parse.py 105 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
36
from typing import Callable, cast, List, Tuple, Set, Iterator, Dict, \
    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
eckhart's avatar
eckhart committed
45
from DHParser.toolkit import sane_parser_name, escape_control_characters, re, cython
46

47

48
__all__ = ('Parser',
49
           'UnknownParserError',
50
51
           'GrammarErrorType',
           'GrammarError',
52
           'Grammar',
Eckhart Arnold's avatar
Eckhart Arnold committed
53
           'EMPTY_NODE',
54
           'PreprocessorToken',
55
           'Token',
56
           'DropToken',
57
           'RegExp',
58
59
           'RE',
           'TKN',
60
           'Whitespace',
61
           'DropWhitespace',
62
           'mixin_comment',
63
           'MetaParser',
eckhart's avatar
eckhart committed
64
65
           'UnaryParser',
           'NaryParser',
66
67
68
69
70
71
72
73
74
           'Synonym',
           'Option',
           'ZeroOrMore',
           'OneOrMore',
           'Series',
           'Alternative',
           'AllOf',
           'SomeOf',
           'Unordered',
eckhart's avatar
eckhart committed
75
           'Required',
76
77
78
79
80
81
82
83
84
85
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
           'last_value',
           'counterpart',
           'accumulate',
           'Capture',
           'Retrieve',
           'Pop',
86
           'Forward')
87
88
89
90


########################################################################
#
91
# Parser base class
92
93
94
95
#
########################################################################


96
EMPTY_NODE = FrozenNode(':EMPTY__', '')
97
98


Eckhart Arnold's avatar
Eckhart Arnold committed
99
100
101
class ParserError(Exception):
    """
    A `ParserError` is thrown for those parser errors that allow the
102
    controlled re-entrance of the parsing process after the error occurred.
Eckhart Arnold's avatar
Eckhart Arnold committed
103
    If a reentry-rule has been configured for the parser where the error
104
    occurred, the parser guard can resume the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
105
106
107
108
109

    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.
    """
110
111
112
113
    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]
114
        self.first_throw = first_throw  # type: bool
Eckhart Arnold's avatar
Eckhart Arnold committed
115

116
    def __str__(self):
117
        return "%i: %s    %s" % (self.node.pos, str(self.rest[:25]), repr(self.node))
118

Eckhart Arnold's avatar
Eckhart Arnold committed
119

120
ResumeList = List[Union[str, Any]]  # list of strings or regular expressiones
Eckhart Arnold's avatar
Eckhart Arnold committed
121
122


123
def reentry_point(rest: StringView, rules: ResumeList, comment_regex) -> int:
Eckhart Arnold's avatar
Eckhart Arnold committed
124
125
    """
    Finds the point where parsing should resume after a ParserError has been caught.
126
    Makes sure that this reentry-point does not lie inside a comment.
Eckhart Arnold's avatar
Eckhart Arnold committed
127
128
129
130
131
132
    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.
133
        comment_regex: A regular expression object that matches comments.
Eckhart Arnold's avatar
Eckhart Arnold committed
134
    Returns:
135
136
        The integer index of the closest reentry point or -1 if no reentry-point
        was found.
Eckhart Arnold's avatar
Eckhart Arnold committed
137
    """
138
    upper_limit = len(rest) + 1
139
    closest_match = upper_limit
140
141
142
143
144
145
146
147
148
149
150
151
152
    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

153
154
155
156
157
158
    def str_search(s, start: int = 0) -> Tuple[int, int]:
        nonlocal rest
        return rest.find(s), len(rule)

    def rx_search(rx, start: int = 0) -> Tuple[int, int]:
        nonlocal rest
159
160
161
        m = rest.search(rx, start)
        if m:
            start, end = m.span()
162
            return rest.index(start), end - start
163
164
        return -1, 0

165
166
167
168
169
170
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()
        while a <= k < b:
            k, length = search_func(search_rule, k + length)
            while a < b <= k:
                a, b = next_comment()
        return k if k >= 0 else upper_limit

