parse.py 147 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, Iterator
37

eckhart's avatar
eckhart committed
38
from DHParser.configuration import get_config_value
39
from DHParser.error import Error, ErrorCode, is_error, 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_control_characters, re, cython, \
53
    abbreviate_middle, RX_NEVER_MATCH, RxPatternType, linebreaks, line_col
54

55
__all__ = ('ParserError',
eckhart's avatar
eckhart committed
56
57
58
           'ApplyFunc',
           'FlagFunc',
           'ParseFunc',
59
           'Parser',
60
           'UnknownParserError',
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
72
           'RE',
           'TKN',
73
           'Whitespace',
74
           'DropRegExp',
75
           'mixin_comment',
di68kap's avatar
di68kap committed
76
           'mixin_nonempty',
di68kap's avatar
di68kap committed
77
           'CombinedParser',
eckhart's avatar
eckhart committed
78
79
           'UnaryParser',
           'NaryParser',
80
           'Drop',
81
82
83
84
           'Synonym',
           'Option',
           'ZeroOrMore',
           'OneOrMore',
85
           'NO_MANDATORY',
Eckhart Arnold's avatar
Eckhart Arnold committed
86
           'MandatoryNary',
87
88
           'Series',
           'Alternative',
89
           'INFINITE',
90
           'Counted',
91
           'Interleave',
eckhart's avatar
eckhart committed
92
           'Required',
93
94
95
96
97
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
           'last_value',
98
99
           'optional_last_value',
           'matching_bracket',
100
101
102
           'Capture',
           'Retrieve',
           'Pop',
103
           'Forward')
104
105
106
107


########################################################################
#
108
# ParserError class
109
110
111
112
#
########################################################################


Eckhart Arnold's avatar
Eckhart Arnold committed
113
114
115
class ParserError(Exception):
    """
    A `ParserError` is thrown for those parser errors that allow the
116
    controlled re-entrance of the parsing process after the error occurred.
Eckhart Arnold's avatar
Eckhart Arnold committed
117
    If a reentry-rule has been configured for the parser where the error
118
    occurred, the parser guard can resume the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
119
120

    Currently, the only case when a `ParserError` is thrown (and not some
121
    different kind of error like `UnknownParserError`) is when a `Series`-
eckhart's avatar
eckhart committed
122
    or `Interleave`-parser detects a missing mandatory element.
Eckhart Arnold's avatar
Eckhart Arnold committed
123
    """
eckhart's avatar
eckhart committed
124
    def __init__(self, node: Node, rest: StringView, error: Error, first_throw: bool):
eckhart's avatar
eckhart committed
125
        assert node is not None
eckhart's avatar
eckhart committed
126
127
128
        self.node = node    # type: Node
        self.rest = rest    # type: StringView
        self.error = error  # type: Error
129
        self.first_throw = first_throw  # type: bool
eckhart's avatar
eckhart committed
130
        self.frozen_callstack = tuple()  # type: Tuple[CallItem, ...]  # tag_name, location
Eckhart Arnold's avatar
Eckhart Arnold committed
131

132
    def __str__(self):
133
        return "%i: %s    %s" % (self.node.pos, str(self.rest[:25]), repr(self.node))
134

Eckhart Arnold's avatar
Eckhart Arnold committed
135

136
ResumeList = List[Union[RxPatternType, str, Callable]]  # list of strings or regular expressions
137
138
ReentryPointAlgorithm = Callable[[StringView, int, int], Tuple[int, int]]
# (text, start point, end point) => (reentry point, match length)
139
# 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
140
141


eckhart's avatar
eckhart committed
142
@cython.locals(upper_limit=cython.int, closest_match=cython.int, pos=cython.int)
143
144
145
def reentry_point(rest: StringView,
                  rules: ResumeList,
                  comment_regex,
146
                  search_window: int = -1) -> int:
Eckhart Arnold's avatar
Eckhart Arnold committed
147
148
    """
    Finds the point where parsing should resume after a ParserError has been caught.
149
    The algorithm makes sure that this reentry-point does not lie inside a comment.
150
151
152
    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.)
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178

    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
179
    """
180
    upper_limit = len(rest) + 1
181
    closest_match = upper_limit
182
183
184
    comments = None  # type: Optional[Iterator]
    if search_window < 0:
        search_window = len(rest)
185

eckhart's avatar
eckhart committed
186
    @cython.locals(a=cython.int, b=cython.int)
187
    def next_comment() -> Tuple[int, int]:
188
189
190
191
        """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.
        """
192
193
194
195
196
197
198
199
200
201
        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

