parse.py 87.1 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

Eckhart Arnold's avatar
Eckhart Arnold committed
33
from collections import defaultdict, OrderedDict
34
import copy
35

36
from DHParser.error import Error, linebreaks, line_col
37
38
from DHParser.log import is_logging, HistoryRecord
from DHParser.preprocess import BEGIN_TOKEN, END_TOKEN, RX_TOKEN_NAME
39
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
di68kap's avatar
di68kap committed
40
41
from DHParser.syntaxtree import Node, RootNode, WHITESPACE_PTYPE, \
    TOKEN_PTYPE, ZOMBIE, ResultType
42
from DHParser.toolkit import sane_parser_name, escape_control_characters, re, typing
Eckhart Arnold's avatar
Eckhart Arnold committed
43
from typing import Callable, cast, List, Tuple, Set, Dict, DefaultDict, Union, Optional, Any
44

45

46
__all__ = ('Parser',
47
           'UnknownParserError',
48
49
           'Grammar',
           'PreprocessorToken',
50
           'Token',
51
           'RegExp',
52
53
           'RE',
           'TKN',
54
           'Whitespace',
55
           'mixin_comment',
eckhart's avatar
eckhart committed
56
57
           'UnaryOperator',
           'NaryOperator',
58
59
60
61
62
63
64
65
66
           'Synonym',
           'Option',
           'ZeroOrMore',
           'OneOrMore',
           'Series',
           'Alternative',
           'AllOf',
           'SomeOf',
           'Unordered',
eckhart's avatar
eckhart committed
67
           'Required',
68
69
70
71
72
73
74
75
76
77
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
           'last_value',
           'counterpart',
           'accumulate',
           'Capture',
           'Retrieve',
           'Pop',
78
           'Forward')
79
80
81
82


########################################################################
#
83
# Parser base class
84
85
86
87
88
89
90
91
92
93
94
#
########################################################################


LEFT_RECURSION_DEPTH = 8  # type: int
# because of python's recursion depth limit, this value ought not to be
# set too high. PyPy allows higher values than CPython
MAX_DROPOUTS = 3  # type: int
# stop trying to recover parsing after so many errors


Eckhart Arnold's avatar
Eckhart Arnold committed
95
96
97
class ParserError(Exception):
    """
    A `ParserError` is thrown for those parser errors that allow the
98
    controlled re-entrance of the parsing process after the error occurred.
Eckhart Arnold's avatar
Eckhart Arnold committed
99
    If a reentry-rule has been configured for the parser where the error
100
    occurred, the parser guard can resume the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
101
102
103
104
105

    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.
    """
106
    def __init__(self, node: Node, rest: StringView, first_throw: bool):
Eckhart Arnold's avatar
Eckhart Arnold committed
107
108
        self.node = node  # type: Node
        self.rest = rest  # type: StringView
109
        self.first_throw = first_throw  # type: bool
Eckhart Arnold's avatar
Eckhart Arnold committed
110
111


112
ResumeList = List[Union[str, Any]]  # list of strings or regular expressiones
Eckhart Arnold's avatar
Eckhart Arnold committed
113
114


115
def reentry_point(rest: StringView, rules: ResumeList) -> int:
Eckhart Arnold's avatar
Eckhart Arnold committed
116
117
118
119
120
121
122
123
124
    """
    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:
125
126
        The integer index of the closest reentry point or -1 if no reentry-point
        was found.
Eckhart Arnold's avatar
Eckhart Arnold committed
127
    """
128
    upper_limit = len(rest) + 1
Eckhart Arnold's avatar
Eckhart Arnold committed
129
130
131
132
    i = upper_limit
    #find closest match
    for rule in rules:
        if isinstance(rule, str):
133
134
            k = rest.find(rule)
            i = min(k if k >= 0 else upper_limit, i)
Eckhart Arnold's avatar
Eckhart Arnold committed
135
136
137
        else:
            m = rest.search(rule)
            if m:
138
                i = min(rest.index(m.start()), i)
139
    # in case no rule matched return -1
140
    if i == upper_limit:
141
142
143
        i = -1
    return i
    # return Node(None, rest[:i]), rest[i:]
Eckhart Arnold's avatar
Eckhart Arnold committed
144
145


eckhart's avatar
eckhart committed
146
147
148
149
ApplyFunc = Callable[['Parser'], None]
FlagFunc = Callable[[ApplyFunc, Set[ApplyFunc]], bool]


