21.10.2021, 9:00 - 11:00: Due to updates GitLab may be unavailable for some minutes between 09:00 and 11:00.

parser.py 81 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
from DHParser.toolkit import is_logging, log_dir, logfile_basename, escape_re, sane_parser_name, load_if_file, \
        re, typing
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
67
68
from DHParser.syntaxtree import Node, TransformationFunc, ParserBase, WHITESPACE_PTYPE, TOKEN_PTYPE, \
    ZOMBIE_PARSER
69
from DHParser.error import Error, is_error, has_errors, linebreaks, line_col
70
from typing import Any, Callable, cast, Dict, Iterator, List, Set, Tuple, Union, Optional
Eckhart Arnold's avatar
Eckhart Arnold committed
71

72
__all__ = ('PreprocessorFunc',
73
           'HistoryRecord',
Eckhart Arnold's avatar
Eckhart Arnold committed
74
           'Parser',
75
           'Grammar',
76
77
78
           'RX_PREPROCESSOR_TOKEN',
           'BEGIN_TOKEN',
           'END_TOKEN',
Eckhart Arnold's avatar
Eckhart Arnold committed
79
           'make_token',
80
81
           'nil_preprocessor',
           'PreprocessorToken',
Eckhart Arnold's avatar
Eckhart Arnold committed
82
83
84
85
           'RegExp',
           'RE',
           'Token',
           'mixin_comment',
86
87
88
           # 'UnaryOperator',
           # 'NaryOperator',
           'Synonym',
89
           'Option',
Eckhart Arnold's avatar
Eckhart Arnold committed
90
91
           'ZeroOrMore',
           'OneOrMore',
92
           'Series',
Eckhart Arnold's avatar
Eckhart Arnold committed
93
94
95
96
97
98
99
           'Alternative',
           'FlowOperator',
           'Required',
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
100
101
102
           'last_value',
           'counterpart',
           'accumulate',
Eckhart Arnold's avatar
Eckhart Arnold committed
103
104
105
106
           'Capture',
           'Retrieve',
           'Pop',
           'Forward',
107
           'Compiler',
108
           'compile_source')
Eckhart Arnold's avatar
Eckhart Arnold committed
109

110

111

112
113
114
115
116
117
118
########################################################################
#
# Grammar and parsing infrastructure
#
########################################################################


119
PreprocessorFunc = Union[Callable[[str], str], partial]
120
121


122
123
124
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
125
MAX_DROPOUTS = 3  # type: int
126
# stop trying to recover parsing after so many errors
127

128

129
130
131
132
133
134
135
136
137
138
139
140
141
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.
    """
142
    __slots__ = ('call_stack', 'node', 'remaining', 'line_col')
143
144
145
146
147

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

148
    def __init__(self, call_stack: List['Parser'], node: Node, remaining: int) -> None:
149
        # copy call stack, dropping uninformative Forward-Parsers
Eckhart Arnold's avatar
Eckhart Arnold committed
150
        self.call_stack = [p for p in call_stack if p.ptype != ":Forward"]  # type: List['Parser']
151
152
        self.node = node                # type: Node
        self.remaining = remaining      # type: int
153
154
        self.line_col = (1, 1)          # type: Tuple[int, int]
        if call_stack:
155
156
157
158
            grammar = call_stack[-1].grammar
            document = grammar.document__
            lbreaks = grammar.document_lbreaks__
            self.line_col = line_col(lbreaks, len(document) - remaining)
159
160
161
162

    def __str__(self):
        return 'line %i, column %i:  %s  "%s"' % \
               (self.line_col[0], self.line_col[1], self.stack, str(self.node))
163

164
    def err_msg(self) -> str:
165
        return self.ERROR + ": " + "; ".join(str(e) for e in self.node._errors).replace('\n', '\\')
166

167
    @property
168
    def stack(self) -> str:
169
        return "->".join((p.repr if p.ptype == ':RegExp' else p.name or p.ptype)
170
                         for p in self.call_stack)
171
172

    @property
173
    def status(self) -> str:
174
        return self.FAIL if self.node is None else \
175
            self.err_msg() if has_errors(self.node._errors) else self.MATCH
176
177