202
203
    @cython.locals(start=cython.int)
    def str_search(s, start: int = 0) -> Tuple[int, int]:
204
205
206
207
208
        """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.
        """
209
210
        nonlocal rest
        return rest.find(s, start, start + search_window), len(s)
211

eckhart's avatar
eckhart committed
212
    @cython.locals(start=cython.int, end=cython.int)
213
    def rx_search(rx, start: int = 0) -> Tuple[int, int]:
214
215
216
217
218
        """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.
        """
219
        nonlocal rest
220
        m = rest.search(rx, start, start + search_window)
221
        if m:
222
223
            begin, end = m.span()
            return rest.index(begin), end - begin
224
225
        return -1, 0

226
227
228
229
    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.
        """
230
        nonlocal rest
231
        return func(rest, start, start + search_window)
232

eckhart's avatar
eckhart committed
233
    @cython.locals(a=cython.int, b=cython.int, k=cython.int, length=cython.int)
234
    def entry_point(search_func, search_rule) -> int:
235
236
237
        """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."""
238
239
        a, b = next_comment()
        k, length = search_func(search_rule)
eckhart's avatar
eckhart committed
240
        while a < b <= k + length:
241
            a, b = next_comment()
242
243
244
        # 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)
245
246
            while a < b <= k:
                a, b = next_comment()
247
        return k + length if k >= 0 else upper_limit
248

249
    # find closest match
Eckhart Arnold's avatar
Eckhart Arnold committed
250
    for rule in rules:
251
        comments = rest.finditer(comment_regex)
252
253
254
255
256
257
258
        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)
259
        closest_match = min(pos, closest_match)
260

261
    # in case no rule matched return -1
262
263
264
    if closest_match == upper_limit:
        closest_match = -1
    return closest_match
Eckhart Arnold's avatar
Eckhart Arnold committed
265
266


267
268
269
270
271
272
273
########################################################################
#
# Parser base class
#
########################################################################


274
ApplyFunc = Callable[[List['Parser']], Optional[bool]]
275
# The return value of `True` stops any further application
eckhart's avatar
eckhart committed
276
FlagFunc = Callable[[ApplyFunc, Set[ApplyFunc]], bool]
eckhart's avatar
eckhart committed
277
ParseFunc = Callable[[StringView], Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
278
279


280
class Parser:
281
282
283
284
285
286
287
288
289
290
291
292
293
294
    """
    (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
295
    1. *Named parsers* for which a name is set in field `parser.pname`.
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
       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:
319

320
321
        pname:  The parser's name or a (possibly empty) alias name in case
                of an anonymous parser.
322

323
        anonymous: A property indicating that the parser remains anonymous
324
325
326
327
                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] == ":"`.
328

329
330
331
332
        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
333
334
335
                dropped from the concrete syntax tree already. Only
                anonymous (or pseudo-anonymous) parsers are allowed to
                drop content.
336

337
        tag_name: The tag_name for the nodes that are created by
338
339
                the parser. If the parser is named, this is the same as
                `pname`, otherwise it is the name of the parser's type.
340

341
342
343
344
345
346
347
348
349
350
        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.

351
352
353
354
        proxied: The original `_parse()`-method is stored here, if a
                proxy (e.g. a tracing debugger) is installed via the
                `set_proxy()`-method.

355
        _grammar:  A reference to the Grammar object to which the parser
356
                is attached.
357
358
359
360
361

        _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.
362
363
    """

364
    def __init__(self) -> None:
365
        # assert isinstance(name, str), str(name)
366
        self.pname = ''               # type: str
367
        self.anonymous = True         # type: bool
368
        self.drop_content = False     # type: bool
369
370
        self.tag_name = self.ptype    # type: str
        self.cycle_detection = set()  # type: Set[ApplyFunc]
371
        # this indirection is required for Cython-compatibility
372
        self._parse_proxy = self._parse  # type: ParseFunc
eckhart's avatar
eckhart committed
373
        # self.proxied = None           # type: Optional[ParseFunc]
374
375
376
377
        try:
            self._grammar = GRAMMAR_PLACEHOLDER  # type: Grammar
        except NameError:
            pass
378
        self._symbol = ''             # type: str
379
380
381
        self.reset()

    def __deepcopy__(self, memo):
382
        """Deepcopy method of the parser. Upon instantiation of a Grammar-
383
        object, parsers will be deep-copied to the Grammar object. If a
384
        derived parser-class changes the signature of the `__init__`-constructor,
385
386
387
        `__deepcopy__`-method must be replaced (i.e. overridden without
        calling the same method from the superclass) by the derived class.
        """
388
        duplicate = self.__class__()
389
        copy_parser_base_attrs(self, duplicate)
390
        return duplicate
391

