# Copyright 2013 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 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU # 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 # http://www.gnu.org/licenses/. =head1 NAME Marpa::R2 - Release 2 of Marpa =head1 Synopsis =for Marpa::R2::Display name: Landing page synopsis normalize-whitespace: 1 use Marpa::R2; my $dsl = <<'END_OF_DSL'; :default ::= action => ::first :start ::= Expression Expression ::= Term Term ::= Factor | Term '+' Term action => do_add Factor ::= Number | Factor '*' Factor action => do_multiply Number ~ digits digits ~ [\d]+ :discard ~ whitespace whitespace ~ [\s]+ END_OF_DSL my $grammar = Marpa::R2::Scanless::G->new( { source => \$dsl } ); my $recce = Marpa::R2::Scanless::R->new( { grammar => $grammar, semantics_package => 'My_Actions' } ); my $input = '42 * 1 + 7'; $recce->read( \$input ); my $value_ref = $recce->value; my $value = $value_ref ? ${$value_ref} : 'No Parse'; sub My_Actions::do_add { my ( undef, $t1, undef, $t2 ) = @_; return $t1 + $t2; } sub My_Actions::do_multiply { my ( undef, $t1, undef, $t2 ) = @_; return $t1 * $t2; } =for Marpa::R2::Display::End =head1 Description =head2 Overview Marpa parses any language whose grammar can be written in BNF. That includes recursive grammars, ambiguous grammars, infinitely ambiguous grammars and grammars with useless or empty productions. Marpa does both left- and right-recursion in linear time -- in fact if a grammar is in any class currently in practical use, Marpa will parse it in linear time. This document centers around a short tutorial of the Scanless interface (SLIF). This is the interface most suitable for beginners. The SLIF is the most suitable interface for most advanced uses as well. =head1 A simple calculator The synopsis shows the code for an extremely simple calculator. It handles only addition and multiplication of integers. The sections which follow explain, line by line, how it works. The explanation will assume that the reader understands BNF and the basics of grammars -- what rules are, what symbols are, what the start symbol of a grammar is, etc. =head2 Marpa::R2::Scanless::G::new =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 my $dsl = <<'END_OF_DSL'; :default ::= action => ::first :start ::= Expression Expression ::= Term Term ::= Factor | Term '+' Term action => do_add Factor ::= Number | Factor '*' Factor action => do_multiply Number ~ digits digits ~ [\d]+ :discard ~ whitespace whitespace ~ [\s]+ END_OF_DSL my $grammar = Marpa::R2::Scanless::G->new( { source => \$dsl } ); =for Marpa::R2::Display::End The code first creates a new SLIF grammar. SLIF grammars are L|Marpa::R2::Scanless::G> objects. They are created with the L constructor. The arguments to L 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. In the example, there is only one named argument to the SLIF grammar constructor: C. The value of C must be a reference to a string in the L (DSL). In this example, the DSL consists of several rules and pseudo-rules. =head2 The default pseudo-rule =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 :default ::= action => ::first =for Marpa::R2::Display::End The default pseudo-rule specifies (as you might suspect) defaults for the other rules. Here it sets the default semantics to C<::first>. What this means we'll explain when we discuss L. =head2 The start pseudo-rule =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 :start ::= Expression =for Marpa::R2::Display::End The C pseudo-rule is is required. Its right hand side (RHS) is the grammar's start symbol. =head2 A G1 rule Next follows a G1, or structural rule. Structural rules are the kinds of rules typically seen in BNF -- they describe the symbols which provide the structure of the grammar, but leave out details of whitespace. The SLIF also handles the lexical details in this example, but it does it via L0 rules, which we will see shortly. =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 Expression ::= Term =for Marpa::R2::Display::End As is normal for BNF rules, this rule consists of a left hand side symbol ("C"), the BNF operator ("C<::=>") and a series of right hand side (RHS) symbols. There is always exactly one left hand side (LHS) symbol. There may be any number of RHS symbols. In the case of an empty rule, the number of RHS symbols would be zero. In this rule, there is one RHS symbol, "C". The BNF operator ("C<::=>") is what makes this rule a G1 (structural) rule. Later we will see lexical rules, which will use the match operator ("C<~>"). =head2 More complicated G1 rules =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 Term ::= Factor | Term '+' Term action => do_add =for Marpa::R2::Display::End This rule says that a C may be one of two alternatives: either a C or two C's separated by an addition operator. The second alternative has an C adverb associated with it. We'll explain its purpose shortly. Immediately following is another G1 rule defining a C. It is very similar in form to the one for C. =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 Factor ::= Number | Factor '*' Factor action => do_multiply =for Marpa::R2::Display::End =head2 L0 rules The structural rules define the high-level structure of the grammar, and ignore details of whitespace, comments, etc. Now we look at how the low-level, lexical issues are handled. This very simple calculator language does not allow comments, but it does define whitespace. =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 :discard ~ whitespace whitespace ~ [\s]+ =for Marpa::R2::Display::End The C<:discard> rule is a pseudo-rule, which tells Marpa to use whatever it matches to separate G1 symbols, but otherwise to ignore it -- to "discard" it. C is defined in the next rule as a sequence of one or more spaces. Note the match operator ("C<~>") in the rule defining whitespace. It tells Marpa that this rule is lexical and should be interpreted exactly as written, character by character. The C rule is a special kind of rule in two respects. First, its RHS is following by a quanitifer ("C<+>"), which makes it a sequence rule. Aside from the quantifier, sequence rules may only have a single symbol or character class on their RHS. The plus quanitifer ("C<+>") means a sequence of one or more items. The star quanitifer ("C<*>") is also allowed, and it indicates a sequence of zero or more items. The whitespace items are defined by a character class: C<[\s]>. Marpa supports the same character classes, and the same character class syntax, as Perl does. The next pair of L0 rules define the C symbol =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 Number ~ digits digits ~ [\d]+ =for Marpa::R2::Display::End The above two rules say that a C is a sequence of one or more digits. C is a lexeme -- a G1 symbol which is defined and recognized at the lexical (L0) level. In this example, there are three other lexemes: C, and the addition and multiplication operators. We've already looked at the C lexeme, which will be discarded without being seen by G1. The addition and multiplication operators were defined with single quoted strings in the G1 rules. As a reminder, here's the rule for C again: =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 Term ::= Factor | Term '+' Term action => do_add =for Marpa::R2::Display::End In the above rule, the single-quoted string C<'+'> implicitly defines a L0 lexeme. Something similar happens with the C<'*'> string in the rule for a C. The SLIF's lexer mostly "does what you mean". While the input is being read, it looks for all lexemes defined in the DSL. The search is a longest tokens match -- if mulitple tokens could match, the longest match is the winner. Marpa tolerates ambiguity, so one feature special to Marpa is that it is a longest B match -- if more than one token is longest, all of them are considered in the parse. The logic of SLIF lexing is described with more precision in the L. =head2 Marpa::R2::Scanless::R::new =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 my $recce = Marpa::R2::Scanless::R->new( { grammar => $grammar, semantics_package => 'My_Actions' } ); =for Marpa::R2::Display::End C creates a new SLIF recognizer. Its arguments are references to hashes of named arguments. In this example the first named argument is the required argument: "C". The value of the C named argument must be a Marpa::R2 SLIF grammar. The second argument is optional, but you will use it frequently. The "C" named argument tells Marpa in which Perl package to look for the closures implementing the semantics for this grammar. We will talk more about this below. =head2 Marpa::R2::Scanless::R::read =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 my $input = '42 * 1 + 7'; $recce->read( \$input ); =for Marpa::R2::Display::End To parse a string, we use the C method. In its simplest form, as here, the C takes a reference to a string containing the input stream as its argument. =head2 Marpa::R2::Scanless::R::value =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 my $value_ref = $recce->value; my $value = $value_ref ? ${$value_ref} : 'No Parse'; =for Marpa::R2::Display::End The C method returns a reference to the parse result's value, if there was a parse result. If there was no parse result, C returns C. We have yet to describe how the Marpa SLIF computes the value of a parse. In fact, up to this point, we have been skipping everything that had to do with semantics. Now it is time to go back to those features. =head2 Semantics The value of the parse result, as returned via the C method, is determined by the parse's B. Marpa's semantics are the traditional ones: The input is seen as a tree which takes its structure from the G1 rules. (This is why the G1 rules are called structural.) The value of the parse results from repeatedly evaluating nodes of this tree, starting at the bottom, with the results of child nodes made available to their parent node when the parent node is evaluated. Parse trees are usually drawn upside-down with their root at the top, and their "leaves" at the bottom. In Marpa::R2's SLIF, the "leaves" are the symbols that the G1 (structural) rules share with the L0 (lexical) rules. The symbols shared by L0 and G1 are those lexemes which are not discarded. In this example, the lexemes visible to G1 are C and the two operators which specified with a quoted string: "C<+>" and "C<*>". Marpa assigns values to the nodes of the tree, starting with the leaves. Marpa's "leaves" will always be L0 symbols, and their value by default is the literal value at their location in the input stream. In the case of the two operators described by quoted string, the value is that quoted string. That is, the value of 'C<+>' is 'C<+>', and the value of 'C<*>' is 'C<*>'. The value of C will be the portion of the input that matched the C<[\d]+> pattern. Starting with the values for leaves, Marpa::R2 moves recursively "up" the tree to its root, assigning a value to each node of the tree based on the value of its child nodes. Each non-leaf node corresponds to a G1 rule, and the children of the non-leaf node correspond to the RHS symbols of the rule. When the non-leaf node is valued, its value becomes the value of its LHS symbol, and this value will become the value of a RHS symbol of another node with one exception. The one exception, the node with a LHS symbol that does not become a RHS symbol, is the value of the top (or "root") node. The value of the top node becomes the value of the parse, and this is the parse result value to which the C method returns a reference. Previously we touched lightly, or skipped over entirely, the parts of this example dealing with semantics. We are now ready to look at them. =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 :default ::= action => ::first =for Marpa::R2::Display::End Each non-leaf node determines its value with an action. The default pseudo-rule allows you to specify the default action. (It is a pseudo-rule because its LHS, "C<:default>", is a pseudo-symbol, not a real one.) Often actions are Perl functions, which in this context are called Perl semantic closures. =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 my $recce = Marpa::R2::Scanless::R->new( { grammar => $grammar, semantics_package => 'My_Actions' } ); =for Marpa::R2::Display::End Above we saw the C named argument used when constructing the SLIF recognizer. As we noted, this specifies the package that is used to find the Perl semantic closures. In this example the default semantics, as specified by the C named argument, come from a "reserved action" named "C<::first>". (The initial double colon indicates a reserved action.) The "C<::first>" action indicates that the value of a rule is to be the value of its first child, that is, the value corresponding to the first symbol of the rule's RHS. (In the case of an empty rule, the value would be a Perl C). The other way we specify semantics in this example is by using an C adverb for a RHS alternative. We've seen the C adverb twice, but skipped over it. Now it is time to look at it. =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 Term ::= Factor | Term '+' Term action => do_add Factor ::= Number | Factor '*' Factor action => do_multiply =for Marpa::R2::Display::End The action for the second RHS alternative defining C is C, and the action for the second RHS alternative defining C is C. To implement these actions, we need to "resolve" their names -- map the action names into the Perl closures which actually carry out the semantics. The C specified the package where we can find the actions: "C". So, to resolve the C action, Marpa looks for a closure whose fully qualified name is C, which it finds: =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 sub My_Actions::do_multiply { my ( undef, $t1, undef, $t2 ) = @_; return $t1 * $t2; } =for Marpa::R2::Display::End The C action is resolved to a Perl semantic closure in much the same way: =for Marpa::R2::Display name: Landing page synopsis partial: 1 normalize-whitespace: 1 sub My_Actions::do_add { my ( undef, $t1, undef, $t2 ) = @_; return $t1 + $t2; } =for Marpa::R2::Display::End The Perl semantic closures are callbacks. They are called as each node in a parse tree is evaluated. Each Perl semantic closure is called with one or more arguments. The first argument to a value action is always a per-parse-tree object, which the callbacks can use as a scratchpad. In this example, the per-parse-tree object is not used. The remaining arguments will be the values of the node's "children" -- in other words, the values computed for each of its RHS symbols, in order. If the action is for an empty rule, the per-parse-tree 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 parse result. =head1 Tainted data Marpa::R2 exists to allow its input to alter execution in flexible and powerful ways. Marpa should not be used with untrusted input. In Perl' s taint mode, it is a fatal error to use Marpa's SLIF interface with a tainted grammar, a tainted input string, or tainted token values. =head1 Threads When used in a thread-safe Perl, Marpa::R2 should be thread-safe, with one important restriction: All Marpa objects that share the same grammar must be created and used within a single thread. This restriction may be lifted someday, but in practice it does not seem onerous. Note that you can use the same grammar in different threads by creating grammars that are exact copies of each other, one grammar per thread. =head1 The Marpa:: namespace The C top-level namespace is reserved. For extensions to Marpa, one appropriate place is the C namespace. This practice helps avoid namespace collisions, and follows a CPAN standard, as exemplified by the C C and C which are for extensions of, respectively, DBI, LWP and Moose. =head1 Other documents This document gives a semi-tutorial overview of Marpa's Scanless interface (SLIF). For more details about the SLIF, there is an L, and pages describing L, L, and L. Marpa has two other interfaces. L provides direct access to the underlying Libmarpa C library. Of the Perl interfaces to Marpa, the thin interface is the most low-level. The thin interface offers efficient access to the full power of the Marpa parse engine, but it requires the application to do a lot of the work itself. Now discouraged, the L was Marpa::R2's first interface. It is a more traditional, middle level interface which uses Perl calls instead of a DSL. L is intended as a quick refresher in parsing terminology, emphasizing how the standard terms are used in the Marpa context. Marpa's standard semantics are fully described in the L document. Techniques for tracing and for debugging your Marpa grammars are described in the L document and the L document. For those with a theoretical bent, my sources, and other useful references, are described in L. =head1 Author Jeffrey Kegler =head2 Why is it called "Marpa"? Marpa is the name of the greatest of the Tibetan "translators". In his time (the 11th century AD) Indian Buddhism was at its height. Marpa's generation of scholars was devoted to producing Tibetan versions of Buddhism's Sanskrit scriptures. Marpa became the greatest of them, and today is known as Marpa Lotsawa: "Marpa the Translator". =head2 Blatant plug Marpa is a character in my novel, B. B centers around Kurt GEdel's proof of God's existence. Yes, I Kurt GEdel, and yes, he really did work out a God Proof (it's in his I, Vol. 3, pp. 403-404). B is available as a free download (L). It can be purchased in print form at Amazon.com: L. =head1 Support L comes without warranty. Support is provided on a volunteer basis through the standard mechanisms for CPAN modules. L has details. =head1 Copyright and License =for Marpa::R2::Display ignore: 1 Copyright 2013 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 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 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 http://www.gnu.org/licenses/. =for Marpa::R2::Display::End =cut # Local Variables: # mode: cperl # cperl-indent-level: 4 # fill-column: 100 # End: # vim: expandtab shiftwidth=4: