parser.py 76.2 KB
Newer Older
1
"""parsers.py - parser combinators for 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
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46

Module ``parsers.py`` contains a number of classes that together
make up parser combinators for left-recursive grammers. For each
element of the extended Backus-Naur-Form as well as for a regular
expression token a class is defined. The set of classes can be used to
define a parser for (ambiguous) left-recursive grammers.


References and Acknowledgements:

Dominikus Herzberg: Objekt-orientierte Parser-Kombinatoren in Python,
Blog-Post, September, 18th 2008 on denkspuren. gedanken, ideen,
anregungen und links rund um informatik-themen, URL:
http://denkspuren.blogspot.de/2008/09/objekt-orientierte-parser-kombinatoren.html

Dominikus Herzberg: Eine einfache Grammatik für LaTeX, Blog-Post,
September, 18th 2008 on denkspuren. gedanken, ideen, anregungen und
links rund um informatik-themen, URL:
http://denkspuren.blogspot.de/2008/09/eine-einfache-grammatik-fr-latex.html

Dominikus Herzberg: Uniform Syntax, Blog-Post, February, 27th 2007 on
denkspuren. gedanken, ideen, anregungen und links rund um
informatik-themen, URL:
http://denkspuren.blogspot.de/2007/02/uniform-syntax.html

Richard A. Frost, Rahmatullah Hafiz and Paul Callaghan: Parser
Combinators for Ambiguous Left-Recursive Grammars, in: P. Hudak and
D.S. Warren (Eds.): PADL 2008, LNCS 4902, pp. 167–181, Springer-Verlag
Berlin Heidelberg 2008.

47
48
49
50
Elizabeth Scott and Adrian Johnstone, GLL Parsing,
in: Electronic Notes in Theoretical Computer Science 253 (2010) 177–189,
http://dotat.at/tmp/gll.pdf

Eckhart Arnold's avatar
Eckhart Arnold committed
51
52
Juancarlo Añez: grako, a PEG parser generator in Python,
https://bitbucket.org/apalala/grako
53
54
55

Vegard Øye: General Parser Combinators in Racket, 2012,
https://epsil.github.io/gll/
56
57
"""

58

Eckhart Arnold's avatar
Eckhart Arnold committed
59
import abc
60
import copy
61
import os
Eckhart Arnold's avatar
Eckhart Arnold committed
62
63
from functools import partial

64
65
66
67
try:
    import regex as re
except ImportError:
    import re
68
try:
69
    from typing import Any, Callable, cast, Dict, Iterator, List, Set, Tuple, Union, Optional
70
71
72
73
    # try:
    #     from typing import Collection
    # except ImportError:
    #     pass
74
except ImportError:
75
    from .typing34 import Any, Callable, cast, Dict, Iterator, List, Set, Tuple, Union, Optional
76

77
from DHParser.toolkit import is_logging, log_dir, logfile_basename, escape_re, sane_parser_name
Eckhart Arnold's avatar
Eckhart Arnold committed
78
79
from DHParser.syntaxtree import WHITESPACE_PTYPE, TOKEN_PTYPE, ZOMBIE_PARSER, ParserBase, \
    Node, TransformationFunc
80
81
from DHParser.toolkit import StringView, EMPTY_STRING_VIEW, sv_match, sv_index, sv_search, \
    load_if_file, error_messages, line_col
Eckhart Arnold's avatar
Eckhart Arnold committed
82

83
__all__ = ('PreprocessorFunc',
84
           'HistoryRecord',
Eckhart Arnold's avatar
Eckhart Arnold committed
85
           'Parser',
86
           'Grammar',
87
88
89
           'RX_PREPROCESSOR_TOKEN',
           'BEGIN_TOKEN',
           'END_TOKEN',
Eckhart Arnold's avatar
Eckhart Arnold committed
90
           'make_token',
91
92
           'nil_preprocessor',
           'PreprocessorToken',
Eckhart Arnold's avatar
Eckhart Arnold committed
93
94
95
96
           'RegExp',
           'RE',
           'Token',
           'mixin_comment',
97
98
99
           # 'UnaryOperator',
           # 'NaryOperator',
           'Synonym',
Eckhart Arnold's avatar
Eckhart Arnold committed
100
101
102
           'Optional',
           'ZeroOrMore',
           'OneOrMore',
103
           'Series',
Eckhart Arnold's avatar
Eckhart Arnold committed
104
105
106
107
108
109
110
           'Alternative',
           'FlowOperator',
           'Required',
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
111
112
113
           'last_value',
           'counterpart',
           'accumulate',
Eckhart Arnold's avatar
Eckhart Arnold committed
114
115
116
117
           'Capture',
           'Retrieve',
           'Pop',
           'Forward',
118
           'Compiler',
119
           'compile_source')
Eckhart Arnold's avatar
Eckhart Arnold committed
120

121

122

123
124
125
126
127
128
129
########################################################################
#
# Grammar and parsing infrastructure
#
########################################################################


130
PreprocessorFunc = Union[Callable[[str], str], partial]
131
132


133
134
135
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
Eckhart Arnold's avatar
Eckhart Arnold committed
136
MAX_DROPOUTS = 3  # type: int
137
# stop trying to recover parsing after so many errors
138

139

140
141
142
143
144
145
146
147
148
149
150
151
152
class HistoryRecord:
    """
    Stores debugging information about one completed step in the
    parsing history. 