392
    def __repr__(self):
di68kap's avatar
di68kap committed
393
        return self.pname + self.ptype
394
395

    def __str__(self):
di68kap's avatar
di68kap committed
396
        return self.pname + (' = ' if self.pname else '') + repr(self)
397

398
399
400
401
402
403
    @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__

404
405
406
407
408
    @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
409
410
411
412
413
414
            try:
                self._symbol = self.grammar.associated_symbol(self).pname
            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 ''
415
416
        return self._symbol

417
418
419
    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
di68kap's avatar
di68kap committed
420
        return self.pname if self.pname else self.__repr__()
421

422
423
424
425
    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."""
426
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
427

eckhart's avatar
eckhart committed
428
    @cython.locals(location=cython.int, gap=cython.int, i=cython.int)
429
430
431
    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
432
433
        done in the overridden method `_parse()`. This wrapper-method can be thought of
        as a "parser guard", because it guards the parsing process.
434
        """
435
        grammar = self._grammar
di68kap's avatar
di68kap committed
436
        location = grammar.document_length__ - text._len  # faster then len(text)?
437
438

        try:
439
440
            # rollback variable changing operation if parser backtracks to a position
            # before or at the location where the variable changing operation occurred
441
            if location <= grammar.last_rb__loc__:
442
443
                grammar.rollback_to__(location)

444
            # if location has already been visited by the current parser, return saved result
445
446
            visited = self.visited  # using local variable for better performance
            if location in visited:
447
                # no history recording in case of memoized results!
448
                return visited[location]
449

450
451
452
            recursion_state = grammar.returning_from_recursion__
            grammar.returning_from_recursion__ = False

453
            # now, the actual parser call!
454
            try:
Eckhart Arnold's avatar
Eckhart Arnold committed
455
                node, rest = self._parse_proxy(text)
456
            except ParserError as pe:
457
                # catching up with parsing after an error occurred
458
                gap = len(text) - len(pe.rest)
di68kap's avatar
di68kap committed
459
                rules = grammar.resume_rules__.get(self.pname, [])
460
                rest = pe.rest[len(pe.node):]
461
462
                i = reentry_point(rest, rules, grammar.comment_rx__,
                                  grammar.reentry_search_window__)
463
                if i >= 0 or self == grammar.start_parser__:
eckhart's avatar
eckhart committed
464
465
                    # either a reentry point was found or the
                    # error has fallen through to the first level
466
                    assert pe.node.children or (not pe.node.result)
467
                    # apply reentry-rule or catch error at root-parser
eckhart's avatar
eckhart committed
468
                    if i < 0:  i = 0
469
                    try:
470
                        zombie = pe.node.pick_child(ZOMBIE_TAG)  # type: Optional[Node]
471
                    except (KeyError, ValueError):
472
473
474
                        zombie = None
                    if zombie and not zombie.result:
                        zombie.result = rest[:i]
eckhart's avatar
eckhart committed
475
                        tail = tuple()  # type: ChildrenType
476
477
                    else:
                        nd = Node(ZOMBIE_TAG, rest[:i]).with_pos(location)
eckhart's avatar
eckhart committed
478
                        # nd.attr['err'] = pe.error.message
479
                        tail = (nd,)
480
                    rest = rest[i:]
481
482
                    if pe.first_throw:
                        node = pe.node
483
                        node.result = node.children + tail
484
                    else:
485
486
                        node = Node(
                            self.tag_name,
di68kap's avatar
di68kap committed
487
488
                            (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) + tail) \
                            .with_pos(location)
eckhart's avatar
eckhart committed
489
                # if no re-entry point was found, do any of the following:
490
                elif pe.first_throw:
eckhart's avatar
eckhart committed
491
                    # just fall through
eckhart's avatar
eckhart committed
492
                    # TODO: Is this case still needed with module "trace"?
493
                    raise ParserError(pe.node, pe.rest, pe.error, first_throw=False)
494
                elif grammar.tree__.errors[-1].code == MANDATORY_CONTINUATION_AT_EOF:
eckhart's avatar
eckhart committed
495
496
                    # try to create tree as faithful as possible
                    node = Node(self.tag_name, pe.node).with_pos(location)
497
                else:
eckhart's avatar
eckhart committed
498
                    # fall through but skip the gap
499
500
                    result = (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) if gap \
                        else pe.node  # type: ResultType
501
502
                    raise ParserError(Node(self.tag_name, result).with_pos(location),
                                      text, pe.error, first_throw=False)
503

