NAME

Marpa::Grammar - Marpa Grammar Objects

SYNOPSIS

my $grammar = Marpa::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'
            },
        ],
    }
);

$grammar->precompute();

DESCRIPTION

Marpa's grammar objects are created with the new method. Rules and other named arguments may be specified when the grammar is created, or later using the set method.

The precompute method adds data structures that the recognizer will need to the grammar object. After precomputation a grammar is "frozen". The set method and some tracing accessors may be called after precomputation, but no new rules may added and most other named arguments are no longer valid.

Symbol Names

Symbol names that end in right square brackets are reserved for Marpa's internal use. Any other valid Perl string is an acceptable symbol name.

Terminal Symbols

If a grammar has no empty rules, all symbols are terminals by default. Unlike most parser generators, Marpa will allow terminals to appear on the left hand side of rules. Marpa defines a terminal as a symbol which is valid as an input token symbol.

Marpa allows direct marking of terminals using the terminals named argument, and the terminal property. If any terminal is directly marked, only directly marked symbols will be terminals.

Marpa's behavior can be changed by unsetting the lhs_terminals named argument When the lhs_terminals named argument is unset, only symbols which do not appear on the LHS of a rule can be terminals. If no symbols are directly marked, Marpa will implicitly mark all the non-LHS symbols as terminals.

Marpa is stricter for grammars with empty rules -- it does not allow them to take the default. There is an efficiency hit whenever the LHS of an empty rule is a terminal as well, so Marpa does not allow this to happen by default -- the user has to be explicitly indicate that that is what she wants. If a grammar has any empty rules, it must either directly mark its terminals, or must unset the lhs_terminals named argument.

For most grammars, the default can be taken, in which case all symbols are terminals. One advantage of not taking the default is efficiency. Precomputation is faster for grammars which mark their terminals. Also, the recognizer checks that input tokens are terminals, so that being selective about which symbols are terminals and which are not can allow better input checking.

Summary

The following summarizes the logic which determines whether a symbol S is terminal. As a reminder, a symbol is said to be directly marked as a terminal if it is one of the symbols in the terminals named argument, or if the symbol has the terminal property set.

• If any symbol is directly marked as a terminal, then a symbol S is a terminal if and only if it is also directly marked as a terminal.

• If no symbol is directly marked as a terminal, and the lhs_terminals named argument is unset, then all non-LHS symbols are terminals.

• If the grammar contains no empty rules, no symbol is directly marked as a terminal, and the lhs_terminals named argument is set or left at its default, then all symbols are terminals.

• The only case not covered above is that in which a grammar contains one or more empty rules, no symbol is directly marked as a terminal, and the lhs_terminals named argument is set or left at its default. This is a fatal error.

Sequence Rules

It is very common in a grammar for one symbol to produce a repeating sequence. Marpa allows a shorthand for this: sequence rules. The RHS of a sequence rule will be repeated, as specified by the min rule property. In sequence rules the RHS must always be one symbol in length, and that symbol may not be a nullable symbol.

A rule is a sequence rule if the min rule property is defined. min can be 0 or 1, and specifies the minimum number of times that the sequence is allowed to repeat. As of this writing, the maximum number of repetitions is always infinite.

{ lhs => 'sequence', rhs => ['item'], min => 0 }

A min of zero indicates a sequence that repeats zero or more times. This is the equivalent of using the star quantifier ("item*") in the standard regular expression notation.

{ lhs => 'sequence', rhs => ['item'], min => 1 }

A min of one indicates a sequence that repeats one or more times. This is the equivalent of using the plus quantifier ("item+") in the standard regular expression notation.

Sequences can have a separator, specified with the separator rule property. By default, separation is Perl-style: trailing separators are allowed. In "proper" separation, a separator must actually separate two sequence items and therefore is not allowed after the last item of a sequence. If you prefer "proper" separation, you can set the proper rule property.

Advantages of Sequence Rules

You are never forced to use sequence rules, but it's usually better if you do. When a sequence is written as a sequence rule, Marpa optimizes it.

When a sequence is written using non-sequence rules, the semantics typically wind up being spread over two or three Perl closures. The semantic action for a sequence rule is a single Perl closure. Putting the semantics into a single Perl closure often results in simpler and more natural code. See the section on sequences in the semantics document.

