. If we’d used relative pathnames, we’d need to add another cd command at the end of the loop to get back to the starting directory before the next pass through the loop.)

On line 5, $(echo *.o) asks the shell to expand that wildcard. If the shell returns a literal *.o, there were no matching object files, so the test returns a “true” (zero) status and the && operator executes continue, which re-starts the loop (gets another directory name from find). Line 6 is executed only if there are some object files (if the script didn’t branch at line 5). test -z tests for empty output from find *.o. If none of those object files have been accessed in more than 180 days, find produces no output, the test -z succeeds, and the continue re-starts the loop with the next directory. But if test -z fails, there are old files, so lines 7 and 8 run to echo the name of the directory to the terminal and do make clean.

Line 9 redirects the output of the loop — and the output of all commands inside it — to the less pager. Before the pipe to less, the 2>&1 operator merges all error messages from the standard error channel (file descriptor 2) onto the standard output (file descriptor 1) so all messages are piped to less. This setup makes sure that you can see all of the loop outputs; they won’t scroll off.

This long-winded explanation may make the technique seem like a lot of work, but if you understand the shell techniques used, you can type a loop like this with only a minute or two of thought. (That’s a lot faster than clicking through lots of subdirectories with a GUI file browser.)

To enhance the script, you might add more tests: for instance, skipping RCS directories, checking that a makefile exists, and so on. You could do a dry run, too, using ls -lut to list the object files, or run make -n clean to show what make would do without doing it.

Because this is typed on the command line and on-the-fly, you probably won’t add comments or apply clear coding techniques that you’d use in a script file that will stay around for a while. However, command line scripts like this one can easily be captured (say, using cut-and-paste from an xterm window) and refined and saved for use again and again.

Power Tip: Being Nice

A process that takes a lot of system resources can “steal” those resources from other processes that need them more. From a shell, you can tell Linux to give certain programs fewer system resources. Simply put nice before the command name. (You probably shouldn’t start interactive or GUI programs this way. If your system gets busy, a niced program can “freeze.”)

Here’s a rewritten image-processing loop with the CPU-intensive commands niced:

$ for file in *.tif
> do
>   echo "===== doing $file ===="
>   nice tifftopnm $file |
>   nice pnmscale -height 100 |
>   nice pnmtopng > ${file%.*}.png
> done

It’s a bit more efficient to nice the entire loop. Put the loop into a file (like do_thumbnails). Then either make the file executable and type nice ./do_thumbnails or start a niced shell to read the file, as in nice bash do_thumbnails.

Jerry Peek is a freelance writer and instructor who has used Unix and Linux for over 20 years. He’s happy to hear from readers; see

[Read previous article] [Read next article]
[Read Jerry’s other Linux Magazine articles]