View on
MetaCPAN is shutting down
For details read Perl NOC. After June 25th this page will redirect to
Jeffrey Kegler > Marpa-R2 > Marpa::R2::Semantics::Null


Annotate this POD



Open  0
View/Report Bugs
Source   Latest Release: Marpa-R2-5.043_043


Marpa::R2::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:

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 {
        return 'L(' . ( join q{;}, map { $_ // '[ERROR!]' } @_ ) . ')';

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

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

    sub do_X { return 'X(' . $_[1] . ')'; }
    sub do_Y { return 'Y(' . $_[1] . ')'; }

    ## no critic (Variables::ProhibitPackageVars)
    our $null_A = 'null A';
    our $null_B = 'null B';
    our $null_L = 'null L';
    our $null_R = 'null R';
    our $null_X = 'null X';
    our $null_Y = 'null Y';
    ## use critic

    my $slg = Marpa::R2::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

    my $slr = Marpa::R2::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

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 ^

  Copyright 2015 Jeffrey Kegler
  This file is part of Marpa::R2.  Marpa::R2 is free software: you can
  redistribute it and/or modify it under the terms of the GNU Lesser
  General Public License as published by the Free Software Foundation,
  either version 3 of the License, or (at your option) any later version.

  Marpa::R2 is distributed in the hope that it will be useful,
  but WITHOUT ANY WARRANTY; without even the implied warranty of
  Lesser General Public License for more details.

  You should have received a copy of the GNU Lesser
  General Public License along with Marpa::R2.  If not, see
syntax highlighting: