NAME

MarpaX::Grammar::Parser - Converts a Marpa grammar into a tree using Tree::DAG_Node

Synopsis

        use MarpaX::Grammar::Parser;

        my(%option) =
        (               # Inputs:
                marpa_bnf_file   => 'share/metag.bnf',
                user_bnf_file    => 'share/stringparser.bnf',
                        # Outputs:
                cooked_tree_file => 'share/stringparser.cooked.tree',
                raw_tree_file    => 'share/stringparser.raw.tree',
        );

        MarpaX::Grammar::Parser -> new(%option) -> run;

For more help, run:

         perl scripts/bnf2tree.pl -h

See share/*.bnf for input files and share/*.tree for output files.

Installation includes copying all files from the share/ directory, into a dir chosen by File::ShareDir. Run scripts/find.grammars.pl to display the name of that dir.

Description

MarpaX::Grammar::Parser uses Marpa::R2 to convert a user's BNF into a tree of Marpa-style attributes, (see "raw_tree()"), and then post-processes that (see "compress_tree()") to create another tree, this time containing just the original grammar (see "cooked_tree()").

The nature of these trees is discussed in the "FAQ". The trees are managed by Tree::DAG_Node.

Lastly, the major purpose of the cooked tree is to serve as input to MarpaX::Grammar::GraphViz2, which graphs such cooked trees. That module has its own demo page.

Installation

Install MarpaX::Grammar::Parser as you would for any Perl module:

Run:

        cpanm MarpaX::Grammar::Parser

or run:

        sudo cpan MarpaX::Grammar::Parser

or unpack the distro, and then either:

        perl Build.PL
        ./Build
        ./Build test
        sudo ./Build install

or:

        perl Makefile.PL
        make (or dmake or nmake)
        make test
        make install

Constructor and Initialization

new() is called as my($parser) = MarpaX::Grammar::Parser -> new(k1 => v1, k2 => v2, ...).

It returns a new object of type MarpaX::Grammar::Parser.

Key-value pairs accepted in the parameter list (see also the corresponding methods [e.g. "marpa_bnf_file([$bnf_file_name])"]):

o bind_attributes => Boolean

Include (1) or exclude (0) attributes in the tree file(s) output.

Default: 0.

o cooked_tree_file => aTextFileName

The name of the text file to write containing the grammar as a cooked tree.

If '', the file is not written.

Default: ''.

Note: The bind_attributes option/method affects the output.

o logger => aLog::HandlerObject

By default, an object of type Log::Handler is created which prints to STDOUT.

See maxlevel and minlevel below.

Set logger to '' (the empty string) to stop a logger being created.

Default: undef.

o marpa_bnf_file => aMarpaBNFFileName

Specify the name of Marpa's own BNF file. This distro ships it as share/metag.bnf.

This option is mandatory.

Default: ''.

o maxlevel => $level

This option is only used if this module creates an object of type Log::Handler.

See Log::Handler::Levels.

Nothing is printed by default.

Default: 'notice'.

o minlevel => $level

This option affects Log::Handler objects.

See the Log::Handler::Levels docs.

Default: 'error'.

No lower levels are used.

o raw_tree_file => aTextFileName

The name of the text file to write containing the grammar as a raw tree.

If '', the file is not written.

Default: ''.

Note: The bind_attributes option/method affects the output.

o user_bnf_file => aUserBNFFileName

Specify the name of the file containing your Marpa::R2-style grammar.

See share/stringparser.bnf for a sample.

This option is mandatory.

Default: ''.

Methods

bind_attributes([$Boolean])

Here, the [] indicate an optional parameter.

Get or set the option which includes (1) or excludes (0) node attributes from the output cooked_tree_file and raw_tree_file.

Note: bind_attributes is a parameter to new().

compress_tree()

Called automatically by "run()".

Converts the raw tree into the cooked tree.

Output is the tree returned by "cooked_tree()".

cooked_tree()

Returns the root node, of type Tree::DAG_Node, of the cooked tree of items in the user's grammar.

By cooked tree, I mean as post-processed from the raw tree so as to include just the original user's BNF tokens.

The cooked tree is optionally written to the file name given by "cooked_tree_file([$output_file_name])".

The nature of this tree is discussed in the "FAQ".

See also "raw_tree()".

cooked_tree_file([$output_file_name])

Here, the [] indicate an optional parameter.

Get or set the name of the file to which the cooked tree form of the user's grammar will be written.

If no output file is supplied, nothing is written.

See share/stringparser.cooked.tree for the output of post-processing Marpa's analysis of share/stringparser.bnf.

This latter file is the grammar used in MarpaX::Demo::StringParser.

Note: cooked_tree_file is a parameter to new().

Note: The bind_attributes option/method affects the output.

log($level, $s)

Calls $self -> logger -> log($level => $s) if ($self -> logger).

logger([$logger_object])

Here, the [] indicate an optional parameter.

Get or set the logger object.

To disable logging, just set logger to the empty string.

Note: logger is a parameter to new().

marpa_bnf_file([$bnf_file_name])

Here, the [] indicate an optional parameter.

Get or set the name of the file to read Marpa's grammar from.

Note: marpa_bnf_file is a parameter to new().

maxlevel([$$level])

Here, the [] indicate an optional parameter.

Get or set the value used by the logger object.

This option is only used if an object of type Log::Handler is created. See Log::Handler::Levels.

Note: maxlevel is a parameter to new().

minlevel([$$level])

Here, the [] indicate an optional parameter.

Get or set the value used by the logger object.

This option is only used if an object of type Log::Handler is created. See Log::Handler::Levels.

Note: minlevel is a parameter to new().

new()

The constructor. See "Constructor and Initialization".

raw_tree()

Returns the root node, of type Tree::DAG_Node, of the raw tree of items in the user's grammar.

By raw tree, I mean as derived directly from Marpa.

The raw tree is optionally written to the file name given by "raw_tree_file([$output_file_name])".

The nature of this tree is discussed in the "FAQ".

See also "cooked_tree()".

raw_tree_file([$output_file_name])

Here, the [] indicate an optional parameter.

Get or set the name of the file to which the raw tree form of the user's grammar will be written.

If no output file is supplied, nothing is written.

See share/stringparser.raw.tree for the output of Marpa's analysis of share/stringparser.bnf.

This latter file is the grammar used in MarpaX::Demo::StringParser.

Note: raw_tree_file is a parameter to new().

Note: The bind_attributes option/method affects the output.

run()

The method which does all the work.

See "Synopsis" and scripts/bnf2tree.pl for sample code.

run() returns 0 for success and 1 for failure.

user_bnf_file([$bnf_file_name])

Here, the [] indicate an optional parameter.

Get or set the name of the file to read the user's grammar's BNF from. The whole file is slurped in as a single string.

See share/stringparser.bnf for a sample. It is the grammar used in MarpaX::Demo::StringParser.

Note: user_bnf_file is a parameter to new().

Files Shipped with this Module

Data Files

o share/c.ast.bnf

This is part of MarpaX::Languages::C::AST, by Jean-Damien Durand. It's 1,883 lines long.

The outputs are share/c.ast.cooked.tree and share/c.ast.raw.tree.

o share/c.ast.cooked.tree

This is the output from post-processing Marpa's analysis of share/c.ast.bnf.

The command to generate this file is:

        scripts/bnf2tree.sh c.ast
o share/c.ast.raw.tree

This is the output from processing Marpa's analysis of share/c.ast.bnf. It's 86,057 lines long, which indicates the complexity of Jean-Damien's grammar for C.

The command to generate this file is:

        scripts/bnf2tree.sh c.ast
o share/json.1.bnf

It is part of MarpaX::Demo::JSONParser, written as a gist by Peter Stuifzand.

See https://gist.github.com/pstuifzand/4447349.

The command to process this file is:

        scripts/bnf2tree.sh json.1

The outputs are share/json.1.cooked.tree and share/json.1.raw.tree.

o share/json.2.bnf

It also is part of MarpaX::Demo::JSONParser, written by Jeffrey Kegler as a reply to the gist above from Peter.

The command to process this file is:

        scripts/bnf2tree.sh json.2

The outputs are share/json.2.cooked.tree and share/json.2.raw.tree.

o share/json.3.bnf

The is yet another JSON grammar written by Jeffrey Kegler.

The command to process this file is:

        scripts/bnf2tree.sh json.3

The outputs are share/json.3.cooked.tree and share/json.3.raw.tree.

o share/metag.bnf.

This is a copy of Marpa::R2's BNF. That is, it's the file which Marpa uses to validate both its own metag.bnf (self-reflexively), and any user's BNF file.

See "marpa_bnf_file([$bnf_file_name])" above.

The command to process this file is:

        scripts/bnf2tree.sh metag

The outputs are share/metag.cooked.tree and share/metag.raw.tree.

o share/numeric.expressions.bnf

This BNF was extracted from MarpaX::Demo::SampleScripts's examples/ambiguous.grammar.01.pl.

It helped me debug the handling of '|' and '||' between right-hand-side alternatives.

o share/stringparser.bnf.

This is a copy of MarpaX::Demo::StringParser's BNF.

See "user_bnf_file([$bnf_file_name])" above.

The command to process this file is:

        scripts/bnf2tree.sh stringparser

The outputs are share/stringparser.cooked.tree and share/stringparser.raw.tree.

o share/termcap.info.bnf

It is part of MarpaX::Database::Terminfo, written by Jean-Damien Durand.

The command to process this file is:

        scripts/bnf2tree.sh termcap.info

The outputs are share/termcap.info.cooked.tree and share/termcap.info.raw.tree.

Scripts

These scripts are all in the scripts/ directory.

o bnf2tree.pl

This is a neat way of using this module. For help, run:

        perl scripts/bnf2tree.pl -h

Of course you are also encouraged to include the module directly in your own code.

o bnf2tree.sh

This is a quick way for me to run bnf2tree.pl.

o find.grammars.pl

This prints the path to a grammar file. After installation of the module, run it with any of these parameters:

        scripts/find.grammars.pl (Defaults to json.1.bnf)
        scripts/find.grammars.pl c.ast.bnf
        scripts/find.grammars.pl json.1.bnf
        scripts/find.grammars.pl json.2.bnf
        scripts/find.grammars.pl json.3.bnf
        scripts/find.grammars.pl stringparser.bnf
        scripts/find.grammars.pl termcap.inf.bnf

It will print the name of the path to given grammar file.

o metag.pl

This is Jeffrey Kegler's code. See the "FAQ" for more.

o pod2html.sh

This lets me quickly proof-read edits to the docs.

FAQ

What is this BNF (SLIF-DSL) thingy?

Marpa's grammars are written in what we call a SLIF-DSL. Here, SLIF stands for Marpa's Scanless Interface, and DSL is Domain-specific Language.

Many programmers will have heard of BNF. Well, Marpa's SLIF-DSL is an extended BNF. That is, it includes special tokens which only make sense within the context of a Marpa grammar. Hence the 'Domain Specific' part of the name.

In practice, this means you express your grammar in a string, and Marpa treats that as a set of rules as to how you want Marpa to process your input stream.

Marpa's docs for its SLIF-DSL are here.

What is the difference between the cooked tree and the raw tree?

The raw tree is generated by processing the output of Marpa's parse of the user's grammar file. It contains Marpa's view of that grammar. This raw tree is output by Tree::DAG_Node.

The cooked tree is generated by post-processing the raw tree, to extract just the user's grammar's tokens. It contains the user's view of their grammar. This cooked tree is output by this module.

And yes, the cooked tree can be used to reproduce (apart from formatting details) the user's BNF file.

The cooked tree can be graphed with MarpaX::Grammar::GraphViz2. That module has its own demo page.

The following items explain this in more detail.

What are the details of the nodes in the cooked tree?

Under the root (whose name is 'Cooked tree'), there are a set of nodes:

o $n1 nodes, 1 per statement (BNF rule) in the grammar

Each of these $n1 nodes has the name 'statement', and each also has a sub-tree of $n2 daughter nodes of its own.

So, each 'statement' node is the root of a sub-tree describing that statement (rule).

These sub-trees' nodes are:

o 1 node for the left-hand side of the rule
o 1 node for the separator between the left and right sides of the statement

So, this node's name is one of: '=' '::=' or '~'.

o 1 node per token from the right-hand side of the statement

These nodes' names are the tokens themselves.

So, for a rule like:

        array ::= ('[' ']') | ('[') elements (']') action => ::first

The nodes will be (see share/json.2.cooked.tree):

        :
    |--- statement. Attributes: {token => "statement"}
    |    |--- lhs. Attributes: {token => "array"}
    |    |--- parenthesized_rhs_primary_list. Attributes: {token => "("}
    |    |    |--- rhs. Attributes: {token => "'['"}
    |    |    |--- rhs. Attributes: {token => "']'"}
    |    |--- parenthesized_rhs_primary_list. Attributes: {token => ")"}
    |    |--- alternative. Attributes: {token => "|"}
    |    |--- parenthesized_rhs_primary_list. Attributes: {token => "("}
    |    |    |--- rhs. Attributes: {token => "'['"}
    |    |--- parenthesized_rhs_primary_list. Attributes: {token => ")"}
    |    |--- rhs. Attributes: {token => "elements"}
    |    |--- parenthesized_rhs_primary_list. Attributes: {token => "("}
    |    |    |--- rhs. Attributes: {token => "']'"}
    |    |--- parenthesized_rhs_primary_list. Attributes: {token => ")"}
    |    |--- action. Attributes: {token => "action"}
    |         |--- reserved_action_name. Attributes: {token => "::first"}
        :

Firstly, strip off the first 2 daughters. They are the rule name and the separator.

