text

Regular expressions

POSIX extended regular expressions as used by grep -E, with notes on where Perl-style engines differ.

Updated 2026-10-14

This sheet describes POSIX extended regular expressions (ERE), the flavor selected by grep -E. Unless a card says otherwise, the syntax is portable ERE; Perl-style differences are collected in the last section.

Building blocks

Ordinary and special characters #

Pattern	Matches
`cat`	The literal text `cat`
`.`	Any single character
`\.`	A literal dot
`\\`	A literal backslash
`a\+b`	Literal `a+b`

Outside brackets these need a backslash to be literal: ^ . [ $ ( ) | * + ? { \.

Anchors #

Pattern	Matches
`^Error`	`Error` at the start of a line
`;$`	`;` at the end of a line
`^$`	An empty line
`^[0-9]+$`	A line made only of digits

Alternation and grouping #

red|green|blue          one of three words
gr(a|e)y                gray or grey
(ab)+                   ab, abab, ababab, ...
^(GET|POST) /api/       request lines for two methods

| has the lowest precedence, so ^a|b$ means ^a or b$; group to change that.

Repetition

Quantifiers #

Piece	Repeats the atom
`x*`	Zero or more times
`x+`	One or more times
`x?`	Zero or one time
`x{3}`	Exactly three times
`x{2,}`	Two or more times
`x{2,5}`	Two to five times

A quantifier applies to the single atom before it: a character, a bracket expression or a group.

Bounds in practice #

[0-9]{4}-[0-9]{2}-[0-9]{2}        a date such as 2026-10-14
[A-Fa-f0-9]{6}                    six hex digits
(na){2,}                          nana, nanana, ...

A { that is not followed by a digit is an ordinary character on macOS. Bounds may not exceed 255 there.

Leftmost, then longest #

Note

pattern:  cat|category
input:    category
POSIX:    category        (longest match at the earliest position)
Perl/JS:  cat             (first alternative that succeeds)

POSIX engines choose the earliest starting position, then the longest match from there. Backtracking engines stop at the first alternative that works.

Bracket expressions

Sets and ranges #

Pattern	Matches one character
`[aeiou]`	Any listed vowel
`[^aeiou]`	Anything except those
`[a-z]`	A range in the collating order
`[]x]`	`]` or `x`; put `]` first
`[a-]`	`a` or `-`; put `-` first or last
`[*.]`	A literal star or dot

Inside brackets, \ and most other special characters are ordinary.

Character classes #

Class	Members
`[[:digit:]]`	Decimal digits
`[[:alpha:]]`	Letters
`[[:alnum:]]`	Letters and digits
`[[:space:]]`	Whitespace, including tab and newline
`[[:upper:]]`, `[[:lower:]]`	Upper- or lowercase letters
`[[:punct:]]`	Punctuation
`[[:xdigit:]]`	Hexadecimal digits

Classes are locale-aware and only valid inside a bracket expression, so write [[:digit:]], not [:digit:].

Ranges are locale-dependent #

Warning

[[:lower:]]+       prefer a class
[a-z]+             may differ between locales

Ranges follow the collating sequence, so prefer named classes in portable patterns.

Matching with grep

Useful options #

Option	Effect
`-E`	Use extended (ERE) syntax
`-i`	Ignore case
`-v`	Select lines that do not match
`-o`	Print only the matched parts
`-w`, `-x`	Match whole words or whole lines
`-c`, `-n`	Count matches or show line numbers
`-q`	No output; exit status only

More search patterns live on the grep sheet.

Everyday patterns #

grep -E '^[[:space:]]*#' app.conf
grep -Ev '^[[:space:]]*(#|$)' app.conf
grep -Eo '[0-9]{1,3}(\.[0-9]{1,3}){3}' access.log
grep -Ei 'warn(ing)?|error' build.log

Quote patterns in single quotes so the shell leaves $, * and | alone.

Basic vs extended syntax #

ERE (`grep -E`)	BRE (`grep`)
`a+`, `a?`	`aa*`, `a\{0,1\}`
`(ab)`	`$ab$`
`a{2,3}`	`a\{2,3\}`
`a\|b`	no portable equivalent

Back-references such as \1 belong to BRE; strict ERE has none.

Other flavors

Perl-style shortcuts #

Perl-style (PCRE, JavaScript)	POSIX ERE spelling
`\d`, `\D`	`[[:digit:]]`, `[^[:digit:]]`
`\s`, `\S`	`[[:space:]]`, `[^[:space:]]`
`\w`, `\W`	`[[:alnum:]_]`, `[^[:alnum:]_]`
`\b`	no ERE equivalent; try `grep -w`

macOS accepts these shortcuts only in its "enhanced" mode, and spells word edges [[:<:]] and [[:>:]].

Features without an ERE equivalent #

(?:abc)          group without capturing
a+?  a*?         lazy (shortest) repetition
(?=next)         lookahead
(?<year>\d{4})   named group
(?i)abc          inline case-insensitive flag

These belong to backtracking engines such as PCRE, JavaScript or the Oniguruma engine used by jq. Test such patterns in the tool that will run them.