504
            if node is not None:
505
                node._pos = location
506
507
            if (grammar.memoization__
                    and not grammar.returning_from_recursion__
508
509
510
                    # Variable-manipulating parsers will not be entered into the cache,
                    # because caching would interfere with changes of variable state.
                    # See `_rollback_location()` for the added compensation term.
511
                    and location > grammar.last_rb__loc__ + int(text._len == rest._len)):
512
513
514
515
516
                visited[location] = (node, rest)

            if not grammar.returning_from_recursion__:
                grammar.returning_from_recursion__ = recursion_state

517
        except RecursionError:
518
            node = Node(ZOMBIE_TAG, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
519
            node._pos = location
520
521
522
523
            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)
            grammar.most_recent_error__ = ParserError(node, text, error, first_throw=False)
524
525
526
            rest = EMPTY_STRING_VIEW

        return node, rest
527
528
529
530

    def __add__(self, other: 'Parser') -> 'Series':
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
di68kap's avatar
di68kap committed
531
532
        if isinstance(other, Series):
            return cast('Series', other).__radd__(self)
533
534
535
536
537
538
        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
539
540
        if isinstance(other, Alternative):
            return cast('Alternative', other).__ror__(self)
541
542
        return Alternative(self, other)

543
    def __mul__(self, other: 'Parser') -> 'Interleave':
di68kap's avatar
di68kap committed
544
545
546
547
548
549
550
551
        """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)

552
553
554
555
556
557
    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
558
559
560
561
562
563
564
    def is_optional(self) -> Optional[bool]:
        """Returns `True`, if the parser can never fails, i.e. never yields
        `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

565
    def set_proxy(self, proxy: Optional[ParseFunc]):
566
567
568
        """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`.
569
570
        """
        if proxy is None:
Eckhart Arnold's avatar
Eckhart Arnold committed
571
            self._parse_proxy = self._parse
572
        else:
eckhart's avatar
eckhart committed
573
574
            if not isinstance(proxy, type(self._parse)):
                # assume that proxy is a function and bind it to self
575
576
                proxy = proxy.__get__(self, type(self))
            else:
eckhart's avatar
eckhart committed
577
                # if proxy is a method it must be a method of self
578
                assert proxy.__self__ == self
Eckhart Arnold's avatar
Eckhart Arnold committed
579
            self._parse_proxy = cast(ParseFunc, proxy)
580

581
    @property
eckhart's avatar
eckhart committed
582
    def grammar(self) -> 'Grammar':
583
        try:
eckhart's avatar
eckhart committed
584
            if self._grammar is not GRAMMAR_PLACEHOLDER:
585
586
                return self._grammar
            else:
eckhart's avatar
eckhart committed
587
                raise ValueError('Grammar has not yet been set!')
588
589
        except (AttributeError, NameError):
            raise AttributeError('Parser placeholder does not have a grammar!')
590
591
592

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
593
594
595
596
597
        try:
            if self._grammar == GRAMMAR_PLACEHOLDER:
                self._grammar = grammar
                # self._grammar_assigned_notifier()
            elif self._grammar != grammar:
eckhart's avatar
eckhart committed
598
599
                raise AssertionError("Parser has already been assigned"
                                     "to a different Grammar object!")
600
601
        except AttributeError:
            pass  # ignore setting of grammar attribute for placeholder parser
602
        except NameError:  # Cython: No access to GRAMMAR_PLACEHOLDER, yet :-(
603
            self._grammar = grammar
604

eckhart's avatar
eckhart committed
605
    def sub_parsers(self) -> Tuple['Parser', ...]:
606
607
608
        """Returns the list of sub-parsers if there are any.
        Overridden by Unary, Nary and Forward.
        """
eckhart's avatar
eckhart committed
609
        return tuple()
610

611
    def _apply(self, func: ApplyFunc, context: List['Parser'], flip: FlagFunc) -> bool:
612
613
        """
        Applies function `func(parser)` recursively to this parser and all
614
615
        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
616
617
618
619
620
621
622
623

        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!
624
        """
625
626
        if not flip(func, self.cycle_detection):
            if func(context + [self]):
627
628
629
                return True
            else:
                for parser in self.sub_parsers():
630
                    if parser._apply(func, context + [self], flip):
631
632
633
                        return True
                return False
        return False
634

635
    def apply(self, func: ApplyFunc) -> Optional[bool]:
eckhart's avatar
eckhart committed
636
637
        """
        Applies function `func(parser)` recursively to this parser and all
638
639
640
641
642
643
644
645
646
647
648
649
650
651
        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
652
        """
653
        def positive_flip(f: ApplyFunc, flagged: Set[ApplyFunc]) -> bool:
eckhart's avatar
eckhart committed
654
655
656
657
658
659
660
661
662
            """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

663
        def negative_flip(f: ApplyFunc, flagged: Set[ApplyFunc]) -> bool:
eckhart's avatar
eckhart committed
664
665
666
667
668
669
670
671
672
673
            """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:
674
            return self._apply(func, [], negative_flip)
eckhart's avatar
eckhart committed
675
        else:
676
            return self._apply(func, [], positive_flip)
677

678
    def static_error(self, msg: str, code: ErrorCode) -> 'AnalysisError':
679
        return self.symbol, self, Error(msg, 0, code)
680

eckhart's avatar
eckhart committed
681
    def static_analysis(self) -> List['AnalysisError']:
682
683
        """Analyses the parser for logical errors after the grammar has been
        instantiated."""
eckhart's avatar
eckhart committed
684
        return []
eckhart's avatar
eckhart committed
685

686

687
def copy_parser_base_attrs(src: Parser, duplicate: Parser):
688
    """Duplicates all attributes of the Parser-class from `src` to `duplicate`."""
Eckhart Arnold's avatar
Eckhart Arnold committed
689
690
691
692
693
694
    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
695
696
def Drop(parser: Parser) -> Parser:
    """Returns the parser with the `parser.drop_content`-property set to `True`."""
Eckhart Arnold's avatar
Eckhart Arnold committed
697
    assert parser.anonymous, "Parser must be anonymous to be allowed to drop ist content."
eckhart's avatar
eckhart committed
698
699
700
    if isinstance(parser, Forward):
        cast(Forward, parser).parser.drop_content = True
    parser.drop_content = True
Eckhart Arnold's avatar
Eckhart Arnold committed
701
    return parser
eckhart's avatar
eckhart committed
702
703


704
PARSER_PLACEHOLDER = Parser()
705
706


eckhart's avatar
eckhart committed
707
708
709
710
711
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
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
# 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


732
733
734
735
736
737
########################################################################
#
# Grammar class, central administration of all parser of a grammar
#
########################################################################

738
739
def mixin_comment(whitespace: str, comment: str) -> str:
    """
740
    Returns a regular expression pattern that merges comment and whitespace
741
    regexps. Thus comments can occur wherever whitespace is allowed
742
743
744
745
746
747
    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).
    """
748
    if comment:
eckhart's avatar
eckhart committed
749
750
        whitespace = '(?:' + whitespace + ')'
        comment = '(?:' + comment + ')'
751
752
        return '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return whitespace
753
754


di68kap's avatar
di68kap committed
755
def mixin_nonempty(whitespace: str) -> str:
Eckhart Arnold's avatar
Eckhart Arnold committed
756
    r"""
757
758
759
    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
760
761
    If `whitespace`  does not match the empty string '', anyway,
    then it will be returned unaltered.
762
763
764
765
766
767
768
769
770

    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
771
772
773
774
775
    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".
776
777
778
779
780
781
782
783
784
785

    :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


786
class UnknownParserError(KeyError):
eckhart's avatar
eckhart committed
787
    """UnknownParserError is raised if a Grammar object is called with a
788
    parser that does not exist or if in the course of parsing a parser
eckhart's avatar
eckhart committed
789
    is referred to that does not exist."""
790
791


792
793
AnalysisError = Tuple[str, Parser, Error]      # pname, parser, error
# TODO: replace with a named tuple?
794
795
796
797
798
799


class GrammarError(Exception):
    """GrammarError will be raised if static analysis reveals errors
    in the grammar.
    """
800
    def __init__(self, static_analysis_result: List[AnalysisError]):
801
802
803
804
805
806
807
808
809
810
        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))


811
812
813
814
815
816
817
818
819
820
821
822
823
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
824
    Example for direct instantiation of a grammar::
825

826
        >>> number = RE(r'\d+') + RE(r'\.') + RE(r'\d+') | RE(r'\d+')
827
828
829
830
831
832
833
834
835
836
837
838
839
840
        >>> 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.
841
       See classmethod `Grammar._assign_parser_names__()`
842
843

    3. The parsers in the class do not necessarily need to be connected
844
845
       to one single root parser, which is helpful for testing and when
       building up a parser gradually from several components.
846
847
848
849
850
851
852

    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.

853
    Example::
854
855
856
857
858
859
860
861
862

        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+')
863
864
865
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
866
867
868
869
            root__ = expression

    Upon instantiation the parser objects are deep-copied to the
    Grammar object and assigned to object variables of the same name.
870
871
872
873
874
    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.
875
    If one and the same parser is assigned to several class variables
876
    such as, for example, the parser `expression` in the example above,
877
    which is also assigned to `root__`, the first name sticks.
878
879
880
881
882
883
884
885
886
887
888
889

    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:

890
    Class Attributes:
891
892
893
894
        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.
895

896
897
898
        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
899

900
901
902
903
904
        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
905
906
                parsers that are not directly assigned to a class field.

eckhart's avatar
eckhart committed
907
        parser_initialization__:  Before the grammar class (!) has been initialized,
908
                 which happens upon the first time it is instantiated (see
909
                 `:func:_assign_parser_names()` for an explanation), this class
910
911
                 field contains a value other than "done". A value of "done" indicates
                 that the class has already been initialized.
912

913
914
915
916
917
918
919
        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.
920

921
922
923
        static_analysis_errors__: A list of errors and warnings that were found in the
                static analysis

eckhart's avatar
eckhart committed
924
925
926
927
        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`).

928
    Instance Attributes:
929
930
        all_parsers__:  A set of all parsers connected to this grammar object

931
932
        comment_rx__:  The compiled regular expression for comments. If no
                comments have been defined, it defaults to RX_NEVER_MATCH
933
934
                This instance-attribute will only be defined if a class-attribute
                with the same name does not already exist!
935

eckhart's avatar
eckhart committed
936
937
938
939
        start_parser__:  During parsing, the parser with which the parsing process
                was started (see method `__call__`) or `None` if no parsing process
                is running.

940
941
942
943
944
945
946
        _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
947
948
        document_length__:  the length of the document.

949
950
        document_lbreaks__ (property):  list of linebreaks within the document,
                starting with -1 and ending with EOF. This helps generating line
951
                and column number for history recording and will only be
952
                initialized if :attr:`history_tracking__` is true.
953

954
955
956
957
958
        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.

959
        _reversed__:  the same text in reverse order - needed by the `Lookbehind`-
960
961
962
                parsers.

        variables__:  A mapping for variable names to a stack of their respective
963
964
                string values - needed by the :class:`Capture`-, :class:`Retrieve`-
                and :class:`Pop`-parsers.
965
966

        rollback__:  A list of tuples (location, rollback-function) that are
967
968
969
970
                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.
971
972
973

        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
974
                backtracks to a location at or before last_rb__loc__ (i.e.
975
                location < last_rb__loc__) then a rollback of all variable
976
977
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
978
                calling method :func:`rollback_to__(location)`.
979

980
981
982
983
984
985
986
        returning_from_recursion__: A flag that is true if - during parsing -
                a parser returns from a recursive call, in which case its
                results should not be memoized, because it might be tried
                again with an increased recursion depth. This flag is needed by
                the left-recursion handling algorithm. See `Parser.__call__`
                ans `Forward.__call__`.

987
988
989
990
991
        left_recursion__: Turns on left-recursion handling. This prevents the
                recursive descent parser to get caught in an infinite loop
                (resulting in a maximum recursion depth reached error) when
                the grammar definition contains left recursions.

992
993
994
995
996
        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)

997
998
999
1000
        # mirrored class attributes:

        static_analysis_pending__: A pointer to the class attribute of the same name.
                (See the description above.) If the class is instantiated with a