Clearly, to process the remaining daughters (if any) you must start by examining them from the end, looking for triplets of the form ($a, '=>', $b). $a will be a reserved word (an adverb). This then is the adverb list.

What's left, if anything, is a '|' or '||' separated list of right-hand side alternatives.

See share/json.2.cooked.tree, or any file share/*.cooked.tree.

Did you know there can be multiple 'statement' nodes with the same rule (1st daughter) name?

E.g.: Parsing share/metag.bnf produces cases like this:

        :
        |--- statement
        |    |--- <start rule>
        |    |--- ::=
        |    |--- (
        |    |--- ':start'
        |    |--- <op declare bnf>
        |    |--- )
        |    |--- symbol
        |--- statement
        |    |--- <start rule>
        |    |--- ::=
        |    |--- (
        |    |--- 'start'
        |    |--- 'symbol'
        |    |--- 'is'
        |    |--- )
        |    |--- symbol
        :

See share/metag.cooked.tree.

Did you know rules do not have to have right-hand sides?

E.g.: Parsing share/metag.bnf produces cases like this:

        :
        |--- statement
        |    |--- <event initializer>
        |    |--- ::=
        |--- statement
        :

See share/metag.cooked.tree.

What happened to my use of 'forgiving'?

It is an alias for 'latm' (Longest Acceptable Token Match), so this module always outputs 'latm'.

This is deemed to be a feature.

What are the details of the nodes in the raw tree?

The first few nodes are:

        Marpa value()
            |--- Class = MarpaX::Grammar::Parser::statements [BLESS 1]
                 |--- 0 = [] [ARRAY 2]
                      |--- 0 = 0 [SCALAR 3]
                      |--- 1 = 2460 [SCALAR 4]

This says the input text offsets are from 0 to 2460. I.e. share/stringparser.bnf is 2461 bytes long.

After this there are a set of nodes like this, one per statement:

        |--- Class = MarpaX::Grammar::Parser::statement [BLESS 5]
        |    |--- 2 = [] [ARRAY 6]
        |         |--- 0 = 0 [SCALAR 7]
        |         |--- 1 = 34 [SCALAR 8]
        |         |--- Class = MarpaX::Grammar::Parser::default_rule [BLESS 9]
        :         :

For complex statements, these node can be nested to considerable depth.

This says the first statement in the BNF is at offsets 0 .. 34, and happens to be the default rule (':default ::= action => [values]').

See share/stringparser.raw.tree, or any file share/*.raw.tree.

Where did the basic code come from?

Jeffrey Kegler wrote it, and posted it on the Google Group dedicated to Marpa, on 2013-07-22, in the thread 'Low-hanging fruit'. I modified it slightly for a module context.

The original code is shipped as scripts/metag.pl.

Why did you use Data::RenderAsTree?

It offered the output which was most easily parsed of the modules I tested. The others were Data::TreeDump, Data::Dumper, Data::TreeDraw, Data::TreeDumper and Data::Printer.

Where is Marpa's Homepage?

http://savage.net.au/Marpa.html.

See Also

MarpaX::Demo::JSONParser.

MarpaX::Demo::SampleScripts.

MarpaX::Demo::StringParser.

MarpaX::Grammar::GraphViz2.

MarpaX::Languages::C::AST.

MarpaX::Languages::Perl::PackUnpack.

MarpaX::Languages::SVG::Parser.

Data::RenderAsTree.

Log::Handler.

Machine-Readable Change Log

The file Changes was converted into Changelog.ini by Module::Metadata::Changes.

Version Numbers

Version numbers < 1.00 represent development versions. From 1.00 up, they are production versions.

Repository

https://github.com/ronsavage/MarpaX-Grammar-Parser

Support

Email the author, or log a bug on RT:

https://rt.cpan.org/Public/Dist/Display.html?Name=MarpaX::Grammar::Parser.

Author

MarpaX::Grammar::Parser was written by Ron Savage <ron@savage.net.au> in 2013.

Marpa's homepage: http://savage.net.au/Marpa.html.

Homepage: http://savage.net.au/.

Copyright

Australian copyright (c) 2013, Ron Savage.

        All Programs of mine are 'OSI Certified Open Source Software';
        you can redistribute them and/or modify them under the terms of
        The Artistic License 2.0, a copy of which is available at:
        http://www.opensource.org/licenses/index.html