sed

Stream editor: applies edit commands to each line. GNU sed on Ubuntu.

The one command you'll use 95% of the time

sed 's/old/new/'          # substitute, first match per line
sed 's/old/new/g'         # all matches per line
sed 's/old/new/2'         # only the 2nd match per line
sed -i 's/old/new/g' f    # edit file in place
sed -i.bak 's/x/y/' f     # in place, keeps f.bak
sed -E 's/(a)(b)/\2\1/'   # extended regex + backrefs

Any delimiter works: useful for paths/URLs: sed 's|/a/b|/x/y|g'. & in replacement = the whole match: sed 's/foo/[&]/g' → [foo].

Addressing: //which// lines get the command

sed '5d' f                    # delete line 5
sed '5,10d' f                 # delete 5-10
sed '$d' f                    # last line
sed '2~3d' f                  # every 3rd line starting at 2
sed '/error/d' f              # lines matching /error/
sed '/START/,/END/d' f        # inclusive pattern range
sed '/START/,/END/!d' f       # keep ONLY the range (! inverts)
sed -n '5,10p' f              # print just lines 5-10 (-n + p)
sed -n '/error/p' f           # print only matches (grep-like)

The commands beyond `s`

d   delete line          p   print line (with -n)
a\  append after line    i\  insert before line
c\  change whole line    y/abc/ABC/  transliterate chars

sed '3i\inserted text' f        # insert before line 3
sed '/pattern/a\appended' f     # append after matches
sed 'y/abc/ABC/' f              # like tr, per char
sed -n 's/x/y/p' f              # print only changed lines

Multiple commands

sed -e 's/a/b/' -e '5d' f
sed 's/a/b/; s/c/d/' f          # ; separates commands
sed '/skip/!s/a/b/' f           # s on lines NOT matching

Useful regex bits (GNU sed BRE/ERE)

Default is BRE: +, ?, (, ), | need backslashes.

sed 's/\(ab\)\+/X/: or just always use sed -E (ERE) and forget it.

\b word boundary, \w/\W, \s/\S work in GNU sed.
\( \) groups become \1-\9 backrefs: sed -E 's/([0-9]+)x([0-9]+)/\2 by \1/'.
.* is greedy: sed -E 's/“(.*)”/\1/' grabs the LAST quote pair.

Non-greedy needs [^“]*: sed -E 's/”([^“]*)”/\1/'.

Classic one-liners

sed 's/^[[:space:]]*//;s/[[:space:]]*$//' f   # trim whitespace
sed 's/^/# /' f                              # comment out every line
sed -n 'x;$p' f || tail -1 f                 # last line (use tail)
sed -e :a -e 'N;$!ba' -e 's/\n/ /g' f        # join all lines
sed 's/./&\n/g' f                            # one char per line
seq 5 | sed 's/^/item/'                      # generate

Gotchas

-i without backup is irreversible: test without -i first.
sed -i '' is macOS syntax; GNU is sed -i (no arg). These pages assume

GNU sed on Linux.

sed -z treats input as NUL-separated: for filenames from find -print0.
It processes line-by-line; multi-line matching needs N, P, D: at that

point reach for perl -0777 -pe or awk.

Table of Contents

sed