parser.py 70.9 KB
Newer Older
1
"""parsers.py - parser combinators for for DHParser
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

Copyright 2016  by Eckhart Arnold (arnold@badw.de)
                Bavarian Academy of Sciences an Humanities (badw.de)

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied.  See the License for the specific language governing
permissions and limitations under the License.
Eckhart Arnold's avatar
Eckhart Arnold committed
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46

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


References and Acknowledgements:

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

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

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

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

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

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

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

58

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

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

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

82
__all__ = ('PreprocessorFunc',
83
           'HistoryRecord',
Eckhart Arnold's avatar
Eckhart Arnold committed
84
           'Parser',
85
           'Grammar',
86
87
88
           'RX_PREPROCESSOR_TOKEN',
           'BEGIN_TOKEN',
           'END_TOKEN',
Eckhart Arnold's avatar
Eckhart Arnold committed
89
           'make_token',
90
91
           'nil_preprocessor',
           'PreprocessorToken',
Eckhart Arnold's avatar
Eckhart Arnold committed
92
93
94
95
           'RegExp',
           'RE',
           'Token',
           'mixin_comment',
96
97
98
           # 'UnaryOperator',
           # 'NaryOperator',
           'Synonym',
Eckhart Arnold's avatar
Eckhart Arnold committed
99
100
101
           'Optional',
           'ZeroOrMore',
           'OneOrMore',
102
           'Series',
Eckhart Arnold's avatar
Eckhart Arnold committed
103
104
105
106
107
108
109
           'Alternative',
           'FlowOperator',
           'Required',
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
110
111
112
           'last_value',
           'counterpart',
           'accumulate',
Eckhart Arnold's avatar
Eckhart Arnold committed
113
114
115
116
           'Capture',
           'Retrieve',
           'Pop',
           'Forward',
117
           'Compiler',
118
           'compile_source')
Eckhart Arnold's avatar
Eckhart Arnold committed
119

120

121

122
123
124
125
126
127
128
########################################################################
#
# Grammar and parsing infrastructure
#
########################################################################


129
PreprocessorFunc = Union[Callable[[str], str], partial]
130
131


132
133
134
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
135
MAX_DROPOUTS = 3  # type: int
136
# stop trying to recover parsing after so many errors
137

138

139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
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.
    """
    __slots__ = ('call_stack', 'node', 'remaining')

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

158
159
160
161
    def __init__(self, call_stack: List['Parser'], node: Node, remaining: int) -> None:
        self.call_stack = call_stack    # type: List['Parser']
        self.node = node                # type: Node
        self.remaining = remaining      # type: int
162

163
    def err_msg(self) -> str:
164
165
        return self.ERROR + ": " + "; ".join(self.node._errors).replace('\n', '\\')

166
    @property
167
    def stack(self) -> str:
168
169
        return "->".join((repr(p) if p.ptype == ':RegExp' else p.name or p.ptype)
                         for p in self.call_stack)
170
171

    @property
172
    def status(self) -> str:
173
174
        return self.FAIL if self.node is None else \
            self.err_msg() if self.node._errors else self.MATCH
175
176

    @property
177
178
179
    def extent(self) -> slice:
        return (slice(-self.remaining - self.node.len, -self.remaining) if self.node
                else slice(-self.remaining, None))
180
181
182


def add_parser_guard(parser_func):
183
184
185
186
187
    """
    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.
    """
188
    def guarded_call(parser: 'Parser', text: str) -> Tuple[Node, str]:
189
190
        try:
            location = len(text)
191
192
            grammar = parser.grammar  # grammar may be 'None' for unconnected parsers!

193
            if not grammar.moving_forward__:
194
                # rollback variable changes from discarded parser passes
195
196
197
                if grammar.last_rb__loc__ <= location:
                    grammar.rollback_to__(location)
                grammar.moving_forward__ = True
198
                grammar.left_recursion_encountered__ = False
199

200
201
202
            # if location has already been visited by the current parser,
            # return saved result
            if location in parser.visited:
203
                return parser.visited[location]
204

205
206
            # break left recursion at the maximum allowed depth
            if parser.recursion_counter.setdefault(location, 0) > LEFT_RECURSION_DEPTH:
207
                grammar.left_recursion_encountered__ = True
208
209
                return None, text

Eckhart Arnold's avatar
Eckhart Arnold committed
210
211
212
            if grammar.history_tracking__:
                grammar.call_stack__.append(parser)

