...one of the most highly
regarded and expertly designed C++ library projects in the
world.
— Herb Sutter and Andrei
Alexandrescu, C++
Coding Standards
The Perl regular expression syntax is based on that used by the programming
language Perl . Perl regular expressions are the default behavior in Boost.Regex
or you can pass the flag perl
to the basic_regex
constructor, for example:
// e1 is a case sensitive Perl regular expression: // since Perl is the default option there's no need to explicitly specify the syntax used here: boost::regex e1(my_expression); // e2 a case insensitive Perl regular expression: boost::regex e2(my_expression, boost::regex::perl|boost::regex::icase);
In Perl regular expressions, all characters match themselves except for the following special characters:
.[{}()\*+?|^$
Other characters are special only in certain situations - for example ]
is special only after an opening [
.
The single character '.' when used outside of a character set will match any single character except:
match_not_dot_null
is passed to the matching
algorithms.
match_not_dot_newline
is passed to the matching
algorithms.
A '^' character shall match the start of a line.
A '$' character shall match the end of a line.
A section beginning (
and ending )
acts as a marked sub-expression. Whatever matched the sub-expression is split
out in a separate field by the matching algorithms. Marked sub-expressions
can also repeated, or referred to by a back-reference.
A marked sub-expression is useful to lexically group part of a regular expression,
but has the side-effect of spitting out an extra field in the result. As
an alternative you can lexically group part of a regular expression, without
generating a marked sub-expression by using (?:
and )
, for example (?:ab)+
will repeat ab
without splitting out any separate sub-expressions.
Any atom (a single character, a marked sub-expression, or a character class)
can be repeated with the *
, +
, ?
,
and {}
operators.
The *
operator will match the preceding atom zero or more
times, for example the expression a*b
will match any of
the following:
b ab aaaaaaaab
The +
operator will match the preceding atom one or more
times, for example the expression a+b
will match any of
the following:
ab aaaaaaaab
But will not match:
b
The ?
operator will match the preceding atom zero or one
times, for example the expression ca?b will match any of the following:
cb cab
But will not match:
caab
An atom can also be repeated with a bounded repeat:
a{n}
Matches 'a' repeated exactly n times.
a{n,}
Matches 'a' repeated n or more times.
a{n, m}
Matches 'a' repeated between n and m times inclusive.
For example:
^a{2,3}$
Will match either of:
aa aaa
But neither of:
a aaaa
Note that the "{" and "}" characters will treated as ordinary literals when used in a context that is not a repeat: this matches Perl 5.x behavior. For example in the expressions "ab{1", "ab1}" and "a{b}c" the curly brackets are all treated as literals and no error will be raised.
It is an error to use a repeat operator, if the preceding construct can not be repeated, for example:
a(*)
Will raise an error, as there is nothing for the *
operator
to be applied to.
The normal repeat operators are "greedy", that is to say they will consume as much input as possible. There are non-greedy versions available that will consume as little input as possible while still producing a match.
*?
Matches the previous atom zero or more times, while
consuming as little input as possible.
+?
Matches the previous atom one or more times, while
consuming as little input as possible.
??
Matches the previous atom zero or one times, while
consuming as little input as possible.
{n,}?
Matches the previous atom n or more times, while
consuming as little input as possible.
{n,m}?
Matches the previous atom between n and m times,
while consuming as little input as possible.
By default when a repeated pattern does not match then the engine will backtrack until a match is found. However, this behaviour can sometime be undesireble so there are also "possessive" repeats: these match as much as possible and do not then allow backtracking if the rest of the expression fails to match.
*+
Matches the previous atom zero or more times, while
giving nothing back.
++
Matches the previous atom one or more times, while
giving nothing back.
?+
Matches the previous atom zero or one times, while
giving nothing back.
{n,}+
Matches the previous atom n or more times, while
giving nothing back.
{n,m}+
Matches the previous atom between n and m times,
while giving nothing back.
An escape character followed by a digit n, where n is in the range 1-9, matches the same string that was matched by sub-expression n. For example the expression:
^(a*)[^a]*\1$
Will match the string:
aaabbaaa
But not the string:
aaabba
You can also use the \g escape for the same function, for example:
Escape |
Meaning |
---|---|
|
Match whatever matched sub-expression 1 |
|
Match whatever matched sub-expression 1: this form allows for safer
parsing of the expression in cases like |
|
Match whatever matched the last opened sub-expression |
|
Match whatever matched the last but one opened sub-expression |
|
Match whatever matched the sub-expression named "one" |
Finally the \k escape can be used to refer to named subexpressions, for example
\k<two>
will match whatever matched the subexpression
named "two".
The |
operator will match either of its arguments, so
for example: abc|def
will match either "abc"
or "def".
Parenthesis can be used to group alternations, for example: ab(d|ef)
will match either of "abd" or "abef".
Empty alternatives are not allowed (these are almost always a mistake), but
if you really want an empty alternative use (?:)
as a
placeholder, for example:
|abc
is not a valid expression, but
(?:)|abc
is and is equivalent, also the expression:
(?:abc)??
has exactly the same effect.
A character set is a bracket-expression starting with [] and ending
with
, it defines a set of characters, and matches
any single character that is a member of that set.
A bracket expression may contain any combination of the following:
For example [abc]
, will match any of the characters 'a',
'b', or 'c'.
For example [a-c]
will match any single character in the
range 'a' to 'c'. By default, for Perl regular expressions, a character x
is within the range y to z, if the code point of the character lies within
the codepoints of the endpoints of the range. Alternatively, if you set the
collate
flag when constructing the regular expression, then ranges are locale
sensitive.
If the bracket-expression begins with the ^ character, then it matches the
complement of the characters it contains, for example [^a-c]
matches any character that is not in the range a-c
.
An expression of the form [[:name:]]
matches the named
character class "name", for example [[:lower:]]
matches any lower case character. See character
class names.
An expression of the form [[.col.]]
matches the collating
element col. A collating element is any single character,
or any sequence of characters that collates as a single unit. Collating elements
may also be used as the end point of a range, for example: [[.ae.]-c]
matches the character sequence "ae", plus any single character
in the range "ae"-c, assuming that "ae" is treated as
a single collating element in the current locale.
As an extension, a collating element may also be specified via it's symbolic name, for example:
[[.NUL.]]
matches a \0
character.
An expression of the form [[=col=]]
, matches any character
or collating element whose primary sort key is the same as that for collating
element col, as with collating elements the name col
may be a symbolic name.
A primary sort key is one that ignores case, accentation, or locale-specific
tailorings; so for example [[=a=]]
matches
any of the characters: a, À, Á, Â, Ã, Ä, Å, A, à, á, â, ã, ä and å. Unfortunately implementation
of this is reliant on the platform's collation and localisation support;
this feature can not be relied upon to work portably across all platforms,
or even all locales on one platform.
All the escape sequences that match a single character, or a single character
class are permitted within a character class definition. For example [\[\]]
would match either of [
or ]
while [\W\d]
would match any character that is either a "digit", or
is not a "word" character.
All of the above can be combined in one character set declaration, for example:
[[:digit:]a-c[.NUL.]]
.
Any special character preceded by an escape shall match itself.
The following escape sequences are all synonyms for single characters:
Escape |
Character |
---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
An ASCII escape sequence - the character whose code point is X % 32 |
|
A hexadecimal escape sequence - matches the single character whose code point is 0xdd. |
|
A hexadecimal escape sequence - matches the single character whose code point is 0xdddd. |
|
An octal escape sequence - matches the single character whose code point is 0ddd. |
|
Matches the single character which has the symbolic
name name. For example |
Any escaped character x, if x is the name of a character class shall match any character that is a member of that class, and any escaped character X, if x is the name of a character class, shall match any character not in that class.
The following are supported by default:
Escape sequence |
Equivalent to |
---|---|
|
|
|
|
|
|
|
|
|
|
|
Horizontal whitespace |
|
Vertical whitespace |
|
|
|
|
|
|
|
|
|
|
|
Not Horizontal whitespace |
|
Not Vertical whitespace |
The character property names in the following table are all equivalent to the names used in character classes.
Form |
Description |
Equivalent character set form |
---|---|---|
|
Matches any character that has the property X. |
|
|
Matches any character that has the property Name. |
|
|
Matches any character that does not have the property X. |
|
|
Matches any character that does not have the property Name. |
|
For example \pd
matches any "digit" character,
as does \p{digit}
.
The following escape sequences match the boundaries of words:
<
Matches the start of a word.
>
Matches the end of a word.
\b
Matches a word boundary (the start or end of a word).
\B
Matches only when not at a word boundary.
The following match only at buffer boundaries: a "buffer" in this context is the whole of the input text that is being matched against (note that ^ and $ may match embedded newlines within the text).
\` Matches at the start of a buffer only.
\' Matches at the end of a buffer only.
\A Matches at the start of a buffer only (the same as \`
).
\z Matches at the end of a buffer only (the same as \'
).
\Z Matches a zero-width assertion consisting of an optional sequence of newlines
at the end of a buffer: equivalent to the regular expression (?=\v*\z)
.
Note that this is subtly different from Perl which behaves as if matching
(?=\n?\z)
.
The sequence \G
matches only at the end of the last match
found, or at the start of the text being matched if no previous match was
found. This escape useful if you're iterating over the matches contained
within a text, and you want each subsequence match to start where the last
one ended.
The escape sequence \Q
begins a "quoted sequence":
all the subsequent characters are treated as literals, until either the end
of the regular expression or \E is found. For example the expression: \Q*+\Ea+
would match either of:
\*+a \*+aaa
\C
Matches a single code point: in Boost regex this has
exactly the same effect as a "." operator. \X
Matches a combining character sequence: that is any non-combining character
followed by a sequence of zero or more combining characters.
The escape sequence \R
matches any line ending character
sequence, specifically it is identical to the expression (?>\x0D\x0A?|[\x0A-\x0C\x85\x{2028}\x{2029}])
.
\K
Resets the start location of $0 to the current text
position: in other words everything to the left of \K is "kept back"
and does not form part of the regular expression match. $` is updated accordingly.
For example foo\Kbar
matched against the text "foobar"
would return the match "bar" for $0 and "foo" for $`.
This can be used to simulate variable width lookbehind assertions.
Any other escape sequence matches the character that is escaped, for example \@ matches a literal '@'.
Perl-specific extensions to the regular expression syntax all start with
(?
.
You can create a named subexpression using:
(?<NAME>expression)
Which can be then be referred to by the name NAME. Alternatively you can delimit the name using 'NAME' as in:
(?'NAME'expression)
These named subexpressions can be referred to in a backreference using either
\g{NAME}
or \k<NAME>
and can
also be referred to by name in a Perl
format string for search and replace operations, or in the match_results
member functions.
(?# ... )
is treated as a comment, it's contents are ignored.
(?imsx-imsx ... )
alters which of the perl modifiers are
in effect within the pattern, changes take effect from the point that the
block is first seen and extend to any enclosing )
. Letters
before a '-' turn that perl modifier on, letters afterward, turn it off.
(?imsx-imsx:pattern)
applies the specified modifiers to
pattern only.
(?:pattern)
lexically groups pattern, without generating
an additional sub-expression.
(?|pattern)
resets the subexpression count at the start
of each "|" alternative within pattern.
The sub-expression count following this construct is that of whichever branch had the largest number of sub-expressions. This construct is useful when you want to capture one of a number of alternative matches in a single sub-expression index.
In the following example the index of each sub-expression is shown below the expression:
# before ---------------branch-reset----------- after / ( a ) (?| x ( y ) z | (p (q) r) | (t) u (v) ) ( z ) /x # 1 2 2 3 2 3 4
(?=pattern)
consumes zero characters, only if pattern
matches.
(?!pattern)
consumes zero characters, only if pattern
does not match.
Lookahead is typically used to create the logical AND of two regular expressions, for example if a password must contain a lower case letter, an upper case letter, a punctuation symbol, and be at least 6 characters long, then the expression:
(?=.*[[:lower:]])(?=.*[[:upper:]])(?=.*[[:punct:]]).{6,}
could be used to validate the password.
(?<=pattern)
consumes zero characters, only if pattern
could be matched against the characters preceding the current position (pattern
must be of fixed length).
(?<!pattern)
consumes zero characters, only if pattern
could not be matched against the characters preceding the current position
(pattern must be of fixed length).
(?>pattern)
pattern is matched
independently of the surrounding patterns, the expression will never backtrack
into pattern. Independent sub-expressions are typically
used to improve performance; only the best possible match for pattern will
be considered, if this doesn't allow the expression as a whole to match then
no match is found at all.
(?N) (?-N) (?+N)
(?R) (?0) (?&NAME)
(?R)
and (?0)
recurse to the start
of the entire pattern.
(?N)
executes sub-expression N
recursively, for example (?2)
will recurse to sub-expression
2.
(?-N)
and (?+N)
are relative recursions, so for example (?-1)
recurses
to the last sub-expression to be declared, and (?+1)
recurses
to the next sub-expression to be declared.
(?&NAME)
recurses to named sub-expression NAME.