176
    # find closest match
Eckhart Arnold's avatar
Eckhart Arnold committed
177
    for rule in rules:
178
        comments = rest.finditer(comment_regex)
Eckhart Arnold's avatar
Eckhart Arnold committed
179
        if isinstance(rule, str):
180
181
182
183
            pos = entry_point(str_search, rule)
        else:  # rule is a compiled regular expression
            pos = entry_point(rx_search, rule)
        closest_match = min(pos, closest_match)
184

185
    # in case no rule matched return -1
186
187
188
    if closest_match == upper_limit:
        closest_match = -1
    return closest_match
Eckhart Arnold's avatar
Eckhart Arnold committed
189
190


eckhart's avatar
eckhart committed
191
192
193
194
ApplyFunc = Callable[['Parser'], None]
FlagFunc = Callable[[ApplyFunc, Set[ApplyFunc]], bool]


195
class Parser:
196
197
198
199
200
201
202
203
204
205
206
207
208
209
    """
    (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
210
    1. *Named parsers* for which a name is set in field `parser.pname`.
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
       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:
234
235
236
237
238
        pname:    The parser name or the empty string in case the parser
                remains anonymous.
        tag_name:  The tag_name for the nodes that are created by
                the parser. If the parser is named, this is the same as
                `pname`, otherwise it is the name of the parser's type.
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
        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.

254
        _grammar:  A reference to the Grammar object to which the parser
255
256
257
                is attached.
    """

258
    def __init__(self) -> None:
259
        # assert isinstance(name, str), str(name)
260
261
262
        self.pname = ''               # type: str
        self.tag_name = self.ptype    # type: str
        self.cycle_detection = set()  # type: Set[ApplyFunc]
263
264
265
266
        try:
            self._grammar = GRAMMAR_PLACEHOLDER  # type: Grammar
        except NameError:
            pass
267
268
269
        self.reset()

    def __deepcopy__(self, memo):
270
        """        Deepcopy method of the parser. Upon instantiation of a Grammar-
271
272
273
274
275
        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.
        """
276
        duplicate = self.__class__()
di68kap's avatar
di68kap committed
277
        duplicate.pname = self.pname
278
        duplicate.tag_name = self.tag_name
279
        return duplicate
280

281
    def __repr__(self):
di68kap's avatar
di68kap committed
282
        return self.pname + self.ptype
283
284

    def __str__(self):
di68kap's avatar
di68kap committed
285
        return self.pname + (' = ' if self.pname else '') + repr(self)
286

287
288
289
290
291
292
    @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__

293
294
295
    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
di68kap's avatar
di68kap committed
296
        return self.pname if self.pname else self.__repr__()
297

298
299
300
301
    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."""
302
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
303
        self.recursion_counter = defaultdict(int)  # type: DefaultDict[int, int]
304

eckhart's avatar
eckhart committed
305
    @cython.locals(location=cython.int, gap=cython.int, i=cython.int)
306
307
308
    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
309
310
        done in the overridden method `_parse()`. This wrapper-method can be thought of
        as a "parser guard", because it guards the parsing process.
311
        """
312
313
314
315
316
317
318
319
320
        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

321
        grammar = self._grammar
322
323
324
        location = grammar.document_length__ - len(text)

        try:
325
326
            # rollback variable changing operation if parser backtracks
            # to a position before the variable changing operation occurred
327
328
329
            if grammar.last_rb__loc__ >= location:
                grammar.rollback_to__(location)

330
            # if location has already been visited by the current parser, return saved result
331
332
            visited = self.visited  # using local variable for better performance
            if location in visited:
333
                # no history recording in case of memoized results
334
                return visited[location]
335
336

            # break left recursion at the maximum allowed depth
337
338
339
            left_recursion_depth__ = grammar.left_recursion_depth__
            if left_recursion_depth__:
                if self.recursion_counter[location] > left_recursion_depth__:
340
341
342
343
                    grammar.recursion_locations__.add(location)
                    return None, text
                self.recursion_counter[location] += 1

344
            # write current step to call stack, if history tracking is configured
345
346
            history_tracking__ = grammar.history_tracking__
            if history_tracking__:
di68kap's avatar
di68kap committed
347
                grammar.call_stack__.append(
348
349
                    ((self.repr if self.tag_name in (':RegExp', ':Token', ':DropToken')
                      else self.tag_name), location))
350
                grammar.moving_forward__ = True
351
                error = None
352

