Redirection

Revision as of 16:13, 13 September 2006

Redirection is sending the output of a program to somewhere other than where it would otherwise go - for example you can redirect the output of an ls command to a text file for later processing or to the grep command for filtering. Shell redirection operators include: >, >>, <, <<, and the ever popular |.

>  sends output to a file (may include special files such as /dev/null)
>> appends output to a file (without overwriting it)
<  read file to stdin
<< read to stdin from <<delimiter to delimiter (a HERE doc).
 |  sends output to a program (frequently, a system command like grep)

If you're using the bash or bourne shells, you also have some special options available to you: you can redirect standard input, standard output and standard error messages with far greater flexibility and reliability. Other shells such as csh are notably limited in redirection capability, making them better suited to interactive use than to shell programming or other complex uses.

Shell pipelines and redirection

There are three standard input/output file descriptors (fd) that are preconnected to the shell process running on your FreeBSD machine. Most commands that you would run from the command line expect these file descriptors to be open and accessible. The first (fd 0) provides a stream of input to your program, the second (fd 1) provides a stream of output, and the third (fd 2) provides a stream of diagnostic messages (usually to your terminal).

When you open a terminal, before running the shell the terminal device is opened three times to preconnect these file descriptors. The shell then inherits the file descriptors, and passes them on to each process run from the shell.

On FreeBSD systems it looks like this:

The example below demonstrates that by default, all of these input/output streams are directed to your terminal (color is added).

$ for i in stdin stdout stderr; do echo $i stuff > /dev/$i; done
 stdin stuff
stdout stuff
stderr stuff

Redirection means that, the file descriptor is temporarily reassigned to somewhere other than the terminal device (a file, a pipe, another file descriptor). When the next process inherits the open fd from the shell the stream of data is passed along, for example, by writing to stdout, which has been temporarily reassigned to the stdin for the next process by means of a pipe (|). This handing off of open file descriptors by the shell, from one sub-process to another, is called a "shell pipeline".

There are fd n (0-9) descriptors potentially available, but only 0-2 are preconnected to the shell; the others must be created from the shell. In some shells, a standard file descriptor can be detached and reassigned to another file descriptor for as long as the terminal device is open, using the exec command (examples appear below). This capability is just one case among many of why sh is useful for scripting, and tcsh (which lacks it) is not (see the FAQ).

Redirection in sh compared to tcsh

The c-shells (tcsh or csh) and the Bourne shells (sh or bash) do not handle redirection or piping in quite the same way.

tcsh and sh

1. Write output to a file

    % mycmd > out.txt

   stderr stuff

2. Append output to a file

    % mycmd >> out.txt

   stderr stuff

3. Redirect the output of a remote command to local.txt.

    % localcmd "remotecmd" > local.txt

   stderr stuff

4. Same command as above, showing only the changes compared to local.txt.
   Note: Many programs recognize '-' as a shortcut for '/dev/stdin'. These two commands are equivalent.

    % localcmd "remotecmd" | diff /dev/stdin local.txt

    % localcmd "remotecmd" | diff - local.txt

   stderr stuff

5. Direct stdout+stderr to file

    % mycmd >& out.txt

6. Sort lines of jumble.txt into sorted.txt
   Note: the sequence in which redirection appears is not important. All of the following are exactly equivalent.

    % <jumble.txt sort >sorted.txt

    % >sorted.txt sort <jumble.txt

    % sort < jumble.txt >sorted.txt

    % <jumble.txt>sorted.txt sort

   stderr stuff

7. Sort unique lines of jumble.txt into sorted.txt

    % <jumble.txt sort | uniq >sorted.txt

   stderr stuff

8. Sort HERE doc delimited by "lines"

    % <<lines sort
   

   ? a second line
   

   ? a first line
   

   lines
   a first line
   a second line
   stderr stuff

tcsh only

1. Discard errors, watch output (probably evil)
   Note: There is no reliable way to do this in tcsh. Here we exploit the fact that terminal reads from stdin.

    % ( mycmd > /dev/stdin ) > & /dev/null

   stdout stuff

    % ( mycmd > /dev/tty ) > & /dev/null

   stdout stuff

2. Append output to out.txt; discard messages

    % (mycmd >> out.txt) >& /dev/null

3. Write output to out.txt; store and watch errors
   Note: Compare the same task in sh.

    % ( mycmd > out.txt ) | & tee err.txt

   stderr stuff

sh only

1. Discard errors, watch output

   $ mycmd 2> /dev/null

   stdout stuff

2. Append output to out.txt; discard messages

   $ mycmd 2> /dev/null >> out.txt

3. Write output to out.txt; store and watch errors
   Note: Compare the same task in tcsh.

   $ mycmd 2>&1 > out.txt | tee err.txt

   stderr stuff

4. Write messages to err.txt; write output to out.txt and copy output to terminal

   $ mycmd 2> err.txt | tee out.txt

   stdout stuff

5. Assign a variable from stored.txt

   $ <stored.txt read var; mycmd $var

   stderr stuff

   stdout stuff

6. Assign first three lines of stored.txt to three different variables

   $ exec 3<&0; exec <stored.txt; read v1; read v2; read v3; exec 0<&3 3<&-; echo $v1 $v2 $v3

   stderr stuff

   stdout stuff

7. Use all unique lines in stored.txt as an argument variable for mycmd, appending to result.txt

   $ exec 3<&0; exec <stored.txt; sort | uniq | while read line; do mycmd $line >> result.txt ; done; exec 0<&3 3<&-;

   stderr stuff

Note: By default, redirection pointed right represents stdout, so that these two commands are exactly equivalent:

$ mycmd 1> out.txt

$ mycmd > out.txt

stderr stuff

Note: By default, redirection pointed left represents stdin, so that these two commands are exactly equivalent:

$ mycmd 0< in.txt

$ mycmd < in.txt

stderr stuff

stdout stuff

Note: To close a file descriptor, say n<&-

$ mycmd >out.txt 1<&-;

-sh: fails without messages

-bash: mycmd: write error: Bad file descriptor

A little more about fd n in sh

Let's say you want to send output to your screen and errors to another command. You can't just do

samizdata# myprogram 1>&2 2>&1 | command > errors.txt

because when you switched output to the messages file descriptor, it's done right away. When you tried to switch errors to output there's nothing left to pipe. It's exactly as though you had said

samizdata# myprogram 1>/dev/stderr | command > errors.txt

command receives no data, because it's all been sent to your screen. errors.txt will be empty. If you had tried it the other way, it's just as though you would have said:

samizdata# myprogram 2>/dev/stdout | command > errors.txt

In which case, errors and output are both piped to command, which was not what you wanted.

This is where the other, normally unused, file descriptors 3-9 come in. You can use them as place-holders, like this:

samizdata# myprogram 3>&2 2>&1 1>&3 | command > errors.txt

Let's step through what this does.

1. fd 3 points to stderr
2. fd 2 points to stdout
3. fd 1 points to fd 3 (which is connected to stderr)
4. | passes the open file descriptor connected to stdout (fd 2) making it stdin for command
5. > redirects the output of command to errors.txt.

So, what you'll see on your screen is the output of myprogram on stderr (not passed to command), followed by errors (if any) from command. In short, what's been accomplished is, a swap of errors for output, from myprogram to command, the output of which is redirected to errors.txt. There is no reliable way to do the same thing in tcsh.

Also, see the man page for mkfifo, a utility for creating arbitrary FIFO files.