213
214
215
216
217
            parser.recursion_counter[location] += 1

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

218
            if node is None:
Eckhart Arnold's avatar
Eckhart Arnold committed
219
220
221
222
223
224
225
                # retrieve an earlier match result (from left recursion)
                # if it exists
                node, rest = parser.visited.get(location, (None, rest))
                # 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!
                if node is None and not grammar.left_recursion_encountered__:
                    # ortherwise also cache None-results
226
227
                    parser.visited[location] = None, rest
            else:
228
                # variable manipulating parsers will be excluded, though,
229
                # because caching would interfere with changes of variable state
230
                if grammar.last_rb__loc__ > location:
Eckhart Arnold's avatar
Eckhart Arnold committed
231
232
                    # in case of left recursion, the first recursive step that
                    # matches will store its result in the cache
233
                    parser.visited[location] = (node, rest)
234
235
236

            parser.recursion_counter[location] -= 1

237
            if grammar.history_tracking__:
238
                # don't track returning parsers except in case an error has occurred
239
240
241
                if grammar.moving_forward__ or (node and node._errors):
                    record = HistoryRecord(grammar.call_stack__.copy(), node, len(rest))
                    grammar.history__.append(record)
242
                    # print(record.stack, record.status, rest[:20].replace('\n', '|'))
243
244
                grammar.call_stack__.pop()
            grammar.moving_forward__ = False
245

246
247
248
249
250
251
252
253
254
255
256
        except RecursionError:
            node = Node(None, text[:min(10, max(1, text.find("\n")))] + " ...")
            node.add_error("maximum recursion depth of parser reached; "
                           "potentially due to too many errors!")
            rest = ''

        return node, rest

    return guarded_call


Eckhart Arnold's avatar
Eckhart Arnold committed
257
class ParserMetaClass(abc.ABCMeta):
258
259
260
261
262
    """
    ParserMetaClass adds a wrapper to the __call__ method of parser
    objects during initialization that takes care of memoizing,
    left recursion and tracing.
    """
263
    def __init__(cls, name, bases, attrs):
264
265
        guarded_parser_call = add_parser_guard(cls.__call__)
        # The following check is necessary for classes that don't override
266
267
268
269
270
271
272
        # 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
273
class Parser(ParserBase, metaclass=ParserMetaClass):
274
275
276
277
    """
    (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.
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294

    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
295
296

    Parser objects are callable and parsing is done by calling a parser
Eckhart Arnold's avatar
Eckhart Arnold committed
297
298
299
300
301
302
303
304
305
306
307
308
309
    object with the text to parse.

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

    If the parser does not match it returns `(None, text). **Note** that
    this is not the same as an empty match `("", text)`. Any empty match
    can for example be returned by the `ZeroOrMore`-parser in case the
    contained parser is repeated zero times.
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325

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

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

        cycle_detection: The apply()-method uses this variable to make
                sure that one and the same function will not be applied
                (recursively) a second time, if it has already been
                applied to this parser.
326
327
    """

328
329
    ApplyFunc = Callable[['Parser'], None]

330
331
    def __init__(self, name: str = '') -> None:
        # assert isinstance(name, str), str(name)
Eckhart Arnold's avatar
Eckhart Arnold committed
332
        super(Parser, self).__init__(name)
333
        self._grammar = None  # type: 'Grammar'
334
335
        self.reset()

Eckhart Arnold's avatar
Eckhart Arnold committed
336
    def __deepcopy__(self, memo):
Eckhart Arnold's avatar
Eckhart Arnold committed
337
338
339
340
341
342
        """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.
        """
343
344
        return self.__class__(self.name)

345
    def reset(self):
Eckhart Arnold's avatar
Eckhart Arnold committed
346
347
348
        """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."""
349
        self.visited = dict()            # type: Dict[int, Tuple[Node, str]]
350
        self.recursion_counter = dict()  # type: Dict[int, int]
351
        self.cycle_detection = set()     # type: Set[Callable]
352
        return self
353

354
    def __call__(self, text: str) -> Tuple[Node, str]:
Eckhart Arnold's avatar
Eckhart Arnold committed
355
356
357
        """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."""
358
359
        return None, text  # default behaviour: don't match

360
    def __add__(self, other: 'Parser') -> 'Series':
361
362
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
363
        return Series(self, other)
364