353
            # finally, the actual parser call!
354
355
            try:
                node, rest = self._parse(text)
356
            except ParserError as pe:
357
                # catching up with parsing after an error occurred
358
                gap = len(text) - len(pe.rest)
di68kap's avatar
di68kap committed
359
                rules = grammar.resume_rules__.get(self.pname, [])
360
                rest = pe.rest[len(pe.node):]
361
                i = reentry_point(rest, rules, grammar.comment_rx__)
362
363
364
                if i >= 0 or self == grammar.start_parser__:
                    # apply reentry-rule or catch error at root-parser
                    if i < 0:
365
                        i = 1
366
                    nd = Node(ZOMBIE_TAG, rest[:i]).with_pos(location)
367
                    nd.attr['err'] = pe.error.message
368
                    rest = rest[i:]
369
370
371
                    assert pe.node.children or (not pe.node.result)
                    if pe.first_throw:
                        node = pe.node
eckhart's avatar
eckhart committed
372
                        node.result = node.children + (nd,)
373
                    else:
374
                        node = Node(self.tag_name,
375
                                    (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node, nd))
376
377
                elif pe.first_throw:
                    raise ParserError(pe.node, pe.rest, pe.error, first_throw=False)
378
                elif grammar.tree__.errors[-1].code == Error.MANDATORY_CONTINUATION_AT_EOF:
379
                    node = pe.node
380
                else:
381
382
                    result = (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) if gap \
                        else pe.node  # type: ResultType
383
384
                    raise ParserError(Node(self.tag_name, result).with_pos(location),
                                      text, pe.error, first_throw=False)
385
386
                error = pe.error  # needed for history tracking

387

388
            if left_recursion_depth__:
389
390
391
392
393
394
                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__:
395
396
                    if location in visited:
                        node, rest = visited[location]
397
                        if location != grammar.last_recursion_location__:
398
399
400
                            grammar.tree__.add_error(
                                node, Error("Left recursion encountered. "
                                            "Refactor grammar to avoid slow parsing.",
401
402
                                            node.pos if node else location,
                                            Error.LEFT_RECURSION_WARNING))
403
                            error_id = id(node)
404
                            grammar.last_recursion_location__ = location
405
406
407
408
                    # 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
409
                    visited[location] = (None, rest)
410
            else:
411
                # assert node._pos < 0 or node == EMPTY_NODE
412
                # if node._pos != EMPTY_NODE:
413
                node._pos = location
414
415
                # assert node._pos >= 0 or node == EMPTY_NODE, \
                #     str("%i < %i" % (grammar.document_length__, location))
416
417
418
419
420
421
422
                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?
423
                    visited[location] = (node, rest)
424
425
426

            # Mind that memoized parser calls will not appear in the history record!
            # Does this make sense? Or should it be changed?
427
            if history_tracking__:
428
429
                # don't track returning parsers except in case an error has occurred
                # remaining = len(rest)
430
                if grammar.moving_forward__:
431
                    record = HistoryRecord(grammar.call_stack__, node, text,
di68kap's avatar
di68kap committed
432
                                           grammar.line_col__(text))
433
                    grammar.history__.append(record)
434
435
436
437
                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
438
                                           grammar.line_col__(text),
439
440
                                           [error])
                    grammar.history__.append(record)
441
442
443
444
                grammar.moving_forward__ = False
                grammar.call_stack__.pop()

        except RecursionError:
445
            node = Node(ZOMBIE_TAG, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
446
447
448
449
450
451
            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
452
453
454
455
456
457
458
459
460
461
462
463

    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
464

465
466
467
468
469
470
    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

471
    @property
eckhart's avatar
eckhart committed
472
    def grammar(self) -> 'Grammar':
473
474
475
476
477
        try:
            if self._grammar != GRAMMAR_PLACEHOLDER:
                return self._grammar
            else:
                raise AssertionError('Grammar has not yet been set!')
478
479
        except (AttributeError, NameError):
            raise AttributeError('Parser placeholder does not have a grammar!')
480
481
482

