<- Tests and Conditionals | Compound Commands ->

Input And Output

This basic principle of computer science applies just as well to applications started through BASH. BASH makes it fairly easy to play around with the input and output of commands, which gives us great flexibility and incredible opportunities for automation.

1. File Descriptors

Input and output from and to processes occurs via so-called File Descriptors (in short: FDs). FDs are kind of like pointers to sources of data. When something reads from or writes to that FD, the data is being read from or written to the FD's data source. FDs can point to regular files, but they can also point to other data sources, like sockets, pipes, or devices.

By default, every new process starts with three FDs. They are referred to by the names Standard Input, Standard Output and Standard Error. In short, they are respectively called stdin, stdout and stderr. In an interactive shell, or in a script running on a terminal, the Standard Input is how bash sees the characters you type on your keyboard. The Standard Output is where the program sends most of its normal information so that the user can see it, and the Standard Error is where the program sends its error messages. Be aware that GUI applications work in the same way; but the actual GUI doesn't work via these FDs. GUI applications can still read and write from and to the standard FDs, but they usually don't. Usually, they do all the user interaction via that GUI; making it hard for BASH to control. As a result, we'll stick to simple terminal applications. Those, we can easily feed data on their "Standard Input", and read data from on their "Standard Output" and "Standard Error".

Let's make these definitions a little more concrete. Here's a demonstration of how "Standard Input" and "Standard Output" work:

    $ read -p "What is your name? " name; echo "Good day, $name.  Would you like some tea?"
    What is your name? lhunath
    Good day, lhunath.  Would you like some tea?

read is a command that reads information from stdin and stores it in a variable. We specified name to be that variable. Once read has read a line of information from stdin, it finishes and lets echo display a message. echo uses stdout to send its output to. stdin is connected to your terminal's input device, which is probably your keyboard. stdout is connected to your terminal's output device, which we assume is your computer monitor. As a result, you can type in your name and are then greeted with a friendly message on your monitor, offering you a cup of tea.

So what is stderr? Let's demonstrate:

    $ rm secrets
    rm: cannot remove `secrets': No such file or directory

Unless you have a file called secrets in your current directory, that rm command will fail and show an error message explaining what went wrong. Error messages like these are by convention displayed on stderr. stderr is also connected to your terminal's output device, just like stdout. As a result, error messages display on your monitor just like the messages on stdout. However, this separation makes it easy to keep errors separated from the application's normal messages. Some people like to use wrappers to make all the output on stderr red, so that they can see the error messages more clearly. This is not generally advisable, but it is a simple example of the many options this separation provides us with. Another common practice is writing stderr to a special log file.

• Good Practice:
  Remember that when you create scripts, you should send your custom error messages to the stderr FD. This is a convention and it is very convenient when applications follow the convention. As such, so should you! You're about to learn redirection soon, but let me show you quickly how it's done:

    echo "Uh oh.  Something went really bad.." >&2

• File Descriptor: A numeric index referring to one of a process's open files. Each command has at least three basic descriptors: FD 0 is stdin, FD 1 is stdout and FD 2 is stderr.

2. Redirection

The most basic form of input/output manipulation in BASH is Redirection. Redirection is used to change the data source or destination of an application's FDs. That way, you can send the application's output to a file instead of the terminal, or have the application read from a file instead of from the keyboard.

Redirection, too, comes in different shapes. There's File Redirection, File Descriptor manipulation, Heredocs and Herestrings.

• In The Manual: Redirections

• Redirection: This is the practice of changing a certain FD to read its input from or send its output to elsewhere.

2.1. File Redirection

File Redirection is the most basic form of redirection. I'll start with this so you can grasp the concept of redirection well.

    $ echo "The story of William Tell.
    >
    > It was a cold december night.  Too cold to write." > story
    $ cat story
    The story of William Tell.

    It was a cold december night.  Too cold to write.

As a result, the echo command will not send its output to the terminal; rather, the > story redirection changes the destination of the stdout FD so that it now points to a file called story. Be aware that before the echo command is executed, BASH first checks to see whether that file story actually exists. If it doesn't, it is created as an empty file, so that the FD can be pointed to it. This behaviour can be toggled with Shell Options (see later).

It should be noted that this redirection is in effect only for the single echo command it was applied to. Other commands executed after that will continue sending their output to the script's stdout location.