    A parsing step is "completed" when the last one of a nested
    sequence of parser-calls returns. The call stack including
    the last parser call will be frozen in the ``HistoryRecord``-
    object. In addition a reference to the generated leaf node
    (if any) will be stored and the result status of the last
    parser call, which ist either MATCH, FAIL (i.e. no match)
    or ERROR.
    """
153
    __slots__ = ('call_stack', 'node', 'remaining', 'line_col')
154
155
156
157
158

    MATCH = "MATCH"
    ERROR = "ERROR"
    FAIL = "FAIL"

159
    def __init__(self, call_stack: List['Parser'], node: Node, remaining: int) -> None:
160
        # copy call stack, dropping uninformative Forward-Parsers
Eckhart Arnold's avatar
Eckhart Arnold committed
161
        self.call_stack = [p for p in call_stack if p.ptype != ":Forward"]  # type: List['Parser']
162
163
        self.node = node                # type: Node
        self.remaining = remaining      # type: int
164
        document = call_stack[-1].grammar.document__.text if call_stack else ''
165
166
167
168
169
        self.line_col = line_col(document, len(document) - remaining)  # type: Tuple[int, int]

    def __str__(self):
        return 'line %i, column %i:  %s  "%s"' % \
               (self.line_col[0], self.line_col[1], self.stack, str(self.node))
170

171
    def err_msg(self) -> str:
172
173
        return self.ERROR + ": " + "; ".join(self.node._errors).replace('\n', '\\')

174
    @property
175
    def stack(self) -> str:
176
        return "->".join((p.repr if p.ptype == ':RegExp' else p.name or p.ptype)
177
                         for p in self.call_stack)
178
179

    @property
180
    def status(self) -> str:
181
182
        return self.FAIL if self.node is None else \
            self.err_msg() if self.node._errors else self.MATCH
183
184

    @property
185
    def extent(self) -> slice:
186
        return (slice(-self.remaining - len(self.node), -self.remaining) if self.node
187
                else slice(-self.remaining, None))
188

189
    @staticmethod
Eckhart Arnold's avatar
Eckhart Arnold committed
190
    def last_match(history: List['HistoryRecord']) -> Union['HistoryRecord', None]:
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
        """
        Returns the last match from the parsing-history.
        Args:
            history:  the parsing-history as a list of HistoryRecord objects

        Returns:
            the history record of the last match or none if either history is
            empty or no parser could match
        """
        for record in reversed(history):
            if record.status == HistoryRecord.MATCH:
                return record
        return None

    @staticmethod
Eckhart Arnold's avatar
Eckhart Arnold committed
206
    def most_advanced_match(history: List['HistoryRecord']) -> Union['HistoryRecord', None]:
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
        """
        Returns the closest-to-the-end-match from the parsing-history.
        Args:
            history:  the parsing-history as a list of HistoryRecord objects

        Returns:
            the history record of the closest-to-the-end-match or none if either history is
            empty or no parser could match
        """
        remaining = -1
        result = None
        for record in history:
            if (record.status == HistoryRecord.MATCH and
                    (record.remaining < remaining or remaining < 0)):
                result = record
                remaining = record.remaining
        return result


226
def add_parser_guard(parser_func):
227
228
229
230
231
    """
    Add a wrapper function to a parser functions (i.e. Parser.__call__ method)
    that takes care of memoizing, left recursion and optionally tracing
    (aka "history tracking") of parser calls. Returns the wrapped call.
    """
232
    def guarded_call(parser: 'Parser', text: StringView) -> Tuple[Node, StringView]:
di68kap's avatar
di68kap committed
233
234
235
236
237
238
239
240
241
242
243
        # def memoized(parser, location):
        #     node = parser.visited[location]
        #     rlen = location - (0 if node is None else node.len)
        #     rest = grammar.document__[-rlen:] if rlen else EMPTY_STRING_VIEW
        #     return node, rest
        #     # NOTE: An older and simpler implementation of memoization
        #     # relied on `parser.visited[location] == node, rest`. Although,
        #     # rest is really just a substring of one and the same document,
        #     # this resulted in an explosion of memory usage. Seems that
        #     # `rext = text[i:]` really copies the sub-string. See:
        #     # https://mail.python.org/pipermail/python-dev/2008-May/079699.html
244

245
        try:
246
            location = len(text)    # mind that location is always the distance to the end
247
248
            grammar = parser.grammar  # grammar may be 'None' for unconnected parsers!