365
    def __or__(self, other: 'Parser') -> 'Alternative':
366
367
368
        """The | operator generates an alternative parser that applies
        the first parser and, if that does not match, the second parser.
        """
369
370
        return Alternative(self, other)

371
    @property
372
    def grammar(self) -> 'Grammar':
373
374
375
        return self._grammar

    @grammar.setter
376
    def grammar(self, grammar: 'Grammar'):
Eckhart Arnold's avatar
Eckhart Arnold committed
377
378
379
380
381
382
        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!"
383
384

    def _grammar_assigned_notifier(self):
385
386
        """A function that notifies the parser object that it has been
        assigned to a grammar."""
387
388
        pass

389
    def apply(self, func: ApplyFunc):
390
391
        """
        Applies function `func(parser)` recursively to this parser and all
Eckhart Arnold's avatar
Eckhart Arnold committed
392
        descendant parsers if any exist. The same function can never
393
394
395
396
397
398
399
400
401
402
        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


403
def mixin_comment(whitespace: str, comment: str) -> str:
404
405
    """
    Returns a regular expression that merges comment and whitespace
406
407
408
409
410
411
412
413
414
415
416
    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


417
class Grammar:
418
    """
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
    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
434
        >>> number_parser("3.1416").content()
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
        '3.1416'

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

    1. It keeps the namespace clean.

    2. The parser names of named parsers do not need to be passed to the
       constructor of the Parser object explicitly, but it suffices to
       assign them to class variables.

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

    As a consequence, though, it is highly recommended that a Grammar
    class should not define any other variables or methods with names
    that are legal parser names. A name ending with a double
    underscore '__' is *not* a legal parser name and can safely be
    used.
456
457

    Example:
458

459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
        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,
477
478
479
    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.
480
481
482
483
484
485
486
487
488
489
490
    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.

491
    Grammar classes contain a few special class fields for implicit
492
493
    whitespace and comments that should be overwritten, if the defaults
    (no comments, horizontal right aligned whitespace) don't fit:
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
    COMMENT__:  regular expression string for matching comments
    WSP__:   regular expression for whitespace and comments
    wspL__:  regular expression string for left aligned whitespace,
             which either equals WSP__ or is empty.
    wspR__:  regular expression string for right aligned whitespace,
             which either equals WSP__ or is empty.
    root__:  The root parser of the grammar. Theoretically, all parsers of the
             grammar should be reachable by the root parser. However, for testing
             of yet incomplete grammars class Grammar does not assume that this
             is the case.
    parser_initializiation__:  Before the parser class (!) has been initialized,
             which happens upon the first time it is instantiated (see doctring for
             method `_assign_parser_names()` for an explanation), this class
             field contains a value other than "done". A value of "done" indicates
             that the class has already been initialized.

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

        hostory_tracking:  A flag indicating that the parsing history shall
                be tracked

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

        _dirty_flag__:  A flag indicating that the Grammar has been called at
                least once so that the parsing-variables need to be reset
                when it is called again.

        document__:  the text that has most recently been parsed or that is
                currently being parsed.
        _reversed__:  the same text in reverse order - needed by the `Lookbehind'-
                parsers.
        variables__:  A mapping for variable names to a stack of their respective
                string values - needed by the `Capture`-, `Retrieve`- and `Pop`-
                parsers.
        rollback__:  A list of tuples (location, rollback-function) that are
                deposited by the `Capture`- and `Pop`-parsers. If the parsing
                process reaches a dead end then all rollback-functions up to
                the point to which it retreats will be called and the state
                of the variable stack restored accordingly.
        call_stack__:  A stack of all parsers that have been called. This
                is required for recording the parser history (for debugging)
                and, eventually, i.e. one day in the future, for tracing through
                the parsing process.
        history__:  A list of parser-call-stacks. A parser-call-stack is
                appended to the list each time a parser either matches, fails
                or if a parser-error occurs.
        moving_forward__:  This flag indicates that the parsing process is currently
                moving forward. This information is needed among other thins to
                trigger the roolback of variables, which happens stepwise when the
                parser is reatreating form a dead end, i.e. not moving forward.
                (See `add_parser_guard` and its local function `guarded_call`)
        left_recursion_encountered__:  This flag indicates that left recursion has
                been encountered and triggers the left-recursion algorithm.
556
    """
557
558
    root__ = None  # type: Union[Parser, None]
    # root__ must be overwritten with the root-parser by grammar subclass
Eckhart Arnold's avatar
Eckhart Arnold committed
559
    parser_initialization__ = "pending"  # type: str
560
561
562
563
564
    # some default values
    COMMENT__ = r''  # r'#.*(?:\n|$)'
    WSP__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)
    wspL__ = ''
    wspR__ = WSP__
565

