コード例 #1
0
ファイル: parser.py プロジェクト: pombredanne/eazytext
class ETParser(object):
    ENDMARKER = u'<{<{}>}>'

    def __init__(
            self,
            etxconfig={},
            outputdir=None,
            # PLY-lexer options
            lex_optimize=False,
            lextab=LEXTAB,
            lex_debug=False,
            # PLY-parser options
            yacc_optimize=False,
            yacctab=YACCTAB,
            yacc_debug=False,
            debug=None):
        self.debug = lex_debug or yacc_debug or debug
        self._etxconfig = etxconfig
        # Build Lexer
        self.etlex = ETLexer(error_func=self._lex_error_func)
        kwargs = {'optimize': lex_optimize} if lex_optimize != None else {}
        kwargs.update(debug=lex_debug) if lex_debug else None
        kwargs.update(lextab=lextab)
        self.etlex.build(**kwargs)
        self.tokens = self.etlex.tokens

        # Build Parser
        kwargs = {'optimize': yacc_optimize} if yacc_optimize != None else {}
        kwargs.update(debug=yacc_debug) if yacc_debug else None
        kwargs.update(outputdir=outputdir) if outputdir else None
        kwargs.update(tabmodule=yacctab)
        self.parser = ply.yacc.yacc(module=self, **kwargs)
        self.parser.etparser = self  # For AST nodes to access `this`

        self._initialize()

    def _initialize(self, etxfile=None, etxconfig={}):
        self.etxfile = etxfile
        self.etxconfig = deepcopy(self._etxconfig)
        self.etxconfig.update(etxconfig)
        self.ctx = Context()

    def _fetchskin(self, skinfile):
        if skinfile == None:
            skincss = ''
        elif skinfile.endswith('.css'):
            skincss = codecs.open(join(rootdir, 'skins', skinfile),
                                  encoding='utf-8').read()
        else:
            skincss = skincss
        return skincss

    def _preprocess(self, text):
        """The text to be parsed is pre-parsed to remove and fix unwanted
        side effects in the parser.
        Return the preprossed text"""
        # Replace `a string ESCAPE characters`
        #text = re.compile( r'\\+', re.MULTILINE | re.UNICODE ).sub(r'\\', text)
        text = text.rstrip(r'\\')  # Remove trailing ESCAPE char
        # Replace `\ Escaped new lines`
        #text = text.replace( '\\\n', '' )
        return text

    def parse(self, text, etxfile=None, etxconfig={}, debuglevel=0):
        """Parses eazytext-markup //text// and creates an AST tree. For every
        parsing invocation, the same lex, yacc, and objects will
        be used.

        : text ::
            A string containing the Wiki text
        : etxfile ::
            Name of the file being parsed (for meaningful error messages)
        : debuglevel ::
            Debug level to yacc
        """
        # Initialize
        self._initialize(etxfile=etxfile, etxconfig=etxconfig)
        self.text = text
        self.etlex.filename = etxfile
        enc = self.etxconfig['input_encoding']
        self.hashtext = sha1(text.encode(enc)).hexdigest()

        self.skincss = self._fetchskin(
            self.etxconfig['skinfile']
        ) if self.etxconfig['include_skin'] else ''

        # Pre-process the text, massage them for prasing.
        self.pptext = self._preprocess(text)

        # parse and get the Translation Unit
        debuglevel = self.debug or debuglevel
        self.pptext += u'\n' + self.ENDMARKER
        tu = self.parser.parse(self.pptext, lexer=self.etlex, debug=debuglevel)
        # Restart the parser and lexer
        self.etlex.reset_lineno()
        self.parser.restart()
        return tu

    # ------------------------- Private functions -----------------------------

    def _lex_error_func(self, msg, line, column):
        self._parse_error(msg, self._coord(line, column))

    def _coord(self, lineno, column=None):
        return Coord(file=self.etlex.filename, line=lineno, column=column)

    def _parse_error(self, msg, coord):
        raise ParseError("%s: %s" % (coord, msg))

    def _printparse(self, p):
        print p[0], "  : ",
        for i in range(1, len(p)):
            print p[i],
        print

    # ---------- Precedence and associativity of operators --------------------

    precedence = (('left', 'PREC_LINK', 'PREC_MACRO', 'PREC_HTML'), )

    def _buildterms(self, p, terms):
        rc = []
        for t in terms:
            if t == None:
                rc.append(None)
                continue
            elif isinstance(t, tuple):
                cls, idx = t
                rc.append(cls(p.parser, p[idx]))
            else:
                rc.append(t)
        return rc