    @property
178
    def extent(self) -> slice:
179
        return (slice(-self.remaining - len(self.node), -self.remaining) if self.node
180
                else slice(-self.remaining, None))
181

182
    @staticmethod
Eckhart Arnold's avatar
Eckhart Arnold committed
183
    def last_match(history: List['HistoryRecord']) -> Union['HistoryRecord', None]:
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
        """
        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
199
    def most_advanced_match(history: List['HistoryRecord']) -> Union['HistoryRecord', None]:
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
        """
        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


219
def add_parser_guard(parser_func):
220
221
222
223
224
    """
    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.
    """
225
    def guarded_call(parser: 'Parser', text: StringView) -> Tuple[Node, StringView]:
di68kap's avatar
di68kap committed
226
227
228
229
230
231
232
233
234
235
236
        # 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
237

238
        try:
239
            location = len(text)    # mind that location is always the distance to the end
240
241
            grammar = parser.grammar  # grammar may be 'None' for unconnected parsers!

242
243
            if grammar.last_rb__loc__ <= location:
                grammar.rollback_to__(location)
244

245
246
247
            # if location has already been visited by the current parser,
            # return saved result
            if location in parser.visited:
di68kap's avatar
di68kap committed
248
                return parser.visited[location]
249

Eckhart Arnold's avatar
Eckhart Arnold committed
250
251
            if grammar.history_tracking__:
                grammar.call_stack__.append(parser)
252
                grammar.moving_forward__ = True
Eckhart Arnold's avatar
Eckhart Arnold committed
253

254
255
256
257
258
259
            # break left recursion at the maximum allowed depth
            if grammar.left_recursion_handling__:
                if parser.recursion_counter.setdefault(location, 0) > LEFT_RECURSION_DEPTH:
                    grammar.recursion_locations__.add(location)
                    return None, text
                parser.recursion_counter[location] += 1
260
261
262
263

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

264
265
266
            if grammar.left_recursion_handling__:
                parser.recursion_counter[location] -= 1

267
            if node is None:
268
                # retrieve an earlier match result (from left recursion) if it exists
269
270
                if location in grammar.recursion_locations__:
                    if location in parser.visited:
di68kap's avatar
di68kap committed
271
                        node, rest = parser.visited[location]
272
                        # TODO: maybe add a warning about occurrence of left-recursion here?
273
274
275
                    # 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__:
276
                    # otherwise also cache None-results
di68kap's avatar
di68kap committed
277
                    parser.visited[location] = (None, rest)
Eckhart Arnold's avatar
Eckhart Arnold committed
278
279
            elif (grammar.last_rb__loc__ > location
                  and (grammar.memoization__ or location in grammar.recursion_locations__)):
280
281
282
283
                # - 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
284
                parser.visited[location] = (node, rest)
285

286
            if grammar.history_tracking__:
287
                # don't track returning parsers except in case an error has occurred
288
                remaining = len(rest)
289
                if grammar.moving_forward__ or (node and node._errors):
290
                    record = HistoryRecord(grammar.call_stack__, node, remaining)
291
                    grammar.history__.append(record)
292
                    # print(record.stack, record.status, rest[:20].replace('\n', '|'))
293
                grammar.moving_forward__ = False
294
                grammar.call_stack__.pop()
295

296
        except RecursionError:
297
            node = Node(None, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
298
299
            node.add_error("maximum recursion depth of parser reached; "
                           "potentially due to too many errors!")
300
            rest = EMPTY_STRING_VIEW
301
302
303
304
305
306

        return node, rest

    return guarded_call


Eckhart Arnold's avatar
Eckhart Arnold committed
307
class ParserMetaClass(abc.ABCMeta):
308
309
310
311
312
    """
    ParserMetaClass adds a wrapper to the __call__ method of parser
    objects during initialization that takes care of memoizing,
    left recursion and tracing.
    """
313
    def __init__(cls, name, bases, attrs):
314
315
        guarded_parser_call = add_parser_guard(cls.__call__)
        # The following check is necessary for classes that don't override
316
317
318
319
320
321
322
        # 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
323
class Parser(ParserBase, metaclass=ParserMetaClass):
324
325
326
327
    """
    (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.
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344

    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
345
346

    Parser objects are callable and parsing is done by calling a parser
Eckhart Arnold's avatar
Eckhart Arnold committed
347
348
349
350
351
352
    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
353
    `ZeroOrMore` or `Option`). If `i > 0` then the parser has "moved
Eckhart Arnold's avatar
Eckhart Arnold committed
354
355
356
357
358
359
    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.
360
361

    Attributes:
362
        visited:  Mapping of places this parser has already been to
di68kap's avatar
di68kap committed
363
                during the current parsing process onto the results the
364
365
366
367
368
369
370
371
                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.

