parse.py 91.2 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 DHParser.error import Error, linebreaks, line_col
37
38
from DHParser.log import is_logging, HistoryRecord
from DHParser.preprocess import BEGIN_TOKEN, END_TOKEN, RX_TOKEN_NAME
39
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
40
41
from DHParser.syntaxtree import Node, FrozenNode, RootNode, WHITESPACE_PTYPE, \
    TOKEN_PTYPE, ZOMBIE_TAG, ResultType
eckhart's avatar
eckhart committed
42
from DHParser.toolkit import sane_parser_name, escape_control_characters, re, typing, cython
Eckhart Arnold's avatar
Eckhart Arnold committed
43
from typing import Callable, cast, List, Tuple, Set, Dict, DefaultDict, Union, Optional, Any
44

45

46
__all__ = ('Parser',
47
           'UnknownParserError',
48
           'Grammar',
Eckhart Arnold's avatar
Eckhart Arnold committed
49
           'EMPTY_NODE',
50
           'PreprocessorToken',
51
           'Token',
52
           'DropToken',
53
           'RegExp',
54
55
           'RE',
           'TKN',
56
           'Whitespace',
57
           'DropWhitespace',
58
           'mixin_comment',
59
           'MetaParser',
eckhart's avatar
eckhart committed
60
61
           'UnaryParser',
           'NaryParser',
62
63
64
65
66
67
68
69
70
           'Synonym',
           'Option',
           'ZeroOrMore',
           'OneOrMore',
           'Series',
           'Alternative',
           'AllOf',
           'SomeOf',
           'Unordered',
eckhart's avatar
eckhart committed
71
           'Required',
72
73
74
75
76
77
78
79
80
81
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
           'last_value',
           'counterpart',
           'accumulate',
           'Capture',
           'Retrieve',
           'Pop',
82
           'Forward')
83
84


85
86
87
88

EMPTY_NODE = FrozenNode(':EMPTY__', '')


89
90
########################################################################
#
91
# Parser base class
92
93
94
95
96
97
98
99
100
101
102
#
########################################################################


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


Eckhart Arnold's avatar
Eckhart Arnold committed
103
104
105
class ParserError(Exception):
    """
    A `ParserError` is thrown for those parser errors that allow the
106
    controlled re-entrance of the parsing process after the error occurred.
Eckhart Arnold's avatar
Eckhart Arnold committed
107
    If a reentry-rule has been configured for the parser where the error
108
    occurred, the parser guard can resume the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
109
110
111
112
113

    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.
    """
114
    def __init__(self, node: Node, rest: StringView, first_throw: bool):
Eckhart Arnold's avatar
Eckhart Arnold committed
115
116
        self.node = node  # type: Node
        self.rest = rest  # type: StringView
117
        self.first_throw = first_throw  # type: bool
Eckhart Arnold's avatar
Eckhart Arnold committed
118
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) -> int:
Eckhart Arnold's avatar
Eckhart Arnold committed
124
125
126
127
128
129
130
131
132
    """
    Finds the point where parsing should resume after a ParserError has been caught.
    Args:
        rest:  The rest of the parsed text or, in other words, the point where
                a ParserError was thrown.
        rules: A list of strings or regular expressions. The rest of the text is
                searched for each of these. The closest match is the point where
                parsing will be resumed.
    Returns:
133
134
        The integer index of the closest reentry point or -1 if no reentry-point
        was found.
Eckhart Arnold's avatar
Eckhart Arnold committed
135
    """
136
    upper_limit = len(rest) + 1
Eckhart Arnold's avatar
Eckhart Arnold committed
137
138
139
140
    i = upper_limit
    #find closest match
    for rule in rules:
        if isinstance(rule, str):
141
142
            k = rest.find(rule)
            i = min(k if k >= 0 else upper_limit, i)
Eckhart Arnold's avatar
Eckhart Arnold committed
143
144
145
        else:
            m = rest.search(rule)
            if m:
146
                i = min(rest.index(m.start()), i)
147
    # in case no rule matched return -1
148
    if i == upper_limit:
149
150
        i = -1
    return i
151
    # return Node(ZOMBIE_TAG, rest[:i]), rest[i:]
Eckhart Arnold's avatar
Eckhart Arnold committed
152
153


eckhart's avatar
eckhart committed
154
155
156
157
ApplyFunc = Callable[['Parser'], None]
FlagFunc = Callable[[ApplyFunc, Set[ApplyFunc]], bool]


