parser.py 84.3 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
from functools import partial
63
from typing import Any, Callable, cast, Dict, List, Set, Tuple, Union
Eckhart Arnold's avatar
Eckhart Arnold committed
64

65
from DHParser.error import Error, is_error, has_errors, linebreaks, line_col
66
67
68
69
70
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
from DHParser.syntaxtree import Node, TransformationFunc, ParserBase, WHITESPACE_PTYPE, \
    TOKEN_PTYPE, ZOMBIE_PARSER
from DHParser.toolkit import is_logging, log_dir, logfile_basename, escape_re, sane_parser_name, \
    load_if_file, re
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
           'Alternative',
94
95
           'AllOf',
           'SomeOf',
Eckhart Arnold's avatar
Eckhart Arnold committed
96
97
98
99
100
101
           'FlowOperator',
           'Required',
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
102
103
104
           'last_value',
           'counterpart',
           'accumulate',
Eckhart Arnold's avatar
Eckhart Arnold committed
105
106
107
108
           'Capture',
           'Retrieve',
           'Pop',
           'Forward',
109
           'Compiler',
110
           'compile_source')
Eckhart Arnold's avatar
Eckhart Arnold committed
111

112

113

114
115
116
117
118
119
120
########################################################################
#
# Grammar and parsing infrastructure
#
########################################################################


121
PreprocessorFunc = Union[Callable[[str], str], partial]
122
123


124
125
126
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
127
MAX_DROPOUTS = 3  # type: int
128
# stop trying to recover parsing after so many errors
129

130

131
132
133
134
135
136
137
138
139
140
141
142
143
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.
    """
144
    __slots__ = ('call_stack', 'node', 'remaining', 'line_col')
145
146
147
148
149

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

150
    def __init__(self, call_stack: List['Parser'], node: Node, remaining: int) -> None:
151
        # copy call stack, dropping uninformative Forward-Parsers
Eckhart Arnold's avatar
Eckhart Arnold committed
152
        self.call_stack = [p for p in call_stack if p.ptype != ":Forward"]  # type: List['Parser']
153
154
        self.node = node                # type: Node
        self.remaining = remaining      # type: int
155
156
        self.line_col = (1, 1)          # type: Tuple[int, int]
        if call_stack:
157
158
159
160
            grammar = call_stack[-1].grammar
            document = grammar.document__
            lbreaks = grammar.document_lbreaks__
            self.line_col = line_col(lbreaks, len(document) - remaining)
161
162
163
164

    def __str__(self):
        return 'line %i, column %i:  %s  "%s"' % \
               (self.line_col[0], self.line_col[1], self.stack, str(self.node))
165

166
    def err_msg(self) -> str:
167
        return self.ERROR + ": " + "; ".join(str(e) for e in self.node._errors).replace('\n', '\\')
168

169
    @property
170
    def stack(self) -> str:
171
        return "->".join((p.repr if p.ptype == ':RegExp' else p.name or p.ptype)
172
                         for p in self.call_stack)
173
174

    @property
175
    def status(self) -> str:
176
        return self.FAIL if self.node is None else \
177
            self.err_msg() if has_errors(self.node._errors) else self.MATCH
178
179

    @property
180
    def extent(self) -> slice:
181
        return (slice(-self.remaining - len(self.node), -self.remaining) if self.node
182
                else slice(-self.remaining, None))
183

184
    @staticmethod
Eckhart Arnold's avatar
Eckhart Arnold committed
185
    def last_match(history: List['HistoryRecord']) -> Union['HistoryRecord', None]:
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
        """
        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
201
    def most_advanced_match(history: List['HistoryRecord']) -> Union['HistoryRecord', None]:
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
        """
        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


221
def add_parser_guard(parser_func):
222
223
224
225
226
    """
    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.
    """
227
    def guarded_call(parser: 'Parser', text: StringView) -> Tuple[Node, StringView]:
di68kap's avatar
di68kap committed
228
229
230
231
232
233
234
235
236
237
238
        # 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
239

240
        try:
241
            location = len(text)    # mind that location is always the distance to the end
242
243
            grammar = parser.grammar  # grammar may be 'None' for unconnected parsers!