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
483
484
485
486
487
488
489
490
491
        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
492
493
        except NameError:  # Cython: No access to GRAMMA_PLACEHOLDER, yet :-(
            self._grammar = grammar
494

eckhart's avatar
eckhart committed
495
    def _apply(self, func: ApplyFunc, flip: FlagFunc) -> bool:
496
497
        """
        Applies function `func(parser)` recursively to this parser and all
eckhart's avatar
eckhart committed
498
499
500
501
502
503
504
505
506
        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!
507
        """
eckhart's avatar
eckhart committed
508
        if flip(func, self.cycle_detection):
509
510
511
512
513
            return False
        else:
            func(self)
            return True

eckhart's avatar
eckhart committed
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
    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)

544

545
PARSER_PLACEHOLDER = Parser()
546
547
548
549
550
551
552
553


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

554
555
RX_NEVER_MATCH = re.compile(r'..(?<=^)')

556

557
558
559
560
561
562
563
564
565
566
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).
    """
567
568
569
    if comment:
        return '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return whitespace
570
571


572
class UnknownParserError(KeyError):
eckhart's avatar
eckhart committed
573
    """UnknownParserError is raised if a Grammar object is called with a
574
    parser that does not exist or if in the course of parsing a parser
eckhart's avatar
eckhart committed
575
    is referred to that does not exist."""
576
577


578
GrammarErrorType = Tuple[str, Parser, Error]      # TODO: replace with a named tuple?
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595


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))


596
597
598
599
600
601
602
603
604
605
606
607
608
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
609
    Example for direct instantiation of a grammar::
610

611
        >>> number = RE(r'\d+') + RE(r'\.') + RE(r'\d+') | RE(r'\d+')
612
613
614
615
616
617
618
619
620
621
622
623
624
625
        >>> 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.
626
       See classmethod `Grammar._assign_parser_names__()`
627
628

    3. The parsers in the class do not necessarily need to be connected
629
630
       to one single root parser, which is helpful for testing and when
       building up a parser gradually from several components.
631
632
633
634
635
636
637

    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.

638
    Example::
639
640
641
642
643
644
645
646
647

        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+')
648
649
650
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
651
652
653
654
655
            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
656
    'named' parser and its field `parser.pname` contains the variable
657
658
    name after instantiation of the Grammar class. All other parsers,
    i.e. parsers that are defined within a `named` parser, remain
659
    "anonymous parsers" where `parser.pname` is the empty string.
660
    If one and the same parser is assigned to several class variables
661
    such as, for example, the parser `expression` in the example above,
662
663
664
665
666
667
668
669
670
671
672
673
674
    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:

675
    Class Attributes:
676
677
678
679
        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.
680

Eckhart Arnold's avatar
Eckhart Arnold committed
681
        resume_rules__: A mapping of parser names to a list of regular expressions or search
682
                strings that act as rules to find the the reentry point if a ParserError
Eckhart Arnold's avatar
Eckhart Arnold committed
683
684
                was thrown during the execution of the parser with the respective name.

685
686
687
688
689
        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.
690

691
692
693
694
695
696
697
        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.
698

eckhart's avatar
eckhart committed
699
700
701
702
        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`).

703
    Instance Attributes:
704
705
        all_parsers__:  A set of all parsers connected to this grammar object

706
707
708
        comment_rx__:  The compiled regular expression for comments. If no
                comments have been defined, it defaults to RX_NEVER_MATCH

eckhart's avatar
eckhart committed
709
710
711
712
        start_parser__:  During parsing, the parser with which the parsing process
                was started (see method `__call__`) or `None` if no parsing process
                is running.

713
714
715
716
717
718
719
720
721
722
        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
723
724
        document_length__:  the length of the document.

725
726
727
        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
728
                initialized if :attr:`history_tracking__` is true.
729

730
731
732
733
734
        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.