158
class Parser:
159
160
161
162
163
164
165
166
167
168
169
170
171
172
    """
    (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
173
    1. *Named parsers* for which a name is set in field `parser.pname`.
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
       The results produced by these parsers can later be retrieved in
       the AST by the parser name.

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

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

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

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

    Attributes and Properties:
        visited:  Mapping of places this parser has already been to
                during the current parsing process onto the results the
                parser returned at the respective place. This dictionary
                is used to implement memoizing.

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

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

212
        _grammar:  A reference to the Grammar object to which the parser
213
214
215
                is attached.
    """

216
    def __init__(self) -> None:
217
        # assert isinstance(name, str), str(name)
di68kap's avatar
di68kap committed
218
        self.pname = ''  # type: str
219
        self.tag_name = self.ptype  # type: str
220
221
222
223
        try:
            self._grammar = GRAMMAR_PLACEHOLDER  # type: Grammar
        except NameError:
            pass
224
225
226
        self.reset()

    def __deepcopy__(self, memo):
227
        """        Deepcopy method of the parser. Upon instantiation of a Grammar-
228
229
230
231
232
        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.
        """
233
        duplicate = self.__class__()
di68kap's avatar
di68kap committed
234
        duplicate.pname = self.pname
235
        duplicate.tag_name = self.tag_name
236
        return duplicate
237

238
    def __repr__(self):
di68kap's avatar
di68kap committed
239
        return self.pname + self.ptype
240
241

    def __str__(self):
di68kap's avatar
di68kap committed
242
        return self.pname + (' = ' if self.pname else '') + repr(self)
243

244
245
246
247
248
249
    @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__

250
251
252
    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
di68kap's avatar
di68kap committed
253
        return self.pname if self.pname else self.__repr__()
254

255
256
257
258
    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."""
259
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
260
        self.recursion_counter = defaultdict(int)  # type: DefaultDict[int, int]
eckhart's avatar
eckhart committed
261
        self.cycle_detection = set()  # type: Set[ApplyFunc]
262

eckhart's avatar
eckhart committed
263
    @cython.locals(location=cython.int, gap=cython.int, i=cython.int)
264
265
266
267
268
    def __call__(self: 'Parser', text: StringView) -> Tuple[Optional[Node], StringView]:
        """Applies the parser to the given text. This is a wrapper method that adds
        the business intelligence that is common to all parsers. The actual parsing is
        done in the overridden method `_parse()`.
        """
269
        grammar = self._grammar
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
        location = grammar.document_length__ - len(text)

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

            # if location has already been visited by the current parser,
            # return saved result
            if location in self.visited:
                # no history recording in case of memoized results
                return self.visited[location]

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

            if grammar.history_tracking__:
290
291
                grammar.call_stack__.append(self.repr if self.tag_name in (':RegExp', ':Token')
                                            else self.tag_name)
292
293
294
295
296
297
298
299
                grammar.moving_forward__ = True

