++ed by:

6 PAUSE users
10 non-PAUSE users.

Damian Conway


Text::Balanced - Extract delimited text sequences from strings.


 use Text::Balanced qw (



 # Extract the initial substring of $text that is delimited by
 # two (unescaped) instances of the first character in $delim.

        ($extracted, $remainder) = extract_delimited($text,$delim);

 # Extract the initial substring of $text that is bracketed
 # with a delimiter(s) specified by $delim (where the string
 # in $delim contains one or more of '(){}[]<>').

        ($extracted, $remainder) = extract_bracketed($text,$delim);

 # Extract the initial substring of $text that is bounded by
 # an HTML/XML tag.

        ($extracted, $remainder) = extract_tagged($text);

 # Extract the initial substring of $text that is bounded by
 # a C<BEGIN>...C<END> pair. Don't allow nested C<BEGIN> tags

        ($extracted, $remainder) =

 # Extract the initial substring of $text that represents a
 # Perl "quote or quote-like operation"

        ($extracted, $remainder) = extract_quotelike($text);

 # Extract the initial substring of $text that represents a block
 # of Perl code, bracketed by any of character(s) specified by $delim
 # (where the string $delim contains one or more of '(){}[]<>').

        ($extracted, $remainder) = extract_codeblock($text,$delim);

 # Extract the initial substrings of $text that would be extracted by
 # one or more sequential applications of the specified functions.

        @extracted = extract_multiple($text,
                                      [ \&extract_bracketed,

# Create a string representing an optimized pattern (a la Friedl) # that matches a substring delimited by any of the specified characters # (in this case: any type of quote or a slash)

        $patstring = delimited_pat(q{'"`/});

# Generate a reference to an anonymous sub that is just like extract_tagged # but pre-compiled and optimized for a specific pair of tags, and consequently # much faster (i.e. 3 times faster). It uses qr// for better performance on # repeated calls, so it only works under Perl 5.005 or later.

        $extract_head = gen_extract_tagged('<HEAD>','</HEAD>');

        ($extracted, $remainder) = $extract_head->($text);


The various extract_... subroutines may be used to extract a leading delimited string (possibily after skipping a specified prefix string).

In a list context, they all return a list, the first three elements of which are always: the extracted string (including the specified delimiters) , the remainder of the input string, the skipped prefix. Note that the original input text (the first argument) is not modified in any way.

In a scalar context, the extracted string is returned, having first been removed from the input text.

In a void context, the input text has the extracted string (and any specified prefix) removed from it.


The extract_delimited function formalizes the common idiom of extracting a single-character-delimited substring from the start of a string. For example, to extract a single-quote delimited string, the following code is typically used:

        ($remainder = $text) =~ s/\A('(\\'|[^'])*')//s;
        $extracted = $1;

but with extract_delimited it can be simplified to:

        ($extracted,$remainder) = extract_delimited $text, "'";

extract_delimited takes up to four scalars (the input text, the delimiters, a prefix pattern to be skipped, and any escape characters) and extracts the initial substring of the text that is appropriately delimited. If the delimiter string has multiple characters, the first one encountered in the text is taken to delimit the substring. The third argument specifies a prefix pattern that is to be skipped (but must be present!) before the substring is extracted. The final argument specifies the escape character to be used for each delimiter.

All arguments are optional. If the escape characters are not specified, every delimiter is escaped with a backslash (\). If the prefix is not specified, the pattern '\s*' - optional whitespace - is used. If the delimiter set is also not specified, the set /["'`]/ is used. If the text to be processed is not specified either, $_ is used.

In list context, extract_delimited returns a array of three elements, the extracted substring (including the surrounding delimiters), the remainder of the text, and the skipped prefix (if any). If a suitable delimited substring is not found, the first element of the array is the empty string, the second is the complete original text, and the prefix returned in the third element is an empty string.

In a scalar context, just the extracted substring is returned. In a void context, the extracted substring (and any prefix) are simply removed from the beginning of the first argument.


