parse.py 102 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

37
from DHParser.error import Error, linebreaks, line_col
38
39
from DHParser.log import is_logging, HistoryRecord
from DHParser.preprocess import BEGIN_TOKEN, END_TOKEN, RX_TOKEN_NAME
40
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
41
42
from DHParser.syntaxtree import Node, FrozenNode, RootNode, WHITESPACE_PTYPE, \
    TOKEN_PTYPE, ZOMBIE_TAG, ResultType
43
from DHParser.toolkit import sane_parser_name, escape_control_characters, get_config_value, \
eckhart's avatar
eckhart committed
44
    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
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
       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.

210
        _grammar:  A reference to the Grammar object to which the parser
211
212
213
                is attached.
    """

214
    def __init__(self) -> None:
215
        # assert isinstance(name, str), str(name)
216
217
218
        self.pname = ''               # type: str
        self.tag_name = self.ptype    # type: str
        self.cycle_detection = set()  # type: Set[ApplyFunc]
219
220
221
222
        try:
            self._grammar = GRAMMAR_PLACEHOLDER  # type: Grammar
        except NameError:
            pass
223
224
225
        self.reset()

    def __deepcopy__(self, memo):
226
        """        Deepcopy method of the parser. Upon instantiation of a Grammar-
227
228
229
230
231
        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.
        """
232
        duplicate = self.__class__()
di68kap's avatar
di68kap committed
233
        duplicate.pname = self.pname
234
        duplicate.tag_name = self.tag_name
235
        return duplicate
236

237
    def __repr__(self):
di68kap's avatar
di68kap committed
238
        return self.pname + self.ptype
239
240

    def __str__(self):
di68kap's avatar
di68kap committed
241
        return self.pname + (' = ' if self.pname else '') + repr(self)
242

243
244
245
246
247
248
    @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__

249
250
251
    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
di68kap's avatar
di68kap committed
252
        return self.pname if self.pname else self.__repr__()
253

254
255
256
257
    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."""
258
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
259
        self.recursion_counter = defaultdict(int)  # type: DefaultDict[int, int]
260

eckhart's avatar
eckhart committed
261
    @cython.locals(location=cython.int, gap=cython.int, i=cython.int)
262
263
264
265
266
    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()`.
        """
267
268
269
270
271
272
273
274
275
        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

276
        grammar = self._grammar
277
278
279
280
281
282
283
284
        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
285
286
            visited = self.visited  # using local variable for better performance
            if location in visited:
287
                # no history recording in case of memoized results
288
                return visited[location]
289
290

            # break left recursion at the maximum allowed depth
291
292
293
            left_recursion_depth__ = grammar.left_recursion_depth__
            if left_recursion_depth__:
                if self.recursion_counter[location] > left_recursion_depth__:
294
295
296
297
                    grammar.recursion_locations__.add(location)
                    return None, text
                self.recursion_counter[location] += 1

298
299
            history_tracking__ = grammar.history_tracking__
            if history_tracking__:
300
301
                grammar.call_stack__.append(self.repr if self.tag_name in
                                                         (':RegExp', ':Token', ':DropToken')
302
                                            else self.tag_name)
303
                grammar.moving_forward__ = True
304
                error = None
305
306
307
308

            try:
                # PARSER CALL: run _parse() method
                node, rest = self._parse(text)
309
            except ParserError as pe:
310
                # does this play well with variable setting? add rollback clause here? tests needed...
311
                gap = len(text) - len(pe.rest)
di68kap's avatar
di68kap committed
312
                rules = grammar.resume_rules__.get(self.pname, [])
313
                rest = pe.rest[len(pe.node):]
314
315
316
317
                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:
318
                        i = 1
319
                    nd = Node(ZOMBIE_TAG, rest[:i]).with_pos(location)
320
                    nd.attr['err'] = pe.error.message
321
                    rest = rest[i:]
322
323
324
                    assert pe.node.children or (not pe.node.result)
                    if pe.first_throw:
                        node = pe.node
eckhart's avatar
eckhart committed
325
                        node.result = node.children + (nd,)
326
                    else:
327
                        node = Node(self.tag_name,
328
                                    (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node, nd))
329
330
                elif pe.first_throw:
                    raise ParserError(pe.node, pe.rest, pe.error, first_throw=False)
331
332
                elif grammar.tree__.errors[-1].code == Error.MANDATORY_CONTINUATION_AT_EOF:  # EXPERIMENTAL!!
                    node = pe.node
333
                else:
334
335
                    result = (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) if gap \
                        else pe.node  # type: ResultType
336
337
                    raise ParserError(Node(self.tag_name, result).with_pos(location),
                                      text, pe.error, first_throw=False)
338
339
                error = pe.error  # needed for history tracking