            try:
                # PARSER CALL: run _parse() method
                node, rest = self._parse(text)
            except ParserError as error:
                # does this play well with variable setting? add rollback clause here? tests needed...
                gap = len(text) - len(error.rest)
di68kap's avatar
di68kap committed
300
                rules = grammar.resume_rules__.get(self.pname, [])
301
302
303
304
305
306
                rest = error.rest[len(error.node):]
                i = reentry_point(rest, rules)
                if i >= 0 or self == grammar.start_parser__:
                    # apply reentry-rule or catch error at root-parser
                    if i < 0:
                        i = 1
307
                    nd = Node(ZOMBIE_TAG, rest[:i])
308
                    rest = rest[i:]
eckhart's avatar
eckhart committed
309
                    assert error.node.children or (not error.node.result)
310
311
                    if error.first_throw:
                        node = error.node
eckhart's avatar
eckhart committed
312
                        node.result = node.children + (nd,)
313
                    else:
di68kap's avatar
di68kap committed
314
315
                        # TODO: ggf. Fehlermeldung, die sagt, wo es weitergeht anfügen
                        #       dürfte allerdings erst an den nächsten(!) Knoten angehängt werden (wie?)
316
                        node = Node(self.tag_name, (Node(ZOMBIE_TAG, text[:gap]), error.node, nd))
317
318
319
                elif error.first_throw:
                    raise ParserError(error.node, error.rest, first_throw=False)
                else:
320
                    result = (Node(ZOMBIE_TAG, text[:gap]), error.node) if gap else error.node  # type: ResultType
321
                    raise ParserError(Node(self.tag_name, result), text, first_throw=False)
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338

            if grammar.left_recursion_handling__:
                self.recursion_counter[location] -= 1
                # don't clear recursion_locations__ !!!

            if node is None:
                # retrieve an earlier match result (from left recursion) if it exists
                if location in grammar.recursion_locations__:
                    if location in self.visited:
                        node, rest = self.visited[location]
                        # TODO: maybe add a warning about occurrence of left-recursion here?
                    # don't overwrite any positive match (i.e. node not None) in the cache
                    # and don't add empty entries for parsers returning from left recursive calls!
                elif grammar.memoization__:
                    # otherwise also cache None-results
                    self.visited[location] = (None, rest)
            else:
339
                # assert node._pos < 0 or node == EMPTY_NODE
340
                node._pos = location
341
342
                # assert node._pos >= 0 or node == EMPTY_NODE, \
                #     str("%i < %i" % (grammar.document_length__, location))
343
344
345
346
347
348
349
350
351
352
353
354
355
356
                if (grammar.last_rb__loc__ < location
                        and (grammar.memoization__ or location in grammar.recursion_locations__)):
                    # - variable manipulating parsers will not be entered into the cache,
                    #   because caching would interfere with changes of variable state
                    # - in case of left recursion, the first recursive step that
                    #   matches will store its result in the cache
                    # TODO: need a unit-test concerning interference of variable manipulation and left recursion algorithm?
                    self.visited[location] = (node, rest)

            # Mind that memoized parser calls will not appear in the history record!
            # Does this make sense? Or should it be changed?
            if grammar.history_tracking__:
                # don't track returning parsers except in case an error has occurred
                # remaining = len(rest)
357
                if grammar.moving_forward__:
eckhart's avatar
eckhart committed
358
                    record = HistoryRecord(grammar.call_stack__, node or EMPTY_NODE, text,
359
                                           grammar.line_col__(text))
360
                    grammar.history__.append(record)
361
362
363
364
365
366
367
                elif node:
                    nid = id(node)  # type: int
                    if nid in grammar.tree__.error_nodes:
                        record = HistoryRecord(grammar.call_stack__, node or EMPTY_NODE, text,
                                               grammar.line_col__(text),
                                               grammar.tree__.error_nodes[nid])
                        grammar.history__.append(record)
368
369
370
371
                grammar.moving_forward__ = False
                grammar.call_stack__.pop()

        except RecursionError:
372
            node = Node(ZOMBIE_TAG, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
373
374
375
376
377
378
            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
379
380
381
382
383
384
385
386
387
388
389
390

    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
391

392
393
394
395
396
397
    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

398
    @property
eckhart's avatar
eckhart committed
399
    def grammar(self) -> 'Grammar':
400
401
402
403
404
        try:
            if self._grammar != GRAMMAR_PLACEHOLDER:
                return self._grammar
            else:
                raise AssertionError('Grammar has not yet been set!')
405
406
        except (AttributeError, NameError):
            raise AttributeError('Parser placeholder does not have a grammar!')
407
408
409

