parse.py 103 KB
Newer Older
1
# parse.py - parser combinators for DHParser
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
#
# 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.
Eckhart Arnold's avatar
Eckhart Arnold committed
17

18
19

"""
20
Module ``parse`` contains the python classes and functions for
21
22
DHParser's packrat-parser. It's central class is the
``Grammar``-class, which is the base class for any concrete
23
24
25
26
27
28
29
30
31
Grammar. Grammar-objects are callable and parsing is done by
calling a Grammar-object with a source text as argument.

The different parsing functions are callable descendants of class
``Parser``. Usually, they are organized in a tree and defined
within the namespace of a grammar-class. See ``ebnf.EBNFGrammar``
for an example.
"""

32

33
from collections import defaultdict
34
import copy
eckhart's avatar
eckhart committed
35
from typing import Callable, cast, List, Tuple, Set, Dict, DefaultDict, Union, Optional, Any
36

eckhart's avatar
eckhart committed
37
from DHParser.configuration import get_config_value
38
from DHParser.error import Error, linebreaks, line_col
39
40
from DHParser.log import is_logging, HistoryRecord
from DHParser.preprocess import BEGIN_TOKEN, END_TOKEN, RX_TOKEN_NAME
41
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
42
43
from DHParser.syntaxtree import Node, FrozenNode, RootNode, WHITESPACE_PTYPE, \
    TOKEN_PTYPE, ZOMBIE_TAG, ResultType
eckhart's avatar
eckhart committed
44
from DHParser.toolkit import sane_parser_name, escape_control_characters, re, cython
45

46

47
__all__ = ('Parser',
48
           'UnknownParserError',
49
50
           'GrammarErrorType',
           'GrammarError',
51
           'Grammar',
Eckhart Arnold's avatar
Eckhart Arnold committed
52
           'EMPTY_NODE',
53
           'PreprocessorToken',
54
           'Token',
55
           'DropToken',
56
           'RegExp',
57
58
           'RE',
           'TKN',
59
           'Whitespace',
60
           'DropWhitespace',
61
           'mixin_comment',
62
           'MetaParser',
eckhart's avatar
eckhart committed
63
64
           'UnaryParser',
           'NaryParser',
65
66
67
68
69
70
71
72
73
           'Synonym',
           'Option',
           'ZeroOrMore',
           'OneOrMore',
           'Series',
           'Alternative',
           'AllOf',
           'SomeOf',
           'Unordered',
eckhart's avatar
eckhart committed
74
           'Required',
75
76
77
78
79
80
81
82
83
84
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
           'last_value',
           'counterpart',
           'accumulate',
           'Capture',
           'Retrieve',
           'Pop',
85
           'Forward')
86
87
88
89


########################################################################
#
90
# Parser base class
91
92
93
94
#
########################################################################


95
EMPTY_NODE = FrozenNode(':EMPTY__', '')
96
97


Eckhart Arnold's avatar
Eckhart Arnold committed
98
99
100
class ParserError(Exception):
    """
    A `ParserError` is thrown for those parser errors that allow the
101
    controlled re-entrance of the parsing process after the error occurred.
Eckhart Arnold's avatar
Eckhart Arnold committed
102
    If a reentry-rule has been configured for the parser where the error
103
    occurred, the parser guard can resume the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
104
105
106
107
108

    Currently, the only case when a `ParserError` is thrown (and not some
    different kind of error like `UnknownParserError`, is when a `Series`-
    detects a missing mandatory element.
    """
109
110
111
112
    def __init__(self, node: Node, rest: StringView, error: Optional[Error], first_throw: bool):
        self.node = node   # type: Node
        self.rest = rest   # type: StringView
        self.error = error # type: Optional[Error]
113
        self.first_throw = first_throw  # type: bool
Eckhart Arnold's avatar
Eckhart Arnold committed
114

115
    def __str__(self):
116
        return "%i: %s    %s" % (self.node.pos, str(self.rest[:25]), repr(self.node))
117

Eckhart Arnold's avatar
Eckhart Arnold committed
118

119
ResumeList = List[Union[str, Any]]  # list of strings or regular expressiones
Eckhart Arnold's avatar
Eckhart Arnold committed
120
121


122
def reentry_point(rest: StringView, rules: ResumeList) -> int:
Eckhart Arnold's avatar
Eckhart Arnold committed
123
124
125
126
127
128
129
130
131
    """
    Finds the point where parsing should resume after a ParserError has been caught.
    Args:
        rest:  The rest of the parsed text or, in other words, the point where
                a ParserError was thrown.
        rules: A list of strings or regular expressions. The rest of the text is
                searched for each of these. The closest match is the point where
                parsing will be resumed.
    Returns:
132
133
        The integer index of the closest reentry point or -1 if no reentry-point
        was found.
Eckhart Arnold's avatar
Eckhart Arnold committed
134
    """
135
    upper_limit = len(rest) + 1
Eckhart Arnold's avatar
Eckhart Arnold committed
136
137
138
139
    i = upper_limit
    #find closest match
    for rule in rules:
        if isinstance(rule, str):
140
141
            k = rest.find(rule)
            i = min(k if k >= 0 else upper_limit, i)
Eckhart Arnold's avatar
Eckhart Arnold committed
142
143
144
        else:
            m = rest.search(rule)
            if m:
145
                i = min(rest.index(m.start()), i)
146
    # in case no rule matched return -1
147
    if i == upper_limit:
148
149
        i = -1
    return i
Eckhart Arnold's avatar
Eckhart Arnold committed
150
151


eckhart's avatar
eckhart committed
152
153
154
155
ApplyFunc = Callable[['Parser'], None]
FlagFunc = Callable[[ApplyFunc, Set[ApplyFunc]], bool]


156
class Parser:
157
158
159
160
161
162
163
164
165
166
167
168
169
170
    """
    (Abstract) Base class for Parser combinator parsers. Any parser
    object that is actually used for parsing (i.e. no mock parsers)
    should should be derived from this class.

    Since parsers can contain other parsers (see classes UnaryOperator
    and NaryOperator) they form a cyclical directed graph. A root
    parser is a parser from which all other parsers can be reached.
    Usually, there is one root parser which serves as the starting
    point of the parsing process. When speaking of "the root parser"
    it is this root parser object that is meant.

    There are two different types of parsers:

di68kap's avatar
di68kap committed
171
    1. *Named parsers* for which a name is set in field `parser.pname`.
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
       The results produced by these parsers can later be retrieved in
       the AST by the parser name.

    2. *Anonymous parsers* where the name-field just contains the empty
       string. AST-transformation of Anonymous parsers can be hooked
       only to their class name, and not to the individual parser.

    Parser objects are callable and parsing is done by calling a parser
    object with the text to parse.

    If the parser matches it returns a tuple consisting of a node
    representing the root of the concrete syntax tree resulting from the
    match as well as the substring `text[i:]` where i is the length of
    matched text (which can be zero in the case of parsers like
    `ZeroOrMore` or `Option`). If `i > 0` then the parser has "moved
    forward".

    If the parser does not match it returns `(None, text). **Note** that
    this is not the same as an empty match `("", text)`. Any empty match
    can for example be returned by the `ZeroOrMore`-parser in case the
    contained parser is repeated zero times.

    Attributes and Properties:
195
196
197
198
199
        pname:    The parser name or the empty string in case the parser
                remains anonymous.
        tag_name:  The tag_name for the nodes that are created by
                the parser. If the parser is named, this is the same as
                `pname`, otherwise it is the name of the parser's type.
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
        visited:  Mapping of places this parser has already been to
                during the current parsing process onto the results the
                parser returned at the respective place. This dictionary
                is used to implement memoizing.

        recursion_counter:  Mapping of places to how often the parser
                has already been called recursively at this place. This
                is needed to implement left recursion. The number of
                calls becomes irrelevant once a resault has been memoized.

        cycle_detection:  The apply()-method uses this variable to make
                sure that one and the same function will not be applied
                (recursively) a second time, if it has already been
                applied to this parser.

215
        _grammar:  A reference to the Grammar object to which the parser
216
217
218
                is attached.
    """

219
    def __init__(self) -> None:
220
        # assert isinstance(name, str), str(name)
221
222
223
        self.pname = ''               # type: str
        self.tag_name = self.ptype    # type: str
        self.cycle_detection = set()  # type: Set[ApplyFunc]
224
225
226
227
        try:
            self._grammar = GRAMMAR_PLACEHOLDER  # type: Grammar
        except NameError:
            pass
228
229
230
        self.reset()

    def __deepcopy__(self, memo):
231
        """        Deepcopy method of the parser. Upon instantiation of a Grammar-
232
233
234
235
236
        object, parsers will be deep-copied to the Grammar object. If a
        derived parser-class changes the signature of the constructor,
        `__deepcopy__`-method must be replaced (i.e. overridden without
        calling the same method from the superclass) by the derived class.
        """
237
        duplicate = self.__class__()
di68kap's avatar
di68kap committed
238
        duplicate.pname = self.pname
239
        duplicate.tag_name = self.tag_name
240
        return duplicate
241

242
    def __repr__(self):
di68kap's avatar
di68kap committed
243
        return self.pname + self.ptype
244
245

    def __str__(self):
di68kap's avatar
di68kap committed
246
        return self.pname + (' = ' if self.pname else '') + repr(self)
247

248
249
250
251
252
253
    @property
    def ptype(self) -> str:
        """Returns a type name for the parser. By default this is the name of
        the parser class with an added leading colon ':'. """
        return ':' + self.__class__.__name__

254
255
256
    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
di68kap's avatar
di68kap committed
257
        return self.pname if self.pname else self.__repr__()
258

259
260
261
262
    def reset(self):
        """Initializes or resets any parser variables. If overwritten,
        the `reset()`-method of the parent class must be called from the
        `reset()`-method of the derived class."""
263
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
264
        self.recursion_counter = defaultdict(int)  # type: DefaultDict[int, int]
265

eckhart's avatar
eckhart committed
266
    @cython.locals(location=cython.int, gap=cython.int, i=cython.int)
267
268
269
    def __call__(self: 'Parser', text: StringView) -> Tuple[Optional[Node], StringView]:
        """Applies the parser to the given text. This is a wrapper method that adds
        the business intelligence that is common to all parsers. The actual parsing is
270
271
        done in the overridden method `_parse()`. This wrapper-method can be thought of
        as a "parser guard", because it guards the parsing process.
272
        """
273
274
275
276
277
278
279
280
281
        def get_error_node_id(error_node: Node, root_node: RootNode) -> int:
            if error_node:
                error_node_id = id(error_node)
                while error_node_id not in grammar.tree__.error_nodes and error_node.children:
                    error_node = error_node.result[-1]
                    error_node_id = id(error_node)
            else:
                error_node_id = 0