340

341
            if left_recursion_depth__:
342
343
344
345
346
347
                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__:
348
349
                    if location in visited:
                        node, rest = visited[location]
350
                        if location != grammar.last_recursion_location__:
351
352
353
                            grammar.tree__.add_error(
                                node, Error("Left recursion encountered. "
                                            "Refactor grammar to avoid slow parsing.",
354
355
                                            node.pos if node else location,
                                            Error.LEFT_RECURSION_WARNING))
356
                            error_id = id(node)
357
                            grammar.last_recursion_location__ = location
358
359
360
361
                    # 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
362
                    visited[location] = (None, rest)
363
            else:
364
                # assert node._pos < 0 or node == EMPTY_NODE
365
                node._pos = location
366
367
                # assert node._pos >= 0 or node == EMPTY_NODE, \
                #     str("%i < %i" % (grammar.document_length__, location))
368
369
370
371
372
373
374
                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?
375
                    visited[location] = (node, rest)
376
377
378

            # Mind that memoized parser calls will not appear in the history record!
            # Does this make sense? Or should it be changed?
379
            if history_tracking__:
380
381
                # don't track returning parsers except in case an error has occurred
                # remaining = len(rest)
382
                if grammar.moving_forward__:
383
                    record = HistoryRecord(grammar.call_stack__, node, text,
384
                                           grammar.line_col__(text))
385
                    grammar.history__.append(record)
386
387
388
389
390
391
392
                elif error:
                    # error_nid = id(node)  # type: int
                    # if error_nid in grammar.tree__.error_nodes:
                    record = HistoryRecord(grammar.call_stack__, node, text,
                                           grammar.line_col__(text),
                                           [error])
                    grammar.history__.append(record)
393
394
395
396
                grammar.moving_forward__ = False
                grammar.call_stack__.pop()

        except RecursionError:
397
            node = Node(ZOMBIE_TAG, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
398
399
400
401
402
403
            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
404
405
406
407
408
409
410
411
412
413
414
415

    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
416

417
418
419
420
421
422
    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

423
    @property
eckhart's avatar
eckhart committed
424
    def grammar(self) -> 'Grammar':
425
426
427
428
429
        try:
            if self._grammar != GRAMMAR_PLACEHOLDER:
                return self._grammar
            else:
                raise AssertionError('Grammar has not yet been set!')
430
431
        except (AttributeError, NameError):
            raise AttributeError('Parser placeholder does not have a grammar!')
432
433
434

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
435
436
437
438
439
440
441
442
443
        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
444
445
        except NameError:  # Cython: No access to GRAMMA_PLACEHOLDER, yet :-(
            self._grammar = grammar
446

eckhart's avatar
eckhart committed
447
    def _apply(self, func: ApplyFunc, flip: FlagFunc) -> bool:
448
449
        """
        Applies function `func(parser)` recursively to this parser and all
eckhart's avatar
eckhart committed
450
451
452
453
454
455
456
457
458
        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!
459
        """
eckhart's avatar
eckhart committed
460
        if flip(func, self.cycle_detection):
461
462
463
464
465
            return False
        else:
            func(self)
            return True

eckhart's avatar
eckhart committed
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
    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)