249
250
            if grammar.last_rb__loc__ <= location:
                grammar.rollback_to__(location)
251

252
253
254
            # if location has already been visited by the current parser,
            # return saved result
            if location in parser.visited:
di68kap's avatar
di68kap committed
255
                return parser.visited[location]
256

257
258
            # break left recursion at the maximum allowed depth
            if parser.recursion_counter.setdefault(location, 0) > LEFT_RECURSION_DEPTH:
259
                grammar.recursion_locations__.add(location)
260
261
                return None, text

Eckhart Arnold's avatar
Eckhart Arnold committed
262
263
            if grammar.history_tracking__:
                grammar.call_stack__.append(parser)
264
                grammar.moving_forward__ = True
Eckhart Arnold's avatar
Eckhart Arnold committed
265

266
267
268
269
270
            parser.recursion_counter[location] += 1

            # run original __call__ method
            node, rest = parser_func(parser, text)

271
            if node is None:
272
                # retrieve an earlier match result (from left recursion) if it exists
273
274
                if location in grammar.recursion_locations__:
                    if location in parser.visited:
di68kap's avatar
di68kap committed
275
                        node, rest = parser.visited[location]
276
277
278
                    # 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__:
279
                    # otherwise also cache None-results
di68kap's avatar
di68kap committed
280
                    parser.visited[location] = (None, rest)
281
282
283
284
285
286
            elif ((grammar.memoization__ or location in grammar.recursion_locations__)
                  and grammar.last_rb__loc__ > location):
                # - 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
di68kap's avatar
di68kap committed
287
                parser.visited[location] = (node, rest)
288
289
290

            parser.recursion_counter[location] -= 1

291
            if grammar.history_tracking__:
292
                # don't track returning parsers except in case an error has occurred
293
                remaining = len(rest)
294
                if grammar.moving_forward__ or (node and node._errors):
295
                    record = HistoryRecord(grammar.call_stack__, node, remaining)
296
                    grammar.history__.append(record)
297
                    # print(record.stack, record.status, rest[:20].replace('\n', '|'))
298
                grammar.moving_forward__ = False
299
                grammar.call_stack__.pop()
300

301
302
303
304
        except RecursionError:
            node = Node(None, text[:min(10, max(1, text.find("\n")))] + " ...")
            node.add_error("maximum recursion depth of parser reached; "
                           "potentially due to too many errors!")
305
            rest = EMPTY_STRING_VIEW
306
307
308
309
310
311

        return node, rest

    return guarded_call


Eckhart Arnold's avatar
Eckhart Arnold committed
312
class ParserMetaClass(abc.ABCMeta):
313
314
315
316
317
    """
    ParserMetaClass adds a wrapper to the __call__ method of parser
    objects during initialization that takes care of memoizing,
    left recursion and tracing.
    """
318
    def __init__(cls, name, bases, attrs):
319
320
        guarded_parser_call = add_parser_guard(cls.__call__)
        # The following check is necessary for classes that don't override
321
322
323
324
325
326
327
        # the __call__() method, because in these cases the non-overridden
        # __call__()-method would be substituted a second time!
        if cls.__call__.__code__ != guarded_parser_call.__code__:
            cls.__call__ = guarded_parser_call
        super(ParserMetaClass, cls).__init__(name, bases, attrs)


Eckhart Arnold's avatar
Eckhart Arnold committed
328
class Parser(ParserBase, metaclass=ParserMetaClass):
329
330
331
332
    """
    (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.
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349

    Since parsers can contain other parsers (see classes UnaryOperator
    and NaryOperator) they form a cyclical directed graph. A root
    parser is a parser from which all other parsers can be reached.
    Usually, there is one root parser which serves as the starting
    point of the parsing process. When speaking of "the root parser"
    it is this root parser object that is meant.

    There are two different types of parsers:

    1. *Named parsers* for which a name is set in field parser.name.
       The results produced by these parsers can later be retrieved in
       the AST by the parser name.

    2. *Anonymous parsers* where the name-field just contains the empty
       string. AST-transformation of Anonymous parsers can be hooked
       only to their class name, and not to the individual parser.
Eckhart Arnold's avatar
Eckhart Arnold committed
350
351

    Parser objects are callable and parsing is done by calling a parser
Eckhart Arnold's avatar
Eckhart Arnold committed
352
353
354
355
356
357
358
359
360
361
362
363
364
    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 `Optional`). 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.
365
366

    Attributes:
367
        visited:  Mapping of places this parser has already been to
di68kap's avatar
di68kap committed
368
                during the current parsing process onto the results the
369
370
371
372
373
374
375
376
                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.

377
        cycle_detection:  The apply()-method uses this variable to make
378
379
380
                sure that one and the same function will not be applied
                (recursively) a second time, if it has already been
                applied to this parser.
381
382
383

        grammar:  A reference to the Grammar object to which the parser
                is attached.
384
385
    """

386
387
    ApplyFunc = Callable[['Parser'], None]

388
389
    def __init__(self, name: str = '') -> None:
        # assert isinstance(name, str), str(name)
Eckhart Arnold's avatar
Eckhart Arnold committed
390
        super(Parser, self).__init__(name)
391
        self._grammar = None  # type: 'Grammar'
392
393
        self.reset()

Eckhart Arnold's avatar
Eckhart Arnold committed
394
    def __deepcopy__(self, memo):
Eckhart Arnold's avatar
Eckhart Arnold committed
395
396
397
398
399
400
        """Deepcopy method of the parser. Upon instantiation of a Grammar-
        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.
        """
401
402
        return self.__class__(self.name)