244
245
            if grammar.last_rb__loc__ <= location:
                grammar.rollback_to__(location)
246

247
248
249
            # if location has already been visited by the current parser,
            # return saved result
            if location in parser.visited:
di68kap's avatar
di68kap committed
250
                return parser.visited[location]
251

Eckhart Arnold's avatar
Eckhart Arnold committed
252
253
            if grammar.history_tracking__:
                grammar.call_stack__.append(parser)
254
                grammar.moving_forward__ = True
Eckhart Arnold's avatar
Eckhart Arnold committed
255

256
257
258
259
260
261
            # 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
262
263
264
265

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

266
267
268
            if grammar.left_recursion_handling__:
                parser.recursion_counter[location] -= 1

269
            if node is None:
270
                # retrieve an earlier match result (from left recursion) if it exists
271
272
                if location in grammar.recursion_locations__:
                    if location in parser.visited:
di68kap's avatar
di68kap committed
273
                        node, rest = parser.visited[location]
274
                        # TODO: maybe add a warning about occurrence of left-recursion here?
275
276
277
                    # 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__:
278
                    # otherwise also cache None-results
di68kap's avatar
di68kap committed
279
                    parser.visited[location] = (None, rest)
Eckhart Arnold's avatar
Eckhart Arnold committed
280
281
            elif (grammar.last_rb__loc__ > location
                  and (grammar.memoization__ or location in grammar.recursion_locations__)):
282
283
284
285
                # - 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
286
                parser.visited[location] = (node, rest)
287

288
            if grammar.history_tracking__:
289
                # don't track returning parsers except in case an error has occurred
290
                remaining = len(rest)
291
                if grammar.moving_forward__ or (node and node._errors):
292
                    record = HistoryRecord(grammar.call_stack__, node, remaining)
293
                    grammar.history__.append(record)
294
                    # print(record.stack, record.status, rest[:20].replace('\n', '|'))
295
                grammar.moving_forward__ = False
296
                grammar.call_stack__.pop()
297

298
        except RecursionError:
299
            node = Node(None, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
300
301
            node.add_error("maximum recursion depth of parser reached; "
                           "potentially due to too many errors!")
302
            rest = EMPTY_STRING_VIEW
303
304
305
306
307
308

        return node, rest

    return guarded_call


Eckhart Arnold's avatar
Eckhart Arnold committed
309
class ParserMetaClass(abc.ABCMeta):
310
311
312
313
314
    """
    ParserMetaClass adds a wrapper to the __call__ method of parser
    objects during initialization that takes care of memoizing,
    left recursion and tracing.
    """
315
    def __init__(cls, name, bases, attrs):
316
317
        guarded_parser_call = add_parser_guard(cls.__call__)
        # The following check is necessary for classes that don't override
318
319
320
321
322
323
324
        # 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
325
class Parser(ParserBase, metaclass=ParserMetaClass):
326
327
328
329
    """
    (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.
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346

    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
347
348

    Parser objects are callable and parsing is done by calling a parser
Eckhart Arnold's avatar
Eckhart Arnold committed
349
350
351
352
353
354
    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
355
    `ZeroOrMore` or `Option`). If `i > 0` then the parser has "moved
Eckhart Arnold's avatar
Eckhart Arnold committed
356
357
358
359
360
361
    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.
362
363

    Attributes:
364
        visited:  Mapping of places this parser has already been to
di68kap's avatar
di68kap committed
365
                during the current parsing process onto the results the
366
367
368
369
370
371
372
373
                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.

374
        cycle_detection:  The apply()-method uses this variable to make
375
376
377
                sure that one and the same function will not be applied
                (recursively) a second time, if it has already been
                applied to this parser.
378
379
380

        grammar:  A reference to the Grammar object to which the parser
                is attached.
381
382
    """

383
384
    ApplyFunc = Callable[['Parser'], None]

385
386
    def __init__(self, name: str = '') -> None:
        # assert isinstance(name, str), str(name)
Eckhart Arnold's avatar
Eckhart Arnold committed
387
        super(Parser, self).__init__(name)
388
        self._grammar = None  # type: 'Grammar'
389
390
        self.reset()

Eckhart Arnold's avatar
Eckhart Arnold committed
391
    def __deepcopy__(self, memo):
Eckhart Arnold's avatar
Eckhart Arnold committed
392
393
394
395
396
397
        """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.
        """
398
399
        return self.__class__(self.name)

400
    def reset(self):
Eckhart Arnold's avatar
Eckhart Arnold committed
401
402
403
        """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
404
        self.visited = dict()            # type: Dict[int, Tuple[Node, StringView]]
405
        self.recursion_counter = dict()  # type: Dict[int, int]
406
        self.cycle_detection = set()     # type: Set[Callable]
407
        return self
408

409
    def __call__(self, text: StringView) -> Tuple[Node, StringView]:
Eckhart Arnold's avatar
Eckhart Arnold committed
410
411
412
        """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."""
413
414
        return None, text  # default behaviour: don't match

415
    def __add__(self, other: 'Parser') -> 'Series':
416
417
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
418
        return Series(self, other)
419

420
    def __or__(self, other: 'Parser') -> 'Alternative':
421
422
423
        """The | operator generates an alternative parser that applies
        the first parser and, if that does not match, the second parser.
        """
424
425
        return Alternative(self, other)

426
    @property
427
    def grammar(self) -> 'Grammar':
428
429
430
        return self._grammar

    @grammar.setter
431
    def grammar(self, grammar: 'Grammar'):
Eckhart Arnold's avatar
Eckhart Arnold committed
432
433
434
435
436
437
        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!"
438
439

    def _grammar_assigned_notifier(self):
440
441
        """A function that notifies the parser object that it has been
        assigned to a grammar."""
442
443
        pass

444
    def apply(self, func: ApplyFunc):
445
446
        """
        Applies function `func(parser)` recursively to this parser and all
Eckhart Arnold's avatar
Eckhart Arnold committed
447
        descendant parsers if any exist. The same function can never
448
449
450
451
452
453
454
455
456
457
        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


458
def mixin_comment(whitespace: str, comment: str) -> str:
459
460
    """
    Returns a regular expression that merges comment and whitespace
461
462
463
464
465
466
467
468
469
470
471
    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


472
class Grammar:
473
    """
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
    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
489
        >>> number_parser("3.1416").content()
490
491
        '3.1416'

Eckhart Arnold's avatar
Eckhart Arnold committed
492
    Collecting the parsers that define a grammar in a descendant class of
493
494
495
496
497
498
499
    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
500
501
       assign them to class variables, which results in better
       readability of the Python code.
502

Eckhart Arnold's avatar
Eckhart Arnold committed
503
504
505
    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.
506
507
508
509
510
511

    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.
512
513

    Example:
514

515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
        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,
533
534
535
    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.
536
537
538
539
540
541
542
543
544
545
546
    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.

547
    Grammar classes contain a few special class fields for implicit
548
549
    whitespace and comments that should be overwritten, if the defaults
    (no comments, horizontal right aligned whitespace) don't fit:
550
551
    COMMENT__:  regular expression string for matching comments
    WSP__:   regular expression for whitespace and comments
Eckhart Arnold's avatar
Eckhart Arnold committed
552

553
554
    wspL__:  regular expression string for left aligned whitespace,
             which either equals WSP__ or is empty.
Eckhart Arnold's avatar
Eckhart Arnold committed
555

556
557
    wspR__:  regular expression string for right aligned whitespace,
             which either equals WSP__ or is empty.
Eckhart Arnold's avatar
Eckhart Arnold committed
558

559
560
561
562
    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
563

564
565
566
567
568
569
570
571
572
    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
573
        history_tracking__:  A flag indicating that the parsing history shall
574
575
576
577
578
579
580
                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
581

582
583
584
585
586
587
588
589
590
591
592
        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
593

594
595
596
597
598
        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.

599
600
        _reversed__:  the same text in reverse order - needed by the `Lookbehind'-
                parsers.
Eckhart Arnold's avatar
Eckhart Arnold committed
601

602
603
604
        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
605

606
607
608
609
610
        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
611

612
613
614
615
616
617
618
619
        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
620

621
622
623
624
        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
625

626
627
628
        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
629

630
631
632
633
634
        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
635

636
637
638
639
640
        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
641

642
643
644
        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
645
646
647
648
649
                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.
650
    """
651
652
    root__ = None  # type: Union[Parser, None]
    # root__ must be overwritten with the root-parser by grammar subclass
Eckhart Arnold's avatar
Eckhart Arnold committed
653
    parser_initialization__ = "pending"  # type: str
654
    # some default values
Eckhart Arnold's avatar
Eckhart Arnold committed
655
656
657
658
    COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    WSP__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
    wspL__ = ''     # type: str
    wspR__ = WSP__  # type: str
659