282
        grammar = self._grammar
283
284
285
286
287
288
289
290
        location = grammar.document_length__ - len(text)

        try:
            if grammar.last_rb__loc__ >= location:
                grammar.rollback_to__(location)

            # if location has already been visited by the current parser,
            # return saved result
291
292
            visited = self.visited  # using local variable for better performance
            if location in visited:
293
                # no history recording in case of memoized results
294
                return visited[location]
295
296

            # break left recursion at the maximum allowed depth
297
298
299
            left_recursion_depth__ = grammar.left_recursion_depth__
            if left_recursion_depth__:
                if self.recursion_counter[location] > left_recursion_depth__:
300
301
302
303
                    grammar.recursion_locations__.add(location)
                    return None, text
                self.recursion_counter[location] += 1

304
305
            history_tracking__ = grammar.history_tracking__
            if history_tracking__:
di68kap's avatar
di68kap committed
306
                grammar.call_stack__.append(
307
308
                    ((self.repr if self.tag_name in (':RegExp', ':Token', ':DropToken')
                      else self.tag_name), location))
309
                grammar.moving_forward__ = True
310
                error = None
311
312
313
314

            try:
                # PARSER CALL: run _parse() method
                node, rest = self._parse(text)
315
            except ParserError as pe:
316
                # does this play well with variable setting? add rollback clause here? tests needed...
317
                gap = len(text) - len(pe.rest)
di68kap's avatar
di68kap committed
318
                rules = grammar.resume_rules__.get(self.pname, [])
319
                rest = pe.rest[len(pe.node):]
320
321
322
323
                i = reentry_point(rest, rules)
                if i >= 0 or self == grammar.start_parser__:
                    # apply reentry-rule or catch error at root-parser
                    if i < 0:
324
                        i = 1
325
                    nd = Node(ZOMBIE_TAG, rest[:i]).with_pos(location)
326
                    nd.attr['err'] = pe.error.message
327
                    rest = rest[i:]
328
329
330
                    assert pe.node.children or (not pe.node.result)
                    if pe.first_throw:
                        node = pe.node
eckhart's avatar
eckhart committed
331
                        node.result = node.children + (nd,)
332
                    else:
333
                        node = Node(self.tag_name,
334
                                    (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node, nd))
335
336
                elif pe.first_throw:
                    raise ParserError(pe.node, pe.rest, pe.error, first_throw=False)
337
338
                elif grammar.tree__.errors[-1].code == Error.MANDATORY_CONTINUATION_AT_EOF:  # EXPERIMENTAL!!
                    node = pe.node
339
                else:
340
341
                    result = (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) if gap \
                        else pe.node  # type: ResultType
342
343
                    raise ParserError(Node(self.tag_name, result).with_pos(location),
                                      text, pe.error, first_throw=False)
344
345
                error = pe.error  # needed for history tracking

346

347
            if left_recursion_depth__:
348
349
350
351
352
353
                self.recursion_counter[location] -= 1
                # don't clear recursion_locations__ !!!

            if node is None:
                # retrieve an earlier match result (from left recursion) if it exists
                if location in grammar.recursion_locations__:
354
355
                    if location in visited:
                        node, rest = visited[location]
356
                        if location != grammar.last_recursion_location__:
357
358
359
                            grammar.tree__.add_error(
                                node, Error("Left recursion encountered. "
                                            "Refactor grammar to avoid slow parsing.",
360
361
                                            node.pos if node else location,
                                            Error.LEFT_RECURSION_WARNING))
362
                            error_id = id(node)
363
                            grammar.last_recursion_location__ = location
364
365
366
367
                    # don't overwrite any positive match (i.e. node not None) in the cache
                    # and don't add empty entries for parsers returning from left recursive calls!
                elif grammar.memoization__:
                    # otherwise also cache None-results
368
                    visited[location] = (None, rest)
369
            else:
370
                # assert node._pos < 0 or node == EMPTY_NODE
371
                # if node._pos != EMPTY_NODE:
372
                node._pos = location
373
374
                # assert node._pos >= 0 or node == EMPTY_NODE, \
                #     str("%i < %i" % (grammar.document_length__, location))
375
376
377
378
379
380
381
                if (grammar.last_rb__loc__ < location
                        and (grammar.memoization__ or location in grammar.recursion_locations__)):
                    # - variable manipulating parsers will not be entered into the cache,
                    #   because caching would interfere with changes of variable state
                    # - in case of left recursion, the first recursive step that
                    #   matches will store its result in the cache
                    # TODO: need a unit-test concerning interference of variable manipulation and left recursion algorithm?
382
                    visited[location] = (node, rest)
383
384
385

            # Mind that memoized parser calls will not appear in the history record!
            # Does this make sense? Or should it be changed?
386
            if history_tracking__:
387
388
                # don't track returning parsers except in case an error has occurred
                # remaining = len(rest)
389
                if grammar.moving_forward__:
390
                    record = HistoryRecord(grammar.call_stack__, node, text,
di68kap's avatar
di68kap committed
391
                                           grammar.line_col__(text))
392
                    grammar.history__.append(record)
393
394
395
396
                elif error:
                    # error_nid = id(node)  # type: int
                    # if error_nid in grammar.tree__.error_nodes:
                    record = HistoryRecord(grammar.call_stack__, node, text,
di68kap's avatar
di68kap committed
397
                                           grammar.line_col__(text),
398
399
                                           [error])
                    grammar.history__.append(record)
400
401
402
403
                grammar.moving_forward__ = False
                grammar.call_stack__.pop()

        except RecursionError:
404
            node = Node(ZOMBIE_TAG, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
405
406
407
408
409
410
            node._pos = location
            grammar.tree__.new_error(node, "maximum recursion depth of parser reached; "
                                           "potentially due to too many errors!")
            rest = EMPTY_STRING_VIEW

        return node, rest
411
412
413
414
415
416
417
418
419
420
421
422

    def __add__(self, other: 'Parser') -> 'Series':
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
        return Series(self, other)

    def __or__(self, other: 'Parser') -> 'Alternative':
        """The | operator generates an alternative parser that applies
        the first parser and, if that does not match, the second parser.
        """
        return Alternative(self, other)

eckhart's avatar
eckhart committed
423

424
425
426
427
428
429
    def _parse(self, text: StringView) -> Tuple[Optional[Node], StringView]:
        """Applies the parser to the given `text` and returns a node with
        the results or None as well as the text at the position right behind
        the matching string."""
        raise NotImplementedError

430
    @property
eckhart's avatar
eckhart committed
431
    def grammar(self) -> 'Grammar':
432
433
434
435
436
        try:
            if self._grammar != GRAMMAR_PLACEHOLDER:
                return self._grammar
            else:
                raise AssertionError('Grammar has not yet been set!')
437
438
        except (AttributeError, NameError):
            raise AttributeError('Parser placeholder does not have a grammar!')
439
440
441

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
442
443
444
445
446
447
448
449
450
        try:
            if self._grammar == GRAMMAR_PLACEHOLDER:
                self._grammar = grammar
                # self._grammar_assigned_notifier()
            elif self._grammar != grammar:
                  raise AssertionError("Parser has already been assigned"
                                       "to a different Grammar object!")
        except AttributeError:
            pass  # ignore setting of grammar attribute for placeholder parser
451
452
        except NameError:  # Cython: No access to GRAMMA_PLACEHOLDER, yet :-(
            self._grammar = grammar
453

eckhart's avatar
eckhart committed
454
    def _apply(self, func: ApplyFunc, flip: FlagFunc) -> bool:
455
456
        """
        Applies function `func(parser)` recursively to this parser and all
eckhart's avatar
eckhart committed
457
458
459
460
461
462
463
464
465
        descendant parsers, if any exist.

        In order to break cycles, function `flip` is called, which should
        return `True`, if this parser has already been visited. If not, it
        flips the cycle detection flag and returns `False`.

        This is a protected function and should not called from outside
        class Parser or any of its descendants. The entry point for external
        calls is the method `apply()` without underscore!
466
        """
eckhart's avatar
eckhart committed
467
        if flip(func, self.cycle_detection):
468
469
470
471
472
            return False
        else:
            func(self)
            return True

eckhart's avatar
eckhart committed
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
    def apply(self, func: ApplyFunc):
        """
        Applies function `func(parser)` recursively to this parser and all
        descendant parsers, if any exist. Traversal is pre-order.
        """
        def positive_flip(f: ApplyFunc, flagged: Set[Callable]) -> bool:
            """Returns True, if function `f` has already been applied to this
            parser and sets the flag accordingly. Interprets `f in flagged == True`
            as meaning that `f` has already been applied."""
            if f in flagged:
                return True
            else:
                flagged.add(f)
                return False

        def negative_flip(f: ApplyFunc, flagged: Set[Callable]) -> bool:
            """Returns True, if function `f` has already been applied to this
            parser and sets the flag accordingly. Interprets `f in flagged == False`
            as meaning that `f` has already been applied."""
            if f not in flagged:
                return True
            else:
                flagged.remove(f)
                return False

        if func in self.cycle_detection:
            self._apply(func, negative_flip)
        else:
            self._apply(func, positive_flip)

503

504
PARSER_PLACEHOLDER = Parser()
505
506
507
508
509
510
511
512
513


########################################################################
#
# Grammar class, central administration of all parser of a grammar
#
########################################################################


514
515
516
517
518
519
520
521
522
523
def mixin_comment(whitespace: str, comment: str) -> str:
    """
    Returns a regular expression that merges comment and whitespace
    regexps. Thus comments cann occur whereever whitespace is allowed
    and will be skipped just as implicit whitespace.

    Note, that because this works on the level of regular expressions,
    nesting comments is not possible. It also makes it much harder to
    use directives inside comments (which isn't recommended, anyway).
    """
524
525
526
    if comment:
        return '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return whitespace
527
528


529
class UnknownParserError(KeyError):
eckhart's avatar
eckhart committed
530
    """UnknownParserError is raised if a Grammar object is called with a
531
    parser that does not exist or if in the course of parsing a parser
eckhart's avatar
eckhart committed
532
    is referred to that does not exist."""
533
534


535
GrammarErrorType = Tuple[str, Parser, Error]      # TODO: replace with a named tuple?
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552


class GrammarError(Exception):
    """GrammarError will be raised if static analysis reveals errors
    in the grammar.
    """
    def __init__(self, static_analysis_result: List[GrammarErrorType]):
        assert static_analysis_result  # must not be empty
        self.errors = static_analysis_result

    def __str__(self):
        if len(self.errors) == 1:
            return str(self.errors[0][2])
        return '\n' + '\n'.join(("%i. " % (i + 1) + str(err_tuple[2]))
                                for i, err_tuple in enumerate(self.errors))


553
554
555
556
557
558
559
560
561
562
563
564
565
class Grammar:
    r"""
    Class Grammar directs the parsing process and stores global state
    information of the parsers, i.e. state information that is shared
    accross parsers.

