NAME

Marpa::R3::Semantics::Null - How the SLIF evaluates null rules and symbols

Overview

In Marpa parses, rules and symbols can be nulled -- in other words they can derive the zero-length, or null, string. Which symbols can be, or are, nulled, depends on the grammar and the input. When a symbol or rule is not nulled, the symbol is said to be visible.

Even the start symbol can be nulled, in which case the entire parse derives the null string. A parse in which the start symbol is nulled is called a null parse.

When evaluating a parse, nulled rules and symbols are assigned values as described in the semantics document. This document provides additional detail on the assignment of values to nulled symbols.

Description

Null values come from rules

Nulled subtrees are pruned back to their topmost symbol. Lexemes are never nulled, so a nulled symbol is always the LHS of a rule instance, and the action is determined from the rule alternative.

A complication arises if the symbol appears on the LHS of more than one nullable rule alternative. Because the symbol is nulled, the input is no help in determining which rule alternative to use. The rule alternative whose semantics are used for a nulled symbol is determined as follows:

• If all nullable rule alternatives have the same semantics, that semantics is used.

• If one of the nullable rule alternatives is empty (that is, has a zero-length RHS), then the empty alternative's semantics are used.

• In the remaining case, two or more of the rule alternatives have different action names, but none of the alternatives has a zero-length RHS. When this happens, Marpa throws an exception. One easy way to fix the issue, is to add an empty rule with the intended semantics.

In determining whether the semantics of two nullable rule alternatives is "the same", the blessing is taken into account. Two rule alternatives are considered to have different semantics if they are blessed differently.

The "lost" semantics of the non-topmost symbols and rules of null subtrees are usually not missed. Nulled subtrees cannot contain input, and therefore do not contain token symbols. So no token values are lost when nulled subtrees are pruned. As bushy as a null subtree might be, all of its symbols and rules are nulled.

Since nulled symbols and rules correspond to zero-length strings, so we are literally dealing here with the "semantics of nothing". In theory the semantics of nothing can be arbitrarily complex. In practice it should be possible to keep them simple.

Example

As already stated, Marpa prunes every null subtree back to its topmost null symbol. Here is an example:

sub do_L {
    my (undef, $values) = @_;
    return 'L(' . ( join q{;}, map { $_ // '[ERROR!]' } @{$values} ) . ')';
}

sub do_R {
    return 'R(): I will never be called';
}

sub do_S {
    my (undef, $values) = @_;
    return 'S(' . ( join q{;}, map { $_ // '[ERROR!]' } @{$values} ) . ')';
}

sub do_X { return 'X(' . $_[1]->[0] . ')'; }
sub do_Y { return 'Y(' . $_[1]->[0] . ')'; }
sub null_A { return 'null A'; }
sub null_B { return 'null B'; }
sub null_L { return 'null L'; }
sub null_R { return 'null R'; }
sub null_X { return 'null X'; }
sub null_Y { return 'null Y'; }

my $slg = Marpa::R3::Scanless::G->new(
    {   source => \<<'END_OF_DSL',
:start ::= S
S ::= L R     action => do_S
L ::= A B X   action => do_L
L ::=         action => null_L
R ::= A B Y   action => do_R
R ::=         action => null_R
A ::=         action => null_A
B ::=         action => null_B
X ::=         action => null_X
X ::= 'x'     action => do_X
Y ::=         action => null_Y
Y ::= 'y'     action => do_Y
END_OF_DSL
    }
);

my $slr = Marpa::R3::Scanless::R->new(
    {   grammar           => $slg,
        semantics_package => 'main',
    }
);

$slr->read( \'x' );

If we write the unpruned parse tree in pre-order, depth-first, indenting children below their parents, we get something like this:

0: Visible Rule: S := L R
     1: Visible Rule L := A B X
         1.1: Nulled Symbol A
         1.2: Nulled Symbol B
         1.3: Token, Value is 'x'
     2: Nulled Rule, Rule R := A B Y
         2.1: Nulled Symbol A
         2.2: Nulled Symbol B
         2.3: Nulled Symbol Y

In this example, five symbols and a rule are nulled. The rule and three of the symbols are in a single subtree: 2, 2.1, 2.2 and 2.3. Marpa prunes every null subtree back to its topmost symbol, which in this case is the LHS of the rule numbered 2.

The pruned tree looks like this

0: Visible Rule: S := L R
     1: Visible Rule L := A B X
         1.1: Nulled Symbol A
         1.2: Nulled Symbol B
         1.3: Token, Value is 'x'
     2: LHS of Nulled Rule, Symbol R

Nulled nodes 1.1, 1.2 and 2 were all kept, because they are topmost in their nulled subtree. All the other nulled nodes were discarded.

Here is the output:

S(L(null A;null B;X(x));null R)

In the output we see

• The null value for symbol 1.1: "null A". This comes from the empty rule for A.

• The null value for symbol 1.2: "null B". This comes from the empty rule for B.

• The token value for symbol 1.3: "x".

• An application of the rule evaluation closure for the rule L := A B X.

• The null value for rule 2: "null R". This comes from the empty rule for R.

• An application of the rule evaluation closure for the rule S := L R

We do not see any output for symbols 2.1 (A), 2.2 (B), or 2.3 (Y) because they were not topmost in the pruned subtree. We do not see an application of the rule evaluation closure for rule R := A B Y, because there is an empty rule for R, and that takes priority.

COPYRIGHT AND LICENSE

Marpa::R3 is Copyright (C) 2016, Jeffrey Kegler.

This module is free software; you can redistribute it and/or modify it
under the same terms as Perl 5.10.1. For more details, see the full text
of the licenses in the directory LICENSES.

This program is distributed in the hope that it will be
useful, but without any warranty; without even the implied
warranty of merchantability or fitness for a particular purpose.