Caveats

Marpa throws an exception if you try to use a nullable symbol as the right hand side of a sequence rule, or as the separator for a sequence rule. The ban on nullables in sequences only applies to sequences when they are written using sequence rules. Nothing prevents you from specifying a sequence of nullables using non-sequence rules. But usually there is no good reason to do this, and for efficiency reasons, it is a good thing to avoid.

To keep things simple, the right hand side of a sequence rule must be a single symbol. Of course, applications will often want to repeat sequences of multiple symbols. That is easy to do indirectly:

{ lhs => 'sequence', rhs => [qw(item)], min => 0 },
{ lhs => 'item', rhs => [qw(part1 part2)], },

CONSTRUCTOR

new

my $grammar = Marpa::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::Grammar::new returns a new Marpa grammar object or throws an exception. The arguments to Marpa::Grammar::new are references to hashes of named arguments. In each key/value pair of this hash, the hash key is the argument name and the hash value is the value of the named argument. The available named arguments are described below.

MUTATORS

precompute

$grammar->precompute();

The precompute method compiles data structures that the recognizer will need. It returns the grammar object or throws an exception.

set

The arguments to the set method are references to hashes of named arguments. The available named arguments are described below. set either returns true or throws an exception.

ACCESSORS

check_terminal

Returns a Perl true when its argument is the name of a terminal symbol. Otherwise, returns a Perl false. Not often needed, but a lexer may find this the most convenient way to determine if a symbol is a terminal.

TRACE ACCESSORS

show_AHFA

print $grammar->show_AHFA()
    or die "print failed: $ERRNO";

Returns a multi-line string listing the states of the AHFA with the LR(0) items, LR(0) NFA states, and the transitions for each. Not useful before the grammar is precomputed. Useful in debugging, but requires knowledge of Marpa internals.

show_problems

print $grammar->show_problems()
    or die "print failed: $ERRNO";

Returns a string describing any serious but non-fatal problems a grammar had in the precomputation phase. A serious problem is one that will prevent parsing. Warnings are not serious problems in this sense. If there were no serious problems, returns a string saying so. This method is not useful before precomputation.

In Marpa, most serious grammar problems are not immediately thrown as exceptions. This is because there can be a number of serious problems in a grammar, particularly one that is large or in an early draft. If each serious problem caused an immediate exception, the user would have to fix them one at a time -- very tedious.

Usually the application does not call this method directly. The recognizer throws an exception when the user attempts to create a parse from a grammar with serious problems. When that happens, the string returned by show_problems is part of the error message.

show_rules

print $grammar->show_rules()
    or die "print failed: $ERRNO";

Returns a string listing the rules. Each rule is shown with comments which indicate either rule properties or internal settings. show_rules is often useful in debugging grammars. Much of its information requires no knowledge of the Marpa internals to interpret.

Marpa does extensive rewriting of its grammars, and both the original rules and the rewritten rules appear in the show_rules list. When a rule is rewritten, the original rule is often not used. In that case, "!used" will be one of the comments for the original rule. The "!used" comment also marks rules not used for reasons other than rewrites. For example, inaccessible and unproductive rules are also marked "!used".

The "discard_sep" comment indicates that the rule discards separators This is only relevant in sequence rules. Other comments indicate whether rules were nullable, unproductive, inaccessible, or empty.

The vlhs, vrhs and real comments show rule settings relevant in tracking "virtual" internal symbols. Details on virtual symbols can be found in the implementation document.

show_symbols

print $grammar->show_symbols()
    or die "print failed: $ERRNO";

Returns a string listing the symbols, along with comments indicating whether they were terminal, nulling, nullable, unproductive or inaccessible. Also shown is a list of rules with that symbol on the left hand side, and a list of rules which have that symbol anywhere on the right hand side. Often useful and much of the information requires no knowledge of the Marpa internals to interpret.

