parse.py 111 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
35
from typing import Callable, cast, List, Tuple, Set, Container, Dict, \
36
    DefaultDict, Union, Optional, Any
37

eckhart's avatar
eckhart committed
38
from DHParser.configuration import get_config_value
39
from DHParser.error import Error, linebreaks, line_col
40
41
from DHParser.log import is_logging, HistoryRecord
from DHParser.preprocess import BEGIN_TOKEN, END_TOKEN, RX_TOKEN_NAME
42
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
43
44
from DHParser.syntaxtree import Node, FrozenNode, RootNode, WHITESPACE_PTYPE, \
    TOKEN_PTYPE, ZOMBIE_TAG, ResultType
45
from DHParser.toolkit import sane_parser_name, escape_control_characters, re, cython, \
46
    RX_NEVER_MATCH, RxPatternType
47

48

49
__all__ = ('Parser',
50
           'UnknownParserError',
51
52
           'GrammarErrorType',
           'GrammarError',
53
           'Grammar',
Eckhart Arnold's avatar
Eckhart Arnold committed
54
           'EMPTY_NODE',
55
           'PreprocessorToken',
56
           'Token',
57
           'DropToken',
58
           'RegExp',
59
60
           'RE',
           'TKN',
61
           'Whitespace',
62
           'DropRegExp',
63
           'mixin_comment',
Eckhart Arnold's avatar
Eckhart Arnold committed
64
           'mixin_noempty',
65
           'MetaParser',
eckhart's avatar
eckhart committed
66
67
           'UnaryParser',
           'NaryParser',
68
           'Drop',
69
70
71
72
73
74
75
76
77
           'Synonym',
           'Option',
           'ZeroOrMore',
           'OneOrMore',
           'Series',
           'Alternative',
           'AllOf',
           'SomeOf',
           'Unordered',
eckhart's avatar
eckhart committed
78
           'Required',
79
80
81
82
83
84
85
86
87
88
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
           'last_value',
           'counterpart',
           'accumulate',
           'Capture',
           'Retrieve',
           'Pop',
89
           'Forward')
90
91
92
93


########################################################################
#
94
# Parser base class
95
96
97
98
#
########################################################################


99
EMPTY_NODE = FrozenNode(':EMPTY__', '')
100
101


Eckhart Arnold's avatar
Eckhart Arnold committed
102
103
104
class ParserError(Exception):
    """
    A `ParserError` is thrown for those parser errors that allow the
105
    controlled re-entrance of the parsing process after the error occurred.
Eckhart Arnold's avatar
Eckhart Arnold committed
106
    If a reentry-rule has been configured for the parser where the error
107
    occurred, the parser guard can resume the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
108
109
110
111
112

    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.
    """
113
114
115
116
    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]
117
        self.first_throw = first_throw  # type: bool
Eckhart Arnold's avatar
Eckhart Arnold committed
118

119
    def __str__(self):
120
        return "%i: %s    %s" % (self.node.pos, str(self.rest[:25]), repr(self.node))
121

Eckhart Arnold's avatar
Eckhart Arnold committed
122

123
ResumeList = List[RxPatternType]  # list of regular expressiones
Eckhart Arnold's avatar
Eckhart Arnold committed
124
125


126
def reentry_point(rest: StringView, rules: ResumeList, comment_regex) -> int:
Eckhart Arnold's avatar
Eckhart Arnold committed
127
128
    """
    Finds the point where parsing should resume after a ParserError has been caught.
129
    The algorithm makes sure that this reentry-point does not lie inside a comment.
130
131
132
    The re-entry point is always the point after the end of the match of the regular
    expression defining the re-entry point. (Use look ahead, if you wand to define
    the re-entry point by what follows rather than by what text precedes the point.)
Eckhart Arnold's avatar
Eckhart Arnold committed
133
134
    Args:
        rest:  The rest of the parsed text or, in other words, the point where
135
136
137
138
            a ParserError was thrown.
        rules: A list of regular expressions. The rest of the text is searched for
            each of these. The closest match is the point where parsing will be
            resumed.
139
        comment_regex: A regular expression object that matches comments.
Eckhart Arnold's avatar
Eckhart Arnold committed
140
    Returns:
141
142
        The integer index of the closest reentry point or -1 if no reentry-point
        was found.
Eckhart Arnold's avatar
Eckhart Arnold committed
143
    """
144
    upper_limit = len(rest) + 1
145
    closest_match = upper_limit
146
147
148
149
150
151
152
153
154
155
156
157
158
    comments = None  # typ: Optional[Iterator]

