ebnf.py 36 KB
Newer Older
1
"""ebnf.py - EBNF -> Python-Parser compilation for DHParser
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Copyright 2016  by Eckhart Arnold (arnold@badw.de)
                Bavarian Academy of Sciences an Humanities (badw.de)

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied.  See the License for the specific language governing
permissions and limitations under the License.
"""

19
import keyword
20
from collections import OrderedDict
21
from functools import partial
22

23 24 25 26
try:
    import regex as re
except ImportError:
    import re
27
try:
Eckhart Arnold's avatar
Eckhart Arnold committed
28
    from typing import Callable, Dict, List, Set, Tuple, Union
29
except ImportError:
Eckhart Arnold's avatar
Eckhart Arnold committed
30
    from .typing34 import Callable, Dict, List, Set, Tuple, Union
31

32
from DHParser.toolkit import load_if_file, escape_re, md5, sane_parser_name
33
from DHParser.parser import Grammar, mixin_comment, nil_preprocessor, Forward, RE, NegativeLookahead, \
34
    Alternative, Series, Option, Required, OneOrMore, ZeroOrMore, Token, Compiler, \
35
    PreprocessorFunc
36 37
from DHParser.syntaxtree import Node, TransformationFunc, WHITESPACE_PTYPE, TOKEN_PTYPE
from DHParser.error import Error
38
from DHParser.transform import TransformationDict, traverse, remove_brackets, \
39
    reduce_single_child, replace_by_single_child, remove_expendables, \
40
    remove_tokens, flatten, forbid, assert_content, remove_infix_operator
41
from DHParser.versionnumber import __version__
42

43
__all__ = ('get_ebnf_preprocessor',
44 45 46 47
           'get_ebnf_grammar',
           'get_ebnf_transformer',
           'get_ebnf_compiler',
           'EBNFGrammar',
48
           'EBNFTransform',
Eckhart Arnold's avatar
Eckhart Arnold committed
49
           'EBNFCompilerError',
50
           'EBNFCompiler',
51
           'grammar_changed',
52
           'PreprocessorFactoryFunc',
53 54
           'ParserFactoryFunc',
           'TransformerFactoryFunc',
55
           'CompilerFactoryFunc')
56 57


Eckhart Arnold's avatar
Eckhart Arnold committed
58 59 60 61 62 63 64
########################################################################
#
# EBNF scanning
#
########################################################################


65 66
def get_ebnf_preprocessor() -> PreprocessorFunc:
    return nil_preprocessor
Eckhart Arnold's avatar
Eckhart Arnold committed
67 68 69 70 71 72 73 74


########################################################################
#
# EBNF parsing
#
########################################################################

75

76
class EBNFGrammar(Grammar):
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105
    r"""Parser for an EBNF source file, with this grammar:

    # EBNF-Grammar in EBNF

    @ comment    =  /#.*(?:\n|$)/                    # comments start with '#' and eat all chars up to and including '\n'
    @ whitespace =  /\s*/                            # whitespace includes linefeed
    @ literalws  =  right                            # trailing whitespace of literals will be ignored tacitly

    syntax     =  [~//] { definition | directive } §EOF
    definition =  symbol §"=" expression
    directive  =  "@" §symbol §"=" ( regexp | literal | list_ )

    expression =  term { "|" term }
    term       =  { factor }+
    factor     =  [flowmarker] [retrieveop] symbol !"="   # negative lookahead to be sure it's not a definition
                | [flowmarker] literal
                | [flowmarker] regexp
                | [flowmarker] group
                | [flowmarker] oneormore
                | repetition
                | option

    flowmarker =  "!"  | "&"  | "§" |                # '!' negative lookahead, '&' positive lookahead, '§' required
                  "-!" | "-&"                        # '-' negative lookbehind, '-&' positive lookbehind
    retrieveop =  "::" | ":"                         # '::' pop, ':' retrieve

    group      =  "(" expression §")"
    oneormore  =  "{" expression "}+"
    repetition =  "{" expression §"}"
106 107
    option     =  "[" expression §"]"

108 109 110 111 112 113
    symbol     =  /(?!\d)\w+/~                       # e.g. expression, factor, parameter_list
    literal    =  /"(?:[^"]|\\")*?"/~                # e.g. "(", '+', 'while'
                | /'(?:[^']|\\')*?'/~                # whitespace following literals will be ignored tacitly.
    regexp     =  /~?\/(?:[^\/]|(?<=\\)\/)*\/~?/~    # e.g. /\w+/, ~/#.*(?:\n|$)/~
                                                     # '~' is a whitespace-marker, if present leading or trailing
                                                     # whitespace of a regular expression will be ignored tacitly.
114
    list_      =  /\w+/~ { "," /\w+/~ }              # comma separated list of symbols, e.g. BEGIN_LIST, END_LIST,
115 116 117 118
                                                     # BEGIN_QUOTE, END_QUOTE ; see CommonMark/markdown.py for an exmaple
    EOF =  !/./
    """
    expression = Forward()
119
    source_hash__ = "a410e1727fb7575e98ff8451dbf8f3bd"
120
    parser_initialization__ = "upon instantiation"
121 122
    COMMENT__ = r'#.*(?:\n|$)'
    WSP__ = mixin_comment(whitespace=r'\s*', comment=r'#.*(?:\n|$)')
123
    wspL__ = ''
124
    wspR__ = WSP__
125
    EOF = NegativeLookahead(RE('.', wR=''))
126
    list_ = Series(RE('\\w+'), ZeroOrMore(Series(Token(","), RE('\\w+'))))
127
    regexp = RE(r'~?/(?:\\/|[^/])*?/~?')  # RE('~?/(?:[^/]|(?<=\\\\)/)*/~?')
128 129
    literal = Alternative(RE('"(?:[^"]|\\\\")*?"'), RE("'(?:[^']|\\\\')*?'"))
    symbol = RE('(?!\\d)\\w+')
130 131 132 133
    option = Series(Token("["), expression, Required(Token("]")))
    repetition = Series(Token("{"), expression, Required(Token("}")))
    oneormore = Series(Token("{"), expression, Token("}+"))
    group = Series(Token("("), expression, Required(Token(")")))
134 135
    retrieveop = Alternative(Token("::"), Token(":"))
    flowmarker = Alternative(Token("!"), Token("&"), Token("§"), Token("-!"), Token("-&"))
136 137 138
    factor = Alternative(Series(Option(flowmarker), Option(retrieveop), symbol, NegativeLookahead(Token("="))),
                         Series(Option(flowmarker), literal), Series(Option(flowmarker), regexp),
                         Series(Option(flowmarker), group), Series(Option(flowmarker), oneormore),
139
                         repetition, option)
140
    term = OneOrMore(factor)
141 142 143
    expression.set(Series(term, ZeroOrMore(Series(Token("|"), term))))
    directive = Series(Token("@"), Required(symbol), Required(Token("=")), Alternative(regexp, literal, list_))
    definition = Series(symbol, Required(Token("=")), expression)
144
    syntax = Series(Option(RE('', wR='', wL=WSP__)), ZeroOrMore(Alternative(definition, directive)), Required(EOF))
145 146 147
    root__ = syntax


148
def grammar_changed(grammar_class, grammar_source: str) -> bool:
Eckhart Arnold's avatar
Eckhart Arnold committed
149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167
    """Returns ``True`` if ``grammar_class`` does not reflect the latest
    changes of ``grammar_source``

    Parameters:
        grammar_class:  the parser class representing the grammar
            or the file name of a compiler suite containing the grammar
        grammar_source:  File name or string representation of the
            EBNF code of the grammar

    Returns (bool):
        True, if the source text of the grammar is different from the
        source from which the grammar class was generated
    """
    grammar = load_if_file(grammar_source)
    chksum = md5(grammar, __version__)
    if isinstance(grammar_class, str):
        # grammar_class = load_compiler_suite(grammar_class)[1]
        with open(grammar_class, 'r', encoding='utf8') as f:
            pycode = f.read()
168
        m = re.search('class \w*\(Grammar\)', pycode)
Eckhart Arnold's avatar
Eckhart Arnold committed
169 170 171 172 173 174 175 176 177 178
        if m:
            m = re.search('    source_hash__ *= *"([a-z0-9]*)"',
                          pycode[m.span()[1]:])
            return not (m and m.groups() and m.groups()[-1] == chksum)
        else:
            return True
    else:
        return chksum != grammar_class.source_hash__


179
def get_ebnf_grammar() -> EBNFGrammar:
Eckhart Arnold's avatar
Eckhart Arnold committed
180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195
    global thread_local_ebnf_grammar_singleton
    try:
        grammar = thread_local_ebnf_grammar_singleton
        return grammar
    except NameError:
        thread_local_ebnf_grammar_singleton = EBNFGrammar()
        return thread_local_ebnf_grammar_singleton


########################################################################
#
# EBNF concrete to abstract syntax tree transformation and validation
#
########################################################################


196
EBNF_AST_transformation_table = {
197
    # AST Transformations for EBNF-grammar
198
    "+":
199
        remove_expendables,
200
    "syntax":
201
        [],  # otherwise '"*": replace_by_single_child' would be applied
202
    "directive, definition":
203
        remove_tokens('@', '='),
Eckhart Arnold's avatar
Eckhart Arnold committed
204
    "expression":
205
        [replace_by_single_child, flatten, remove_tokens('|')],  # remove_infix_operator],
206 207 208 209 210
    "term":
        [replace_by_single_child, flatten],  # supports both idioms:  "{ factor }+" and "factor { factor }"
    "factor, flowmarker, retrieveop":
        replace_by_single_child,
    "group":
211
        [remove_brackets, replace_by_single_child],
212
    "oneormore, repetition, option":
213 214
        [reduce_single_child, remove_brackets,
         forbid('repetition', 'option', 'oneormore'), assert_content(r'(?!§)')],
215
    "symbol, literal, regexp":
216
        reduce_single_child,
217
    (TOKEN_PTYPE, WHITESPACE_PTYPE):
218
        reduce_single_child,
219
    "list_":
220
        [flatten, remove_infix_operator],
221
    "*":
222
        replace_by_single_child
223 224
}

225

Eckhart Arnold's avatar
Eckhart Arnold committed
226
def EBNFTransform() -> TransformationFunc:
227
    return partial(traverse, processing_table=EBNF_AST_transformation_table.copy())
di68kap's avatar
di68kap committed
228

229
def get_ebnf_transformer() -> TransformationFunc:
230 231 232 233 234 235 236
    global thread_local_EBNF_transformer_singleton
    try:
        transformer = thread_local_EBNF_transformer_singleton
    except NameError:
        thread_local_EBNF_transformer_singleton = EBNFTransform()
        transformer = thread_local_EBNF_transformer_singleton
    return transformer
Eckhart Arnold's avatar
Eckhart Arnold committed
237 238 239 240 241 242 243 244


########################################################################
#
# EBNF abstract syntax tree to Python parser compilation
#
########################################################################

245

246
PreprocessorFactoryFunc = Callable[[], PreprocessorFunc]
247
ParserFactoryFunc = Callable[[], Grammar]
248
TransformerFactoryFunc = Callable[[], TransformationFunc]
249 250
CompilerFactoryFunc = Callable[[], Compiler]

251 252 253
PREPROCESSOR_FACTORY = '''
def get_preprocessor() -> PreprocessorFunc:
    return {NAME}Preprocessor
254 255 256 257
'''


GRAMMAR_FACTORY = '''
258
def get_grammar() -> {NAME}Grammar:
259 260 261 262 263
    global thread_local_{NAME}_grammar_singleton
    try:
        grammar = thread_local_{NAME}_grammar_singleton
    except NameError:
        thread_local_{NAME}_grammar_singleton = {NAME}Grammar()
264 265
        grammar = thread_local_{NAME}_grammar_singleton
    return grammar
266 267 268 269
'''


TRANSFORMER_FACTORY = '''
270 271 272
def {NAME}Transform() -> TransformationDict:
    return partial(traverse, processing_table={NAME}_AST_transformation_table.copy())

273
def get_transformer() -> TransformationFunc:
274 275 276 277 278 279 280
    global thread_local_{NAME}_transformer_singleton
    try:
        transformer = thread_local_{NAME}_transformer_singleton
    except NameError:
        thread_local_{NAME}_transformer_singleton = {NAME}Transform()
        transformer = thread_local_{NAME}_transformer_singleton
    return transformer
281 282 283 284
'''


COMPILER_FACTORY = '''
285
def get_compiler(grammar_name="{NAME}", grammar_source="") -> {NAME}Compiler:
286 287 288 289 290 291 292
    global thread_local_{NAME}_compiler_singleton
    try:
        compiler = thread_local_{NAME}_compiler_singleton
        compiler.set_grammar_name(grammar_name, grammar_source)
    except NameError:
        thread_local_{NAME}_compiler_singleton = \\
            {NAME}Compiler(grammar_name, grammar_source)
293 294
        compiler = thread_local_{NAME}_compiler_singleton
    return compiler
295 296
'''

Eckhart Arnold's avatar
Eckhart Arnold committed
297

298 299
class EBNFCompilerError(Exception):
    """Error raised by `EBNFCompiler` class. (Not compilation errors
300
    in the strict sense, see `CompilationError` in module ``dsl.py``)"""
301 302 303
    pass


304
class EBNFCompiler(Compiler):
305 306
    """
    Generates a Parser from an abstract syntax tree of a grammar specified
307
    in EBNF-Notation.
308 309 310 311 312 313 314 315 316 317 318 319 320

    Instances of this class must be called with the root-node of the
    abstract syntax tree from an EBNF-specification of a formal language.
    The returned value is the Python-source-code of a Grammar class for
    this language that can be used to parse texts in this language.
    See classes `parser.Compiler` and `parser.Grammar` for more information.

    Addionally, class EBNFCompiler provides helper methods to generate
    code-skeletons for a preprocessor, AST-transformation and full
    compilation of the formal language. These method's names start with
    the prefix `gen_`.