    #---- Page

    def p_wikipage(self, p):
        """wikipage         : paragraphs
                            | paragraphs ENDMARKER"""
        p[0] = WikiPage(p.parser, p[1])

    def p_paragraphs_1(self, p):
        """paragraphs       : paragraph_separator"""
        p[0] = Paragraphs(p.parser, None, None, p[1])

    def p_paragraphs_2(self, p):
        """paragraphs       : paragraph"""
        p[0] = Paragraphs(p.parser, None, p[1], None)

    def p_paragraphs_3(self, p):
        """paragraphs       : paragraph paragraph_separator"""
        p[0] = Paragraphs(p.parser, None, p[1], p[2])

    def p_paragraphs_4(self, p):
        """paragraphs       : paragraphs paragraph paragraph_separator"""
        p[0] = Paragraphs(p.parser, p[1], p[2], p[3])

    def p_paragraphs_5(self, p):
        """paragraphs       : paragraphs paragraph"""
        p[0] = Paragraphs(p.parser, p[1], p[2], None)

    def p_paragraph_1(self, p):
        """paragraph        : textlines"""
        p[0] = Paragraph(p.parser, p[1])

    def p_paragraph_2(self, p):
        """paragraph        : specialpara"""
        p[0] = p[1]

    def p_specialpara(self, p):
        """specialpara      : nowikiblock
                            | heading
                            | section
                            | horizontalrule
                            | btable
                            | table
                            | mixedlists
                            | definitions
                            | blockquotes"""
        p[0] = Paragraph(p.parser, p[1])

    #---- Nowiki

    def p_nowiki_1(self, p):
        """nowikiblock  : NOWIKI_OPEN NEWLINE nowikilines NOWIKI_CLOSE NEWLINE"""
        terms = [(NOWIKI_OPEN, 1), (NEWLINE, 2), p[3], (NOWIKI_CLOSE, 4),
                 (NEWLINE, 5)]
        p[0] = NoWiki(p.parser, *self._buildterms(p, terms))

    def p_nowiki_2(self, p):
        """nowikiblock  : NOWIKI_OPEN NEWLINE nowikilines ENDMARKER"""
        terms = [(NOWIKI_OPEN, 1), (NEWLINE, 2), p[3], None, None]
        p[0] = NoWiki(p.parser, *self._buildterms(p, terms))

    def p_nowikilines_1(self, p):
        """nowikilines  : NEWLINE"""
        p[0] = NowikiLines(p.parser, None, None, NEWLINE(p.parser, p[1]))

    def p_nowikilines_2(self, p):
        """nowikilines  : NOWIKITEXT NEWLINE"""
        terms = [None, (NOWIKITEXT, 1), (NEWLINE, 2)]
        p[0] = NowikiLines(p.parser, *self._buildterms(p, terms))

    def p_nowikilines_3(self, p):
        """nowikilines  : nowikilines NEWLINE"""
        p[0] = NowikiLines(p.parser, p[1], None, NEWLINE(p.parser, p[2]))

    def p_nowikilines_4(self, p):
        """nowikilines  : nowikilines NOWIKITEXT NEWLINE"""
        terms = [p[1], (NOWIKITEXT, 2), (NEWLINE, 3)]
        p[0] = NowikiLines(p.parser, *self._buildterms(p, terms))

    def p_nowikilines_5(self, p):
        """nowikilines  : """
        p[0] = NowikiLines(p.parser, None, None, None)

    #---- Heading, horizontal and textlines

    def p_heading(self, p):
        """heading      : HEADING text_contents NEWLINE
                        | HEADING NEWLINE"""
        terms = [(HEADING, 1), p[2],
                 (NEWLINE, 3)] if len(p) == 4 else [(HEADING, 1), None,
                                                    (NEWLINE, 2)]
        p[0] = Heading(p.parser, *self._buildterms(p, terms))

    def p_section(self, p):
        """section      : SECTION text_contents NEWLINE
                        | SECTION NEWLINE"""
        terms = [(SECTION, 1), p[2],
                 (NEWLINE, 3)] if len(p) == 4 else [(SECTION, 1), None,
                                                    (NEWLINE, 2)]
        p[0] = Section(p.parser, *self._buildterms(p, terms))

    def p_horizontalrule(self, p):
        """horizontalrule : HORIZONTALRULE"""
        p[0] = HorizontalRule(p.parser, HORIZONTALRULE(p.parser, p[1]))

    def p_textlines(self, p):
        """textlines    : textline
                        | textlines textline"""
        args = [p[1], p[2]] if len(p) == 3 else [None, p[1]]
        p[0] = TextLines(p.parser, *args)

    def p_textline(self, p):
        """textline     : text_contents NEWLINE"""
        p[0] = TextLine(p.parser, p[1], NEWLINE(p.parser, p[2]))

    #---- Bigtable rows

    def p_btable(self, p):
        """btable       : btableblocks"""
        p[0] = BTable(p.parser, p[1])

    def p_btableblocks(self, p):
        """btableblocks : btableblock
                        | btableblocks btableblock"""
        args = [p[1], p[2]] if len(p) == 3 else [None, p[1]]
        p[0] = BigtableBlocks(p.parser, *args)

    def p_btableblock_1(self, p):
        """btableblock  : BTABLE_START text_contents NEWLINE
                        | BTABLE_START NEWLINE"""
        terms = [None, (BTABLE_START, 1), p[2],
                 (NEWLINE, 3)] if len(p) == 4 else [
                     None, (BTABLE_START, 1), None, (NEWLINE, 2)
                 ]
        p[0] = BigtableBlock(p.parser, *self._buildterms(p, terms))

    def p_btableblock_2(self, p):
        """btableblock  : btableblock text_contents NEWLINE"""
        p[0] = BigtableBlock(p.parser, p[1], None, p[2],
                             NEWLINE(p.parser, p[3]))