566

567
    @classmethod
568
    def _assign_parser_names__(cls):
569
570
        """
        Initializes the `parser.name` fields of those
571
572
        Parser objects that are directly assigned to a class field with
        the field's name, e.g.
573
            class Grammar(Grammar):
574
575
576
                ...
                symbol = RE('(?!\\d)\\w+')
        After the call of this method symbol.name == "symbol"
577
578
579
580
        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()``
581
582
583
584

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

585
586
587
588
589
        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. 
590
        """
591
592
593
594
595
596
597
598
599
        if cls.parser_initialization__ != "done":
            cdict = cls.__dict__
            for entry, parser in cdict.items():
                if isinstance(parser, Parser) and sane_parser_name(entry):
                    if not parser.name:
                        parser.name = entry
                    if (isinstance(parser, Forward) and (not parser.parser.name)):
                        parser.parser.name = entry
            cls.parser_initialization__ = "done"
600

601

Eckhart Arnold's avatar
Eckhart Arnold committed
602
603
604
    def __init__(self, root: Parser=None) -> None:
        # if not hasattr(self.__class__, 'parser_initialization__'):
        #     self.__class__.parser_initialization__ = "pending"
605
606
607
608
        # if not hasattr(self.__class__, 'wspL__'):
        #     self.wspL__ = ''
        # if not hasattr(self.__class__, 'wspR__'):
        #     self.wspR__ = ''
609
        self.all_parsers__ = set()  # type: Set[Parser]
610
611
        self._dirty_flag__ = False  # type: bool
        self.history_tracking__ = False  # type: bool
612
        self._reset__()
613

Eckhart Arnold's avatar
Eckhart Arnold committed
614
        # prepare parsers in the class, first
615
        self._assign_parser_names__()
616

Eckhart Arnold's avatar
Eckhart Arnold committed
617
618
619
620
        # 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
621
        self.root__ = copy.deepcopy(root) if root else copy.deepcopy(self.__class__.root__)
622

623
        if self.wspL__:
Eckhart Arnold's avatar
Eckhart Arnold committed
624
            self.wsp_left_parser__ = Whitespace(self.wspL__)  # type: ParserBase
625
            self.wsp_left_parser__.grammar = self
626
            self.all_parsers__.add(self.wsp_left_parser__)  # don't you forget about me...
627
628
629
        else:
            self.wsp_left_parser__ = ZOMBIE_PARSER
        if self.wspR__:
Eckhart Arnold's avatar
Eckhart Arnold committed
630
            self.wsp_right_parser__ = Whitespace(self.wspR__)  # type: ParserBase
631
            self.wsp_right_parser__.grammar = self
632
            self.all_parsers__.add(self.wsp_right_parser__)  # don't you forget about me...
633
634
        else:
            self.wsp_right_parser__ = ZOMBIE_PARSER
635
        self.root__.apply(self._add_parser__)
636

637

638
    def __getitem__(self, key):
639
640
641
        try:
            return self.__dict__[key]
        except KeyError:
642
643
            parser_template = getattr(self, key, None)
            if parser_template:
Eckhart Arnold's avatar
Eckhart Arnold committed
644
                # add parser to grammar object on the fly...
645
                parser = copy.deepcopy(parser_template)
646
                parser.apply(self._add_parser__)
647
                # assert self[key] == parser
Eckhart Arnold's avatar
Eckhart Arnold committed
648
                return self[key]
649
            raise KeyError('Unknown parser "%s" !' % key)
650

651

652
    def _reset__(self):
653
        self.document__ = ""          # type: str
Eckhart Arnold's avatar
Eckhart Arnold committed
654
        self._reversed__ = ""         # type: str
655
        # variables stored and recalled by Capture and Retrieve parsers
656
657
        self.variables__ = dict()     # type: Dict[str, List[str]]
        self.rollback__ = []          # type: List[Tuple[int, Callable]]
658
        self.last_rb__loc__ = -1  # type: int
659
        # support for call stack tracing
660
        self.call_stack__ = []        # type: List[Parser]
661
        # snapshots of call stacks
662
        self.history__ = []           # type: List[HistoryRecord]
663
        # also needed for call stack tracing
664
        self.moving_forward__ = True  # type: bool
665
        self.left_recursion_encountered__ = False  # type: bool
666

667

Eckhart Arnold's avatar
Eckhart Arnold committed
668
669
670
671
672
673
674
    @property
    def reversed__(self) -> str:
        if not self._reversed__:
            self._reversed__ = self.document__[::-1]
        return self._reversed__


675
    def _add_parser__(self, parser: Parser) -> None:
676
677
        """
        Adds the particular copy of the parser object to this
678
        particular instance of Grammar.
679
        """
680
        if parser.name:
Eckhart Arnold's avatar
Eckhart Arnold committed
681
682
683
            # 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__), \
684
685
                ('Cannot add parser "%s" because a field with the same name '
                 'already exists in grammar object!' % parser.name)
686
            setattr(self, parser.name, parser)
687
        self.all_parsers__.add(parser)
688
689
        parser.grammar = self

690

Eckhart Arnold's avatar
Eckhart Arnold committed
691
    def __call__(self, document: str, start_parser="root__") -> Node:
692
693
        """
        Parses a document with with parser-combinators.
694
695
696

        Args:
            document (str): The source text to be parsed.
697
698
699
            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.)
700
701
702
        Returns:
            Node: The root node ot the parse tree.
        """
703
        # assert isinstance(document, str), type(document)
704
705
        if self.root__ is None:
            raise NotImplementedError()
706
        if self._dirty_flag__:
707
            self._reset__()
708
            for parser in self.all_parsers__:
709
710
                parser.reset()
        else:
711
            self._dirty_flag__ = True
712
713
        self.history_tracking__ = is_logging()
        self.document__ = document
714
        self.last_rb__loc__ = len(document) + 1  # rollback location
Eckhart Arnold's avatar
Eckhart Arnold committed
715
        parser = self[start_parser] if isinstance(start_parser, str) else start_parser
716
717
        assert parser.grammar == self, "Cannot run parsers from a different grammar object!" \
                                       " %s vs. %s" % (str(self), str(parser.grammar))
718
        stitches = []  # type: List[Node]
719
        rest = document
720
721
        if not rest:
            result, ignore = parser(rest)
722
723
724
725
726
727
728
729
730
731
        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:
                    error_msg = "Parser did not match! Invalid source file?"
                else:
                    stitches.append(result)
                    error_msg = "Parser stopped before end" + \
732
733
                                (("! trying to recover" +
                                  (" but stopping history recording at this point."
734
                                   if self.history_tracking__ else "..."))
735
736
737
738
                                 if len(stitches) < MAX_DROPOUTS
                                 else " too often! Terminating parser.")
                stitches.append(Node(None, skip))
                stitches[-1].add_error(error_msg)
739
                if self.history_tracking__:
740
741
742
743
                    # some parsers may have matched and left history records with nodes != None.
                    # Because these are not connected to the stiched root node, their pos
                    # properties will not be initialized by setting the root node's pos property
                    # to zero. Therefore, their pos properties need to be initialized here
744
                    for record in self.history__:
745
746
                        if record.node and record.node._pos < 0:
                            record.node.pos = 0
747
748
749
                    record = HistoryRecord(self.call_stack__.copy(), stitches[-1], len(rest))
                    self.history__.append(record)
                    self.history_tracking__ = False
750
751
752
753
        if stitches:
            if rest:
                stitches.append(Node(None, rest))
            result = Node(None, tuple(stitches))
754
        if any(self.variables__.values()):
Eckhart Arnold's avatar
Eckhart Arnold committed
755
756
757
758
759
760
761
762
763
764
765
            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)
766
767
768
        result.pos = 0  # calculate all positions
        return result

769

770
    def push_rollback__(self, location, func):
771
772
        """
        Adds a rollback function that either removes or re-adds
773
774
775
776
        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.
        """
777
778
779
        self.rollback__.append((location, func))
        self.last_rb__loc__ = location

780

781
    def rollback_to__(self, location):
782
783
        """
        Rolls back the variable stacks (`self.variables`) to its
784
785
        state at an earlier location in the parsed document.
        """
786
787
788
789
790
791
792
        while self.rollback__ and self.rollback__[-1][0] <= location:
            loc, rollback_func = self.rollback__.pop()
            assert not loc > self.last_rb__loc__
            rollback_func()
        self.last_rb__loc__ == self.rollback__[-1][0] if self.rollback__ \
            else (len(self.document__) + 1)

793

794
    def log_parsing_history__(self, log_file_name: str = '') -> None:
795
796
        """
        Writes a log of the parsing history of the most recently parsed
797
798
799
        document. 
        """
        def prepare_line(record):
800
            excerpt = self.document__.__getitem__(record.extent)[:25].replace('\n', '\\n')
801
            excerpt = "'%s'" % excerpt if len(excerpt) < 25 else "'%s...'" % excerpt
802
            return record.stack, record.status, excerpt
803
804

        def write_log(history, log_name):
Eckhart Arnold's avatar
Eckhart Arnold committed
805
            path = os.path.join(log_dir(), log_name + "_parser.log")
Eckhart Arnold's avatar
Eckhart Arnold committed
806
807
808
            if os.path.exists(path):
                os.remove(path)
                print('WARNING: Log-file "%s" already existed and was deleted.' % path)
809
810
811
812
            if history:
                with open(path, "w", encoding="utf-8") as f:
                    f.write("\n".join(history))

813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
        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')
830
831


Eckhart Arnold's avatar
Eckhart Arnold committed
832
def dsl_error_msg(parser: Parser, error_str: str) -> str:
833
834
    """
    Returns an error message for errors in the parser configuration,
835
836
837
    e.g. errors that result in infinite loops.

    Args:
Eckhart Arnold's avatar
Eckhart Arnold committed
838
        parser (Parser):  The parser where the error was noticed. Note
839
            that this is not necessarily the parser that caused the
840
            error but only where the error became apparent.
841
842
843
844
845
        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.
    """
846
    msg = ["DSL parser specification error:", error_str, 'Caught by parser "%s".' % str(parser)]
847
848
    if parser.grammar.history__:
        msg.extend(["\nCall stack:", parser.grammar.history__[-1].stack])
849
850
851
852
853
    else:
        msg.extend(["\nEnable history tracking in Grammar object to display call stack."])
    return " ".join(msg)


854

855
856
857
858
859
860
861
########################################################################
#
# Token and Regular Expression parser classes (i.e. leaf classes)
#
########################################################################


862
863
864
RX_PREPROCESSOR_TOKEN = re.compile('\w+')
BEGIN_TOKEN = '\x1b'
END_TOKEN = '\x1c'
865
866


867
def make_token(token: str, argument: str = '') -> str:
868
869
    """
    Turns the ``token`` and ``argument`` into a special token that
870
    will be caught by the `PreprocessorToken`-parser.
871

872
873
    This function is a support function that should be used by
    preprocessors to inject preprocessor tokens into the source text.
874
    """
875
876
877
    assert RX_PREPROCESSOR_TOKEN.match(token)
    assert argument.find(BEGIN_TOKEN) < 0
    assert argument.find(END_TOKEN) < 0
878

879
    return BEGIN_TOKEN + token + argument + END_TOKEN
880
881


882
def nil_preprocessor(text: str) -> str:
Eckhart Arnold's avatar
Eckhart Arnold committed
883
    return text
884
885


886
class PreprocessorToken(Parser):
887
    """
888
    Parses tokens that have been inserted by a preprocessor.
889
    
890
    Preprocessors can generate Tokens with the ``make_token``-function.
891
    These tokens start and end with magic characters that can only be
892
893
894
    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.
895
    """
896

897
898
899
900
    def __init__(self, token: str) -> None:
        assert token and token.isupper()
        assert RX_PREPROCESSOR_TOKEN.match(token)
        super(PreprocessorToken, self).__init__(token)
901

902
    def __call__(self, text: str) -> Tuple[Node, str]:
903
904
        if text[0:1] == BEGIN_TOKEN:
            end = text.find(END_TOKEN, 1)
905
906
            if end < 0:
                node = Node(self, '').add_error(
907
908
                    'END_TOKEN delimiter missing from preprocessor token. '
                    '(Most likely due to a preprocessor bug!)')  # type: Node
909
910
911
                return node, text[1:]
            elif end == 0:
                node = Node(self, '').add_error(
912
913
                    'Preprocessor-token cannot have zero length. '
                    '(Most likely due to a preprocessor bug!)')
914
                return node, text[2:]
915
            elif text.find(BEGIN_TOKEN, 1, end) >= 0:
916
917
                node = Node(self, text[len(self.name) + 1:end])
                node.add_error(
918
919
920
                    'Preprocessor-tokens must not be nested or contain '
                    'BEGIN_TOKEN delimiter as part of their argument. '
                    '(Most likely due to a preprocessor bug!)')
921
922
923
924
925
926
927
928
                return node, text[end:]
            if text[1:len(self.name) + 1] == self.name:
                return Node(self, text[len(self.name) + 1:end]), \
                       text[end + 1:]
        return None, text


class RegExp(Parser):
Eckhart Arnold's avatar
Eckhart Arnold committed
929
    """Regular expression parser.
930
931
932
933
934
    
    The RegExp-parser parses text that matches a regular expression.
    RegExp can also be considered as the "atomic parser", because all
    other parsers delegate part of the parsing job to other parsers,
    but do not match text directly.
Eckhart Arnold's avatar
Eckhart Arnold committed
935
936
937
938
939
940
941
942

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

    EBNF-Notation:  `/ ... /`
    EBNF-Example:   `word = /\w+/`
943
    """
944
945

    def __init__(self, regexp, name: str = '') -> None:
946
        super(RegExp, self).__init__(name)
947
948
949
        self.regexp = re.compile(regexp) if isinstance(regexp, str) else regexp

    def __deepcopy__(self, memo):
Eckhart Arnold's avatar
Eckhart Arnold committed
950
        # `regex` supports deep copies, but not `re`
951
        try:
Eckhart Arnold's avatar
Eckhart Arnold committed
952
            regexp = copy.deepcopy(self.regexp, memo)
953
954
        except TypeError:
            regexp = self.regexp.pattern
955
        return RegExp(regexp, self.name)
956

957
    def __call__(self, text: str) -> Tuple[Node, str]:
958
        match = text[0:1] != BEGIN_TOKEN and self.regexp.match(text)  # ESC starts a preprocessor token.
959
960
961
962
963
        if match:
            end = match.end()
            return Node(self, text[:end]), text[end:]
        return None, text

964
    def __repr__(self):
965
        return '/%s/' % self.regexp.pattern
966

967

968
969
class Whitespace(RegExp):
    assert WHITESPACE_PTYPE == ":Whitespace"
970
971


972
class RE(Parser):
973
974
    """
    Regular Expressions with optional leading or trailing whitespace.
975
976
    
    The RE-parser parses pieces of text that match a given regular
Eckhart Arnold's avatar
Eckhart Arnold committed
977
    expression. Other than the ``RegExp``-Parser it can also skip
978
979
    "implicit whitespace" before or after the matched text.
    
Eckhart Arnold's avatar
Eckhart Arnold committed
980
981
982
983
984
    The whitespace is in turn defined by a regular expression. It should
    be made sure that this expression also matches the empty string,
    e.g. use r'\s*' or r'[\t ]+', but not r'\s+'. If the respective
    parameters in the constructor are set to ``None`` the default
    whitespace expression from the Grammar object will be used.
Eckhart Arnold's avatar
Eckhart Arnold committed
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999

    Example (allowing whitespace on the right hand side, but not on
    the left hand side of a regular expression):
    >>> word = RE(r'\w+', wR=r'\s*')
    >>> parser = Grammar(word)
    >>> result = parser('Haus ')
    >>> result.content()
    'Haus '
    >>> result.structure()
    '(:RE (:RegExp "Haus") (:Whitespace " "))'
    >>> parser(' Haus').content()
    ' <<< Error on " Haus" | Parser did not match! Invalid source file? >>> '

    EBNF-Notation:  `/ ... /~`  or  `~/ ... /`  or  `~/ ... /~`
    EBNF-Example:   `word = /\w+/~`
1000
    """
1001
    def __init__(self, regexp, wL=None, wR=None, name=''):
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
        """Constructor for class RE.
                
        Args:
            regexp (str or regex object):  The regular expression to be
                used for parsing. 
            wL (str or regexp):  Left whitespace regular expression, 
                i.e. either ``None``, the empty string or a regular
                expression (e.g. "\s*") that defines whitespace. An 
                empty string means no whitespace will be skipped,
                ``None`` means that the default whitespace will be 
                used.
            wR (str or regexp):  Right whitespace regular expression.
                See above.
            name:  The optional name of the parser.
        """
1017
        super(RE, self).__init__(name)
1018
1019
        self.wL = wL
        self.wR = wR
1020
1021
        self.wspLeft = Whitespace(wL) if wL else ZOMBIE_PARSER
        self.wspRight = Whitespace(wR) if wR else ZOMBIE_PARSER
1022
1023
        self.main = RegExp(regexp)

Eckhart Arnold's avatar
Eckhart Arnold committed
1024
1025
1026
1027
1028
    def __deepcopy__(self, memo={}):
        try:
            regexp = copy.deepcopy(self.main.regexp, memo)
        except TypeError:
            regexp = self.main.regexp.pattern
1029
        return self.__class__(regexp, self.wL, self.wR, self.name)
Eckhart Arnold's avatar
Eckhart Arnold committed
1030

1031
    def __call__(self, text: str) -> Tuple[Node, str]:
1032
        # assert self.main.regexp.pattern != "@"
1033
        t = text    # type: str
1034
1035
1036
1037
1038
1039
1040
1041
1042
        wL, t = self.wspLeft(t)
        main, t = self.main(t)
        if main:
            wR, t = self.wspRight(t)
            result = tuple(nd for nd in (wL, main, wR)
                           if nd and nd.result != '')
            return Node(self, result), t
        return None, text

1043
    def __repr__(self):
1044
1045
        wL = '~' if self.wspLeft != ZOMBIE_PARSER else ''
        wR = '~' if self.wspRight != ZOMBIE_PARSER else ''
1046
        return wL + '/%s/' % self.main.regexp.pattern + wR
1047

1048
1049
    def _grammar_assigned_notifier(self):
        if self.grammar:
1050
            # use default whitespace parsers if not otherwise specified
1051
1052
1053
1054
1055
            if self.wL is None:
                self.wspLeft = self.grammar.wsp_left_parser__
            if self.wR is None:
                self.wspRight = self.grammar.wsp_right_parser__

1056
    def apply(self, func: Parser.ApplyFunc):
1057
1058
1059
1060
1061
1062
1063
1064
        if super(RE, self).apply(func):
            if self.wL:
                self.wspLeft.apply(func)
            if self.wR:
                self.wspRight.apply(func)
            self.main.apply(func)


1065
class Token(RE):
1066
    """
Eckhart Arnold's avatar
Eckhart Arnold committed
1067
1068
    Class Token parses simple strings. Any regular regular expression
    commands will be interpreted as simple sequence of characters.
1069
1070
1071
1072
1073
1074

    Other than that class Token is essentially a renamed version of
    class RE. Because tokens often have a particular semantic different
    from other REs, parsing them with a separate parser class allows to
    distinguish them by their parser type.
    """
1075
1076
    assert TOKEN_PTYPE == ":Token"

1077
    def __init__(self, token: str, wL=None, wR=None, name: str = '') -> None:
1078
1079
1080
1081
1082
        self.token = token
        super(Token, self).__init__(escape_re(token), wL, wR, name)

    def __deepcopy__(self, memo={}):
        return self.__class__(self.token, self.wL, self.wR, self.name)
1083

1084
    def __repr__(self):
1085
        return '"%s"' % self.token if self.token.find('"') < 0 else "'%s'" % self.token
1086

1087
1088
1089

########################################################################
#
Eckhart Arnold's avatar
Eckhart Arnold committed
1090
1091
# Containing parser classes, i.e. parsers that contain other parsers
# to which they delegate (i.e. trunk classes)
1092
1093
1094
1095
1096
#
########################################################################


class UnaryOperator(Parser):
1097
    """
Eckhart Arnold's avatar
Eckhart Arnold committed
1098
1099
    Base class of all unary parser operators, i.e. parser that contains
    one and only one other parser, like the optional parser for example.
1100
1101
1102
1103
1104
1105

    The UnaryOperator base class supplies __deepcopy__ and apply
    methods for unary parser operators. The __deepcopy__ method needs
    to be overwritten, however, if the constructor of a derived class
    has additional parameters.
    """
1106
    def __init__(self, parser: Parser, name: str = '') -> None:
1107
        super(UnaryOperator, self).__init__(name)
Eckhart Arnold's avatar
Eckhart Arnold committed
1108
        assert isinstance(parser, Parser), str(parser)