parser.py 65.1 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

            if grammar.history_tracking__:
                grammar.call_stack__.append(parser)
202

203
204
205
            # if location has already been visited by the current parser,
            # return saved result
            if location in parser.visited:
206
                return parser.visited[location]
207

208
209
            # break left recursion at the maximum allowed depth
            if parser.recursion_counter.setdefault(location, 0) > LEFT_RECURSION_DEPTH:
210
                grammar.left_recursion_encountered__ = True
211
212
213
214
215
216
217
                return None, text

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

314
315
    def __init__(self, name: str = '') -> None:
        # assert isinstance(name, str), str(name)
Eckhart Arnold's avatar
Eckhart Arnold committed
316
        super(Parser, self).__init__(name)
317
        self._grammar = None  # type: 'Grammar'
318
319
        self.reset()

Eckhart Arnold's avatar
Eckhart Arnold committed
320
    def __deepcopy__(self, memo):
Eckhart Arnold's avatar
Eckhart Arnold committed
321
322
323
324
325
326
        """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.
        """
327
328
        return self.__class__(self.name)

329
    def reset(self):
Eckhart Arnold's avatar
Eckhart Arnold committed
330
331
332
        """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."""
333
        self.visited = dict()            # type: Dict[int, Tuple[Node, str]]
334
        self.recursion_counter = dict()  # type: Dict[int, int]
335
        self.cycle_detection = set()     # type: Set[Callable]
336
        return self
337

338
    def __call__(self, text: str) -> Tuple[Node, str]:
Eckhart Arnold's avatar
Eckhart Arnold committed
339
340
341
        """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."""
342
343
        return None, text  # default behaviour: don't match

344
    def __add__(self, other: 'Parser') -> 'Series':
345
346
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
347
        return Series(self, other)
348

349
    def __or__(self, other: 'Parser') -> 'Alternative':
350
351
352
        """The | operator generates an alternative parser that applies
        the first parser and, if that does not match, the second parser.
        """
353
354
        return Alternative(self, other)

355
    @property
356
    def grammar(self) -> 'Grammar':
357
358
359
        return self._grammar

    @grammar.setter
360
    def grammar(self, grammar: 'Grammar'):
Eckhart Arnold's avatar
Eckhart Arnold committed
361
362
363
364
365
366
        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!"
367
368

    def _grammar_assigned_notifier(self):
369
370
        """A function that notifies the parser object that it has been
        assigned to a grammar."""
371
372
        pass

373
    def apply(self, func: ApplyFunc):
374
375
        """
        Applies function `func(parser)` recursively to this parser and all
Eckhart Arnold's avatar
Eckhart Arnold committed
376
        descendant parsers if any exist. The same function can never
377
378
379
380
381
382
383
384
385
386
        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


387
def mixin_comment(whitespace: str, comment: str) -> str:
388
389
    """
    Returns a regular expression that merges comment and whitespace
390
391
392
393
394
395
396
397
398
399
400
    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


401
class Grammar:
402
    """
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
    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
418
        >>> number_parser("3.1416").content()
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
        '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.
440
441

    Example:
442

443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
        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,
461
462
463
    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.
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
    If one and the same parser is assigned to several class variables
    such as, for example the parser `expression` in the example above,
    the first name sticks.

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

    Grammar objects contain a few special fields for implicit
    whitespace and comments that should be overwritten, if the defaults
    (no comments, horizontal right aligned whitespace) don't fit:
    COMMENT__   - regular expression string for matching comments
    wspL__      - regular expression string for left aligned whitespace
    wspR__      - regular expression string for right aligned whitespace
481
    root__      - the root parser of the grammar
482
483
    """

484
485
    root__ = None  # type: Union[Parser, None]
    # root__ must be overwritten with the root-parser by grammar subclass
Eckhart Arnold's avatar
Eckhart Arnold committed
486
    parser_initialization__ = "pending"  # type: str
487
488
489
490
491
    # some default values
    COMMENT__ = r''  # r'#.*(?:\n|$)'
    WSP__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)
    wspL__ = ''
    wspR__ = WSP__
492

493

494
    @classmethod
495
    def _assign_parser_names__(cls):
496
497
        """
        Initializes the `parser.name` fields of those
498
499
        Parser objects that are directly assigned to a class field with
        the field's name, e.g.
500
            class Grammar(Grammar):
501
502
503
                ...
                symbol = RE('(?!\\d)\\w+')
        After the call of this method symbol.name == "symbol"
504
505
506
507
        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()``
508
509
510
511