    Attributes:
321
        current_symbols:  During compilation, a list containing the root
322 323 324 325
                node of the currently compiled definition as first element
                and then the nodes of the symbols that are referred to in
                the currently compiled definition.

326
        rules:  Dictionary that maps rule names to a list of Nodes that
327 328 329 330 331 332 333 334 335
                contain symbol-references in the definition of the rule.
                The first item in the list is the node of the rule-
                definition itself. Example:

                           `alternative = a | b`

                Now `[str(node) for node in self.rules['alternative']]`
                yields `['alternative = a | b', 'a', 'b']`

336
        symbols:  A mapping of symbol names to their first usage (not
337 338
                their definition!) in the EBNF source.

339
        variables:  A set of symbols names that are used with the
340 341 342 343
                Pop or Retrieve operator. Because the values of these
                symbols need to be captured they are called variables.
                See `test_parser.TestPopRetrieve` for an example.

344
        recursive:  A set of symbols that are used recursively and
345 346
                therefore require a `Forward`-operator.

347
        definitions:  A dictionary of definitions. Other than `rules`
348 349
                this maps the symbols to their compiled definienda.

350
        deferred_taks:  A list of callables that is filled during
351 352 353 354 355
                compilatation, but that will be executed only after
                compilation has finished. Typically, it contains
                sementatic checks that require information that
                is only available upon completion of compilation.

356
        root:   The name of the root symbol.
357

358
        directives:  A dictionary of all directives and their default
359
                values.
360 361 362

        re_flags:  A set of regular expression flags to be added to all
                regular expressions found in the current parsing process
363 364
    """
    COMMENT_KEYWORD = "COMMENT__"
365
    WHITESPACE_KEYWORD = "WSP__"
Eckhart Arnold's avatar
Eckhart Arnold committed
366 367
    RAW_WS_KEYWORD = "WHITESPACE__"
    RESERVED_SYMBOLS = {WHITESPACE_KEYWORD, RAW_WS_KEYWORD, COMMENT_KEYWORD}
368
    AST_ERROR = "Badly structured syntax tree. " \
Eckhart Arnold's avatar
Eckhart Arnold committed
369
                "Potentially due to erroneous AST transformation."
370 371 372 373
    PREFIX_TABLE = {'§': 'Required',
                    '&': 'Lookahead', '!': 'NegativeLookahead',
                    '-&': 'Lookbehind', '-!': 'NegativeLookbehind',
                    '::': 'Pop', ':': 'Retrieve'}
374 375 376
    WHITESPACE = {'horizontal': r'[\t ]*',  # default: horizontal
                  'linefeed': r'[ \t]*\n?(?!\s*\n)[ \t]*',
                  'vertical': r'\s*'}
377

378

379
    def __init__(self, grammar_name="", grammar_source=""):
Eckhart Arnold's avatar
Eckhart Arnold committed
380
        super(EBNFCompiler, self).__init__(grammar_name, grammar_source)
381 382
        self._reset()

383

384
    def _reset(self):
385
        super(EBNFCompiler, self)._reset()
386
        self._result = ''           # type: str
387
        self.re_flags = set()       # type: Set[str]
388 389 390
        self.rules = OrderedDict()  # type: OrderedDict[str, List[Node]]
        self.current_symbols = []   # type: List[Node]
        self.symbols = {}           # type: Dict[str, Node]
391 392
        self.variables = set()      # type: Set[str]
        self.recursive = set()      # type: Set[str]
393
        self.definitions = {}       # type: Dict[str, str]
394
        self.deferred_tasks = []    # type: List[Callable]
395
        self.root_symbol = ""  # type: str
396
        self.directives = {'whitespace': self.WHITESPACE['horizontal'],
397
                           'comment': '',
398
                           'literalws': ['right'],
399 400
                           'tokens': set(),  # alt. 'preprocessor_tokens'
                           'filter': dict(),  # alt. 'filter'
Eckhart Arnold's avatar
Eckhart Arnold committed
401
                           'ignorecase': False}
402

Eckhart Arnold's avatar
Eckhart Arnold committed
403
    @property
404
    def result(self) -> str:
Eckhart Arnold's avatar
Eckhart Arnold committed
405 406
        return self._result

407
    # methods for generating skeleton code for preprocessor, transformer, and compiler
408

409
    def gen_preprocessor_skeleton(self) -> str:
410 411 412 413
        """
        Returns Python-skeleton-code for a preprocessor-function for
        the previously compiled formal language.
        """
414
        name = self.grammar_name + "Preprocessor"
415
        return "def %s(text):\n    return text\n" % name \
416
               + PREPROCESSOR_FACTORY.format(NAME=self.grammar_name)
417

418

419
    def gen_transformer_skeleton(self) -> str:
420 421 422 423
        """
        Returns Python-skeleton-code for the AST-transformation for the
        previously compiled formal language.
        """
424
        if not self.rules:
Eckhart Arnold's avatar
Eckhart Arnold committed
425 426
            raise EBNFCompilerError('Compiler must be run before calling '
                                    '"gen_transformer_Skeleton()"!')
427
        tt_name = self.grammar_name + '_AST_transformation_table'
di68kap's avatar
di68kap committed
428
        transtable = [tt_name + ' = {',
Eckhart Arnold's avatar
Eckhart Arnold committed
429
                      '    # AST Transformations for the ' + self.grammar_name + '-grammar']
Eckhart Arnold's avatar
Eckhart Arnold committed
430
        transtable.append('    "+": remove_empty,')
431
        for name in self.rules:
432 433 434 435 436 437 438
            tf = '[]'
            rule = self.definitions[name]
            if rule.startswith('Alternative'):
                tf = '[replace_or_reduce]'
            elif rule.startswith('Synonym'):
                tf = '[replace_by_single_child]'
            transtable.append('    "' + name + '": %s,' % tf)
439
        transtable.append('    ":Token, :RE": reduce_single_child,')
440
        transtable += ['    "*": replace_by_single_child', '}', '']
441
        transtable += [TRANSFORMER_FACTORY.format(NAME=self.grammar_name)]
442 443
        return '\n'.join(transtable)

444

445
    def gen_compiler_skeleton(self) -> str:
446 447 448 449
        """
        Returns Python-skeleton-code for a Compiler-class for the
        previously compiled formal language.
        """
450
        if not self.rules:
451 452
            raise EBNFCompilerError('Compiler has not been run before calling '
                                    '"gen_Compiler_Skeleton()"!')
453
        compiler = ['class ' + self.grammar_name + 'Compiler(Compiler):',
454 455 456 457
                    '    """Compiler for the abstract-syntax-tree of a ' +
                    self.grammar_name + ' source file.',
                    '    """', '',
                    '    def __init__(self, grammar_name="' +
Eckhart Arnold's avatar
Eckhart Arnold committed
458
                    self.grammar_name + '", grammar_source=""):',
459
                    '        super(' + self.grammar_name +
Eckhart Arnold's avatar
Eckhart Arnold committed
460
                    'Compiler, self).__init__(grammar_name, grammar_source)',
461
                    "        assert re.match('\w+\Z', grammar_name)", '']
462
        for name in self.rules:
463
            method_name = Compiler.method_name(name)
464
            if name == self.root_symbol:
465
                compiler += ['    def ' + method_name + '(self, node):',
466 467
                             '        return node', '']
            else:
468
                compiler += ['    def ' + method_name + '(self, node):',
469
                             '        pass', '']
470
        compiler += [COMPILER_FACTORY.format(NAME=self.grammar_name)]
471
        return '\n'.join(compiler)
472

473

474 475 476 477 478
    def assemble_parser(self, definitions: List[Tuple[str, str]], root_node: Node) -> str:
        """
        Creates the Python code for the parser after compilation of
        the EBNF-Grammar
        """
479 480 481 482 483 484 485 486 487 488

        # execute deferred tasks, for example semantic checks that cannot
        # be done before the symbol table is complete

        for task in self.deferred_tasks:
            task()

        # provide for capturing of symbols that are variables, i.e. the
        # value of will be retrieved at some point during the parsing process

489 490 491
        if self.variables:
            for i in range(len(definitions)):
                if definitions[i][0] in self.variables:
492
                    definitions[i] = (definitions[i][0], 'Capture(%s)' % definitions[i][1])
493

494 495
        # add special fields for Grammar class

496
        definitions.append(('wspR__', self.WHITESPACE_KEYWORD
Eckhart Arnold's avatar
Eckhart Arnold committed
497
                            if 'right' in self.directives['literalws'] else "''"))
498
        definitions.append(('wspL__', self.WHITESPACE_KEYWORD
Eckhart Arnold's avatar
Eckhart Arnold committed
499
                            if 'left' in self.directives['literalws'] else "''"))
500
        definitions.append((self.WHITESPACE_KEYWORD,
Eckhart Arnold's avatar
Eckhart Arnold committed
501 502 503
                            ("mixin_comment(whitespace=" + self.RAW_WS_KEYWORD +
                             ", comment=" + self.COMMENT_KEYWORD + ")")))
        definitions.append((self.RAW_WS_KEYWORD, "r'{whitespace}'".format(**self.directives)))
504 505 506 507
        definitions.append((self.COMMENT_KEYWORD, "r'{comment}'".format(**self.directives)))

