parse.py 160 KB
Newer Older
1
# parse.py - parser combinators 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
Module ``parse`` contains the python classes and functions for
21
22
DHParser's packrat-parser. It's central class is the
``Grammar``-class, which is the base class for any concrete
23
24
25
26
27
28
29
30
31
Grammar. Grammar-objects are callable and parsing is done by
calling a Grammar-object with a source text as argument.

The different parsing functions are callable descendants of class
``Parser``. Usually, they are organized in a tree and defined
within the namespace of a grammar-class. See ``ebnf.EBNFGrammar``
for an example.
"""

32

33
from collections import defaultdict
34
import copy
35
from typing import Callable, cast, List, Tuple, Set, AbstractSet, Dict, \
36
    DefaultDict, Sequence, Union, Optional, Iterator
37

eckhart's avatar
eckhart committed
38
from DHParser.configuration import get_config_value
Eckhart Arnold's avatar
Eckhart Arnold committed
39
from DHParser.error import Error, ErrorCode, MANDATORY_CONTINUATION, \
40
    UNDEFINED_RETRIEVE, PARSER_LOOKAHEAD_FAILURE_ONLY, \
eckhart's avatar
eckhart committed
41
    PARSER_LOOKAHEAD_MATCH_ONLY, PARSER_STOPPED_BEFORE_END, PARSER_NEVER_TOUCHES_DOCUMENT, \
42
43
44
    MALFORMED_ERROR_STRING, MANDATORY_CONTINUATION_AT_EOF, DUPLICATE_PARSERS_IN_ALTERNATIVE, \
    CAPTURE_WITHOUT_PARSERNAME, CAPTURE_DROPPED_CONTENT_WARNING, LOOKAHEAD_WITH_OPTIONAL_PARSER, \
    BADLY_NESTED_OPTIONAL_PARSER, BAD_ORDER_OF_ALTERNATIVES, BAD_MANDATORY_SETUP, \
45
46
    OPTIONAL_REDUNDANTLY_NESTED_WARNING, CAPTURE_STACK_NOT_EMPTY, BAD_REPETITION_COUNT, \
    AUTORETRIEVED_SYMBOL_NOT_CLEARED, RECURSION_DEPTH_LIMIT_HIT
eckhart's avatar
eckhart committed
47
from DHParser.log import CallItem, HistoryRecord
48
from DHParser.preprocess import BEGIN_TOKEN, END_TOKEN, RX_TOKEN_NAME
49
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
eckhart's avatar
eckhart committed
50
from DHParser.syntaxtree import ChildrenType, Node, RootNode, WHITESPACE_PTYPE, \
51
    TOKEN_PTYPE, ZOMBIE_TAG, EMPTY_NODE, ResultType
52
from DHParser.toolkit import sane_parser_name, escape_ctrl_chars, re, cython, \
53
    abbreviate_middle, RX_NEVER_MATCH, RxPatternType, linebreaks, line_col, identity
54

55

56
__all__ = ('ParserError',
eckhart's avatar
eckhart committed
57
58
59
           'ApplyFunc',
           'FlagFunc',
           'ParseFunc',
60
           'Parser',
61
           'AnalysisError',
62
           'GrammarError',
63
           'Grammar',
64
65
           'Always',
           'Never',
eckhart's avatar
eckhart committed
66
           'AnyChar',
67
           'PreprocessorToken',
68
69
           'Text',
           'DropText',
70
           'RegExp',
71
           'update_scanner',
72
73
           'RE',
           'TKN',
74
           'Whitespace',
75
           'DropRegExp',
76
           'mixin_comment',
di68kap's avatar
di68kap committed
77
           'mixin_nonempty',
di68kap's avatar
di68kap committed
78
           'CombinedParser',
79
           'TreeReduction',
eckhart's avatar
eckhart committed
80
81
           'UnaryParser',
           'NaryParser',
82
           'Drop',
83
84
85
86
           'Synonym',
           'Option',
           'ZeroOrMore',
           'OneOrMore',
87
           'NO_MANDATORY',
Eckhart Arnold's avatar
Eckhart Arnold committed
88
           'MandatoryNary',
89
90
           'Series',
           'Alternative',
91
           'INFINITE',
92
           'Counted',
93
           'Interleave',
eckhart's avatar
eckhart committed
94
           'Required',
95
96
97
98
99
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
           'last_value',
100
101
           'optional_last_value',
           'matching_bracket',
102
103
104
           'Capture',
           'Retrieve',
           'Pop',
105
           'Forward')
106
107
108
109


########################################################################
#
110
# ParserError class
111
112
113
114
#
########################################################################


Eckhart Arnold's avatar
Eckhart Arnold committed
115
116
117
class ParserError(Exception):
    """
    A `ParserError` is thrown for those parser errors that allow the
118
    controlled re-entrance of the parsing process after the error occurred.
Eckhart Arnold's avatar
Eckhart Arnold committed
119
    If a reentry-rule has been configured for the parser where the error
120
    occurred, the parser guard can resume the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
121
122

    Currently, the only case when a `ParserError` is thrown (and not some
123
    different kind of error like `UnknownParserError`) is when a `Series`-
eckhart's avatar
eckhart committed
124
    or `Interleave`-parser detects a missing mandatory element.
Eckhart Arnold's avatar
Eckhart Arnold committed
125
    """
126
127
128
129
130
131
    def __init__(self,
                 parser: 'Parser',
                 node: Node,
                 rest: StringView,
                 error: Error, *,
                 first_throw: bool):
eckhart's avatar
eckhart committed
132
        assert node is not None
133
134
135
136
137
138
        self.parser = parser  # type: 'Parser'
        self.node = node      # type: Node
        self.rest = rest      # type: StringView
        self.error = error    # type: Error
        self.first_throw = first_throw   # type:
        self.attributes_locked = frozenset({'parser', 'node', 'rest', 'error', 'first_throw'})
eckhart's avatar
eckhart committed
139
        self.frozen_callstack = tuple()  # type: Tuple[CallItem, ...]  # tag_name, location
Eckhart Arnold's avatar
Eckhart Arnold committed
140

141
142
143
144
145
146
147
148
149
    def __setattr__(self, name, value):
        if name == "attributes_locked":
            self.__dict__[name] = value
        elif "attributes_locked" not in self.__dict__ \
                or name not in self.__dict__['attributes_locked']:
            self.__dict__[name] = value
        else:
            raise TypeError('Attribute %s of ParserError-object must not be reassigned!' % name)

150
    def __str__(self):
151
152
        return "%i: %s    %s (%s)" \
               % (self.node.pos, str(self.rest[:25]), repr(self.node), str(self.error))
153

Eckhart Arnold's avatar
Eckhart Arnold committed
154

155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
    def new_PE(self, **kwargs):
        """Returns a new ParserError object with the same attribute values
        as `self`, except those that are reassigned in `**kwargs`.

        >>> pe = ParserError(Parser(), Node('test', ""), StringView(""), Error("", 0), first_throw=True)
        >>> pe_derived = pe.new_PE(first_throw = False)
        >>> pe.first_throw
        True
        >>> pe_derived.first_throw
        False
        """
        args = { "parser": self.parser,
                 "node": self.node,
                 "rest": self.rest,
                 "error": self.error,
                 "first_throw": self.first_throw }
        assert len(kwargs.keys() - args.keys()) == 0, str(kwargs.keys() - args.keys())
        args.update(kwargs)
        pe = ParserError(**args)
        pe.frozen_callstack = self.frozen_callstack
        return pe


178
179
180
PatternMatchType = Union[RxPatternType, str, Callable]
ErrorMessagesType = List[Tuple[PatternMatchType, str]]
ResumeList = List[PatternMatchType]  # list of strings or regular expressions
181
182
ReentryPointAlgorithm = Callable[[StringView, int, int], Tuple[int, int]]
# (text, start point, end point) => (reentry point, match length)
183
# A return value of (-1, x) means that no reentry point before the end of the document was found
Eckhart Arnold's avatar
Eckhart Arnold committed
184

185

186
@cython.returns(cython.int)
eckhart's avatar
eckhart committed
187
@cython.locals(upper_limit=cython.int, closest_match=cython.int, pos=cython.int)
188
189
190
def reentry_point(rest: StringView,
                  rules: ResumeList,
                  comment_regex,
191
                  search_window: int = -1) -> int:
Eckhart Arnold's avatar
Eckhart Arnold committed
192
193
    """
    Finds the point where parsing should resume after a ParserError has been caught.
194
    The algorithm makes sure that this reentry-point does not lie inside a comment.
195
196
197
    The re-entry point is always the point after the end of the match of the regular
    expression defining the re-entry point. (Use look ahead, if you wand to define
    the re-entry point by what follows rather than by what text precedes the point.)
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223

    REMARK: The algorithm assumes that any stretch of the document that matches
    `comment_regex` is actually a comment. It is possible to define grammars,
    where the use of comments is restricted to certain areas and that allow to
    use constructs that look like comments (i.e. will be matched by `comment_regex`)
    but are none in other areas. For example::

            my_string = "# This is not a comment"; foo()  # This is a comment
            bar()

    Here the reentry-algorithm would overlook `foo()` and jump directly to `bar()`.
    However, since the reentry-algorithm only needs to be good enough to do its
    work, this seems acceptable.

    :param rest:  The rest of the parsed text or, in other words, the point where
        a ParserError was thrown.
    :param rules: A list of strings, regular expressions or search functions.
        The rest of the text is searched for each of these. The closest match
        is the point where parsing will be resumed.
    :param comment_regex: A regular expression object that matches comments.
    :param search_window: The maximum size of the search window for finding the
        reentry-point. A value smaller than zero means that the complete remaining
        text will be searched. A value of zero effectively turns of resuming after
        error.
    :return: The integer index of the closest reentry point or -1 if no
        reentry-point was found.
Eckhart Arnold's avatar
Eckhart Arnold committed
224
    """
225
    upper_limit = len(rest) + 1
226
    closest_match = upper_limit
227
228
229
    comments = None  # type: Optional[Iterator]
    if search_window < 0:
        search_window = len(rest)
230

eckhart's avatar
eckhart committed
231
    @cython.locals(a=cython.int, b=cython.int)
232
    def next_comment() -> Tuple[int, int]:
233
234
235
236
        """Returns the [start, end[ intervall of the next comment in the text.
        The comment-iterator start at the beginning of the `rest` of the
        document and is reset for each search rule.
        """
237
238
239
240
241
242
243
244
245
246
        nonlocal rest, comments
        if comments:
            try:
                m = next(comments)
                a, b = m.span()
                return rest.index(a), rest.index(b)
            except StopIteration:
                comments = None
        return -1, -2

247
248
    @cython.locals(start=cython.int)
    def str_search(s, start: int = 0) -> Tuple[int, int]:
249
250
251
252
253
        """Returns the starting position of the next occurrence of `s` in
        the `rest` of the document beginning with `start` and the length
        of the match, which in this case is always the length of `s` itself.
        If their is no match, the returned starting position will be -1.
        """
254
255
        nonlocal rest
        return rest.find(s, start, start + search_window), len(s)
256

eckhart's avatar
eckhart committed
257
    @cython.locals(start=cython.int, end=cython.int)
258
    def rx_search(rx, start: int = 0) -> Tuple[int, int]:
259
260
261
262
263
        """Returns the staring position and the length of the next match of
        the regular expression `rx` in the `rest` of the document, starting
        with `start`.
        If their is no match, the returned starting position will be -1.
        """
264
        nonlocal rest
265
        m = rest.search(rx, start, start + search_window)
266
        if m:
267
268
            begin, end = m.span()
            return rest.index(begin), end - begin
269
270
        return -1, 0

271
272
273
274
    def algorithm_search(func: ReentryPointAlgorithm, start: int = 0):
        """Returns the next match as a tuple of position and length that
        the reentry-point-search-function `func` yields.
        """
275
        nonlocal rest
276
        return func(rest, start, start + search_window)
277

278
    @cython.returns(cython.int)
eckhart's avatar
eckhart committed
279
    @cython.locals(a=cython.int, b=cython.int, k=cython.int, length=cython.int)
280
    def entry_point(search_func, search_rule) -> int:
281
282
283
        """Returns the next reentry-point outside a comment that `search_func`
        yields. If no reentry point is found, the first position after the
        end of the text ("upper limit") is returned."""
284
285
        a, b = next_comment()
        k, length = search_func(search_rule)
eckhart's avatar
eckhart committed
286
        while a < b <= k + length:
287
            a, b = next_comment()
288
289
290
        # find next as long as start or end point of resume regex are inside a comment
        while (a < k < b) or (a < k + length < b):
            k, length = search_func(search_rule, b)
291
292
            while a < b <= k:
                a, b = next_comment()
293
        return k + length if k >= 0 else upper_limit
294

295
    # find closest match
Eckhart Arnold's avatar
Eckhart Arnold committed
296
    for rule in rules:
297
        comments = rest.finditer(comment_regex)
298
299
300
301
302
303
304
        if callable(rule):
            search_func = algorithm_search
        elif isinstance(rule, str):
            search_func = str_search
        else:
            search_func = rx_search
        pos = entry_point(search_func, rule)
305
        closest_match = min(pos, closest_match)
306

307
    # in case no rule matched return -1
308
309
310
    if closest_match == upper_limit:
        closest_match = -1
    return closest_match
Eckhart Arnold's avatar
Eckhart Arnold committed
311
312


313
314
315
316
317
318
319
########################################################################
#
# Parser base class
#
########################################################################


320
ApplyFunc = Callable[[List['Parser']], Optional[bool]]
321
# The return value of `True` stops any further application
eckhart's avatar
eckhart committed
322
FlagFunc = Callable[[ApplyFunc, Set[ApplyFunc]], bool]
eckhart's avatar
eckhart committed
323
ParseFunc = Callable[[StringView], Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
324
325


326
class Parser:
327
328
329
330
331
332
333
334
335
336
337
338
339
340
    """
    (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.

    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:

di68kap's avatar
di68kap committed
341
    1. *Named parsers* for which a name is set in field `parser.pname`.
342
343
344
       The results produced by these parsers can later be retrieved in
       the AST by the parser name.

345
346
    2. *Disposable parsers* where the name-field just contains the empty
       string. AST-transformation of disposable parsers can be hooked
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
       only to their class name, and not to the individual parser.

    Parser objects are callable and parsing is done by calling a parser
    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 `Option`). 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.

    Attributes and Properties:
365

366
        pname:  The parser's name.
367

368
369
        disposable: A property indicating that the parser returns
                anonymous nodes. For performance
370
                reasons this is implemented as an object variable rather
371
                than a property. This property should always be equal to
372
                `self.tag_name[0] == ":"`.
373

374
375
376
377
        drop_content: A property (for performance reasons implemented as
                simple field) that, if set, induces the parser not to return
                the parsed content or sub-tree if it has matched but the
                dummy `EMPTY_NODE`. In effect the parsed content will be
378
379
380
                dropped from the concrete syntax tree already. Only
                anonymous (or pseudo-anonymous) parsers are allowed to
                drop content.
381

382
        tag_name: The tag_name for the nodes that are created by
383
                the parser. If the parser is named, this is the same as
di68kap's avatar
di68kap committed
384
385
                `pname`, otherwise it is the name of the parser's type
                prefixed with a colon ":".
386

387
388
389
390
391
392
393
394
395
396
        visited:  Mapping of places this parser has already been to
                during the current parsing process onto the results the
                parser returned at the respective place. This dictionary
                is used to implement memoizing.

