Regex Clinic
Regex Clinic
net @a
ConFoo 2010
terminology
Regex
a pattern describing a set of strings
abcdef
Friday, March 12, 2010
terminology
apple
Subject String
text that the regex is applied to
Friday, March 12, 2010
terminology
a apple
Match
a portion of the string that is successfully described by the regex
Friday, March 12, 2010
terminology
Engine
A program or a library that obtains matches given a regex and a string
PCRE
Friday, March 12, 2010
color legend
regular expression
subject string
match
Syntax
characters
Special set is a well-dened subset of ASCII Ordinary set consist of all characters not designated special Special characters are also called metacharacters
a 4 0 K x . ! * ? ^
matching literals
123
Friday, March 12, 2010
The most basic regex consists of a single ordinary character It matches the rst occurrence of that character in the string Characters can be added together to form longer regexes
extended characters
To match an extended character, use \xhh notation where hh are hexadecimal digits To match Unicode characters (in UTF-8 mode) mode use \x{hhh..} notation
metacharacters
To use one of these literally, escape it, that is prepend it with a backslash
\$
Friday, March 12, 2010
. [] () ^$ *+? {} |
metacharacters
To escape a sequence of characters, put them between \Q and \E
Price is \Q$12.36\E
will match
Price is $12.36
. [] () ^$ *+? {} |
metacharacters
So will the backslashed version
Price is \$12\.36
will match
Price is $12.36
. [] () ^$ *+? {} |
character classes
[]
Consist of a set of characters placed inside square brackets Matches one and only one of the characters specied inside the class
character classes
matches an English vowel (lowercase)
[]
[aeiou] [st]urf
negated classes
Placing a caret as the rst character after the opening bracket negates the class Will match any character not in the class, including newlines [^<>] would match a character that is not left or right bracket
[^]
character ranges
Placing a dash (-) between two characters creates a range from the rst one to the second one Useful for abbreviating a list of characters
[-]
[a-z]
Friday, March 12, 2010
character ranges
Ranges can be reversed
[-]
[z-a]
Friday, March 12, 2010
character ranges
Ranges can be reversed A class can have more than one range and combine ranges with normal lists
[-]
[a-z0-9:]
Friday, March 12, 2010
\w \d \s \W \D \S
word character decimal digit whitespace not a word character not a decimal digit not whitespace
[-]
] ^
Friday, March 12, 2010
\ -
Inside a character class, most metacharacters lose their meaning Exceptions are:
] ^
Friday, March 12, 2010
\ -
Inside a character class, most metacharacters lose their meaning Exceptions are:
closing bracket
] ^
Friday, March 12, 2010
\ -
Inside a character class, most metacharacters lose their meaning Exceptions are:
closing bracket backslash
] ^
Friday, March 12, 2010
\ -
Inside a character class, most metacharacters lose their meaning Exceptions are:
closing bracket backslash caret
] ^
Friday, March 12, 2010
\ -
Inside a character class, most metacharacters lose their meaning Exceptions are:
closing bracket backslash caret dash
To use them literally, either escape them with a backslash or put them where they do not have special meaning
dot metacharacter
By default matches any single character
dot metacharacter
By default matches any single character Except a newline
\n
dot metacharacter
Is equivalent to
[^\n]
dot metacharacter
12345
Use dot carefully - it might match something you did not intend 12.45 will match literal 12.45 But it will also match these:
.
12945 12a45 12-45 78812 45839
quantiers
quantiers ?
Quantiers help us deal with this uncertainty
* + {}
quantiers ?
They specify how many times a regex component must repeat in order for the match to be successful
* + {}
repeatable components a
literal character
.
dot metacharacter
\w \d \s \W \D \S
range shortcuts
[]
character class
Friday, March 12, 2010
subpattern backreference
zero-or-one
Indicates that the preceding component is optional Regex welcome!? will match either welcome or welcome!
Regex super\s?strong means that super and strong may have an optional whitespace character between them Regex hello[!?]? Will match hello, hello!, or hello?