660

661
    @classmethod
662
    def _assign_parser_names__(cls):
663
664
        """
        Initializes the `parser.name` fields of those
665
666
        Parser objects that are directly assigned to a class field with
        the field's name, e.g.
667
            class Grammar(Grammar):
668
669
670
                ...
                symbol = RE('(?!\\d)\\w+')
        After the call of this method symbol.name == "symbol"
671
672
673
674
        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()``
675
676
677
678

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

679
680
681
682
683
        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. 
684
        """
685
686
687
688
689
        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:
690
691
692
                        parser._name = entry
                    if (isinstance(parser, Forward) and (not parser.parser._name)):
                        parser.parser._name = entry
693
            cls.parser_initialization__ = "done"
694

Eckhart Arnold's avatar
Eckhart Arnold committed
695
696
697
    def __init__(self, root: Parser=None) -> None:
        # if not hasattr(self.__class__, 'parser_initialization__'):
        #     self.__class__.parser_initialization__ = "pending"
698
699
700
701
        # if not hasattr(self.__class__, 'wspL__'):
        #     self.wspL__ = ''
        # if not hasattr(self.__class__, 'wspR__'):
        #     self.wspR__ = ''
702
703
704
705
        self.all_parsers__ = set()             # type: Set[Parser]
        self._dirty_flag__ = False             # type: bool
        self.history_tracking__ = False        # type: bool
        self.memoization__ = True              # type: bool
706
        self.left_recursion_handling__ = True  # type: bool
707
        self._reset__()
708

Eckhart Arnold's avatar
Eckhart Arnold committed
709
        # prepare parsers in the class, first
710
        self._assign_parser_names__()
711

Eckhart Arnold's avatar
Eckhart Arnold committed
712
713
714
715
        # 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
716
        self.root__ = copy.deepcopy(root) if root else copy.deepcopy(self.__class__.root__)
717

718
        if self.wspL__:
Eckhart Arnold's avatar
Eckhart Arnold committed
719
            self.wsp_left_parser__ = Whitespace(self.wspL__)  # type: ParserBase
720
            self.wsp_left_parser__.grammar = self
721
            self.all_parsers__.add(self.wsp_left_parser__)  # don't you forget about me...
722
723
724
        else:
            self.wsp_left_parser__ = ZOMBIE_PARSER
        if self.wspR__:
Eckhart Arnold's avatar
Eckhart Arnold committed
725
            self.wsp_right_parser__ = Whitespace(self.wspR__)  # type: ParserBase
726
            self.wsp_right_parser__.grammar = self
727
            self.all_parsers__.add(self.wsp_right_parser__)  # don't you forget about me...
728
729
        else:
            self.wsp_right_parser__ = ZOMBIE_PARSER
730
        self.root__.apply(self._add_parser__)
731

732

733
    def __getitem__(self, key):
734
735
736
        try:
            return self.__dict__[key]
        except KeyError:
737
738
            parser_template = getattr(self, key, None)
            if parser_template:
Eckhart Arnold's avatar
Eckhart Arnold committed
739
                # add parser to grammar object on the fly...
740
                parser = copy.deepcopy(parser_template)
741
                parser.apply(self._add_parser__)
742
                # assert self[key] == parser
Eckhart Arnold's avatar
Eckhart Arnold committed
743
                return self[key]
744
            raise KeyError('Unknown parser "%s" !' % key)
745

746

747
    def _reset__(self):
748
749
        self.document__ = EMPTY_STRING_VIEW   # type: StringView
        self._reversed__ = EMPTY_STRING_VIEW  # type: StringView
750
        self.document_lbreaks__ = []          # type: List[int]
751
        # variables stored and recalled by Capture and Retrieve parsers
752
753
        self.variables__ = dict()     # type: Dict[str, List[str]]
        self.rollback__ = []          # type: List[Tuple[int, Callable]]
754
        self.last_rb__loc__ = -1  # type: int
755
        # support for call stack tracing
756
        self.call_stack__ = []        # type: List[Parser]
757
        # snapshots of call stacks
758
        self.history__ = []           # type: List[HistoryRecord]
759
        # also needed for call stack tracing
760
761
        self.moving_forward__ = False  # type: bool
        self.recursion_locations__ = set()  # type: Set[int]
762

763

Eckhart Arnold's avatar
Eckhart Arnold committed
764
    @property
Eckhart Arnold's avatar
Eckhart Arnold committed
765
    def reversed__(self) -> StringView:
Eckhart Arnold's avatar
Eckhart Arnold committed
766
        if not self._reversed__:
767
            self._reversed__ = StringView(self.document__.text[::-1])
Eckhart Arnold's avatar
Eckhart Arnold committed
768
769
770
        return self._reversed__


771
    def _add_parser__(self, parser: Parser) -> None:
772
773
        """
        Adds the particular copy of the parser object to this
774
        particular instance of Grammar.
775
        """
776
        if parser.name:
Eckhart Arnold's avatar
Eckhart Arnold committed
777
778
779
            # 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__), \
780
781
                ('Cannot add parser "%s" because a field with the same name '
                 'already exists in grammar object!' % parser.name)
782
            setattr(self, parser.name, parser)
783
        self.all_parsers__.add(parser)
784
785
        parser.grammar = self

786

Eckhart Arnold's avatar
Eckhart Arnold committed
787
    def __call__(self, document: str, start_parser="root__") -> Node:
788
789
        """
        Parses a document with with parser-combinators.
790
791
792

        Args:
            document (str): The source text to be parsed.
793
794
795
            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.)
796
797
798
        Returns:
            Node: The root node ot the parse tree.
        """
799
        # assert isinstance(document, str), type(document)
800
801
        if self.root__ is None:
            raise NotImplementedError()
802
        if self._dirty_flag__:
803
            self._reset__()
804
            for parser in self.all_parsers__:
805
806
                parser.reset()
        else:
807
            self._dirty_flag__ = True
808
        self.history_tracking__ = is_logging()
809
        self.document__ = StringView(document)
810
        self.document_lbreaks__ = linebreaks(document) if self.history_tracking__ else []
811
        self.last_rb__loc__ = len(self.document__) + 1  # rollback location
Eckhart Arnold's avatar
Eckhart Arnold committed
812
        parser = self[start_parser] if isinstance(start_parser, str) else start_parser
813
814
        assert parser.grammar == self, "Cannot run parsers from a different grammar object!" \
                                       " %s vs. %s" % (str(self), str(parser.grammar))
815
        stitches = []  # type: List[Node]
816
        rest = self.document__
817
818
        if not rest:
            result, ignore = parser(rest)
di68kap's avatar
di68kap committed
819
820
821
            if result is None:
                result = Node(None, '')
                result.add_error('Parser "%s" did not match empty document.' % str(parser))
822
823
824
825
826
827
        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:
828
829
830
831
                    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__)))
832
833
834
                else:
                    stitches.append(result)
                    error_msg = "Parser stopped before end" + \
835
836
                                (("! trying to recover" +
                                  (" but stopping history recording at this point."
837
                                   if self.history_tracking__ else "..."))
838
839
840
841
                                 if len(stitches) < MAX_DROPOUTS
                                 else " too often! Terminating parser.")
                stitches.append(Node(None, skip))
                stitches[-1].add_error(error_msg)
842
                if self.history_tracking__:
843
                    # some parsers may have matched and left history records with nodes != None.
844
                    # Because these are not connected to the stitched root node, their pos-
845
846
                    # properties will not be initialized by setting the root node's pos property
                    # to zero. Therefore, their pos properties need to be initialized here
847
                    for record in self.history__:
848
849
                        if record.node and record.node._pos < 0:
                            record.node.pos = 0
850
851
852
                    record = HistoryRecord(self.call_stack__.copy(), stitches[-1], len(rest))
                    self.history__.append(record)
                    self.history_tracking__ = False
853
854
855
856
        if stitches:
            if rest:
                stitches.append(Node(None, rest))
            result = Node(None, tuple(stitches))
857
        if any(self.variables__.values()):
Eckhart Arnold's avatar
Eckhart Arnold committed
858
859
860
861
862
863
864
865
866
867
868
            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)
869
        result.pos = 0  # calculate all positions
870
        # result.collect_errors(self.document__)
871
872
        return result

873

874
    def push_rollback__(self, location, func):
875
876
        """
        Adds a rollback function that either removes or re-adds
877
878
879
880
        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.
        """
881
        self.rollback__.append((location, func))
882
        # print("push:  line %i, col %i" % line_col(self.document__, len(self.document__) - location))
883
884
        self.last_rb__loc__ = location

885

886
    def rollback_to__(self, location):
887
888
        """
        Rolls back the variable stacks (`self.variables`) to its
889
890
        state at an earlier location in the parsed document.
        """
891
        # print("rollback:  line %i, col %i" % line_col(self.document__, len(self.document__) - location))
892
893
        while self.rollback__ and self.rollback__[-1][0] <= location:
            loc, rollback_func = self.rollback__.pop()
894
895
896
897
            # 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__))
898
            rollback_func()
899
            # print("rb to:  line %i, col %i" % line_col(self.document__, len(self.document__) - loc))
900
901
902
        self.last_rb__loc__ == self.rollback__[-1][0] if self.rollback__ \
            else (len(self.document__) + 1)

903

904
    def log_parsing_history__(self, log_file_name: str = '') -> None:
905
906
        """
        Writes a log of the parsing history of the most recently parsed
907
908
909
        document. 
        """
        def prepare_line(record):
910
            excerpt = self.document__.text.__getitem__(record.extent)[:25].replace('\n', '\\n')
911
            excerpt = "'%s'" % excerpt if len(excerpt) < 25 else "'%s...'" % excerpt
912
            return record.stack, record.status, excerpt
913
914

        def write_log(history, log_name):
Eckhart Arnold's avatar
Eckhart Arnold committed
915
            path = os.path.join(log_dir(), log_name + "_parser.log")
Eckhart Arnold's avatar
Eckhart Arnold committed
916
917
918
            if os.path.exists(path):
                os.remove(path)
                print('WARNING: Log-file "%s" already existed and was deleted.' % path)
919
920
921
922
            if history:
                with open(path, "w", encoding="utf-8") as f:
                    f.write("\n".join(history))

923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
        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')
940
941


Eckhart Arnold's avatar
Eckhart Arnold committed
942
def dsl_error_msg(parser: Parser, error_str: str) -> str:
943
944
    """
    Returns an error message for errors in the parser configuration,
945
946
947
    e.g. errors that result in infinite loops.

    Args:
Eckhart Arnold's avatar
Eckhart Arnold committed
948
        parser (Parser):  The parser where the error was noticed. Note
949
            that this is not necessarily the parser that caused the
950
            error but only where the error became apparent.
951
952
953
954
955
        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.
    """
956
    msg = ["DSL parser specification error:", error_str, 'Caught by parser "%s".' % str(parser)]
957
958
    if parser.grammar.history__:
        msg.extend(["\nCall stack:", parser.grammar.history__[-1].stack])
959
960
961
962
963
    else:
        msg.extend(["\nEnable history tracking in Grammar object to display call stack."])
    return " ".join(msg)


964

965
966
967
968
969
970
971
########################################################################
#
# Token and Regular Expression parser classes (i.e. leaf classes)
#
########################################################################


972
973
974
RX_PREPROCESSOR_TOKEN = re.compile('\w+')
BEGIN_TOKEN = '\x1b'
END_TOKEN = '\x1c'
975
976


977
def make_token(token: str, argument: str = '') -> str:
978
979
    """
    Turns the ``token`` and ``argument`` into a special token that
980
    will be caught by the `PreprocessorToken`-parser.
981

982
983
    This function is a support function that should be used by
    preprocessors to inject preprocessor tokens into the source text.
984
    """
985
986
987
    assert RX_PREPROCESSOR_TOKEN.match(token)
    assert argument.find(BEGIN_TOKEN) < 0
    assert argument.find(END_TOKEN) < 0
988

989
    return BEGIN_TOKEN + token + argument + END_TOKEN
990
991


992
def nil_preprocessor(text: str) -> str:
Eckhart Arnold's avatar
Eckhart Arnold committed
993
    return text
994
995


996
class PreprocessorToken(Parser):
997
    """
998
    Parses tokens that have been inserted by a preprocessor.
999
    
1000
    Preprocessors can generate Tokens with the ``make_token``-function.
1001
    These tokens start and end with magic characters that can only be
1002
1003
1004
    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.
1005
    """
1006

1007
1008
1009
1010
    def __init__(self, token: str) -> None:
        assert token and token.isupper()
        assert RX_PREPROCESSOR_TOKEN.match(token)
        super(PreprocessorToken, self).__init__(token)
1011

1012
    def __call__(self, text: StringView) -> Tuple[Node, StringView]:
1013
1014
        if text[0:1] == BEGIN_TOKEN:
            end = text.find(END_TOKEN, 1)
1015
1016
            if end < 0:
                node = Node(self, '').add_error(
1017
1018
                    'END_TOKEN delimiter missing from preprocessor token. '
                    '(Most likely due to a preprocessor bug!)')  # type: Node
1019
1020
1021
                return node, text[1:]
            elif end == 0:
                node = Node(self, '').add_error(
1022
1023