496

497
PARSER_PLACEHOLDER = Parser()
498
499
500
501
502
503
504
505
506


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


507
508
509
510
511
512
513
514
515
516
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).
    """
517
518
519
    if comment:
        return '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return whitespace
520
521


522
class UnknownParserError(KeyError):
eckhart's avatar
eckhart committed
523
    """UnknownParserError is raised if a Grammar object is called with a
524
    parser that does not exist or if in the course of parsing a parser
eckhart's avatar
eckhart committed
525
    is referred to that does not exist."""
526
527


528
GrammarErrorType = Tuple[str, Parser, Error]      # TODO: replace with a named tuple?
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545


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))


546
547
548
549
550
551
552
553
554
555
556
557
558
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
559
    Example for direct instantiation of a grammar::
560

561
        >>> number = RE('\d+') + RE('\.') + RE('\d+') | RE('\d+')
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
        >>> 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.

587
    Example::
588
589
590
591
592
593
594
595
596

        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+')
597
598
599
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
600
601
602
603
604
            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
605
    'named' parser and its field `parser.pname` contains the variable
606
607
    name after instantiation of the Grammar class. All other parsers,
    i.e. parsers that are defined within a `named` parser, remain
di68kap's avatar
di68kap committed
608
    "anonymous parsers" where `parser.pname` is the empty string, unless
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
    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:

625
    Class Attributes:
626
627
628
629
        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.
630

Eckhart Arnold's avatar
Eckhart Arnold committed
631
        resume_rules__: A mapping of parser names to a list of regular expressions or search
632
                strings that act as rules to find the the reentry point if a ParserError
Eckhart Arnold's avatar
Eckhart Arnold committed
633
634
                was thrown during the execution of the parser with the respective name.

635
636
637
638
639
        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.
640

641
642
643
644
645
646
647
        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.
648

eckhart's avatar
eckhart committed
649
650
651
652
        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`).

653
    Instance Attributes:
654
655
        all_parsers__:  A set of all parsers connected to this grammar object

eckhart's avatar
eckhart committed
656
657
658
659
        start_parser__:  During parsing, the parser with which the parsing process
                was started (see method `__call__`) or `None` if no parsing process
                is running.

660
661
662
663
664
665
666
667
668
669
        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
670
671
        document_length__:  the length of the document.

672
673
674
        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
675
                initialized if :attr:`history_tracking__` is true.
676

677
678
679
680
681
        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.

682
        _reversed__:  the same text in reverse order - needed by the `Lookbehind`-
683
684
685
                parsers.

        variables__:  A mapping for variable names to a stack of their respective
686
687
                string values - needed by the :class:`Capture`-, :class:`Retrieve`-
                and :class:`Pop`-parsers.
688
689

        rollback__:  A list of tuples (location, rollback-function) that are
690
691
692
693
                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.
694
695
696

        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
697
698
                backtracks to a location at or before last_rb__loc__ (i.e.
                location <= last_rb__loc__) then a rollback of all variable
699
700
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
701
                calling method :func:`rollback_to__(location)`.
702

703
704
        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)
705
706
707
708
709
710
711
712
713
714
                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
715
716
                recording is turned off! (See :func:`add_parser_guard` and its local
                function :func:`guarded_call`)
717
718
719
720

        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
721
722
723
724
725
                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.
726
727
728
729
730
731

        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)

732
733
734
735
736
        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.
737
738
739
740
741
742
743
744

        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
745
    """
eckhart's avatar
eckhart committed
746
    python_src__ = ''  # type: str
747
    root__ = PARSER_PLACEHOLDER  # type: Parser
748
    # root__ must be overwritten with the root-parser by grammar subclass
eckhart's avatar
eckhart committed
749
    parser_initialization__ = ["pending"]  # type: List[str]
eckhart's avatar
eckhart committed
750
    resume_rules__ = dict()  # type: Dict[str, ResumeList]
751
    # some default values
752
753
    # COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    # WSP_RE__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
754
    static_analysis_pending__ = [True]  # type: List[bool]
755
756
757
758
759