NAMED ARGUMENTS

action_object

The action_object named argument specifies a class name to be used in resolving action names to Perl closures. If a new constructor is defined in the action_object package, that constructor is used to create the per-parse variables. Details are in the document on semantics.

actions

actions => 'My_Actions',

The actions named argument specifies the Perl package that Marpa will use when resolving action names to Perl closures. If both an actions named argument and an action_object named argument are specified, the package from the actions named argument is the only one used to resolve action names. The actions package is treated only as a package, and not as a class. Any new constructor in the actions package is ignored. Details are given in the document on semantics.

default_action

default_action => 'first_arg',

The default_action named argument specifies the semantic action for rules without an action property. Details are given in the document on semantics.

default_null_value

The default_null_value named argument specifies the null value for symbols without a null_value property. Details are given in the document on semantics.

inaccessible_ok

The value must be a reference to an array of symbol names. By default, Marpa warns if a symbol is inaccessible, but the warning is suppressed for any symbol named in the array. Setting the inaccessible_ok named argument after grammar precomputation is useless, and itself results in a warning.

Inaccessible symbols are symbols which cannot be derived from the start symbol, and which therefore can never be part of a successful parse. Inaccessible symbols often indicate errors in grammar design. But a user may have plans for these symbols, may wish to keep them as notes, or may simply wish to deal with them later.

infinite_action

Takes as its value a string specifying what Marpa should do if it discovers that its grammar is infinitely ambigious. The value must be one of "fatal", "warn" or "quiet". A grammar is infinitely ambiguous if there is some input for which it produces an endless number of parses.

If the value is "fatal", Marpa throws an exception when it encounters an infinitely ambiguous grammar. This is the default and will usually be what the user wants. In most cases, an infinitely ambiguous grammar is simply a mistake.

"quiet" indicates that the user wants to allow infinitely ambiguous grammars. "warn" indicates that the user wants to allow infinitely ambiguous grammars, but wants a warning message to be printed to the trace file handle.

lhs_terminals

The value of the lhs_terminals named argument is a Boolean. If true, symbols which appear on the LHS of a rule are allowed to be terminals. lhs_terminals is true by default.

If lhs_terminals is unset and no symbols are directly marked as terminals, Marpa will mark all non-LHS symbols as terminals. If any symbol is directly marked, all terminals must be directly marked. If lhs_terminals is unset, but some symbols are directly marked, it is a fatal error for a terminal to appear on the LHS of a rule. For more, see the discussion of terminals above.

rules

The value of the rules named argument is a reference to an array of rule descriptors. The rules named argument may be specified multiple times, adding new rules to the grammar each time. New rules may be added until the grammar is precomputed. The format of rule descriptors is explained below.

start

start => 'Expression',

The value of the start named argument must be a symbol name. It will be used as the start symbol for the grammar. The start named argument is required.

strip

The value is a Boolean. If true, after precomputation Marpa "strips" the grammar of all data structures not needed for further processing. This saves space and time. Stripping is the default behavior.

If strip is set to false, the grammar object is not stripped. The data that would have been stripped remains available for tracing and debugging.

symbols

The value of the symbols named arguments must be a reference to a hash. In each key/value pair of this hash, the hash key is the symbol property name and the hash value is the symbol descriptor. Symbol descriptors are described below.

Note that the value of symbols named argument is a hash, but the value of the rules named argument is an array. This is because symbol names make convenient hash keys. For rules, there is no equally natural choice for a hash key.

terminals

The value of the terminals named argument must be a reference to an array of symbol names. Specifying symbols in a terminals named argument is one way of directly marking them as terminals. See the discussion of terminals above.

trace_file_handle

The value is a file handle. Trace output and warning messages go to the trace file handle. By default the trace file handle is STDERR.

trace_rules

Traces rules as they are added to the grammar. Useful, but you will usually prefer the show_rules method. Does not require knowledge of Marpa internals.

A trace_rules setting becomes effective within the named argument hash which sets it. A trace message warns the user if he turns on rule tracing when rules have already been added.

unproductive_ok

The value must be a reference to an array of symbol names. By default, Marpa warns if a symbol is unproductive, but the warning is suppressed for any symbol named in the array. Setting the unproductive_ok named argument after grammar precomputation is useless, and itself results in a warning.

