View on
Greg London > Parse-Gnaw-0.52 > Parse::Gnaw



Annotate this POD

View/Report Bugs
Module Version: 0.52   Source   Latest Release: Parse-Gnaw-0.601


Parse::Gnaw - An extensible parser. Define grammars using subroutine calls. Define your own grammar extensions by defining new subroutines. Parse text in memory or from/to files or other streams.


Gnaw is a perl module which implements full regular expressions and full text parsing grammars using nothing but pure perl code limited to subroutine closures and basic perl variables such as scalars, hashes, arrays, and references.

You write your grammar in pure perl. There is no intermediate "parser language" that then gets interpreted into something executable.

When you do a "use Parse::Gnaw", the Gnaw module will import a number of functions directly into your namespace. Yes, this is completely bad form for normal modules. But this is not a normal module. The imported subroutines include regular expression and parsing functions for matching, quantifiers, literals, alternations, character classes, and so on. You build up your grammar by calling these functions. The final call will return a code reference. This code reference is your grammar.

When you dereference that grammar, if it is a "match" function, then you pass in the string you want to parse.

        use Parse::Gnaw;

        # create the grammar
        my $grammar = match('hello');

        # apply the grammar to a string
        if($grammar->('hello world')) {
                print "match\n";
        } else {
                print "no match";

You can also create the grammar and execute it in one step:

        my $texttoparse = "howdy partner";

        if(match('hello', 'world')->($texttoparse)) {
                print "match\n";
        } else {
                print "no match\n";

Note the above example translated into perls regular expression syntax would look something like this:

        my $texttoparse = "howdy partner";

        if($texttoparse =~ m{hello\s*world}) {
                print "match\n";
        } else {
                print "no match\n";

You can build up more complicated grammars fairly easily. This one looks for a sentence about fruits.

        $grammar = match(ql('I would like to buy'), some('a', qa('banana apple pear peach')));

The "match" function looks for a match to the grammar in the string being parsed.

The "ql" function (quoted literal) allows you to put a sequence of literals into a single string. It the splits the string up into individual literals much like perls "qw" function does. Then it puts them into a grammar sequence for you. This saves you from putting quotes around each individual literal.

The "some" function is a quantifier looking for "1 or more" of whatever it surrounds, in this case, a sequence of the literal "a" followed by an alternation of different possible fruit.

The "qa" function (quoted alternation) takes a single string and splits it into individual words, also similar to perls "qw" function. The "qa" function then takes those individual words and creates an alternation that tries to match any individual word in the string as a valid alternate.


The grammars are defined by calling subroutines imported into your namespace from Parse::Gnaw. You can define your own grammar components by declaring a subroutine that performs whatever preprocessing you want and generates whatever set of lower level components you want. Examples of this are provided later on in this pod file.


Please note that this is a beta release. This is more of a proof of concept than something ready for production code or for massive grammars. The interfaces may change completely in the future. When the interfaces have settled, I will release this as a version 1.0+ module. Until then, please do not use this to develop some gigantic parser when the grammar may have to completely change.


Currently, EVERYTHING is exported into the callers namespace. Yes, I know, bad programmer. No biscuit.

The problem is I did not want grammars to be weighed down with a bunch of package prefixes at every step.

        my $grammar = Parse::Gnaw::match(Parse::Gnaw::a(Parse::Gnaw::l('hello'),Parse::Gnaw::l('howdy'));

Gets rather tedious.

Hopefully, no one will have collisions with subroutine names. If they do, then it might make sense to create a new package, declare the grammar in there, then simply pull out the code reference and use that in the rest of your code.

        package mynormalpackage;

        our $grammar;

                package mylocalpackage;

                use Parse::Gnaw; # all subs imported into mylocalpackage

                $mynormalpackage::grammar = match('hello', 'world');



Again, everything may change tomorrow as I sort this out.

ok function ^

In the examples provided, the "ok" function is used repeatedly. This allows the examples to be used directly in the Parse::Gnaw tests directly. The "ok" function is defined there by the Test::More package. Here is a stub for the "ok" function so that you can use the examples independent of Test::More.

        use Parse::Gnaw;

        sub ok {
                my $boolean = shift(@_);
                if($boolean) {
                        print "pass";
                } else {
                        print "FAIL";
                print " : ";
                my $msg = shift(@_);
                if(defined($msg)) {
                        print $msg;
                print "\n";

        my $grammar; # declare it here so other examples can just use it.

The above code is assumed to be contained in all the examples that follow.


So, again, when you do a

        use Parse::Gnaw;

in your code, a number of subroutines are imported into your namespace. You can then call these subroutines to build up a grammar. The grammar is held in the form of a code reference. You then dereference the coderef to call the parser and provide it with the string you want to parse. The dereferenced code reference that is the parser will return a boolean to indicate whether a match was found or not.

        $grammar = match('hello', 'world');
        ok($grammar->('hello world')==1), "match");
        ok($grammar->('howdy world')==0), "NO match");

        > pass : match
        > pass : NO match

If you want your grammar to match a literal string, simply pass that string into the "parse" or "match" function. Strings passed into the Gnaw functions are interpreted to mean that they are literals to look for in the text being parsed.

Top Level and Low Level Functions

The Gnaw module provides a number of "top level" functions to create your grammars. These top level functions return code references which are your grammar. You then dereference these code references to call the grammar and pass in the string you want to parse.

Only top level functions return grammar code references. They include "match", "swap", "parse", and "modify". These functions return code references that are the grammar, code references you dereference to run the grammar on a string.

Low level functions return code references that are called "stitchers". These are low level subroutines used by Parse::Gnaw to build up the grammar. Most of the low level and high level Gnaw functions know how to handle stitchers. A "stitcher" is not a grammar, and if you try to dereference a coderef from a low level function, your script will die with strange errors.

        my $stitcher = a('hello', 'howdy');
        # WRONG! Don't do it!
        $stitcher->("these are not the droids you are looking for");  

Always make sure you use a top level function as your final function of your grammar.



The "match" function is similar to perls =~ m/// regexp function.

Since "m" is a reserved perl function, we had to use a different name.

The "match" function will find the first match of the grammar in the string. It will start at the first character in the string, attempt to find a match from there, and if it fails to find a match, it will move forward one character and attempt to find a match there. If no match is found, it will move forward one character and try the pattern, repeating until it reaches the end of the string.

match examples:

        my $grammar = match('hello', 'world');
        ok($grammar->("these are not the droids you are looking for")==0, 'no match');
        ok($grammar->("why did you say hello world just now?")==1, 'match found');


The "swap" function is similar to perls =~ s/// regexp function.

Since "s" is a reserved perl function, we had to use a different name.

The "swap" function allows you to define a pattern to look for in the string and allows you to define some replacement text if a match is found. The replacement text can be defined as a literal string or as a callback.

The last parameter in the "swap" function is either a string or a subroutine reference. This last parameter will be used as the replacement value. All other parameters are low level functions to define the pattern.

If you want to swap "Alice" for "Bob", then just do this:

        swap('Alice', 'Bob')

A working example might look like this:

        my $input = "hello Alice ! How are you?";

        # find 'Alice' and replace with 'Bob', use string literal
        ok(swap('Alice', 'Bob') -> ($input)==1, "checking match was found");

        ok($input eq 'hello Bob ! How are you?', "checking substitution/swap took place");

If the last parameter to the "swap" funciton is a subroutine reference, then if a match is found, then the parser will call the subroutine, pass the matching value into the subroutine as the first parameter, and whatever that subroutine returns is the replacement text the parser will use.

