The Bash shell

The ubiquitous shell on Linux systems, warts and all. Not to be confused with /bin/sh, which is often a less-fully-featured shell.

Cheatsheet

Strings

${s%suffix} # remove suffix
${s#prefix} # remove prefix
${s/from/to} # replace first occurrence
${s//from/to} # replace all occurrences
${s:-whatever} # take value of $s or 'whatever' if unset

${x:-default} # variable or default value

Redirection

• https://www.gnu.org/software/bash/manual/html_node/Redirections.html

# redirect stdout to stderr
echo hello >&2

# redirect both stdout and stderr
echo hello &>log.txt

Order matters! echo hello >log.txt 2>&1 redirects both standard output and standard error to the file log.txt. Think of the redirection operator as an assignment.

Heredocs

• https://linuxize.com/post/bash-heredoc/

cat << EOF
Line 1
Line 2
EOF

cat << "EOF"
No parameter expansion is done.
EOF

    cat <<- EOF
    The <<- operator lets you indent your code.
    EOF

s=$(cat << EOF
Line 1
Line 2
EOF
)

Control flow

for p in *.csv; do
  # ...
done

if [[ ... ]]; then
fi

case "$x" in
  pattern)
    ;;
  *)
    ;;
esac

Conditions

[[ "$s" == "prefix"* ]] # string has prefix
[[ -z "$s" ]] # string is empty
[[ -n "$s" ]] # string is not empty

[[ -e "$f" ]] # file exists
[[ -L "$f" ]] # symlink exists (regardless of whether target exists)
[[ -d "$f" ]] # file is directory
[[ -f "$f" ]] # file is regular file
[[ -h "$f" ]] # file is a symlink
[[ -x "$f" ]] # file is executable

(( x == 100 )) # no needs for '$'

Arrays

arr=()
arr+=("$x")
${#arr[@]} # array length
${arr[@]:1} # slice starting from first element
for x in "${arr[@]}"; do .. done

Associative arrays

• https://xeiaso.net/notes/2024/bash-hashmap/

declare -A mymap
mymap["key"]="value"
echo "${mymap[key]}"
unset mymap["key"]

# check if a key is set (even if empty)
# NOTE that it's `mymap["key"]`, not `${mymap["key"]}`
[[ -v mymap["key"] ]] && echo exists

# iterate over keys
for key in "${!mymap[@]}"; do
  # ...
done

# iterate over values
for key in "${mymap[@]}"; do
  # ...
done

# pre-declare and make read-only (-r)
declare -A -r mymap=(
  ["key1"]="value1"
  ["key2"]="value2"
)

Temporary files and directories

tmp="$(mktemp)"
tmpdir="$(mktemp -d)"

# with specific extension
# https://stackoverflow.com/a/59638023/3934904
tmpdir="$(mktemp -d)"
tmp="$tmpdir/test.md"

Clean-up functions

cleanup() { ... }
trap cleanup EXIT

Get script path

SCRIPT_PATH="$(realpath "$0")"

Temporarily allow unset variables

if [[ $- = *u* ]]; then
  restore_unset="1"
else
  restore_unset="0"
fi
set +u

# ...

if (( restore_unset == 1 )); then
  set -u
fi

tar

# create a gzipped archive of a directory with relative paths
# NOTE: '.' at the end
tar -czf OUTFILE --directory=DIR .

# exclude a directory
# NOTE: exclude path is relative to DIR, not the current working directory
tar -czf OUTFILE --exclude=EXCLUDE --directory=DIR .

# extract tarball
tar -xzf TARBALL -C DIR

Tips

• set -euo pipefail
  • -e: exit when a command exits with non-zero status
  • -u: exit on undefined variable
  • -o pipefail: set return value of cmd1 | cmd2 | cmd3 to last command to exit with non-zero status (instead of exit status of cmd3 by default)
    • Though this is not always what you want; see "Be careful" below.
    • Use ${PIPESTATUS[@]} to retrieve the exit status of each command.
• set -x to print each command as it is run
• Wrap the script in main() { ... } and put main "$@" at the bottom.
  • Forces Bash to load and parse entire file before beginning execution.
• Use /usr/bin/env bash as the shebang for your scripts, not /bin/bash. On macOS, /bin/bash is an ancient version and some users have installed a newer version via Homebrew.

Be careful

• If you press Ctrl+Z in a compound command (cmd1 && cmd2, a for loop, etc.), then when you run fg only the command that was interrupted will be resumed, not the whole compound command. Try it yourself: sleep 5 && echo done
• ln -sf target name is supposed to replace the symlink at name with one pointing to target, but if name is a directory, it instead creates a symlink name/target that points at target.
• -o pipefail can cause flaky failures with commands that truncate the input (like head); in cmd | head, if head exits before cmd is finished writing, the kernel will deliver a SIGPIPE signal to cmd, killing it with a non-zero exit status.
• MYVAR=hello echo "$MYVAR" seems like it should work, but doesn't; MYVAR is defined when the program executes, but not when the command line is parsed. You can do MYVAR=hello; echo "$MYVAR" (note the semicolon), though this leaves MYVAR defined in your shell after echo exits.

Posts

• "Robustly parsing flags in Bash scripts" (August 2025)

Bibliography

Reference:

• ⭐️ Bash guide (Greg's Wiki)
  • Bash pitfalls
• Bash reference manual (GNU)
• bash(1) manpage
• Bash cheatsheet (devhints.io)

Writing robust scripts:

• "Writing Shell Scripts" (Signs of Triviality, 2016)
• "How to write idempotent Bash scripts" (Fatih Arslan, 2019)

Minutiae:

• "Seven Surprising Bash Variables" (zwischenzugs, 2019)