403
    def reset(self):
Eckhart Arnold's avatar
Eckhart Arnold committed
404
405
406
        """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."""
di68kap's avatar
di68kap committed
407
        self.visited = dict()            # type: Dict[int, Tuple[Node, StringView]]
408
        self.recursion_counter = dict()  # type: Dict[int, int]
409
        self.cycle_detection = set()     # type: Set[Callable]
410
        return self
411

412
    def __call__(self, text: StringView) -> Tuple[Node, StringView]:
Eckhart Arnold's avatar
Eckhart Arnold committed
413
414
415
        """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."""
416
417
        return None, text  # default behaviour: don't match

418
    def __add__(self, other: 'Parser') -> 'Series':
419
420
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
421
        return Series(self, other)
422

423
    def __or__(self, other: 'Parser') -> 'Alternative':
424
425
426
        """The | operator generates an alternative parser that applies
        the first parser and, if that does not match, the second parser.
        """
427
428
        return Alternative(self, other)

429
    @property
430
    def grammar(self) -> 'Grammar':
431
432
433
        return self._grammar

    @grammar.setter
434
    def grammar(self, grammar: 'Grammar'):
Eckhart Arnold's avatar
Eckhart Arnold committed
435
436
437
438
439
440
        if self._grammar is None:
            self._grammar = grammar
            self._grammar_assigned_notifier()
        else:
            assert self._grammar == grammar, \
                "Parser has already been assigned to a different Grammar object!"
441
442

    def _grammar_assigned_notifier(self):
443
444
        """A function that notifies the parser object that it has been
        assigned to a grammar."""
445
446
        pass

447
    def apply(self, func: ApplyFunc):
448
449
        """
        Applies function `func(parser)` recursively to this parser and all
Eckhart Arnold's avatar
Eckhart Arnold committed
450
        descendant parsers if any exist. The same function can never
451
452
453
454
455
456
457
458
459
460
        be applied twice between calls of the ``reset()``-method!
        """
        if func in self.cycle_detection:
            return False
        else:
            self.cycle_detection.add(func)
            func(self)
            return True


461
def mixin_comment(whitespace: str, comment: str) -> str:
462
463
    """
    Returns a regular expression that merges comment and whitespace
464
465
466
467
468
469
470
471
472
473
474
    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


475
class Grammar:
476
    """
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
    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.

    Example for direct instantian of a grammar:

        >>> number = RE('\d+') + RE('\.') + RE('\d+') | RE('\d+')
        >>> number_parser = Grammar(number)
Eckhart Arnold's avatar
Eckhart Arnold committed
492
        >>> number_parser("3.1416").content()
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
        '3.1416'

    Collecting the parsers that define a grammar in a descentand 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.

    3. The parsers in 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.
514
515

    Example:
516

517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
        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+')
            factor = INTEGER | Token("(") + expression + Token(")")
            term = factor + ZeroOrMore((Token("*") | Token("/")) + factor)
            expression.set(term + ZeroOrMore((Token("+") | Token("-")) + term))
            root__ = expression

    Upon instantiation the parser objects are deep-copied to the
    Grammar object and assigned to object variables of the same name.
    Any parser that is directly assigned to a class variable is a
    'named' parser and its field `parser.name` contains the variable
    name after instantiation of the Grammar class. All other parsers,
535
536
537
    i.e. parsers that are defined within a `named` parser, remain
    "anonymous parsers" where `parser.name` is the empty string, unless
    a name has been passed explicitly upon instantiation.
538
539
540
541
542
543
544
545
546
547
548
    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.

549
    Grammar classes contain a few special class fields for implicit
550
551
    whitespace and comments that should be overwritten, if the defaults
    (no comments, horizontal right aligned whitespace) don't fit:
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
    COMMENT__:  regular expression string for matching comments
    WSP__:   regular expression for whitespace and comments
    wspL__:  regular expression string for left aligned whitespace,
             which either equals WSP__ or is empty.
    wspR__:  regular expression string for right aligned whitespace,
             which either equals WSP__ or is empty.
    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.
    parser_initializiation__:  Before the parser class (!) has been initialized,
             which happens upon the first time it is instantiated (see doctring for
             method `_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.

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

571
        hostory_tracking__:  A flag indicating that the parsing history shall
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
                be tracked

        wsp_left_parser__:  A parser for the default left-adjacent-whitespace
                or the zombie-parser (see `syntaxtree.ZOMBIE_PARSER`) if the
                default is empty. The default whitespace will be used by parsers
                `Token` and, if no other parsers are passed to its constructor,
                by parser `RE'.
        wsp_right_parser__: The same for the default right-adjacent-whitespace.
                Both wsp_left_parser__ and wsp_right_parser__ merely serve the
                purpose to avoid having to specify the default whitespace
                explicitly every time an `RE`-parser-object is created.

        _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.
        _reversed__:  the same text in reverse order - needed by the `Lookbehind'-
                parsers.
        variables__:  A mapping for variable names to a stack of their respective
                string values - needed by the `Capture`-, `Retrieve`- and `Pop`-
                parsers.
        rollback__:  A list of tuples (location, rollback-function) that are
                deposited by the `Capture`- and `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.
600
601
602
603
604
605
606
607
        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
                backtracks to a location at or before `last_rb__loc__` (which,
                since locations are counted from the reverse, means:
                `location >= last_rb__loc__`) then a rollback of all variable
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
                calling method `.rollback_to__(location)`.
608
609
610
611
612
613
614
        call_stack__:  A stack of all parsers that have been called. This
                is required for recording the parser history (for debugging)
                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.
615
616
617
618
619
620
621
622
623
624
        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
                recording is turned off! (See `add_parser_guard` and its local
                function `guarded_call`)
        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
                if full memoization is enabled. (See `add_parser_guard` and its
                local function `guarded_call`)
625
626
627
628
        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.
629
    """
630
631
    root__ = None  # type: Union[Parser, None]
    # root__ must be overwritten with the root-parser by grammar subclass
Eckhart Arnold's avatar
Eckhart Arnold committed
632
    parser_initialization__ = "pending"  # type: str
633
    # some default values
Eckhart Arnold's avatar
Eckhart Arnold committed
634
635
636
637
    COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    WSP__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
    wspL__ = ''     # type: str
    wspR__ = WSP__  # type: str
638

639

640
    @classmethod
641
    def _assign_parser_names__(cls):
642
643
        """
        Initializes the `parser.name` fields of those
644
645
        Parser objects that are directly assigned to a class field with
        the field's name, e.g.
646
            class Grammar(Grammar):
647
648
649
                ...
                symbol = RE('(?!\\d)\\w+')
        After the call of this method symbol.name == "symbol"
650
651
652
653
        holds. Names assigned via the ``name``-parameter of the
        constructor will not be overwritten. Parser names starting or
        ending with a double underscore like ``root__`` will be
        ignored. See ``toolkit.sane_parser_name()``
654
655
656
657

        This is done only once, upon the first instantiation of the
        grammar class!

658
659
660
661
662
        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. 
663
        """
664
665
666
667
668
669
670
671
672
        if cls.parser_initialization__ != "done":
            cdict = cls.__dict__
            for entry, parser in cdict.items():
                if isinstance(parser, Parser) and sane_parser_name(entry):
                    if not parser.name:
                        parser.name = entry
                    if (isinstance(parser, Forward) and (not parser.parser.name)):
                        parser.parser.name = entry
            cls.parser_initialization__ = "done"
673

674

Eckhart Arnold's avatar
Eckhart Arnold committed
675
676
677
    def __init__(self, root: Parser=None) -> None:
        # if not hasattr(self.__class__, 'parser_initialization__'):
        #     self.__class__.parser_initialization__ = "pending"
678
679
680
681
        # if not hasattr(self.__class__, 'wspL__'):
        #     self.wspL__ = ''
        # if not hasattr(self.__class__, 'wspR__'):
        #     self.wspR__ = ''
682
        self.all_parsers__ = set()  # type: Set[Parser]
683
684
        self._dirty_flag__ = False  # type: bool
        self.history_tracking__ = False  # type: bool
685
        self.memoization__ = True  # type: bool
686
        self._reset__()
687

Eckhart Arnold's avatar
Eckhart Arnold committed
688
        # prepare parsers in the class, first
689
        self._assign_parser_names__()
690

Eckhart Arnold's avatar
Eckhart Arnold committed
691
692
693
694
        # 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 Arnold's avatar
Eckhart Arnold committed
695
        self.root__ = copy.deepcopy(root) if root else copy.deepcopy(self.__class__.root__)
696

697
        if self.wspL__:
Eckhart Arnold's avatar
Eckhart Arnold committed
698
            self.wsp_left_parser__ = Whitespace(self.wspL__)  # type: ParserBase
699
            self.wsp_left_parser__.grammar = self
700
            self.all_parsers__.add(self.wsp_left_parser__)  # don't you forget about me...
701
702
703
        else:
            self.wsp_left_parser__ = ZOMBIE_PARSER
        if self.wspR__:
Eckhart Arnold's avatar
Eckhart Arnold committed
704
            self.wsp_right_parser__ = Whitespace(self.wspR__)  # type: ParserBase
705
            self.wsp_right_parser__.grammar = self
706
            self.all_parsers__.add(self.wsp_right_parser__)  # don't you forget about me...
707
708
        else:
            self.wsp_right_parser__ = ZOMBIE_PARSER
709
        self.root__.apply(self._add_parser__)
710

711

712
    def __getitem__(self, key):
713
714
715
        try:
            return self.__dict__[key]
        except KeyError:
716
717
            parser_template = getattr(self, key, None)
            if parser_template:
Eckhart Arnold's avatar
Eckhart Arnold committed
718
                # add parser to grammar object on the fly...
719
                parser = copy.deepcopy(parser_template)
720
                parser.apply(self._add_parser__)
721
                # assert self[key] == parser
Eckhart Arnold's avatar
Eckhart Arnold committed
722
                return self[key]
723
            raise KeyError('Unknown parser "%s" !' % key)
724

725

726
    def _reset__(self):
727
728
        self.document__ = EMPTY_STRING_VIEW   # type: StringView
        self._reversed__ = EMPTY_STRING_VIEW  # type: StringView
729
        # variables stored and recalled by Capture and Retrieve parsers
730
731
        self.variables__ = dict()     # type: Dict[str, List[str]]
        self.rollback__ = []          # type: List[Tuple[int, Callable]]
732
        self.last_rb__loc__ = -1  # type: int
733
        # support for call stack tracing
734
        self.call_stack__ = []        # type: List[Parser]
735
        # snapshots of call stacks
736
        self.history__ = []           # type: List[HistoryRecord]
737
        # also needed for call stack tracing
738
739
        self.moving_forward__ = False  # type: bool
        self.recursion_locations__ = set()  # type: Set[int]
740

741

Eckhart Arnold's avatar
Eckhart Arnold committed
742
    @property
Eckhart Arnold's avatar
Eckhart Arnold committed
743
    def reversed__(self) -> StringView:
Eckhart Arnold's avatar
Eckhart Arnold committed
744
        if not self._reversed__:
745
            self._reversed__ = StringView(self.document__.text[::-1])
Eckhart Arnold's avatar
Eckhart Arnold committed
746
747
748
        return self._reversed__


749
    def _add_parser__(self, parser: Parser) -> None:
750
751
        """
        Adds the particular copy of the parser object to this
752
        particular instance of Grammar.
753
        """
754
        if parser.name:
Eckhart Arnold's avatar
Eckhart Arnold committed
755
756
757
            # prevent overwriting instance variables or parsers of a different class
            assert parser.name not in self.__dict__ or \
                   isinstance(self.__dict__[parser.name], parser.__class__), \
758
759
                ('Cannot add parser "%s" because a field with the same name '
                 'already exists in grammar object!' % parser.name)
760
            setattr(self, parser.name, parser)
761
        self.all_parsers__.add(parser)
762
763
        parser.grammar = self

764

Eckhart Arnold's avatar
Eckhart Arnold committed
765
    def __call__(self, document: str, start_parser="root__") -> Node:
766
767
        """
        Parses a document with with parser-combinators.
768
769
770

        Args:
            document (str): The source text to be parsed.
771
772
773
            start_parser (str): The name of the parser with which to
                start. This is useful for testing particular parsers
                (i.e. particular parts of the EBNF-Grammar.)
774
775
776
        Returns:
            Node: The root node ot the parse tree.
        """
777
        # assert isinstance(document, str), type(document)
778
779
        if self.root__ is None:
            raise NotImplementedError()
780
        if self._dirty_flag__:
781
            self._reset__()
782
            for parser in self.all_parsers__:
783
784
                parser.reset()
        else:
785
            self._dirty_flag__ = True
786
        self.history_tracking__ = is_logging()
787
788
        self.document__ = StringView(document)
        self.last_rb__loc__ = len(self.document__) + 1  # rollback location
Eckhart Arnold's avatar
Eckhart Arnold committed
789
        parser = self[start_parser] if isinstance(start_parser, str) else start_parser
790
791
        assert parser.grammar == self, "Cannot run parsers from a different grammar object!" \
                                       " %s vs. %s" % (str(self), str(parser.grammar))
792
        stitches = []  # type: List[Node]
793
        rest = self.document__
794
795
        if not rest:
            result, ignore = parser(rest)
di68kap's avatar
di68kap committed
796
797
798
            if result is None:
                result = Node(None, '')
                result.add_error('Parser "%s" did not match empty document.' % str(parser))
799
800
801
802
803
804
        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:
805
806
807
808
                    error_msg = 'Parser did not match! Invalid source file?' \
                                '\n    Most advanced: %s\n    Last match:    %s;' % \
                                (str(HistoryRecord.most_advanced_match(self.history__)),
                                 str(HistoryRecord.last_match(self.history__)))
809
810
811
                else:
                    stitches.append(result)
                    error_msg = "Parser stopped before end" + \
812
813
                                (("! trying to recover" +
                                  (" but stopping history recording at this point."
814
                                   if self.history_tracking__ else "..."))
815
816
817
818
                                 if len(stitches) < MAX_DROPOUTS
                                 else " too often! Terminating parser.")
                stitches.append(Node(None, skip))
                stitches[-1].add_error(error_msg)
819
                if self.history_tracking__:
820
821
822
823
                    # some parsers may have matched and left history records with nodes != None.
                    # Because these are not connected to the stiched 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
824
                    for record in self.history__:
825
826
                        if record.node and record.node._pos < 0:
                            record.node.pos = 0
827
828
829
                    record = HistoryRecord(self.call_stack__.copy(), stitches[-1], len(rest))
                    self.history__.append(record)
                    self.history_tracking__ = False
830
831
832
833
        if stitches:
            if rest:
                stitches.append(Node(None, rest))
            result = Node(None, tuple(stitches))
834
        if any(self.variables__.values()):
Eckhart Arnold's avatar
Eckhart Arnold committed
835
836
837
838
839
840
841
842
843
844
845
            error_str = "Capture-retrieve-stack not empty after end of parsing: " + \
                            str(self.variables__)
            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.
                error_node = Node(ZOMBIE_PARSER, '')
                error_node.add_error(error_str)
                result.result = result.children + (error_node,)
            else:
                result.add_error(error_str)
846
847
848
        result.pos = 0  # calculate all positions
        return result

