This is ALPHA software

Overview

This document contains a top-level overview of the API for the Marpa parse engine. The two examples in this document show the typical flows of Marpa method calls. This document will use these examples to describe the basic features of Marpa in semi-tutorial fashion. Marpa's advanced features, and full reference details of all features, can be found in the other Marpa API documents.

The Three Phases

• Accept a grammar.
• Read input.
• Return values from the parses, according to a semantics.

In Marpa these three tasks are, for the most part, distinct phases. Grammars are "Marpa::PP::Grammar" objects. The reading of input and the evaluation of the parse according to the semantics is performed by "Marpa::PP::Recognizer" objects.

Marpa::PP::Grammar::new

    my $grammar = Marpa::PP::Grammar->new(
        {   start   => 'Expression',
            actions => 'My_Actions',
            default_action => 'first_arg',
            rules   => [
                { lhs => 'Expression', rhs => [qw/Term/] },
                { lhs => 'Term', rhs => [qw/Factor/] },
                { lhs => 'Factor', rhs => [qw/Number/] },
                { lhs => 'Term', rhs => [qw/Term Add Term/], action => 'do_add' },
                {   lhs    => 'Factor',
                    rhs    => [qw/Factor Multiply Factor/],
                    action => 'do_multiply'
                },
            ],
        }
    );

Marpa grammars are "Marpa::PP::Grammar" objects. They are created with the Marpa::PP::Grammar::new constructor. The arguments to Marpa::PP::Grammar::new are references to hashes of named arguments. In the key/value pairs of these hashes, the hash key is the name of the argument, and the hash value is the value of the named argument.

The start Named Argument

    start => 'Expression',

The "start" named argument is required. Its value is a string containing the name of the grammar's start symbol.

Named Arguments for the Semantics

            actions => 'My_Actions',
            default_action => 'first_arg',

The "actions" and "default_action" named arguments specify semantics. Their argument values are strings, which acquire their semantics during evaluation.

Evaluation will be described later. Peeking ahead, the "default_action" named argument will be interpreted as an action name. This action name will resolve to an action -- a Perl closure that implements semantics. The action specified by "default_action" is used as the action for rules with no action of their own. "actions" provides the name of a Perl package where Marpa will look for its actions.

The rules Named Argument

    rules => [
        { lhs => 'Expression', rhs => [qw/Term/] },
        { lhs => 'Term',       rhs => [qw/Factor/] },
        { lhs => 'Factor',     rhs => [qw/Number/] },
        { lhs => 'Term', rhs => [qw/Term Add Term/], action => 'do_add' },
        {   lhs    => 'Factor',
            rhs    => [qw/Factor Multiply Factor/],
            action => 'do_multiply'
        },
    ],

The value of the "rules" named argument is a reference to an array of rule descriptors. In this example, all the rule descriptors are in the "long" form -- they are references to hashes of rule properties. In each key/value pair of a rule descriptor hash, the key is the name of a rule property, and the hash value is the value of that rule property.

The lhs Property

The value of the "lhs" rule property must be a string containing the name of the rule's left hand side symbol. Every Marpa rule must have a left hand side symbol.

The rhs Property

The value of the "rhs" property is a reference to an array of strings containing names of the rule's right hand symbols, in order. This array may be zero length, in which case this is an empty rule -- a rule with no symbols on the right hand side. There are no empty rules in this example.

The action Property

The value of the "action" rule property is a string. Peeking ahead, each "action" property string will be interpreted as an action name. This action name will be resolved to a Perl closure that implements the rule's semantics.

Marpa::PP::Grammar::precompute

    $grammar->precompute();

Before a Marpa grammar object can be used by a Marpa recognizer, it must be precomputed. Precomputation compiles data structures that the recognizer will need.

Marpa::PP::Recognizer::new

    my $recce = Marpa::PP::Recognizer->new( { grammar => $grammar } );

"Marpa::PP::Recognizer::new" creates a new recognizer. Its arguments are references to hashes of named arguments. In this example the only named argument is the required argument: ""grammar"". The value of the "grammar" named argument must be a precomputed Marpa grammar.

Marpa::PP::Recognizer::read

    $recce->read( 'Number', 42 );
    $recce->read( 'Multiply', );
    $recce->read( 'Number', 1 );
    $recce->read( 'Add', );
    $recce->read( 'Number', 7 );

The "Marpa::PP::Recognizer::read" method takes two arguments, a token name and a token value. The token name must be the name of a valid terminal symbol in the grammar. By default all symbols are valid as terminal symbols, unless a grammar contains empty rules.