You could return a hardcoded value:

        swap('Alice', sub{'Bob'})

You could take the matching value, and modify it:

        swap('Alice', sub{uc(shift(@_))})

A working example might look like this:

        my $input = 'Hello Alice';
        ok(swap('Alice', sub{uc(shift(@_))})->($input)==1, "match found");
        ok($input eq 'Hello ALICE', "text swap occurred");

Here's another swap example:

        # example 2
        my $input = "please call 555-1212 for more details";

        # find a phone number and replace with phone number surrounded by asterisks. Use subroutine.
        swap(some(dgt),'-',some(dgt), sub{return "***".shift(@_)."***";}) -> ($input);
        ok($input eq 'please call ***555-1212*** for more details');


The "modify" function operates similar to the "match" function in that it starts at the beginning of the text, tries to find a pattern match, and moves forward through the text until it finds a match.

The difference between "modify" and "match" is that "modify" allows the "get" function to modify the text in flight. If the "get" function returns anything other than undef, the original text will be modified to whatever the "get" function return value is.

Think of "modify" as "match" but with the "write protect" bit turned off. And think of "get" as the way to "write" new text.


The "parse" function is similar to perls =~ m/\A ... / regular expression function. The "parse" function only attempts to match the pattern starting at the beginning of the text. It does not move forward through the string to find a match.

        $grammar = parse('Alice');

        ok($grammar->('Alice Bob Charlie')==1, "match found");
        ok($grammar->('Bob Alice Charlie')==0, "match not found");

If you're defining a grammar to parse a file or some formal language, you probably want to use "parse" instead of "match".


The "az" function requires the text being parsed match the pattern from beginning to end.

        az( blah )   ==>  =~ m/\A blah \Z/



Most of the Parse::Gnaw subroutines will recognize strings passed in as parameters as string literals to become part of the grammar. We already saw this in the above example:

        $grammar = match('hello', 'world');

This example will create a grammar that looks for the literal 'hello' followed by the literal 'world'.

Actually, the grammar will look for the literal 'hello', it will then Skip over any text that the Skip function tells it to Skip (by default, it will skip over any whitespace), and then look for the literal 'world'.

See the "Skip" function for more information.

Strings passed into Parse::Gnaw functions are perl strings, meaning that they follow the normal perl rules for string evaluation. i.e. Single quotes do NOT get evaluated. Double quotes do.

If you want to match a tab character, then be sure to use "\t" instead of '\t'.

List Context

Before we go too far into talking about Gnaw, we need to address a meta issue around Parse::Gnaw.

The thing to keep in mind is that all Parse::Gnaw subroutines are...

... perl subroutines.

This means that they are susceptible to list context issues that occur in perl when you call the functions, and that they are susceptible to list context issues and return mechanism issues when you declare your own subroutines to wrap up a subrule of your grammar.

When you call any perl subroutine, the parameters passed in are flattend into one gigantic list by the list context mechanism in perl.

        # all parameters are flattened. subroutine receives 6 parameters.
        my_perl_sub( (1,2,3),('a', 'b', 'c') ); 

When the my_perl_sub subroutine is called, perl takes all the parameters and flattens them into a single list context. This means in the above example that my_perl_sub receives 6 parameters:

        1, 2, 3, 'a', 'b', 'c'

This also means that my_perl_sub has no idea that the numbers had been grouped together in one set of parens and the letters had been grouped together in another set of parens. The my_perl_sub subroutine sees nothing but a flat list of parameters being passed in.

If you want to call a subroutine and have it treat parameters in groups, then the most common way to do that is to group the parameters you want grouped by putting them into arrays and passing references to those arrays.

        # my_perl_sub receives two parameters, two array references.
        my_perl_sub( [1,2,3] , ['a', 'b', 'c'] ); 

This exact same list context mechanism affects Parse::Gnaw. And the solution within Parse::Gnaw is the same as for general perl: wrap groups of items in anonymous arrays.

Example. You want a grammar with two possible alternates. The first alternate is "hello" followed by "world". The second alternate is "howdy" followed by "planet". If you try to put them into the alternate function (the "a" function), you might try this:

        $grammar = match( a( 
                        "hello", "world",
                        "howdy", "planet"

The problem is that the "a" function has no way of knowing how it is supposed to group the parameters. Instead, the above example will look for four possible alternates:


The solution is to group the parameters with an anonymous array reference. The above example can be rewritten as follows:

        $grammar = match( a( ["hello", "world"],["howdy", "planet"]     ));

This will look for two possible alternates:

        "hello" "world"
        "howdy" "planet"

Most of the Parse::Gnaw functions (such as the "a" (alternate) function) are smart enough to recognize array references and treat them as a series of grammar components meant to be addressed as a single unit.

declare subroutines for your own sub-rules

The other issue around perl list context and Parse::Gnaw occurs around perl and its "return" function in subroutines. In Parse::Gnaw, if you have a complex grammar, you can break it up into rules which take the form of perl subroutines you declare. The subroutine is some subset of the whole grammar, and must return something that the rest of the grammar can use.

This means you must be aware of perls "return" functionality when declaring subroutines as your rules.

For example, you can declare the following subroutine in any perl code:

        sub mysub { 'hello', 'there' };

When you call a perl subroutine, it will return whatever was the last expression evaluated. And in the above example, "there" is the last expression evaluated, and "there" is what mysub will return.

        my $retval = tester;
        print "retval is '$retval'\n";

        > retval is 'there'

Again this is an issue relating to perls list context. If you want a subroutine to return a list of items, then enclose them in parenthesis to force list context on the return value.

        sub fruit { ('apple', 'banana', 'pear') } # parens force list context

You could then call this subroutine and pass it to the Parse::Gnaw "a" function as a list of possible alternates.

        $grammar = match(a( fruit ));

However, if you want to declare a subroutine which is intended to contain a sequence of items, then you will want to wrap them in an anonymous array so that a single array reference is returned and is treated as a sequence by the rest of the grammar.

        # square brackets groups return value
        sub greetworld { ['hello', 'world'] }   
        sub greetplanet{ ['howdy', 'planet'] }

        $grammar = match(a( greetworld, greetplanet ));

Note that the two subroutines define a sequence of literals that get passed into the "a" function. To keep the "a" function from flattening the return value of the two subroutines, the subroutines need to group their sequences into array references.

Always keep in mind that Parse::Gnaw functions are perl functions and are susceptible to list context issues relating to the things being passed into a function as well as what is returned by the function.

As a general rule, when declaring subroutines to describe pieces of your grammar, if that subroutine contains more than one grammar component, you should wrap the return value in parenthesis if you intend to use the components in list context (a list of different alternates to pass in to "a"), or you should wrap the return value in square brackets if you intend to use the components as one sequence of components.

If the subroutine contains ONLY one Parse::Gnaw grammar component, then you do not have to wrap that component in parens or square brackets. Parse::Gnaw low level functions always return one scalar, a stitcher code reference that can be used by other Parse::Gnaw functions.

Since Parse::Gnaw grammar functions always return one scalar, you do not have to worry about forcing scalar context with square brackets. Just call the Gnaw function and perl will return the stitcher for you.

        # return value of 'a' function is a scalar, 
        # (a scalar containing a code ref that is a stitcher function)
        # therefore, subroutine "fruit" doesn't have to put 
        # the "a" function in square brackets to force scalar context on
        # return value.
        sub fruit { a('apple', 'banana', 'pear') }

summary of return values in subroutines/subrules

Always using parens or square brackets in your subroutines will show how you intend the subroutine to be used in a larger grammar.

        # parens forces return in list context. items may be used in parallel
        sub fruit { ('apple', 'pear', 'banana') } 

        # square brackets forces return in scalar context. items are seqential.
        sub greeting { ['may', 'the', 'force', 'be', 'with', 'you'] }

Do not use a wrapper (no parens, no square brackets) if subroutine only calls a single Gnaw grammar component.

        # 'name' subroutine only calls one function 'a'. don't have to wrap.
        sub name { a('Alice', 'Bob', 'Charlie') }

Now back to the grammar functions.

cc (character classes)

The Parse::Gnaw "cc" function is used to declare a character class. The input to the "cc" function is a single string which lists all the characters that are to be part of the character class. The class will match any single character that is listed in the string.