    def next_comment() -> Tuple[int, int]:
        nonlocal rest, comments
        if comments:
            try:
                m = next(comments)
                a, b = m.span()
                return rest.index(a), rest.index(b)
            except StopIteration:
                comments = None
        return -1, -2

159
160
161
    # def str_search(s, start: int = 0) -> Tuple[int, int]:
    #     nonlocal rest
    #     return rest.find(s, start), len(s)
162
163
164

    def rx_search(rx, start: int = 0) -> Tuple[int, int]:
        nonlocal rest
165
166
167
        m = rest.search(rx, start)
        if m:
            start, end = m.span()
168
            return rest.index(start), end - start
169
170
        return -1, 0

171
172
173
174
175
    def entry_point(search_func, search_rule) -> int:
        a, b = next_comment()
        k, length = search_func(search_rule)
        while a < b <= k:
            a, b = next_comment()
176
177
178
        # find next as long as start or end point of resume regex are inside a comment
        while (a < k < b) or (a < k + length < b):
            k, length = search_func(search_rule, b)
179
180
            while a < b <= k:
                a, b = next_comment()
181
        return k + length if k >= 0 else upper_limit
182

183
    # find closest match
Eckhart Arnold's avatar
Eckhart Arnold committed
184
    for rule in rules:
185
        comments = rest.finditer(comment_regex)
186
187
188
        assert not isinstance(rule, str), \
            'Strings not allowed as search rules, use a regular expression instead.'
        pos = entry_point(rx_search, rule)
189
        closest_match = min(pos, closest_match)
190

191
    # in case no rule matched return -1
192
193
194
    if closest_match == upper_limit:
        closest_match = -1
    return closest_match
Eckhart Arnold's avatar
Eckhart Arnold committed
195
196


eckhart's avatar
eckhart committed
197
198
199
200
ApplyFunc = Callable[['Parser'], None]
FlagFunc = Callable[[ApplyFunc, Set[ApplyFunc]], bool]


201
202
203
204
205
206
207
def copy_parser_attrs(src: 'Parser', duplicate: 'Parser'):
    """Duplicates all parser attributes from source to dest."""
    duplicate.pname = src.pname
    duplicate.anonymous = src.anonymous
    duplicate.tag_name = src.tag_name


208
class Parser:
209
210
211
212
213
214
215
216
217
218
219
220
221
222
    """
    (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
223
    1. *Named parsers* for which a name is set in field `parser.pname`.
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
       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:
247
248
249
250
251
252
253
        pname:  The parser's name or a (possibly empty) alias name in case
                of an anonymous parser.
        anonymous: A property indicating that the parser remains anynomous
                anonymous with respect to the nodes it returns. For performance
                reasons this is implemented as an object variable rather
                than a property. This property must always be equal to
                `self.tag_name[0] == ":"`.
254
255
256
257
258
        drop_content: A property (for performance reasons implemented as
                simple field) that, if set, induces the parser not to return
                the parsed content or sub-tree if it has matched but the
                dummy `EMPTY_NODE`. In effect the parsed content will be
                dropped from the concrete syntax tree already.
259
        tag_name: The tag_name for the nodes that are created by
260
261
                the parser. If the parser is named, this is the same as
                `pname`, otherwise it is the name of the parser's type.
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
        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.