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

512
513
514
515
516
        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. 
517
518
519
520
521
        """
        if cls.parser_initialization__ == "done":
            return
        cdict = cls.__dict__
        for entry, parser in cdict.items():
522
            if isinstance(parser, Parser) and sane_parser_name(entry):
523
                if not parser.name:
524
                    parser.name = entry
525
                if (isinstance(parser, Forward) and (not parser.parser.name)):
526
527
528
                    parser.parser.name = entry
        cls.parser_initialization__ = "done"

529

Eckhart Arnold's avatar
Eckhart Arnold committed
530
531
532
    def __init__(self, root: Parser=None) -> None:
        # if not hasattr(self.__class__, 'parser_initialization__'):
        #     self.__class__.parser_initialization__ = "pending"
533
534
535
536
        # if not hasattr(self.__class__, 'wspL__'):
        #     self.wspL__ = ''
        # if not hasattr(self.__class__, 'wspR__'):
        #     self.wspR__ = ''
537
538
539
        self.all_parsers__ = set()  # type: Set[Parser]
        self.dirty_flag__ = False
        self.history_tracking__ = False
540
        self._reset__()
541

Eckhart Arnold's avatar
Eckhart Arnold committed
542
        # prepare parsers in the class, first
543
        self._assign_parser_names__()
544

Eckhart Arnold's avatar
Eckhart Arnold committed
545
546
547
548
        # 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
549
        self.root__ = copy.deepcopy(root) if root else copy.deepcopy(self.__class__.root__)
550

551
        if self.wspL__:
Eckhart Arnold's avatar
Eckhart Arnold committed
552
            self.wsp_left_parser__ = Whitespace(self.wspL__)  # type: ParserBase
553
            self.wsp_left_parser__.grammar = self
554
            self.all_parsers__.add(self.wsp_left_parser__)  # don't you forget about me...
555
556
557
        else:
            self.wsp_left_parser__ = ZOMBIE_PARSER
        if self.wspR__:
Eckhart Arnold's avatar
Eckhart Arnold committed
558
            self.wsp_right_parser__ = Whitespace(self.wspR__)  # type: ParserBase
559
            self.wsp_right_parser__.grammar = self
560
            self.all_parsers__.add(self.wsp_right_parser__)  # don't you forget about me...
561
562
        else:
            self.wsp_right_parser__ = ZOMBIE_PARSER
563
        self.root__.apply(self._add_parser__)
564

565

566
    def __getitem__(self, key):
567
568
569
        try:
            return self.__dict__[key]
        except KeyError:
570
571
            parser_template = getattr(self, key, None)
            if parser_template:
Eckhart Arnold's avatar
Eckhart Arnold committed
572
                # add parser to grammar object on the fly...
573
                parser = copy.deepcopy(parser_template)
574
                parser.apply(self._add_parser__)
575
                # assert self[key] == parser
Eckhart Arnold's avatar
Eckhart Arnold committed
576
                return self[key]
577
            raise KeyError('Unknown parser "%s" !' % key)
578

579

580
    def _reset__(self):
581
        self.document__ = ""          # type: str
Eckhart Arnold's avatar
Eckhart Arnold committed
582
        self._reversed__ = ""         # type: str
583
        # variables stored and recalled by Capture and Retrieve parsers
584
585
        self.variables__ = dict()     # type: Dict[str, List[str]]
        self.rollback__ = []          # type: List[Tuple[int, Callable]]
586
        self.last_rb__loc__ = -1  # type: int
587
        # support for call stack tracing
588
        self.call_stack__ = []        # type: List[Parser]
589
        # snapshots of call stacks
590
        self.history__ = []           # type: List[HistoryRecord]
591
        # also needed for call stack tracing
592
        self.moving_forward__ = True  # type: bool
593
        self.left_recursion_encountered__ = False  # type: bool
594

595

Eckhart Arnold's avatar
Eckhart Arnold committed
596
597
598
599
600
601
602
    @property
    def reversed__(self) -> str:
        if not self._reversed__:
            self._reversed__ = self.document__[::-1]
        return self._reversed__


603
    def _add_parser__(self, parser: Parser) -> None:
604
605
        """
        Adds the particular copy of the parser object to this
606
        particular instance of Grammar.
607
        """
608
        if parser.name:
Eckhart Arnold's avatar
Eckhart Arnold committed
609
610
611
            # 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__), \
612
613
                ('Cannot add parser "%s" because a field with the same name '
                 'already exists in grammar object!' % parser.name)
614
            setattr(self, parser.name, parser)
615
        self.all_parsers__.add(parser)
616
617
        parser.grammar = self

618

Eckhart Arnold's avatar
Eckhart Arnold committed
619
    def __call__(self, document: str, start_parser="root__") -> Node:
620
621
        """
        Parses a document with with parser-combinators.
622
623
624

        Args:
            document (str): The source text to be parsed.
625
626
627
            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.)
628
629
630
        Returns:
            Node: The root node ot the parse tree.
        """
631
        # assert isinstance(document, str), type(document)
632
633
        if self.root__ is None:
            raise NotImplementedError()
634
        if self.dirty_flag__:
635
            self._reset__()
636
            for parser in self.all_parsers__:
637
638
                parser.reset()
        else:
639
640
641
            self.dirty_flag__ = True
        self.history_tracking__ = is_logging()
        self.document__ = document
642
        self.last_rb__loc__ = len(document) + 1  # rollback location
Eckhart Arnold's avatar
Eckhart Arnold committed
643
        parser = self[start_parser] if isinstance(start_parser, str) else start_parser
644
645
        assert parser.grammar == self, "Cannot run parsers from a different grammar object!" \
                                       " %s vs. %s" % (str(self), str(parser.grammar))
646
        stitches = []  # type: List[Node]
647
        rest = document
648
649
        if not rest:
            result, ignore = parser(rest)
650
651
652
653
654
655
656
657
658
659
        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" + \
660
661
                                (("! trying to recover" +
                                  (" but stopping history recording at this point."
662
                                   if self.history_tracking__ else "..."))
663
664
665
666
                                 if len(stitches) < MAX_DROPOUTS
                                 else " too often! Terminating parser.")
                stitches.append(Node(None, skip))
                stitches[-1].add_error(error_msg)
667
                if self.history_tracking__:
668
669
670
671
                    # 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
672
                    for record in self.history__:
673
674
                        if record.node and record.node._pos < 0:
                            record.node.pos = 0
675
676
677
                    record = HistoryRecord(self.call_stack__.copy(), stitches[-1], len(rest))
                    self.history__.append(record)
                    self.history_tracking__ = False
678
679
680
681
        if stitches:
            if rest:
                stitches.append(Node(None, rest))
            result = Node(None, tuple(stitches))
682
        if any(self.variables__.values()):
Eckhart Arnold's avatar
Eckhart Arnold committed
683
684
685
686
687
688
689
690
691
692
693
            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)
694
695
696
        result.pos = 0  # calculate all positions
        return result

697

698
    def push_rollback__(self, location, func):
699
700
        """
        Adds a rollback function that either removes or re-adds
701
702
703
704
        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.
        """
705
706
707
        self.rollback__.append((location, func))
        self.last_rb__loc__ = location

708

709
    def rollback_to__(self, location):
710
711
        """
        Rolls back the variable stacks (`self.variables`) to its
712
713
        state at an earlier location in the parsed document.
        """
714
715
716
717
718
719
720
        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)

721

722
    def log_parsing_history__(self, log_file_name: str = '') -> None:
723
724
        """
        Writes a log of the parsing history of the most recently parsed
725
726
727
        document. 
        """
        def prepare_line(record):
728
            excerpt = self.document__.__getitem__(record.extent)[:25].replace('\n', '\\n')
729
            excerpt = "'%s'" % excerpt if len(excerpt) < 25 else "'%s...'" % excerpt
730
            return record.stack, record.status, excerpt
731
732

        def write_log(history, log_name):
Eckhart Arnold's avatar
Eckhart Arnold committed
733
            path = os.path.join(log_dir(), log_name + "_parser.log")
734
735
736
737
738
739
            if history:
                with open(path, "w", encoding="utf-8") as f:
                    f.write("\n".join(history))
            elif os.path.exists(path):
                os.remove(path)

740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
        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')
757
758


Eckhart Arnold's avatar
Eckhart Arnold committed
759
def dsl_error_msg(parser: Parser, error_str: str) -> str:
760
761
    """
    Returns an error message for errors in the parser configuration,
762
763
764
    e.g. errors that result in infinite loops.

    Args:
Eckhart Arnold's avatar
Eckhart Arnold committed
765
        parser (Parser):  The parser where the error was noticed. Note
766
            that this is not necessarily the parser that caused the
767
            error but only where the error became apparent.
768
769
770
771
772
        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.
    """
773
    msg = ["DSL parser specification error:", error_str, 'Caught by parser "%s".' % str(parser)]
774
775
    if parser.grammar.history__:
        msg.extend(["\nCall stack:", parser.grammar.history__[-1].stack])
776
777
778
779
780
    else:
        msg.extend(["\nEnable history tracking in Grammar object to display call stack."])
    return " ".join(msg)


781

782
783
784
785
786
787
788
########################################################################
#
# Token and Regular Expression parser classes (i.e. leaf classes)
#
########################################################################


789
790
791
RX_PREPROCESSOR_TOKEN = re.compile('\w+')
BEGIN_TOKEN = '\x1b'
END_TOKEN = '\x1c'
792
793


794
def make_token(token: str, argument: str = '') -> str:
795
796
    """
    Turns the ``token`` and ``argument`` into a special token that
797
    will be caught by the `PreprocessorToken`-parser.
798

799
800
    This function is a support function that should be used by
    preprocessors to inject preprocessor tokens into the source text.
801
    """
802
803
804
    assert RX_PREPROCESSOR_TOKEN.match(token)
    assert argument.find(BEGIN_TOKEN) < 0
    assert argument.find(END_TOKEN) < 0
805

806
    return BEGIN_TOKEN + token + argument + END_TOKEN
807
808


809
def nil_preprocessor(text: str) -> str:
Eckhart Arnold's avatar
Eckhart Arnold committed
810
    return text
811
812


813
class PreprocessorToken(Parser):
814
    """
815
    Parses tokens that have been inserted by a preprocessor.
816
    
817
    Preprocessors can generate Tokens with the ``make_token``-function.
818
    These tokens start and end with magic characters that can only be
819
820
821
    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.
822
    """
823

824
825
826
827
    def __init__(self, token: str) -> None:
        assert token and token.isupper()
        assert RX_PREPROCESSOR_TOKEN.match(token)
        super(PreprocessorToken, self).__init__(token)
828

829
    def __call__(self, text: str) -> Tuple[Node, str]:
830
831
        if text[0:1] == BEGIN_TOKEN:
            end = text.find(END_TOKEN, 1)
832
833
            if end < 0:
                node = Node(self, '').add_error(
834
835
                    'END_TOKEN delimiter missing from preprocessor token. '
                    '(Most likely due to a preprocessor bug!)')  # type: Node
836
837
838
                return node, text[1:]
            elif end == 0:
                node = Node(self, '').add_error(
839
840
                    'Preprocessor-token cannot have zero length. '
                    '(Most likely due to a preprocessor bug!)')
841
                return node, text[2:]
842
            elif text.find(BEGIN_TOKEN, 1, end) >= 0:
843
844
                node = Node(self, text[len(self.name) + 1:end])
                node.add_error(
845
846
847
                    'Preprocessor-tokens must not be nested or contain '
                    'BEGIN_TOKEN delimiter as part of their argument. '
                    '(Most likely due to a preprocessor bug!)')
848
849
850
851
852
853
854
855
                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
856
    """Regular expression parser.
857
858
859
860
861
    
    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
862
863
864
865
866
867
868
869

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

    EBNF-Notation:  `/ ... /`
    EBNF-Example:   `word = /\w+/`
870
    """
871
872

    def __init__(self, regexp, name: str = '') -> None:
873
        super(RegExp, self).__init__(name)
874
875
876
        self.regexp = re.compile(regexp) if isinstance(regexp, str) else regexp

    def __deepcopy__(self, memo):
Eckhart Arnold's avatar
Eckhart Arnold committed
877
        # `regex` supports deep copies, but not `re`
878
        try:
Eckhart Arnold's avatar
Eckhart Arnold committed
879
            regexp = copy.deepcopy(self.regexp, memo)
880
881
        except TypeError:
            regexp = self.regexp.pattern
882
        return RegExp(regexp, self.name)
883

884
    def __call__(self, text: str) -> Tuple[Node, str]:
885
        match = text[0:1] != BEGIN_TOKEN and self.regexp.match(text)  # ESC starts a preprocessor token.
886
887
888
889
890
        if match:
            end = match.end()
            return Node(self, text[:end]), text[end:]
        return None, text

891
    def __repr__(self):
892
        return '/%s/' % self.regexp.pattern
893

894

895
896
class Whitespace(RegExp):
    assert WHITESPACE_PTYPE == ":Whitespace"
897
898


899
class RE(Parser):
900
901
    """
    Regular Expressions with optional leading or trailing whitespace.
902
903
    
    The RE-parser parses pieces of text that match a given regular
Eckhart Arnold's avatar
Eckhart Arnold committed
904
    expression. Other than the ``RegExp``-Parser it can also skip
905
906
    "implicit whitespace" before or after the matched text.
    
Eckhart Arnold's avatar
Eckhart Arnold committed
907
908
909
910
911
    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
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926

    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+/~`
927
    """
928
    def __init__(self, regexp, wL=None, wR=None, name=''):
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
        """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.
        """
944
        super(RE, self).__init__(name)
945
946
        self.wL = wL
        self.wR = wR
947
948
        self.wspLeft = Whitespace(wL) if wL else ZOMBIE_PARSER
        self.wspRight = Whitespace(wR) if wR else ZOMBIE_PARSER
949
950
        self.main = RegExp(regexp)

Eckhart Arnold's avatar
Eckhart Arnold committed
951
952
953
954
955
    def __deepcopy__(self, memo={}):
        try:
            regexp = copy.deepcopy(self.main.regexp, memo)
        except TypeError:
            regexp = self.main.regexp.pattern
956
        return self.__class__(regexp, self.wL, self.wR, self.name)
Eckhart Arnold's avatar
Eckhart Arnold committed
957

958
    def __call__(self, text: str) -> Tuple[Node, str]:
959
        # assert self.main.regexp.pattern != "@"
960
        t = text    # type: str
961
962
963
964
965
966
967
968
969
        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

970
    def __repr__(self):
971
972
        wL = '~' if self.wspLeft != ZOMBIE_PARSER else ''
        wR = '~' if self.wspRight != ZOMBIE_PARSER else ''
973
        return wL + '/%s/' % self.main.regexp.pattern + wR
974

975
976
    def _grammar_assigned_notifier(self):
        if self.grammar:
977
            # use default whitespace parsers if not otherwise specified
978
979
980
981
982
            if self.wL is None:
                self.wspLeft = self.grammar.wsp_left_parser__
            if self.wR is None:
                self.wspRight = self.grammar.wsp_right_parser__

983
    def apply(self, func: Parser.ApplyFunc):
984
985
986
987
988
989
990
991
        if super(RE, self).apply(func):
            if self.wL:
                self.wspLeft.apply(func)
            if self.wR:
                self.wspRight.apply(func)
            self.main.apply(func)


992
class Token(RE):
993
    """
Eckhart Arnold's avatar
Eckhart Arnold committed
994
995
    Class Token parses simple strings. Any regular regular expression
    commands will be interpreted as simple sequence of characters.
996
997
998
999
1000
1001

    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.
    """
1002
1003
    assert TOKEN_PTYPE == ":Token"

1004
    def __init__(self, token: str, wL=None, wR=None, name: str = '') -> None:
1005
1006
1007
1008
1009
        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)
1010

1011
    def __repr__(self):
1012
        return '"%s"' % self.token if self.token.find('"') < 0 else "'%s'" % self.token
1013

1014
1015
1016

########################################################################
#
Eckhart Arnold's avatar
Eckhart Arnold committed
1017
1018
# Containing parser classes, i.e. parsers that contain other parsers
# to which they delegate (i.e. trunk classes)
1019
1020
1021
1022
1023
#
########################################################################


class UnaryOperator(Parser):
1024
    """
Eckhart Arnold's avatar
Eckhart Arnold committed
1025
1026
    Base class of all unary parser operators, i.e. parser that contains
    one and only one other parser, like the optional parser for example.
1027
1028
1029
1030
1031
1032

    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.
    """
1033
    def __init__(self, parser: Parser, name: str = '') -> None:
1034
        super(UnaryOperator, self).__init__(name)
1035
        # assert isinstance(parser, Parser)
1036
        self.parser = parser  # type: Parser
1037

Eckhart Arnold's avatar
Eckhart Arnold committed
1038
1039
    def __deepcopy__(self, memo):
        parser = copy.deepcopy(self.parser, memo)
1040
        return self.__class__(parser, self.name)
Eckhart Arnold's avatar
Eckhart Arnold committed
1041

1042
    def apply(self, func: Parser.ApplyFunc):
1043
1044
1045
1046
1047
        if super(UnaryOperator, self).apply(func):
            self.parser.apply(func)


class NaryOperator(Parser):
1048
1049
1050
1051
1052
    """
    Base class of all Nnary parser operators, i.e. parser that
    contains one or more other parsers, like the alternative
    parser for example.

Eckhart Arnold's avatar
Eckhart Arnold committed
1053
1054
1055
1056
    The NnaryOperator 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.
1057
    """
1058
    def __init__(self, *parsers: Parser, name: str = '') -> None:
1059
        super(NaryOperator, self).__init__(name)
1060
        # assert all([isinstance(parser, Parser) for parser in parsers]), str(parsers)