The Grammar file

"Comments"

Through all your files, comments are either Perl style, introduced by # up to the end of line, or C style, enclosed between /* and */.

"Tokens and string literals"

Through all the grammar files, two kind of symbols may appear: Non-terminal symbols, called also left-hand-side symbols, which are the names of your rules, and Terminal symbols, called also Tokens.

Tokens are the symbols your lexer function will feed your parser with (see below). They are of two flavours: symbolic tokens and string literals.

Non-terminals and symbolic tokens share the same identifier syntax:

                [A-Za-z][A-Za-z0-9_]*
    

String literals are enclosed in single quotes and can contain almost anything. They will be output to your parser file double-quoted, making any special character as such. '"', '$' and '@' will be automatically quoted with '\', making their writing more natural. On the other hand, if you need a single quote inside your literal, just quote it with '\'.

You cannot have a literal 'error' in your grammar as it would confuse the driver with the error token. Use a symbolic token instead. In case you inadvertently use it, this will produce a warning telling you you should have written it error and will treat it as if it were the error token, which is certainly NOT what you meant.

"Grammar file syntax"

It is very close to yacc syntax (in fact, Parse::Yapp should compile a clean yacc grammar without any modification, whereas the opposite is not true).

This file is divided in three sections, separated by "%%":

        header section
        %%
        rules section
        %%
        footer section
    

The Rule Section contains your grammar rules:

A rule is made of a left-hand-side symbol, followed by a ':' and one or more right-hand-sides separated by '|' and terminated by a ';':

    exp:    exp '+' exp
        |   exp '-' exp
        ;
    

A right hand side may be empty:

    input:  #empty
        |   input line
        ;
    

(if you have more than one empty rhs, Parse::Yapp will issue a warning, as this is usually a mistake, and you will certainly have a reduce/reduce conflict)

A rhs may be followed by an optional %prec directive, followed by a token, giving the rule an explicit precedence (see yacc manuals for its precise meaning) and optional semantic action code block (see below).

    exp:   '-' exp %prec NEG { -$_[1] }
        |  exp '+' exp       { $_[1] + $_[3] }
        |  NUM
        ;
    

Note that in Parse::Yapp, a lhs cannot appear more than once as a rule name (This differs from yacc).

"The footer section"

may contain any valid Perl code and will be appended at the very end of your parser module. Here you can write your lexer, error report subs and anything relevant to you parser.

"Semantic actions"

Semantic actions are run every time a reduction occurs in the parsing flow and they must return a semantic value.

They are (usually, but see below "In rule actions") written at the very end of the rhs, enclosed with "{ }", and are copied verbatim to your parser file, inside of the rules table.

Be aware that matching braces in Perl is much more difficult than in C: inside strings they don't need to match. While in C it is very easy to detect the beginning of a string construct, or a single character, it is much more difficult in Perl, as there are so many ways of writing such literals. So there is no check for that today. If you need a brace in a double-quoted string, just quote it ("\{" or "\}"). For single-quoted strings, you will need to make a comment matching it in the right order. Sorry for the inconvenience.

    {
        "{ My string block }".
        "\{ My other string block \}".
        qq/ My unmatched brace \} /.
        # Force the match: {
        q/ for my closing brace } /
        q/ My opening brace { /
        # must be closed: }
    }
    

All of these constructs should work.

In Parse::Yapp, semantic actions are called like normal Perl sub calls, with their arguments passed in @_, and their semantic value are their return values.

$_[1] to $_[n] are the parameters just as $1 to $n in yacc, while $_[0] is the parser object itself.

Having $_[0] being the parser object itself allows you to call parser methods. That's how the yacc macros are implemented:

        yyerrok is done by calling $_[0]->YYErrok
        YYERROR is done by calling $_[0]->YYError
        YYACCEPT is done by calling $_[0]->YYAccept
        YYABORT is done by calling $_[0]->YYAbort
    

All those methods explicitly return undef, for convenience.

    YYRECOVERING is done by calling $_[0]->YYRecovering
    

Four useful methods in error recovery sub

    $_[0]->YYCurtok
    $_[0]->YYCurval
    $_[0]->YYExpect
    $_[0]->YYLexer
    

return respectively the current input token that made the parse fail, its semantic value (both can be used to modify their values too, but know what you are doing! See Error reporting routine section for an example), a list which contains the tokens the parser expected when the failure occurred and a reference to the lexer routine.

Note that if "$_[0]->YYCurtok" is declared as a %nonassoc token, it can be included in "$_[0]->YYExpect" list whenever the input try to use it in an associative way. This is not a bug: the token IS expected to report an error if encountered.

To detect such a thing in your error reporting sub, the following example should do the trick:

        grep { $_[0]->YYCurtok eq $_ } $_[0]->YYExpect
    and do {
        #Non-associative token used in an associative expression
    };
    

Accessing semantics values on the left of your reducing rule is done through the method

    $_[0]->YYSemval( index )
    

where index is an integer. Its value being 1 .. n returns the same values than $_[1] .. $_[n], but -n .. 0 returns values on the left of the rule being reduced (It is related to $-n .. $0 .. $n in yacc, but you cannot use $_[0] or $_[-n] constructs in Parse::Yapp for obvious reasons)

There is also a provision for a user data area in the parser object, accessed by the method:

    $_[0]->YYData
    

which returns a reference to an anonymous hash, which let you have all of your parsing data held inside the object (see the Calc.yp or ParseYapp.yp files in the distribution for some examples). That's how you can make you parser module reentrant: all of your module states and variables are held inside the parser object.