277
        _grammar:  A reference to the Grammar object to which the parser
278
279
280
                is attached.
    """

281
    def __init__(self) -> None:
282
        # assert isinstance(name, str), str(name)
283
        self.pname = ''               # type: str
284
        self.anonymous = True         # type: bool
285
        self.drop_content = False     # type: bool
286
287
        self.tag_name = self.ptype    # type: str
        self.cycle_detection = set()  # type: Set[ApplyFunc]
288
289
290
291
        try:
            self._grammar = GRAMMAR_PLACEHOLDER  # type: Grammar
        except NameError:
            pass
292
293
294
        self.reset()

    def __deepcopy__(self, memo):
295
        """        Deepcopy method of the parser. Upon instantiation of a Grammar-
296
297
298
299
300
        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.
        """
301
        duplicate = self.__class__()
302
        copy_parser_attrs(self, duplicate)
303
        return duplicate
304

305
    def __repr__(self):
di68kap's avatar
di68kap committed
306
        return self.pname + self.ptype
307
308

    def __str__(self):
di68kap's avatar
di68kap committed
309
        return self.pname + (' = ' if self.pname else '') + repr(self)
310

311
312
313
314
315
316
    @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__

317
318
319
    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
di68kap's avatar
di68kap committed
320
        return self.pname if self.pname else self.__repr__()
321

322
323
324
325
    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."""
326
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
327
        self.recursion_counter = defaultdict(int)  # type: DefaultDict[int, int]
328

eckhart's avatar
eckhart committed
329
    @cython.locals(location=cython.int, gap=cython.int, i=cython.int)
330
331
332
    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
333
334
        done in the overridden method `_parse()`. This wrapper-method can be thought of
        as a "parser guard", because it guards the parsing process.
335
        """
336
337
338
339
340
341
342
343
344
        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

345
        grammar = self._grammar
346
347
348
        location = grammar.document_length__ - len(text)

        try:
349
350
            # rollback variable changing operation if parser backtracks
            # to a position before the variable changing operation occurred
351
352
353
            if grammar.last_rb__loc__ >= location:
                grammar.rollback_to__(location)

354
            # if location has already been visited by the current parser, return saved result
355
356
            visited = self.visited  # using local variable for better performance
            if location in visited:
357
                # no history recording in case of memoized results
358
                return visited[location]
359
360

            # break left recursion at the maximum allowed depth
361
362
363
            left_recursion_depth__ = grammar.left_recursion_depth__
            if left_recursion_depth__:
                if self.recursion_counter[location] > left_recursion_depth__:
364
365
366
367
                    grammar.recursion_locations__.add(location)
                    return None, text
                self.recursion_counter[location] += 1

368
            # write current step to call stack, if history tracking is configured
369
370
            history_tracking__ = grammar.history_tracking__
            if history_tracking__:
di68kap's avatar
di68kap committed
371
                grammar.call_stack__.append(
372
                    ((self.repr if self.tag_name in (':RegExp', ':Token', ':DropToken')
373
                      else (self.pname or self.tag_name)), location))
374
                grammar.moving_forward__ = True
375
                error = None
376

377
            # finally, the actual parser call!
378
379
            try:
                node, rest = self._parse(text)
380
            except ParserError as pe:
381
                # catching up with parsing after an error occurred
382
                gap = len(text) - len(pe.rest)
di68kap's avatar
di68kap committed
383
                rules = grammar.resume_rules__.get(self.pname, [])
384
                rest = pe.rest[len(pe.node):]
385
                i = reentry_point(rest, rules, grammar.comment_rx__)
386
                if i >= 0 or self == grammar.start_parser__:
387
                    assert pe.node.children or (not pe.node.result)
388
389
                    # apply reentry-rule or catch error at root-parser
                    if i < 0:
390
                        i = 1
391
392
393
394
395
396
397
398
399
400
401
                    try:
                        zombie = pe.node[ZOMBIE_TAG]
                    except KeyError:
                        zombie = None
                    if zombie and not zombie.result:
                        zombie.result = rest[:i]
                        tail = tuple()
                    else:
                        nd = Node(ZOMBIE_TAG, rest[:i]).with_pos(location)
                        nd.attr['err'] = pe.error.message
                        tail = (nd,)
402
                    rest = rest[i:]
403
404
                    if pe.first_throw:
                        node = pe.node
405
                        node.result = node.children + tail
406
                    else:
407
408
409
                        node = Node(
                            self.tag_name,
                            (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) + tail)
410
411
                elif pe.first_throw:
                    raise ParserError(pe.node, pe.rest, pe.error, first_throw=False)
412
                elif grammar.tree__.errors[-1].code == Error.MANDATORY_CONTINUATION_AT_EOF:
413
                    node = Node(self.tag_name, pe.node).with_pos(location)  # try to create tree as faithful as possible
414
                else:
415
416
                    result = (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) if gap \
                        else pe.node  # type: ResultType
417
418
                    raise ParserError(Node(self.tag_name, result).with_pos(location),
                                      text, pe.error, first_throw=False)
419
420
                error = pe.error  # needed for history tracking

421
            if left_recursion_depth__:
422
423
424
425
426
427
                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__:
428
429
                    if location in visited:
                        node, rest = visited[location]
430
                        if location != grammar.last_recursion_location__:
431
432
433
                            grammar.tree__.add_error(
                                node, Error("Left recursion encountered. "
                                            "Refactor grammar to avoid slow parsing.",
434
435
                                            node.pos if node else location,
                                            Error.LEFT_RECURSION_WARNING))
436
                            error_id = id(node)
437
                            grammar.last_recursion_location__ = location
438
439
440
441
                    # 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
442
                    visited[location] = (None, rest)
443
            else:
444
                # assert node._pos < 0 or node == EMPTY_NODE
445
                # if node._pos != EMPTY_NODE:
446
                node._pos = location
447
448
                # assert node._pos >= 0 or node == EMPTY_NODE, \
                #     str("%i < %i" % (grammar.document_length__, location))
449
450
451
452
453
454
455
                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?
456
                    visited[location] = (node, rest)
457
458
459

            # Mind that memoized parser calls will not appear in the history record!
            # Does this make sense? Or should it be changed?
460
            if history_tracking__:
461
462
                # don't track returning parsers except in case an error has occurred
                # remaining = len(rest)
463
                if grammar.moving_forward__:
464
                    record = HistoryRecord(grammar.call_stack__, node, text,
di68kap's avatar
di68kap committed
465
                                           grammar.line_col__(text))
466
                    grammar.history__.append(record)
467
468
469
470
                elif error:
                    # error_nid = id(node)  # type: int
                    # if error_nid in grammar.tree__.error_nodes:
                    record = HistoryRecord(grammar.call_stack__, node, text,
di68kap's avatar
di68kap committed
471
                                           grammar.line_col__(text),
472
473
                                           [error])
                    grammar.history__.append(record)
474
475
476
477
                grammar.moving_forward__ = False
                grammar.call_stack__.pop()

        except RecursionError:
478
            node = Node(ZOMBIE_TAG, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
479
480
481
482
483
484
            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
485
486
487
488
489
490
491
492
493
494
495
496

    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
497

498
499
500
501
502
503
    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

504
    @property
eckhart's avatar
eckhart committed
505
    def grammar(self) -> 'Grammar':
506
507
508
509
510
        try:
            if self._grammar != GRAMMAR_PLACEHOLDER:
                return self._grammar
            else:
                raise AssertionError('Grammar has not yet been set!')
511
512
        except (AttributeError, NameError):
            raise AttributeError('Parser placeholder does not have a grammar!')
513
514
515