        # Remove a single-quoted substring from the very beginning of $text:

                $substring = extract_delimited($text, "'", '');

        # Remove a single-quoted Pascalish substring (i.e. one in which
        # doubling the quote character escapes it) from the very
        # beginning of $text:

                $substring = extract_delimited($text, "'", '', "'");

        # Extract a single- or double- quoted substring from the
        # beginning of $text, optionally after some whitespace
        # (note the list context to protect $text from modification):

                ($substring) = extract_delimited $text, q{"'};

        # Delete the substring delimited by the first '/' in $text:

                $text = join '', (extract_delimited($text,'/','[^/]*')[2,1];

Note that this last example is not the same as deleting the first quote-like pattern. For instance, if $text contained the string:

        "if ('./cmd' =~ m/$UNIXCMD/s) { $cmd = $1; }"

then after the deletion it would contain:

        "if ('.$UNIXCMD/s) { $cmd = $1; }"


        "if ('./cmd' =~ ms) { $cmd = $1; }"

See "extract_quotelike" for a (partial) solution to this problem.


Like "extract_delimited", the extract_bracketed function takes up to three optional scalar arguments: a string to extract from, a delimiter specifier, and a prefix pattern. As before, a missing prefix defaults to optional whitespace and a missing text defaults to $_. However, a missing delimiter specifier defaults to '{}()[]<>' (see below).

extract_bracketed extracts a balanced-bracket-delimited substring (using any one (or more) of the user-specified delimiter brackets: '(..)', '{..}', '[..]', or '<..>'). Optionally it will also respect quoted unbalanced brackets (see below).

A "delimiter bracket" is a bracket in list of delimiters passed as extract_bracketed's second argument. Delimiter brackets are specified by giving either the left or right (or both!) versions of the required bracket(s). Note that the order in which two or more delimiter brackets are specified is not significant.

A "balanced-bracket-delimited substring" is a substring bounded by matched brackets, such that any other (left or right) delimiter bracket within the substring is also matched by an opposite (right or left) delimiter bracket at the same level of nesting. Any type of bracket not in the delimiter list is treated as an ordinary character.

In other words, each type of bracket specified as a delimiter must be balanced and correctly nested within the substring, and any other kind of ("non-delimiter") bracket in the substring is ignored.

For example, given the string:

        $text = "{ an '[irregularly :-(] {} parenthesized >:-)' string }";

then a call to extract_bracketed in a list context:

        @result = extract_bracketed( $text, '{}' );

would return:

        ( "{ an '[irregularly :-(] {} parenthesized >:-)' string }" , "" , "" )

since both sets of '{..}' brackets are properly nested and evenly balanced. (In a scalar context just the first element of the array would be returned. In a void context, $text would be replaced by an empty string.)

Likewise the call in:

        @result = extract_bracketed( $text, '{[' );

would return the same result, since all sets of both types of specified delimiter brackets are correctly nested and balanced.

However, the call in:

        @result = extract_bracketed( $text, '{([<' );

would fail, returning:

        ( undef , "{ an '[irregularly :-(] {} parenthesized >:-)' string }"  );

because the embedded pairs of '(..)'s and '[..]'s are "cross-nested" and the embedded '>' is unbalanced. (In a scalar context, this call would return an empty string. In a void context, $text would be unchanged.)

Note that the embedded single-quotes in the string don't help in this case, since they have not been specified as acceptable delimiters and are therefore treated as non-delimiter characters (and ignored).

However, if a particular species of quote character is included in the delimiter specification, then that type of quote will be correctly handled. for example, if $text is:

        $text = '<A HREF=">>>>">link</A>';


        @result = extract_bracketed( $text, '<">' );


        ( '<A HREF=">>>>">', 'link</A>', "" )

as expected. Without the specification of " as an embedded quoter:

        @result = extract_bracketed( $text, '<>' );

the result would be:

        ( '<A HREF=">', '>>>">link</A>', "" )

In addition to the quote delimiters ', ", and `, full Perl quote-like quoting (i.e. q{string}, qq{string}, etc) can be specified by including the letter 'q' as a delimiter. Hence:

        @result = extract_bracketed( $text, '<q>' );

would correctly match something like this:

        $text = '<leftop: conj /and/ conj>';

See also: "extract_quotelike" and "extract_codeblock".


extract_tagged extracts and segments text between (balanced) specified tags.

The subroutine takes up to five optional arguments:

  1. A string to be processed ($_ if the string is omitted or undef)

  2. A string specifying a pattern to be matched as the opening tag. If the pattern string is omitted (or undef) then a pattern that matches any standard HTML/XML tag is used.

  3. A string specifying a pattern to be matched at the closing tag. If the pattern string is omitted (or undef) then the closing tag is constructed by inserting a / after any leading bracket characters in the actual opening tag that was matched (not the pattern that matched the tag). For example, if the opening tag pattern is specified as '{{\w+}}' and actually matched the opening tag "{{DATA}}", then the constructed closing tag would be "{{/DATA}}".

  4. A string specifying a pattern to be matched as a prefix (which is to be skipped). If omitted, optional whitespace is skipped.

  5. A hash reference containing various parsing options (see below)

The various options that can be specified are:

reject => $listref

The list reference contains one or more strings specifying patterns that must not appear within the tagged text.

For example, to extract an HTML link (which should not contain nested links) use:

        extract_tagged($text, '<A>', '</A>', undef, {reject => ['<A>']} );
ignore => $listref

The list reference contains one or more strings specifying patterns that are not be be treated as nested tags within the tagged text (even if they would match the start tag pattern).

For example, to extract an arbitrary XML tag, but ignore "empty" elements:

        extract_tagged($text, undef, undef, undef, {ignore => ['<[^>]*/>']} );

(also see "delimited_pat" below).

fail => $str

The fail option indicates the action to be taken if a matching end tag is not encountered (i.e. before the end of the string or some reject pattern matches). By default, a failure to match a closing tag causes extract_tagged to immediately fail.

However, if the string value associated with <reject> is "MAX", then extract_tagged returns the complete text up to the point of failure. If the string is "PARA", extract_tagged returns only the first paragraph after the tag (up to the first line that is either empty or contains only whitespace characters). If the string is "", the the default behaviour (i.e. failure) is reinstated.

For example, suppose the start tag "/para" introduces a paragraph, which then continues until the next "/endpara" tag or until another "/para" tag is encountered:

        $text = "/para line 1\n\nline 3\n/para line 4";

        extract_tagged($text, '/para', '/endpara', undef,
                                {reject => '/para', fail => MAX );

        # EXTRACTED: "/para line 1\n\nline 3\n"

Suppose instead, that if no matching "/endpara" tag is found, the "/para" tag refers only to the immediately following paragraph:

        $text = "/para line 1\n\nline 3\n/para line 4";

        extract_tagged($text, '/para', '/endpara', undef,
                        {reject => '/para', fail => MAX );

        # EXTRACTED: "/para line 1\n"

Note that the specified fail behaviour applies to nested tags as well.

On success in a list context, an array of 6 elements is returned. The elements are:


the extracted tagged substring (including the outermost tags),


the remainder of the input text,


the prefix substring (if any),


the opening tag


the text between the opening and closing tags


the closing tag (or "" if no closing tag was found)

On failure, all of these values (except the remaining text) are undef.

In a scalar context, extract_tagged returns just the complete substring that matched a tagged text (including the start and end tags). undef is returned on failure. In addition, the original input text has the returned substring (and any prefix) removed from it.

In a void context, the input text just has the matched substring (and any specified prefix) removed.


(Note: This subroutine is only available under Perl5.005)

gen_extract_tagged generates a new anonymous subroutine which extracts and segments text between (balanced) specified tags. In other words, it generates a function identical in function to extract_tagged.

The difference between extract_tagged and the subroutines generated by gen_extract_tagged, is that those anonymous subroutines:

  • are limited to the actual tags specified to gen_extract_tagged (whereas extract_tagged can automatically select tags as they are encountered during extraction);

  • do not have to reparse tag specification or parsing options every time they are called (whereas extract_tagged has to effectively rebuild its tag parser on every call);

  • make use of the new qr// construct to pre-compile the regexes they use (whereas extract_tagged uses standard string variable interpolation to create tag-matching patterns).

The subroutine takes up to four optional arguments (the same set as extract_tagged except for the string to be processed). It returns a reference to a subroutine which in turn takes a single argument (the text to be extracted from).

In other words, the implementation of extract_tagged is exactly equivalent to:

        sub extract_tagged
                my $text = shift;
                $extractor = gen_extract_tagged(@_);
                return $extractor->($text);

(although extract_tagged is not currently implemented that way, in order to preserve pre-5.005 compatibility).

