Marpa::R2::NAIF::Recognizer - NAIF recognizers


    my $recce = Marpa::R2::Recognizer->new( { grammar => $grammar } );
    $recce->read( 'Number', 42 );
    $recce->read( 'Number', 1 );
    $recce->read( 'Number', 7 );


This document describes recognizers for Marpa's named argument interface (NAIF). If you are a beginner, or are not sure which interface you are interested in, or do not know what the NAIF interfaces is, you probably are looking for the document on recognizers for the SLIF interface.

To create a recognizer object, use the new method.

To read input, use the read method.

To evaluate a parse tree, based on the input, use the value method.

Token streams

By default, Marpa uses the token-stream model of input. The token-stream model is standard -- so standard the most documents about parsing do not bother to describe it. In the token-stream model, each read adds a token at the current location, then advances the current location by one. The location before any input is numbered 0 and if N tokens are parsed, they fill the locations from 1 to N.

This document will describe only the token-stream model of input. Marpa allows other models of the input, but their use requires special method calls, which are described in the document on alternative input models.



    my $recce = Marpa::R2::Recognizer->new( { grammar => $grammar } );

The new method creates a recognizer object. The new method either returns a new recognizer object or throws an exception.

The arguments to the new method are references to hashes of named arguments. In each key/value pair of these hashes, the key is the argument name, and the hash value is the value of the argument. The named arguments are described below.



    my $is_symbol_a_terminal = $recce->check_terminal('Document');

Returns a Perl true when its argument is the name of a terminal symbol. Otherwise, returns a Perl false. Not often needed.


    my @expected_symbols =
        map { $_->[1]; }
        grep { $_->[0] eq 'SYMBOL_EXPECTED' } @{ $recce->events() };

Returns a reference to an array of the events from the last "read()" method call. Each element of the array is a subarray or 1 or 2 elements. The first element of the subarray is the name of an event type, as described in "Recognizer events". The second element is the event value of the event, where that is applicable. For more detail, see "Recognizer events".


        $recce->exhausted() and die 'Recognizer exhausted';

The exhausted method returns a Perl true if parsing in a recognizer is exhausted, and a Perl false otherwise. Parsing is exhausted when the recognizer will not accept any further input. By default, a recognizer event occurs if parsing is exhausted. An attempt to read input into an exhausted parser causes an exception to be thrown. The recognizer event and the exception are all that many applications require, but this method allows the recognizer's exhaustion status to be discovered directly.


    my $latest_earley_set = $recce->latest_earley_set();

Return the location of the latest (in other words, the most recent) Earley set. In the places where it is most often needed, the latest Earley set is the default, and there is usually no need to request the explicit value of the latest Earley set.


Given the location (Earley set ID) as its argument, returns an array that describes the parse progress at that location. Details on progress reports can be found in their own document.


    my $terminals_expected = $recce->terminals_expected();

Returns a reference to a list of strings, where the strings are the names of the terminals acceptable at the current location. In the default input model, the presence of a terminal in this list means that terminal will be acceptable in the next read method call. This is highly useful for Ruby Slippers parsing.



    $recce->expected_symbol_event_set( 'endmark', 1 );

Marpa can generate a recognizer event when a symbol is expected at the current earleme. This method takes a symbol name as its first argument, and turns the expected-symbol event for that symbol on or off, according to whether its second argument is 1 or 0. Always succeeds or throws an exception.

Events can occur at location 0 -- when the recognizer is first created. However, the event setting of expected_symbol_event_set() cannot have an effect until after the first token is read -- after location 0. In cases where this is an issue, the "event_if_expected" named argument of the "new()" method can be used to set an expected-symbol event.


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

The read method reads one token at the current parse location. It then advances the current location by 1.

read takes two arguments: a token name and a token value. The token name is required. It must be the name of a valid terminal symbol. The token value is optional. It defaults to a Perl undef. For details about terminal symbols, see "Terminal symbols" in Marpa::R2::NAIF::Grammar.

The parser may accept or reject the token. If the parser accepted the token, the read method returns the number of recognizer events that occurred during the read. For more about events, see "Recognizer events".

Marpa may reject a token because it is not one of those acceptable at the current location. When this happens, read returns a Perl undef. A rejected token need not end parsing -- it is perfectly possible to retry the read call with another token. This is, in fact, an important technique in Ruby Slippers parsing. For details, see the section on Ruby Slippers parsing.

For other failures, including an attempt to read a token into an exhausted parser, Marpa throws an exception.


    $recce->set( { max_parses => 10, } );

The set method's arguments are references to hashes of named arguments. The set method can be used to set or change named arguments after the recognizer has been created. Details of the named arguments are below.


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

Because Marpa parses ambiguous grammars, every parse is a series of zero or more parse trees. There are zero parse trees if there was no valid parse of the input according to the grammar.

The value method call evaluates the next parse tree in the parse series, and returns a reference to the parse result for that parse tree. If there are no more parse trees, the value method returns undef.


        $recce->set( { end => $loc, max_parses => 999, } );

The reset_evaluation() method ends a parse series, and starts another. It can be used to "restart" the parse series. Restarting the parse series with the reset_evaluation() method allows the application to specify new values for the closures, end and ranking_method named arguments. Once a parse series is underway, these values cannot be changed.

The most common use for reset_evaluation() method is to parse a single input stream at different end points. This can also be done by creating a new recognizer and re-reading the input from the beginning, but it is much more efficient to evaluate a single recognizer run several times, using different parse end locations. After the parse is restarted using the reset_evaluation() method, the recognizer's set() method and its end named argument can be used to change the parse end location.

Trace accessors


    print $recce->show_progress()
        or die "print failed: $ERRNO";

Returns a string describing the progress of the parse. With no arguments, the string contains reports for the current location. With a single integer argument N, the string contains reports for location N. With two numeric arguments, N and M, the arguments are interpreted as a range of locations and the returned string contains reports for all locations in the range. ("Location" as referred to in this section, and elsewhere in this document, is what is also called the Earley set ID.)

If an argument is negative, -N, it indicates the Nth location counting backward from the furthest location of the parse. For example, if 42 was the furthest location, -1 would be location 42 and -2 would be location 41. For example, the method call $recce->show_progress(-3, -1) returns reports for the last three locations of the parse. The method call $recce->show_progress(0, -1) will print progress reports for the entire parse.

show_progress is Marpa's most powerful tool for debugging application grammars. It can also be used to track the progress of a parse or to investigate how a parse works. A much fuller description, with an example, is in the document on debugging Marpa grammars.

Named arguments

The recognizer's named arguments are accepted by its new and set methods.


The value of closures named argument must be a reference to a hash. In each key/value pair of this hash, the key must be an action name. The hash value must be a CODE ref. The closures named argument is not allowed once evaluation has begun.

When an action name is a key in the closures named argument, the usual action resolution mechanism of the semantics is bypassed. One common use of the closures named argument is to allow anonymous subroutines to be semantic actions. For more details, see the document on semantics.


The end named argument specifies the parse end location. The default is for the parse to end where the input did, so that the parse returned is of the entire input. The end named argument is not allowed once evaluation has begun. "Location" as referred to here and elsewhere in this document is what is also called an Earley set ID.


The value of the event_if_expected named argument must be a reference to an array of symbol names. Expected-symbol events will be turned on for those symbol names. Expected-symbol events may be turned off (or back on) using the "expected_symbol_event_set()" method. The advantage of the event_if_expected named argument is that it takes effect as soon as the recognizer is created, while events set using the "expected_symbol_event_set()" method cannot occur until after the first token is read.


The new method is required to have a grammar named argument. Its value must be a precomputed Marpa grammar object. The grammar named argument is not allowed anywhere else.


If non-zero, causes a fatal error when that number of parse results is exceeded. max_parses is useful to limit CPU usage and output length when testing and debugging. Stable and production applications may prefer to count the number of parses, and take a less Draconian response when the count is exceeded.

The value must be an integer. If it is zero, there will be no limit on the number of parse results returned. The default is for there to be no limit.


The value must be a string: one of "none", "rule", or "high_rule_only". When the value is "none", Marpa returns the parse results in arbitrary order. This is the default. The ranking_method named argument is not allowed once evaluation has begun.

The "rule" and "high_rule_only" ranking methods allows the user to control the order in which parse results are returned by the value method, and to exclude some parse results from the parse series. For details, see the document on parse order.


The too_many_earley_items argument is optional. If specified, it sets the Earley item warning threshold. If an Earley set becomes larger than the Earley item warning threshold, a recognizer event is generated, and a warning is printed to the trace file handle.

Marpa parses from any BNF, and can handle grammars and inputs which produce large Earley sets. But parsing that involves large Earley sets can be slow. Large Earley sets are something most applications can, and will wish to, avoid.