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
516
517
518
519
520
521
522
523
524
        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
525
526
        except NameError:  # Cython: No access to GRAMMA_PLACEHOLDER, yet :-(
            self._grammar = grammar
527

eckhart's avatar
eckhart committed
528
    def _apply(self, func: ApplyFunc, flip: FlagFunc) -> bool:
529
530
        """
        Applies function `func(parser)` recursively to this parser and all
eckhart's avatar
eckhart committed
531
532
533
534
535
536
537
538
539
        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!
540
        """
eckhart's avatar
eckhart committed
541
        if flip(func, self.cycle_detection):
542
543
544
545
546
            return False
        else:
            func(self)
            return True

eckhart's avatar
eckhart committed
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
    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)

577

578
PARSER_PLACEHOLDER = Parser()
579
580
581
582
583
584
585
586


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

587
588
def mixin_comment(whitespace: str, comment: str) -> str:
    """
589
    Returns a regular expression pattern that merges comment and whitespace
590
    regexps. Thus comments can occur wherever whitespace is allowed
591
592
593
594
595
596
    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).
    """
597
598
599
    if comment:
        return '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return whitespace
600
601


Eckhart Arnold's avatar
Eckhart Arnold committed
602
603
def mixin_noempty(whitespace: str) -> str:
    r"""
604
605
606
607
608
609
610
611
612
613
614
615
616
617
    Returns a regular expression pattern that matches only if the regular
    expression pattern `whitespace` matches AND if the match is not empty.

    If `whitespace` already matches the empty string '', then it will be
    returned unaltered.

    WARNING: `non_empty_ws` does not work regular expressions the matched
    strings of which can be followed by a symbol that can also occur at
    the start of the regular expression.

    In particular, it does not work for fixed size regular expressions,
    that ist / / or /   / or /\t/ won't work, but / */ or /\s*/ or /\s+/
    do work. There is no test for this. Fixed sizes regular expressions
    run through `non_empty_ws` will not match at any more if they are applied
Eckhart Arnold's avatar
Eckhart Arnold committed
618
619
620
621
622
    to the beginning or the middle of a sequence of whitespaces!

    In order to be safe, you whitespace regular expressions should follow
    the rule: "Whitespace cannot be followed by whitespace" or "Either
    grab it all or leave it all".
623
624
625
626
627
628
629
630
631
632

    :param whitespace: a regular expression pattern
    :return: new regular expression pattern that does not match the empty
        string '' any more.
    """
    if re.match(whitespace, ''):
        return r'(?:(?=(.|\n))' + whitespace + r'(?!\1))'
    return whitespace


633
class UnknownParserError(KeyError):
eckhart's avatar
eckhart committed
634
    """UnknownParserError is raised if a Grammar object is called with a
635
    parser that does not exist or if in the course of parsing a parser
eckhart's avatar
eckhart committed
636
    is referred to that does not exist."""
637
638


639
GrammarErrorType = Tuple[str, Parser, Error]      # TODO: replace with a named tuple?
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656


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))


657
658
659
660
661
662
663
664
665
666
667
668
669
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
670
    Example for direct instantiation of a grammar::
671

672
        >>> number = RE(r'\d+') + RE(r'\.') + RE(r'\d+') | RE(r'\d+')
673
674
675
676
677
678
679
680
681
682
683
684
685
686
        >>> 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.
687
       See classmethod `Grammar._assign_parser_names__()`
688
689

    3. The parsers in the class do not necessarily need to be connected
690
691
       to one single root parser, which is helpful for testing and when
       building up a parser gradually from several components.
692
693
694
695
696
697
698

    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.

699
    Example::
700
701
702
703
704
705
706
707
708

        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+')
709
710
711
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
712
713
714
715
            root__ = expression

    Upon instantiation the parser objects are deep-copied to the
    Grammar object and assigned to object variables of the same name.
716
717
718
719
720
    For any parser that is directly assigned to a class variable the
    field `parser.pname` contains the variable name after instantiation
    of the Grammar class. The parser will never the less remain anonymous
    with respect to the tag names of the nodes it generates, if its name
    is matched by the `anonymous__` regular expression.
721
    If one and the same parser is assigned to several class variables
722
    such as, for example, the parser `expression` in the example above,
723
    which is also assigned to `root__`, the first name sticks.
724
725
726
727
728
729
730
731
732
733
734
735

    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:

736
    Class Attributes:
737
738
739
740
        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.
741

742
743
744
        resume_rules__: A mapping of parser names to a list of regular expressions
                that act as rules to find the reentry point if a ParserError was
                thrown during the execution of the parser with the respective name.
Eckhart Arnold's avatar
Eckhart Arnold committed
745

746
747
748
749
750
        anonymous__: A regular expression to identify names of parsers that are
                assigned to class fields but shall never the less yield anonymous
                nodes (i.e. nodes the tag name of which starts with a colon ":"
                followed by the parser's class name). The default is to treat all
                parsers starting with an underscore as anonymous in addition to those
751
752
                parsers that are not directly assigned to a class field.

eckhart's avatar
eckhart committed
753
        parser_initialization__:  Before the grammar class (!) has been initialized,
754
                 which happens upon the first time it is instantiated (see
755
                 `:func:_assign_parser_names()` for an explanation), this class
756
757
                 field contains a value other than "done". A value of "done" indicates
                 that the class has already been initialized.
758

759
760
761
762
763
764
765
        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.
766

eckhart's avatar
eckhart committed
767
768
769
770
        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`).

771
    Instance Attributes:
772
773
        all_parsers__:  A set of all parsers connected to this grammar object

774
775
        comment_rx__:  The compiled regular expression for comments. If no
                comments have been defined, it defaults to RX_NEVER_MATCH
776
777
                This instance-attribute will only be defined if a class-attribute
                with the same name does not already exist!
778

eckhart's avatar
eckhart committed
779
780
781
782
        start_parser__:  During parsing, the parser with which the parsing process
                was started (see method `__call__`) or `None` if no parsing process
                is running.

783
784
785
786
787
788
789
790
791
792
        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
793
794
        document_length__:  the length of the document.

795
796
797
        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
798
                initialized if :attr:`history_tracking__` is true.
799

800
801
802
803
804
        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.

805
        _reversed__:  the same text in reverse order - needed by the `Lookbehind`-
806
807
808
                parsers.

        variables__:  A mapping for variable names to a stack of their respective
809
810
                string values - needed by the :class:`Capture`-, :class:`Retrieve`-
                and :class:`Pop`-parsers.
811
812

        rollback__:  A list of tuples (location, rollback-function) that are
813
814
815
816
                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.
817
818
819

        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
820
821
                backtracks to a location at or before last_rb__loc__ (i.e.
                location <= last_rb__loc__) then a rollback of all variable
822
823
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
824
                calling method :func:`rollback_to__(location)`.
825

di68kap's avatar
di68kap committed
826
827
828
        call_stack__:  A stack of the tag names and locations of all parsers
                in the call chain to the currently processed parser during
                parsing. The call stack can be thought of as a breadcrumb trail.
829
                This is required for recording the parser history (for debugging)
830
831
832
833
834
835
836
837
838
839
                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
840
                recording is turned off! (See :func:`Parser.__call__`)
841
842
843
844

        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
845
846
847
848
849
                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.
850
851
852
853
854
855

        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)

856
857
858
859
860
        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.
861
862
863
864
865
866
867
868

        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
869
    """
eckhart's avatar
eckhart committed
870
    python_src__ = ''  # type: str
871
    root__ = PARSER_PLACEHOLDER  # type: Parser
872
    # root__ must be overwritten with the root-parser by grammar subclass
eckhart's avatar
eckhart committed
873
    parser_initialization__ = ["pending"]  # type: List[str]
eckhart's avatar
eckhart committed
874
    resume_rules__ = dict()  # type: Dict[str, ResumeList]
875
    anonymous__ = re.compile(r'_')  # type: RxPatternType
876
    # some default values
877
878
    # COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    # WSP_RE__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
879
    static_analysis_pending__ = [True]  # type: List[bool]
880
881
882
883
884