        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.

397
398
399
400
        proxied: The original `_parse()`-method is stored here, if a
                proxy (e.g. a tracing debugger) is installed via the
                `set_proxy()`-method.

401
        _grammar:  A reference to the Grammar object to which the parser
402
                is attached.
403
404
405
406
407

        _symbol:  The name of the closest named parser to which this
                parser is connected in a grammar. If pname is not the
                empty string, this will become the same as pname, when
                the property `symbol` is read for the first time.
408
409
    """

410
    def __init__(self) -> None:
411
        # assert isinstance(name, str), str(name)
412
        self.pname = ''               # type: str
413
        self.disposable = True         # type: bool
414
        self.drop_content = False     # type: bool
415
416
        self.tag_name = self.ptype    # type: str
        self.cycle_detection = set()  # type: Set[ApplyFunc]
417
        # this indirection is required for Cython-compatibility
418
        self._parse_proxy = self._parse  # type: ParseFunc
eckhart's avatar
eckhart committed
419
        # self.proxied = None           # type: Optional[ParseFunc]
420
421
422
423
        try:
            self._grammar = GRAMMAR_PLACEHOLDER  # type: Grammar
        except NameError:
            pass
424
        self._symbol = ''             # type: str
425
426
427
        self.reset()

    def __deepcopy__(self, memo):
428
        """Deepcopy method of the parser. Upon instantiation of a Grammar-
429
        object, parsers will be deep-copied to the Grammar object. If a
430
        derived parser-class changes the signature of the `__init__`-constructor,
431
432
433
        `__deepcopy__`-method must be replaced (i.e. overridden without
        calling the same method from the superclass) by the derived class.
        """
434
        duplicate = self.__class__()
435
        copy_parser_base_attrs(self, duplicate)
436
        return duplicate
437

438
    def __repr__(self):
di68kap's avatar
di68kap committed
439
        return self.pname + self.ptype
440
441

    def __str__(self):
di68kap's avatar
di68kap committed
442
        return self.pname + (' = ' if self.pname else '') + repr(self)
443

444
445
446
447
448
449
    @property
    def ptype(self) -> str:
        """Returns a type name for the parser. By default this is the name of
        the parser class with an added leading colon ':'. """
        return ':' + self.__class__.__name__

450
451
452
453
454
    @property
    def symbol(self) -> str:
        """Returns the symbol with which the parser is associated in a grammar.
        This is the closest parser with a pname that contains this parser."""
        if not self._symbol:
eckhart's avatar
eckhart committed
455
            try:
456
                self._symbol = self.grammar.associated_symbol__(self).pname
eckhart's avatar
eckhart committed
457
458
459
460
            except AttributeError:
                # return an empty string, if parser is not connected to grammar,
                # but be sure not to save the empty string in self._symbol
                return ''
461
462
        return self._symbol

463
464
465
    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
di68kap's avatar
di68kap committed
466
        return self.pname if self.pname else self.__repr__()
467

468
469
470
471
    def reset(self):
        """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."""
472
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
473

eckhart's avatar
eckhart committed
474
    @cython.locals(location=cython.int, gap=cython.int, i=cython.int)
475
476
477
    def __call__(self: 'Parser', text: StringView) -> Tuple[Optional[Node], StringView]:
        """Applies the parser to the given text. This is a wrapper method that adds
        the business intelligence that is common to all parsers. The actual parsing is
478
479
        done in the overridden method `_parse()`. This wrapper-method can be thought of
        as a "parser guard", because it guards the parsing process.
480
        """
481
        grammar = self._grammar
di68kap's avatar
di68kap committed
482
        location = grammar.document_length__ - text._len  # faster then len(text)?
483
484

        try:
485
486
            # rollback variable changing operation if parser backtracks to a position
            # before or at the location where the variable changing operation occurred
487
            if location <= grammar.last_rb__loc__:
488
489
                grammar.rollback_to__(location)

490
            # if location has already been visited by the current parser, return saved result
491
492
            visited = self.visited  # using local variable for better performance
            if location in visited:
493
                # no history recording in case of memoized results!
494
                return visited[location]
495

496
497
            memoization_state = grammar.suspend_memoization__
            grammar.suspend_memoization__ = False
498