        # prepare parser class header and docstring and
        # add EBNF grammar to the doc string of the parser class
508

509
        article = 'an ' if self.grammar_name[0:1] in "AaEeIiOoUu" else 'a '  # what about 'hour', 'universe' etc.?
510
        declarations = ['class ' + self.grammar_name +
511
                        'Grammar(Grammar):',
512 513
                        'r"""Parser for ' + article + self.grammar_name +
                        ' source file' +
514
                        (', with this grammar:' if self.grammar_source else '.')]
515
        definitions.append(('parser_initialization__', '"upon instantiation"'))
516
        if self.grammar_source:
517
            definitions.append(('source_hash__',
518
                                '"%s"' % md5(self.grammar_source, __version__)))
519
            declarations.append('')
520
            declarations += [line for line in self.grammar_source.split('\n')]
521 522 523 524 525
            while declarations[-1].strip() == '':
                declarations = declarations[:-1]
        declarations.append('"""')

        # turn definitions into declarations in reverse order
526

527
        self.root_symbol = definitions[0][0] if definitions else ""
528 529 530 531 532 533 534 535
        definitions.reverse()
        declarations += [symbol + ' = Forward()'
                         for symbol in sorted(list(self.recursive))]
        for symbol, statement in definitions:
            if symbol in self.recursive:
                declarations += [symbol + '.set(' + statement + ')']
            else:
                declarations += [symbol + ' = ' + statement]
536 537 538 539 540 541 542

        # check for symbols used but never defined

        defined_symbols = set(self.rules.keys()) | self.RESERVED_SYMBOLS
        for symbol in self.symbols:
            if symbol not in defined_symbols:
                self.symbols[symbol].add_error("Missing definition for symbol '%s'" % symbol)
543
                # root_node.error_flag = True
544 545 546

        # check for unconnected rules

Eckhart Arnold's avatar
Eckhart Arnold committed
547 548 549 550 551 552 553 554 555 556 557 558
        defined_symbols.difference_update(self.RESERVED_SYMBOLS)

        def remove_connections(symbol):
            if symbol in defined_symbols:
                defined_symbols.remove(symbol)
                for related in self.rules[symbol][1:]:
                    remove_connections(str(related))

        remove_connections(self.root_symbol)
        for leftover in defined_symbols:
            self.rules[leftover][0].add_error(('Rule "%s" is not connected to '
                'parser root "%s" !') % (leftover, self.root_symbol), Error.WARNING)
559

560
        # set root_symbol parser and assemble python grammar definition
561

562 563
        if self.root_symbol and 'root__' not in self.rules:
            declarations.append('root__ = ' + self.root_symbol)
564
        declarations.append('')
Eckhart Arnold's avatar
Eckhart Arnold committed
565 566 567
        self._result = '\n    '.join(declarations) \
                       + GRAMMAR_FACTORY.format(NAME=self.grammar_name)
        return self._result
568

569 570 571

    ## compilation methods

572
    def on_syntax(self, node: Node) -> str:
573
        definitions = []  # type: List[Tuple[str, str]]
574 575

        # drop the wrapping sequence node
576 577
        if len(node.children) == 1 and not node.children[0].parser.name:
            node = node.children[0]
578 579

        # compile definitions and directives and collect definitions
580
        for nd in node.children:
581
            if nd.parser.name == "definition":
582
                definitions.append(self.compile(nd))
583
            else:
584
                assert nd.parser.name == "directive", nd.as_sxpr()
585
                self.compile(nd)
586
                node.error_flag = max(node.error_flag, nd.error_flag)
587
        self.definitions.update(definitions)
588

589
        return self.assemble_parser(definitions, node)
590

591

592
    def on_definition(self, node: Node) -> Tuple[str, str]:
593
        rule = str(node.children[0])
594
        if rule in self.rules:
Eckhart Arnold's avatar
Eckhart Arnold committed
595 596 597 598 599
            first = self.rules[rule][0]
            if not first._errors:
                first.add_error('First definition of rule "%s" '
                                'followed by illegal redefinitions.' % rule)
            node.add_error('A rule with name "%s" has already been defined earlier.' % rule)
600
        elif rule in EBNFCompiler.RESERVED_SYMBOLS:
601 602 603 604
            node.add_error('Symbol "%s" is a reserved symbol.' % rule)
        elif not sane_parser_name(rule):
            node.add_error('Illegal symbol "%s". Symbols must not start or '
                           ' end with a doube underscore "__".' % rule)
605
        elif rule in self.directives['tokens']:
606
            node.add_error('Symbol "%s" has already been defined as '
607
                           'a preprocessor token.' % rule)
608 609
        elif keyword.iskeyword(rule):
            node.add_error('Python keyword "%s" may not be used as a symbol. '
610
                           % rule + '(This may change in the future.)')
611
        try:
612 613
            self.current_symbols = [node]
            self.rules[rule] = self.current_symbols
614
            defn = self.compile(node.children[1])
615
            if rule in self.variables:
616
                defn = 'Capture(%s)' % defn
617
                self.variables.remove(rule)
618 619 620
            elif defn.find("(") < 0:
                # assume it's a synonym, like 'page = REGEX_PAGE_NR'
                defn = 'Synonym(%s)' % defn
621
        except TypeError as error:
622
            errmsg = EBNFCompiler.AST_ERROR + " (" + str(error) + ")\n" + node.as_sxpr()
623 624
            node.add_error(errmsg)
            rule, defn = rule + ':error', '"' + errmsg + '"'
Eckhart Arnold's avatar
Eckhart Arnold committed
625
        return rule, defn
626