    #---- table

    def p_table(self, p):
        """table        : table_rows"""
        p[0] = Table(p.parser, p[1])

    def p_table_rows(self, p):
        """table_rows   : table_row
                        | table_rows table_row"""
        args = [p[1], p[2]] if len(p) == 3 else [None, p[1]]
        p[0] = TableRows(p.parser, *args)

    def p_table_row(self, p):
        """table_row    : table_cells NEWLINE"""
        p[0] = TableRow(p.parser, p[1], NEWLINE(p.parser, p[2]))

    def p_table_cells(self, p):
        """table_cells  : table_cell
                        | table_cells table_cell"""
        args = [p[1], p[2]] if len(p) == 3 else [None, p[1]]
        p[0] = TableCells(p.parser, *args)

    def p_table_cell(self, p):
        """table_cell   : TABLE_CELLSTART
                        | TABLE_CELLSTART text_contents"""
        args = [TABLE_CELLSTART(p.parser, p[1]), p[2]
                ] if len(p) == 3 else [TABLE_CELLSTART(p.parser, p[1]), None]
        p[0] = TableCell(p.parser, *args)

    #---- Lists

    def p_mixedlists_1(self, p):
        """mixedlists   : orderedlists
                        | mixedlists unorderedlists"""
        args = [p[1], p[2], None] if len(p) == 3 else [None, p[1], None]
        p[0] = MixedLists(p.parser, *args)

    def p_mixedlists_2(self, p):
        """mixedlists   : unorderedlists
                        | mixedlists orderedlists"""
        args = [p[1], None, p[2]] if len(p) == 3 else [None, None, p[1]]
        p[0] = MixedLists(p.parser, *args)

    def p_orderedlists(self, p):
        """orderedlists : orderedlist
                        | orderedlists orderedlist"""
        args = [p[1], p[2]] if len(p) == 3 else [None, p[1]]
        p[0] = Lists(p.parser, 'ol', *args)

    def p_unorderedlists(self, p):
        """unorderedlists   : unorderedlist
                            | unorderedlists unorderedlist"""
        args = [p[1], p[2]] if len(p) == 3 else [None, p[1]]
        p[0] = Lists(p.parser, 'ul', *args)

    def p_orderedlist(self, p):
        """orderedlist  : orderedlistbegin
                        | orderedlist text_contents NEWLINE"""
        args = [None, p[1], p[2], NEWLINE(p.parser, p[3])
                ] if len(p) == 4 else [p[1], None, None, None]
        p[0] = List(p.parser, 'ol', *args)

    def p_unorderedlist(self, p):
        """unorderedlist    : unorderedlistbegin
                            | unorderedlist text_contents NEWLINE"""
        args = [None, p[1], p[2], NEWLINE(p.parser, p[3])
                ] if len(p) == 4 else [p[1], None, None, None]
        p[0] = List(p.parser, 'ul', *args)

    def p_orderedlistbegin(self, p):
        """orderedlistbegin : ORDLIST_START text_contents NEWLINE
                            | ORDLIST_START NEWLINE"""
        terms = [(ORDLIST_START, 1), p[2],
                 (NEWLINE, 3)] if len(p) == 4 else [(ORDLIST_START, 1), None,
                                                    (NEWLINE, 2)]
        p[0] = ListBegin(p.parser, 'ol', *self._buildterms(p, terms))

    def p_unorderedlistbegin(self, p):
        """unorderedlistbegin   : UNORDLIST_START text_contents NEWLINE
                                | UNORDLIST_START NEWLINE"""
        terms = [(UNORDLIST_START, 1), p[2],
                 (NEWLINE, 3)] if len(p) == 4 else [(UNORDLIST_START, 1), None,
                                                    (NEWLINE, 2)]
        p[0] = ListBegin(p.parser, 'ul', *self._buildterms(p, terms))

    #---- Definition

    def p_definitions(self, p):
        """definitions  : definition
                        | definitions definition"""
        args = [p[1], p[2]] if len(p) == 3 else [None, p[1]]
        p[0] = Definitions(p.parser, *args)

    def p_definition(self, p):
        """definition   : definitionbegin
                        | definition text_contents NEWLINE"""
        args = [None, p[1], p[2], NEWLINE(p.parser, p[3])
                ] if len(p) == 4 else [p[1], None, None, None]
        p[0] = Definition(p.parser, *args)

    def p_definitionbegin(self, p):
        """definitionbegin  : DEFINITION_START text_contents NEWLINE
                            | DEFINITION_START NEWLINE"""
        terms = [(DEFINITION_START, 1), p[2],
                 (NEWLINE, 3)] if len(p) == 4 else [(DEFINITION_START,
                                                     1), None, (NEWLINE, 2)]
        p[0] = DefinitionBegin(p.parser, *self._buildterms(p, terms))

    #---- Blockquotes

    def p_blockquotes(self, p):  # BQuotes
        """blockquotes  : blockquote
                        | blockquotes blockquote"""
        args = [p[1], p[2]] if len(p) == 3 else [None, p[1]]
        p[0] = BlockQuotes(p.parser, *args)

    def p_blockquote(self, p):  # BQuote
        """blockquote   : BQUOTE_START text_contents NEWLINE
                        | BQUOTE_START NEWLINE"""
        terms = [(BQUOTE_START, 1), p[2],
                 (NEWLINE, 3)] if len(p) == 4 else [(BQUOTE_START, 1), None,
                                                    (NEWLINE, 2)]
        p[0] = BlockQuote(p.parser, *self._buildterms(p, terms))

    #---- Text contents

    def p_text_contents(self, p):  # TextContents
        """text_contents    : basictext
                            | markuptext
                            | link
                            | macro
                            | html
                            | text_contents basictext
                            | text_contents markuptext
                            | text_contents link
                            | text_contents macro
                            | text_contents html"""
        args = [p[1], p[2]] if len(p) == 3 else [None, p[1]]
        p[0] = TextContents(p.parser, *args)

    def p_link(self, p):
        """link         : LINK %prec PREC_LINK"""
        p[0] = Link(p.parser, LINK(p.parser, p[1]))

    #def p_nestedlink( self, p ):
    #    """nestedlink   : NESTEDLINK %prec PREC_LINK"""
    #    p[0] = NestedLink( p.parser, NESTEDLINK(p.parser, p[1]) )
    #    pass

    def p_macro(self, p):
        """macro        : MACRO %prec PREC_MACRO"""
        p[0] = Macro(p.parser, MACRO(p.parser, p[1]))

    def p_html(self, p):
        """html         : HTML %prec PREC_HTML"""
        p[0] = Html(p.parser, HTML(p.parser, p[1]))

    #---- Basic Text

    def p_basictext_0(self, p):
        """basictext    : TEXT"""
        p[0] = BasicText(p.parser, TEXT(p.parser, p[1]))

    def p_basictext_1(self, p):
        """basictext    : ESCAPED_TEXT"""
        p[0] = BasicText(p.parser, TEXT(p.parser, p[1]))

    def p_basictext_2(self, p):
        """basictext    : SPECIALCHARS"""
        p[0] = BasicText(p.parser, SPECIALCHARS(p.parser, p[1]))

    def p_basictext_3(self, p):
        """basictext    : HTTP_URI"""
        p[0] = BasicText(p.parser, HTTP_URI(p.parser, p[1]))

    def p_basictext_4(self, p):
        """basictext    : HTTPS_URI"""
        p[0] = BasicText(p.parser, HTTPS_URI(p.parser, p[1]))

    def p_basictext_5(self, p):
        """basictext    : WWW_URI"""
        p[0] = BasicText(p.parser, WWW_URI(p.parser, p[1]))

    def p_basictext_6(self, p):
        """basictext    : LINEBREAK"""
        p[0] = BasicText(p.parser, LINEBREAK(p.parser, p[1]))

    def p_markuptext_1(self, p):
        """markuptext   : M_SPAN"""
        p[0] = MarkupText(p.parser, M_SPAN(p.parser, p[1]))

    def p_markuptext_2(self, p):
        """markuptext   : M_BOLD"""
        p[0] = MarkupText(p.parser, M_BOLD(p.parser, p[1]))

    def p_markuptext_3(self, p):
        """markuptext   : M_ITALIC"""
        p[0] = MarkupText(p.parser, M_ITALIC(p.parser, p[1]))

    def p_markuptext_4(self, p):
        """markuptext   : M_UNDERLINE"""
        p[0] = MarkupText(p.parser, M_UNDERLINE(p.parser, p[1]))

    def p_markuptext_5(self, p):
        """markuptext   : M_SUPERSCRIPT"""
        p[0] = MarkupText(p.parser, M_SUPERSCRIPT(p.parser, p[1]))

    def p_markuptext_6(self, p):
        """markuptext   : M_SUBSCRIPT"""
        p[0] = MarkupText(p.parser, M_SUBSCRIPT(p.parser, p[1]))

    def p_markuptext_7(self, p):
        """markuptext   : M_BOLDITALIC"""
        p[0] = MarkupText(p.parser, M_BOLDITALIC(p.parser, p[1]))

    def p_markuptext_8(self, p):
        """markuptext   : M_BOLDUNDERLINE"""
        p[0] = MarkupText(p.parser, M_BOLDUNDERLINE(p.parser, p[1]))

    def p_markuptext_9(self, p):
        """markuptext   : M_ITALICUNDERLINE"""
        p[0] = MarkupText(p.parser, M_ITALICUNDERLINE(p.parser, p[1]))

    def p_markuptext_10(self, p):
        """markuptext   : M_BOLDITALICUNDERLINE"""
        p[0] = MarkupText(p.parser, M_BOLDITALICUNDERLINE(p.parser, p[1]))

    #---- sub-grammars

    def p_paragraph_seperator_1(self, p):
        """paragraph_separator  : NEWLINE"""
        p[0] = ParagraphSeparator(p.parser, None, NEWLINE(p.parser, p[1]))

    def p_paragraph_seperator_2(self, p):
        """paragraph_separator  : paragraph_separator NEWLINE"""
        p[0] = ParagraphSeparator(p.parser, p[1], NEWLINE(p.parser, p[2]))

    def p_error(self, p):
        if p:
            column = self.etlex._find_tok_column(p)
            self._parse_error('before: %s ' % p.value,
                              self._coord(p.lineno, column))
        else:
            self._parse_error('At end of input', '')
コード例 #2
0
def setUpModule() :
    global etlex
    etlex = ETLexer()
    etlex.build()
コード例 #3
0
ファイル: parser.py プロジェクト: prataprc/eazytext
class ETParser( object ):
    ENDMARKER = u'<{<{}>}>'
    def __init__(   self,
                    etxconfig={},
                    outputdir=None,
                    # PLY-lexer options
                    lex_optimize=False,
                    lextab=LEXTAB,
                    lex_debug=False,
                    # PLY-parser options
                    yacc_optimize=False,
                    yacctab=YACCTAB,
                    yacc_debug=False,
                    debug=None
                ):
        self.debug = lex_debug or yacc_debug or debug
        self._etxconfig = etxconfig
        # Build Lexer
        self.etlex = ETLexer( error_func=self._lex_error_func )
        kwargs = {'optimize' : lex_optimize} if lex_optimize != None else {}
        kwargs.update( debug=lex_debug ) if lex_debug else None
        kwargs.update( lextab=lextab )
        self.etlex.build( **kwargs )
        self.tokens = self.etlex.tokens

        # Build Parser
        kwargs = {'optimize' : yacc_optimize} if yacc_optimize != None else {}
        kwargs.update( debug=yacc_debug    ) if yacc_debug else None
        kwargs.update( outputdir=outputdir ) if outputdir else None
        kwargs.update( tabmodule=yacctab )
        self.parser = ply.yacc.yacc( module=self, **kwargs )
        self.parser.etparser = self   # For AST nodes to access `this`

        self._initialize()

    def _initialize( self, etxfile=None, etxconfig={} ):
        self.etxfile = etxfile
        self.etxconfig = deepcopy( self._etxconfig )
        self.etxconfig.update( etxconfig )
        self.ctx = Context()

    def _fetchskin( self, skinfile ):
        if skinfile == None :
            skincss = ''
        elif skinfile.endswith( '.css' ) :
            skincss = codecs.open(
                join(rootdir, 'skins', skinfile), encoding='utf-8' ).read()
        else :
            skincss = skincss
        return skincss

    def _preprocess( self, text ) :
        """The text to be parsed is pre-parsed to remove and fix unwanted
        side effects in the parser.
        Return the preprossed text"""
        # Replace `a string ESCAPE characters`
        #text = re.compile( r'\\+', re.MULTILINE | re.UNICODE ).sub(r'\\', text)
        text = text.rstrip( r'\\' )   # Remove trailing ESCAPE char
        # Replace `\ Escaped new lines`
        #text = text.replace( '\\\n', '' )
        return text

    def parse( self, text, etxfile=None, etxconfig={}, debuglevel=0 ):
        """Parses eazytext-markup //text// and creates an AST tree. For every
        parsing invocation, the same lex, yacc, and objects will
        be used.

        : text ::
            A string containing the Wiki text
        : etxfile ::
            Name of the file being parsed (for meaningful error messages)
        : debuglevel ::
            Debug level to yacc
        """
        # Initialize
        self._initialize( etxfile=etxfile, etxconfig=etxconfig )
        self.text = text
        self.etlex.filename = etxfile
        enc = self.etxconfig['input_encoding']
        self.hashtext = sha1( text.encode(enc) ).hexdigest()

        self.skincss = self._fetchskin( self.etxconfig['skinfile'] 
                       ) if self.etxconfig['include_skin'] else ''

        # Pre-process the text, massage them for prasing.
        self.pptext = self._preprocess( text )

        # parse and get the Translation Unit
        debuglevel = self.debug or debuglevel
        self.pptext += u'\n' + self.ENDMARKER
        tu = self.parser.parse(self.pptext, lexer=self.etlex, debug=debuglevel)
        # Restart the parser and lexer
        self.etlex.reset_lineno()
        self.parser.restart()
        return tu

    # ------------------------- Private functions -----------------------------

    def _lex_error_func( self, msg, line, column ):
        self._parse_error( msg, self._coord( line, column ))
    
    def _coord( self, lineno, column=None ):
        return Coord( file=self.etlex.filename, 
                      line=lineno,
                      column=column
               )
    
    def _parse_error(self, msg, coord):
        raise ParseError("%s: %s" % (coord, msg))

    def _printparse( self, p ) :
        print p[0], "  : ",
        for i in range(1,len(p)) :
            print p[i],
        print

    # ---------- Precedence and associativity of operators --------------------

    precedence = (
        ( 'left', 'PREC_LINK', 'PREC_MACRO', 'PREC_HTML' ),
    )
    
    def _buildterms( self, p, terms ):
        rc = []
        for t in terms :
            if t == None :
                rc.append( None )
                continue
            elif isinstance(t, tuple) :
                cls, idx = t
                rc.append( cls(p.parser, p[idx]) )
            else :
                rc.append(t)
        return rc


    #---- Page

    def p_wikipage( self, p ):
        """wikipage         : paragraphs
                            | paragraphs ENDMARKER"""
        p[0] = WikiPage( p.parser, p[1] )

    def p_paragraphs_1( self, p ):
        """paragraphs       : paragraph_separator"""
        p[0] = Paragraphs( p.parser, None, None, p[1] )

    def p_paragraphs_2( self, p ):
        """paragraphs       : paragraph"""
        p[0] = Paragraphs( p.parser, None, p[1], None )

    def p_paragraphs_3( self, p ):
        """paragraphs       : paragraph paragraph_separator"""
        p[0] = Paragraphs( p.parser, None, p[1], p[2] )

    def p_paragraphs_4( self, p ):
        """paragraphs       : paragraphs paragraph paragraph_separator"""
        p[0] = Paragraphs( p.parser, p[1], p[2], p[3] )

    def p_paragraphs_5( self, p ):
        """paragraphs       : paragraphs paragraph"""
        p[0] = Paragraphs( p.parser, p[1], p[2], None )

    def p_paragraph_1( self, p ):
        """paragraph        : textlines"""
        p[0] = Paragraph( p.parser, p[1] )

    def p_paragraph_2( self, p ):
        """paragraph        : specialpara"""
        p[0] = p[1]

    def p_specialpara( self, p ):
        """specialpara      : nowikiblock
                            | heading
                            | section
                            | horizontalrule
                            | btable
                            | table
                            | mixedlists
                            | definitions
                            | blockquotes"""
        p[0] = Paragraph( p.parser, p[1] )

    #---- Nowiki

    def p_nowiki_1( self, p ):
        """nowikiblock  : NOWIKI_OPEN NEWLINE nowikilines NOWIKI_CLOSE NEWLINE"""
        terms = [ (NOWIKI_OPEN,1), (NEWLINE,2), p[3], (NOWIKI_CLOSE,4), (NEWLINE,5) ]
        p[0] = NoWiki( p.parser, *self._buildterms( p, terms ) )

    def p_nowiki_2( self, p ):
        """nowikiblock  : NOWIKI_OPEN NEWLINE nowikilines ENDMARKER"""
        terms = [ (NOWIKI_OPEN,1), (NEWLINE,2), p[3], None, None ]
        p[0] = NoWiki( p.parser, *self._buildterms( p, terms ) )

    def p_nowikilines_1( self, p ):
        """nowikilines  : NEWLINE"""
        p[0] = NowikiLines( p.parser, None, None, NEWLINE(p.parser, p[1]) )

    def p_nowikilines_2( self, p ):
        """nowikilines  : NOWIKITEXT NEWLINE"""
        terms = [ None, (NOWIKITEXT,1), (NEWLINE,2) ]
        p[0] = NowikiLines( p.parser, *self._buildterms(p, terms) )

    def p_nowikilines_3( self, p ):
        """nowikilines  : nowikilines NEWLINE"""
        p[0] = NowikiLines( p.parser, p[1], None, NEWLINE(p.parser, p[2]) )

    def p_nowikilines_4( self, p ):
        """nowikilines  : nowikilines NOWIKITEXT NEWLINE"""
        terms = [ p[1], (NOWIKITEXT,2), (NEWLINE,3) ]
        p[0] = NowikiLines( p.parser, *self._buildterms(p, terms) )

    def p_nowikilines_5( self, p ):
        """nowikilines  : """
        p[0] = NowikiLines( p.parser, None, None, None )

    #---- Heading, horizontal and textlines

    def p_heading( self, p ):
        """heading      : HEADING text_contents NEWLINE
                        | HEADING NEWLINE"""
        terms = [ (HEADING,1), p[2], (NEWLINE,3) 
                ] if len(p) == 4 else [ (HEADING,1), None, (NEWLINE,2) ]
        p[0] = Heading( p.parser, *self._buildterms( p, terms ))

    def p_section( self, p ):
        """section      : SECTION text_contents NEWLINE
                        | SECTION NEWLINE"""
        terms = [ (SECTION,1), p[2], (NEWLINE,3) 
                ] if len(p) == 4 else [ (SECTION,1), None, (NEWLINE,2) ]
        p[0] = Section( p.parser, *self._buildterms( p, terms ))

    def p_horizontalrule( self, p ):
        """horizontalrule : HORIZONTALRULE"""
        p[0] = HorizontalRule( p.parser, HORIZONTALRULE(p.parser, p[1]) )

    def p_textlines( self, p ):
        """textlines    : textline
                        | textlines textline"""
        args = [ p[1], p[2] ] if len(p)==3 else [ None, p[1] ]
        p[0] = TextLines( p.parser, *args )

    def p_textline( self, p ):
        """textline     : text_contents NEWLINE"""
        p[0] = TextLine( p.parser, p[1], NEWLINE(p.parser, p[2]) )

    #---- Bigtable rows

    def p_btable( self, p ):
        """btable       : btableblocks"""
        p[0] = BTable( p.parser, p[1] )

    def p_btableblocks( self, p ):
        """btableblocks : btableblock
                        | btableblocks btableblock"""
        args = [ p[1], p[2] ] if len(p) == 3 else [ None, p[1] ]
        p[0] = BigtableBlocks( p.parser, *args )

    def p_btableblock_1( self, p ):
        """btableblock  : BTABLE_START text_contents NEWLINE
                        | BTABLE_START NEWLINE"""
        terms = [ None, (BTABLE_START,1), p[2], (NEWLINE,3)
                ] if len(p) == 4 else [ None, (BTABLE_START,1), None, (NEWLINE,2) ]
        p[0] = BigtableBlock( p.parser, *self._buildterms( p, terms ) )

    def p_btableblock_2( self, p ):
        """btableblock  : btableblock text_contents NEWLINE"""
        p[0] = BigtableBlock( p.parser, p[1], None, p[2], NEWLINE(p.parser, p[3] ))

    #---- table

    def p_table( self, p ):
        """table        : table_rows"""
        p[0] = Table( p.parser, p[1] )

    def p_table_rows( self, p):
        """table_rows   : table_row
                        | table_rows table_row"""
        args = [ p[1], p[2] ] if len(p)==3 else [ None, p[1] ]
        p[0] = TableRows( p.parser, *args )

    def p_table_row( self, p ):
        """table_row    : table_cells NEWLINE"""
        p[0] = TableRow( p.parser, p[1], NEWLINE(p.parser, p[2]) )

    def p_table_cells( self, p ):
        """table_cells  : table_cell
                        | table_cells table_cell"""
        args = [ p[1], p[2] ] if len(p) == 3 else [ None, p[1] ]
        p[0] = TableCells( p.parser, *args )

    def p_table_cell( self, p ):
        """table_cell   : TABLE_CELLSTART
                        | TABLE_CELLSTART text_contents"""
        args = [ TABLE_CELLSTART(p.parser, p[1]), p[2]
               ] if len(p) == 3 else [ TABLE_CELLSTART(p.parser, p[1]), None ]
        p[0] = TableCell( p.parser, *args )

    #---- Lists

    def p_mixedlists_1( self, p ):
        """mixedlists   : orderedlists
                        | mixedlists unorderedlists"""
        args = [ p[1], p[2], None ] if len(p)==3 else [ None, p[1], None ]
        p[0] = MixedLists( p.parser, *args )

    def p_mixedlists_2( self, p ):
        """mixedlists   : unorderedlists
                        | mixedlists orderedlists"""
        args = [ p[1], None, p[2] ] if len(p)==3 else [ None, None, p[1] ]
        p[0] = MixedLists( p.parser, *args )

    def p_orderedlists( self, p ):
        """orderedlists : orderedlist
                        | orderedlists orderedlist"""
        args = [ p[1], p[2] ] if len(p)==3 else [ None, p[1] ]
        p[0] = Lists( p.parser, 'ol', *args )

    def p_unorderedlists( self, p ):
        """unorderedlists   : unorderedlist
                            | unorderedlists unorderedlist"""
        args = [ p[1], p[2] ] if len(p)==3 else [ None, p[1] ]
        p[0] = Lists( p.parser, 'ul', *args )

    def p_orderedlist( self, p ):
        """orderedlist  : orderedlistbegin
                        | orderedlist text_contents NEWLINE"""
        args = [ None, p[1], p[2], NEWLINE(p.parser, p[3]) 
               ] if len(p) == 4 else [ p[1], None, None, None ]
        p[0] = List( p.parser, 'ol', *args )

    def p_unorderedlist( self, p ):
        """unorderedlist    : unorderedlistbegin
                            | unorderedlist text_contents NEWLINE"""
        args = [ None, p[1], p[2], NEWLINE(p.parser, p[3]) 
               ] if len(p) == 4 else [ p[1], None, None, None ]
        p[0] = List( p.parser, 'ul', *args )

    def p_orderedlistbegin( self, p ):
        """orderedlistbegin : ORDLIST_START text_contents NEWLINE
                            | ORDLIST_START NEWLINE"""
        terms = [ (ORDLIST_START,1), p[2], (NEWLINE,3) 
                ] if len(p) == 4 else [ (ORDLIST_START,1), None, (NEWLINE,2) ]
        p[0] = ListBegin( p.parser, 'ol', *self._buildterms(p, terms) )

    def p_unorderedlistbegin( self, p ):
        """unorderedlistbegin   : UNORDLIST_START text_contents NEWLINE
                                | UNORDLIST_START NEWLINE"""
        terms = [ (UNORDLIST_START,1), p[2], (NEWLINE,3) 
                ] if len(p) == 4 else [ (UNORDLIST_START,1), None, (NEWLINE,2) ]
        p[0] = ListBegin( p.parser, 'ul', *self._buildterms(p, terms) )

    #---- Definition

    def p_definitions( self, p ):
        """definitions  : definition
                        | definitions definition"""
        args = [ p[1], p[2] ]if len(p) == 3 else [ None, p[1] ]
        p[0] = Definitions( p.parser, *args )

    def p_definition( self, p ):
        """definition   : definitionbegin
                        | definition text_contents NEWLINE"""
        args = [ None, p[1], p[2], NEWLINE(p.parser, p[3]) 
               ] if len(p) == 4 else [ p[1], None, None, None ]
        p[0] = Definition( p.parser, *args )

    def p_definitionbegin( self, p ):
        """definitionbegin  : DEFINITION_START text_contents NEWLINE
                            | DEFINITION_START NEWLINE"""
        terms = [ (DEFINITION_START,1), p[2], (NEWLINE,3)
                ] if len(p) == 4 else [ (DEFINITION_START,1), None, (NEWLINE,2) ]
        p[0] = DefinitionBegin( p.parser, *self._buildterms(p, terms) )

    #---- Blockquotes

    def p_blockquotes( self, p ):                       # BQuotes
        """blockquotes  : blockquote
                        | blockquotes blockquote"""
        args = [ p[1], p[2] ] if len(p) == 3 else [ None, p[1] ]
        p[0] = BlockQuotes( p.parser, *args )

    def p_blockquote( self, p ):                        # BQuote
        """blockquote   : BQUOTE_START text_contents NEWLINE
                        | BQUOTE_START NEWLINE"""
        terms = [ (BQUOTE_START,1), p[2], (NEWLINE,3)
                ] if len(p) == 4 else [ (BQUOTE_START,1), None, (NEWLINE,2) ]
        p[0] = BlockQuote( p.parser, *self._buildterms( p, terms ) )

    #---- Text contents

    def p_text_contents( self, p ) :                    # TextContents
        """text_contents    : basictext
                            | markuptext
                            | link
                            | macro
                            | html
                            | text_contents basictext
                            | text_contents markuptext
                            | text_contents link
                            | text_contents macro
                            | text_contents html"""
        args = [ p[1], p[2] ] if len(p)==3 else [ None, p[1] ]
        p[0] = TextContents( p.parser, *args )

    def p_link( self, p ):
        """link         : LINK %prec PREC_LINK"""
        p[0] = Link( p.parser, LINK(p.parser, p[1]) )

    #def p_nestedlink( self, p ):
    #    """nestedlink   : NESTEDLINK %prec PREC_LINK"""
    #    p[0] = NestedLink( p.parser, NESTEDLINK(p.parser, p[1]) )
    #    pass

    def p_macro( self, p ):
        """macro        : MACRO %prec PREC_MACRO"""
        p[0] = Macro( p.parser, MACRO(p.parser, p[1]) )

    def p_html( self, p ):
        """html         : HTML %prec PREC_HTML"""
        p[0] = Html( p.parser, HTML(p.parser, p[1]) )


    #---- Basic Text

    def p_basictext_0( self, p ):
        """basictext    : TEXT"""
        p[0] = BasicText( p.parser, TEXT(p.parser, p[1]) )

    def p_basictext_1( self, p ):
        """basictext    : ESCAPED_TEXT"""
        p[0] = BasicText( p.parser, TEXT(p.parser, p[1]) )

    def p_basictext_2( self, p ):
        """basictext    : SPECIALCHARS"""
        p[0] = BasicText( p.parser, SPECIALCHARS(p.parser, p[1]) )

    def p_basictext_3( self, p ):
        """basictext    : HTTP_URI"""
        p[0] = BasicText( p.parser, HTTP_URI(p.parser, p[1]) )

    def p_basictext_4( self, p ):
        """basictext    : HTTPS_URI"""
        p[0] = BasicText( p.parser, HTTPS_URI(p.parser, p[1]) )

    def p_basictext_5( self, p ):
        """basictext    : WWW_URI"""
        p[0] = BasicText( p.parser, WWW_URI(p.parser, p[1]) )

    def p_basictext_6( self, p ):
        """basictext    : LINEBREAK"""
        p[0] = BasicText( p.parser, LINEBREAK(p.parser, p[1]) )

    def p_markuptext_1( self, p ):
        """markuptext   : M_SPAN"""
        p[0] = MarkupText( p.parser, M_SPAN(p.parser, p[1]) )

    def p_markuptext_2( self, p ):
        """markuptext   : M_BOLD"""
        p[0] = MarkupText( p.parser, M_BOLD(p.parser, p[1]) )

    def p_markuptext_3( self, p ):
        """markuptext   : M_ITALIC"""
        p[0] = MarkupText( p.parser, M_ITALIC(p.parser, p[1]) )

    def p_markuptext_4( self, p ):
        """markuptext   : M_UNDERLINE"""
        p[0] = MarkupText( p.parser, M_UNDERLINE(p.parser, p[1]) )

    def p_markuptext_5( self, p ):
        """markuptext   : M_SUPERSCRIPT"""
        p[0] = MarkupText( p.parser, M_SUPERSCRIPT(p.parser, p[1]) )

    def p_markuptext_6( self, p ):
        """markuptext   : M_SUBSCRIPT"""
        p[0] = MarkupText( p.parser, M_SUBSCRIPT(p.parser, p[1]) )

    def p_markuptext_7( self, p ):
        """markuptext   : M_BOLDITALIC"""
        p[0] = MarkupText( p.parser, M_BOLDITALIC(p.parser, p[1]) )

    def p_markuptext_8( self, p ):
        """markuptext   : M_BOLDUNDERLINE"""
        p[0] = MarkupText( p.parser, M_BOLDUNDERLINE(p.parser, p[1]) )

    def p_markuptext_9( self, p ):
        """markuptext   : M_ITALICUNDERLINE"""
        p[0] = MarkupText( p.parser, M_ITALICUNDERLINE(p.parser, p[1]) )

    def p_markuptext_10( self, p ):
        """markuptext   : M_BOLDITALICUNDERLINE"""
        p[0] = MarkupText( p.parser, M_BOLDITALICUNDERLINE(p.parser, p[1]) )

    #---- sub-grammars

    def p_paragraph_seperator_1( self, p ) :
        """paragraph_separator  : NEWLINE"""
        p[0] = ParagraphSeparator( p.parser, None, NEWLINE(p.parser, p[1]) )

    def p_paragraph_seperator_2( self, p ) :
        """paragraph_separator  : paragraph_separator NEWLINE"""
        p[0] = ParagraphSeparator( p.parser, p[1], NEWLINE(p.parser, p[2]) )

    def p_error( self, p ):
        if p:
            column = self.etlex._find_tok_column( p )
            self._parse_error( 'before: %s ' % p.value,
                               self._coord(p.lineno, column) )
        else:
            self._parse_error( 'At end of input', '' )