372
        cycle_detection:  The apply()-method uses this variable to make
373
374
375
                sure that one and the same function will not be applied
                (recursively) a second time, if it has already been
                applied to this parser.
376
377
378

        grammar:  A reference to the Grammar object to which the parser
                is attached.
379
380
    """

381
382
    ApplyFunc = Callable[['Parser'], None]

383
384
    def __init__(self, name: str = '') -> None:
        # assert isinstance(name, str), str(name)
Eckhart Arnold's avatar
Eckhart Arnold committed
385
        super(Parser, self).__init__(name)
386
        self._grammar = None  # type: 'Grammar'
387
388
        self.reset()

Eckhart Arnold's avatar
Eckhart Arnold committed
389
    def __deepcopy__(self, memo):
Eckhart Arnold's avatar
Eckhart Arnold committed
390
391
392
393
394
395
        """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.
        """
396
397
        return self.__class__(self.name)

398
    def reset(self):
Eckhart Arnold's avatar
Eckhart Arnold committed
399
400
401
        """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
402
        self.visited = dict()            # type: Dict[int, Tuple[Node, StringView]]
403
        self.recursion_counter = dict()  # type: Dict[int, int]
404
        self.cycle_detection = set()     # type: Set[Callable]
405
        return self
406

407
    def __call__(self, text: StringView) -> Tuple[Node, StringView]:
Eckhart Arnold's avatar
Eckhart Arnold committed
408
409
410
        """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."""
411
412
        return None, text  # default behaviour: don't match

413
    def __add__(self, other: 'Parser') -> 'Series':
414
415
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
416
        return Series(self, other)
417

418
    def __or__(self, other: 'Parser') -> 'Alternative':
419
420
421
        """The | operator generates an alternative parser that applies
        the first parser and, if that does not match, the second parser.
        """
422
423
        return Alternative(self, other)

424
    @property
425
    def grammar(self) -> 'Grammar':
426
427
428
        return self._grammar

    @grammar.setter
429
    def grammar(self, grammar: 'Grammar'):
Eckhart Arnold's avatar
Eckhart Arnold committed
430
431
432
433
434
435
        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!"
436
437

    def _grammar_assigned_notifier(self):
438
439
        """A function that notifies the parser object that it has been
        assigned to a grammar."""
440
441
        pass

442
    def apply(self, func: ApplyFunc):
443
444
        """
        Applies function `func(parser)` recursively to this parser and all
Eckhart Arnold's avatar
Eckhart Arnold committed
445
        descendant parsers if any exist. The same function can never
446
447
448
449
450
451
452
453
454
455
        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


456
def mixin_comment(whitespace: str, comment: str) -> str:
457
458
    """
    Returns a regular expression that merges comment and whitespace
459
460
461
462
463
464
465
466
467
468
469
    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


470
class Grammar:
471
    """
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
    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
487
        >>> number_parser("3.1416").content()
488
489
        '3.1416'

Eckhart Arnold's avatar
Eckhart Arnold committed
490
    Collecting the parsers that define a grammar in a descendant class of
491
492
493
494
495
496
497
    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
Eckhart Arnold's avatar
Eckhart Arnold committed
498
499
       assign them to class variables, which results in better
       readability of the Python code.
500

Eckhart Arnold's avatar
Eckhart Arnold committed
501
502
503
    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.
504
505
506
507
508
509

    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.
510
511

    Example:
512

513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
        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,
531
532
533
    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.
534
535
536
537
538
539
540
541
542
543
544
    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.