150
class Parser:
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
    """
    (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:

    1. *Named parsers* for which a name is set in field `parser.name`.
       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:
        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.

        grammar:  A reference to the Grammar object to which the parser
                is attached.
    """

208
    def __init__(self) -> None:
209
        # assert isinstance(name, str), str(name)
210
211
212
        self.name = ''  # type: str
        self.ptype = ':' + self.__class__.__name__  # type: str
        self.tag_name = self.ptype  # type: str
213
214
        if not hasattr(self.__class__, 'alive'):
            self._grammar = ZOMBIE_GRAMMAR  # type: Grammar
215
216
217
        self.reset()

    def __deepcopy__(self, memo):
218
        """        Deepcopy method of the parser. Upon instantiation of a Grammar-
219
220
221
222
223
        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.
        """
224
225
        duplicate = self.__class__()
        duplicate.name = self.name
eckhart's avatar
eckhart committed
226
        duplicate.ptype = self.ptype
227
        duplicate.tag_name = self.tag_name
228
        return duplicate
229

230
231
232
233
234
235
236
237
238
239
240
    def __repr__(self):
        return self.name + self.ptype

    def __str__(self):
        return self.name + (' = ' if self.name else '') + repr(self)

    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
        return self.name if self.name else self.__repr__()

241
242
243
244
    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."""
245
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
246
        self.recursion_counter = defaultdict(lambda: 0)  # type: DefaultDict[int, int]
eckhart's avatar
eckhart committed
247
        self.cycle_detection = set()  # type: Set[ApplyFunc]
248

249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
    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
        done in the overridden method `_parse()`.
        """
        grammar = self._grammar  # read protected member instead property to avoid function call
        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
            if location in self.visited:
                # no history recording in case of memoized results
                return self.visited[location]

            # break left recursion at the maximum allowed depth
            if grammar.left_recursion_handling__:
                if self.recursion_counter[location] > LEFT_RECURSION_DEPTH:
                    grammar.recursion_locations__.add(location)
                    return None, text
                self.recursion_counter[location] += 1

            if grammar.history_tracking__:
275
276
                grammar.call_stack__.append(self.repr if self.tag_name in (':RegExp', ':Token')
                                            else self.tag_name)
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
                grammar.moving_forward__ = True

            try:
                # PARSER CALL: run _parse() method
                node, rest = self._parse(text)
            except ParserError as error:
                # does this play well with variable setting? add rollback clause here? tests needed...
                gap = len(text) - len(error.rest)
                rules = grammar.resume_rules__.get(self.name, [])
                rest = error.rest[len(error.node):]
                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:
                        i = 1
                    nd = Node(None, rest[:i])
                    rest = rest[i:]
                    assert error.node.children
                    if error.first_throw:
                        node = error.node
                        node.result += (nd,)
                    else:
299
                        node = Node(self.tag_name, (Node(None, text[:gap]), error.node, nd))
300
301
302
303
                elif error.first_throw:
                    raise ParserError(error.node, error.rest, first_throw=False)
                else:
                    result = (Node(None, text[:gap]), error.node) if gap else error.node  # type: ResultType
304
                    raise ParserError(Node(self.tag_name, result), text, first_throw=False)
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339

            if grammar.left_recursion_handling__:
                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__:
                    if location in self.visited:
                        node, rest = self.visited[location]
                        # TODO: maybe add a warning about occurrence of left-recursion here?
                    # 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
                    self.visited[location] = (None, rest)
            else:
                assert node._pos < 0
                node._pos = location
                assert node._pos >= 0, str("%i < %i" % (grammar.document_length__, location))
                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?
                    self.visited[location] = (node, rest)