Note: unfortunately, method calls in Perl have a lot of overhead,
and when YYData is used, it may be called a huge number
of times. If you are not a real purist, and efficiency
is your concern, you may access directly the user-space
in the object: $parser->{USER} which is a reference to an
anonymous hash array, and then benchmark.

If no action is specified for a rule, the equivalant of a default action is run, which returns the first parameter:

   { $_[1] }
    

"In rule actions"

It is also possible to embed semantic actions inside of a rule:

    typedef:    TYPE { $type = $_[1] } identlist { ... } ;
    

When the Parse::Yapp's parser encounter such an embedded action, it modifies the grammar as if you wrote (although @x-1 is not a legal lhs value):

    @x-1:   /* empty */ { $type = $_[1] };
    typedef:    TYPE @x-1 identlist { ... } ;
    

where x is a sequential number incremented for each "in rule" action, and -1 represents the "dot position" in the rule where the action arises.

In such actions, you can use $_[1]..$_[n] variables, which are the semantic values on the left of your action.

Be aware that the way Parse::Yapp modifies your grammar because of in-rule actions can produce, in some cases, spurious conflicts that wouldn't happen otherwise.

"Generating the Parser Module"

Now that you grammar file is written, you can use yapp on it to generate your parser module:

    yapp -v Calc.yp
    

will create two files Calc.pm, your parser module, and Calc.output a verbose output of your parser rules, conflicts, warnings, states and summary.

What your are missing now is a lexer routine.

"The Lexer sub"

is called each time the parser need to read the next token.

It is called with only one argument that is the parser object itself, so you can access its methods, specially the

    $_[0]->YYData
    

data area.

It is its duty to return the next token and value to the parser. They "must" be returned as a list of two variables, the first one is the token known by the parser (symbolic or literal), the second one being anything you want (usually the content of the token, or the literal value) from a simple scalar value to any complex reference, as the parsing driver only uses it to call semantic actions:

    ( 'NUMBER', $num )
or
    ( '>=', '>=' )
or
    ( 'ARRAY', [ @values ] )
    

When the lexer reach the end of input, it must return the '' empty token with an undef value:

     ( '', undef )
    

Note that your lexer should never return 'error' as token value: for the driver, this is the error token used for error recovery and would lead to odd reactions.

Now that you have your lexer written, maybe you will need to output meaningful error messages, instead of the default which is to print 'Parse error.' on STDERR.

So you will need an Error reporting sub.

"Error reporting routine"

If you want one, write it knowing that it is passed as parameter the parser object. So you can share information with the lexer routine quite easily.

You can also use the "$_[0]->YYErrok" method in it, which will resume parsing as if no error occurred. Of course, since the invalid token is still invalid, you're supposed to fix the problem by yourself.

The method "$_[0]->YYLexer" may help you, as it returns a reference to the lexer routine, and can be called as

    ($tok,$val)=&{$_[0]->Lexer}
    

to get the next token and semantic value from the input stream. To make them current for the parser, use:

    ($_[0]->YYCurtok, $_[0]->YYCurval) = ($tok, $val)
    

and know what you're doing...

"Parsing"

Now you've got everything to do the parsing.

First, use the parser module:

    use Calc;
    

Then create the parser object:

    $parser=new Calc;
    

Now, call the YYParse method, telling it where to find the lexer and error report subs:

    $result=$parser->YYParse(yylex => \&Lexer,
                           yyerror => \&ErrorReport);
    

(assuming Lexer and ErrorReport subs have been written in your current package)

The order in which parameters appear is unimportant.

Et voila.

The YYParse method will do the parse, then return the last semantic value returned, or "undef" if error recovery cannot recover.

If you need to be sure the parse has been successful (in case your last returned semantic value is "undef"), make a call to:

    $parser->YYNberr()
    

which returns the total number of time the error reporting sub has been called.

"Error Recovery"

in Parse::Yapp is implemented the same way it is in yacc.

"Debugging Parser"

To debug your parser, you can call the YYParse method with a debug parameter:

    $parser->YYParse( ... , yydebug => value, ... )
    

where value is a bitfield, each bit representing a specific debug output:

    Bit Value    Outputs
    0x01         Token reading (useful for Lexer debugging)
    0x02         States information
    0x04         Driver actions (shifts, reduces, accept...)
    0x08         Parse Stack dump
    0x10         Error Recovery tracing
    

To have a full debugging output, use

    debug => 0x1F
    

Debugging output is sent to STDERR, and be aware that it can produce "huge" outputs.

"Standalone Parsers"

By default, the parser modules generated will need the Parse::Yapp module installed on the system to run. They use the Parse::Yapp::Driver which can be safely shared between parsers in the same script.

In the case you'd prefer to have a standalone module generated, use the "-s" switch with yapp: this will automagically copy the driver code into your module so you can use/distribute it without the need of the Parse::Yapp module, making it really a "Standalone Parser".

If you do so, please remember to include Parse::Yapp's copyright notice in your main module copyright, so others can know about Parse::Yapp module.

"Source file line numbers"

by default will be included in the generated parser module, which will help to find the guilty line in your source file in case of a syntax error. You can disable this feature by compiling your grammar with yapp using the "-n" switch.