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
410
411
412
413
414
415
416
417
418
        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
419
420
        except NameError:  # Cython: No access to GRAMMA_PLACEHOLDER, yet :-(
            self._grammar = grammar
421
422
423
424
425

    # def _grammar_assigned_notifier(self):
    #     """A function that notifies the parser object that it has been
    #     assigned to a grammar."""
    #     pass
426

eckhart's avatar
eckhart committed
427
    def _apply(self, func: ApplyFunc, flip: FlagFunc) -> bool:
428
429
        """
        Applies function `func(parser)` recursively to this parser and all
eckhart's avatar
eckhart committed
430
431
432
433
434
435
436
437
438
        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!
439
        """
eckhart's avatar
eckhart committed
440
        if flip(func, self.cycle_detection):
441
442
443
444
445
            return False
        else:
            func(self)
            return True

eckhart's avatar
eckhart committed
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
    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)

476

477
PARSER_PLACEHOLDER = Parser()
478
479
480
481
482
483
484
485
486


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


487
488
489
490
491
492
493
494
495
496
497
498
499
500
def mixin_comment(whitespace: str, comment: str) -> str:
    """
    Returns a regular expression that merges comment and whitespace
    regexps. Thus comments cann occur whereever whitespace is allowed
    and will be skipped just as implicit whitespace.

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


501
class UnknownParserError(KeyError):
eckhart's avatar
eckhart committed
502
    """UnknownParserError is raised if a Grammar object is called with a
503
    parser that does not exist or if in the course of parsing a parser
eckhart's avatar
eckhart committed
504
    is referred to that does not exist."""
505
506


507
508
509
510
511
512
513
514
515
516
517
518
519
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
520
    Example for direct instantiation of a grammar::
521

522
        >>> number = RE('\d+') + RE('\.') + RE('\d+') | RE('\d+')
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
        >>> number_parser = Grammar(number)
        >>> number_parser("3.1416").content
        '3.1416'

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

    1. It keeps the namespace clean.

    2. The parser names of named parsers do not need to be passed to the
       constructor of the Parser object explicitly, but it suffices to
       assign them to class variables, which results in better
       readability of the Python code.

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

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

548
    Example::
549
550
551
552
553
554
555
556
557

        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+')
558
559
560
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
561
562
563
564
565
            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
566
    'named' parser and its field `parser.pname` contains the variable
567
568
    name after instantiation of the Grammar class. All other parsers,
    i.e. parsers that are defined within a `named` parser, remain
di68kap's avatar
di68kap committed
569
    "anonymous parsers" where `parser.pname` is the empty string, unless
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
    a name has been passed explicitly upon instantiation.
    If one and the same parser is assigned to several class variables
    such as, for example the parser `expression` in the example above,
    the first name sticks.

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

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

586
587
588
589
590
    Attributes:
        root__:  The root parser of the grammar. Theoretically, all parsers of the
                 grammar should be reachable by the root parser. However, for testing
                 of yet incomplete grammars class Grammar does not assume that this
                 is the case.
591

Eckhart Arnold's avatar
Eckhart Arnold committed
592
        resume_rules__: A mapping of parser names to a list of regular expressions or search
593
                strings that act as rules to find the the reentry point if a ParserError
Eckhart Arnold's avatar
Eckhart Arnold committed
594
595
                was thrown during the execution of the parser with the respective name.

596
597
598
599
600
        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.
601

eckhart's avatar
eckhart committed
602
603
604
605
        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`).

606
607
608
    Attributes:
        all_parsers__:  A set of all parsers connected to this grammar object

eckhart's avatar
eckhart committed
609
610
611
612
        start_parser__:  During parsing, the parser with which the parsing process
                was started (see method `__call__`) or `None` if no parsing process
                is running.

613
614
615
616
617
618
619
620
621
622
        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
623
624
        document_length__:  the length of the document.

625
626
627
        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
628
                initialized if :attr:`history_tracking__` is true.
629

630
631
632
633
634
        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.