Indicates that the preceding component has to appear once or more Regex a+h will match ah, aah, aaah, etc Regex -\d+ will match negative integers, such as -33 Regex [^]+ means to match a sequence (more than one) of characters until the next quote
one-or-more
Friday, March 12, 2010
zero-or-more
Indicates that the preceding component can match zero or more times Regex \d+\.\d* will match 2., 3.1, 0.001 Regex <[a-z][a-z0-9]*> will match an opening HTML tag with no attributes, such as <b> or <h2>, but not <> or </i>
general repetition
{}
Species the minimum and the maximum number of times a component has to match Regex ha{1,3} matches ha, haa, haaa Regex \d{8} matches exactly 8 digits If second number is omitted, no upper range is set Regex go{2,}al matches gooal, goooal, gooooal, etc
general repetition
{}
? + *
= = =
greediness
greediness
PHP 5? \d{2,4}
Friday, March 12, 2010
10/26/2004 2004
greediness
Quantiers try to grab as much as possible by default Applying <.+> to <i>greediness</i> matches the whole string rather than just <i>
greediness
If the entire match fails because they consumed too much, then they are forced to give up as much as needed to make the rest of regex succeed
greediness
To nd words ending in ness, you will probably use \w+ness On the rst run \w+ takes the whole word But since ness still has to match, it gives up the last 4 characters and the match succeeds
overcoming greediness
The simplest solution is to make the repetition operators non-greedy, or lazy Lazy quantiers grab as little as possible If the overall match fails, they grab a little more and the match is tried again
overcoming greediness
*? +? { , }? ??
Friday, March 12, 2010
To make a greedy quantier lazy, append ? Note that this use of the question mark is different from its use as a regular quantier
overcoming greediness
*?
Applying <.+?>
+?
to <i> <i>greediness</i>
{ , }? ??
Friday, March 12, 2010
gets us <i>
overcoming greediness
Another option is to use negated character classes More efcient and clearer than lazy repetition
overcoming greediness
<.+?> can be turned into <[^>]+> Note that the second version will match tags spanning multiple lines Single-line version: <[^>\r\n]+>
The most common type of an assertion is an anchor Anchor matches a certain position in the subject string
caret
^
^F
Caret, or circumex, is an anchor that matches at the beginning of the subject string ^F basically means that the subject string has to start with an F
Fandango F
dollar sign
Dollar sign is an anchor that matches at the end of the subject string or right before the string-ending newline \d$ means that the subject string has to end with a digit The string may be top 10 or top 10\n, but either one will match
$
\d$
top 10 0
multiline matching
Often subject strings consist of multiple lines If the multiline option is set:
Caret (^) also matches immediately after any newlines Dollar sign ($) also matches immediately before any newlines
^t.+
absolute start/end
Sometimes you really want to match the absolute start or end of the subject string when in the multiline mode These assertions are always valid: \A matches only at the very beginning \z matches only at the very end \Z matches like $ used in single-line
mode
\At.+
word boundaries
\b \B
\bto\b
A word boundary is a position in the string with a word character (\w) on one side and a non-word character (or string boundary) on the other \b matches when the current position is a word boundary
word boundaries
\b \B
\B2\B
A word boundary is a position in the string with a word character (\w) on one side and a non-word character (or string boundary) on the other \b matches when the current position is a word boundary
doc2html 2
subpatterns
()
Parentheses can be used group a part of the regex together, creating a subpattern You can apply regex operators to a subpattern as a whole
grouping
()
Regex is(land)? matches both is and island Regex (\d\d,)*\d\d will match a commaseparated list of double-digit numbers
capturing subpatterns
()
All subpatterns by default are capturing A capturing subpattern stores the corresponding matched portion of the subject string in memory for later use
capturing subpatterns
Subpatterns are numbered by counting their opening parentheses from left to right Regex (\d\d-(\w+)-\d{4}) has two subpatterns
()
(\d\d-(\w+)-\d{4})
12-May-2004
capturing subpatterns
Subpatterns are numbered by counting their opening parentheses from left to right Regex (\d\d-(\w+)-\d{4}) has two subpatterns When run against 12-May-2004 the second subpattern will capture May
()
(\d\d-(\w+)-\d{4}) (\w+)
12-May-2004 May
non-capturing subpatterns
The capturing aspect of subpatterns is not always necessary It requires more memory and more processing time
non-capturing subpatterns
Using ?: after the opening parenthesis makes a subpattern be a purely grouping one Regex box(?:ers)? will match boxers but will not capture anything The (?:) subpatterns are not included in the subpattern numbering
named subpatterns
It can be hard to keep track of subpattern numbers in a complicated regex Using ?P<name> after the opening parenthesis creates a named subpattern Named subpatterns are still assigned numbers Pattern (?P<number>\d+) will match and capture 99 into subpattern named number when run against 99 bottles