499
            # now, the actual parser call!
500
            try:
Eckhart Arnold's avatar
Eckhart Arnold committed
501
                node, rest = self._parse_proxy(text)
502
            except ParserError as pe:
503
                # catching up with parsing after an error occurred
504
                gap = len(text) - len(pe.rest)
505
                rules = grammar.resume_rules__.get(
506
                    self.pname or grammar.associated_symbol__(self).pname, [])
507
                rest = pe.rest[len(pe.node):]
508
509
                i = reentry_point(rest, rules, grammar.comment_rx__,
                                  grammar.reentry_search_window__)
510
                if i >= 0 or self == grammar.start_parser__:
eckhart's avatar
eckhart committed
511
512
                    # either a reentry point was found or the
                    # error has fallen through to the first level
513
                    assert pe.node.children or (not pe.node.result)
514
                    # apply reentry-rule or catch error at root-parser
eckhart's avatar
eckhart committed
515
                    if i < 0:  i = 0
516
                    try:
517
                        zombie = pe.node.pick_child(ZOMBIE_TAG)  # type: Optional[Node]
518
                    except (KeyError, ValueError):
519
520
521
                        zombie = None
                    if zombie and not zombie.result:
                        zombie.result = rest[:i]
eckhart's avatar
eckhart committed
522
                        tail = tuple()  # type: ChildrenType
523
524
                    else:
                        nd = Node(ZOMBIE_TAG, rest[:i]).with_pos(location)
eckhart's avatar
eckhart committed
525
                        # nd.attr['err'] = pe.error.message
526
                        tail = (nd,)
527
                    rest = rest[i:]
528
529
                    if pe.first_throw:
                        node = pe.node
530
                        node.result = node.children + tail
531
                    else:
532
533
                        node = Node(
                            self.tag_name,
di68kap's avatar
di68kap committed
534
535
                            (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) + tail) \
                            .with_pos(location)
eckhart's avatar
eckhart committed
536
                # if no re-entry point was found, do any of the following:
537
                elif pe.first_throw:
eckhart's avatar
eckhart committed
538
                    # just fall through
eckhart's avatar
eckhart committed
539
                    # TODO: Is this case still needed with module "trace"?
540
                    raise pe.new_PE(first_throw=False)
541
                elif grammar.tree__.errors[-1].code == MANDATORY_CONTINUATION_AT_EOF:
eckhart's avatar
eckhart committed
542
543
                    # try to create tree as faithful as possible
                    node = Node(self.tag_name, pe.node).with_pos(location)
544
                else:
eckhart's avatar
eckhart committed
545
                    # fall through but skip the gap
546
547
                    result = (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) if gap \
                        else pe.node  # type: ResultType
548
549
                    raise pe.new_PE(node=Node(self.tag_name, result).with_pos(location),
                                    rest=text, first_throw=False)
550

551
            if node is not None:
552
                node._pos = location
553
554
555
            if not grammar.suspend_memoization__:
                visited[location] = (node, rest)
                grammar.suspend_memoization__ = memoization_state
556

557
        except RecursionError:
558
            node = Node(ZOMBIE_TAG, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
559
            node._pos = location
560
561
562
            error = Error("maximum recursion depth of parser reached; potentially due to too many "
                          "errors or left recursion!", location, RECURSION_DEPTH_LIMIT_HIT)
            grammar.tree__.add_error(node, error)
563
            grammar.most_recent_error__ = ParserError(self, node, text, error, first_throw=False)
564
565
566
            rest = EMPTY_STRING_VIEW

        return node, rest
567
568
569
570

    def __add__(self, other: 'Parser') -> 'Series':
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
di68kap's avatar
di68kap committed
571
572
        if isinstance(other, Series):
            return cast('Series', other).__radd__(self)
573
574
575
576
577
578
        return Series(self, other)

    def __or__(self, other: 'Parser') -> 'Alternative':
        """The | operator generates an alternative parser that applies
        the first parser and, if that does not match, the second parser.
        """
di68kap's avatar
di68kap committed
579
580
        if isinstance(other, Alternative):
            return cast('Alternative', other).__ror__(self)
581
582
        return Alternative(self, other)

583
    def __mul__(self, other: 'Parser') -> 'Interleave':
di68kap's avatar
di68kap committed
584
585
586
587
588
589
590
591
        """The * operator generates an interleave parser that applies
        the first parser and the second parser in any possible order
        until both match.
        """
        if isinstance(other, Interleave):
            return cast(Interleave, other).__rmul__(self)
        return Interleave(self, other)

592
593
594
595
596
597
    def _parse(self, text: StringView) -> Tuple[Optional[Node], StringView]:
        """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."""
        raise NotImplementedError

eckhart's avatar
eckhart committed
598
    def is_optional(self) -> Optional[bool]:
599
        """Returns `True`, if the parser can never fail, i.e. never yields
eckhart's avatar
eckhart committed
600
601
602
603
604
        `None`, instead of a node. Returns `False`, if the parser can fail.
        Returns `None` if it is not known whether the parser can fail.
        """
        return None

605
    def set_proxy(self, proxy: Optional[ParseFunc]):
606
607
608
        """Sets a proxy that replaces the _parse()-method. Call `set_proxy`
        with `None` to remove a previously set proxy. Typical use case is
        the installation of a tracing debugger. See module `trace`.
609
610
        """
        if proxy is None:
Eckhart Arnold's avatar
Eckhart Arnold committed
611
            self._parse_proxy = self._parse
612
        else:
eckhart's avatar
eckhart committed
613
614
            if not isinstance(proxy, type(self._parse)):
                # assume that proxy is a function and bind it to self
615
616
                proxy = proxy.__get__(self, type(self))
            else:
eckhart's avatar
eckhart committed
617
                # if proxy is a method it must be a method of self
618
                assert proxy.__self__ == self
Eckhart Arnold's avatar
Eckhart Arnold committed
619
            self._parse_proxy = cast(ParseFunc, proxy)
620

621
    def name(self, pname: str) -> 'Parser':
622
623
624
        """Sets the parser name to `pname` and returns `self`."""
        self.pname = pname
        self.tag_name = pname or self.ptype
625
        self.disposable = not self.pname
626
627
        return self

628
    @property
eckhart's avatar
eckhart committed
629
    def grammar(self) -> 'Grammar':
630
        try:
eckhart's avatar
eckhart committed
631
            if self._grammar is not GRAMMAR_PLACEHOLDER:
632
633
                return self._grammar
            else:
eckhart's avatar
eckhart committed
634
                raise ValueError('Grammar has not yet been set!')
635
636
        except (AttributeError, NameError):
            raise AttributeError('Parser placeholder does not have a grammar!')
637
638
639

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
640
641
642
643
644
        try:
            if self._grammar == GRAMMAR_PLACEHOLDER:
                self._grammar = grammar
                # self._grammar_assigned_notifier()
            elif self._grammar != grammar:
eckhart's avatar
eckhart committed
645
646
                raise AssertionError("Parser has already been assigned"
                                     "to a different Grammar object!")
647
648
        except AttributeError:
            pass  # ignore setting of grammar attribute for placeholder parser
649
        except NameError:  # Cython: No access to GRAMMAR_PLACEHOLDER, yet :-(
650
            self._grammar = grammar
651

eckhart's avatar
eckhart committed
652
    def sub_parsers(self) -> Tuple['Parser', ...]:
653
654
655
        """Returns the list of sub-parsers if there are any.
        Overridden by Unary, Nary and Forward.
        """
eckhart's avatar
eckhart committed
656
        return tuple()
657

658
    def _apply(self, func: ApplyFunc, context: List['Parser'], flip: FlagFunc) -> bool:
659
660
        """
        Applies function `func(parser)` recursively to this parser and all
661
662
        descendant parsers as long as `func()` returns `None` or `False`.
        Otherwise stops the further application of `func` and returns `True`.
eckhart's avatar
eckhart committed
663
664
665
666
667
668
669
670

        In order to break cycles, function `flip` is called, which should
        return `True`, if this parser has already been visited. If not, it
        flips the cycle detection flag and returns `False`.

        This is a protected function and should not called from outside
        class Parser or any of its descendants. The entry point for external
        calls is the method `apply()` without underscore!
671
        """
672
673
        if not flip(func, self.cycle_detection):
            if func(context + [self]):
674
675
676
                return True
            else:
                for parser in self.sub_parsers():
677
                    if parser._apply(func, context + [self], flip):
678
679
680
                        return True
                return False
        return False
681

682
    def apply(self, func: ApplyFunc) -> Optional[bool]:
eckhart's avatar
eckhart committed
683
684
        """
        Applies function `func(parser)` recursively to this parser and all
685
686
687
688
689
690
691
692
693
694
695
696
697
698
        descendant parsers as long as `func()` returns `None` or `False`.
        Traversal is pre-order. Stops the further application of `func` and
        returns `True` once `func` has returned `True`.

        If `func` has been applied to all descendant parsers without issuing
        a stop signal by returning `True`, `False` is returned.

        This use of the return value allows to use the `apply`-method both
        to issue tests on all descendant parsers (including self) which may be
        decided already after some parsers have been visited without any need
        to visit further parsers. At the same time `apply` can be used to simply
        `apply` a procedure to all descendant parsers (including self) without
        worrying about forgetting the return value of procedure, because a
        return value of `None` means "carry on".
eckhart's avatar
eckhart committed
699
        """
700
        def positive_flip(f: ApplyFunc, flagged: AbstractSet[ApplyFunc]) -> bool:
eckhart's avatar
eckhart committed
701
702
703
704
705
706
707
708
709
            """Returns True, if function `f` has already been applied to this
            parser and sets the flag accordingly. Interprets `f in flagged == True`
            as meaning that `f` has already been applied."""
            if f in flagged:
                return True
            else:
                flagged.add(f)
                return False

710
        def negative_flip(f: ApplyFunc, flagged: AbstractSet[ApplyFunc]) -> bool:
eckhart's avatar
eckhart committed
711
712
713
714
715
716
717
718
719
720
            """Returns True, if function `f` has already been applied to this
            parser and sets the flag accordingly. Interprets `f in flagged == False`
            as meaning that `f` has already been applied."""
            if f not in flagged:
                return True
            else:
                flagged.remove(f)
                return False

        if func in self.cycle_detection:
721
            return self._apply(func, [], negative_flip)
eckhart's avatar
eckhart committed
722
        else:
723
            return self._apply(func, [], positive_flip)
724

725
    def static_error(self, msg: str, code: ErrorCode) -> 'AnalysisError':
726
        return self.symbol, self, Error(msg, 0, code)
727

eckhart's avatar
eckhart committed
728
    def static_analysis(self) -> List['AnalysisError']:
729
730
        """Analyses the parser for logical errors after the grammar has been
        instantiated."""
eckhart's avatar
eckhart committed
731
        return []
eckhart's avatar
eckhart committed
732

733

734
def copy_parser_base_attrs(src: Parser, duplicate: Parser):
735
    """Duplicates all attributes of the Parser-class from `src` to `duplicate`."""
Eckhart Arnold's avatar
Eckhart Arnold committed
736
    duplicate.pname = src.pname
737
    duplicate.disposable = src.disposable
Eckhart Arnold's avatar
Eckhart Arnold committed
738
739
740
741
    duplicate.drop_content = src.drop_content
    duplicate.tag_name = src.tag_name


eckhart's avatar
eckhart committed
742
743
def Drop(parser: Parser) -> Parser:
    """Returns the parser with the `parser.drop_content`-property set to `True`."""
744
    assert parser.disposable, "Parser must be anonymous to be allowed to drop ist content."
eckhart's avatar
eckhart committed
745
746
747
    if isinstance(parser, Forward):
        cast(Forward, parser).parser.drop_content = True
    parser.drop_content = True
Eckhart Arnold's avatar
Eckhart Arnold committed
748
    return parser
eckhart's avatar
eckhart committed
749
750


751
PARSER_PLACEHOLDER = Parser()
752
753


eckhart's avatar
eckhart committed
754
755
756
757
758
def is_parser_placeholder(parser: Optional[Parser]) -> bool:
    """Returns True, if `parser` is `None` or merely a placeholder for a parser."""
    return not parser or parser.ptype == ":Parser"


eckhart's avatar
eckhart committed
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
# functions for analysing the parser tree/graph ###


def has_non_autocaptured_symbols(context: List[Parser]) -> Optional[bool]:
    """Returns True, if the context contains a Capture-Parser that is not
    shielded by a Retrieve-Parser. This is the case for captured symbols
    that are not "auto-captured" by a Retrieve-Parser.
    """
    for parser in context:
        if parser.ptype == ":Retrieve":
            break
        elif parser.ptype == ":Capture":
            p = cast(UnaryParser, parser).parser
            while p.ptype in (":Synonym", ":Forward"):
                p = cast(UnaryParser, p).parser
            if not isinstance(p, Retrieve):
                return True
    return None


779
780
781
782
783
784
########################################################################
#
# Grammar class, central administration of all parser of a grammar
#
########################################################################

785
786
def mixin_comment(whitespace: str, comment: str) -> str:
    """
787
    Returns a regular expression pattern that merges comment and whitespace
788
    regexps. Thus comments can occur wherever whitespace is allowed
789
790
791
792
793
794
    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).
    """
795
    if comment:
eckhart's avatar
eckhart committed
796
797
        whitespace = '(?:' + whitespace + ')'
        comment = '(?:' + comment + ')'
798
799
        return '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return whitespace
800
801


di68kap's avatar
di68kap committed
802
def mixin_nonempty(whitespace: str) -> str:
Eckhart Arnold's avatar
Eckhart Arnold committed
803
    r"""
804
805
806
    Returns a regular expression pattern that matches only if the regular
    expression pattern `whitespace` matches AND if the match is not empty.

di68kap's avatar
di68kap committed
807
808
    If `whitespace`  does not match the empty string '', anyway,
    then it will be returned unaltered.
809
810
811
812
813
814
815
816
817

    WARNING: `non_empty_ws` does not work regular expressions the matched
    strings of which can be followed by a symbol that can also occur at
    the start of the regular expression.

    In particular, it does not work for fixed size regular expressions,
    that ist / / or /   / or /\t/ won't work, but / */ or /\s*/ or /\s+/
    do work. There is no test for this. Fixed sizes regular expressions
    run through `non_empty_ws` will not match at any more if they are applied
Eckhart Arnold's avatar
Eckhart Arnold committed
818
819
820
821
822
    to the beginning or the middle of a sequence of whitespaces!

    In order to be safe, you whitespace regular expressions should follow
    the rule: "Whitespace cannot be followed by whitespace" or "Either
    grab it all or leave it all".
823
824
825
826
827
828
829
830
831
832

    :param whitespace: a regular expression pattern
    :return: new regular expression pattern that does not match the empty
        string '' any more.
    """
    if re.match(whitespace, ''):
        return r'(?:(?=(.|\n))' + whitespace + r'(?!\1))'
    return whitespace


833
834
AnalysisError = Tuple[str, Parser, Error]      # pname, parser, error
# TODO: replace with a named tuple?
835
836
837
838
839
840


class GrammarError(Exception):
    """GrammarError will be raised if static analysis reveals errors
    in the grammar.
    """
841
    def __init__(self, static_analysis_result: List[AnalysisError]):
842
843
844
845
846
847
848
849
850
851
        assert static_analysis_result  # must not be empty
        self.errors = static_analysis_result

    def __str__(self):
        if len(self.errors) == 1:
            return str(self.errors[0][2])
        return '\n' + '\n'.join(("%i. " % (i + 1) + str(err_tuple[2]))
                                for i, err_tuple in enumerate(self.errors))


852
853
854
855
856
857
858
859
860
861
862
863
864
class Grammar:
    r"""
    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.

eckhart's avatar
eckhart committed
865
    Example for direct instantiation of a grammar::
866

867
        >>> number = RE(r'\d+') + RE(r'\.') + RE(r'\d+') | RE(r'\d+')
868
869
870
871
872
873
874
875
876
877
878
879
880
881
        >>> number_parser = Grammar(number)
        >>> number_parser("3.1416").content
        '3.1416'

    Collecting the parsers that define a grammar in a descendant 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, which results in better
       readability of the Python code.
882
       See classmethod `Grammar._assign_parser_names__()`
883
884

    3. The parsers in the class do not necessarily need to be connected
885
886
       to one single root parser, which is helpful for testing and when
       building up a parser gradually from several components.
887
888
889
890
891
892
893

    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.

894
    Example::
895
896
897
898
899
900
901
902
903

        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+')
904
905
906
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
907
908
909
910
            root__ = expression

    Upon instantiation the parser objects are deep-copied to the
    Grammar object and assigned to object variables of the same name.
911
912
913
914
    For any parser that is directly assigned to a class variable the
    field `parser.pname` contains the variable name after instantiation
    of the Grammar class. The parser will never the less remain anonymous
    with respect to the tag names of the nodes it generates, if its name
915
    is matched by the `disposable__` regular expression.
916
    If one and the same parser is assigned to several class variables
917
    such as, for example, the parser `expression` in the example above,
918
    which is also assigned to `root__`, the first name sticks.
919
920
921
922
923
924
925
926
927
928
929
930

    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 classes contain a few special class fields for implicit
    whitespace and comments that should be overwritten, if the defaults
    (no comments, horizontal right aligned whitespace) don't fit:

931
    Class Attributes:
932
933
934
935
        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.
936

937
938
939
        resume_rules__: A mapping of parser names to a list of regular expressions
                that act as rules to find the reentry point if a ParserError was
                thrown during the execution of the parser with the respective name.
Eckhart Arnold's avatar
Eckhart Arnold committed
940

941
942
943
944
945
946
947
948
949
950
951
952
953
        skip_rules__: A mapping of parser names to a list of regular expressions
                that act as rules to find the reentry point if a ParserError was
                thrown during the execution of the parser with the respective name.

        error_messages__: A mapping of parser names to a Tuple of regalar expressions
                and error messages. If a mandatory violation error occurs on a
                specific symbol (i.e. parser name) and any of the regular expressions
                matches the error message of the first matching expression is used
                instead of the generic manadtory violation error messages. This
                allows to answer typical kinds of errors (say putting a colon ","
                where a semi-colon ";" is expected) with more informative error
                messages.

954
        disposable__: A regular expression to identify names of parsers that are
955
956
                assigned to class fields but shall never the less yield anonymous
                nodes (i.e. nodes the tag name of which starts with a colon ":"
957
                followed by the parser's class name).
958

eckhart's avatar
eckhart committed
959
        parser_initialization__:  Before the grammar class (!) has been initialized,
960
                 which happens upon the first time it is instantiated (see
961
                 `:func:_assign_parser_names()` for an explanation), this class
962
963
                 field contains a value other than "done". A value of "done" indicates
                 that the class has already been initialized.
964

965
966
967
968
969
970
971
        static_analysis_pending__: True as long as no static analysis (see the method
                with the same name for more information) has been done to check
                parser tree for correctness. Static analysis
                is done at instantiation and the flag is then set to false, but it
                can also be carried out once the class has been generated
                (by DHParser.ebnf.EBNFCompiler) and then be set to false in the
                definition of the grammar class already.
972

973
974
975
        static_analysis_errors__: A list of errors and warnings that were found in the
                static analysis

976
        python_src__:  For the purpose of debugging and inspection, this field can
eckhart's avatar
eckhart committed
977
978
979
                 take the python src of the concrete grammar class
                 (see `dsl.grammar_provider`).

980
    Instance Attributes:
981
982
        all_parsers__:  A set of all parsers connected to this grammar object

983
984
        comment_rx__:  The compiled regular expression for comments. If no
                comments have been defined, it defaults to RX_NEVER_MATCH
985
986
                This instance-attribute will only be defined if a class-attribute
                with the same name does not already exist!
987

eckhart's avatar
eckhart committed
988
989
990
991
        start_parser__:  During parsing, the parser with which the parsing process
                was started (see method `__call__`) or `None` if no parsing process
                is running.

992
993
994
995
996
997
998
        _dirty_flag__:  A flag indicating that the Grammar has been called at
                least once so that the parsing-variables need to be reset
                when it is called again.

        document__:  the text that has most recently been parsed or that is
                currently being parsed.

eckhart's avatar
eckhart committed
999
1000
        document_length__:  the length of the document.