        $grammar = match( cc('0123456789') ); # any single digit. 0,1,2,3,4,5,6,7,8,9

The "cc" function recognizes the '-' character as a meta character which operates the same as the '-' in perl character classes, creating a character class with all the characters between the two characters (inclusive) separated by the '-' character.

        $grammar = match( cc('0-9') ); # any single digit. 0,1,2,3,4,5,6,7,8,9

If you want the '-' character to be part of your character class, make it the first character in the string.

        $grammar = match( cc('-aeiou') ); # a hyphen or any vowel

Character class shortcuts are provided for common character classes.

        dgt => 0-9
        wrd => a-zA-Z0-9_
        spc => \n \t \r \f or literal space

        $grammar = match( dgt ); # any single digit. 0,1,2,3,4,5,6,7,8,9

Note that the "cc" function and its related shortcuts are one of the functions in Parse::Gnaw that do NOT handle array references, nor do they handle multiple strings being passed into them in list context. The only valid parameter to pass to the "cc" function is a single string containing all the characters that you want in the character class.

The cc shortcuts take no parameters at all.

CC (inverted character class)

The Parse::Gnaw "CC" function is an inverted character class. The input is a single string which defines the characters that you do NOT want to be part of the character class. Note that the "CC" function is uppercase and the "cc" function is lower case.

        $grammar = match( CC('0-9') ); # any single character EXCEPT a digit

Short cuts are provided for common inverted character classes:

        DGT => anything BUT 0-9
        WRD => anything BUT a-zA-Z0-9_
        SPC => anything BUT \n \t \r \f or literal space

        $grammar = match( DGT ); # any single character EXCEPT a digit

Note that the "CC" function has the same parameter restrictions as "cc". Only one string containing the characters defining the NOT character class. No more than a single parameter. No array references. The shortcuts take no parameters at all.


The Parse::Gnaw "thing" function is a special character class that matches any single character. It is equivalent to the perl regular expression '.' character. But perl does not allow '.' to be the name of a subroutine, so I called it "thing". Part of the reason for this name was because of how it works well with quantifier shortcuts of "any" and "some", described below.

The "thing" function takes no parameters.

        $grammar=match('b', thing, 'b');

        ok($grammar->('bob')==1, "bob matches");
        ok($grammar->('bb')==0, "bb does not match");

a (alternation)

The "a" function is an alternation function. It interprets each of its parameters to be one of several possible alternates.

The following will look for one of any possible fruit in your text.

        $grammar=match(a('apple', 'banana', 'pear', 'peach'));

        ok($grammar->("she is the apple of my eye")==1, "found apple");

qa (quoted alternation)

Because listing a bunch of alternate literals is fairly common, the "qa" function allows you to do that with fewer quotation marks. The "qa" function takes a single string, breaks it up into individual literals based on whitespace, and then treats each literal as an alternate in the grammar.

The above example could be shortened to this:

        $grammar = match(qa('apple banana pear peach'));

        ok($grammar->("she is the apple of my eye")==1, "found apple");

Grouping of a sequence of components into one alternate is done by putting the sequence in an array reference, see "list context" issues described above for a more detailed explanation of why this is.

The following example looks for two possible alternates. The first alternate is the literal 'hello' followed by the literal 'world'. The second alternate is the literal 'howdy' followed by 'partner'.

        $grammar = match(a( ['hello', 'world'], ['howdy', 'partner'] ));
        ok($grammar->("why howdy partner !")==1, "found alternate");

The alternates can be any sequence of grammar components wrapped in an array reference.

greedy and thrifty quantifiers

The "greedy" and "thrifty" functions are quantifier functions. They allow you to define a grammar that looks for some pattern that repeats for some number of times.

The format for a quantifier call looks like this:

        greedy ( [min, max(?)], grammar component(, component)* )

The first parameter to the "greedy" or "thrifty" function is an array reference. This contains the parameters that control how the quantifier works. The "min" value determines the minimum number of times the pattern must match before considered a pass. The "max" value is optional. If max is not defined, it is assumed to be infinite, and the parser will attempt to fit as many patterns into memory as can fit.

The remaining one or more parameters define the pattern to which the quantifier will be applied. If more than one are supplied, they will be applied as one large sequence.

The difference between "greedy" and "thrifty" is how quickly the quantifier will try to match. The "greedy" function is greedy and will try to match as many patterns as possible. If it fails to match, the greedy quantifier will start backing off pattern counts and try to let the rest of the grammar find a match.

The "thrifty" function is thrifty and will try to match as few patterns as possible. If the rest of the grammar fails to match, the thrifty pattern will try to match another pattern, before trying the rest of the grammar.

        # seventeen or more 'g' or 'a' or 't' or 'c' characters.
        $grammar = match( greedy([17], cc('gatc')) ); 

        # one or more "word" characters
        $grammar = match( greedy([1], wrd) );

        # zero or more "digit" characters
        $grammar = match( greedy([0], dgt) );

        # at least 3 and up to 8 letters
        $grammar = match( greedy([3,8], cc('a-zA-Z')));

Note that the "greedy" and "thrifty" quantifiers are functions that can accept one or more parameters passed into them. These parameters can be other grammar components, string literals, or array references for grouped bits of grammar.

        # one or more 'hello' followed by 'world'
        $grammar = match( greedy([1], 'hello', 'world') )

Shorcuts are provided for thrifty([0],...) and thrifty([1], ...), called "any" and "some" respectively

any function

some function

Because thrifty quantifiers of "zero or more" and "one or more" are fairly common, shortcuts 'any' and 'some' have been provided.

        $grammar = match( any(dgt) ); # zero or more (thrifty) digits

        $grammar = match( some(wrd) ); # one or more (thrifty) word characters.

The "any" and "some" functions take any number of input parameters, and the parameters can be string literals, array references, and other grammar components.

        # one or more of "hello" followed by "world"
        $grammar = match( some('hello', 'world') );

If you have trouble reading that just insert "some number of" for "some". And insert "any number of" for "any".

anything function

something function

The character set "thing" matches any single character. The "any" and "some" functions are quantifiers of zero or more and one or more. These two functions are commonly combined, therefore a shortcut has been provided:

        anything  => g([0], thing)
        something => g([1], thing)

You can use these in your grammars to make them slightly more readable.

        # literal 'hello' followed by 1 or more characters, followed by
        # literal 'how', literal 'are', and literal 'you'
        $grammar = match( 'hello', something, 'how', 'are', 'you' );

maybe function

The maybe function is a shortcut that wraps the "greedy" function with the quantity hardcoded to [0,1]. This means whatever you put into the "maybe" function is optional. It may be there or it may not and the parse will still pass.

        $grammar = match('why', maybe('hello','there'), 'Alice');
        ok($grammar->("why hello there Alice") == 1, 'match');
        ok($grammar->("why Alice") == 1, 'match');
        ok($grammar->("why hello Alice") == 0, 'no match');
        ok($grammar->("why there Alice") == 0, 'no match');

slist (separated list)

Say you have a list of separated items, such as:

        red, orange, yellow, green, blue, indigo, violet

You might first try coding it up as:

        sub color {qa("red orange yellow green blue indigo violet")}
        sub comma {','}
        $grammar=match(some(color, comma));

The problem is this will require that the last color be followed by a comma:

        red, green, blue,

Many formally defined languages have comma separated lists, but do not allow a comma to follow the last item.

The "slist" function allows you to define an item and a separator, and the resulting parser will look for items separated by separators, but will not look for a trailing separator after the last item. The general format looks like this:

        slist( item, separator )

The "slist" function then returns a grammar component that will look for any of the following patterns.

        item separator item
        item separator item separator item
        item (separator item)*

If item is more than one grammar component, enclose it in square brackets to group it into an array reference. If the separator is more than one component, enclose it in square brackets to group it into an array reference.