635
        _reversed__:  the same text in reverse order - needed by the `Lookbehind`-
636
637
638
                parsers.

        variables__:  A mapping for variable names to a stack of their respective
639
640
                string values - needed by the :class:`Capture`-, :class:`Retrieve`-
                and :class:`Pop`-parsers.
641
642

        rollback__:  A list of tuples (location, rollback-function) that are
643
644
645
646
                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.
647
648
649

        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
650
651
                backtracks to a location at or before last_rb__loc__ (i.e.
                location <= last_rb__loc__) then a rollback of all variable
652
653
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
654
                calling method :func:`rollback_to__(location)`.
655

656
657
        call_stack__:  A stack of the tag names of all parsers that have been called.
                This is required for recording the parser history (for debugging)
658
659
660
661
662
663
664
665
666
667
                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
668
669
                recording is turned off! (See :func:`add_parser_guard` and its local
                function :func:`guarded_call`)
670
671
672
673

        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
674
675
                if full memoization is enabled. (See :func:`add_parser_guard` and its
                local function :func:`guarded_call`)
676
677
678
679
680
681
682
683
684
685

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

        left_recursion_handling__:  Turns left recursion handling on or off.
                If turned off, a recursion error will result in case of left
                recursion.
    """
eckhart's avatar
eckhart committed
686
    python_src__ = ''  # type: str
687
    root__ = PARSER_PLACEHOLDER  # type: Parser
688
    # root__ must be overwritten with the root-parser by grammar subclass
eckhart's avatar
eckhart committed
689
    parser_initialization__ = ["pending"]  # type: List[str]
eckhart's avatar
eckhart committed
690
    resume_rules__ = dict()  # type: Dict[str, ResumeList]
691
    # some default values
692
693
    # COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    # WSP_RE__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
694
695
696
697
698


    @classmethod
    def _assign_parser_names__(cls):
        """
di68kap's avatar
di68kap committed
699
        Initializes the `parser.pname` fields of those
700
        Parser objects that are directly assigned to a class field with
701
        the field's name, e.g.::
702

703
704
            class Grammar(Grammar):
                ...
705
                symbol = RE('(?!\\d)\\w+')
706

di68kap's avatar
di68kap committed
707
        After the call of this method symbol.pname == "symbol" holds.
di68kap's avatar
di68kap committed
708
709
        Parser names starting or ending with a double underscore like
        ``root__`` will be ignored. See :func:`sane_parser_name()`
710
711
712
713
714
715
716
717
718
719

        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
720
        if cls.parser_initialization__[0] != "done":
721
722
723
            cdict = cls.__dict__
            for entry, parser in cdict.items():
                if isinstance(parser, Parser) and sane_parser_name(entry):
di68kap's avatar
di68kap committed
724
                    if isinstance(parser, Forward):
di68kap's avatar
di68kap committed
725
726
727
728
                        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
729
            cls.parser_initialization__[0] = "done"
730
731
732


    def __init__(self, root: Parser = None) -> None:
733
734
        self.all_parsers__ = set()             # type: Set[Parser]
        self.start_parser__ = None             # type: Optional[Parser]
735
736
737
738
739
740
741
742
743
744
745
746
747
        self._dirty_flag__ = False             # type: bool
        self.history_tracking__ = False        # type: bool
        self.memoization__ = True              # type: bool
        self.left_recursion_handling__ = True  # type: bool
        self._reset__()

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

        # then deep-copy the parser tree from class to instance;
        # parsers not connected to the root object will be copied later
        # on demand (see Grammar.__getitem__()). Usually, the need to
        # do so only arises during testing.
eckhart's avatar
eckhart committed
748
749
750
751
        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__']
752
753
754
755
756
757


    def __getitem__(self, key):
        try:
            return self.__dict__[key]
        except KeyError:
eckhart's avatar
eckhart committed
758
            parser_template = getattr(self.__class__, key, None)
759
760
761
762
            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
763
                assert self[key] == parser
764
                return self[key]
765
            raise UnknownParserError('Unknown parser "%s" !' % key)
766

