Regular expressions (regexps) are patterns which describe the contents of a string. They’re used for testing whether a string contains a given pattern, or extracting the portions that match. They are created with the /pat/ and %r{pat} literals or the Regexp.new constructor.

A regexp is usually delimited with forward slashes (/). For example:

/hay/ =~ 'haystack'   #=> 0
/y/.match('haystack') #=> #<MatchData "y">

If a string contains the pattern it is said to match. A literal string matches itself.

Here ‘haystack’ does not contain the pattern ‘needle’, so it doesn’t match:

/needle/.match('haystack') #=> nil

Here ‘haystack’ contains the pattern ‘hay’, so it matches:

/hay/.match('haystack')    #=> #<MatchData "hay">

Specifically, /st/ requires that the string contains the letter s followed by the letter t, so it matches haystack, also.

Note that any Regexp matching will raise a RuntimeError if timeout is set and exceeded. See “Timeout” section in detail.

Regexp Interpolation

A regexp may contain interpolated strings; trivially:

foo = 'bar'
/#{foo}/ # => /bar/

=~ and Regexp#match

Pattern matching may be achieved by using =~ operator or Regexp#match method.

=~ Operator

=~ is Ruby’s basic pattern-matching operator. When one operand is a regular expression and the other is a string then the regular expression is used as a pattern to match against the string. (This operator is equivalently defined by Regexp and String so the order of String and Regexp do not matter. Other classes may have different implementations of =~.) If a match is found, the operator returns index of first match in string, otherwise it returns nil.

/hay/ =~ 'haystack'   #=> 0
'haystack' =~ /hay/   #=> 0
/a/   =~ 'haystack'   #=> 1
/u/   =~ 'haystack'   #=> nil

Using =~ operator with a String and Regexp the $~ global variable is set after a successful match. $~ holds a MatchData object. Regexp.last_match is equivalent to $~.

Regexp#match Method

The match method returns a MatchData object:

/st/.match('haystack')   #=> #<MatchData "st">

Metacharacters and Escapes

The following are metacharacters (, ), [, ], {, }, ., ?, +, *. They have a specific meaning when appearing in a pattern. To match them literally they must be backslash-escaped. To match a backslash literally, backslash-escape it: \\.

/1 \+ 2 = 3\?/.match('Does 1 + 2 = 3?') #=> #<MatchData "1 + 2 = 3?">
/a\\\\b/.match('a\\\\b')                    #=> #<MatchData "a\\b">

Patterns behave like double-quoted strings and can contain the same backslash escapes (the meaning of \s is different, however, see below).

/\s\u{6771 4eac 90fd}/.match("Go to 東京都")
    #=> #<MatchData " 東京都">

Arbitrary Ruby expressions can be embedded into patterns with the #{...} construct.

place = "東京都"
/#{place}/.match("Go to 東京都")
    #=> #<MatchData "東京都">

Character Classes

A character class is delimited with square brackets ([, ]) and lists characters that may appear at that point in the match. /[ab]/ means a or b, as opposed to /ab/ which means a followed by b.

/W[aeiou]rd/.match("Word") #=> #<MatchData "Word">

Within a character class the hyphen (-) is a metacharacter denoting an inclusive range of characters. [abcd] is equivalent to [a-d]. A range can be followed by another range, so [abcdwxyz] is equivalent to [a-dw-z]. The order in which ranges or individual characters appear inside a character class is irrelevant.

/[0-9a-f]/.match('9f') #=> #<MatchData "9">
/[9f]/.match('9f')     #=> #<MatchData "9">

If the first character of a character class is a caret (^) the class is inverted: it matches any character except those named.

/[^a-eg-z]/.match('f') #=> #<MatchData "f">

A character class may contain another character class. By itself this isn’t useful because [a-z[0-9]] describes the same set as [a-z0-9]. However, character classes also support the && operator which performs set intersection on its arguments. The two can be combined as follows:

/[a-w&&[^c-g]z]/ # ([a-w] AND ([^c-g] OR z))

This is equivalent to:

/[abh-w]/

The following metacharacters also behave like character classes:

POSIX bracket expressions are also similar to character classes. They provide a portable alternative to the above, with the added benefit that they encompass non-ASCII characters. For instance, /\d/ matches only the ASCII decimal digits (0-9); whereas /[[:digit:]]/ matches any character in the Unicode Nd category.

Ruby also supports the following non-POSIX character classes:

Repetition

The constructs described so far match a single character. They can be followed by a repetition metacharacter to specify how many times they need to occur. Such metacharacters are called quantifiers.

At least one uppercase character (‘H’), at least one lowercase character (‘e’), two ‘l’ characters, then one ‘o’:

"Hello".match(/[[:upper:]]+[[:lower:]]+l{2}o/) #=> #<MatchData "Hello">

Greedy Match

Repetition is greedy by default: as many occurrences as possible are matched while still allowing the overall match to succeed. By contrast, lazy matching makes the minimal amount of matches necessary for overall success. Most greedy metacharacters can be made lazy by following them with ?. For the {n} pattern, because it specifies an exact number of characters to match and not a variable number of characters, the ? metacharacter instead makes the repeated pattern optional.

Both patterns below match the string. The first uses a greedy quantifier so ‘.+’ matches ‘<a><b>’; the second uses a lazy quantifier so ‘.+?’ matches ‘<a>’:

/<.+>/.match("<a><b>")  #=> #<MatchData "<a><b>">
/<.+?>/.match("<a><b>") #=> #<MatchData "<a>">

Possessive Match

A quantifier followed by + matches possessively: once it has matched it does not backtrack. They behave like greedy quantifiers, but having matched they refuse to “give up” their match even if this jeopardises the overall match.

/<.*><.+>/.match("<a><b>") #=> #<MatchData "<a><b>">
/<.*+><.+>/.match("<a><b>") #=> nil
/<.*><.++>/.match("<a><b>") #=> nil

Capturing

Parentheses can be used for capturing. The text enclosed by the nth group of parentheses can be subsequently referred to with n. Within a pattern use the backreference \n (e.g. \1); outside of the pattern use MatchData[n] (e.g. MatchData[1]).

In this example, 'at' is captured by the first group of parentheses, then referred to later with \1:

/[csh](..) [csh]\1 in/.match("The cat sat in the hat")
    #=> #<MatchData "cat sat in" 1:"at">

Regexp#match returns a MatchData object which makes the captured text available with its [] method:

/[csh](..) [csh]\1 in/.match("The cat sat in the hat")[1] #=> 'at'

While Ruby supports an arbitrary number of numbered captured groups, only groups 1-9 are supported using the \n backreference syntax.

Ruby also supports \0 as a special backreference, which references the entire matched string. This is also available at MatchData[0]. Note that the \0 backreference cannot be used inside the regexp, as backreferences can only be used after the end of the capture group, and the \0 backreference uses the implicit capture group of the entire match. However, you can use this backreference when doing substitution:

"The cat sat in the hat".gsub(/[csh]at/, '\0s')
  # => "The cats sats in the hats"

Named Captures

Capture groups can be referred to by name when defined with the (?<name>) or (?'name') constructs.

/\$(?<dollars>\d+)\.(?<cents>\d+)/.match("$3.67")
    #=> #<MatchData "$3.67" dollars:"3" cents:"67">
/\$(?<dollars>\d+)\.(?<cents>\d+)/.match("$3.67")[:dollars] #=> "3"

Named groups can be backreferenced with \k<name>, where name is the group name.

/(?<vowel>[aeiou]).\k<vowel>.\k<vowel>/.match('ototomy')
    #=> #<MatchData "ototo" vowel:"o">

Note: A regexp can’t use named backreferences and numbered backreferences simultaneously. Also, if a named capture is used in a regexp, then parentheses used for grouping which would otherwise result in a unnamed capture are treated as non-capturing.

/(\w)(\w)/.match("ab").captures # => ["a", "b"]
/(\w)(\w)/.match("ab").named_captures # => {}

/(?<c>\w)(\w)/.match("ab").captures # => ["a"]
/(?<c>\w)(\w)/.match("ab").named_captures # => {"c"=>"a"}

When named capture groups are used with a literal regexp on the left-hand side of an expression and the =~ operator, the captured text is also assigned to local variables with corresponding names.

/\$(?<dollars>\d+)\.(?<cents>\d+)/ =~ "$3.67" #=> 0
dollars #=> "3"

Grouping

Parentheses also group the terms they enclose, allowing them to be quantified as one atomic whole.