545
    Grammar classes contain a few special class fields for implicit
546
547
    whitespace and comments that should be overwritten, if the defaults
    (no comments, horizontal right aligned whitespace) don't fit:
548
549
    COMMENT__:  regular expression string for matching comments
    WSP__:   regular expression for whitespace and comments
Eckhart Arnold's avatar
Eckhart Arnold committed
550

551
552
    wspL__:  regular expression string for left aligned whitespace,
             which either equals WSP__ or is empty.
Eckhart Arnold's avatar
Eckhart Arnold committed
553

554
555
    wspR__:  regular expression string for right aligned whitespace,
             which either equals WSP__ or is empty.
Eckhart Arnold's avatar
Eckhart Arnold committed
556

557
558
559
560
    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.
Eckhart Arnold's avatar
Eckhart Arnold committed
561

562
563
564
565
566
567
568
569
570
    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

Eckhart Arnold's avatar
Eckhart Arnold committed
571
        history_tracking__:  A flag indicating that the parsing history shall
572
573
574
575
576
577
578
                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'.
Eckhart Arnold's avatar
Eckhart Arnold committed
579

580
581
582
583
584
585
586
587
588
589
590
        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.
Eckhart Arnold's avatar
Eckhart Arnold committed
591

592
593
594
595
596
        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
                initialized if `history_tracking__` is true.

597
598
        _reversed__:  the same text in reverse order - needed by the `Lookbehind'-
                parsers.
Eckhart Arnold's avatar
Eckhart Arnold committed
599

600
601
602
        variables__:  A mapping for variable names to a stack of their respective
                string values - needed by the `Capture`-, `Retrieve`- and `Pop`-
                parsers.
Eckhart Arnold's avatar
Eckhart Arnold committed
603

604
605
606
607
608
        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.
Eckhart Arnold's avatar
Eckhart Arnold committed
609

610
611
612
613
614
615
616
617
        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)`.
Eckhart Arnold's avatar
Eckhart Arnold committed
618

619
620
621
622
        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.
Eckhart Arnold's avatar
Eckhart Arnold committed
623

624
625
626
        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.
Eckhart Arnold's avatar
Eckhart Arnold committed
627

628
629
630
631
632
        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`)
Eckhart Arnold's avatar
Eckhart Arnold committed
633

634
635
636
637
638
        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`)
Eckhart Arnold's avatar
Eckhart Arnold committed
639

640
641
642
        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
643
644
645
646
647
                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.
648
    """
649
650
    root__ = None  # type: Union[Parser, None]
    # root__ must be overwritten with the root-parser by grammar subclass
Eckhart Arnold's avatar
Eckhart Arnold committed
651
    parser_initialization__ = "pending"  # type: str
652
    # some default values
Eckhart Arnold's avatar
Eckhart Arnold committed
653
654
655
656
    COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    WSP__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
    wspL__ = ''     # type: str
    wspR__ = WSP__  # type: str
657

658

659
    @classmethod
660
    def _assign_parser_names__(cls):
661
662
        """
        Initializes the `parser.name` fields of those
663
664
        Parser objects that are directly assigned to a class field with
        the field's name, e.g.
665
            class Grammar(Grammar):
666
667
668
                ...
                symbol = RE('(?!\\d)\\w+')
        After the call of this method symbol.name == "symbol"
669
670
671
672
        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()``
673
674
675
676

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

677
678
679
680
681
        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. 
682
        """
683
684
685
686
687
        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:
688
689
690
                        parser._name = entry
                    if (isinstance(parser, Forward) and (not parser.parser._name)):
                        parser.parser._name = entry
691
            cls.parser_initialization__ = "done"
692

693

Eckhart Arnold's avatar
Eckhart Arnold committed
694
695
696
    def __init__(self, root: Parser=None) -> None:
        # if not hasattr(self.__class__, 'parser_initialization__'):
        #     self.__class__.parser_initialization__ = "pending"
697
698
699
700
        # if not hasattr(self.__class__, 'wspL__'):
        #     self.wspL__ = ''
        # if not hasattr(self.__class__, 'wspR__'):
        #     self.wspR__ = ''
701
702
703
704
        self.all_parsers__ = set()             # type: Set[Parser]
        self._dirty_flag__ = False             # type: bool
        self.history_tracking__ = False        # type: bool
        self.memoization__ = True              # type: bool
705
        self.left_recursion_handling__ = True  # type: bool
706
        self._reset__()
707

Eckhart Arnold's avatar
Eckhart Arnold committed
708
        # prepare parsers in the class, first
709
        self._assign_parser_names__()
710

Eckhart Arnold's avatar
Eckhart Arnold committed
711
712
713
714
        # 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
715
        self.root__ = copy.deepcopy(root) if root else copy.deepcopy(self.__class__.root__)
716

717
        if self.wspL__:
Eckhart Arnold's avatar
Eckhart Arnold committed
718
            self.wsp_left_parser__ = Whitespace(self.wspL__)  # type: ParserBase
719
            self.wsp_left_parser__.grammar = self
720
            self.all_parsers__.add(self.wsp_left_parser__)  # don't you forget about me...
721
722
723
        else:
            self.wsp_left_parser__ = ZOMBIE_PARSER
        if self.wspR__:
Eckhart Arnold's avatar
Eckhart Arnold committed
724
            self.wsp_right_parser__ = Whitespace(self.wspR__)  # type: ParserBase
725
            self.wsp_right_parser__.grammar = self
726
            self.all_parsers__.add(self.wsp_right_parser__)  # don't you forget about me...
727
728
        else:
            self.wsp_right_parser__ = ZOMBIE_PARSER
729
        self.root__.apply(self._add_parser__)
730

731

732
    def __getitem__(self, key):
733
734
735
        try:
            return self.__dict__[key]
        except KeyError:
736
737
            parser_template = getattr(self, key, None)
            if parser_template:
Eckhart Arnold's avatar
Eckhart Arnold committed
738
                # add parser to grammar object on the fly...
739
                parser = copy.deepcopy(parser_template)
740
                parser.apply(self._add_parser__)
741
                # assert self[key] == parser
Eckhart Arnold's avatar
Eckhart Arnold committed
742
                return self[key]
743
            raise KeyError('Unknown parser "%s" !' % key)
744

745

746
    def _reset__(self):
747
748
        self.document__ = EMPTY_STRING_VIEW   # type: StringView
        self._reversed__ = EMPTY_STRING_VIEW  # type: StringView
749
        self.document_lbreaks__ = []          # type: List[int]
750
        # variables stored and recalled by Capture and Retrieve parsers
751
752
        self.variables__ = dict()     # type: Dict[str, List[str]]
        self.rollback__ = []          # type: List[Tuple[int, Callable]]
753
        self.last_rb__loc__ = -1  # type: int
754
        # support for call stack tracing
755
        self.call_stack__ = []        # type: List[Parser]
756
        # snapshots of call stacks
757
        self.history__ = []           # type: List[HistoryRecord]
758
        # also needed for call stack tracing
759
760
        self.moving_forward__ = False  # type: bool
        self.recursion_locations__ = set()  # type: Set[int]
761

762

Eckhart Arnold's avatar
Eckhart Arnold committed
763
    @property
Eckhart Arnold's avatar
Eckhart Arnold committed
764
    def reversed__(self) -> StringView:
Eckhart Arnold's avatar
Eckhart Arnold committed
765
        if not self._reversed__:
766
            self._reversed__ = StringView(self.document__.text[::-1])
Eckhart Arnold's avatar
Eckhart Arnold committed
767
768
769
        return self._reversed__


770
    def _add_parser__(self, parser: Parser) -> None:
771
772
        """
        Adds the particular copy of the parser object to this
773
        particular instance of Grammar.
774
        """
775
        if parser.name:
Eckhart Arnold's avatar
Eckhart Arnold committed
776
777
778
            # 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__), \
779
780
                ('Cannot add parser "%s" because a field with the same name '
                 'already exists in grammar object!' % parser.name)
781
            setattr(self, parser.name, parser)
782
        self.all_parsers__.add(parser)
783
784
        parser.grammar = self

785

Eckhart Arnold's avatar
Eckhart Arnold committed
786
    def __call__(self, document: str, start_parser="root__") -> Node:
787
788
        """
        Parses a document with with parser-combinators.
789
790
791

        Args:
            document (str): The source text to be parsed.
792
793
794
            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.)
795
796
797
        Returns:
            Node: The root node ot the parse tree.
        """
798
        # assert isinstance(document, str), type(document)
799
800
        if self.root__ is None:
            raise NotImplementedError()
801
        if self._dirty_flag__:
802
            self._reset__()
803
            for parser in self.all_parsers__:
804
805
                parser.reset()
        else:
806
            self._dirty_flag__ = True
807
        self.history_tracking__ = is_logging()
808
        self.document__ = StringView(document)
809
        self.document_lbreaks__ = linebreaks(document) if self.history_tracking__ else []
810
        self.last_rb__loc__ = len(self.document__) + 1  # rollback location
Eckhart Arnold's avatar
Eckhart Arnold committed
811
        parser = self[start_parser] if isinstance(start_parser, str) else start_parser
812
813
        assert parser.grammar == self, "Cannot run parsers from a different grammar object!" \
                                       " %s vs. %s" % (str(self), str(parser.grammar))
814
        stitches = []  # type: List[Node]
815
        rest = self.document__
816
817
        if not rest:
            result, ignore = parser(rest)
di68kap's avatar
di68kap committed
818
819
820
            if result is None:
                result = Node(None, '')
                result.add_error('Parser "%s" did not match empty document.' % str(parser))
821
822
823
824
825
826
        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:
827
828
829
830
                    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__)))
831
832
833
                else:
                    stitches.append(result)
                    error_msg = "Parser stopped before end" + \
834
835
                                (("! trying to recover" +
                                  (" but stopping history recording at this point."
836
                                   if self.history_tracking__ else "..."))
837
838
839
840
                                 if len(stitches) < MAX_DROPOUTS
                                 else " too often! Terminating parser.")
                stitches.append(Node(None, skip))
                stitches[-1].add_error(error_msg)
841
                if self.history_tracking__:
842
                    # some parsers may have matched and left history records with nodes != None.
843
                    # Because these are not connected to the stitched root node, their pos-
844
845
                    # properties will not be initialized by setting the root node's pos property
                    # to zero. Therefore, their pos properties need to be initialized here
846
                    for record in self.history__:
847
848
                        if record.node and record.node._pos < 0:
                            record.node.pos = 0
849
850
851
                    record = HistoryRecord(self.call_stack__.copy(), stitches[-1], len(rest))
                    self.history__.append(record)
                    self.history_tracking__ = False
852
853
854
855
        if stitches:
            if rest:
                stitches.append(Node(None, rest))
            result = Node(None, tuple(stitches))
856
        if any(self.variables__.values()):
Eckhart Arnold's avatar
Eckhart Arnold committed
857
858
859
860
861
862
863
864
865
866
867
            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)
868
        result.pos = 0  # calculate all positions
869
        # result.collect_errors(self.document__)
870
871
        return result

872

873
    def push_rollback__(self, location, func):
874
875
        """
        Adds a rollback function that either removes or re-adds
876
877
878
879
        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.
        """
880
        self.rollback__.append((location, func))
881
        # print("push:  line %i, col %i" % line_col(self.document__, len(self.document__) - location))
882
883
        self.last_rb__loc__ = location

884

885
    def rollback_to__(self, location):
886
887
        """
        Rolls back the variable stacks (`self.variables`) to its
888
889
        state at an earlier location in the parsed document.
        """
890
        # print("rollback:  line %i, col %i" % line_col(self.document__, len(self.document__) - location))
891
892
        while self.rollback__ and self.rollback__[-1][0] <= location:
            loc, rollback_func = self.rollback__.pop()
893
894
895
896
            # 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__))
897
            rollback_func()
898
            # print("rb to:  line %i, col %i" % line_col(self.document__, len(self.document__) - loc))
899
900
901
        self.last_rb__loc__ == self.rollback__[-1][0] if self.rollback__ \
            else (len(self.document__) + 1)

902

903
    def log_parsing_history__(self, log_file_name: str = '') -> None:
904
905
        """
        Writes a log of the parsing history of the most recently parsed
906
907
908
        document. 
        """
        def prepare_line(record):
909
            excerpt = self.document__.text.__getitem__(record.extent)[:25].replace('\n', '\\n')
910
            excerpt = "'%s'" % excerpt if len(excerpt) < 25 else "'%s...'" % excerpt
911
            return record.stack, record.status, excerpt
912
913

        def write_log(history, log_name):
Eckhart Arnold's avatar
Eckhart Arnold committed
914
            path = os.path.join(log_dir(), log_name + "_parser.log")
Eckhart Arnold's avatar
Eckhart Arnold committed
915
916
917
            if os.path.exists(path):
                os.remove(path)
                print('WARNING: Log-file "%s" already existed and was deleted.' % path)
918
919
920
921
            if history:
                with open(path, "w", encoding="utf-8") as f:
                    f.write("\n".join(history))

922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
        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')
939
940


Eckhart Arnold's avatar
Eckhart Arnold committed
941
def dsl_error_msg(parser: Parser, error_str: str) -> str:
942
943
    """
    Returns an error message for errors in the parser configuration,
944
945
946
    e.g. errors that result in infinite loops.

    Args:
Eckhart Arnold's avatar
Eckhart Arnold committed
947
        parser (Parser):  The parser where the error was noticed. Note
948
            that this is not necessarily the parser that caused the
949
            error but only where the error became apparent.
950
951
952
953
954
        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.
    """
955
    msg = ["DSL parser specification error:", error_str, 'Caught by parser "%s".' % str(parser)]
956
957
    if parser.grammar.history__:
        msg.extend(["\nCall stack:", parser.grammar.history__[-1].stack])
958
959
960
961
962
    else:
        msg.extend(["\nEnable history tracking in Grammar object to display call stack."])
    return " ".join(msg)


963

964
965
966
967
968
969
970
########################################################################
#
# Token and Regular Expression parser classes (i.e. leaf classes)
#
########################################################################


971
972
973
RX_PREPROCESSOR_TOKEN = re.compile('\w+')
BEGIN_TOKEN = '\x1b'
END_TOKEN = '\x1c'
974
975


976
def make_token(token: str, argument: str = '') -> str:
977
978
    """
    Turns the ``token`` and ``argument`` into a special token that
979
    will be caught by the `PreprocessorToken`-parser.
980

981
982
    This function is a support function that should be used by
    preprocessors to inject preprocessor tokens into the source text.
983
    """
984
985
986
    assert RX_PREPROCESSOR_TOKEN.match(token)
    assert argument.find(BEGIN_TOKEN) < 0
    assert argument.find(END_TOKEN) < 0
987

988
    return BEGIN_TOKEN + token + argument + END_TOKEN
989
990


991
def nil_preprocessor(text: str) -> str:
Eckhart Arnold's avatar
Eckhart Arnold committed
992
    return text
993
994


995
class PreprocessorToken(Parser):
996
    """
997
    Parses tokens that have been inserted by a preprocessor.
998
    
999
    Preprocessors can generate Tokens with the ``make_token``-function.
1000
    These tokens start and end with magic characters that can only be
1001
1002
1003
    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.
1004
    """
1005

1006
1007
1008
1009
    def __init__(self, token: str) -> None:
        assert token and token.isupper()
        assert RX_PREPROCESSOR_TOKEN.match(token)
        super(PreprocessorToken, self).__init__(token)
1010

1011
    def __call__(self, text: StringView) -> Tuple[Node, StringView]:
1012
1013
        if text[0:1] == BEGIN_TOKEN:
            end = text.find(END_TOKEN, 1)
1014
1015
            if end < 0:
                node = Node(self, '').add_error(
1016
1017
                    'END_TOKEN delimiter missing from preprocessor token. '
                    '(Most likely due to a preprocessor bug!)')  # type: Node
1018
1019
1020
                return node, text[1:]
            elif end == 0:
                node = Node(self, '').add_error(
1021
1022
                    'Preprocessor-token cannot have zero length. '
                    '(Most likely due to a preprocessor bug!)')
1023
                return node, text[2:]
1024
            elif text.find(BEGIN_TOKEN, 1, end) >= 0:
1025
1026
                node = Node(self, text[len(self.name) + 1:end])
                node.add_error(
1027
1028
1029
                    'Preprocessor-tokens must not be nested or contain '
                    'BEGIN_TOKEN delimiter as part of their argument. '
                    '(Most likely due to a preprocessor bug!)')