767
768
    def __contains__(self, key):
        return key in self.__dict__ or hasattr(self, key)
769
770

    def _reset__(self):
771
        self.tree__ = RootNode()              # type: RootNode
772
773
        self.document__ = EMPTY_STRING_VIEW   # type: StringView
        self._reversed__ = EMPTY_STRING_VIEW  # type: StringView
eckhart's avatar
eckhart committed
774
        self.document_length__ = 0            # type: int
775
776
        self.document_lbreaks__ = []          # type: List[int]
        # variables stored and recalled by Capture and Retrieve parsers
eckhart's avatar
eckhart committed
777
        self.variables__ = defaultdict(lambda: [])  # type: DefaultDict[str, List[str]]
778
779
780
        self.rollback__ = []                  # type: List[Tuple[int, Callable]]
        self.last_rb__loc__ = -1              # type: int
        # support for call stack tracing
781
        self.call_stack__ = []                # type: List[str]
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
        # snapshots of call stacks
        self.history__ = []                   # type: List[HistoryRecord]
        # also needed for call stack tracing
        self.moving_forward__ = False         # type: bool
        self.recursion_locations__ = set()    # type: Set[int]


    @property
    def reversed__(self) -> StringView:
        """
        Returns a reversed version of the currently parsed document. As
        about the only case where this is needed is the Lookbehind-parser,
        this is done lazily.
        """
        if not self._reversed__:
797
            self._reversed__ = StringView(self.document__.get_text()[::-1])
798
799
800
801
802
803
804
805
        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
806
        if parser.pname:
807
            # prevent overwriting instance variables or parsers of a different class
di68kap's avatar
di68kap committed
808
809
            assert parser.pname not in self.__dict__ or \
                   isinstance(self.__dict__[parser.pname], parser.__class__), \
810
                ('Cannot add parser "%s" because a field with the same name '
di68kap's avatar
di68kap committed
811
                 'already exists in grammar object: %s!'
di68kap's avatar
di68kap committed
812
813
814
                 % (parser.pname, str(self.__dict__[parser.pname])))
            setattr(self, parser.pname, parser)
        parser.tag_name = parser.pname or parser.ptype
815
816
817
818
        self.all_parsers__.add(parser)
        parser.grammar = self


eckhart's avatar
eckhart committed
819
820
    def __call__(self,
                 document: str,
eckhart's avatar
eckhart committed
821
                 start_parser: Union[str, Parser] = "root_parser__",
eckhart's avatar
eckhart committed
822
                 track_history: bool = False) -> RootNode:
823
824
825
826
827
        """
        Parses a document with with parser-combinators.

        Args:
            document (str): The source text to be parsed.
di68kap's avatar
di68kap committed
828
829
            start_parser (str or Parser): The name of the parser with which
                to start. This is useful for testing particular parsers
830
                (i.e. particular parts of the EBNF-Grammar.)
di68kap's avatar
di68kap committed
831
832
833
834
835
            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.
836
        Returns:
di68kap's avatar
di68kap committed
837
            Node: The root node to the parse tree.
838
        """
839

eckhart's avatar
eckhart committed
840
        def tail_pos(predecessors: Union[List[Node], Tuple[Node, ...], None]) -> int:
eckhart's avatar
eckhart committed
841
842
            """Adds the position after the last node in the list of
            predecessors to the node."""
eckhart's avatar
eckhart committed
843
            return predecessors[-1].pos + len(predecessors[-1]) if predecessors else 0
eckhart's avatar
eckhart committed
844