We then use the application cat to print out the contents of that file. cat is an application that reads the contents of all the files you pass it as arguments. It then outputs each file one after another on stdout. In essence, it concatenates the contents of all the files you pass it as arguments.

Warning: Far too many code examples and shell tutorials on the Internet tell you to use cat whenever you need to read the contents of a file. This is highly ill-advised! cat only serves well to concatenate contents of multiple files together, or as a quick tool on the shell prompt to see what's inside a file. You should NOT use cat to read from files in your scripts. There will almost always be far better ways to do this. Please keep this warning in mind. Useless use of cat will merely result in an extra process to create, and often results in poorer read speed because cat cannot determine the context of what it's reading and the purpose for that data.

When we use cat without passing any kind of arguments, it obviously doesn't know what files to read. In this case, cat will just read from stdin instead of from a file (much like read). Since stdin is normally not a regular file, starting cat without any arguments will seem to do nothing:

    $ cat

It doesn't even give you back your shell prompt! What's going on? cat is still reading from stdin, which is your keyboard. Anything you type now will be sent to cat. As soon as you hit the Enter key, cat will do what it normally does: it will display what it reads on stdout, just the same way as when it displayed our story on stdout:

    $ cat
    test?
    test?

Why does it say test? twice now? Well, as you type, your terminal shows you all the characters that you send to stdin before sending them there. That results in the first test? that you see. As soon as you hit Enter, cat has read a line from stdin, and shows it on stdout, which is also your terminal; hence, resulting in the second line: test?. You can press Ctrl+D to send your terminal the End of File character. That'll cause cat to think stdin has closed. It will stop reading from it and return you to your prompt. Let's use file redirection to attach a file to stdin, so that stdin is no longer reading from our keyboard, but instead, now reads from the file:

    $ cat < story
    The story of William Tell.

    It was a cold december night.  Too cold to write.

The result of this is exactly the same as the result from our previous cat story; except this time, the way it works is a little different. In our first example, cat opened an FD to the file story and read its contents through that FD. In the second example, cat simply reads from stdin, just like it did when it was reading from our keyboard. However, this time, the < story operation has modified stdin so that its data source is the file story rather than our keyboard.

Redirection operators can be preceded by a number. That number denotes the FD that it changes.

Let's summarize with some examples:

• command > file: Send the stdout of command to file.

• command 1> file: Send the stdout of command to file. Since stdout was open on FD 1, that's the number we precede the redirection operator with. Note, the effect of this is identical to the previous example, because FD 1 is the default FD for the > redirection operator.

• command < file: Use the contents of file when command reads from stdin.

• command 0< file: Use the contents of file when command reads from stdin, exactly as in the previous example, since FD 0 (stdin) is the default for the < redirection operator.

The number for the stderr FD is 2. So, let's try sending the output of stderr to a file:

    $ for homedir in /home/*
    > do rm "$homedir/secret"
    > done 2> errors

In this example, we're looping over each directory (or file) in /home. We then try to delete the file secret in each of them. Some homedirs may not have a secret, or we may not have permission to remove it. As a result, the rm operation will fail and send an error message on stderr.

You may have noticed that our redirection operator isn't on rm, but it's on that done thing. Why is that? Well, this way, the redirection applies to all output to stderr made inside the whole loop.

Let's see what the result of our loop was:

    $ cat errors
    rm: cannot remove `/home/axxo/secret': No such file or directory
    rm: cannot remove `/home/lhunath/secret': No such file or directory

Two error messages in our error log file. Two people that didn't have a secret file in their home directory.

If you're writing a script, and you expect that running a certain command may fail on occasion, but don't want the script's user to be bothered by the possible error messages that command may produce, you can silence an FD. Silencing it is as easy as normal File Redirection. We're just going to send all output to that FD into the system's black hole:

    $ for homedir in /home/*
    > do rm "$homedir/secret"
    > done 2> /dev/null

The file /dev/null is always empty, no matter what you write or read from it. As such, when we write our error messages to it, they just disappear. The /dev/null file remains as empty as ever before. That's because it's not a normal file; it's a virtual device. Some people call /dev/null the bit bucket.