The pattern below matches a vowel followed by 2 word characters:

/[aeiou]\w{2}/.match("Caenorhabditis elegans") #=> #<MatchData "aen">

Whereas the following pattern matches a vowel followed by a word character, twice, i.e. [aeiou]\w[aeiou]\w: ‘enor’.

/([aeiou]\w){2}/.match("Caenorhabditis elegans")
    #=> #<MatchData "enor" 1:"or">

The (?:) construct provides grouping without capturing. That is, it combines the terms it contains into an atomic whole without creating a backreference. This benefits performance at the slight expense of readability.

The first group of parentheses captures ‘n’ and the second ‘ti’. The second group is referred to later with the backreference \2:

/I(n)ves(ti)ga\2ons/.match("Investigations")
    #=> #<MatchData "Investigations" 1:"n" 2:"ti">

The first group of parentheses is now made non-capturing with ‘?:’, so it still matches ‘n’, but doesn’t create the backreference. Thus, the backreference \1 now refers to ‘ti’.

/I(?:n)ves(ti)ga\1ons/.match("Investigations")
    #=> #<MatchData "Investigations" 1:"ti">

Atomic Grouping

Grouping can be made atomic with (?>pat). This causes the subexpression pat to be matched independently of the rest of the expression such that what it matches becomes fixed for the remainder of the match, unless the entire subexpression must be abandoned and subsequently revisited. In this way pat is treated as a non-divisible whole. Atomic grouping is typically used to optimise patterns so as to prevent the regular expression engine from backtracking needlessly.

The " in the pattern below matches the first character of the string, then .* matches Quote“. This causes the overall match to fail, so the text matched by .* is backtracked by one position, which leaves the final character of the string available to match "

/".*"/.match('"Quote"')     #=> #<MatchData "\"Quote\"">

If .* is grouped atomically, it refuses to backtrack Quote“, even though this means that the overall match fails

/"(?>.*)"/.match('"Quote"') #=> nil

Subexpression Calls

The \g<name> syntax matches the previous subexpression named name, which can be a group name or number, again. This differs from backreferences in that it re-executes the group rather than simply trying to re-match the same text.

This pattern matches a ( character and assigns it to the paren group, tries to call that the paren sub-expression again but fails, then matches a literal ):

/\A(?<paren>\(\g<paren>*\))*\z/ =~ '()'

/\A(?<paren>\(\g<paren>*\))*\z/ =~ '(())' #=> 0
# ^1
#      ^2
#           ^3
#                 ^4
#      ^5
#           ^6
#                      ^7
#                       ^8
#                       ^9
#                           ^10
  1. Matches at the beginning of the string, i.e. before the first character.

  2. Enters a named capture group called paren

  3. Matches a literal (, the first character in the string

  4. Calls the paren group again, i.e. recurses back to the second step

  5. Re-enters the paren group

  6. Matches a literal (, the second character in the string

  7. Try to call paren a third time, but fail because doing so would prevent an overall successful match

  8. Match a literal ), the third character in the string. Marks the end of the second recursive call

  9. Match a literal ), the fourth character in the string

  10. Match the end of the string

Alternation

The vertical bar metacharacter (|) combines several expressions into a single one that matches any of the expressions. Each expression is an alternative.

/\w(and|or)\w/.match("Feliformia") #=> #<MatchData "form" 1:"or">
/\w(and|or)\w/.match("furandi")    #=> #<MatchData "randi" 1:"and">
/\w(and|or)\w/.match("dissemblance") #=> nil

Character Properties

The \p{} construct matches characters with the named property, much like POSIX bracket classes.

A Unicode character’s General Category value can also be matched with \p{Ab} where Ab is the category’s abbreviation as described below:

Lastly, \p{} matches a character’s Unicode script. The following scripts are supported: Arabic, Armenian, Balinese, Bengali, Bopomofo, Braille, Buginese, Buhid, Canadian_Aboriginal, Carian, Cham, Cherokee, Common, Coptic, Cuneiform, Cypriot, Cyrillic, Deseret, Devanagari, Ethiopic, Georgian, Glagolitic, Gothic, Greek, Gujarati, Gurmukhi, Han, Hangul, Hanunoo, Hebrew, Hiragana, Inherited, Kannada, Katakana, Kayah_Li, Kharoshthi, Khmer, Lao, Latin, Lepcha, Limbu, Linear_B, Lycian, Lydian, Malayalam, Mongolian, Myanmar, New_Tai_Lue, Nko, Ogham, Ol_Chiki, Old_Italic, Old_Persian, Oriya, Osmanya, Phags_Pa, Phoenician, Rejang, Runic, Saurashtra, Shavian, Sinhala, Sundanese, Syloti_Nagri, Syriac, Tagalog, Tagbanwa, Tai_Le, Tamil, Telugu, Thaana, Thai, Tibetan, Tifinagh, Ugaritic, Vai, and Yi.