849

850
    def push_rollback__(self, location, func):
851
852
        """
        Adds a rollback function that either removes or re-adds
853
854
855
856
        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.
        """
857
        self.rollback__.append((location, func))
858
        # print("push:  line %i, col %i" % line_col(self.document__, len(self.document__) - location))
859
860
        self.last_rb__loc__ = location

861

862
    def rollback_to__(self, location):
863
864
        """
        Rolls back the variable stacks (`self.variables`) to its
865
866
        state at an earlier location in the parsed document.
        """
867
        # print("rollback:  line %i, col %i" % line_col(self.document__, len(self.document__) - location))
868
869
        while self.rollback__ and self.rollback__[-1][0] <= location:
            loc, rollback_func = self.rollback__.pop()
870
871
872
873
            # 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__))
874
            rollback_func()
875
            # print("rb to:  line %i, col %i" % line_col(self.document__, len(self.document__) - loc))
876
877
878
        self.last_rb__loc__ == self.rollback__[-1][0] if self.rollback__ \
            else (len(self.document__) + 1)

879

880
    def log_parsing_history__(self, log_file_name: str = '') -> None:
881
882
        """
        Writes a log of the parsing history of the most recently parsed
883
884
885
        document. 
        """
        def prepare_line(record):
886
            excerpt = self.document__.text.__getitem__(record.extent)[:25].replace('\n', '\\n')
887
            excerpt = "'%s'" % excerpt if len(excerpt) < 25 else "'%s...'" % excerpt
888
            return record.stack, record.status, excerpt
889
890

        def write_log(history, log_name):
Eckhart Arnold's avatar
Eckhart Arnold committed
891
            path = os.path.join(log_dir(), log_name + "_parser.log")
Eckhart Arnold's avatar
Eckhart Arnold committed
892
893
894
            if os.path.exists(path):
                os.remove(path)
                print('WARNING: Log-file "%s" already existed and was deleted.' % path)
895
896
897
898
            if history:
                with open(path, "w", encoding="utf-8") as f:
                    f.write("\n".join(history))

899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
        if is_logging():
            assert self.history__, \
                "Parser did not yet run or logging was turned off when running parser!"
            if not log_file_name:
                name = self.__class__.__name__
                log_file_name = name[:-7] if name.lower().endswith('grammar') else name
            full_history, match_history, errors_only = [], [], []
            for record in self.history__:
                line = ";  ".join(prepare_line(record))
                full_history.append(line)
                if record.node and record.node.parser.ptype != WHITESPACE_PTYPE:
                    match_history.append(line)
                    if record.node.error_flag:
                        errors_only.append(line)
            write_log(full_history, log_file_name + '_full')
            write_log(match_history, log_file_name + '_match')
            write_log(errors_only, log_file_name + '_errors')
916
917


Eckhart Arnold's avatar
Eckhart Arnold committed
918
def dsl_error_msg(parser: Parser, error_str: str) -> str:
919
920
    """
    Returns an error message for errors in the parser configuration,
921
922
923
    e.g. errors that result in infinite loops.

    Args:
Eckhart Arnold's avatar
Eckhart Arnold committed
924
        parser (Parser):  The parser where the error was noticed. Note
925
            that this is not necessarily the parser that caused the
926
            error but only where the error became apparent.
927
928
929
930
931
        error_str (str):  A short string describing the error.
    Returns:  
        str: An error message including the call stack if history 
        tacking has been turned in the grammar object.
    """
932
    msg = ["DSL parser specification error:", error_str, 'Caught by parser "%s".' % str(parser)]
933
934
    if parser.grammar.history__:
        msg.extend(["\nCall stack:", parser.grammar.history__[-1].stack])
935
936
937
938
939
    else:
        msg.extend(["\nEnable history tracking in Grammar object to display call stack."])
    return " ".join(msg)


940

941
942
943
944
945
946
947
########################################################################
#
# Token and Regular Expression parser classes (i.e. leaf classes)
#
########################################################################


948
949
950
RX_PREPROCESSOR_TOKEN = re.compile('\w+')
BEGIN_TOKEN = '\x1b'
END_TOKEN = '\x1c'
951
952


953
def make_token(token: str, argument: str = '') -> str:
954
955
    """
    Turns the ``token`` and ``argument`` into a special token that
956
    will be caught by the `PreprocessorToken`-parser.
957

958
959
    This function is a support function that should be used by
    preprocessors to inject preprocessor tokens into the source text.
960
    """
961
962
963
    assert RX_PREPROCESSOR_TOKEN.match(token)
    assert argument.find(BEGIN_TOKEN) < 0
    assert argument.find(END_TOKEN) < 0
964