There is one last thing you should learn about File Redirection. It's interesting that you can make error log files like this to keep your error messages; but as I mentioned before, BASH makes sure that the file exists before trying to redirect to it. BASH also makes sure the file is empty before redirecting to it. As a result, each time we run our loop to delete secret files, our log file will be truncated empty before we fill it up again with new error messages. What if we'd like to keep a record of any error messages generated by our loop? What if we don't want that file to be truncated each time we start our loop? The solution is achieved by doubling the redirection operator. > becomes >>. >> will not empty a file, it will just append new data to the end of it!

    $ for homedir in /home/*
    > do rm "$homedir/secret"
    > done 2>> errors

Hooray!

• Good Practice:
  It's a good idea to use redirection whenever an application needs file data and is built to read data from stdin. A lot of bad examples on the Internet tell you to pipe (see later) the output of cat into processes; but this is nothing more than a very bad idea.
  When designing an application that could be fed data from a variety of different sources, it is often best simply to have your application read from stdin; that way, the user can use redirection to feed it whatever data she wishes. An application that reads standard input in a generalized way is called a filter.

• In The Manual: Redirecting Input, Redirecting Output, Appending Redirected Output, Redirecting Standard Output and Standard Error

2.2. File Descriptor Manipulation

Now that you know how to manipulate process input and output by sending it to and reading it from files, let's make it a little more interesting still.

It's possible to change the source and desination of FDs to point to or from files, as you know. It's also possible to copy one FD to another. Let's prepare a simple testbed:

    $ echo "I am a proud sentence." > file

We've made a file called file, and written a proud sentence into it. It's time I introduce a new application to you. Its name is grep, and it's incredibly powerful. grep is that one thing that you need more than anything else in your household. It basically takes a search string as its first argument and one or more files as extra arguments. Just like cat, grep also uses stdin if you don't specify any files as extra arguments. grep reads the files (or stdin if none were provided) and searches for the search string you gave it. Most versions of grep even support a -r switch, which makes it take directories as well as files as extra arguments, and then searches all the files and directories in those directories that you gave it. Here's an example of how grep can work:

    $ ls house/
    drawer  closet  dustbin  sofa
    $ grep -r socks house/
    house/sofa:socks

In this silly example we have a directory called house with several pieces of furniture in it as files. If we're looking for our socks in each of those files, we send grep to search the directory house/. grep will search everything in there, open each file and look through its contents. In our example, grep finds socks in the file house/sofa; presumably tucked away under a pillow. You want a more realistic example? Sure:

    $ grep "$HOSTNAME" /etc/*
    /etc/hosts:127.0.0.1       localhost Lyndir

Here we instruct grep to search for whatever $HOSTNAME expands to in whatever files /etc/* expands to. It finds my hostname, which is Lyndir, in the file /etc/hosts, and shows me the line in that file that contains the search string.

OK, now that you understand grep, let's continue with our File Descriptor Manipulation. Remember that we created a file called file, and wrote a proud sentence to it? Let's use grep to find where that proud sentence is now:

    $ grep proud *
    file:I am a proud sentence.

Good! grep found our sentence in file. It writes the result of its operation to stdout which is shown on our terminal. Now let's see if we can make grep send an error message, too:

    $ grep proud file 'not a file'
    file:I am a proud sentence.
    grep: not a file: No such file or directory

This time, we instruct grep to search for the string proud in the files 'file' and 'not a file'. file exists, and the sentence is in there, so grep happily writes the result to stdout. It moves on to the next file to scan, which is 'not a file'. grep can't open this file to read its content, because it doesn't exist. As a result, grep emits an error message on stderr which is still connected to our terminal.

Now, how would you go about silencing this grep statement completely? We'd like to send all the output that appears on the terminal to a file instead; let's call it proud.log:

    $ grep proud file 'not a file' > proud.log 2> proud.log

Does that look about right? We first use > to send stdout to proud.log, and then use 2> to send stderr to proud.log as well. Almost, but not quite. If you run this command (at least on some computers), and then look in proud.log, you'll see there's only an error message, not the output from stdout. We've created a very bad condition here. We've created two FDs that both point to the same file, independently of each other. The results of this are not well-defined. Depending on how the operating system handles FDs, some information written via one FD may clobber information written through the other FD.

    $ echo "I am a very proud sentence with a lot of words in it, all for you." > file2
    $ grep proud file2 'not a file' > proud.log 2> proud.log
    $ cat proud.log
    grep: not a file: No such file or directory
    of words in it, all for you.

What happened here? grep opened file2 first, found what we told it to look for, and then wrote our very proud sentence to stdout (FD 1). FD 1 pointed to proud.log, so the information was written to that file. However, we also had another FD (FD 2) pointed to this same file, and specifically, pointed to the beginning of this file. When grep tried to open 'not a file' to read it, it couldn't. Then, it wrote an error message to stderr (FD 2), which was pointing to the beginning of proud.log. As a result, the second write operation overwrote information from the first one!

We need to prevent having two FDs working on the same destination or source. We can do this by duplicating FDs:

    $ grep proud file 'not a file' > proud.log 2>&1

In order to understand these, you need to remember: always read file redirections from left to right. This is the order in which BASH processes them. First, stdout is changed so that it points to our proud.log. Then, we use the >& syntax to duplicate FD 1 and put this duplicate in FD 2.

A duplicate FD works differently from having two independent FDs pointing to the same place. Write operations that go through either one of them are exactly the same. There won't be a mix-up with one FD pointing to the start of the file while the other has already moved on.

Be careful not to confuse the order:

    $ grep proud file 'not a file' 2>&1 > proud.log

This will duplicate stderr to where stdout points (which is the terminal), and then stdout will be redirected to proud.log. As a result, stdout's messages will be logged, but the error messages will still go to the terminal. Oops.

Note:
For convenience, BASH also makes yet another form of redirection available to you. The &> redirection operator is actually just a shorter version of what we did here; redirecting both stdout and stderr to a file :

    $ grep proud file 'not a file' &> proud.log

This is the same as > proud.log 2>&1, but not portable to BourneShell. It is not recommended practice, but you should recognize it if you see it used in someone else's scripts.

TODO: Moving FDs and Opening FDs RW.

• In The Manual: Duplicating File Descriptors, Moving File Descriptors, Opening File Descriptors for Reading and Writing

• In the FAQ:
  How can I redirect the output of multiple commands at once?

• How can I redirect the output of 'time' to a variable or file?

• How do I use dialog to get input from the user?

• How can I redirect stderr to a pipe?

• Tell me all about 2>&1 -- what's the difference between 2>&1 >foo and >foo 2>&1, and when do I use which?

2.3. Heredocs And Herestrings

Files aren't all that. They're boring, really. Strings are so much more interesting. They're not permanent like files on a hard disk, but they're easy to work with, easy to make and easy to manipulate.

Heredocs and Herestrings allow you to perform Redirection as you would with files, just by using strings instead. Let's try it out!

    $ grep proud <<END
    > I am a proud sentence.
    > END
    I am a proud sentence.

This is a Heredoc (or Here Document). Heredocs aren't really useful unless you're trying to embed long strings of several lines inside your scripts, which is bad practice. You should keep your logic (your code) and your input (your data) separated, preferably in different files.

The way Heredocs work, is by adding the <<STRING redirection to a command. That'll instruct that command's stdin that it has to start reading from the script (or the command line, if you're not in a script). The input of the Heredoc stops as soon as you repeat whatever string you added to the end of the <<. In the example above, I used the string END, but it can really be any single word.

Beware that all text following the Heredoc operator is sent to the command's stdin almost literally ($ and backticks have to be escaped with \ or will be expanded). That also means any spaces you use for indenting your script. The terminator string (in our case END) must be in the beginning of the line.

    echo "Let's test abc:"
    if [[ abc = a* ]]; then
        cat <<END
            abc seems to start with an a!
    END
    fi

Will result in:

    Let's test abc:
            abc seems to start with an a!

You can avoid this by temporarily removing the indentation for the lines of your Heredocs. However, that distorts your pretty and consistent indentation. There is an alternative. If you use <<-END instead of <<END as your Heredoc operator, BASH removes any tab characters in the beginning of each line of your Heredoc content before sending it to the command. That way you can still use tabs to indent your Heredoc content with the rest of your code. Those tabs will not be sent to the command that receives your Heredoc. This also means you can use tabs to indent your terminator string.

Finally, note that if you quote the word that you're using to delimit your Heredoc, BASH won't perform any substitutions on the contents. Try this example with and without the quote characters, to see the difference:

    $ cat <<'XYZ'
    > My home directory is $HOME
    > XYZ
    My home directory is $HOME

Now let's check out the very similar but more compact Herestrings:

    $ grep proud <<<"I am a proud sentence"
    I am a proud sentence.

This time, stdin reads its information straight from the string you put after the <<< operator. This is very convenient to send data that's in variables into processes:

    $ grep proud <<<"$USER sits proudly on his throne in $HOSTNAME."
    lhunath sits proudly on his throne in Lyndir.

Herestrings are shorter, less intrusive and overall more convenient than their bulky Heredoc counterpart. However, they are not portable to the Bourne shell.

Later on, you will learn about pipes and how they can be used to send the output of a command into another command's stdin. Many people use pipes to send the output of a variable as stdin into a command. However, for this purpose, Herestrings should be preferred. They do not create a subshell and are lighter both to the shell and to the style of your shell script:

    $ echo 'Wrap this silly sentence.' | fmt -t -w 20
    Wrap this silly
       sentence.
    $ fmt -t -w 20 <<< 'Wrap this silly sentence.'
    Wrap this silly
       sentence.

• Good Practice:
  Long heredocs are usually a bad idea because scripts should contain logic, not data. If you have a large document that your script needs, you should ship it in a separate file along with your script. Herestrings, however, come in handy quite often, especially for sending variable content (rather than files) to filters like grep or sed.

• In The Manual: Here Documents, Here Strings

3. Pipes

Now that you can effortlessly manipulate File Descriptors to direct certain types of output to certain files, it's time you learn some more ingenious tricks available through I/O redirection.

You can use File Redirection to write output to files or read input from files. But what if you want to connect the output of one application directly to the input of another? That way, you could build a sort of chain to process output. If you already know about FIFOs, you could use something like this to that end:

    $ ls
    $ mkfifo myfifo; ls
    myfifo
    $ grep bea myfifo &
    [1] 32635
    $ echo "rat
    > cow
    > deer
    > bear
    > snake" > myfifo
    bear

We use the mkfifo command to create a new file in the current directory named 'myfifo'. This is no ordinary file, however, but a FIFO (which stands for First In, First Out). FIFOs are special files that serve data on a First In, First Out-basis. When you read from a FIFO, you will only receive data as soon as another process writes to it. As such, a FIFO never really contains any data. So long as no process writes to it, any read operation on the FIFO will block as it waits for data to become available. The same works for writes to the FIFO -- they will block until another process reads from the FIFO.

In our example, the FIFO called myfifo is read from by grep. grep waits for data to become available on the FIFO. That's why we append the grep command with the & operator, which puts it in the background. That way, we can continue typing and executing commands while grep runs and waits for data. Our echo statement feeds data to the FIFO. As soon as this data becomes available, the running grep command reads it in and processes it. The result is displayed. We have successfully sent data from the echo command to the grep command.

But these temporary files are a real annoyance. You may not have write permissions. You need to remember to clean up any temporary files you create. You need to make sure that data is going in and out, or the FIFO might just end up blocking for no reason.

For these reasons, another feature is made available: Pipes. A pipe basically just connects the stdout of one process to the stdin of another, effectively piping the data from one process into another. The entire set of commands that are piped together is called a pipeline. Let's try our above example again, but using pipes:

    $ echo "rat
    > cow
    > deer
    > bear
    > snake" | grep bea
    bear

The pipe is created using the | operator between two commands that are connected with the pipe. The former command's stdout is connected to the latter command's stdin. As a result, grep can read echo's output and display the result of its operation, which is bear.

Pipes are widely used as a means of post-processing application output. FIFOs are, in fact, also referred to as named pipes. They accomplish the same results as the pipe operator, but through a filename.

Note:
The pipe operator creates a subshell environment to run each process in. This is important to know because any variables that you modify or initialize inside the second command will appear unmodified outside of it. Let's illustrate:

    $ message=Test
    $ echo "Salut, le monde!" | { read message; echo "The message is: $message"; }
    The message is: Salut, le monde!
    $ echo "The message is: $message"
    The message is: Test

Once the pipeline ends, so do the subshells that were created for it. Along with those subshells, any modifications made in them are lost. So be careful!