Unicode codepoint U+06E9 is named “ARABIC PLACE OF SAJDAH” and belongs to the Arabic script:

/\p{Arabic}/.match("\u06E9") #=> #<MatchData "\u06E9">

All character properties can be inverted by prefixing their name with a caret (^).

Letter ‘A’ is not in the Unicode Ll (Letter; Lowercase) category, so this match succeeds:

/\p{^Ll}/.match("A") #=> #<MatchData "A">

Anchors

Anchors are metacharacter that match the zero-width positions between characters, anchoring the match to a specific position.

If a pattern isn’t anchored it can begin at any point in the string:

/real/.match("surrealist") #=> #<MatchData "real">

Anchoring the pattern to the beginning of the string forces the match to start there. ‘real’ doesn’t occur at the beginning of the string, so now the match fails:

/\Areal/.match("surrealist") #=> nil

The match below fails because although ‘Demand’ contains ‘and’, the pattern does not occur at a word boundary.

/\band/.match("Demand")

Whereas in the following example ‘and’ has been anchored to a non-word boundary so instead of matching the first ‘and’ it matches from the fourth letter of ‘demand’ instead:

/\Band.+/.match("Supply and demand curve") #=> #<MatchData "and curve">

The pattern below uses positive lookahead and positive lookbehind to match text appearing in tags without including the tags in the match:

/(?<=<b>)\w+(?=<\/b>)/.match("Fortune favours the <b>bold</b>")
    #=> #<MatchData "bold">

Options

The end delimiter for a regexp can be followed by one or more single-letter options which control how the pattern can match.

i, m, and x can also be applied on the subexpression level with the (?on-off) construct, which enables options on, and disables options off for the expression enclosed by the parentheses:

/a(?i:b)c/.match('aBc')   #=> #<MatchData "aBc">
/a(?-i:b)c/i.match('ABC') #=> nil

Additionally, these options can also be toggled for the remainder of the pattern:

/a(?i)bc/.match('abC') #=> #<MatchData "abC">

Options may also be used with Regexp.new:

Regexp.new("abc", Regexp::IGNORECASE)                     #=> /abc/i
Regexp.new("abc", Regexp::MULTILINE)                      #=> /abc/m
Regexp.new("abc # Comment", Regexp::EXTENDED)             #=> /abc # Comment/x
Regexp.new("abc", Regexp::IGNORECASE | Regexp::MULTILINE) #=> /abc/mi

Regexp.new("abc", "i")           #=> /abc/i
Regexp.new("abc", "m")           #=> /abc/m
Regexp.new("abc # Comment", "x") #=> /abc # Comment/x
Regexp.new("abc", "im")          #=> /abc/mi

Free-Spacing Mode and Comments

As mentioned above, the x option enables free-spacing mode. Literal white space inside the pattern is ignored, and the octothorpe (#) character introduces a comment until the end of the line. This allows the components of the pattern to be organized in a potentially more readable fashion.

A contrived pattern to match a number with optional decimal places:

float_pat = /\A
    [[:digit:]]+ # 1 or more digits before the decimal point
    (\.          # Decimal point
        [[:digit:]]+ # 1 or more digits after the decimal point
    )? # The decimal point and following digits are optional
\Z/x
float_pat.match('3.14') #=> #<MatchData "3.14" 1:".14">

There are a number of strategies for matching whitespace:

Comments can be included in a non-x pattern with the (?#comment) construct, where comment is arbitrary text ignored by the regexp engine.

Comments in regexp literals cannot include unescaped terminator characters.

Encoding

Regular expressions are assumed to use the source encoding. This can be overridden with one of the following modifiers.