    @classmethod
    def _assign_parser_names__(cls):
        """
di68kap's avatar
di68kap committed
885
        Initializes the `parser.pname` fields of those
886
        Parser objects that are directly assigned to a class field with
887
        the field's name, e.g.::
888

889
890
            class Grammar(Grammar):
                ...
891
                symbol = RE(r'(?!\\d)\\w+')
892

di68kap's avatar
di68kap committed
893
        After the call of this method symbol.pname == "symbol" holds.
di68kap's avatar
di68kap committed
894
895
        Parser names starting or ending with a double underscore like
        ``root__`` will be ignored. See :func:`sane_parser_name()`
896
897
898
899
900
901
902
903
904
905

        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
906
        if cls.parser_initialization__[0] != "done":
907
908
909
            cdict = cls.__dict__
            for entry, parser in cdict.items():
                if isinstance(parser, Parser) and sane_parser_name(entry):
910
                    anonymous = True if cls.anonymous__.match(entry) else False
di68kap's avatar
di68kap committed
911
                    if isinstance(parser, Forward):
di68kap's avatar
di68kap committed
912
913
                        if not cast(Forward, parser).parser.pname:
                            cast(Forward, parser).parser.pname = entry
914
                            cast(Forward, parser).parser.anonymous = anonymous
di68kap's avatar
di68kap committed
915
916
                    else:   # if not parser.pname:
                        parser.pname = entry
917
                        parser.anonymous = anonymous
eckhart's avatar
eckhart committed
918
            cls.parser_initialization__[0] = "done"
919
920
921


    def __init__(self, root: Parser = None) -> None:
922
        self.all_parsers__ = set()             # type: Set[Parser]
923
924
925
926
927
928
929
        # add compiled regular expression for comments, if it does not already exist
        if not hasattr(self, 'comment_rx__'):
            self.comment_rx__ = re.compile(self.COMMENT__) \
                if hasattr(self, 'COMMENT__') and self.COMMENT__ else RX_NEVER_MATCH
        else:
            assert ((self.COMMENT__ and self.COMMENT__ == self.comment_rx__.pattern)
                     or (not self.COMMENT__ and self.comment_rx__ == RX_NEVER_MATCH))
930
        self.start_parser__ = None             # type: Optional[Parser]
931
932
933
        self._dirty_flag__ = False             # type: bool
        self.history_tracking__ = False        # type: bool
        self.memoization__ = True              # type: bool
934
935
936
        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
937
938
939
940
941
942
943
944
945
        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
946
947
948
949
        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__']
950

951
952
953
954
955
956
957
958
959
        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
960

961
962
963
964
965

    def __getitem__(self, key):
        try:
            return self.__dict__[key]
        except KeyError:
eckhart's avatar
eckhart committed
966
            parser_template = getattr(self.__class__, key, None)
967
968
969
970
            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
971
                assert self[key] == parser
972
                return self[key]
973
            raise UnknownParserError('Unknown parser "%s" !' % key)
974

975

976
977
    def __contains__(self, key):
        return key in self.__dict__ or hasattr(self, key)
978

979

980
    def _reset__(self):
981
        self.tree__ = RootNode()              # type: RootNode
982
983
        self.document__ = EMPTY_STRING_VIEW   # type: StringView
        self._reversed__ = EMPTY_STRING_VIEW  # type: StringView
eckhart's avatar
eckhart committed
984
        self.document_length__ = 0            # type: int
985
986
        self.document_lbreaks__ = []          # type: List[int]
        # variables stored and recalled by Capture and Retrieve parsers
eckhart's avatar
eckhart committed
987
        self.variables__ = defaultdict(lambda: [])  # type: DefaultDict[str, List[str]]
988
989
990
        self.rollback__ = []                  # type: List[Tuple[int, Callable]]
        self.last_rb__loc__ = -1              # type: int
        # support for call stack tracing
991
        self.call_stack__ = []                # type: List[Tuple[str, int]]  # tag_name, location
992
993
994
995
996
        # 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]
997
        self.last_recursion_location__ = -1   # type: int
998
999
1000


    @property