This example defines the item to be the word "color" followed by a color word. The separator is still a comma.

        slist( [ 'color' qa('red green blue') ], ',' )

The above example will match this text:

        color red, color green, color blue

If you allow a trailing comma, set the third parameter to a "1".

        slist( [ 'color' qa('red green blue') ], ',', 1 )

The above example will match this text:

        color red, color green, color blue,   # with or without that last comma

If you allows empty items, enclose the item in a "maybe" function. Note that in this condition, the third parameter becomes irrelevant.

        slist( maybe('color' qa('red green blue')), ',')


The "get" function is used to capture the text that matches a specific portion of the grammar and pass it into a user defined subroutine.

The first parameter of the "get" function can be one of several possible things.

(1) a scalar reference

(2) an array reference

(3) a subroutine reference.

If the first parameter is a (1) scalar reference, the scalar will be used to hold whatever matched the "get" function.

        my $name;

        $grammar = match('hello', get(\$name, some(wrd)), '!' );

        $grammar->('hello Alice !');

        print "name is '$name'\n";

        > name is 'Alice'

If your grammar gets more than one match while it parses, the scalar reference only holds the last match. If you use the same grammar to parse different strings, scalars are initialized to whatever value they held when the grammar was generated, and this value is overwritten only if a match occurs.

        my $name;

        $grammar = match('hello', get(\$name, some(wrd)), '!' );
        $grammar->('hello Alice !');
        $grammar->('hello Bob !');
        $grammar->('hello Charlie !');

        print "name is '$name'\n";

        > name is 'Charlie'

If the first parameter is (2) an array reference, then every match by the "get" function will get pushed into the array. If you call the same grammar multiple times, the array is emptied at the start of parsing. Note this behaviour is different from scalars.

        my @name;

        $grammar = match(some('hello', get(\@name, some(wrd)), '!') );
        $grammar->('hello Alice ! hello Bob ! hello Charlie !');

        print "name is ". Dumper \@name;

        > name is $VAR1 = [
        >       'Alice',
        >       'Bob',
        >       'Charlie'
        > ];

If the first parameter is (3) a subroutine reference, the parser will call this subroutine on every match and pass in as the first parameter the value of the string that matched. You may then do whatever you wish with this string inside the subroutine you define.

        my $name = sub{
                my ($string) = @_;
                # write $string to a file, or whatever
                print "called subroutine, received '$string'\n";

        $grammar = match('hello', get($name, some(wrd)), '!' );

        $grammar->('hello Alice !');

        > called subroutine, received 'Alice'

NOTE: When you use a subroutine reference with "get", you should always finish the subroutine with a "return;" statement. If you're subroutine returns anything other than undef, (and it will if the last expression is anything other than undef), then Parse::Gnaw will treat that return value as a replacement string and substitute that string in place of the string captured by "get".

See the "get with modify" section.

The remaining parameters of the "get" function are one or more grammar components. Literals, character classes, alternations, quantifiers, array refs used for grouping sequences of grammar components, etc.

The "get" function can be nested, like nested capturing parens in perl regular expressions.

        my ($g1, $g2, $g3, $g4, $g5, $g6, $g7, $g8);

        $grammar = parse( 
                        get(\$g1, get(\$g2, thing), get(\$g3, thing) ),
                        get(\$g4, get(\$g5, thing), get(\$g6, thing), get(\$g7, thing) ),

        ok($g2 eq 'a', "g2");
        ok($g3 eq 'b', "g3");
        ok($g1 eq 'ab', "g1");
        ok($g5 eq 'c', "g5");
        ok($g6 eq 'd', "g6");
        ok($g7 eq 'e', "g7");
        ok($g4 eq 'cde', "g4");
        ok($g8 eq 'abcde', "g8");

c (capture)

For short grammars or regular expressions, the "get" function might be a bit of needless overhead. For those who want to save a bit of typing, the "c" function is available.

The "c" function is somewhat like "capture parens" in perl regular expressions in that it captures whatever that matches inside the parens and puts it into a numbered variable. The "c" function uses numbered variables that look like this:

        $c1 $c2 $c3 and so on.

You can have up to 99 capture variables. The variables are declared and imported into your namespace when you use the Parse::Gnaw package.

An exmaple using the C function might look like th"Hello Alice, how are you?"is:

        $grammar = parse( 
                        c( c(thing),c(thing) ),
                        c( c(thing), c(thing), c(thing) ),


        ok($c1 eq 'abcde',      "checking c1 ");
        ok($c2 eq 'ab',         "checking c2 ");
        ok($c3 eq 'a',          "checking c3 ");
        ok($c4 eq 'b',          "checking c4 ");
        ok($c5 eq 'cde',        "checking c5 ");
        ok($c6 eq 'c',          "checking c6 ");
        ok($c7 eq 'd',          "checking c7 ");
        ok($c8 eq 'e',          "checking c8 ");

One significant difference between "c" numbered variables and perl regular expression capture parens is that the "c" function puts the captured results into the numbered variable that corresponds to the number of left parens that actually match the particular string being parsed.

This shows up when your grammar uses the "a" (alternation) function. In the below example, we have a grammar that uses an alternation to look for one of two possible sentences, and then capture different bits of each sentence.

        $grammar = parse( 
                        ['hello', c('world'), 'my', 'name', 'is', c(something)],
                        ['howdy', c(something)]

        ok($grammar->('hello world my name is Alice')==1, "1 match");

        ok($c1 eq 'world', 'c1 is world');
        ok($c2 eq 'Alice', 'c2 is alice');

        ok($grammar->('howdy Bob')==1, "1 match");
        ok($c1 eq 'Bob', 'c1 is Bob');

With the first string we parse, the first alternate matches, therefore the first "c" function is assigned to "$c1" and the second "c" function is assigned to "$c2".

With the second string we parse, the second alternate matches and the c(something) function after 'howdy' is assigned to "$c1". The two "c" functions in the first alternate were not part of the correct interpretation, and therefore were not used to count the numbered variables.

get with modify

The "get" function has a secondary mode when it is used with the "modify" function: It can be used to do inline text substitution.

The example below will look for "Alice" and replace it with "Bob".

        my $text = "Hello Alice, how are you?";
        $grammar=swap("Alice", sub{return "Bob"});
        ok($grammar->($text), "match found");
        ok($text eq "Hello Bob, how are you?", "substitution occurred");

The above example could be done with less typing by using the "swap" function. The "swap" function is nothing more than the "modify" and "get" function wrapped up for you to look more like perls regular expression syntax and to reduce some of the typing you need to do.

However, using "get" to do inline text replacement has it's advantage for more complicated grammars or patterns. Sometimes when you do a substitution, you need to make sure you have the correct text by looking for certain text before and after the pattern. But when you use this in a s/// regular expression, you have to capture the before and after text and put it into the replacement text as $1 and $2.

This is how you might look for two numbers separated by a plus character "+" and replace it with a minus "-" character, using regular expressions:

        my $text = "A+, 123+456, count++";
        ok($text=~s{(\d+)\+(\d+)}{$1-$2}, "match found");
        ok($text eq 'A+, 123-456, count++', "replacement took place");

With Parse::Gnaw, you don't have to capture $1 and $2 and use it in the replacement string. Instead, you can use the "modify" function, define the pattern, and then use "get" to replace just the part you want, without using $1 and $2, and without capturing text before and after.

        my $text = "A+, 123+456, count++";
        ok(modify(some(dgt),get(sub{return'-'},'+'),some(dgt))->($text), "match found");
        ok($text eq 'A+, 123-456, count++', "replacement took place");

now versus defer

The "now" function allows you to define a subroutine to be called every time the parser hits that part of the grammar. The code subroutine will be called whether that part of the grammar eventually matches the text being parsed or not.