(?(condition)yes-pattern|no-pattern)
attempts to match
yes-pattern if the condition is
true, otherwise attempts to match no-pattern.
(?(condition)yes-pattern)
attempts to match yes-pattern
if the condition is true, otherwise matches the NULL
string.
condition may be either: a forward lookahead assert, the index of a marked sub-expression (the condition becomes true if the sub-expression has been matched), or an index of a recursion (the condition become true if we are executing directly inside the specified recursion).
Here is a summary of the possible predicates:
(?(?=assert)yes-pattern|no-pattern)
Executes yes-pattern
if the forward look-ahead assert matches, otherwise executes no-pattern.
(?(?!assert)yes-pattern|no-pattern)
Executes yes-pattern
if the forward look-ahead assert does not match, otherwise executes
no-pattern.
(?(N)yes-pattern|no-pattern)
Executes yes-pattern if subexpression N
has been matched, otherwise executes no-pattern.
(?(<name>)yes-pattern|no-pattern)
Executes yes-pattern if named subexpression name
has been matched, otherwise executes no-pattern.
(?('name')yes-pattern|no-pattern)
Executes yes-pattern if named subexpression name
has been matched, otherwise executes no-pattern.
(?(R)yes-pattern|no-pattern)
Executes yes-pattern
if we are executing inside a recursion, otherwise executes no-pattern.
(?(RN)yes-pattern|no-pattern)
Executes yes-pattern if we are executing inside
a recursion to sub-expression N, otherwise executes
no-pattern.
(?(R&name)yes-pattern|no-pattern)
Executes yes-pattern if we are executing inside
a recursion to named sub-expression name, otherwise
executes no-pattern.
(?(DEFINE)never-exectuted-pattern)
Defines a block
of code that is never executed and matches no characters: this is usually
used to define one or more named sub-expressions which are referred to
from elsewhere in the pattern.
This library has partial support for Perl's backtracking control verbs, in particular (*MARK) is not supported. There may also be detail differences in behaviour between this library and Perl, not least because Perl's behaviour is rather under-documented and often somewhat random in how it behaves in practice. The verbs supported are:
(*PRUNE)
Has no effect unless backtracked onto, in
which case all the backtracking information prior to this point is discarded.
(*SKIP)
Behaves the same as (*PRUNE)
except that it is assumed that no match can possibly occur prior to the
current point in the string being searched. This can be used to optimize
searches by skipping over chunks of text that have already been determined
can not form a match.
(*THEN)
Has no effect unless backtracked onto, in
which case all subsequent alternatives in a group of alternations are
discarded.
(*COMMIT)
Has no effect unless backtracked onto, in
which case all subsequent matching/searching attempts are abandoned.
(*FAIL)
Causes the match to fail unconditionally at
this point, can be used to force the engine to backtrack.
(*ACCEPT)
Causes the pattern to be considered matched
at the current point. Any half-open sub-expressions are closed at the
current point.
The order of precedence for of operators is as follows:
[==]
[::] [..]
\
[]
()
* + ? {m,n}
If you view the regular expression as a directed (possibly cyclic) graph, then the best match found is the first match found by a depth-first-search performed on that graph, while matching the input text.
Alternatively:
The best match found is the leftmost match, with individual elements matched as follows;
Construct |
What gets matched |
---|---|
|
Locates the best match for AtomA that has a following match for AtomB. |
|
If Expresion1 can be matched then returns that match, otherwise attempts to match Expression2. |
|
Matches S repeated exactly N times. |
|
Matches S repeated between N and M times, and as many times as possible. |
|
Matches S repeated between N and M times, and as few times as possible. |
|
The same as |
|
The same as |
|
Matches the best match for S, and only that. |
|
Matches only the best match for S (this is only visible if there are capturing parenthesis within S). |
|
Considers only whether a match for S exists or not. |
|
If condition is true, then only yes-pattern is considered, otherwise only no-pattern is considered. |
The options
normal
, ECMAScript
, JavaScript
and JScript
are all synonyms for perl
.
There are a variety
of flags that may be combined with the perl
option
when constructing the regular expression, in particular note that the newline_alt
option alters the syntax, while the collate
, nosubs
and icase
options modify how the case and locale sensitivity
are to be applied.
The perl smix
modifiers can either be applied using a
(?smix-smix)
prefix to the regular expression, or with
one of the regex-compile
time flags no_mod_m
, mod_x
, mod_s
,
and no_mod_s
.