A regexp can be matched against a string when they either share an encoding, or the regexp’s encoding is US-ASCII and the string’s encoding is ASCII-compatible.

If a match between incompatible encodings is attempted an Encoding::CompatibilityError exception is raised.

The Regexp#fixed_encoding? predicate indicates whether the regexp has a fixed encoding, that is one incompatible with ASCII. A regexp’s encoding can be explicitly fixed by supplying Regexp::FIXEDENCODING as the second argument of Regexp.new:

r = Regexp.new("a".force_encoding("iso-8859-1"),Regexp::FIXEDENCODING)
r =~ "a\u3042"
   # raises Encoding::CompatibilityError: incompatible encoding regexp match
   #         (ISO-8859-1 regexp with UTF-8 string)

Regexp Global Variables

Pattern matching sets some global variables :

Example:

m = /s(\w{2}).*(c)/.match('haystack') #=> #<MatchData "stac" 1:"ta" 2:"c">
$~                                    #=> #<MatchData "stac" 1:"ta" 2:"c">
Regexp.last_match                     #=> #<MatchData "stac" 1:"ta" 2:"c">

$&      #=> "stac"
        # same as m[0]
$`      #=> "hay"
        # same as m.pre_match
$'      #=> "k"
        # same as m.post_match
$1      #=> "ta"
        # same as m[1]
$2      #=> "c"
        # same as m[2]
$3      #=> nil
        # no third group in pattern
$+      #=> "c"
        # same as m[-1]

These global variables are thread-local and method-local variables.

Performance

Certain pathological combinations of constructs can lead to abysmally bad performance.

Consider a string of 25 as, a d, 4 as, and a c.

s = 'a' * 25 + 'd' + 'a' * 4 + 'c'
#=> "aaaaaaaaaaaaaaaaaaaaaaaaadaaaac"

The following patterns match instantly as you would expect:

/(b|a)/ =~ s #=> 0
/(b|a+)/ =~ s #=> 0
/(b|a+)*/ =~ s #=> 0

However, the following pattern takes appreciably longer:

/(b|a+)*c/ =~ s #=> 26

This happens because an atom in the regexp is quantified by both an immediate + and an enclosing * with nothing to differentiate which is in control of any particular character. The nondeterminism that results produces super-linear performance. (Consult Mastering Regular Expressions (3rd ed.), pp 222, by Jeffery Friedl, for an in-depth analysis). This particular case can be fixed by use of atomic grouping, which prevents the unnecessary backtracking:

(start = Time.now) && /(b|a+)*c/ =~ s && (Time.now - start)
   #=> 24.702736882
(start = Time.now) && /(?>b|a+)*c/ =~ s && (Time.now - start)
   #=> 0.000166571

A similar case is typified by the following example, which takes approximately 60 seconds to execute for me:

Match a string of 29 as against a pattern of 29 optional as followed by 29 mandatory as:

Regexp.new('a?' * 29 + 'a' * 29) =~ 'a' * 29

The 29 optional as match the string, but this prevents the 29 mandatory as that follow from matching. Ruby must then backtrack repeatedly so as to satisfy as many of the optional matches as it can while still matching the mandatory 29. It is plain to us that none of the optional matches can succeed, but this fact unfortunately eludes Ruby.

The best way to improve performance is to significantly reduce the amount of backtracking needed. For this case, instead of individually matching 29 optional as, a range of optional as can be matched all at once with a{0,29}:

Regexp.new('a{0,29}' + 'a' * 29) =~ 'a' * 29

Timeout

There are two APIs to set timeout. One is Regexp.timeout=, which is process-global configuration of timeout for Regexp matching.

Regexp.timeout = 3
s = 'a' * 25 + 'd' + 'a' * 4 + 'c'
/(b|a+)*c/ =~ s  #=> This raises an exception in three seconds

The other is timeout keyword of Regexp.new.

re = Regexp.new("(b|a+)*c", timeout: 3)
s = 'a' * 25 + 'd' + 'a' * 4 + 'c'
/(b|a+)*c/ =~ s  #=> This raises an exception in three seconds

When using Regexps to process untrusted input, you should use the timeout feature to avoid excessive backtracking. Otherwise, a malicious user can provide input to Regexp causing Denial-of-Service attack. Note that the timeout is not set by default because an appropriate limit highly depends on an application requirement and context.