735
        _reversed__:  the same text in reverse order - needed by the `Lookbehind`-
736
737
738
                parsers.

        variables__:  A mapping for variable names to a stack of their respective
739
740
                string values - needed by the :class:`Capture`-, :class:`Retrieve`-
                and :class:`Pop`-parsers.
741
742

        rollback__:  A list of tuples (location, rollback-function) that are
743
744
745
746
                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.
747
748
749

        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
750
751
                backtracks to a location at or before last_rb__loc__ (i.e.
                location <= last_rb__loc__) then a rollback of all variable
752
753
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
754
                calling method :func:`rollback_to__(location)`.
755

di68kap's avatar
di68kap committed
756
757
758
        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.
759
                This is required for recording the parser history (for debugging)
760
761
762
763
764
765
766
767
768
769
                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
770
                recording is turned off! (See :func:`Parser.__call__`)
771
772
773
774

        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
775
776
777
778
779
                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.
780
781
782
783
784
785

        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)

786
787
788
789
790
        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.
791
792
793
794
795
796
797
798

        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
799
    """
eckhart's avatar
eckhart committed
800
    python_src__ = ''  # type: str
801
    root__ = PARSER_PLACEHOLDER  # type: Parser
802
    # root__ must be overwritten with the root-parser by grammar subclass
eckhart's avatar
eckhart committed
803
    parser_initialization__ = ["pending"]  # type: List[str]
eckhart's avatar
eckhart committed
804
    resume_rules__ = dict()  # type: Dict[str, ResumeList]
805
    # some default values
806
807
    # COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    # WSP_RE__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
808
    static_analysis_pending__ = [True]  # type: List[bool]
809
810
811
812
813


    @classmethod
    def _assign_parser_names__(cls):
        """
di68kap's avatar
di68kap committed
814
        Initializes the `parser.pname` fields of those
815
        Parser objects that are directly assigned to a class field with
816
        the field's name, e.g.::
817

818
819
            class Grammar(Grammar):
                ...
820
                symbol = RE(r'(?!\\d)\\w+')
821

di68kap's avatar
di68kap committed
822
        After the call of this method symbol.pname == "symbol" holds.
di68kap's avatar
di68kap committed
823
824
        Parser names starting or ending with a double underscore like
        ``root__`` will be ignored. See :func:`sane_parser_name()`
825
826
827
828
829
830
831
832
833
834

        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
835
        if cls.parser_initialization__[0] != "done":
836
837
838
            cdict = cls.__dict__
            for entry, parser in cdict.items():
                if isinstance(parser, Parser) and sane_parser_name(entry):
di68kap's avatar
di68kap committed
839
                    if isinstance(parser, Forward):
di68kap's avatar
di68kap committed
840
841
842
843
                        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
844
            cls.parser_initialization__[0] = "done"
845
846
847


    def __init__(self, root: Parser = None) -> None:
848
        self.all_parsers__ = set()             # type: Set[Parser]
849
850
        self.comment_rx__ = re.compile(self.COMMENT__) \
            if hasattr(self, 'COMMENT__') and self.COMMENT__ else RX_NEVER_MATCH
851
        self.start_parser__ = None             # type: Optional[Parser]
852
853
854
        self._dirty_flag__ = False             # type: bool
        self.history_tracking__ = False        # type: bool
        self.memoization__ = True              # type: bool
855
856
857
        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
858
859
860
861
862
863
864
865
866
        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
867
868
869
870
        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__']
871

872
873
874
875
876
877
878
879
880
        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
881

882
883
884
885
886

    def __getitem__(self, key):
        try:
            return self.__dict__[key]
        except KeyError:
eckhart's avatar
eckhart committed
887
            parser_template = getattr(self.__class__, key, None)
888
889
890
891
            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
892
                assert self[key] == parser
893
                return self[key]
894
            raise UnknownParserError('Unknown parser "%s" !' % key)
895

896

897
898
    def __contains__(self, key):
        return key in self.__dict__ or hasattr(self, key)
899