• Good Practice:
  Pipes are a very attractive means of post-processing application output. You should, however, be careful not to over-use pipes. If you end up making a pipeline that consists of three or more applications, it is time to ask yourself whether you're doing things a smart way. You might be able to use more application features of one of the post-processing applications you've used earlier in the pipe. Each new command in a pipeline causes a new subshell and a new application to be loaded. It also makes it very hard to follow the logic in your script!

• In The Manual: Pipelines

• In the FAQ:
  I set variables in a loop. Why do they suddenly disappear after the loop terminates? Or, why can't I pipe data to read?

• How can two processes communicate using named pipes (fifos)?

• How can I redirect stderr to a pipe?

• How can I read a file line-by-line?

• Tell me all about 2>&1 -- what's the difference between 2>&1 >foo and >foo 2>&1, and when do I use which?

4. Miscellaneous Operators

Aside from the standard I/O operators, bash also provides a few more advanced operators that make life on the shell that much nicer.

4.1. Process Substitution

A cousin of the pipe is the process substitution operator (<(), >()). It's a convenient way to use named pipes without having to create temporary files. Whenever you think you need a temporary file to do something, process substitution might be a better way to handle things.

What it does, is basically run the command inside the brackets. With the <() operator, the command's output is put in a sort of temporary file that's created by bash. The operator itself in your command is replaced by the filename of that file. After your whole command finishes, the file is cleaned up.

Here's how we can put that into action: Imagine a situation where you want to see the difference between the output of two commands. Ordinarily, you'd have to put the two outputs in two files and diff those:

    $ head -n 1 .dictionary > file1
    $ tail -n 1 .dictionary > file2
    $ diff -y file1 file2
    Aachen                                                        | zymurgy
    $ rm file1 file2

Using the Process Substitution operator, we can do all that with a one-liner and no need for manual cleanup:

    $ diff -y <(head -n 1 .dictionary) <(tail -n 1 .dictionary)
    Aachen                                                        | zymurgy

The <(..) part is replaced by the temporary file created by bash, so diff actually sees something like this:

    $ diff -y /dev/fd/63 /dev/fd/62

Here we see how bash runs diff when we use process substitution. It runs our head and tail commands, redirects their respective outputs to the files /dev/fd/63 and /dev/fd/62. Then it runs the diff command, passing those filenames where originally we had put the respective filename's process substitution operator.

The actual implementation of the temporary files depends from system to system. In fact, you can see what the above would actually look like to diff on your box by putting an echo in front of our command:

    $ echo diff -y <(head -n 1 .dictionary) <(tail -n 1 .dictionary)
    diff -y /dev/fd/63 /dev/fd/62

The >(..) operator is much like the <(..) operator, but instead of redirecting the command's output to a file, we redirect the file to the command's input. It's used for cases where you're running a command that writes to a file, but you want it to write to another command instead:

    $ tar -cf >(ssh host tar x) .

4.2. Here Strings

Much like the earlier mentioned Here Documents are the bash Here Strings. A Here String is created by using the operator <<<. It is almost the same as a Here Document, but instead of redirecting multiple lines of text to a command's stdin, we're redirecting one argument to it. Note that the argument may still consist of anything a string can hold; which means it can still be multiple lines long.

It's a great replacement for echo "something" | somecommand, because it doesn't incur the overhead of the subshell, and it's slightly neater in code. Moreover, very often, the echo solution is not an option, because it runs somecommand in a subshell. That means you can't set any bash variables in there!

Let's illustrate:

    $ echo "well, well, what have we here?" | tr "wh" "v'"
    vell, vell, v'at 'ave ve 'ere?
    $ tr "wh" "v'" <<< "well, well, what have we here?"
    vell, vell, v'at 'ave ve 'ere?
    $ echo "$USER" | read name # 'name' is set in a subshell!  Can't access it in the rest of the script.
    $ read name <<< "$USER"    # 'name' is set in the main shell, now it's accessible to the rest of the script.

Combining the <<< operator with the $'' operator is often very handy. The latter is like echo -e, but more reliable. It expands escape codes, such as \n, \a, etc. We can use it for a shorter replacement of heredocs:

    $ grep '[A-C]' <<< $'Olivia\nPeter\nAlfred\nCornelia'
    Alfred
    Cornelia

<- Tests and Conditionals | Compound Commands ->