By default, Marpa calculates an Earley item warning threshold based on the size of the grammar. The default threshold will never be less than 100. If the Earley item warning threshold is set to 0, no recognizer event is generated, and warnings about large Earley sets are turned off.


The trace_actions named argument is a boolean. If the boolean value is true, Marpa prints tracing information as it resolves action names to Perl closures. A boolean value of false turns tracing off, which is the default. Traces are written to the trace file handle.


The value is a file handle. Traces and warning messages go to the trace file handle. By default the trace file handle is inherited from the grammar used to create the recognizer.


Very handy in debugging, and often useful even when the problem is not in the lexing. The value is a trace level. When the trace level is 0, tracing of terminals is off. This is the default.

At a trace level of 1 or higher, Marpa produces a trace message for each terminal as it is accepted or rejected by the recognizer. At a trace level of 2 or higher, the trace messages include, for every location, a list of the terminals expected. In practical grammars, output from trace level 2 can be voluminous.


The trace_values named argument is a numeric trace level. If the numeric trace level is 1, Marpa prints tracing information as values are computed in the evaluation stack. A trace level of 0 turns value tracing off, which is the default. Traces are written to the trace file handle.


The value is a boolean. Warnings are written to the trace file handle. By default, the recognizer's warnings are on. Usually, an application will want to leave them on.

Recognizer events

The recognizer's read() method can generate events. To access events, use the recognizer's "events()" method.

The EARLEY_ITEM_THRESHOLD and The EXHAUSTED events are enabled by default. Events optionally have an "event value", as specified in the description of each event. The following events are possible.


The Earley item threshold was exceeded. For more about the Earley item warning threshold, see "too_many_earley_items". No event value is defined for this event. This event is enabled by default.


"Exhaustion" means that the next read call must fail, because there is no token that will be acceptable to it. More details on "exhaustion" are in a section below. No event value is defined for this event. This event is enabled by default.


A "symbol expected" event means that a symbol is expected at that point. The event value of this event is the symbol whose expectation caused the event. This event is disabled by default. For details, see "expected_symbol_event_set()".

Parse exhaustion

A parse is exhausted when it will accept no more input. An exhausted parse is not necessarily a failed parse. Grammars are often written so that once they "find what they are looking for", no further input is acceptable. Grammars of that kind become exhausted when they succeed.

By default, a recognizer event occurs whenever the parse is exhausted. An application can also check for exhaustion explicitly, using the recognizer's exhausted method.

Ruby Slippers parsing

    $recce = Marpa::R2::Recognizer->new( { grammar => $grammar } );

    my @tokens = (
        [ 'Number', 42 ],
        ['Multiply'], [ 'Number', 1 ],
        ['Add'],      [ 'Number', 7 ],

    TOKEN: for ( my $token_ix = 0; $token_ix <= $#tokens; $token_ix++ ) {
        defined $recce->read( @{ $tokens[$token_ix] } )
            or fix_things( $recce, $token_ix, \@tokens )
            or die q{Don't know how to fix things};

Marpa is able to tell the application which symbols are acceptable as tokens at the next location in the parse. The terminals_expected method returns the list of tokens that will be accepted by the next read. The application can use this information to change the input "on the fly" so that it is acceptable to the parser.

An application can also take a "try it and see" approach. If an application is not sure whether a token is acceptable or not, the application can try to read the dubious token using the read method. If the token is rejected, the read method call will return a Perl undef. At that point, the application can retry the read with a different token.

An example

Marpa's HTML parser, Marpa::HTML, is an example of how Ruby Slippers parsing can help with a non-trivial, real-life application. When a token is rejected in Marpa::HTML, it changes the input to match the parser's expectations by

  • Modifying existing tokens, and

  • Creating new tokens.

The second technique, the creation of new "virtual" tokens, is used by Marpa::HTML to deal with omitted start and end tags. The actual HTML grammar that Marpa::HTML uses takes an oversimplified view of the HTML -- it assumes, even when the HTML standards do not require it, that start and end tags are always present. For most HTML files of interest, this assumption will be contrary to fact.

Ruby Slippers parsing is used to make the grammar's over-simplistic view of the world come true for it. Whenever a token is rejected, Marpa::HTML looks at the expected tokens list. If it sees that a start or end tag is expected, Marpa::HTML creates a token for it -- a completely new "virtual" token that gives the parser exactly what it expects. Marpa::HTML then resumes input at the point in the original input stream where it left off.

Copyright and License

  Copyright 2018 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