Using gen_extract_tagged to create extraction functions for specific tags is a good idea if those functions are going to be called more than once, since their performance is typically three times better than the more general-purpose extract_tagged.


extract_quotelike attempts to recognize, extract, and segment any one of the various Perl quotes and quotelike operators (see perlop(3)) Nested backslashed delimiters, embedded balanced bracket delimiters (for the quotelike operators), and trailing modifiers are all caught. For example, in:

        extract_quotelike 'q # an octothorpe: \# (not the end of the q!) #'
        extract_quotelike '  "You said, \"Use sed\"."  '

        extract_quotelike ' s{([A-Z]{1,8}\.[A-Z]{3})} /\L$1\E/; '

        extract_quotelike ' tr/\\\/\\\\/\\\//ds; '

the full Perl quotelike operations are all extracted correctly.

Note too that, when using the /x modifier on a regex, any comment containing the current pattern delimiter will cause the regex to be immediately terminated. In other words:

        'm /
                (?i)            # CASE INSENSITIVE
                [a-z_]          # LEADING ALPHABETIC/UNDERSCORE
                [a-z0-9]*       # FOLLOWED BY ANY NUMBER OF ALPHANUMERICS

will be extracted as if it were:

        'm /
                (?i)            # CASE INSENSITIVE
                [a-z_]          # LEADING ALPHABETIC/'

This behaviour is identical to that of the Perl 5.004 interpreter.

extract_quotelike takes two arguments: the text to be processed and a prefix to be matched at the very beginning of the text. If no prefix is specified, optional whitespace is the default. If no text is given, $_ is used.

In a list context, an array of 11 elements is returned. The elements are:


the extracted quotelike substring (including trailing modifiers),


the remainder of the input text,


the prefix substring (if any),


the name of the quotelike operator (if any),


the left delimiter of the first block of the operation,


the text of the first block of the operation (that is, the contents of a quote, the regex of a match or substitution or the target list of a translation),


the right delimiter of the first block of the operation,


the left delimiter of the second block of the operation (that is, if it is a s, tr, or y),


the text of the second block of the operation (that is, the replacement of a substitution or the translation list of a translation),


the right delimiter of the second block of the operation (if any),


the trailing modifiers on the operation (if any).

For each of the fields marked "(if any)" the default value on success is an empty string. On failure, all of these values (except the remaining text) are undef.

In a scalar context, extract_quotelike returns just the complete substring that matched a quotelike operation (or undef on failure). In a scalar or void context, the input text has the same substring (and any specified prefix) removed.


        # Remove the first quotelike literal that appears in text

                $quotelike = extract_quotelike($text,'.*?');

        # Replace one or more leading whitespace-separated quotelike
        # literals in $_ with "<QLL>"

                do { $_ = join '<QLL>', (extract_quotelike)[2,1] } until $@;

        # Isolate the search pattern in a quotelike operation from $text

                ($op,$pat) = (extract_quotelike $text)[3,5];
                if ($op =~ /[ms]/)
                        print "search pattern: $pat\n";
                        print "$op is not a pattern matching operation\n";


extract_codeblock attempts to recognize and extract a balanced bracket delimited substring that may contain unbalanced brackets inside Perl quotes or quotelike operations. That is, extract_codeblock is like a combination of "extract_bracketed" and "extract_quotelike".

extract_codeblock takes the same three parameters as extract_bracketed: a text to process, a set of delimiter brackets to look for, and a prefix to match first. No prefix argument implies optional whitespace at the start, no delimiter brackets indicates that only '{' is to be used. No input no text means process $_.

Once the prefix has been disposed of, code blocks are extracted by stepping through the text and trying the following alternatives in sequence:

  1. Try and match a closing delimiter bracket. If the bracket was the same species as the last opening bracket, return the substring to that point. If the bracket was mismatched, return an error.

  2. Try to match a quote or quotelike operator. If found, call extract_quotelike to eat it. If extract_quotelike fails, return the error it returned. Otherwise go back to step 1.

  3. Try to match an opening delimiter bracket. If found, call extract_codeblock recursively to eat the embedded block. If the recursive call fails, return an error. Otherwise, go back to step 1.