The grammars in the examples in this document do not contain empty rules, and therefore are free to use any symbol in the grammar as a token name. For more on terminals, including how to explicitly mark terminal symbols, see "Terminals" in Marpa::PP::Grammar.

The token value must be a Perl scalar, but otherwise its form and semantics are entirely up to the application. If the token value is omitted, it is a Perl "undef". In the calculator example, the values of the ""Add"" and ""Multiply"" tokens are never used, so they are allowed to default to "undef".

Marpa::PP::Recognizer::value

    my $value_ref = $recce->value;
    my $value = $value_ref ? ${$value_ref} : 'No Parse';

The "Marpa::PP::Recognizer::value" method returns a reference to the parse result's value, if there was a parse result. If there was no parse result, "Marpa::PP::Recognizer::value" returns "undef".

Resolving the Semantics

    actions => 'My_Actions',

    { lhs => 'Factor', rhs => [qw/Factor Multiply Factor/], action => 'do_multiply' },

For example, the "action" property for the above rule is ""do_multiply"" and the "actions" named argument to the grammar was ""My_Actions"". So Marpa looks for a closure whose fully qualified name is "My_Actions::do_multiply", which it finds:

    sub My_Actions::do_multiply {
        my ( undef, $t1, undef, $t2 ) = @_;
        return $t1 * $t2;
    }

Rules do not always have "action" properties. That is the case with these rules in this example:

    { lhs => 'Expression', rhs => [qw/Term/] },
    { lhs => 'Term', rhs => [qw/Factor/] },
    { lhs => 'Factor', rhs => [qw/Number/] },

Where there is no "action" rule property, Marpa tries to use the "lhs" property as an action name. When Marpa cannot resolve the "lhs" property as an action name, it will fall back to using the default action for the grammar.

For example, in the first rule in the above display, Marpa will look for a Perl closure with the fully qualified name ""My_Actions::Expression"". When Marpa does not find a closure by that name, Marpa will fall back to trying to use the default action.

The other two rules in the above display work similarly. Marpa will look for Perl closures named ""My_Actions::Term"" and ""My_Actions::Factor"". It will not find them and will fall back to trying to use the default action, as described next.

    default_action => 'first_arg',

The "default_action" named argument is resolved in the same way as are the "action" properties of the rules. In this example, default_action is specified as ""first_arg"" and resolves to "My_Actions::first_arg".

Actions

    sub My_Actions::first_arg { shift; return shift; }

    sub My_Actions::do_add {
        my ( undef, $t1, undef, $t2 ) = @_;
        return $t1 + $t2;
    }

Value actions are Perl closures used as callbacks. Value actions are called when nodes in a parse tree are evaluated. A value action receives one or more arguments. The first argument to a value action is always a per-parse-result object, which the callbacks can use as a scratchpad. In these examples, the per-parse-result object is not used.

For a non-empty rule, the second and any subsequent arguments to the callback are the values, in lexical order, of the symbols on the right hand side of the rule. If the action is for an empty rule, the per-parse-result object will be its only argument.

Every value action is expected to return a value. With one exception, this value is passed up to a parent node as an argument. The exception is the value for the start rule. The return value for the start rule becomes the value of the parse result.

Rules with no action specified for them take their semantics from the "default_action" named argument. If there is no default action for a grammar, rules with no action specified for them return a Perl "undef".

Short Form Rule Descriptors

    rules => [
        [ 'E', [qw/E Add E/],      'do_add' ],
        [ 'E', [qw/E Multiply E/], 'do_multiply' ],
        [ 'E', [qw/Number/], ],
    ],

The rule descriptors in the ambiguous example demonstrate the "short" or array form of rule descriptors. Array form rule descriptors are references to arrays. Here the elements are, in order, the "lhs" property, the "rhs" property, and the "action" property.

Marpa::PP::Recognizer::value

    my @values = ();
    while ( defined( my $ambiguous_value_ref = $ambiguous_recce->value() ) ) {
        push @values, ${$ambiguous_value_ref};
    }

When called more than once, the "Marpa::PP::Recognizer::value" method iterates through the parse results. For each call, it returns a reference to the value of a parse result. At the end of the iteration, after all parse results have been returned, "Marpa::PP::Recognizer::value" returns "undef". If there were no parse results, "Marpa::PP::Recognizer::value" returns "undef" the first time that it is called.

Why is it Called "Marpa"?

Blatant Plug