845
846
847
848
849
850
851
        # assert isinstance(document, str), type(document)
        if self._dirty_flag__:
            self._reset__()
            for parser in self.all_parsers__:
                parser.reset()
        else:
            self._dirty_flag__ = True
di68kap's avatar
di68kap committed
852
853
854
855
        self.history_tracking__ = track_history or is_logging()
        # safe tracking state, because history_tracking__ might be set to false, later,
        # but original tracking state is needed for additional error information.
        track_history = self.history_tracking__
856
        self.document__ = StringView(document)
eckhart's avatar
eckhart committed
857
        self.document_length__ = len(self.document__)
858
        self.document_lbreaks__ = linebreaks(document) if self.history_tracking__ else []
Eckhart Arnold's avatar
Eckhart Arnold committed
859
        self.last_rb__loc__ = -1  # rollback location
860
        parser = self[start_parser] if isinstance(start_parser, str) else start_parser
eckhart's avatar
eckhart committed
861
        self.start_parser__ = parser
862
863
864
865
866
867
868
869
        assert parser.grammar == self, "Cannot run parsers from a different grammar object!" \
                                       " %s vs. %s" % (str(self), str(parser.grammar))
        result = None  # type: Optional[Node]
        stitches = []  # type: List[Node]
        rest = self.document__
        if not rest:
            result, _ = parser(rest)
            if result is None:
Eckhart Arnold's avatar
Eckhart Arnold committed
870
                result = Node(ZOMBIE_TAG, '').with_pos(0)
eckhart's avatar
eckhart committed
871
                self.tree__.new_error(result,
di68kap's avatar
di68kap committed
872
873
                                      'Parser "%s" did not match empty document.' % str(parser),
                                      Error.PARSER_DID_NOT_MATCH)
874
875
876
877
878
879
        while rest and len(stitches) < MAX_DROPOUTS:
            result, rest = parser(rest)
            if rest:
                fwd = rest.find("\n") + 1 or len(rest)
                skip, rest = rest[:fwd], rest[fwd:]
                if result is None:
di68kap's avatar
di68kap committed
880
881
882
883
884
885
                    err_info = '' if not track_history else \
                               '\n    Most advanced: %s\n    Last match:    %s;' % \
                               (str(HistoryRecord.most_advanced_match(self.history__)),
                                str(HistoryRecord.last_match(self.history__)))
                    # Check if a Lookahead-Parser did match. Needed for testing, because
                    # in a test case this is not necessarily an error.
eckhart's avatar
eckhart committed
886
                    last_record = self.history__[-2] if len(self.history__) > 1 else None  # type: Optional[HistoryRecord]
887

eckhart's avatar
eckhart committed
888
                    if last_record and parser != self.root_parser__ \
di68kap's avatar
di68kap committed
889
                            and last_record.status == HistoryRecord.MATCH \
eckhart's avatar
eckhart committed
890
891
                            and last_record.node.pos \
                            + len(last_record.node) >= len(self.document__) \
892
893
894
                            and any(tn in self and isinstance(self[tn], Lookahead)
                                    or tn[0] == ':' and issubclass(eval(tn[1:]), Lookahead)
                                    for tn in last_record.call_stack):
di68kap's avatar
di68kap committed
895
896
897
898
899
                        error_msg = 'Parser did not match except for lookahead! ' + err_info
                        error_code = Error.PARSER_LOOKAHEAD_MATCH_ONLY
                    else:
                        error_msg = 'Parser did not match!' + err_info
                        error_code = Error.PARSER_DID_NOT_MATCH
900
901
                else:
                    stitches.append(result)
eckhart's avatar
eckhart committed
902
903
904
905
906
907
                    error_msg = "Parser stopped before end" \
                        + (("! trying to recover"
                            + (" but stopping history recording at this point."
                               if self.history_tracking__ else "..."))
                            if len(stitches) < MAX_DROPOUTS
                            else " too often! Terminating parser.")
di68kap's avatar
di68kap committed
908
                    error_code = Error.PARSER_STOPPED_BEFORE_END
Eckhart Arnold's avatar
Eckhart Arnold committed
909
                stitches.append(Node(ZOMBIE_TAG, skip).with_pos(tail_pos(stitches)))
di68kap's avatar
di68kap committed
910
                self.tree__.new_error(stitches[-1], error_msg, error_code)
911
                if self.history_tracking__:
eckhart's avatar
eckhart committed
912
913
914
915
916
917
                    # # some parsers may have matched and left history records with nodes != None.
                    # # Because these are not connected to the stitched root node, their pos-
                    # # properties will not be initialized by setting the root node's pos property
                    # # to zero. Therefore, their pos properties need to be initialized here
                    # for record in self.history__:
                    #     if record.node and record.node._pos < 0:
Eckhart Arnold's avatar
Eckhart Arnold committed
918
                    #         record.node.with_pos(0)
919
920
                    record = HistoryRecord(self.call_stack__.copy(), stitches[-1], rest,
                                           self.line_col__(rest))
921
922
923
924
925
                    self.history__.append(record)
                    # stop history tracking when parser returned too early
                    self.history_tracking__ = False
        if stitches:
            if rest:
926
                stitches.append(Node(ZOMBIE_TAG, rest))
927
            #try:
Eckhart Arnold's avatar
Eckhart Arnold committed
928
            result = Node(ZOMBIE_TAG, tuple(stitches)).with_pos(0)
929
930
            # except AssertionError as error:
            #     # some debugging output
931
            #     print(Node(ZOMBIE_TAG, tuple(stitches)).as_sxpr())
932
            #     raise error
933
        if any(self.variables__.values()):
di68kap's avatar
di68kap committed
934
            error_msg = "Capture-retrieve-stack not empty after end of parsing: " \
935
                + str(self.variables__)
di68kap's avatar
di68kap committed
936
            error_code = Error.CAPTURE_STACK_NOT_EMPTY
937
938
939
940
941
            if result:
                if result.children:
                    # add another child node at the end to ensure that the position
                    # of the error will be the end of the text. Otherwise, the error
                    # message above ("...after end of parsing") would appear illogical.
Eckhart Arnold's avatar
Eckhart Arnold committed
942
                    error_node = Node(ZOMBIE_TAG, '').with_pos(tail_pos(result.children))
di68kap's avatar
di68kap committed
943
                    self.tree__.new_error(error_node, error_msg, error_code)
eckhart's avatar
eckhart committed
944
                    result.result = result.children + (error_node,)
945
                else:
di68kap's avatar
di68kap committed
946
                    self.tree__.new_error(result, error_msg, error_code)
eckhart's avatar
eckhart committed
947
        # result.pos = 0  # calculate all positions
948
        # result.errors(self.document__)
eckhart's avatar
eckhart committed
949
950
        if result:
            self.tree__.swallow(result)
eckhart's avatar
eckhart committed
951
        self.start_parser__ = None
952
        return self.tree__
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970


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


    def rollback_to__(self, location):
        """
        Rolls back the variable stacks (`self.variables`) to its
        state at an earlier location in the parsed document.
        """
Eckhart Arnold's avatar
Eckhart Arnold committed
971
        while self.rollback__ and self.rollback__[-1][0] >= location:
972
973
974
975
976
977
978
979
980
981
            _, rollback_func = self.rollback__.pop()
            # assert not loc > self.last_rb__loc__, \
            #     "Rollback confusion: line %i, col %i < line %i, col %i" % \
            #     (*line_col(self.document__, len(self.document__) - loc),
            #      *line_col(self.document__, len(self.document__) - self.last_rb__loc__))
            rollback_func()
        self.last_rb__loc__ == self.rollback__[-1][0] if self.rollback__ \
            else (len(self.document__) + 1)


982
983
984
985
986
987
988
989
990
    def line_col__(self, text):
        """
        Returns the line and column where text is located in the document.
        Does not check, whether text is actually a substring of the currently
        parsed document.
        """
        return line_col(self.document_lbreaks__, self.document_length__ - len(text))


991
992
993
994
995
996
997
998
999
1000
def dsl_error_msg(parser: Parser, error_str: str) -> str:
    """
    Returns an error message for errors in the parser configuration,
    e.g. errors that result in infinite loops.

    Args:
        parser (Parser):  The parser where the error was noticed. Note
            that this is not necessarily the parser that caused the
            error but only where the error became apparent.
        error_str (str):  A short string describing the error.