Alternation operator allows testing several sub-expressions at a given point The branches are tried in order, from left to right, until one succeeds Empty alternatives are permitted Regex sailing|cruising will match either sailing or cruising
alternation
Friday, March 12, 2010
Since alternation has the lowest precedence, grouping is often necessary sixth|seventh sense will match the word sixth or the phrase seventh sense (sixth|seventh) sense will match sixth sense or seventh sense
alternation
Friday, March 12, 2010
Remember that the regex engine is eager It will return a match as soon as it nds one camel|came|camera will only match came when run against camera Put more likely pattern as the rst alternative
alternation
Friday, March 12, 2010
Applying ? to assertions is not permitted but.. The branches may contain assertions, such as anchors for example (^|my|your) friend will match friend at the beginning of the string and after my or your
alternation
Friday, March 12, 2010
backtracking
Also known as if at rst you dont succeed, try, try again When faced with several options it could try to achieve a match, the engine picks one and remembers the others
backtracking
If the picked option does not lead to an overall successful match, the engine backtracks to the decision point and tries another option
backtracking
This continues until an overall match succeeds or all the options are exhausted The decision points include quantiers and alternation
backtracking
Two important rules to remember
With greedy quantiers the engine always attempts the match, and with lazy ones it delays the match If there were several decision points, the engine always goes back to the most recent one
backtracking example
\d+00
12300
start
backtracking example
\d+00
12300 1
add 1
backtracking example
\d+00
12300 12
add 2
backtracking example
\d+00
12300 123
add 3
backtracking example
\d+00
12300 1230
add 0
backtracking example
\d+00
12300
add 0
backtracking example
\d+00
12300
string exhausted still need to match 00
backtracking example
\d+00
12300 1230
give up 0
backtracking example
\d+00
12300 123
give up 0
backtracking example
\d+00
12300
add 00
backtracking example
\d+00
12300
success
backtracking example
\d+ff
123dd
start
backtracking example
\d+ff
123dd 1
add 1
backtracking example
\d+ff
123dd 12
add 2
backtracking example
\d+ff
123 123dd
add 3
backtracking example
\d+ff
123 123dd
cannot match f here
backtracking example
\d+ff
123dd 12
give up 3 still cannot match f
backtracking example
\d+ff
123dd 1
give up 2 still cannot match f
backtracking example
\d+ff
123dd 1
cannot give up more because of +
backtracking example
\d+ff
123dd
failure
atomic grouping
Disabling backtracking can be useful The main goal is to speed up failed matches, especially with nested quantiers
atomic grouping
(?>regex) will treat regex as a single atomic token, no backtracking will occur inside it All the saved states are forgotten
atomic grouping
(?>\d+)ff will lock up all available digits and fail right away if the next two characters are not ff Atomic groups are not capturing
possessive quantiers
Atomic groups can be arbitrarily complex and nested Possessive quantiers are simpler and apply to a single repeated item
possessive quantiers
possessive quantiers
Other ones are *+, ?+, and {m,n}+ Possessive quantiers are always greedy
do not over-optimize
Keep in mind that atomic grouping and possessive quantiers can change the outcome of the match When run against string abcdef
\w+d will match abcd \w++d will not match at all \w+ will match the whole string
backreferences
\n
A backreference is an alias to a capturing subpattern It matches whatever the referent capturing subpattern has matched
backreferences
(re|le)\w+\1 matches words that start with re or le and end with the same thing For example, retire and legible, but not revocable or lecture Reference to a named subpattern can be made with (?P=name)
\n
lookaround
Assertions that test whether the characters before or after the current point match the given regex Consume no characters Do not capture anything Includes lookahead and lookbehind
positive lookahead
(?=)
Tests whether the characters after the current point match the given regex (\w+)(?=:)(.*) matches surng: a sport but colon ends up in the second subpattern
negative lookahead
Tests whether the characters after the current point do not match the given regex sh(?!ing) matches sh not followed by ing Will match sherman and shed
(?!)
negative lookahead
Difcult to do with character classes sh[^i][^n][^g] might work but will consume more than needed and fail on subjects shorter than 7 letters Character classes are no help at all with something like sh(?!hook|ing)
(?!)
positive lookbehind
(?<=)
Tests whether the characters immediately preceding the current point match the given regex The regex must be of xed size, but branches are allowed (?<=foo)bar matches bar only if preceded by foo, e.g. my foobar
negative lookbehind
Tests whether the characters immediately preceding the current point do not match the given regex Once again, regex must be of xed size (?<!foo)bar matches bar only if not preceded by foo, e.g. in the bar but not my foobar
(?<!)
conditionals
Conditionals let you apply a regex selectively or to choose between two regexes depending on a previous match (?(condition)yes-regex) (?(condition)yes-regex|no-regex) There are 3 kinds of conditions
Subpattern match Lookaround assertion Recursive call (not discussed here)
subpattern conditions
(?(n))
This condition is satised if the capturing subpattern number n has previously matched ()? \b\w+\b (?(1)) matches words optionally enclosed by quotes There is a difference between ()? and (?) in this case: the second one will always capture
assertion conditions
This type of condition relies on lookaround assertions to choose one path or the other href=(? (?=[]) ([])\S+\1 | \S+)
Matches href=, then If the next character is single or double quote match a sequence of non-whitespace inside the matching quotes Otherwise just match it without quotes
inline options
The matching can be modied by options you put in the regular expression
enables case-insensitive mode enables multiline matching for ^ and $ makes dot metacharacter match newline also ignores literal whitespace makes quantiers ungreedy (lazy) by default
inline options
(?i) (?m) (?s) (?x) (?U)
Friday, March 12, 2010
Options can be combined and unset (?im-sx) At top level, apply to the whole pattern Localized inside subpatterns (a(?i)b)c
comments
?#
comments
Let me blow that up for you
?#
comments
?#
Most regexes could denitely use some comments (?#) species a comment
comments
If (?x) option is set, anything after # outside a character class and up to the next newline is considered a comment To match literal whitespace, escape it
?#
(?x) \w+ # start with word characters [?!] # and end with ? or !
Regex Toolkit
regex toolkit
In your day-to-day development, you will frequently nd yourself running into situations calling for regular expressions It is useful to have a toolkit from which you can quickly draw the solution It is also important to know how to avoid problems in the regexes themselves
In matching (extraction) the regex must account for boundary conditions In validation your boundary conditions are known the whole string
\b[A-Z][a-zA-Z-]*\b
Validating that a string fullls the same condition
^[A-Z][a-zA-Z-]*$
Do not forget ^ and $ anchors for validation!
extracting markup
Possible to use preg_match_all() for grabbing marked up portions But for tokenizing approach, preg_split() is better
$s = 'a <b><I>test</I></b> of <br /> markup'; $tokens = preg_split( '!( < /? [a-zA-Z][a-zA-Z0-9]* [^/>]* /? > ) !x', $s, -1, PREG_SPLIT_NO_EMPTY | PREG_SPLIT_DELIM_CAPTURE); result is array('a','<b>','<I>','test','</I>', '</b>','of','<br />','markup')
restricting markup
Suppose you want to strip all markup except for some allowed subset. What are your possible approaches?
Use strip_tags() - which has limited functionality Multiple invocations of str_replace() or preg_replace() to remove script blocks, etc Custom tokenizer and processor, or..
restricting markup
$s = preg_replace_callback( '! < (/?) ([a-zA-Z][a-zA-Z0-9]*) ([^/>]*) (/?) > !x', 'my_strip', $s); function my_strip($match) { static $allowed_tags = array('b', 'i', 'p', 'br', 'a'); $tag = $match[2]; $attrs = $match[3]; if (!in_array($tag, $allowed_tags)) return ; if (!empty($match[1])) return "</$tag>"; /* strip evil attributes here */ if ($tag == 'a') { $attrs = ''; } /* any other kind of processing here */ return "<$tag$attrs$match[4]>"; }
matching numbers
Integers are easy: \b\d+\b Floating point numbers:
integer.fractional .fractional
matching numbers
To match both integers and oating point numbers, either combine them with alternation or use: ((\b\d+)?\.)?\b\d+\b [+-]? can be prepended to any of these, if sign matching is needed \b can be substituted by more appropriate assertions based on the required delimiters
Yeah, right The complete regex is about as long as a book page in 10-point type Buy a copy of Jeffrey Friedls book and steal it from there
tips
Dont do everything in regex a lot of tasks are best left to PHP Use string functions for simple tasks Make sure you know how backtracking works
tips
Be aware of the context Capture only what you intend to use Dont use single-character classes
tips
Lazy vs. greedy, be specic Put most likely alternative rst in the alternation list Think!
regex tools
Rubular.com Regex buddy Komodo Regex tester (Rx toolkit) Reggy (reggyapp.com) RegExr (http://www.gskinner.com/RegExr/) http://www.spaweditor.com/scripts/regex/index.php http://regex.larsolavtorvik.com/
Thank You!
Questions?
http://zmievski.org/talks