Unproductive symbols are symbols which can never derive a sentence. (A sentence is a string of zero or more terminals.) That means that unproductive symbols can never be part of a successful parse. Unproductive symbols often indicate errors in grammar design. But a user may have plans for these symbols, may wish to keep them as notes, or may simply wish to deal with them later.

warnings

The value is a boolean. Warnings are written to the trace file handle. By default, warnings are on. Usually, an application will want to leave them on. If warnings are turned off, turning them back on after grammar precomputation is useless, and itself results in a warning.

RULE DESCRIPTORS

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'
    },
],

Rule Descriptors as Hashes

The long form descriptor of a rule is a reference to a hash of rule properties. In each key/value pair of this hash, the hash key is the rule property name and the hash value is the value of that property.

action

The value of the action rule property is a string which specifies the semantics for the rule. For details, see the document on semantics.

The semantics of nulling symbols are dealt with on a per-symbol basis, rather than a per-rule basis. For this reason the action rule property is useless for empty rules. An exception is thrown if an action property is defined for an empty rule.

keep

Separators in sequence rules are usually not semantically significant. By default, Marpa throws away separators during parse tree traversal and before node evaluation time, so that the semantic actions do not see the separators.

If the value of the keep rule property is a Perl true, Marpa keeps separators. This allows the semantic actions to examine them. The downside is that the work of distinguishing sequence separators from sequence items is pushed into the semantic actions. For details about the semantics, see the document on semantics.

lhs

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 lhs plays a role in the semantics. If no action rule property is defined, Marpa looks in either the grammar's actions package or the grammar's action_object for a Perl closure that has the name of the lhs symbol. For details, see the document on semantics.

min

min must be 0, 1, or undefined. If min is 0 or 1, the rule is a sequence rule. If min is undefined, the rule is an ordinary, non-sequence rule.

Only one symbol, called the sequence item, is allowed on the right hand side of a sequence rule. The sequence item may not be a nullable symbol. The input will be required to match the sequence item at least min times and will be allowed to match the sequence item an unlimited number of times.

proper

By default, sequence rules with separators allow trailing separators, Perl-style. If the proper rule property is a Perl true, "proper" separation is enforced. In proper separation, separation must actually separate sequence items, and trailing separators are not allowed.

ranking_action

The ranking_action rule property affects the order in which Marpa returns parse results for an ambiguous parse. For details see the document on Marpa's semantics

rhs

The value of the rhs property is a reference to an array of strings containing the 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. A rule is also empty if there is no rhs specifier in its descriptor.

separator

Any sequence rule may have a separator defined. The value must be a symbol name. By default, Marpa allows trailing separators. This is the usual style in Perl. The separator must not be a nullable symbol.

Rule Descriptors as Arrays

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

Rule descriptors may be given in "short form" -- as a reference to an array. The elements of the array, in order, are the lhs property, the rhs property, and the action property. The last two are optional. Omission of an optional property in a short form descriptor has the same effect that omitting the same optional property would have in the long form.

Duplicate Rules

Marpa throws an exception if a duplicate rule is added. For non-sequence rules, two rules are considered duplicates if

• Both rules have the same left hand symbol.

• Both rules have the same right hand symbols in the same order.

For two sequence rules to be considered duplicates, the same requirements hold as for non-sequence rules, but there is also an additional requirement. For two sequence rules to be considered duplicates, one of the following two conditions must hold

• Both rules have no separator.

• Both rules have identical separators.

SYMBOL DESCRIPTORS

symbols        => {
    L => { null_value => 'null L' },
    R => { null_value => 'null R' },
    A => { null_value => 'null A' },
    B => { null_value => 'null B' },
    X => { null_value => 'null X', terminal => 1 },
    Y => { null_value => 'null Y', terminal => 1 },
},

A symbol descriptor is a hash. In the key/value pairs of this hash, the hash key is the symbol property name and the hash value is the value of that property. The available symbol properties are as follows:

null_value

The null_value symbol property specifies the null value of its symbol. Details are given in the document on semantics.

terminal

A boolean. If true, it marks the symbol as a terminal. If false, it unmarks the symbol as a terminal. For details, see the section on terminals.

LICENSE AND COPYRIGHT

Copyright 2007-2010 Jeffrey Kegler, all rights reserved. Marpa is free software under the Perl license. For details see the LICENSE file in the Marpa distribution.

cpanm Marpa

perl -MCPAN -e shell
install Marpa