Bourne shell idioms

Here are some portable Bourne shell idioms that I find useful to remember for scripting. The Bourne shell does much more than most users realize, and the ksh and bash extensions are rarely essential. (From the command-line, bash and ksh are vastly more useful.)

My favorite reference book is "Portable Shell Programming --- An Extensive Collection of Bourne Shell Examples" by Bruce Blinn from Prentice Hall.

Get information on a built-in bash command with help. It's much easier than reading the full bash man page at http://www.gnu.org/software/bash/manual/bash.html

For Bash suggestions, I recommend this bash FAQ: http://mywiki.wooledge.org/BashFAQ/

See more sections on Gnu/Linux .

§    Text filtering commands

Administrating from scripts and the command-line often benefit from pipes of text filtering commands. Here are some that are easy to overlook or forget.

• mmencode converts to and from base64 and "quoted-printable" formats for email. Search for the metamail package. Unfortunately, this has become hard to find.

  Alternatively uuencode -m converts to base64, and uudecode -m converts from base64.

  Or decode and encode quoted-printable and base64 with

  perl -pe 'use MIME::QuotedPrint; $_=MIME::QuotedPrint::decode($_);' perl -pe 'use MIME::QuotedPrint; $_=MIME::QuotedPrint::encode($_);' perl -pe 'use MIME::Base64; $_=MIME::Base64::encode($_);' perl -pe 'use MIME::Base64; $_=MIME::Base64::decode($_);'

  URL-encode a string with

  perl -ne 'chomp; s/([^-_.~A-Za-z0-9])/sprintf("%%%02X", ord($1))/seg; print "$_\n"'

• Convert utf-8 characters to escaped hexadecimal for html, and back:

  perl -C -pe 's/([^\x00-\x7f])/sprintf("&#%d;", ord($1))/ge;' perl -C -pe 's/&\#(\d+);/chr($1)/ge;s/&\#x([a-fA-F\d]+);/chr(hex($1))/ge;'

• uniq lets you remove duplicated lines from a sorted file.
• Count the number of times a given line occurs with

  sort | uniq -c | sort -n sort | uniq -c | sort -k1,1nr -k2

• Break one word per line with

  perl -pe 's/\s+/\n/g'

• Combine separate lines into a single line of words with

  paste -s -d" "

• comm lets you suppress lines unique to one or both of two files.
• cat -s never prints more than one blank line in a row.
• Remove all blank lines with

  perl -ne 'print if /\S/'

• Print lines starting with one containing FOO and ending with one containing BAR.

  sed -n '/FOO/,/BAR/p'

• merge and diff3 for merging changes in files edited from a common ancestor.
• fold breaks lines to proper width, and fmt will reformat lines into paragraphs.
• dirname and basename let you extract the directory and filenames from a full path to a file.
• namei breaks a pathname into pieces and follows symbolic links.
• expand and col -x replace tabs by spaces.
• col -b removes backspaces from a file.
• cat -v shows non-printing characters as ascii escapes.
• sed '1,10d' deletes the first 10 lines.
• sed -n '3p' and sed -n '3{p;q}' both print the third line, but the latter is more efficient.
• sed '/foo/q' truncates a file after the line containing foo.
• sed -ne '/foo/,/bar/p' prints everything from the line containing foo to the line containing bar.
• Align space-delimited fields into orderly columns with column -t.
• Right justify queries with

  printf "%40s" "Do you want to delete? [y/N] "

• Convert dos text files to unix, and vice versa:

  dos2unix file.txt unix2dos file.txt tr -d \\r < win.txt > unix.txt # if you can't find dos2unix sed -e 's/$/\r/' < unix.txt > win.txt # if you can't find unix2dos

• cat -n and nl numbers lines.
• Both of these perform string substitution, but the latter allows more general regular expressions:

  sed -e 's/oldtext/newtext/g' perl -pe 's/oldtext/newtext/g'

  Here's how to replace double quotes by single quotes for TeX:

  < in.tex perl -pne 's%\B"\b%``%g' | perl -pne "s%\b\"\B%''%g" > out.tex

• Use iconv to convert between character encodings.
• Here are two ways to find string patterns in a file:

  grep 'pattern' filename [file] [< file] perl -ne 'print if /pattern/' [file] [< file]

• Print the first and third columns of each line:

  awk '{print $1,$3}' perl -lane 'print "$F[0] $F[2]"'

• Convert to lower-case:

  tr '[A-Z]' '[a-z]' perl -pe 'tr/[A-Z]/[a-z]/' perl -pe '$_ =lc'

• Simple character substitutions and deletions may be simplest with tr.

  tr -d '\n' # delete newlines tr '\n' '\0' # replace newlines by null characters.

  $ echo 1-2a-3b | tr "[1-9]" "[2-9]" | tr '-' '_' | tr -d 'a' 2_3_4b

• You can pipe into a loop with read -r. Here is a complicated way to cat a text file, piping in and out of a loop.

  cat file | while read -r a; do echo "$a" ; done | cat

• To read lines in pairs from two files try

  paste file1 file2 | while read -r a b ; do echo "$a $b" ; done

• Divide words one per line, then sum them as numbers:

  $ echo 1 2 3.1 | perl -pe 's/\s+/\n/g' | perl -e '$s=0; while (<>) {$s += $_;} ; print "$s\n";' 6.1

• Reverse lines with tac and words with rev.
• Sort a list of dependencies with tsort.

§    Files and directories

• Select text (non-binary) files with one of these

  \ls | perl -lne 'print if -T' perl -le 'for (glob "*") {print if -T }' perl -le 'print for grep -T, <*>'

  The perl algorithm for detecting text files is very good.

• To do something to files with goofy names, including spaces and dashes, delimit the files with null characters instead of whitespace or newlines.

  find . -type f -print0 | xargs -r0 ls

  Or read from one line at a time:

  cd "$dir1" && find . -type f | while read -r f ; do if [ ! -f "$dir2/$f" ] ; then echo "$f is in $dir2 but not in $dir2" fi done

• See if a directory contains any files, including broken links.

  has_files() { set -- "$1"/.[!.]* "$1"/*; test -e "$1" || test -e "$2" || test -L "$1" || test -L "$2"; } if has_files ${dir} ; then echo "${dir} has files" ; else echo "${dir} is empty" ; fi

  See if files of a certain type exist:

  if [ "`printf '%s' *.par`" != '*.par' ] ; then echo "has par files" ; fi [or] test "`printf '%s' *.par`" != '*.par' && echo "has pars" || echo "no pars"

• readlink -f will fully resolve what a symbolic link points to.

  Find all bad symbolic links with

  find . -type l | while read -r f ; do if ! readlink -f "$f" >&/dev/null then echo "$f" ; fi ; done

• To see the canonical path for the current directory, you can use either of these:

  readlink -f . pwd -P

§    Variables

• To see if a variable contains a regular expression, combine if and grep. For example to see if the name of a file begins with a dot, try

  if echo "$filename" | grep '^[.]' >/dev/null then echo yes ; else echo no ; fi

  expr also has a support for limited regular expressions.

  if [ `expr "$filename" : '[.].*'` -ne 0 ] then echo yes ; else echo no ; fi

• Use read -r to avoid tokenizing filenames with spaces. Here's how to find all files containing a space, and replace them by underscores.

  find . -iname '* *' | while read -r f ; do echo mv "$f" "`echo "$f" | sed 's/ */_/g'`" done

• For simple integer arithmetic use expr:

  N=`expr "$N" + 3`

• For arbitrary-precision floating-point math, use bc -l

  # Get pi to 10 places with arctangent (bc man page) PI=`echo "scale=10; 4*a(1)" | bc -l` # Expensive calculation of zero (Craig Artley): ZERO=`echo "c($PI/4)-sqrt(2)/2" | bc -l`

• seq 1 100 generates all integers between 1 and 100. To iterate a loop 100 times, try

  for i in `seq 1 100` ; do ... ; done

• You can set the environment of a subprocess by defining a variable on the same line. The current shell is not affected.

  $ x=doggie sh -c 'echo x=$x' x=doggie $ x=pig ; x=doggie echo x=$x x=pig

• Test that a string has non-zero length with

  if [ -n "$string" ] ; then echo "not empty" ; fi

  The -n is actually the default for a string expression, so you can omit it:

  if [ "$string" ] ; then echo "not empty" ; fi

• There are several good ways to set default values for environmental variables. Many do this

  if [ ! "$VARIABLE" ] ; then VARIABLE="default value" ; fi export VARIABLE

  A simple alternative is

  : ${VARIABLE:="default value"} export VARIABLE

  The colon at the beginning of the line is necessary as a no-op that allows its arguments to be evaluated.

• Rarely you may want to accept a variable defined as an empty string. If so, then omit the colon before the equals when setting the default.

  : ${VARIABLE="default value"} export VARIABLE

  To test whether a string is defined, even if empty test

  if [ "${VARIABLE+x}" ] ; then echo DEFINED ; fi

• To echo all variables starting with X: echo ${!X*}
• To check whether a series of variables are defined, try

  for V in JAVA_HOME SSH_AGENT_PID TEXMFDIR NETHACKOPTIONS ; do eval v="\$$V" if [ ! "$v" ] ; then echo "You must define $V" ; fi done

§    Running commands

• Use "$@" when passing command-line arguments unaltered to subprocesses. This is equivalent to passing "$1" "$2" ..., but the first version works properly for no arguments.
• Test the processing of arguments, like this

  $ set a 'b c' d $ for i in "$@" ; do echo "|$i|" ; done |a| |b c| |d| $ for i in "$*" ; do echo "|$i|" ; done |a b c d| $ for i in $* ; do echo "|$i|" ; done |a| |b| |c| |d|

• See what runtime options you may have set with these

  set -o; bind -p; shopt -p; stty -a

  For example, you can edit a bash command by default in emacs mode. Change to vi with

  set -o vi

  In emacs mode, you can edit your command in your environmental $EDITOR with cntl-x cntl-e

  In vi-mode, use esc-v. See help fc for more.

• Repeat the last argument of the previous command with !$. Repeat all arguments without the command with !*.
• To guarantee that a background process outlives the current shell, add extra parentheses like this:

  ( command & )

  Otherwise, your current shell, by exiting X or ssh, may terminate all processes that have your shell as the parent process. The extra parentheses starts a subshell that exits as soon as the command is spawned in the background. The background process changes its parent process ID to 1. This is a command-line version of the "double fork."

• Repeat until a command succeeds:

  while ! cvs -z 3 -q update -dPA ; do echo -n . ; sleep 60 ; done

• Make a progress bar (loop while waiting on a process)

  sleep 10 & while ps -p $! >/dev/null; do echo -n . ; sleep 1 ; done ; echo or while pidof mozilla-bin > /dev/null ; do echo -n . ; sleep 1 ; done ; echo

  pgrep -f or killall -0 are alternatives to pidof for this purpose.

§    Manipulating paths

• Loop over the elements of a PATH by tokenizing with the character ':'.

  IFS=':' ; for dir in $PATH ; do echo $dir ; done

• Check for the existence of an executable version of a command in your PATH:

  function checkPath() {IFS=':' ; for dir in $PATH ; do if [ -x "$dir/$1" ] ; then return 0; fi ; done; return 1;} if checkPath commandName ; then ... ; fi

• Here is a lovely function to modify a PATH that I found in /etc/profile:

  pathmunge () { if ! echo $PATH | egrep "(^|:)$1($|:)" >/dev/null ; then if [ "$2" = "after" ] ; then PATH=$PATH:$1 else PATH=$1:$PATH fi fi }

§    Common script chores

• Debug the script with set -x.
• Make a script exit immediately after any failed command with set -e.
• Process flags in a script:

  for i in "$@" ; do case $i in -a) FLAG_A=1 shift ;; -b) FLAG_B="$2" shift ; shift ;; --) shift ; break ;; esac done

• Print help from a script:

  if [ $# -lt 1 -o "$1" = "-h" -o "$1" = "-help" -o "$1" = "--help" ] ; then cat <<-END Usage: `basename $0` [-flag] arg1 [arg2] More information. END exit fi

• Handle errors with functions:

  Often an error exit is handled most cleanly with a function.

  print_usage_and_exit() { cat <<-END Usage: `basename $0` arg1 arg2 [arg3] The first two arguments are required. END exit } if [ $# -lt 2 ] ; then print_usage_and_exit fi

• Here's a robust way to locate the directory containing a script, following symbolic links. (Taken from the launch script of FindBugs.)

  program="$0" while [ -h "$program" ]; do link=`ls -ld "$program"` link=`expr "$link" : '.*-> \(.*\)'` if [ "`expr "$link" : '/.*'`" = 0 ]; then dir=`dirname "$program"` program="$dir/$link" else program="$link" fi done script_directory=`dirname $program` script_directory=`cd $script_directory && /bin/pwd`

• Trapping signals to stop scripts:

  Ever try to interrupt your script, then discover that it killed only one command and continued to the next? Force a complete exit by adding the following line early in your script.

  trap "exit 1" 1 2 3 15

  You can also trap normal and error exits:

  # force script to exit when any command fails set -e # Trap on any exit trap "echo Always called before exit" 0 # Trap on error exit only trap "echo Error exit was called " ERR echo "Next command will fail" # Returns error code of 1 false echo "Will not see this comment"

• Process ID's

  Get the process ID of the current shell as $$, of the parent shell with $PPID and $! for the most recently backgrounded child process.

  Interactively, you get see child PID's with jobs -p.

• Here's how to ask a yes or no question, with a default of no. It checks whether the first letter is a y or Y and ignores leading spaces.

  echo -n "Do you want to continue? [y/N]: " read answer if expr "$answer" : ' *[yY].*' > /dev/null; then echo Continuing else echo Quitting exit fi

• Here's how to ask for a password without echoing the characters. The trapping ensures that an interrupt does not leave the echoing off.

  stty -echo trap "stty echo ; echo 'Interrupted' ; exit 1" 1 2 3 15 echo -n "Enter password: " read password echo "Your password is \"$password\"" stty echo

  Gnome and other frameworks often allow simple scripting of GUIs:

  password=`zenity --entry --text "Enter password:"`

§    File descriptors

• Redirecting output file descriptors

  Here are common ways to capture the standard output and standard error of a single command in a log file:

  command >file.log 2>&1 command 2>&1 | tee file.log

• If you have a script with many commands, you can have them all write to the same log file by default:

  # save default standard output in file descriptor 10 exec 10>&1 # redirect standard output to a log file. exec >file.log # redirect standard error to same log file exec 2>&1 # close stdin exec 0<&- # This command will write to log file command # echo to default standard output instead of log file echo "Visible message" 1>&10

  Avoid file descriptor 5, which bash already uses. (ulimit -n should show many available file descriptors.)

• Avoid writing to stdout if it is not connected to a terminal:

  test -t 1 && echo "Connected to a terminal"

• Open a socket

  Associate a file descriptor, say 4, with a socket, and close with

  4< /dev/tcp/$hostname/$port 4<&-

  A more portable solution is to use nc. Listen on a port with

  nc -l -p 3535

  Connect to a remote host port like

  echo 'GET /' | nc hostname 80

  An even more general utility is socat, which also handles Unix sockets.

Return to parent directory.