The "defer" function allows you to define a subroutine to be called only if that branch of the grammar matches the text being parsed. The subroutine handed to the "defer" function is called at the end of parsing when a match occurs or when the "commit" function is called.

This example uses the "now" function:

        sub howdy { [ now(sub{print"trying howdy\n"}), 'howdy' ] }

        sub hi    { [ now(sub{print"trying hi   \n"}), 'hi' ] }

        $grammar = match(a(howdy, hi);


The above example will print

        trying howdy
        trying hi

Even though the alternative "howdy" ended up failing.

The next example uses the "defer" function instead:

        sub howdy { [ defer(sub{print"trying howdy\n"}), 'howdy' ] }

        sub hi    { [ defer(sub{print"trying hi   \n"}), 'hi' ] }

        $grammar = match(a(howdy, hi);


The above example with print only

        trying hi

Because "howdy" ended up failing, the defered subroutine for "howdy" did not get called.

Because the "defer" function is more likely to be useful in more grammars, the short cut to do a "defer" function is to simply pass in a subroutine to any of the other grammar functions. Any code reference will be assumed to be a "defer".

Rewriting the "defer" example above but with the shorthand would look like this:

        sub howdy { [ sub{print"trying howdy\n"}, 'howdy' ] }

        sub hi    { [ sub{print"trying hi   \n"}, 'hi' ] }

        $grammar = match(a(howdy, hi);


As before, the above example with print only

        trying hi

The "defer" function is useful for doing things only when the parser knows it has a correct interpretation of the text being parsed. The "get" function is a "defer" function that has been wrapped.

The "now" function is useful for doing something even if the interpretation might fail later. One example is printing debug statements of a grammar to understand where ther parser is going as it works through a piece of text. The "commit" function is a "now" function that has been wrapped with extra features.


The "commit" function is used to commit the parser to a particular interpretation of a grammar. If the parser is working a grammar and applying it to a text, and the parser hits a "commit" function, then that path in the grammar must match. If the grammar fails to match after a "commit", rather than try any other possible interpretations, the parser will quit with an error.

The "commit" function will cause any currently scheduled "defer" functions to be executed by the parser. The "commit" function can also trigger the text being parsed to be deleted up to the current character being parsed.

Use "commit" if your grammar contains a reserved keyword and you want to force the parser to interpret the text as a keyword. Use the "commit" function if you are parsing a large chunk of text and you want to be able to remove the previous text from memory.

        $grammar = match(some('keyword', commit, 'blah', 'blah', 'blah'));


By default, Parse::Gnaw skips whitespace in between literals and character classes. If you do not want to skip whitespace, use the "noskip" function. It will turn off skipping and look for all the grammar components to not have any extra whitespace in them.

        $grammar = match('x', noskip('hello', 'world'), 'y');

        # turn off skipping after x, no space between literals till skipping turned back on.
        ok($grammar->('xhelloworld  y')==1, "1 match found");

        # this doesn't match because of space between x and hello
        ok($grammar->('x  helloworld  y')==0, "2 match not found");

        # this doesn't match because of space between x and hello
        ok($grammar->('xhello world  y')==0, "3 match not found");

Note that the "noskip" takes effect immediately, so there is no skipping between the 'x' literal and the literal 'hello'.

Also, note that "noskip" must enclose whatever grammar components you want it to be applied to. There is no way for a grammar to change the "skip" value for all components after it. You cannot do something like this:

        $grammar = match(a( ['hello'], ['howdy',noskip] ), 'world' );

In the example above, there are two alternates, 'hello' and 'howdy'. It would appear that in this grammar, if the alternate turns out to be 'howdy', that the parser should not skip anything when it looks for 'world'. This attempts to have the "noskip" function effect the grammar outside of the "noskip" scope. The "noskip" function does not work that way. You must limit where the "noskip" function applies itself by putting those components inside the "noskip" function.

If you really wanted either 'hello' skip 'world', or, 'howdy' noskip 'world', you would have to unroll the code.

        $grammar = match( a( ['hello','world'], ['howdy', noskip('world')] ));


By default, Parse::Gnaw skips whitespace in between literals and character classes. So, why would you need an instruction to tell it to skip? When you are about to call "noskip".

        sub name { some(wrd) }
        $grammar = match( 'hello', name, '!' );
        ok($grammar->("hello Alice !")==1, "expect match");
        ok($grammar->("hello a b c d !")==1, "might not have expected this to match");

In the above example, we use "some" to look for one or more "wrd" characters. We might use this to look for an identifier, a name, or similar. The problem is that the "wrd" function will "skip" whitespace every time it is called. Which means it will skip whitespace every time the "some" function calls it. Which means it will match the string "a b c d", which might not be what we expected.

To force the parser to not skip, we might try putting "noskip" around the "wrd" function.

        sub name { [some(noskip(wrd))] }
        $grammar = match( 'hello', name );
        ok($grammar->("hello Alice")==0, "this won't match");
        ok($grammar->("helloAlice")==1, "this will match");

In the above example, we call "noskip" around the "wrd" function, but that means the parser will not skip the first time it enters the "some" function, which means it will not skip between "hello" and the first "wrd" character in the "some" function. So, the above grammar will only match "helloAlice" or similar situations.

To overcome this, we need to be able to command the parser to skip right now. And then we need to put that skip right now function into the grammar just prior to the "some" function. This will cause the parser to skip whitespace, and then look for "some" "wrd" characters that are separated by no whitespace.

        sub name { [skipnow, some(noskip(wrd))] }
        $grammar = match( 'hello', name );
        ok($grammar->("hello Alice")==0, "this won't match");
        ok($grammar->("helloAlice")==1, "this will match");

In the above example, we tell the parser to "skip" and then we look for some wrd characters. Note that "skip" will skip zero or more whitespace. This means we could also use "anyspace" as a synonym for "skip".

id (identifier) function

Because doing a series of "wrd" characters with no spaces in between them and word boundaries on both ends, with a "skip" function at the front is rather wordy, the "id" function has been provided for you.

It takes one optional parameter, and if you provide one parameter, then the "id" function passes it into the "get" function for you.

The declaration for "id" can be simplified to this:

        sub id { 
                if(scalar(@_)) {
                        my $getthingy=shift(@_);
                        return series(skip, bb, get($getthingy, some(noskip(wrd))), bb);
                } else {
                        return series(skip, bb, some(noskip(wrd)), bb);

You can use it to match but not capture:

        $grammar = match('hello', id, '!');
        ok($grammar->("hello Alice!")==1, "match");

Or to match and capture

        my $name='';
        $grammar = match('hello', id(\$name), '!');
        ok($grammar->("hello Alice!")==1, "match");
        ok($name eq 'Alice', "check capture occurred");


The "token" function is if you want to define a series of items that are to occur with no skipping in between, but might have skipping before and after.

Sort of like "id", but without the "bb" function on either end.

If you wanted to define an operator that used more than one character, but you wanted to require that they not be separated by any spacing (but could allow spacing before and after), then you could use "token" to define that.

Say you were looking for a digit followed by a letter, you could define it like this:

        sub diglet { token(dgt, cc('a-zA-Z') }

A simplified declaration for "token" might look like this:

        sub token {  series( skipnow, noskip( @_ ) ) }

It's not a very complicated function, but it summarizes the idea of a 'token' in parsing well, so it is provided for your convenience.

positional markers



The 'bb' and 'BB' functions are positional assertions. The 'bb' marker is true if the the text being parsed contains a word boundary at that location. The 'BB' function is true if the text being parsed does NOT contain a word boundary at that location.

A word boundary can occur under four different situations:

        1) at a transition from a 'wrd' character to a 'WRD' character
        2) at a transition from a 'WRD' character to a 'wrd' character
        3) at the beginning of the string
        4) at the end of the string

The following grammar will only match text if it contains the literal 'alice' that contains word boundaries before and after the literal. i.e. something must separate 'alice' from the rest of the text.