            # Mind that memoized parser calls will not appear in the history record!
            # Does this make sense? Or should it be changed?
            if grammar.history_tracking__:
                # don't track returning parsers except in case an error has occurred
                # remaining = len(rest)
                if (grammar.moving_forward__ or (node and node.errors)):
340
341
                    record = HistoryRecord(grammar.call_stack__, node, text,
                                           grammar.line_col__(text))
342
343
344
345
346
347
348
349
350
351
352
353
354
                    grammar.history__.append(record)
                    # print(record.stack, record.status, rest[:20].replace('\n', '|'))
                grammar.moving_forward__ = False
                grammar.call_stack__.pop()

        except RecursionError:
            node = Node(None, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
            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
355
356
357
358
359
360
361
362
363
364
365
366

    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)

367
368
369
370
371
372
373
    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
        # return None, text  # default behaviour: don't match

374
    @property
eckhart's avatar
eckhart committed
375
376
    def grammar(self) -> 'Grammar':
        if self._grammar != ZOMBIE_GRAMMAR:
eckhart's avatar
eckhart committed
377
378
379
            return self._grammar
        else:
            raise AssertionError('Grammar has not yet been set!')
380
381
382

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
eckhart's avatar
eckhart committed
383
        if self._grammar == ZOMBIE_GRAMMAR:
384
385
            self._grammar = grammar
            self._grammar_assigned_notifier()
eckhart's avatar
eckhart committed
386
387
388
        elif self._grammar != grammar:
              raise AssertionError("Parser has already been assigned"
                                   "to a different Grammar object!")
389
390
391
392
393
394

    def _grammar_assigned_notifier(self):
        """A function that notifies the parser object that it has been
        assigned to a grammar."""
        pass

eckhart's avatar
eckhart committed
395
    def _apply(self, func: ApplyFunc, flip: FlagFunc) -> bool:
396
397
        """
        Applies function `func(parser)` recursively to this parser and all
eckhart's avatar
eckhart committed
398
399
400
401
402
403
404
405
406
        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!
407
        """
eckhart's avatar
eckhart committed
408
        if flip(func, self.cycle_detection):
409
410
411
412
413
            return False
        else:
            func(self)
            return True

eckhart's avatar
eckhart committed
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
    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)

444

445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
class ZombieParser(Parser):
    """
    Serves as a substitute for a Parser instance.

    ``ZombieParser`` is the class of the singelton object
    ``ZOMBIE_PARSER``. The  ``ZOMBIE_PARSER`` has a name and can be
    called, but it never matches. It serves as a substitute where only
    these (or one of these properties) is needed, but no real Parser-
    object is instantiated.
    """

    alive = False
    __slots__ = ()

    def __init__(self):
460
461
462
463
        super().__init__()
        self.name = ZOMBIE
        self.ptype = ':' + ZOMBIE
        self.tag_name = ZOMBIE
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
        # no need to call super class constructor
        assert not self.__class__.alive, "There can be only one!"
        assert self.__class__ == ZombieParser, "No derivatives, please!"
        self.__class__.alive = True
        self.reset()

    def __copy__(self):
        return self

    def __deepcopy__(self, memo):
        return self

    def __call__(self, text):
        raise AssertionError("Better call Saul ;-)")

    def _grammar_assigned_notifier(self):
        raise AssertionError("No zombies allowed in any grammar!")

    def apply(self, func: ApplyFunc):
        return "Eaten alive..."


ZOMBIE_PARSER = ZombieParser()


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


496
497
498
499
500
501
502
503
504
505
506
507
508
509
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).
    """
    wspc = '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return wspc


510
class UnknownParserError(KeyError):
eckhart's avatar
eckhart committed
511
    """UnknownParserError is raised if a Grammar object is called with a
512
    parser that does not exist or if in the course of parsing a parser
eckhart's avatar
eckhart committed
513
    is referred to that does not exist."""
514
515


516
517
518
519
520
521
522
523
524
525
526
527
528
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
529
    Example for direct instantiation of a grammar::
530

531
        >>> number = RE('\d+') + RE('\.') + RE('\d+') | RE('\d+')
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
        >>> 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.

    3. The parsers in the class do not necessarily need to be connected
       to one single root parser, which is helpful for testing and
       building up a parser successively of several components.

    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.

557
    Example::
558
559
560
561
562
563
564
565
566

        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+')
567
568
569
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
            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
    'named' parser and its field `parser.name` contains the variable
    name after instantiation of the Grammar class. All other parsers,
    i.e. parsers that are defined within a `named` parser, remain
    "anonymous parsers" where `parser.name` is the empty string, unless
    a name has been passed explicitly upon instantiation.
    If one and the same parser is assigned to several class variables
    such as, for example the parser `expression` in the example above,
    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:

595
596
597
598
599
    Attributes:
        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.
600

Eckhart Arnold's avatar
Eckhart Arnold committed
601
        resume_rules__: A mapping of parser names to a list of regular expressions or search
602
                strings that act as rules to find the the reentry point if a ParserError
Eckhart Arnold's avatar
Eckhart Arnold committed
603
604
                was thrown during the execution of the parser with the respective name.

605
606
607
608
609
        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.
610

eckhart's avatar
eckhart committed
611
612
613
614
        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`).

615
616
617
    Attributes:
        all_parsers__:  A set of all parsers connected to this grammar object

eckhart's avatar
eckhart committed
618
619
620
621
        start_parser__:  During parsing, the parser with which the parsing process
                was started (see method `__call__`) or `None` if no parsing process
                is running.

622
623
624
625
626
627
628
629
630
631
        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
632
633
        document_length__:  the length of the document.

634
635
636
        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
637
                initialized if :attr:`history_tracking__` is true.
638

639
640
641
642
643
        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.

644
        _reversed__:  the same text in reverse order - needed by the `Lookbehind`-
645
646
647
                parsers.

        variables__:  A mapping for variable names to a stack of their respective
648
649
                string values - needed by the :class:`Capture`-, :class:`Retrieve`-
                and :class:`Pop`-parsers.
650
651

        rollback__:  A list of tuples (location, rollback-function) that are
652
653
654
655
                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.
656
657
658

        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
659
660
                backtracks to a location at or before last_rb__loc__ (i.e.
                location <= last_rb__loc__) then a rollback of all variable
661
662
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
663
                calling method :func:`rollback_to__(location)`.
664

665
666
        call_stack__:  A stack of the tag names of all parsers that have been called.
                This is required for recording the parser history (for debugging)
667
668
669
670
671
672
673
674
675
676
                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
677
678
                recording is turned off! (See :func:`add_parser_guard` and its local
                function :func:`guarded_call`)
679
680
681
682

        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
683
684
                if full memoization is enabled. (See :func:`add_parser_guard` and its
                local function :func:`guarded_call`)
685
686
687
688
689
690
691
692
693
694

        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)

        left_recursion_handling__:  Turns left recursion handling on or off.
                If turned off, a recursion error will result in case of left
                recursion.
    """
eckhart's avatar
eckhart committed
695
    python_src__ = ''  # type: str
696
    root__ = ZOMBIE_PARSER  # type: Parser
697
698
    # root__ must be overwritten with the root-parser by grammar subclass
    parser_initialization__ = "pending"  # type: str
eckhart's avatar
eckhart committed
699
    resume_rules__ = dict()  # type: Dict[str, ResumeList]
700
    # some default values
701
702
    # COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    # WSP_RE__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
703
704
705
706
707
708
709


    @classmethod
    def _assign_parser_names__(cls):
        """
        Initializes the `parser.name` fields of those
        Parser objects that are directly assigned to a class field with
710
        the field's name, e.g.::
711

712
713
            class Grammar(Grammar):
                ...
714
                symbol = RE('(?!\\d)\\w+')
715

di68kap's avatar
di68kap committed
716
717
718
        After the call of this method symbol.name == "symbol" holds.
        Parser names starting or ending with a double underscore like
        ``root__`` will be ignored. See :func:`sane_parser_name()`
719
720
721
722
723
724
725
726
727
728
729
730
731
732

        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.
        """
        if cls.parser_initialization__ != "done":
            cdict = cls.__dict__
            for entry, parser in cdict.items():
                if isinstance(parser, Parser) and sane_parser_name(entry):
di68kap's avatar
di68kap committed
733
                    if isinstance(parser, Forward):
di68kap's avatar
di68kap committed
734
                        if not cast(Forward, parser).parser.name:
di68kap's avatar
di68kap committed
735
736
                            cast(Forward, parser).parser.name = entry
                    else:   # if not parser.name:
737
                        parser.name = entry
738
739
740
741
            cls.parser_initialization__ = "done"


    def __init__(self, root: Parser = None) -> None:
742
743
        self.all_parsers__ = set()             # type: Set[Parser]
        self.start_parser__ = None             # type: Optional[Parser]
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
        self._dirty_flag__ = False             # type: bool
        self.history_tracking__ = False        # type: bool
        self.memoization__ = True              # type: bool
        self.left_recursion_handling__ = True  # type: bool
        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.
        self.root__ = copy.deepcopy(root) if root else copy.deepcopy(self.__class__.root__)
        self.root__.apply(self._add_parser__)


    def __getitem__(self, key):
        try:
            return self.__dict__[key]
        except KeyError:
            parser_template = getattr(self, key, None)
            if parser_template:
                # add parser to grammar object on the fly...
                parser = copy.deepcopy(parser_template)
                parser.apply(self._add_parser__)
                # assert self[key] == parser
                return self[key]
772
            raise UnknownParserError('Unknown parser "%s" !' % key)
773

774
775
    def __contains__(self, key):
        return key in self.__dict__ or hasattr(self, key)
776
777

    def _reset__(self):
778
        self.tree__ = RootNode()              # type: RootNode
779
780
        self.document__ = EMPTY_STRING_VIEW   # type: StringView
        self._reversed__ = EMPTY_STRING_VIEW  # type: StringView
eckhart's avatar
eckhart committed
781
        self.document_length__ = 0            # type: int
782
783
        self.document_lbreaks__ = []          # type: List[int]
        # variables stored and recalled by Capture and Retrieve parsers
eckhart's avatar
eckhart committed
784
        self.variables__ = defaultdict(lambda: [])  # type: DefaultDict[str, List[str]]
785
786
787
        self.rollback__ = []                  # type: List[Tuple[int, Callable]]
        self.last_rb__loc__ = -1              # type: int
        # support for call stack tracing
788
        self.call_stack__ = []                # type: List[str]
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
        # 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]


    @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__:
804
            self._reversed__ = StringView(self.document__.get_text()[::-1])
805
806
807
808
809
810
811
812
813
814
815
        return self._reversed__


    def _add_parser__(self, parser: Parser) -> None:
        """
        Adds the particular copy of the parser object to this
        particular instance of Grammar.
        """
        if parser.name:
            # prevent overwriting instance variables or parsers of a different class
            assert parser.name not in self.__dict__ or \
816
                isinstance(self.__dict__[parser.name], parser.__class__), \
817
                ('Cannot add parser "%s" because a field with the same name '
di68kap's avatar
di68kap committed
818
819
                 'already exists in grammar object: %s!'
                 % (parser.name, str(self.__dict__[parser.name])))
820
            setattr(self, parser.name, parser)
821
        parser.tag_name = parser.name or parser.ptype
822
823
824
825
        self.all_parsers__.add(parser)
        parser.grammar = self


eckhart's avatar
eckhart committed
826
827
828
829
    def __call__(self,
                 document: str,
                 start_parser: Union[str, Parser] = "root__",
                 track_history: bool = False) -> RootNode:
830
831
832
833
834
        """
        Parses a document with with parser-combinators.

        Args:
            document (str): The source text to be parsed.
di68kap's avatar
di68kap committed
835
836
            start_parser (str or Parser): The name of the parser with which
                to start. This is useful for testing particular parsers
837
                (i.e. particular parts of the EBNF-Grammar.)
di68kap's avatar
di68kap committed
838
839
840
841
842
            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.
843
        Returns:
di68kap's avatar
di68kap committed
844
            Node: The root node to the parse tree.
845
        """
846

eckhart's avatar
eckhart committed
847
        def tail_pos(predecessors: Union[List[Node], Tuple[Node, ...], None]) -> int:
eckhart's avatar
eckhart committed
848
849
            """Adds the position after the last node in the list of
            predecessors to the node."""
eckhart's avatar
eckhart committed
850
            return predecessors[-1].pos + len(predecessors[-1]) if predecessors else 0
eckhart's avatar
eckhart committed
851

852
853
854
855
856
857
858
859
860
        # assert isinstance(document, str), type(document)
        if self.root__ is None:
            raise NotImplementedError()
        if self._dirty_flag__:
            self._reset__()
            for parser in self.all_parsers__:
                parser.reset()
        else:
            self._dirty_flag__ = True
di68kap's avatar
di68kap committed
861
862
863
864
        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__
865
        self.document__ = StringView(document)
eckhart's avatar
eckhart committed
866
        self.document_length__ = len(self.document__)
867
        self.document_lbreaks__ = linebreaks(document) if self.history_tracking__ else []
Eckhart Arnold's avatar
Eckhart Arnold committed
868
        self.last_rb__loc__ = -1  # rollback location
869
        parser = self[start_parser] if isinstance(start_parser, str) else start_parser
eckhart's avatar
eckhart committed
870
        self.start_parser__ = parser
871
872
873
874
875
876
877
878
        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's avatar
eckhart committed
879
                result = Node(None, '').init_pos(0)
eckhart's avatar
eckhart committed
880
                self.tree__.new_error(result,
di68kap's avatar
di68kap committed
881
882
                                      'Parser "%s" did not match empty document.' % str(parser),
                                      Error.PARSER_DID_NOT_MATCH)
883
884
885
886
887
888
        while rest and len(stitches) < MAX_DROPOUTS:
            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
889
890
891
892
893
894
                    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__)))
                    # Check if a Lookahead-Parser did match. Needed for testing, because
                    # in a test case this is not necessarily an error.
eckhart's avatar
eckhart committed
895
                    last_record = self.history__[-2] if len(self.history__) > 1 else None  # type: Optional[HistoryRecord]
896

di68kap's avatar
di68kap committed
897
898
                    if last_record and parser != self.root__ \
                            and last_record.status == HistoryRecord.MATCH \
eckhart's avatar
eckhart committed
899
900
                            and last_record.node.pos \
                            + len(last_record.node) >= len(self.document__) \
901
902
903
                            and any(tn in self and isinstance(self[tn], Lookahead)
                                    or tn[0] == ':' and issubclass(eval(tn[1:]), Lookahead)
                                    for tn in last_record.call_stack):
di68kap's avatar
di68kap committed
904
905
906
907
908
                        error_msg = 'Parser did not match except for lookahead! ' + err_info
                        error_code = Error.PARSER_LOOKAHEAD_MATCH_ONLY
                    else:
                        error_msg = 'Parser did not match!' + err_info
                        error_code = Error.PARSER_DID_NOT_MATCH
909
910
                else:
                    stitches.append(result)
eckhart's avatar
eckhart committed
911
912
913
914
915
916
                    error_msg = "Parser stopped before end" \
                        + (("! trying to recover"
                            + (" but stopping history recording at this point."
                               if self.history_tracking__ else "..."))
                            if len(stitches) < MAX_DROPOUTS
                            else " too often! Terminating parser.")
di68kap's avatar
di68kap committed
917
                    error_code = Error.PARSER_STOPPED_BEFORE_END
eckhart's avatar
eckhart committed
918
                stitches.append(Node(None, skip).init_pos(tail_pos(stitches)))
di68kap's avatar
di68kap committed
919
                self.tree__.new_error(stitches[-1], error_msg, error_code)
920
                if self.history_tracking__:
eckhart's avatar
eckhart committed
921
922
923
924
925
926
927
                    # # some parsers may have matched and left history records with nodes != None.
                    # # Because these are not connected to the stitched root node, their pos-
                    # # properties will not be initialized by setting the root node's pos property
                    # # to zero. Therefore, their pos properties need to be initialized here
                    # for record in self.history__:
                    #     if record.node and record.node._pos < 0:
                    #         record.node.init_pos(0)
928
929
                    record = HistoryRecord(self.call_stack__.copy(), stitches[-1], rest,
                                           self.line_col__(rest))
930
931
932
933
934
                    self.history__.append(record)
                    # stop history tracking when parser returned too early
                    self.history_tracking__ = False
        if stitches:
            if rest:
Eckhart Arnold's avatar
Eckhart Arnold committed
935
                stitches.append(Node(None, rest))
936
937
938
939
940
941
            #try:
            result = Node(None, tuple(stitches)).init_pos(0)
            # except AssertionError as error:
            #     # some debugging output
            #     print(Node(None, tuple(stitches)).as_sxpr())
            #     raise error
942
        if any(self.variables__.values()):
di68kap's avatar
di68kap committed
943
            error_msg = "Capture-retrieve-stack not empty after end of parsing: " \
944
                + str(self.variables__)
di68kap's avatar
di68kap committed
945
            error_code = Error.CAPTURE_STACK_NOT_EMPTY
946
947
948
949
950
            if result:
                if result.children:
                    # add another child node at the end to ensure that the position
                    # of the error will be the end of the text. Otherwise, the error
                    # message above ("...after end of parsing") would appear illogical.
951
                    error_node = Node(ZOMBIE, '').init_pos(tail_pos(result.children))
di68kap's avatar
di68kap committed
952
                    self.tree__.new_error(error_node, error_msg, error_code)
eckhart's avatar
eckhart committed
953
                    result.result = result.children + (error_node,)
954
                else:
di68kap's avatar
di68kap committed
955
                    self.tree__.new_error(result, error_msg, error_code)
eckhart's avatar
eckhart committed
956
        # result.pos = 0  # calculate all positions
957
        # result.collect_errors(self.document__)
eckhart's avatar
eckhart committed
958
959
        if result:
            self.tree__.swallow(result)
eckhart's avatar
eckhart committed
960
        self.start_parser__ = None
961
        return self.tree__
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979


    def push_rollback__(self, location, func):
        """
        Adds a rollback function that either removes or re-adds
        values on the variable stack (`self.variables`) that have been
        added (or removed) by Capture or Pop Parsers, the results of
        which have been dismissed.
        """
        self.rollback__.append((location, func))
        self.last_rb__loc__ = location


    def rollback_to__(self, location):
        """
        Rolls back the variable stacks (`self.variables`) to its
        state at an earlier location in the parsed document.
        """
Eckhart Arnold's avatar
Eckhart Arnold committed
980
        while self.rollback__ and self.rollback__[-1][0] >= location:
981
982
983
984
985
986
987
988
989
990
            _, rollback_func = self.rollback__.pop()
            # assert not loc > self.last_rb__loc__, \
            #     "Rollback confusion: line %i, col %i < line %i, col %i" % \
            #     (*line_col(self.document__, len(self.document__) - loc),
            #      *line_col(self.document__, len(self.document__) - self.last_rb__loc__))
            rollback_func()
        self.last_rb__loc__ == self.rollback__[-1][0] if self.rollback__ \
            else (len(self.document__) + 1)


991
992
993
994
995
996
997
998
999
    def line_col__(self, text):
        """
        Returns the line and column where text is located in the document.
        Does not check, whether text is actually a substring of the currently
        parsed document.
        """
        return line_col(self.document_lbreaks__, self.document_length__ - len(text))


1000
def dsl_error_msg(parser: Parser, error_str: str) -> str: