parse.py 123 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
eckhart's avatar
eckhart committed
35
from typing import Callable, cast, List, Tuple, Set, Dict, \
36
    DefaultDict, Sequence, Union, Optional, Any
37

eckhart's avatar
eckhart committed
38
from DHParser.configuration import get_config_value
39
from DHParser.error import Error
eckhart's avatar
eckhart committed
40
from DHParser.log import CallItem, HistoryRecord
41
from DHParser.preprocess import BEGIN_TOKEN, END_TOKEN, RX_TOKEN_NAME
42
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
eckhart's avatar
eckhart committed
43
from DHParser.syntaxtree import ChildrenType, Node, RootNode, WHITESPACE_PTYPE, \
44
    TOKEN_PTYPE, ZOMBIE_TAG, EMPTY_NODE, ResultType
45
from DHParser.toolkit import sane_parser_name, escape_control_characters, re, cython, \
46
    abbreviate_middle, RX_NEVER_MATCH, RxPatternType, linebreaks, line_col
47

48
__all__ = ('ParserError',
eckhart's avatar
eckhart committed
49
50
51
           'ApplyFunc',
           'FlagFunc',
           'ParseFunc',
52
           'Parser',
53
           'UnknownParserError',
54
55
           'GrammarErrorType',
           'GrammarError',
56
57
           'Grammar',
           'PreprocessorToken',
58
           'Token',
59
           'DropToken',
60
           'RegExp',
61
62
           'RE',
           'TKN',
63
           'Whitespace',
64
           'DropRegExp',
65
           'mixin_comment',
di68kap's avatar
di68kap committed
66
           'mixin_nonempty',
67
           'MetaParser',
eckhart's avatar
eckhart committed
68
69
           'UnaryParser',
           'NaryParser',
70
           'Drop',
71
72
73
74
           'Synonym',
           'Option',
           'ZeroOrMore',
           'OneOrMore',
75
           'NO_MANDATORY',
Eckhart Arnold's avatar
Eckhart Arnold committed
76
           'MandatoryNary',
77
78
79
80
81
           'Series',
           'Alternative',
           'AllOf',
           'SomeOf',
           'Unordered',
82
           'INFINITE',
83
           'Interleave',
eckhart's avatar
eckhart committed
84
           'Required',
85
86
87
88
89
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
           'last_value',
90
91
           'optional_last_value',
           'matching_bracket',
92
93
94
           'Capture',
           'Retrieve',
           'Pop',
95
           'Forward')
96
97
98
99


########################################################################
#
100
# ParserError class
101
102
103
104
#
########################################################################


Eckhart Arnold's avatar
Eckhart Arnold committed
105
106
107
class ParserError(Exception):
    """
    A `ParserError` is thrown for those parser errors that allow the
108
    controlled re-entrance of the parsing process after the error occurred.
Eckhart Arnold's avatar
Eckhart Arnold committed
109
    If a reentry-rule has been configured for the parser where the error
110
    occurred, the parser guard can resume the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
111
112

    Currently, the only case when a `ParserError` is thrown (and not some
113
114
    different kind of error like `UnknownParserError`) is when a `Series`-
    or `AllOf`-parser detects a missing mandatory element.
Eckhart Arnold's avatar
Eckhart Arnold committed
115
    """
eckhart's avatar
eckhart committed
116
    def __init__(self, node: Node, rest: StringView, error: Error, first_throw: bool):
eckhart's avatar
eckhart committed
117
        assert node is not None
eckhart's avatar
eckhart committed
118
119
120
        self.node = node    # type: Node
        self.rest = rest    # type: StringView
        self.error = error  # type: Error
121
        self.first_throw = first_throw  # type: bool
eckhart's avatar
eckhart committed
122
        self.frozen_callstack = tuple()  # type: Tuple[CallItem, ...]  # tag_name, location
Eckhart Arnold's avatar
Eckhart Arnold committed
123

124
    def __str__(self):
125
        return "%i: %s    %s" % (self.node.pos, str(self.rest[:25]), repr(self.node))
126

Eckhart Arnold's avatar
Eckhart Arnold committed
127

128
ResumeList = List[RxPatternType]  # list of regular expressiones
Eckhart Arnold's avatar
Eckhart Arnold committed
129
130


eckhart's avatar
eckhart committed
131
@cython.locals(upper_limit=cython.int, closest_match=cython.int, pos=cython.int)
132
133
134
def reentry_point(rest: StringView,
                  rules: ResumeList,
                  comment_regex,
135
                  search_window: int = -1) -> int:
Eckhart Arnold's avatar
Eckhart Arnold committed
136
137
    """
    Finds the point where parsing should resume after a ParserError has been caught.
138
    The algorithm makes sure that this reentry-point does not lie inside a comment.
139
140
141
    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.)
Eckhart Arnold's avatar
Eckhart Arnold committed
142
143
    Args:
        rest:  The rest of the parsed text or, in other words, the point where
144
145
146
147
            a ParserError was thrown.
        rules: A list of regular expressions. The rest of the text is searched for
            each of these. The closest match is the point where parsing will be
            resumed.
148
        comment_regex: A regular expression object that matches comments.
149
150
        search_window: The maximum size of the search window for finding the
            reentry-point. A value smaller or equal zero means that
Eckhart Arnold's avatar
Eckhart Arnold committed
151
    Returns:
152
153
        The integer index of the closest reentry point or -1 if no reentry-point
        was found.
Eckhart Arnold's avatar
Eckhart Arnold committed
154
    """
155
    upper_limit = len(rest) + 1
156
    closest_match = upper_limit
157
158
    comments = None  # typ: Optional[Iterator]

eckhart's avatar
eckhart committed
159
    @cython.locals(a=cython.int, b=cython.int)
160
161
162
163
164
165
166
167
168
169
170
    def next_comment() -> Tuple[int, int]:
        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

171
172
    # def str_search(s, start: int = 0) -> Tuple[int, int]:
    #     nonlocal rest
173
    #     return rest.find(s, start, start + search_window), len(s)
174

eckhart's avatar
eckhart committed
175
    @cython.locals(start=cython.int, end=cython.int)
176
177
    def rx_search(rx, start: int = 0) -> Tuple[int, int]:
        nonlocal rest
178
        m = rest.search(rx, start, start + search_window)
179
        if m:
180
181
            begin, end = m.span()
            return rest.index(begin), end - begin
182
183
        return -1, 0

eckhart's avatar
eckhart committed
184
    @cython.locals(a=cython.int, b=cython.int, k=cython.int, length=cython.int)
185
186
187
188
189
    def entry_point(search_func, search_rule) -> int:
        a, b = next_comment()
        k, length = search_func(search_rule)
        while a < b <= k:
            a, b = next_comment()
190
191
192
        # 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)
193
194
            while a < b <= k:
                a, b = next_comment()
195
        return k + length if k >= 0 else upper_limit
196

197
    # find closest match
Eckhart Arnold's avatar
Eckhart Arnold committed
198
    for rule in rules:
199
        comments = rest.finditer(comment_regex)
200
201
202
        assert not isinstance(rule, str), \
            'Strings not allowed as search rules, use a regular expression instead.'
        pos = entry_point(rx_search, rule)
203
        closest_match = min(pos, closest_match)
204

205
    # in case no rule matched return -1
206
207
208
    if closest_match == upper_limit:
        closest_match = -1
    return closest_match
Eckhart Arnold's avatar
Eckhart Arnold committed
209
210


211
212
213
214
215
216
217
########################################################################
#
# Parser base class
#
########################################################################


eckhart's avatar
eckhart committed
218
219
ApplyFunc = Callable[['Parser'], None]
FlagFunc = Callable[[ApplyFunc, Set[ApplyFunc]], bool]
eckhart's avatar
eckhart committed
220
ParseFunc = Callable[[StringView], Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
221
222


223
class Parser:
224
225
226
227
228
229
230
231
232
233
234
235
236
237
    """
    (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
238
    1. *Named parsers* for which a name is set in field `parser.pname`.
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
       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.

    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:
262

263
264
        pname:  The parser's name or a (possibly empty) alias name in case
                of an anonymous parser.
265

266
267
268
269
270
        anonymous: A property indicating that the parser remains anynomous
                anonymous with respect to the nodes it returns. For performance
                reasons this is implemented as an object variable rather
                than a property. This property must always be equal to
                `self.tag_name[0] == ":"`.
271

272
273
274
275
        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
276
277
278
                dropped from the concrete syntax tree already. Only
                anonymous (or pseudo-anonymous) parsers are allowed to
                drop content.
279

280
        tag_name: The tag_name for the nodes that are created by
281
282
                the parser. If the parser is named, this is the same as
                `pname`, otherwise it is the name of the parser's type.
283

284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
        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.

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

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

299
300
301
302
        proxied: The original `_parse()`-method is stored here, if a
                proxy (e.g. a tracing debugger) is installed via the
                `set_proxy()`-method.

303
        _grammar:  A reference to the Grammar object to which the parser
304
305
306
                is attached.
    """

307
    def __init__(self) -> None:
308
        # assert isinstance(name, str), str(name)
309
        self.pname = ''               # type: str
310
        self.anonymous = True         # type: bool
311
        self.drop_content = False     # type: bool
312
313
        self.tag_name = self.ptype    # type: str
        self.cycle_detection = set()  # type: Set[ApplyFunc]
314
        # this indirection is required for Cython-compatibility
Eckhart Arnold's avatar
Eckhart Arnold committed
315
        self._parse_proxy = self._parse    # type: ParseFunc
eckhart's avatar
eckhart committed
316
        # self.proxied = None           # type: Optional[ParseFunc]
317
318
319
320
        try:
            self._grammar = GRAMMAR_PLACEHOLDER  # type: Grammar
        except NameError:
            pass
321
322
323
        self.reset()

    def __deepcopy__(self, memo):
324
        """Deepcopy method of the parser. Upon instantiation of a Grammar-
325
        object, parsers will be deep-copied to the Grammar object. If a
326
        derived parser-class changes the signature of the `__init__`-constructor,
327
328
329
        `__deepcopy__`-method must be replaced (i.e. overridden without
        calling the same method from the superclass) by the derived class.
        """
330
        duplicate = self.__class__()
331
        copy_parser_base_attrs(self, duplicate)
332
        return duplicate
333

334
    def __repr__(self):
di68kap's avatar
di68kap committed
335
        return self.pname + self.ptype
336
337

    def __str__(self):
di68kap's avatar
di68kap committed
338
        return self.pname + (' = ' if self.pname else '') + repr(self)
339

340
341
342
343
344
345
    @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__

346
347
348
    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
di68kap's avatar
di68kap committed
349
        return self.pname if self.pname else self.__repr__()
350

351
352
353
354
    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."""
355
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
356
        self.recursion_counter = defaultdict(int)  # type: DefaultDict[int, int]
357

eckhart's avatar
eckhart committed
358
    @cython.locals(location=cython.int, gap=cython.int, i=cython.int)
359
360
361
    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
362
363
        done in the overridden method `_parse()`. This wrapper-method can be thought of
        as a "parser guard", because it guards the parsing process.
364
        """
eckhart's avatar
eckhart committed
365
366
367
368
369
370
371
372
        # def get_error_node_id(error_node: Node, root_node: RootNode) -> int:
        #     if error_node:
        #         error_node_id = id(error_node)
        #         while error_node_id not in grammar.tree__.error_nodes and error_node.children:
        #             error_node = error_node.result[-1]
        #             error_node_id = id(error_node)
        #     else:
        #         error_node_id = 0
373

374
        grammar = self._grammar
di68kap's avatar
di68kap committed
375
        location = grammar.document_length__ - text._len  # faster then len(text)?
376
377

        try:
378
379
            # rollback variable changing operation if parser backtracks
            # to a position before the variable changing operation occurred
380
            if grammar.last_rb__loc__ > location:
381
382
                grammar.rollback_to__(location)

383
            # if location has already been visited by the current parser, return saved result
384
385
            visited = self.visited  # using local variable for better performance
            if location in visited:
386
                # no history recording in case of memoized results
387
                return visited[location]
388
389

            # break left recursion at the maximum allowed depth
390
391
392
            left_recursion_depth__ = grammar.left_recursion_depth__
            if left_recursion_depth__:
                if self.recursion_counter[location] > left_recursion_depth__:
393
394
395
396
                    grammar.recursion_locations__.add(location)
                    return None, text
                self.recursion_counter[location] += 1

397
            # finally, the actual parser call!
398
            try:
Eckhart Arnold's avatar
Eckhart Arnold committed
399
                node, rest = self._parse_proxy(text)
400
            except ParserError as pe:
401
                # catching up with parsing after an error occurred
402
                gap = len(text) - len(pe.rest)
di68kap's avatar
di68kap committed
403
                rules = grammar.resume_rules__.get(self.pname, [])
404
                rest = pe.rest[len(pe.node):]
405
406
                i = reentry_point(rest, rules, grammar.comment_rx__,
                                  grammar.reentry_search_window__)
407
                if i >= 0 or self == grammar.start_parser__:
408
                    assert pe.node.children or (not pe.node.result)
409
                    # apply reentry-rule or catch error at root-parser
eckhart's avatar
eckhart committed
410
                    if i < 0:  i = 0
411
                    try:
412
                        zombie = pe.node.pick_child(ZOMBIE_TAG)  # type: Optional[Node]
413
                    except (KeyError, ValueError):
414
415
416
                        zombie = None
                    if zombie and not zombie.result:
                        zombie.result = rest[:i]
eckhart's avatar
eckhart committed
417
                        tail = tuple()  # type: ChildrenType
418
419
                    else:
                        nd = Node(ZOMBIE_TAG, rest[:i]).with_pos(location)
eckhart's avatar
eckhart committed
420
                        # nd.attr['err'] = pe.error.message
421
                        tail = (nd,)
422
                    rest = rest[i:]
423
424
                    if pe.first_throw:
                        node = pe.node
425
                        node.result = node.children + tail
426
                    else:
427
428
                        node = Node(
                            self.tag_name,
di68kap's avatar
di68kap committed
429
430
                            (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) + tail) \
                            .with_pos(location)
431
                elif pe.first_throw:
eckhart's avatar
eckhart committed
432
                    # TODO: Is this case still needed with module "trace"?
433
                    raise ParserError(pe.node, pe.rest, pe.error, first_throw=False)
434
                elif grammar.tree__.errors[-1].code == Error.MANDATORY_CONTINUATION_AT_EOF:
eckhart's avatar
eckhart committed
435
436
                    # try to create tree as faithful as possible
                    node = Node(self.tag_name, pe.node).with_pos(location)
437
                else:
438
439
                    result = (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) if gap \
                        else pe.node  # type: ResultType
440
441
                    raise ParserError(Node(self.tag_name, result).with_pos(location),
                                      text, pe.error, first_throw=False)
442

443
            if left_recursion_depth__:
444
445
446
447
448
449
                self.recursion_counter[location] -= 1
                # don't clear recursion_locations__ !!!

            if node is None:
                # retrieve an earlier match result (from left recursion) if it exists
                if location in grammar.recursion_locations__:
450
451
                    if location in visited:
                        node, rest = visited[location]
452
                        if location != grammar.last_recursion_location__:
453
454
455
                            grammar.tree__.add_error(
                                node, Error("Left recursion encountered. "
                                            "Refactor grammar to avoid slow parsing.",
456
457
                                            node.pos if node else location,
                                            Error.LEFT_RECURSION_WARNING))
eckhart's avatar
eckhart committed
458
                            # error_id = id(node)
459
                            grammar.last_recursion_location__ = location
460
461
462
463
                    # don't overwrite any positive match (i.e. node not None) in the cache
                    # and don't add empty entries for parsers returning from left recursive calls!
                elif grammar.memoization__:
                    # otherwise also cache None-results
464
                    visited[location] = (None, rest)
465
            else:
466
                # assert node._pos < 0 or node == EMPTY_NODE
467
                # if node._pos != EMPTY_NODE:
468
                node._pos = location
469
470
                # assert node._pos >= 0 or node == EMPTY_NODE, \
                #     str("%i < %i" % (grammar.document_length__, location))
471
472
473
474
475
476
                if (grammar.last_rb__loc__ < location
                        and (grammar.memoization__ or location in grammar.recursion_locations__)):
                    # - variable manipulating parsers will not be entered into the cache,
                    #   because caching would interfere with changes of variable state
                    # - in case of left recursion, the first recursive step that
                    #   matches will store its result in the cache
eckhart's avatar
eckhart committed
477
478
                    # TODO: need a unit-test concerning interference of variable manipulation
                    #       and left recursion algorithm?
479
                    visited[location] = (node, rest)
480
481

        except RecursionError:
482
            node = Node(ZOMBIE_TAG, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
483
484
485
486
487
488
            node._pos = location
            grammar.tree__.new_error(node, "maximum recursion depth of parser reached; "
                                           "potentially due to too many errors!")
            rest = EMPTY_STRING_VIEW

        return node, rest
489
490
491
492

    def __add__(self, other: 'Parser') -> 'Series':
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
di68kap's avatar
di68kap committed
493
494
        if isinstance(other, Series):
            return cast('Series', other).__radd__(self)
495
496
497
498
499
500
        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
501
502
        if isinstance(other, Alternative):
            return cast('Alternative', other).__ror__(self)
503
504
        return Alternative(self, other)

di68kap's avatar
di68kap committed
505
506
507
508
509
510
511
512
513
    def __mul__(self, other: 'Parser') -> 'Alternative':
        """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)

514
515
516
517
518
519
    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

520
521
522
523
524
525
526
527
    def set_proxy(self, proxy: Optional[ParseFunc]):
        """Sets a proxy that replaces the _parse()-method. The original
        parse-method is copied to the `proxied`-filed of the Parser object and
        can be called by the proxy. Call `set_proxy` with `None` to remove
        a previously set proxy. Typical use case is the installation of a
        tracing debugger. See module `trace`.
        """
        if proxy is None:
Eckhart Arnold's avatar
Eckhart Arnold committed
528
            self._parse_proxy = self._parse
529
        else:
eckhart's avatar
eckhart committed
530
531
            if not isinstance(proxy, type(self._parse)):
                # assume that proxy is a function and bind it to self
532
533
                proxy = proxy.__get__(self, type(self))
            else:
eckhart's avatar
eckhart committed
534
                # if proxy is a method it must be a method of self
535
                assert proxy.__self__ == self
Eckhart Arnold's avatar
Eckhart Arnold committed
536
            self._parse_proxy = cast(ParseFunc, proxy)
537

538
    @property
eckhart's avatar
eckhart committed
539
    def grammar(self) -> 'Grammar':
540
541
542
543
544
        try:
            if self._grammar != GRAMMAR_PLACEHOLDER:
                return self._grammar
            else:
                raise AssertionError('Grammar has not yet been set!')
545
546
        except (AttributeError, NameError):
            raise AttributeError('Parser placeholder does not have a grammar!')
547
548
549

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
550
551
552
553
554
        try:
            if self._grammar == GRAMMAR_PLACEHOLDER:
                self._grammar = grammar
                # self._grammar_assigned_notifier()
            elif self._grammar != grammar:
eckhart's avatar
eckhart committed
555
556
                raise AssertionError("Parser has already been assigned"
                                     "to a different Grammar object!")
557
558
        except AttributeError:
            pass  # ignore setting of grammar attribute for placeholder parser
559
560
        except NameError:  # Cython: No access to GRAMMA_PLACEHOLDER, yet :-(
            self._grammar = grammar
561

eckhart's avatar
eckhart committed
562
    def sub_parsers(self) -> Tuple['Parser', ...]:
563
564
565
        """Returns the list of sub-parsers if there are any.
        Overridden by Unary, Nary and Forward.
        """
eckhart's avatar
eckhart committed
566
        return tuple()
567

eckhart's avatar
eckhart committed
568
569
    def _apply(self, func: Callable[['Parser'], None],
               flip: Callable[[Callable, Set[Callable]], bool]) -> bool:
570
571
        """
        Applies function `func(parser)` recursively to this parser and all
eckhart's avatar
eckhart committed
572
573
574
575
576
577
578
579
580
        descendant parsers, if any exist.

        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!
581
        """
eckhart's avatar
eckhart committed
582
        if flip(func, self.cycle_detection):
583
584
585
            return False
        else:
            func(self)
586
587
            for parser in self.sub_parsers():
                parser._apply(func, flip)
588
589
            return True

Eckhart Arnold's avatar
Eckhart Arnold committed
590
    def apply(self, func: Callable[['Parser'], None]):
eckhart's avatar
eckhart committed
591
592
593
594
        """
        Applies function `func(parser)` recursively to this parser and all
        descendant parsers, if any exist. Traversal is pre-order.
        """
Eckhart Arnold's avatar
Eckhart Arnold committed
595
        def positive_flip(f: Callable[['Parser'], None], flagged: Set[Callable]) -> bool:
eckhart's avatar
eckhart committed
596
597
598
599
600
601
602
603
604
            """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

Eckhart Arnold's avatar
Eckhart Arnold committed
605
        def negative_flip(f: Callable[['Parser'], None], flagged: Set[Callable]) -> bool:
eckhart's avatar
eckhart committed
606
607
608
609
610
611
612
613
614
615
616
617
618
619
            """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:
            self._apply(func, negative_flip)
        else:
            self._apply(func, positive_flip)

620

621
622
def copy_parser_base_attrs(src: Parser, duplicate: Parser):
    """Duplicates all attributes of the Parser-class from source to dest."""
Eckhart Arnold's avatar
Eckhart Arnold committed
623
624
625
626
627
628
    duplicate.pname = src.pname
    duplicate.anonymous = src.anonymous
    duplicate.drop_content = src.drop_content
    duplicate.tag_name = src.tag_name


eckhart's avatar
eckhart committed
629
630
def Drop(parser: Parser) -> Parser:
    """Returns the parser with the `parser.drop_content`-property set to `True`."""
Eckhart Arnold's avatar
Eckhart Arnold committed
631
    assert parser.anonymous, "Parser must be anonymous to be allowed to drop ist content."
eckhart's avatar
eckhart committed
632
633
634
    if isinstance(parser, Forward):
        cast(Forward, parser).parser.drop_content = True
    parser.drop_content = True
Eckhart Arnold's avatar
Eckhart Arnold committed
635
    return parser
eckhart's avatar
eckhart committed
636
637


638
PARSER_PLACEHOLDER = Parser()
639
640


eckhart's avatar
eckhart committed
641
642
643
644
645
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"


646
647
648
649
650
651
########################################################################
#
# Grammar class, central administration of all parser of a grammar
#
########################################################################

652
653
def mixin_comment(whitespace: str, comment: str) -> str:
    """
654
    Returns a regular expression pattern that merges comment and whitespace
655
    regexps. Thus comments can occur wherever whitespace is allowed
656
657
658
659
660
661
    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).
    """
662
663
664
    if comment:
        return '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return whitespace
665
666


di68kap's avatar
di68kap committed
667
def mixin_nonempty(whitespace: str) -> str:
Eckhart Arnold's avatar
Eckhart Arnold committed
668
    r"""
669
670
671
    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
672
673
    If `whitespace`  does not match the empty string '', anyway,
    then it will be returned unaltered.
674
675
676
677
678
679
680
681
682

    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
683
684
685
686
687
    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".
688
689
690
691
692
693
694
695
696
697

    :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


698
class UnknownParserError(KeyError):
eckhart's avatar
eckhart committed
699
    """UnknownParserError is raised if a Grammar object is called with a
700
    parser that does not exist or if in the course of parsing a parser
eckhart's avatar
eckhart committed
701
    is referred to that does not exist."""
702
703


704
GrammarErrorType = Tuple[str, Parser, Error]      # TODO: replace with a named tuple?
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721


class GrammarError(Exception):
    """GrammarError will be raised if static analysis reveals errors
    in the grammar.
    """
    def __init__(self, static_analysis_result: List[GrammarErrorType]):
        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))


722
723
724
725
726
727
728
729
730
731
732
733
734
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
735
    Example for direct instantiation of a grammar::
736

737
        >>> number = RE(r'\d+') + RE(r'\.') + RE(r'\d+') | RE(r'\d+')
738
739
740
741
742
743
744
745
746
747
748
749
750
751
        >>> 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.
752
       See classmethod `Grammar._assign_parser_names__()`
753
754

    3. The parsers in the class do not necessarily need to be connected
755
756
       to one single root parser, which is helpful for testing and when
       building up a parser gradually from several components.
757
758
759
760
761
762
763

    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.

764
    Example::
765
766
767
768
769
770
771
772
773

        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+')
774
775
776
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
777
778
779
780
            root__ = expression

    Upon instantiation the parser objects are deep-copied to the
    Grammar object and assigned to object variables of the same name.
781
782
783
784
785
    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
    is matched by the `anonymous__` regular expression.
786
    If one and the same parser is assigned to several class variables
787
    such as, for example, the parser `expression` in the example above,
788
    which is also assigned to `root__`, the first name sticks.
789
790
791
792
793
794
795
796
797
798
799
800

    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:

801
    Class Attributes:
802
803
804
805
        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.
806

807
808
809
        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
810

811
812
813
814
815
        anonymous__: A regular expression to identify names of parsers that are
                assigned to class fields but shall never the less yield anonymous
                nodes (i.e. nodes the tag name of which starts with a colon ":"
                followed by the parser's class name). The default is to treat all
                parsers starting with an underscore as anonymous in addition to those
816
817
                parsers that are not directly assigned to a class field.

eckhart's avatar
eckhart committed
818
        parser_initialization__:  Before the grammar class (!) has been initialized,
819
                 which happens upon the first time it is instantiated (see
820
                 `:func:_assign_parser_names()` for an explanation), this class
821
822
                 field contains a value other than "done". A value of "done" indicates
                 that the class has already been initialized.
823

824
825
826
827
828
829
830
        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.
831

eckhart's avatar
eckhart committed
832
833
834
835
        python__src__:  For the purpose of debugging and inspection, this field can
                 take the python src of the concrete grammar class
                 (see `dsl.grammar_provider`).

836
    Instance Attributes:
837
838
        all_parsers__:  A set of all parsers connected to this grammar object

839
840
        comment_rx__:  The compiled regular expression for comments. If no
                comments have been defined, it defaults to RX_NEVER_MATCH
841
842
                This instance-attribute will only be defined if a class-attribute
                with the same name does not already exist!
843

eckhart's avatar
eckhart committed
844
845
846
847
        start_parser__:  During parsing, the parser with which the parsing process
                was started (see method `__call__`) or `None` if no parsing process
                is running.

848
849
850
851
852
853
854
        _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
855
856
        document_length__:  the length of the document.

857
858
        document_lbreaks__ (property):  list of linebreaks within the document,
                starting with -1 and ending with EOF. This helps generating line
859
                and column number for history recording and will only be
860
                initialized if :attr:`history_tracking__` is true.
861

862
863
864
865
866
        tree__: The root-node of the parsing tree. This variable is available
               for error-reporting already during parsing  via
               ``self.grammar.tree__.add_error``, but it references the full
               parsing tree only after parsing has been finished.

867
        _reversed__:  the same text in reverse order - needed by the `Lookbehind`-
868
869
870
                parsers.

        variables__:  A mapping for variable names to a stack of their respective
871
872
                string values - needed by the :class:`Capture`-, :class:`Retrieve`-
                and :class:`Pop`-parsers.
873
874

        rollback__:  A list of tuples (location, rollback-function) that are
875
876
877
878
                deposited by the :class:`Capture`- and :class:`Pop`-parsers.
                If the parsing process reaches a dead end then all
                rollback-functions up to the point to which it retreats will be
                called and the state of the variable stack restored accordingly.
879
880
881

        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
882
883
                backtracks to a location at or before last_rb__loc__ (i.e.
                location <= last_rb__loc__) then a rollback of all variable
884
885
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
886
                calling method :func:`rollback_to__(location)`.
887

888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
        recursion_locations__:  Stores the locations where left recursion was
                detected. Needed to provide minimal memoization for the left
                recursion detection algorithm, but, strictly speaking, superfluous
                if full memoization is enabled. (See :func:`Parser.__call__()`)

        last_recursion_location__:  Last location where left recursion was
                detected. This is used to avoid reduplicating warning messages
                about left recursion.

        memoization__:  Turns full memoization on or off. Turning memoization off
                results in less memory usage and sometimes reduced parsing time.
                In some situations it may drastically increase parsing time, so
                it is safer to leave it on. (Default: on)


        # tracing and debugging support

        # These parameters are needed by the debugging functions in module
        # `trace.py`. They should not be manipulated by the users of class
        #  Grammar directly.

        history_tracking__:  A flag indicating that the parsing history is
                being tracked. This flag should not be manipulated by the
                user. Use `trace.set_tracer(grammar, trace.trace_history)` to
                turn (full) history tracking on and
                `trace.set_tracer(grammar, None)` to turn it off. Default is off.

        resume_notices__: A flag indicating that resume messages are generated
                in addition to the error messages, in case the parser was able
                to resume after an error. Use `trace.resume_notices(grammar)` to
                turn resume messages on and `trace.set_tracer(grammar, None)`
                to turn resume messages (as well as history recording) off.
                Default is off.

di68kap's avatar
di68kap committed
922
923
924
        call_stack__:  A stack of the tag names and locations of all parsers
                in the call chain to the currently processed parser during
                parsing. The call stack can be thought of as a breadcrumb trail.
925
                This is required for recording the parser history (for debugging)
926
927
928
                and, eventually, i.e. one day in the future, for tracing through
                the parsing process.

929
930
931
932
        history__:  A list of history records. A history record is appended to
                the list each time a parser either matches, fails or if a
                parser-error occurs. See class `log.HistoryRecord`. History
                records store copies of the current call stack.
933
934
935
936

        moving_forward__: This flag indicates that the parsing process is currently
                moving forward . It is needed to reduce noise in history recording
                and should not be considered as having a valid value if history
937
                recording is turned off! (See :func:`Parser.__call__`)
938

939
940
941
        most_recent_error__: The most recent parser error that has occurred
                or `None`. This can be read by tracers. See module `trace`

942

943
944
945
946
        # Configuration parameters.

        # These values of theses parameters are copied from the global configuration
        # in the constructor of the Grammar object. (see mpodule `configuration.py`)
947

948
949
950
        flatten_tree__:  If True (default), anonymous nodes will be flattened
                during parsing already. This greatly reduces the concrete syntax
                tree and simplifies and speeds up abstract syntax tree generation.
951
                Default is on.
952
953
954

        left_recursion_depth__: the maximum allowed depth for left-recursion.
                A depth of zero means that no left recursion handling will
955
                take place. Default is 5.
956
957
958

        max_parser_dropouts__: Maximum allowed number of retries after errors
                where the parser would exit before the complete document has
959
960
                been parsed. Default is 1, as usually the retry-attemts lead
                to a proliferation of senseless error messages.
961
962
963

        reentry_search_window__: The number of following characters that the
                parser considers when searching a reentry point when a syntax error
964
                has been encountered. Default is 10.000 characters.
965
    """
eckhart's avatar
eckhart committed
966
    python_src__ = ''  # type: str
967
    root__ = PARSER_PLACEHOLDER  # type: Parser
968
    # root__ must be overwritten with the root-parser by grammar subclass
eckhart's avatar
eckhart committed
969
    parser_initialization__ = ["pending"]  # type: List[str]
eckhart's avatar
eckhart committed
970
    resume_rules__ = dict()  # type: Dict[str, ResumeList]
971
    anonymous__ = RX_NEVER_MATCH  # type: RxPatternType
972
    # some default values
eckhart's avatar
eckhart committed
973
974
    COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    WSP_RE__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
975
    static_analysis_pending__ = [True]  # type: List[bool]
976
977
978
979
980


    @classmethod
    def _assign_parser_names__(cls):
        """
di68kap's avatar
di68kap committed
981
        Initializes the `parser.pname` fields of those
982
        Parser objects that are directly assigned to a class field with
983
        the field's name, e.g.::
984

985
986
            class Grammar(Grammar):
                ...
987
                symbol = RE(r'(?!\\d)\\w+')
988

di68kap's avatar
di68kap committed
989
        After the call of this method symbol.pname == "symbol" holds.
di68kap's avatar
di68kap committed
990
991
        Parser names starting or ending with a double underscore like
        ``root__`` will be ignored. See :func:`sane_parser_name()`
992
993
994
995
996
997
998
999
1000

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

        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.