965
    return BEGIN_TOKEN + token + argument + END_TOKEN
966
967


968
def nil_preprocessor(text: str) -> str:
Eckhart Arnold's avatar
Eckhart Arnold committed
969
    return text
970
971


972
class PreprocessorToken(Parser):
973
    """
974
    Parses tokens that have been inserted by a preprocessor.
975
    
976
    Preprocessors can generate Tokens with the ``make_token``-function.
977
    These tokens start and end with magic characters that can only be
978
979
980
    matched by the PreprocessorToken Parser. Such tokens can be used to
    insert BEGIN - END delimiters at the beginning or ending of a
    quoted block, for example.
981
    """
982

983
984
985
986
    def __init__(self, token: str) -> None:
        assert token and token.isupper()
        assert RX_PREPROCESSOR_TOKEN.match(token)
        super(PreprocessorToken, self).__init__(token)
987

988
    def __call__(self, text: StringView) -> Tuple[Node, StringView]:
989
990
        if text[0:1] == BEGIN_TOKEN:
            end = text.find(END_TOKEN, 1)
991
992
            if end < 0:
                node = Node(self, '').add_error(
993
994
                    'END_TOKEN delimiter missing from preprocessor token. '
                    '(Most likely due to a preprocessor bug!)')  # type: Node
995
996
997
                return node, text[1:]
            elif end == 0:
                node = Node(self, '').add_error(
998
999
                    'Preprocessor-token cannot have zero length. '
                    '(Most likely due to a preprocessor bug!)')
1000
                return node, text[2:]
1001
            elif text.find(BEGIN_TOKEN, 1, end) >= 0:
1002
1003
                node = Node(self, text[len(self.name) + 1:end])
                node.add_error(
1004
1005
1006
                    'Preprocessor-tokens must not be nested or contain '
                    'BEGIN_TOKEN delimiter as part of their argument. '
                    '(Most likely due to a preprocessor bug!)')
1007
1008
1009
1010
1011
1012
1013
1014
                return node, text[end:]
            if text[1:len(self.name) + 1] == self.name:
                return Node(self, text[len(self.name) + 1:end]), \
                       text[end + 1:]
        return None, text


class RegExp(Parser):
Eckhart Arnold's avatar
Eckhart Arnold committed
1015
    """Regular expression parser.
1016
1017
1018
1019
1020
    
    The RegExp-parser parses text that matches a regular expression.
    RegExp can also be considered as the "atomic parser", because all
    other parsers delegate part of the parsing job to other parsers,
    but do not match text directly.
Eckhart Arnold's avatar
Eckhart Arnold committed
1021
1022
1023
1024
1025
1026
1027
1028

    Example:
    >>> word = RegExp(r'\w+')
    >>> Grammar(word)("Haus").content()
    'Haus'

    EBNF-Notation:  `/ ... /`
    EBNF-Example:   `word = /\w+/`
1029
    """
1030
1031

    def __init__(self, regexp, name: str = '') -> None:
1032
        super(RegExp, self).__init__(name)
1033
1034
1035
        self.regexp = re.compile(regexp) if isinstance(regexp, str) else regexp

    def __deepcopy__(self, memo):
Eckhart Arnold's avatar
Eckhart Arnold committed
1036
        # `regex` supports deep copies, but not `re`
1037
        try:
Eckhart Arnold's avatar
Eckhart Arnold committed
1038
            regexp = copy.deepcopy(self.regexp, memo)
1039
1040
        except TypeError:
            regexp = self.regexp.pattern
1041
        return RegExp(regexp, self.name)
1042

1043
1044
    def __call__(self, text: StringView) -> Tuple[Node, StringView]:
        match = text[0:1] != BEGIN_TOKEN and sv_match(self.regexp, text)  # ESC starts a preprocessor token.
1045
        if match:
1046
            end = sv_index(match.end(), text)
1047
1048
1049
            return Node(self, text[:end]), text[end:]
        return None, text

1050
    def __repr__(self):
1051
        return '/%s/' % self.regexp.pattern
1052

1053

1054
1055
class Whitespace(RegExp):
    assert WHITESPACE_PTYPE == ":Whitespace"
1056
1057


1058
class RE(Parser):
1059
1060
    """
    Regular Expressions with optional leading or trailing whitespace.
1061
1062
    
    The RE-parser parses pieces of text that match a given regular
Eckhart Arnold's avatar
Eckhart Arnold committed
1063
    expression. Other than the ``RegExp``-Parser it can also skip
1064
1065
    "implicit whitespace" before or after the matched text.
    
Eckhart Arnold's avatar
Eckhart Arnold committed
1066
1067
1068
1069
1070
    The whitespace is in turn defined by a regular expression. It should
    be made sure that this expression also matches the empty string,
    e.g. use r'\s*' or r'[\t ]+', but not r'\s+'. If the respective
    parameters in the constructor are set to ``None`` the default
    whitespace expression from the Grammar object will be used.
Eckhart Arnold's avatar
Eckhart Arnold committed
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083