    Grammars are basically collections of parser objects, which are
    connected to an instance object of class Grammar. There exist two
    ways of connecting parsers to grammar objects: Either by passing
    the root parser object to the constructor of a Grammar object
    ("direct instantiation"), or by assigning the root parser to the
    class variable "root__" of a descendant class of class Grammar.

eckhart's avatar
eckhart committed
566
    Example for direct instantiation of a grammar::
567

568
        >>> number = RE(r'\d+') + RE(r'\.') + RE(r'\d+') | RE(r'\d+')
569
570
571
572
573
574
575
576
577
578
579
580
581
582
        >>> number_parser = Grammar(number)
        >>> number_parser("3.1416").content
        '3.1416'

    Collecting the parsers that define a grammar in a descendant class of
    class Grammar and assigning the named parsers to class variables
    rather than global variables has several advantages:

    1. It keeps the namespace clean.

    2. The parser names of named parsers do not need to be passed to the
       constructor of the Parser object explicitly, but it suffices to
       assign them to class variables, which results in better
       readability of the Python code.
583
       See classmethod `Grammar._assign_parser_names__()`
584
585

    3. The parsers in the class do not necessarily need to be connected
586
587
       to one single root parser, which is helpful for testing and when
       building up a parser gradually from several components.
588
589
590
591
592
593
594

    As a consequence, though, it is highly recommended that a Grammar
    class should not define any other variables or methods with names
    that are legal parser names. A name ending with a double
    underscore '__' is *not* a legal parser name and can safely be
    used.

595
    Example::
596
597
598
599
600
601
602
603
604

        class Arithmetic(Grammar):
            # special fields for implicit whitespace and comment configuration
            COMMENT__ = r'#.*(?:\n|$)'  # Python style comments
            wspR__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)

            # parsers
            expression = Forward()
            INTEGER = RE('\\d+')
605
606
607
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
608
609
610
611
612
            root__ = expression

    Upon instantiation the parser objects are deep-copied to the
    Grammar object and assigned to object variables of the same name.
    Any parser that is directly assigned to a class variable is a
di68kap's avatar
di68kap committed
613
    'named' parser and its field `parser.pname` contains the variable
614
615
    name after instantiation of the Grammar class. All other parsers,
    i.e. parsers that are defined within a `named` parser, remain
616
    "anonymous parsers" where `parser.pname` is the empty string.
617
    If one and the same parser is assigned to several class variables
618
    such as, for example, the parser `expression` in the example above,
619
620
621
622
623
624
625
626
627
628
629
630
631
    the first name sticks.

    Grammar objects are callable. Calling a grammar object with a UTF-8
    encoded document, initiates the parsing of the document with the
    root parser. The return value is the concrete syntax tree. Grammar
    objects can be reused (i.e. called again) after parsing. Thus, it
    is not necessary to instantiate more than one Grammar object per
    thread.

    Grammar classes contain a few special class fields for implicit
    whitespace and comments that should be overwritten, if the defaults
    (no comments, horizontal right aligned whitespace) don't fit:

632
    Class Attributes:
633
634
635
636
        root__:  The root parser of the grammar. Theoretically, all parsers of the
                 grammar should be reachable by the root parser. However, for testing
                 of yet incomplete grammars class Grammar does not assume that this
                 is the case.
637

Eckhart Arnold's avatar
Eckhart Arnold committed
638
        resume_rules__: A mapping of parser names to a list of regular expressions or search
639
                strings that act as rules to find the the reentry point if a ParserError
Eckhart Arnold's avatar
Eckhart Arnold committed
640
641
                was thrown during the execution of the parser with the respective name.

642
643
644
645
646
        parser_initializiation__:  Before the parser class (!) has been initialized,
                 which happens upon the first time it is instantiated (see
                 :func:_assign_parser_names()` for an explanation), this class
                 field contains a value other than "done". A value of "done" indicates
                 that the class has already been initialized.
647

648
649
650
651
652
653
654
        static_analysis_pending__: True as long as no static analysis (see the method
                with the same name for more information) has been done to check
                parser tree for correctness. Static analysis
                is done at instantiation and the flag is then set to false, but it
                can also be carried out once the class has been generated
                (by DHParser.ebnf.EBNFCompiler) and then be set to false in the
                definition of the grammar class already.
655

eckhart's avatar
eckhart committed
656
657
658
659
        python__src__:  For the purpose of debugging and inspection, this field can
                 take the python src of the concrete grammar class
                 (see `dsl.grammar_provider`).

660
    Instance Attributes:
661
662
        all_parsers__:  A set of all parsers connected to this grammar object

eckhart's avatar
eckhart committed
663
664
665
666
        start_parser__:  During parsing, the parser with which the parsing process
                was started (see method `__call__`) or `None` if no parsing process
                is running.

667
668
669
670
671
672
673
674
675
676
        history_tracking__:  A flag indicating that the parsing history shall
                be tracked

        _dirty_flag__:  A flag indicating that the Grammar has been called at
                least once so that the parsing-variables need to be reset
                when it is called again.

        document__:  the text that has most recently been parsed or that is
                currently being parsed.

eckhart's avatar
eckhart committed
677
678
        document_length__:  the length of the document.

679
680
681
        document_lbreaks__:  list of linebreaks within the document, starting
                with -1 and ending with EOF. This helps generating line
                and column number for history recording and will only be
682
                initialized if :attr:`history_tracking__` is true.
683

684
685
686
687
688
        tree__: The root-node of the parsing tree. This variable is available
               for error-reporting already during parsing  via
               ``self.grammar.tree__.add_error``, but it references the full
               parsing tree only after parsing has been finished.

689
        _reversed__:  the same text in reverse order - needed by the `Lookbehind`-
690
691
692
                parsers.

        variables__:  A mapping for variable names to a stack of their respective
693
694
                string values - needed by the :class:`Capture`-, :class:`Retrieve`-
                and :class:`Pop`-parsers.
695
696

        rollback__:  A list of tuples (location, rollback-function) that are
697
698
699
700
                deposited by the :class:`Capture`- and :class:`Pop`-parsers.
                If the parsing process reaches a dead end then all
                rollback-functions up to the point to which it retreats will be
                called and the state of the variable stack restored accordingly.
701
702
703

        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
704
705
                backtracks to a location at or before last_rb__loc__ (i.e.
                location <= last_rb__loc__) then a rollback of all variable
706
707
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
708
                calling method :func:`rollback_to__(location)`.
709

di68kap's avatar
di68kap committed
710
711
712
        call_stack__:  A stack of the tag names and locations of all parsers
                in the call chain to the currently processed parser during
                parsing. The call stack can be thought of as a breadcrumb trail.
713
                This is required for recording the parser history (for debugging)
714
715
716
717
718
719
720
721
722
723
                and, eventually, i.e. one day in the future, for tracing through
                the parsing process.

        history__:  A list of parser-call-stacks. A parser-call-stack is
                appended to the list each time a parser either matches, fails
                or if a parser-error occurs.

        moving_forward__: This flag indicates that the parsing process is currently
                moving forward . It is needed to reduce noise in history recording
                and should not be considered as having a valid value if history
724
                recording is turned off! (See :func:`Parser.__call__`)
725
726
727
728

        recursion_locations__:  Stores the locations where left recursion was
                detected. Needed to provide minimal memoization for the left
                recursion detection algorithm, but, strictly speaking, superfluous
729
730
731
732
733
                if full memoization is enabled. (See :func:`Parser.__call__()`)

        last_recursion_location__:  Last location where left recursion was
                detected. This is used to avoid reduplicating warning messages
                about left recursion.
734
735
736
737
738
739

        memoization__:  Turns full memoization on or off. Turning memoization off
                results in less memory usage and sometimes reduced parsing time.
                In some situations it may drastically increase parsing time, so
                it is safer to leave it on. (Default: on)

740
741
742
743
744
        flatten_tree__:  If True (default), anonymous nodes will be flattened
                during parsing already. This greatly reduces the concrete syntax
                tree and simplifies and speeds up abstract syntax tree generation.
                The initial value will be read from the config variable
                'flatten_tree_while_parsing' upon class instantiation.
745
746
747
748
749
750
751
752

        left_recursion_depth__: the maximum allowed depth for left-recursion.
                A depth of zero means that no left recursion handling will
                take place. See 'left_recursion_depth' in config.py.

        max_parser_dropouts__: Maximum allowed number of retries after errors
                where the parser would exit before the complete document has
                been parsed. See config.py
753
    """
eckhart's avatar
eckhart committed
754
    python_src__ = ''  # type: str
755
    root__ = PARSER_PLACEHOLDER  # type: Parser
756
    # root__ must be overwritten with the root-parser by grammar subclass
eckhart's avatar
eckhart committed
757
    parser_initialization__ = ["pending"]  # type: List[str]
eckhart's avatar
eckhart committed
758
    resume_rules__ = dict()  # type: Dict[str, ResumeList]
759
    # some default values
760
761
    # COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    # WSP_RE__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
762
    static_analysis_pending__ = [True]  # type: List[bool]
763
764
765
766
767


    @classmethod
    def _assign_parser_names__(cls):
        """
di68kap's avatar
di68kap committed
768
        Initializes the `parser.pname` fields of those
769
        Parser objects that are directly assigned to a class field with
770
        the field's name, e.g.::
771

772
773
            class Grammar(Grammar):
                ...
774
                symbol = RE(r'(?!\\d)\\w+')
775

di68kap's avatar
di68kap committed
776
        After the call of this method symbol.pname == "symbol" holds.
di68kap's avatar
di68kap committed
777
778
        Parser names starting or ending with a double underscore like
        ``root__`` will be ignored. See :func:`sane_parser_name()`
779
780
781
782
783
784
785
786
787
788

        This is done only once, upon the first instantiation of the
        grammar class!

        Attention: If there exists more than one reference to the same
        parser, only the first one will be chosen for python versions
        greater or equal 3.6.  For python version <= 3.5 an arbitrarily
        selected reference will be chosen. See PEP 520
        (www.python.org/dev/peps/pep-0520/) for an explanation of why.
        """
eckhart's avatar
eckhart committed
789
        if cls.parser_initialization__[0] != "done":
790
791
792
            cdict = cls.__dict__
            for entry, parser in cdict.items():
                if isinstance(parser, Parser) and sane_parser_name(entry):
di68kap's avatar
di68kap committed
793
                    if isinstance(parser, Forward):
di68kap's avatar
di68kap committed
794
795
796
797
                        if not cast(Forward, parser).parser.pname:
                            cast(Forward, parser).parser.pname = entry
                    else:   # if not parser.pname:
                        parser.pname = entry
eckhart's avatar
eckhart committed
798
            cls.parser_initialization__[0] = "done"
799
800
801


    def __init__(self, root: Parser = None) -> None:
802
803
        self.all_parsers__ = set()             # type: Set[Parser]
        self.start_parser__ = None             # type: Optional[Parser]
804
805
806
        self._dirty_flag__ = False             # type: bool
        self.history_tracking__ = False        # type: bool
        self.memoization__ = True              # type: bool
807
808
809
        self.flatten_tree__ = get_config_value('flatten_tree_while_parsing')    # type: bool
        self.left_recursion_depth__ = get_config_value('left_recursion_depth')  # type: int
        self.max_parser_dropouts__ = get_config_value('max_parser_dropouts')    # type: int
810
811
812
813
814
815
816
817
818
        self._reset__()

        # prepare parsers in the class, first
        self._assign_parser_names__()

        # then deep-copy the parser tree from class to instance;
        # parsers not connected to the root object will be copied later
        # on demand (see Grammar.__getitem__()). Usually, the need to
        # do so only arises during testing.
eckhart's avatar
eckhart committed
819
820
821
822
        self.root_parser__ = copy.deepcopy(root) if root else copy.deepcopy(self.__class__.root__)
        self.root_parser__.apply(self._add_parser__)
        assert 'root_parser__' in self.__dict__
        assert self.root_parser__ == self.__dict__['root_parser__']
823

824
825
826
827
828
829
830
831
832
        if self.__class__.static_analysis_pending__ \
                and get_config_value('static_analysis') in {'early', 'late'}:
            try:
                result = self.static_analysis()
                if result:
                    raise GrammarError(result)
                self.__class__.static_analysis_pending__.pop()
            except (NameError, AttributeError):
                pass  # don't fail the initialization of PLACEHOLDER
833

834
835
836
837
838

    def __getitem__(self, key):
        try:
            return self.__dict__[key]
        except KeyError:
eckhart's avatar
eckhart committed
839
            parser_template = getattr(self.__class__, key, None)
840
841
842
843
            if parser_template:
                # add parser to grammar object on the fly...
                parser = copy.deepcopy(parser_template)
                parser.apply(self._add_parser__)
eckhart's avatar
eckhart committed
844
                assert self[key] == parser
845
                return self[key]
846
            raise UnknownParserError('Unknown parser "%s" !' % key)
847

848

849
850
    def __contains__(self, key):
        return key in self.__dict__ or hasattr(self, key)
851

852

853
    def _reset__(self):
854
        self.tree__ = RootNode()              # type: RootNode
855
856
        self.document__ = EMPTY_STRING_VIEW   # type: StringView
        self._reversed__ = EMPTY_STRING_VIEW  # type: StringView
eckhart's avatar
eckhart committed
857
        self.document_length__ = 0            # type: int
858
859
        self.document_lbreaks__ = []          # type: List[int]
        # variables stored and recalled by Capture and Retrieve parsers
eckhart's avatar
eckhart committed
860
        self.variables__ = defaultdict(lambda: [])  # type: DefaultDict[str, List[str]]
861
862
863
        self.rollback__ = []                  # type: List[Tuple[int, Callable]]
        self.last_rb__loc__ = -1              # type: int
        # support for call stack tracing
864
        self.call_stack__ = []                # type: List[Tuple[str, int]]  # tag_name, location
865
866
867
868
869
        # snapshots of call stacks
        self.history__ = []                   # type: List[HistoryRecord]
        # also needed for call stack tracing
        self.moving_forward__ = False         # type: bool
        self.recursion_locations__ = set()    # type: Set[int]
870
        self.last_recursion_location__ = -1   # type: int
871
872
873
874
875
876
877
878
879
880


    @property
    def reversed__(self) -> StringView:
        """
        Returns a reversed version of the currently parsed document. As
        about the only case where this is needed is the Lookbehind-parser,
        this is done lazily.
        """
        if not self._reversed__:
881
            self._reversed__ = StringView(self.document__.get_text()[::-1])
882
883
884
885
886
887
888
889
        return self._reversed__


    def _add_parser__(self, parser: Parser) -> None:
        """
        Adds the particular copy of the parser object to this
        particular instance of Grammar.
        """
di68kap's avatar
di68kap committed
890
        if parser.pname:
891
            # prevent overwriting instance variables or parsers of a different class
di68kap's avatar
di68kap committed
892
893
            assert parser.pname not in self.__dict__ or \
                   isinstance(self.__dict__[parser.pname], parser.__class__), \
894
                ('Cannot add parser "%s" because a field with the same name '
di68kap's avatar
di68kap committed
895
                 'already exists in grammar object: %s!'
di68kap's avatar
di68kap committed
896
897
898
                 % (parser.pname, str(self.__dict__[parser.pname])))
            setattr(self, parser.pname, parser)
        parser.tag_name = parser.pname or parser.ptype
899
900
901
902
        self.all_parsers__.add(parser)
        parser.grammar = self


eckhart's avatar
eckhart committed
903
904
    def __call__(self,
                 document: str,
eckhart's avatar
eckhart committed
905
                 start_parser: Union[str, Parser] = "root_parser__",
eckhart's avatar
eckhart committed
906
                 track_history: bool = False) -> RootNode:
907
908
909
910
911
        """
        Parses a document with with parser-combinators.

        Args:
            document (str): The source text to be parsed.
di68kap's avatar
di68kap committed
912
913
            start_parser (str or Parser): The name of the parser with which
                to start. This is useful for testing particular parsers
914
                (i.e. particular parts of the EBNF-Grammar.)
di68kap's avatar
di68kap committed
915
916
917
918
919
            track_history (bool): If true, the parsing history will be
                recorded in self.history__. If logging is turned on (i.e.
                DHParser.log.is_logging() returns true), the parsing history
                will always be recorded, even if `False` is passed to
                the `track_history` parameter.
920
        Returns:
di68kap's avatar
di68kap committed
921
            Node: The root node to the parse tree.
922
        """
923

eckhart's avatar
eckhart committed
924
        def tail_pos(predecessors: Union[List[Node], Tuple[Node, ...], None]) -> int:
eckhart's avatar
eckhart committed
925
926
            """Adds the position after the last node in the list of
            predecessors to the node."""
eckhart's avatar
eckhart committed
927
            return predecessors[-1].pos + len(predecessors[-1]) if predecessors else 0
eckhart's avatar
eckhart committed
928

929
930
931
932
933
934
        def lookahead_failure_only(parser):
            """EXPERIMENTAL!

            Checks if failure to match document was only due to a succeeding
            lookahead parser, which is a common design pattern that can break test
            cases. (Testing for this case allows to modify the error message, so
935
            that the testing framework knows that the failure is only a
936
937
938
            test-case-artifact and no real failure.
            (See test/test_testing.TestLookahead !)
            """
939
940
941
            def is_lookahead(tag_name: str) -> bool:
                return (tag_name in self and isinstance(self[tag_name], Lookahead)
                        or tag_name[0] == ':' and issubclass(eval(tag_name[1:]), Lookahead))
di68kap's avatar
di68kap committed
942
943
944
945
            # for h in reversed(self.history__[:-1]):
            #     for tn, pos in h.call_stack:
            #         if is_lookahead(tn) and h.status == HistoryRecord.MATCH:
            #             print(h.call_stack, pos, h.line_col)
946
947
            last_record = self.history__[-2] if len(self.history__) > 1 else None  # type: Optional[HistoryRecord]
            return last_record and parser != self.root_parser__ \
di68kap's avatar
di68kap committed
948
                    and any(h.status == HistoryRecord.MATCH
di68kap's avatar
di68kap committed
949
950
                            and any(is_lookahead(tn) and location >= len(self.document__)
                                    for tn, location in h.call_stack)
di68kap's avatar
di68kap committed
951
                            for h in self.history__[:-1])
952

953
954
955
956
957
958
959
        # assert isinstance(document, str), type(document)
        if self._dirty_flag__:
            self._reset__()
            for parser in self.all_parsers__:
                parser.reset()
        else:
            self._dirty_flag__ = True
di68kap's avatar
di68kap committed
960
961
962
963
        self.history_tracking__ = track_history or is_logging()
        # safe tracking state, because history_tracking__ might be set to false, later,
        # but original tracking state is needed for additional error information.
        track_history = self.history_tracking__
964
        self.document__ = StringView(document)
eckhart's avatar
eckhart committed
965
        self.document_length__ = len(self.document__)
966
        self.document_lbreaks__ = linebreaks(document) if self.history_tracking__ else []
Eckhart Arnold's avatar
Eckhart Arnold committed
967
        self.last_rb__loc__ = -1  # rollback location
968
        parser = self[start_parser] if isinstance(start_parser, str) else start_parser
969
        self.start_parser__ = parser.parser if isinstance(parser, Forward) else parser
970
971
972
973
974
975
976
977
        assert parser.grammar == self, "Cannot run parsers from a different grammar object!" \
                                       " %s vs. %s" % (str(self), str(parser.grammar))
        result = None  # type: Optional[Node]
        stitches = []  # type: List[Node]
        rest = self.document__
        if not rest:
            result, _ = parser(rest)
            if result is None:
Eckhart Arnold's avatar
Eckhart Arnold committed
978
                result = Node(ZOMBIE_TAG, '').with_pos(0)
979
980
                if lookahead_failure_only(parser):
                    self.tree__.new_error(
981
982
                        result, 'Parser "%s" only did not match empty document '
                                'because of lookahead' % str(parser),
983
                        Error.PARSER_LOOKAHEAD_FAILURE_ONLY)
984
985
986
987
988
                else:
                    self.tree__.new_error(
                        result, 'Parser "%s" did not match empty document.' % str(parser),
                        Error.PARSER_DID_NOT_MATCH)

989
990
991
        # copy to local variable, so break condition can be triggered manually
        max_parser_dropouts = self.max_parser_dropouts__
        while rest and len(stitches) < max_parser_dropouts:
992
993
994
995
996
            result, rest = parser(rest)
            if rest:
                fwd = rest.find("\n") + 1 or len(rest)
                skip, rest = rest[:fwd], rest[fwd:]
                if result is None:
di68kap's avatar
di68kap committed
997
998
999
1000
                    err_info = '' if not track_history else \
                               '\n    Most advanced: %s\n    Last match:    %s;' % \
                               (str(HistoryRecord.most_advanced_match(self.history__)),
                                str(HistoryRecord.last_match(self.history__)))