        $grammar = match(bb, 'alice', bb);

        # beginning of string is a word boundary
        ok($grammar->('alice bob charlie')==1, "match found"); 

        # spaces are word boundaries
        ok($grammar->('bob alice charlie')==1, "match found");

        # end of string is a word boundary
        ok($grammar->('bob charlie alice')==1, "match found");

        # the '(' and ')' symbols are word boundaries
        ok($grammar->('bob alice(42) charlie')==1, "match found");
        ok($grammar->('bob (71)alice charlie')==1, "match found");

        # 'bob' and 'charlie' are too close to 'alice', therefore 
        # 'alice' is not separated before and after with word boundaries.
        ok($grammar->('bob bobalice charlie')==0, "match not found");
        ok($grammar->('bob alicecharlie charlie')==0, "match not found");
        ok($grammar->('bob bobalicecharlie charlie')==0, "match not found");

end function

The "end" function is a positional assertion that matches only if the parser has reached the end of the input text. Before testing for the end of input, the "end" function will execute the "skip" function, if any is currently active, so that it may skip over any content you deem ignorable. It then checks to see if it has reached the end of the input text. If it has, then it matches.

If it doesn't, then it fails.

Any grammar components after the "end" function will produce undefined behaviour, since nothing can be parsed after the end of the input text.

If you want to see if some text, from beginning to end, matches your pattern, then use the "parse" function, and make the last grammar component the "end" function. For example, this grammar will only match if the input text from beginning to end consists of one or more widgets (where "widgets" is defined by another subroutine of that name.):

        $grammar = parse( some( widgets ), end );

If you use the "match" function, it will start at the beginning but will skip ahead until it finds a match. If you omit the "end" function, any text that does not match will be ignored and the parsing will stop at the last matching widget. To confirm that the whole text is nothing but one or more widgets, use the above syntax.


Since the grammar components are built up as subroutines, it is possible to create your own grammar components as subroutines that call Parse::Gnaw subroutines and encapsulate some higher level functionality or provide shortcuts for you in your grammars.

This means that you are not restricted to writing your grammars using only the low level components available in some parser language. If your grammar repeats the same complex pattern over and over, you can put that patten into a subroutine and call the subroutine wherever you need it.

Note that this is better than a grammar that supports "rules", because extensibility means you can pass other grammar components into your subroutine and manipulate them however you want before generating the final parsing logic.

dgt, wrd, spc

An simple example of extensibility is how the character class shortcuts for digits and so on are extensions of the basic underlying "cc" and "CC" functions, rather than being their own grammar component. Here is how these shortcuts could be declared:

        sub dgt { cc('0-9') }
        sub DGT { CC('0-9') }

        sub wrd { cc('a-zA-Z0-9_') }
        sub WRD { CC('a-zA-Z0-9_') }

        sub spc { cc("\n \t \r \f") }
        sub SPC { CC("\n \t \r \f") }

If you wanted to create even shorter versions of the above shortcuts, you could declare them as subroutines in your own code and then call them in your grammar. For example, if you want to use the letter 'd' every time you want a digit character class, you could declare the following in your code:

        sub d { dgt }


Another example of a grammar extension is the "qa" function. The "a" function is a low level grammar component for alternations. The "qa" function is an extension of "a". It's declaration looks like this:

        sub qa {
                my ($string)=@_;
                my @words = __gnaw__separate_string_into_words($string);
                my $stitcher = a(@words);
                return $stitcher;

The "qa" function takes in a single scalar string, splits that string up into individual words based on whitespace, and creates an array of words. The "qa" function then passes this array of words into the "a" function and returns the stitcher that it creates.

The "qa" function isn't something you can do in a normal parser "rule" because you're passing in a nonstandard parameter, preprocessing it, converting it to standard grammar building blocks, and then returning the parser code.

any some anything something maybe

Another example of extensions are the quantifier shortcut functions. The low level components are the "thrifty" quantifier) and "thing" (any single character) components. From there, we declare the extensions:

        sub any  { thrifty([0], @_) }           # zero or more
        sub some { thrifty([1], @_) }           # one or more

        sub anything  { any (thing) }   # zero or more 'things'
        sub something { some(thing) }   # one or more 'things'

        sub maybe { g([0,1], @_) }

The "any" and "some" functions are simple extensions of the "thrifty" function, taking whatever parameters are passed into "any" and "some" and passing them directly onto the "thrifty" function. Instead of preprocessing the parameters like we did in the "qa" function, instead we hardcode some default values for the quantifier "min" value. i.e. the [0] and [1] parts of the code passed into the "thrifty" function.


The "slist" function is a more complex extension of basic grammar components. The "slist" subroutine can be simplified to something like this:

        sub slist {
                my($item,$separator,$trailingseparatorallowed) = @_;
                my $extension;

                        # an item, 
                        # followed by any number of (separator item) 
                        # and maybe one separator at end
                        $extension = series( $item,  any($separator, $item), maybe($separator) );
                } else {
                        # an item, followed by any number of ( separator item )                 
                        $extension = series( $item, any($separator, $item) );

                return $extension;

The "slist" function is also not something that can be supported by normal parser "rules". The "slist" subroutine is called during the "generation" phase, before the parser is built, and based on the parameters passed in, it decides what sort of grammar to build for a separated list.

az function

The "az" function is a simple extension

        sub az { parse( @_, end ) }

It requires that the string being parsed matches from beginning to end the pattern passed in.

Making Your Own Extensions

The process to making your own grammar extensions can be broken down into a few steps:

        1: deal with any pre-processed arguments passed in
        2: deal with any remaining grammar arguments passed in
        3: return a "stitcher"

preprocessed arguments

preprocessed arguments are arguments passed into your extension subroutine that do not translate directly into grammar components but instead are used by your subroutine in some way to decide what sort of component to generate.

The "qa" (quoted alternation) function takes one preprocessed argument, a single string with space separated literals. It then breaks that string up into an array of literals that can then be passed into an existing grammar component function, the "a" function.

If your extension has some preprocessed arguments, this is the first thing you want to deal with. Does your extension always receive a fixed number of preprocessed arguments? Do they always appear at the same position in the list of arguments (always first or always last or similar)? If so, then you might start by removing these preprocessed arguments from the @_ list. And potentially error checking them in some way, such as calling perls "ref" function on them to make sure they're not a "stitcher" or some other thingy.

If you have a variable number of preprocessed arguments, or if they appear in different positions within the @_ parameter list, then you'll need some way of determining how many and where. Usually, the number of regular grammar components passed into any function is variable, so any pre-processed arguments are usually fixed to make them easier. However you could devise alternative ways to distinguish your preprocessed arguments from the normal grammar component arguments. Since extensions are just perl subroutines, the method is up to you.

Once the preprocessed arguments are removed from @_, process them however your extension needs to deal with them. Usually this entails passing them into some grammar component or another to generate a stitcher.

In the "slist" function, we have two "preprocessed" arguments which are the "item" and the "separator". Because we want to allow the item and separator to be any number of grammar components, we force the number of paramters allowed to exactly two, and then require the user to group multiple components into array references.

The "slist" function then passes the "item" and "separator" pieces directly into other existing components.

remaining grammar arguments

If your extension passes in normal grammar components, such as literal strings, or arrays of literal strings, or functions such as "a" or "bb" or whatever, then if possible, pass these items directly into whatever existing grammar function provides the behaviour pattern you want.

If you want the arguments to be processed in a series, then pass them into the "series" function, which will process them as if they were normal parts of some grammar and return a "stitcher" function.

Since you removed any preprocessed arguments from the @_ array, your extension should be able to pass the remaining @_ array into whatever lower level functional components you want.