    @classmethod
    def _assign_parser_names__(cls):
        """
di68kap's avatar
di68kap committed
760
        Initializes the `parser.pname` fields of those
761
        Parser objects that are directly assigned to a class field with
762
        the field's name, e.g.::
763

764
765
            class Grammar(Grammar):
                ...
766
                symbol = RE('(?!\\d)\\w+')
767

di68kap's avatar
di68kap committed
768
        After the call of this method symbol.pname == "symbol" holds.
di68kap's avatar
di68kap committed
769
770
        Parser names starting or ending with a double underscore like
        ``root__`` will be ignored. See :func:`sane_parser_name()`
771
772
773
774
775
776
777
778
779
780

        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
781
        if cls.parser_initialization__[0] != "done":
782
783
784
            cdict = cls.__dict__
            for entry, parser in cdict.items():
                if isinstance(parser, Parser) and sane_parser_name(entry):
di68kap's avatar
di68kap committed
785
                    if isinstance(parser, Forward):
di68kap's avatar
di68kap committed
786
787
788
789
                        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
790
            cls.parser_initialization__[0] = "done"
791
792
793


    def __init__(self, root: Parser = None) -> None:
794
795
        self.all_parsers__ = set()             # type: Set[Parser]
        self.start_parser__ = None             # type: Optional[Parser]
796
797
798
        self._dirty_flag__ = False             # type: bool
        self.history_tracking__ = False        # type: bool
        self.memoization__ = True              # type: bool
799
800
801
        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
802
803
804
805
806
807
808
809
810
        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
811
812
813
814
        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__']
815

816
817
818
819
820
821
822
823
824
        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
825

826
827
828
829
830

    def __getitem__(self, key):
        try:
            return self.__dict__[key]
        except KeyError:
eckhart's avatar
eckhart committed
831
            parser_template = getattr(self.__class__, key, None)
832
833
834
835
            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
836
                assert self[key] == parser
837
                return self[key]
838
            raise UnknownParserError('Unknown parser "%s" !' % key)
839

840

841
842
    def __contains__(self, key):
        return key in self.__dict__ or hasattr(self, key)
843

844

845
    def _reset__(self):
846
        self.tree__ = RootNode()              # type: RootNode
847
848
        self.document__ = EMPTY_STRING_VIEW   # type: StringView
        self._reversed__ = EMPTY_STRING_VIEW  # type: StringView
eckhart's avatar
eckhart committed
849
        self.document_length__ = 0            # type: int
850
851
        self.document_lbreaks__ = []          # type: List[int]
        # variables stored and recalled by Capture and Retrieve parsers
eckhart's avatar
eckhart committed
852
        self.variables__ = defaultdict(lambda: [])  # type: DefaultDict[str, List[str]]
853
854
855
        self.rollback__ = []                  # type: List[Tuple[int, Callable]]
        self.last_rb__loc__ = -1              # type: int
        # support for call stack tracing
856
        self.call_stack__ = []                # type: List[str]
857
858
859
860
861
        # 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]
862
        self.last_recursion_location__ = -1   # type: int
863
864
865
866
867
868
869
870
871
872


    @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__:
873
            self._reversed__ = StringView(self.document__.get_text()[::-1])
874
875
876
877
878
879
880
881
        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
882
        if parser.pname:
883
            # prevent overwriting instance variables or parsers of a different class
di68kap's avatar
di68kap committed
884
885
            assert parser.pname not in self.__dict__ or \
                   isinstance(self.__dict__[parser.pname], parser.__class__), \
886
                ('Cannot add parser "%s" because a field with the same name '
di68kap's avatar
di68kap committed
887
                 'already exists in grammar object: %s!'
di68kap's avatar
di68kap committed
888
889
890
                 % (parser.pname, str(self.__dict__[parser.pname])))
            setattr(self, parser.pname, parser)
        parser.tag_name = parser.pname or parser.ptype
891
892
893
894
        self.all_parsers__.add(parser)
        parser.grammar = self


eckhart's avatar
eckhart committed
895
896
    def __call__(self,
                 document: str,
eckhart's avatar
eckhart committed
897
                 start_parser: Union[str, Parser] = "root_parser__",
eckhart's avatar
eckhart committed
898
                 track_history: bool = False) -> RootNode:
899
900
901
902
903
        """
        Parses a document with with parser-combinators.