627

628
    def _check_rx(self, node: Node, rx: str) -> str:
629 630
        """
        Checks whether the string `rx` represents a valid regular
631 632 633
        expression. Makes sure that multiline regular expressions are
        prepended by the multiline-flag. Returns the regular expression string.
        """
634
        flags = self.re_flags | {'x'} if rx.find('\n') >= 0 else self.re_flags
635
        if flags:  rx = "(?%s)%s" % ("".join(flags), rx)
636 637 638 639 640 641 642
        try:
            re.compile(rx)
        except Exception as re_error:
            node.add_error("malformed regular expression %s: %s" %
                           (repr(rx), str(re_error)))
        return rx

643

644
    def on_directive(self, node: Node) -> str:
645
        key = str(node.children[0]).lower()
646
        assert key not in self.directives['tokens']
647

648
        if key in {'comment', 'whitespace'}:
649 650
            if node.children[1].parser.name == "list_":
                if len(node.children[1].result) != 1:
Eckhart Arnold's avatar
Eckhart Arnold committed
651
                    node.add_error('Directive "%s" must have one, but not %i values.' %
652
                                   (key, len(node.children[1].result)))
653
                value = self.compile(node.children[1]).pop()
654 655
                if key == 'whitespace' and value in EBNFCompiler.WHITESPACE:
                    value = EBNFCompiler.WHITESPACE[value]  # replace whitespace-name by regex
656
                else:
657
                    node.add_error('Value "%s" not allowed for directive "%s".' % (value, key))
658
            else:
659 660
                value = str(node.children[1]).strip("~")  # cast(str, node.children[1].result).strip("~")
                if value != str(node.children[1]):  # cast(str, node.children[1].result):
661 662 663 664 665 666
                    node.add_error("Whitespace marker '~' not allowed in definition of "
                                   "%s regular expression." % key)
                if value[0] + value[-1] in {'""', "''"}:
                    value = escape_re(value[1:-1])
                elif value[0] + value[-1] == '//':
                    value = self._check_rx(node, value[1:-1])
667 668 669
                if key == 'whitespace' and not re.match(value, ''):
                    node.add_error("Implicit whitespace should always match the empty string, "
                                   "/%s/ does not." % value)
670
            self.directives[key] = value
671

672 673 674 675 676 677
        elif key == 'ignorecase':
            value = str(node.children[1]).lower() not in {"off", "false", "no"}
            self.directives['ignorecase'] == value
            if value:
                self.re_flags.add('i')

Eckhart Arnold's avatar
Eckhart Arnold committed
678 679 680
        # elif key == 'testing':
        #     value = str(node.children[1])
        #     self.directives['testing'] = value.lower() not in {"off", "false", "no"}
681

682
        elif key == 'literalws':
683
            value = {item.lower() for item in self.compile(node.children[1])}
684
            if (len(value - {'left', 'right', 'both', 'none'}) > 0
Eckhart Arnold's avatar
Eckhart Arnold committed
685
                    or ('none' in value and len(value) > 1)):
686 687 688 689 690 691 692
                node.add_error('Directive "literalws" allows the values '
                               '`left`, `right`, `both` or `none`, '
                               'but not `%s`' % ", ".join(value))
            ws = {'left', 'right'} if 'both' in value \
                else {} if 'none' in value else value
            self.directives[key] = list(ws)

693
        elif key in {'tokens', 'preprocessor_tokens'}:
694
            self.directives['tokens'] |= self.compile(node.children[1])
695

696
        elif key.endswith('_filter'):
697
            filter_set = self.compile(node.children[1])
698 699 700 701
            if not isinstance(filter_set, set) or len(filter_set) != 1:
                node.add_error('Directive "%s" accepts exactly on symbol, not %s'
                               % (key, str(filter_set)))
            self.directives['filter'][key[:-7]] = filter_set.pop()
702

703 704 705
        else:
            node.add_error('Unknown directive %s ! (Known ones are %s .)' %
                           (key,
706
                            ', '.join(list(self.directives.keys()))))
707 708
        return ""

709

710
    def non_terminal(self, node: Node, parser_class: str, custom_args: List[str]=[]) -> str:
711 712
        """
        Compiles any non-terminal, where `parser_class` indicates the Parser class
713 714
        name for the particular non-terminal.
        """
715
        arguments = [self.compile(r) for r in node.children] + custom_args
716 717
        return parser_class + '(' + ', '.join(arguments) + ')'

718

719
    def on_expression(self, node) -> str:
720 721
        return self.non_terminal(node, 'Alternative')

722

723
    def on_term(self, node) -> str:
724
        return self.non_terminal(node, 'Series')
725

726

727
    def on_factor(self, node: Node) -> str:
728
        assert node.children
729
        assert len(node.children) >= 2, node.as_sxpr()
730
        prefix = str(node.children[0])  # cast(str, node.children[0].result)
731
        custom_args = []  # type: List[str]
732 733

        if prefix in {'::', ':'}:
734 735
            assert len(node.children) == 2
            arg = node.children[-1]
736
            if arg.parser.name != 'symbol':
Eckhart Arnold's avatar
Eckhart Arnold committed
737
                node.add_error(('Retrieve Operator "%s" requires a symbol, '
738 739
                                'and not a %s.') % (prefix, str(arg.parser)))
                return str(arg.result)
740
            if str(arg) in self.directives['filter']:
741
                custom_args = ['filter=%s' % self.directives['filter'][str(arg)]]
742
            self.variables.add(str(arg))  # cast(str, arg.result)
743

744
        elif len(node.children) > 2:
745 746
            # shift = (Node(node.parser, node.result[1].result),)
            # node.result[1].result = shift + node.result[2:]
747 748 749 750
            node.children[1].result = (Node(node.children[1].parser, node.children[1].result),) \
                                    + node.children[2:]
            node.children[1].parser = node.parser
            node.result = (node.children[0], node.children[1])
751

752
        node.result = node.children[1:]
753 754
        try:
            parser_class = self.PREFIX_TABLE[prefix]
755 756 757 758
            result = self.non_terminal(node, parser_class, custom_args)
            if prefix[:1] == '-':
                def check(node):
                    nd = node
759 760 761 762 763 764 765 766 767 768
                    if len(nd.children) >= 1:
                        nd = nd.children[0]
                    while nd.parser.name == "symbol":
                        symlist = self.rules.get(str(nd), [])
                        if len(symlist) == 2:
                            nd = symlist[1]
                        else:
                            if len(symlist) == 1:
                                nd = symlist[0].children[1]
                            break
769 770 771
                    if (nd.parser.name != "regexp" or str(nd)[:1] != '/'
                        or str(nd)[-1:] != '/'):
                        node.add_error("Lookbehind-parser can only be used with plain RegExp-"
772
                                       "parsers, not with: " + nd.parser.name + nd.parser.ptype)
773 774 775 776

                if not result.startswith('RegExp('):
                    self.deferred_tasks.append(lambda: check(node))
            return result
777 778
        except KeyError:
            node.add_error('Unknown prefix "%s".' % prefix)
779
        return ""
780

781

782
    def on_option(self, node) -> str:
783
        return self.non_terminal(node, 'Option')
784

785

786
    def on_repetition(self, node) -> str:
787 788
        return self.non_terminal(node, 'ZeroOrMore')

789

790
    def on_oneormore(self, node) -> str:
791 792
        return self.non_terminal(node, 'OneOrMore')

793

794
    def on_regexchain(self, node) -> str:
795 796
        raise EBNFCompilerError("Not yet implemented!")

797

798
    def on_group(self, node) -> str:
799 800 801
        raise EBNFCompilerError("Group nodes should have been eliminated by "
                                "AST transformation!")

802

803 804 805
    def on_symbol(self, node: Node) -> str:     # called only for symbols on the right hand side!
        symbol = str(node)  # ; assert result == cast(str, node.result)
        if symbol in self.directives['tokens']:
806
            return 'PreprocessorToken("' + symbol + '")'
807
        else:
808 809
            self.current_symbols.append(node)
            if symbol not in self.symbols:
810
                self.symbols[symbol] = node  # remember first use of symbol
811 812
            if symbol in self.rules:
                self.recursive.add(symbol)
Eckhart Arnold's avatar
Eckhart Arnold committed
813
            if symbol in EBNFCompiler.RESERVED_SYMBOLS:  # (EBNFCompiler.WHITESPACE_KEYWORD, EBNFCompiler.COMMENT_KEYWORD):
Eckhart Arnold's avatar
Eckhart Arnold committed
814
                return "RegExp(%s)" % symbol
815
            return symbol
816

817

818
    def on_literal(self, node) -> str:
819
        return 'Token(' + str(node).replace('\\', r'\\') + ')'  # return 'Token(' + ', '.merge_children([node.result]) + ')' ?
820

821

822
    def on_regexp(self, node: Node) -> str:
823
        rx = str(node)
824
        name = []   # type: List[str]
825 826 827 828 829 830 831 832 833 834 835 836