  4. Unconditionally match a bareword or any other single character, and then go back to step 1.


        # Find a while loop in the text

                if ($text =~ s/.*?while\s*\{/{/)
                        $loop = "while " . extract_codeblock($text);

        # Remove the first round-bracketed list (which may include
        # round- or curly-bracketed code blocks or quotelike operators)

                extract_codeblock $text, "(){}", '[^(]*';


The extract_multiple subroutine takes a string to be processed and a list of extraction subroutines to apply to that string. In an array context extract_multiple returns an array of substrings of the original string, as extracted by the various extraction subroutines. In a scalar context, extract_multiple returns the first substring successfully extracted from the original string. In both scalar and void contexts the original string has the first successfully extracted substring removed from it.

Hence, the aim of of a call to extract_multiple in a list context is to split the processed string into as many non-overlapping fields as possible, by repeatedly applying each of the specified extractor subroutines to the remainder of the string. Thus extract_multiple is a generalized form of Perl's inbuilt split subroutine.

In a scalar and void context, a call to extract_multiple removes the leading substring which would have been removed by any one of the specified extraction subroutines.

The subroutine takes up to four optional arguments:

  1. A string to be processed ($_ if the string is omitted or undef)

  2. A reference to a list of subroutine references, specifying the extractor subroutines to be used to split the string. If this argument is omitted (or undef) the list:

                    sub { extract_variable($_[0], '') },
                    sub { extract_quotelike($_[0],'') },
                    sub { extract_codeblock($_[0],'{}','') },

    is used.

  3. An number specifying the maximum number of fields to return. If this argument is omitted (or undef), split continues as long as possible.

    If the third argument is N, then at most N-1 splits will be performed and the array return value of extract_multiple will contain at most N elements (where the very last element is always the unsplit remainder of the original string)

    Note that in scalar and void contexts the value of this argument is automatically set to 2, unless the fourth argument (see below) is set to true, in which case the maximum is set to 1..

  4. A value indicating whether unmatched substrings (see below) within the text should be skipped or returned as fields. If the value is true, such substrings are skipped. Otherwise, they are returned.

The extraction process works by applying each extractor subroutine in sequence to the text string. Each subroutine is called in a list context and is expected to return two elements: the extract substring and the remainder of the original string. If the subroutine returns an undefined value, it is assumed to have failed to extract.

If an extractor subroutine returns a defined value, that value is immediately treated as the next extracted field and pushed onto the list of fields. The text being processed is then assigned the remainder returned by the successful extractor function. If none of the extractor subroutines succeeds, then one character is extracted from the start of the text and the extraction subroutines reapplied. Characters which are thus removed are accumulated and eventually become the next field (unless the fourth argument is true, in which case they are disgarded).

For example, the following extracts substrings that are valid Perl variables:

        @fields = extract_multiple($text,
                                   [ sub { extract_variable($_[0]) } ],
                                   undef, 1);

This example separates a text into fields which are quote delimited, curly bracketed, and anything else:

        @fields = extract_multiple($text,
                                        sub { extract_delimited($_[0],q{'"}) },
                                        sub { extract_bracketed($_[0],'{}') },

This call extracts the next single substring that is a valid Perl quotelike operator (discarding everything before it):

        $quotelike = extract_multiple($text,
                                        sub { extract_quotelike($_[0]) },

This does the same, but also captures the discarded preliminary substring:

        ($prelim,$quotelike) = extract_multiple($text,
                                        sub { extract_quotelike($_[0]) },

Putting the maximum count in that last example is important because otherwise extract_multiple would process the entire string (possibly extracting many fields) and then discard all but the first two.

Finally, here is yet another way to do comma-separated value parsing:

        @fields = extract_multiple($csv_text,
                                        sub { extract_delimited($_[0],q{'"}) },
                                        sub { $_[0] =~ /\A([^,]+)(.*)/ },

The list in the second argument means: "Try and extract a ' or " delimited string, otherwise extract anything up to a comma...". The undef third argument means: "...as many times as possible...", and the true value in the fourth argument means "...discarding anything else that appears (i.e. the commas)".

If you wanted the commas preserved as separate fields (i.e. like split does if your split pattern has capturing parentheses), you would just make the last parameter undefined (or remove it).


The delimited_pat subroutine takes a single (string) argument and builds a Friedl-style optimized regex that matches a string delimited by any one of the characters in the single argument. For example:


returns the regex:


Note that the specified delimiters are automatically quotemeta'd.

A typical use of delimited_pat would be to build special purpose tags for extract_tagged. For example, to properly ignore "empty" XML elements (which might contain quoted strings):

        my $empty_tag = '<(' . delimited_pat(q{'"}) . '|.)+/>';

        extract_tagged($text, undef, undef, undef, {ignore => [$empty_tag]} );

delimited_pat may also be called with an optional second argument, which specifies the "escape" character(s) to be used for each delimiter. For example to match a Pascal-style string (where ' is the delimiter and '' is a literal ' within the string):


Different escape characters can be specified for different delimiters. For example, to specify that '/' is the escape for single quotes and '%' is the escape for double quotes:


If more delimiters than escape chars are specified, the last escape char is used for the remaining delimiters. If no escape char is specified for a given specified delimiter, '\' is used.


In a list context, all the functions return (undef,$original_text) on failure. In a scalar context, failure is indicated by returning undef (in this case the input text is not modified in any way).

In addition, on failure in any context, one of the following explanatory diagnostic messages is returned in the standard $@ variable (on success the $@ variable is guaranteed to be undef):

Did not find a suitable bracket: "%s"

The delimiter provided to extract_bracketed was not one of '()[]<>{}'.

Did not find prefix: /%s/

A non-optional prefix was specified but wasn't found at the start of the text.

Could not extract "%c"-delimited substring

extract_delimited, extract_quotelike or extract_codeblock couldn't find an initial substring (after the prefix) that was delimited by the delimiter(s) specified.

Did not find opening bracket after prefix: "%s"

extract_bracketed or extract_codeblock was expecting a particular kind of bracket at the start of the text, and didn't find it.

No quotelike operator found after prefix: "%s"

extract_quotelike didn't find one of the quotelike operators q, qq, qw, qx, s, tr or y at the start of the substring it was extracting.

Unmatched closing bracket: "%c"

extract_bracketed, extract_quotelike or extract_codeblock encountered a closing bracket where none was expected.

Unmatched opening bracket(s): "%s"

extract_bracketed, extract_quotelike or extract_codeblock ran out of characters in the text before closing one or more levels of nested brackets.

Mismatched closing bracket: expected "%c" but found "%s"

extract_bracketed, extract_quotelike or extract_codeblock found a valid bracket delimiter, but it was the wrong species. This usually indicates a nesting error, but may indicate incorrect quoting or escaping.

No block delimiter found after quotelike "%s"

extract_quotelike or extract_codeblock found one of the quotelike operators q, qq, qw, qx, s, tr or y without a suitable block after it.

Missing second block for quotelike "%s"

extract_codeblock or extract_quotelike found one of the quotelike operators s, tr or y followed by only one block.

Nested codeblock failed to balance from "%s..."

A block within the substring being extracted by extract_codeblock was incorrectly nested or had a unmatched delimiter in it.

Did not find opening tag: /%s/

extract_tagged did not find a suitable opening tag (after any specified prefix was removed).

Unable to construct closing tag to match: /%s/

extract_tagged matched the specified opening tag and tried to modify the matched text to produce a matching closing tag (because none was specified). It failed to generate the closing tag, almost certainly because the opening tag did not start with a bracket of some kind.

Found invalid nested tag: %s

extract_tagged found a nested tag that appeared in the "reject" list (and the failure mode was not "MAX" or "PARA").

Found unbalanced nested tag: %s

extract_tagged found a nested opening tag that was not matched by a corresponding nested closing tag (and the failure mode was not "MAX" or "PARA").

Did not find closing tag

extract_tagged reached the end of the text without finding a closing tag to match the original opening tag (and the failure mode was not "MAX" or "PARA").


Damian Conway (damian@cs.monash.edu.au)


There are undoubtedly serious bugs lurking somewhere in this code, if only because parts of it give the impression of understanding a great deal more about Perl than they really do.

Bug reports and other feedback are most welcome.


 Copyright (c) 1997-1998, Damian Conway. All Rights Reserved.
 This module is free software. It may be used, redistributed
and/or modified under the terms of the Perl Artistic License
     (see http://www.perl.com/perl/misc/Artistic.html)