        Args:
            document (str): The source text to be parsed.
di68kap's avatar
di68kap committed
904
905
            start_parser (str or Parser): The name of the parser with which
                to start. This is useful for testing particular parsers
906
                (i.e. particular parts of the EBNF-Grammar.)
di68kap's avatar
di68kap committed
907
908
909
910
911
            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.
912
        Returns:
di68kap's avatar
di68kap committed
913
            Node: The root node to the parse tree.
914
        """
915

eckhart's avatar
eckhart committed
916
        def tail_pos(predecessors: Union[List[Node], Tuple[Node, ...], None]) -> int:
eckhart's avatar
eckhart committed
917
918
            """Adds the position after the last node in the list of
            predecessors to the node."""
eckhart's avatar
eckhart committed
919
            return predecessors[-1].pos + len(predecessors[-1]) if predecessors else 0
eckhart's avatar
eckhart committed
920

921
922
923
924
925
926
927
928
929
930
931
932
        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
            that the testing framework can know that the failure is only a
            test-case-artifact and no real failure.
            (See test/test_testing.TestLookahead !)
            """
            last_record = self.history__[-2] if len(self.history__) > 1 else None  # type: Optional[HistoryRecord]
            return last_record and parser != self.root_parser__ \
933
934
935
                    and any(self.history__[i].status == HistoryRecord.MATCH
                            and ((self.history__[i].node.pos + len(self.history__[i].node))
                                 if self.history__[i].node else 0) >= len(self.document__)
936
937
938
939
940
                            and any(tn in self and isinstance(self[tn], Lookahead)
                                    or tn[0] == ':' and issubclass(eval(tn[1:]), Lookahead)
                                    for tn in self.history__[i].call_stack)
                            for i in range(-2, -len(self.history__)-1, -1))

941
942
943
944
945
946
947
        # 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
948
949
950
951
        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__
952
        self.document__ = StringView(document)
eckhart's avatar
eckhart committed
953
        self.document_length__ = len(self.document__)
954
        self.document_lbreaks__ = linebreaks(document) if self.history_tracking__ else []
Eckhart Arnold's avatar
Eckhart Arnold committed
955
        self.last_rb__loc__ = -1  # rollback location
956
        parser = self[start_parser] if isinstance(start_parser, str) else start_parser
957
        self.start_parser__ = parser.parser if isinstance(parser, Forward) else parser
958
959
960
961
962
963
964
965
        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
966
                result = Node(ZOMBIE_TAG, '').with_pos(0)
967
968
                if lookahead_failure_only(parser):
                    self.tree__.new_error(
969
970
                        result, 'Parser "%s" only did not match empty document '
                                'because of lookahead' % str(parser),
971
                        Error.PARSER_LOOKAHEAD_FAILURE_ONLY)
972
973
974
975
976
                else:
                    self.tree__.new_error(
                        result, 'Parser "%s" did not match empty document.' % str(parser),
                        Error.PARSER_DID_NOT_MATCH)

977
978
979
        # 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:
980
981
982
983
984
            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
985
986
987
988
989
990
                    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.
991
                    if lookahead_failure_only(parser):
992
993
                        error_msg = 'Parser "%s" only did not match because of lookahead! ' \
                                    % str(parser) + err_info
994
                        error_code = Error.PARSER_LOOKAHEAD_FAILURE_ONLY
di68kap's avatar
di68kap committed
995
                    else:
996
                        error_msg = 'Parser "%s" did not match!' % str(parser) + err_info
di68kap's avatar
di68kap committed
997
                        error_code = Error.PARSER_DID_NOT_MATCH
998
999
                else:
                    stitches.append(result)
1000
                    h = HistoryRecord.most_advanced_match(self.history__) \