900

901
    def _reset__(self):
902
        self.tree__ = RootNode()              # type: RootNode
903
904
        self.document__ = EMPTY_STRING_VIEW   # type: StringView
        self._reversed__ = EMPTY_STRING_VIEW  # type: StringView
eckhart's avatar
eckhart committed
905
        self.document_length__ = 0            # type: int
906
907
        self.document_lbreaks__ = []          # type: List[int]
        # variables stored and recalled by Capture and Retrieve parsers
eckhart's avatar
eckhart committed
908
        self.variables__ = defaultdict(lambda: [])  # type: DefaultDict[str, List[str]]
909
910
911
        self.rollback__ = []                  # type: List[Tuple[int, Callable]]
        self.last_rb__loc__ = -1              # type: int
        # support for call stack tracing
912
        self.call_stack__ = []                # type: List[Tuple[str, int]]  # tag_name, location
913
914
915
916
917
        # 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]
918
        self.last_recursion_location__ = -1   # type: int
919
920
921
922
923
924
925
926
927
928


    @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__:
929
            self._reversed__ = StringView(self.document__.get_text()[::-1])
930
931
932
933
934
935
936
937
        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
938
        if parser.pname:
939
            # prevent overwriting instance variables or parsers of a different class
di68kap's avatar
di68kap committed
940
941
            assert parser.pname not in self.__dict__ or \
                   isinstance(self.__dict__[parser.pname], parser.__class__), \
942
                ('Cannot add parser "%s" because a field with the same name '
di68kap's avatar
di68kap committed
943
                 'already exists in grammar object: %s!'
di68kap's avatar
di68kap committed
944
945
946
                 % (parser.pname, str(self.__dict__[parser.pname])))
            setattr(self, parser.pname, parser)
        parser.tag_name = parser.pname or parser.ptype
947
948
949
950
        self.all_parsers__.add(parser)
        parser.grammar = self


eckhart's avatar
eckhart committed
951
952
    def __call__(self,
                 document: str,
eckhart's avatar
eckhart committed
953
                 start_parser: Union[str, Parser] = "root_parser__",
eckhart's avatar
eckhart committed
954
                 track_history: bool = False) -> RootNode:
955
956
957
958
959
        """
        Parses a document with with parser-combinators.

        Args:
            document (str): The source text to be parsed.
di68kap's avatar
di68kap committed
960
961
            start_parser (str or Parser): The name of the parser with which
                to start. This is useful for testing particular parsers
962
                (i.e. particular parts of the EBNF-Grammar.)
di68kap's avatar
di68kap committed
963
964
965
966
967
            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.
968
        Returns:
di68kap's avatar
di68kap committed
969
            Node: The root node to the parse tree.
970
        """
971

eckhart's avatar
eckhart committed
972
        def tail_pos(predecessors: Union[List[Node], Tuple[Node, ...], None]) -> int:
eckhart's avatar
eckhart committed
973
974
            """Adds the position after the last node in the list of
            predecessors to the node."""
eckhart's avatar
eckhart committed
975
            return predecessors[-1].pos + len(predecessors[-1]) if predecessors else 0
eckhart's avatar
eckhart committed
976

977
978
979
980
981
982
        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
983
            that the testing framework knows that the failure is only a
984
985
986
            test-case-artifact and no real failure.
            (See test/test_testing.TestLookahead !)
            """
987
988
989
            def is_lookahead(tag_name: str) -> bool:
                return (tag_name in self and isinstance(self[tag_name], Lookahead)
                        or tag_name[0] == ':' and issubclass(eval(tag_name[1:]), Lookahead))
di68kap's avatar
di68kap committed
990
991
992
993
            # for h in reversed(self.history__[:-1]):
            #     for tn, pos in h.call_stack:
            #         if is_lookahead(tn) and h.status == HistoryRecord.MATCH:
            #             print(h.call_stack, pos, h.line_col)
994
995
            last_record = self.history__[-2] if len(self.history__) > 1 else None  # type: Optional[HistoryRecord]
            return last_record and parser != self.root_parser__ \
di68kap's avatar
di68kap committed
996
                    and any(h.status == HistoryRecord.MATCH
di68kap's avatar
di68kap committed
997
998
                            and any(is_lookahead(tn) and location >= len(self.document__)
                                    for tn, location in h.call_stack)
di68kap's avatar
di68kap committed
999
                            for h in self.history__[:-1])
1000