For example, the "some" function is a shortcut to the "thrifty" function, forcing the quantity to be "1 or more". You can pass any number of grammar components into the "some" function. The "some" function simply passes them all directly into the "thrifty" function.

        sub some { thrifty([1], @_) }

The "az" function is an extension that wraps the "parse" function, assumes everything passed in is a grammar component and adds the "end" function to the end of the pattern.

        sub az { parse( @_, end ) }

return a stitcher

Most extensions operate as low level grammar components and need to return a "stitcher" so that the return value of the extension can be used inside a larger grammar. You can use the "some" extension inside a larger grammar and not even know it is an extension because it returns a stitcher just like any other low level component.

In the case of the "some" function, the stitcher is created by the "thrifty" function, and since it is the last expression of the "some" subroutine, the return value of the "thrifty" function is the return value of the "some" function. Therefore the "some" extension returns a stitcher without any explicit return call.

        sub some { g([1], @_) }

The "slist" extension generates one of two possible subgrammars and returns the stitcher for that component. It was coded with an explicit return just so that there is only one return in the subroutine.

        sub slist {
                my($item,$separator,$trailingseparatorallowed) = @_;
                my $extension;

                        # an item, 
                        # followed by any number of (separator item) 
                        # and maybe one separator at end
                        $extension = series( $item,  any($separator, $item), maybe($separator) );
                } else {
                        # an item, followed by any number of ( separator item )                 
                        $extension = series( $item, any($separator, $item) );

                return $extension;

The "slist" function uses the "series" function to generate a stitcher, and then the "slist" function returns that stitcher. The series function is explained in more detail below.

You don't have to return a stitcher if your extension is intended to operate as a top level function. The "az" function is an extension that operates as a top level function wrapping the "parse" function. Just remember that top level functions return coderefs that act as the parser when you dereference them. Top level functions cannot be part of a larger grammar.

If you don't return a stitcher, about the only other thing you can return is a top level code reference.

series function

The "series" function is a low level function in Parse::Gnaw that takes any number of grammar components, literals, array references of grammar components, and so on, and creates a single subgrammar that expects all those components to occur in one long series.

You don't have to use the "series" function in your normal grammars because all the other functions use it for you. For example the "match" function takes any number of input parameters and passes them into the "series" function, it then uses that stitcher to create the "match" function.

If you're creating a grammar extension, however, you might want to explicitely call the "series" function to put several grammar components together into a sequence and generate a single stitcher.

The "slist" extension is an example of an extension that uses the "series" function to generate the stitcher.

trekism extension explained

The trekisms grammar, shown elsewhere in this document, contains an interesting grammar extension. It doesn't really look like a grammar extension so much as it looks like a nice way to bundle some funcitonality that gets reused throughout the grammar.

The grammar identifies a number of bits of dialogue and assumes certain characters say those lines of dialogue. The grammar then prints out the name of the character along with the line they said.

Since this function is done in the grammar at multiple locations, but for different character names, normal grammars would have to execute this functionality as some sort of "action" or defered callback or somehting and then figure out a way to capture the line of dialoge as well as figure out a way to pass in different character names for each instance.

In the case of Parse::Gnaw, we simply make an extension. Here is the entire code for the extension subroutine:

        sub attributeline { 
                my $character_name=shift(@_);
                my @lines_of_dialogue_for_this_character = @_;
                my $callback = sub{$captured.=sprintf ("%8s says: %s\n", $character_name,shift(@_));};
                my $stitcher = get($callback, a(@lines_of_dialogue_for_this_character));
                return $stitcher;

The "attributeline" extension takes one preprocessed argument, a string that is the name of the character for this instance. The remaining arguments are each treated as unique lines of dialogue that the character might say.

It takes the character name and creates a callback for a "get" function. The callback takes the name, the line of dialogue handed to it by "get", creates a pretty-formatted string, and then appends it to the end of a "captured" variable.

The extension then creates a stitcher that is nothing more than the "get" function, the callback, and the lines of dialogue handed into the "a" function, treating each line as a valid altenate for that character.

By passing the @lines of dialogue into "a", we get alternates that we can pass into "get". By passing it into the "get" function, our callback has the line of dialogue available to put into a formatted version. By passing the "name" variable into the callback, the callback becomes a closure and remembers the "name" variable so it can be sent out when the callback is called by the "get" function.

See the "trekisms" grammar in the examples below:



The following grammar shows how one might identify dialogue and attribute it to a particular fictional character.

The "attributeline" subroutine is basically an extension which allows us to bundle the functionality of the output into one spot and then easily use that functionality at several different points in the grammar.

        my $captured="";

        sub attributeline { 
                my $character_name=shift(@_);
                my @lines_of_dialogue_for_this_character = @_;
                my $callback = sub{$captured.=sprintf ("%8s says: %s\n", $character_name,shift(@_));};
                my $stitcher = get($callback, a(@lines_of_dialogue_for_this_character));
                return $stitcher;

        sub trekname { qa('Jim Captain Spock Bones Doctor Scotty') } 
        sub occupation {a('ditch digger', 'bricklayer', 'mechanic')}
        sub mccoy_job { [ql("I'm a doctor, not a"), occupation, a('!', '.')] }
        sub mccoy_diag { [ "He's", 'dead', ',', trekname, a('!', '.') ] }
        sub mccoy_rant1 { [ql('You green-blooded Vulcan'), a('!', '.') ] }
        sub mccoy_isms {
                attributeline('McCoy', mccoy_job, mccoy_diag, mccoy_rant1) 

        sub spock_awe {['Fascinating', ',', trekname, '.']}
        sub spock_logic {['Highly', 'illogical',',', trekname, '.']}
        sub spock_sensors { [ql("It's life ,"), trekname, ql(', but not as we know it .')]}
        sub spock_isms {
                attributeline('Spock', spock_awe, spock_logic, spock_sensors)
        sub kirk_dipolomacy1 {ql('We come in peace .')}
        sub kirk_dipolomacy2 {ql('Shoot to kill .')}
        sub kirk_to_scotty {ql('I need warp speed now , Scotty !')}
        sub kirk_to_spock {ql('What is it , Spock ?')}
        sub kirk_to_bones {ql('Just fix him , Bones')}
        sub kirk_solution {ql('Activate ship self-destruct mechanism .')}
        sub kirk_isms {

        sub scotty_phy101 {ql('Ya kenna change the laws of physics .')}
        sub time_units {qa('minutes hours days weeks')}
        sub scotty_estimate {[ ql("I'll have it ready for you in three"), time_units, '.' ]}
        sub scotty_isms {attributeline('Scotty', scotty_phy101, scotty_estimate)}
        sub alien_isms {attributeline('alien', 'weeboo')}
        sub trek_isms {a(mccoy_isms, spock_isms, kirk_isms, scotty_isms, alien_isms )}
        sub trek_script {some(trek_isms), end}  
        $grammar = parse(  trek_script );
        my $script = <<'SCRIPT';
        What is it, Spock?
        It's life, Jim, but not as we know it.
        We come in peace.
        Shoot to kill.
        I need warp speed now, Scotty!
        I'll have it ready for you in three minutes.
        I need warp speed now, Scotty!
        Ya kenna change the laws of physics.    
        Shoot to kill.
        Shoot to kill.
        I'm a doctor, not a bricklayer.
        Highly illogical, Doctor.
        You green-blooded Vulcan.
        Shoot to kill.
        Shoot to kill.
        He's dead, Jim.
        Activate ship self-destruct mechanism.
        Highly illogical, Captain.

        #print "script is '$script'\n";

        ok($grammar->( $script )==1, "1 match");

        my $expected =  <<'EXPECTED';
            Kirk says: What is it, Spock?
           Spock says: It's life, Jim, but not as we know it.
            Kirk says: We come in peace.
           alien says: weeboo
            Kirk says: Shoot to kill.
           alien says: weeboo
            Kirk says: I need warp speed now, Scotty!
          Scotty says: I'll have it ready for you in three minutes.
           alien says: weeboo
            Kirk says: I need warp speed now, Scotty!
          Scotty says: Ya kenna change the laws of physics.
           alien says: weeboo
           alien says: weeboo
            Kirk says: Shoot to kill.
            Kirk says: Shoot to kill.
           McCoy says: I'm a doctor, not a bricklayer.
           Spock says: Highly illogical, Doctor.
           McCoy says: You green-blooded Vulcan.
            Kirk says: Shoot to kill.
            Kirk says: Shoot to kill.
           McCoy says: He's dead, Jim.
            Kirk says: Activate ship self-destruct mechanism.
           Spock says: Highly illogical, Captain.
        ok($captured eq $expected, "checking captured string matches expected");

The above example is included as a test in the t directory of the tarball from CPAN.

Note that when we call the "attributeline" extension, we pass in a string that isn't part of the grammar to be searched for in the text, it is part of the text output. In the line below, we call "attributeline('McCoy', ...)", but we aren't looking for "McCoy" in the text being parsed, we pass in "McCoy" so we can use that string in the formatted output.

        sub mccoy_isms {
                attributeline('McCoy', mccoy_job, mccoy_diag, mccoy_rant1) 

See the section on "extensions" for more details.


By default, the Parse::Gnaw module works similar to perl regular expressions. It works on the text in memory and loads the entire text when the top level function callback is called and passed a string to parse.

However, the Parse::Gnaw module was designed from the ground up to work with partial texts, to work with a subset of an entire piece of text, and pull into memory only what it needed, and flush out of memory any text that it was finished parsing.

Theoretically, Parse::Gnaw should be able to parse an infinite amount of text as long as the grammar defines rules that only need a finite chunk that will fit in memory to find a match.

To support this feature, Parse::Gnaw has some hooks that allow you to modify its default behaviour so that it only needs to operate on a subset of the entire text at a time.

These functions, including the "commit" function, are defined in this section.


The "commit" function is a lower level grammar component function that you put into your grammar to define points in the grammar where the parser can commit to the current interpretation because no other interpretation is possible.

Normally, the parser will not commit to an interpretation until it has reached the very last component in the grammar. The "commit" function allows you to force the parser to commit to the current interpretation.

This helps parse streams because the "commit" function means that the parser will commit any text parsed so far, call any callbacks on that text, and then flush the text from memory.


After the "commit" function calls any callbacks on the text parsed so far, the parser calls the "flush" function.

        our $__gnaw__flush = undef;

By default, it is undef, which means any committed text is ignored and simply deleted from memory after the "commit" function is called. This might be the functionality you want if you are doing a read-only parse of a block of text.

However, if you are parsing a file, modifying, and want to save the modifications, then you can define $__gnaw__flush to point to a subroutine reference which will write the results to a harddrive or some external device.

If you want a basic append-to-file function, you can call the followng function and pass it the filename:

        __gnaw__flushed_text_is_appended_to_file( 'filename' );

The function will return the filehandle of the file being appended to. You can ignore it if you want or you can use the filehandle to force the file be closed in the middle of your script execution so you can open another file.


At the input side, we need a mechanism for reading a little bit of text from some external source as we need it. This is handled by the "sip" function.

        our $__gnaw__sip = sub{};

Whenever the Parse::Gnaw code hits the end of the text in memory, it will call the code reference in $__gnaw__sip. If this callback wants to add text to the linked list in memory, it can get the text it wants to add and call the __gnaw__insert_string_at_end_of_linked_list() function, passing the new text to that subroutine.

The amount of text you add to the end of the linked list is up to you. If the parser hits the end of text in memory, it will simply call the $__gnaw__sip function again.

When your $__gnaw__sip function has reached the end of its input source file(s), you will want your subroutine to handle further calls to it by doing nothing and returning immediately.

By default, the $__gnaw__sip subref is set to a subroutine reference that doesn't do anything. This is because the default behaviour of Parse::Gnaw is to process the entire text being parsed in memory.


matchall swapall

The match and swap functions only operate on the first instance in the string that matches.

If you want to match, and then keep going, or if you want to swap something and keep going, use the matchall and swapall function.


The "recursive" function has not been implemented yet.

If you define a grammar which is recursive, you must wrap the recursive call in the "recursive" function.

If you declare a recursive grammar without using a "recursive" function, then perl will recursively call the subroutine until it dies a horrible death.

        sub expression;
        sub expression{ a([ '(' expression ')'], 'a+b') }

        # this call to "expression" will recursively call itself
        # until perl crashes.
        my $grammar = match(expression); 

The above example will crash. Note that it crashes during the "generate" stage of creating the grammar. The grammar isnt even being used to try to match a string yet. The mere act of declaring a recursive grammar without the "recursive" function will cause a crash.

What you need to do is wrap any recursive call to a rule in the "recursive" function.

        sub expression;
        sub expression{ a([ '(' recursive(\&expression) ')'], 'a+b') }

        my $grammar = match(expression);


These are notes to developers who may need to understand how the internals of Parse::Gnaw operate so they can create advanced grammars.

Also, these are notes to myself to remember how things function.

Text being parsed.

The text being parsed is held internally in a linked list. This allows Parse::Gnaw to operate on a small subset of the entire string being parsed. Theoretically, Parse::Gnaw can parse an infinite amount of text as long as it can be parsed in chunks small enough that the rules can disambiguate text that fits in memory.

Each element in the linked list is an array.

Array stores the following data:

        0: numeric indicator of what type of element this is
        1: payload
        2: previous element pointer
        3: next element pointer
        4: used for debugging

There are several "types" of elements as indicated by the value in index 0 of an element:

        0: element has been deleted. If code tries to use 
                a deleted element, this should act as a 
                flag that something went wrong
        1: element holds a single text character. 
                letter is stored in payload index.
        2: element holds a "marker". Markers are placed between
                text elements by the parser to keep track of
                where it is in the code, where captured strings
                are located, where to fallback to in the text
                if a particular branch of a parse fails, etc.
                A marker contains no payload.
        3: element contains a callback to be executed on
                successful completion of parsing or a commit.
                Payload contains code ref to callback.
        4: element is a head/tail endpoint
                These two elements are never to be deleted.
        5: word boundary markers inserted by "sip".
                These might indicate transition from one file to another.
                Note that normally moving text pointer forward by one
                will delete all old markers, but it won't delete these markers.

The linked list starts and ends with two signpost elements: 'head' and 'tail'. No text is ever stored in these signpost elements. Any text being parsed is always contained between these two signpost elements.

The 'current' pointer always points to the text that is about to be processed.

When the __gnaw__parse_failed function is called, it will pop the last fallback position off the fallback array, which will change the next instruction to be executed and will also move the text pointer back to teh fallback marker. The last thing the __gnaw__parse_failed function does is delete the fallback marker and any other elements in the text linked list that are not text.

When any instruction moves the text pointer forward, it will also delete any old markers it encounters.

The current pointer may point to the tail signpost element. This may mean that we are at the end of the string, or it may mean that the next time we try to get the current letter that we need to insert more text from a file or something.

All interfaces to the linked list should be through the subroutines designed for working with the linked list. These start with the prefix "__gnaw__" to help prevent subroutine colisions.

All parse level commands should manipulate text based on markers.

No user-level or grammar-level commands are currently supported. You can always call the __gnaw__* functions, but they are currently subject to change.

Generating the Parser



Greg London, <email at>


Please report any bugs or feature requests to bug-parse-gnaw at, or through the web interface at I will be notified, and then you will automatically be notified of progress on your bug as I make changes.


You can find documentation for this module with the perldoc command.

    perldoc Parse::Gnaw

You can also look for information at:



Copyright 2008 Greg London, all rights reserved.

This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.

syntax highlighting: