parse.py 134 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
40
41
42
43
44
from DHParser.error import Error, ErrorCode, is_error, MANDATORY_CONTINUATION, \
    LEFT_RECURSION_WARNING, UNDEFINED_RETRIEVE, PARSER_LOOKAHEAD_FAILURE_ONLY, \
    PARSER_DID_NOT_MATCH, PARSER_LOOKAHEAD_MATCH_ONLY, PARSER_STOPPED_BEFORE_END, \
    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
    OPTIONAL_REDUNDANTLY_NESTED_WARNING, NARY_WITHOUT_PARSERS, CAPTURE_STACK_NOT_EMPTY, BAD_REPETITION_COUNT
eckhart's avatar
eckhart committed
46
from DHParser.log import CallItem, HistoryRecord
47
from DHParser.preprocess import BEGIN_TOKEN, END_TOKEN, RX_TOKEN_NAME
48
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
eckhart's avatar
eckhart committed
49
from DHParser.syntaxtree import ChildrenType, Node, RootNode, WHITESPACE_PTYPE, \
50
    TOKEN_PTYPE, ZOMBIE_TAG, EMPTY_NODE, ResultType
51
from DHParser.toolkit import sane_parser_name, escape_control_characters, re, cython, \
52
    abbreviate_middle, RX_NEVER_MATCH, RxPatternType, linebreaks, line_col
53

54
__all__ = ('ParserError',
eckhart's avatar
eckhart committed
55
56
57
           'ApplyFunc',
           'FlagFunc',
           'ParseFunc',
58
           'Parser',
59
           'UnknownParserError',
60
           'AnalysisError',
61
           'GrammarError',
62
           'Grammar',
63
64
           'Always',
           'Never',
eckhart's avatar
eckhart committed
65
           'AnyChar',
66
           'PreprocessorToken',
67
           'Token',
68
           'DropToken',
69
           'RegExp',
70
71
           'RE',
           'TKN',
72
           'Whitespace',
73
           'DropRegExp',
74
           'mixin_comment',
di68kap's avatar
di68kap committed
75
           'mixin_nonempty',
di68kap's avatar
di68kap committed
76
           'CombinedParser',
eckhart's avatar
eckhart committed
77
78
           'UnaryParser',
           'NaryParser',
79
           'Drop',
80
81
82
83
           'Synonym',
           'Option',
           'ZeroOrMore',
           'OneOrMore',
84
           'NO_MANDATORY',
Eckhart Arnold's avatar
Eckhart Arnold committed
85
           'MandatoryNary',
86
87
           'Series',
           'Alternative',
88
           'INFINITE',
89
           'Interleave',
eckhart's avatar
eckhart committed
90
           'Required',
91
92
93
94
95
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
           'last_value',
96
97
           'optional_last_value',
           'matching_bracket',
98
99
100
           'Capture',
           'Retrieve',
           'Pop',
101
           'Forward')
102
103
104
105


########################################################################
#
106
# ParserError class
107
108
109
110
#
########################################################################


Eckhart Arnold's avatar
Eckhart Arnold committed
111
112
113
class ParserError(Exception):
    """
    A `ParserError` is thrown for those parser errors that allow the
114
    controlled re-entrance of the parsing process after the error occurred.
Eckhart Arnold's avatar
Eckhart Arnold committed
115
    If a reentry-rule has been configured for the parser where the error
116
    occurred, the parser guard can resume the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
117
118

    Currently, the only case when a `ParserError` is thrown (and not some
119
120
    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
121
    """
eckhart's avatar
eckhart committed
122
    def __init__(self, node: Node, rest: StringView, error: Error, first_throw: bool):
eckhart's avatar
eckhart committed
123
        assert node is not None
eckhart's avatar
eckhart committed
124
125
126
        self.node = node    # type: Node
        self.rest = rest    # type: StringView
        self.error = error  # type: Error
127
        self.first_throw = first_throw  # type: bool
eckhart's avatar
eckhart committed
128
        self.frozen_callstack = tuple()  # type: Tuple[CallItem, ...]  # tag_name, location
Eckhart Arnold's avatar
Eckhart Arnold committed
129

130
    def __str__(self):
131
        return "%i: %s    %s" % (self.node.pos, str(self.rest[:25]), repr(self.node))
132

Eckhart Arnold's avatar
Eckhart Arnold committed
133

134
135
136
ResumeList = List[Union[RxPatternType, str, Callable]]  # list of strings or regular expressiones
ReentryPointAlgorithm = Callable[[StringView, int], Tuple[int, int]]
# (text, start point) => (reentry point, match length)
137
# 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
138
139


eckhart's avatar
eckhart committed
140
@cython.locals(upper_limit=cython.int, closest_match=cython.int, pos=cython.int)
141
142
143
def reentry_point(rest: StringView,
                  rules: ResumeList,
                  comment_regex,
144
                  search_window: int = -1) -> int:
Eckhart Arnold's avatar
Eckhart Arnold committed
145
146
    """
    Finds the point where parsing should resume after a ParserError has been caught.
147
    The algorithm makes sure that this reentry-point does not lie inside a comment.
148
149
150
    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
151
152
    Args:
        rest:  The rest of the parsed text or, in other words, the point where
153
            a ParserError was thrown.
154
155
        rules: A list of strings, regular expressions or callable, i.e.
            reentry-point-search-functions. The rest of the text is searched for
156
157
            each of these. The closest match is the point where parsing will be
            resumed.
158
        comment_regex: A regular expression object that matches comments.
159
160
        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
161
    Returns:
162
163
        The integer index of the closest reentry point or -1 if no reentry-point
        was found.
Eckhart Arnold's avatar
Eckhart Arnold committed
164
    """
165
    upper_limit = len(rest) + 1
166
    closest_match = upper_limit
167
168
    comments = None  # typ: Optional[Iterator]

eckhart's avatar
eckhart committed
169
    @cython.locals(a=cython.int, b=cython.int)
170
171
172
173
174
175
176
177
178
179
180
    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

181
182
183
184
    @cython.locals(start=cython.int)
    def str_search(s, start: int = 0) -> Tuple[int, int]:
        nonlocal rest
        return rest.find(s, start, start + search_window), len(s)
185

eckhart's avatar
eckhart committed
186
    @cython.locals(start=cython.int, end=cython.int)
187
188
    def rx_search(rx, start: int = 0) -> Tuple[int, int]:
        nonlocal rest
189
        m = rest.search(rx, start, start + search_window)
190
        if m:
191
192
            begin, end = m.span()
            return rest.index(begin), end - begin
193
194
        return -1, 0

195
196
197
198
    def algorithm_search(func: Callable, start: int = 0):
        nonlocal rest
        return func(rest, start)

eckhart's avatar
eckhart committed
199
    @cython.locals(a=cython.int, b=cython.int, k=cython.int, length=cython.int)
200
201
202
    def entry_point(search_func, search_rule) -> int:
        a, b = next_comment()
        k, length = search_func(search_rule)
eckhart's avatar
eckhart committed
203
        while a < b <= k + length:
204
            a, b = next_comment()
205
206
207
        # 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)
208
209
            while a < b <= k:
                a, b = next_comment()
210
        return k + length if k >= 0 else upper_limit
211

212
    # find closest match
Eckhart Arnold's avatar
Eckhart Arnold committed
213
    for rule in rules:
214
        comments = rest.finditer(comment_regex)
215
216
217
218
219
220
221
        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)
222
        closest_match = min(pos, closest_match)
223

224
    # in case no rule matched return -1
225
226
227
    if closest_match == upper_limit:
        closest_match = -1
    return closest_match
Eckhart Arnold's avatar
Eckhart Arnold committed
228
229


230
231
232
233
234
235
236
########################################################################
#
# Parser base class
#
########################################################################


237
238
ApplyFunc = Callable[[List['Parser']], Optional[bool]]
        # The return value of `True` stops any further application
eckhart's avatar
eckhart committed
239
FlagFunc = Callable[[ApplyFunc, Set[ApplyFunc]], bool]
eckhart's avatar
eckhart committed
240
ParseFunc = Callable[[StringView], Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
241
242


243
class Parser:
244
245
246
247
248
249
250
251
252
253
254
255
256
257
    """
    (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
258
    1. *Named parsers* for which a name is set in field `parser.pname`.
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
       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:
282

283
284
        pname:  The parser's name or a (possibly empty) alias name in case
                of an anonymous parser.
285

286
287
288
289
290
        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] == ":"`.
291

292
293
294
295
        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
296
297
298
                dropped from the concrete syntax tree already. Only
                anonymous (or pseudo-anonymous) parsers are allowed to
                drop content.
299

300
        tag_name: The tag_name for the nodes that are created by
301
302
                the parser. If the parser is named, this is the same as
                `pname`, otherwise it is the name of the parser's type.
303

304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
        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.

319
320
321
322
        proxied: The original `_parse()`-method is stored here, if a
                proxy (e.g. a tracing debugger) is installed via the
                `set_proxy()`-method.

323
        _grammar:  A reference to the Grammar object to which the parser
324
                is attached.
325
326
327
328
329

        _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.
330
331
    """

332
    def __init__(self) -> None:
333
        # assert isinstance(name, str), str(name)
334
        self.pname = ''               # type: str
335
        self.anonymous = True         # type: bool
336
        self.drop_content = False     # type: bool
337
338
        self.tag_name = self.ptype    # type: str
        self.cycle_detection = set()  # type: Set[ApplyFunc]
339
        # this indirection is required for Cython-compatibility
340
        self._parse_proxy = self._parse  # type: ParseFunc
eckhart's avatar
eckhart committed
341
        # self.proxied = None           # type: Optional[ParseFunc]
342
343
344
345
        try:
            self._grammar = GRAMMAR_PLACEHOLDER  # type: Grammar
        except NameError:
            pass
346
        self._symbol = ''             # type: str
347
348
349
        self.reset()

    def __deepcopy__(self, memo):
350
        """Deepcopy method of the parser. Upon instantiation of a Grammar-
351
        object, parsers will be deep-copied to the Grammar object. If a
352
        derived parser-class changes the signature of the `__init__`-constructor,
353
354
355
        `__deepcopy__`-method must be replaced (i.e. overridden without
        calling the same method from the superclass) by the derived class.
        """
356
        duplicate = self.__class__()
357
        copy_parser_base_attrs(self, duplicate)
358
        return duplicate
359

360
    def __repr__(self):
di68kap's avatar
di68kap committed
361
        return self.pname + self.ptype
362
363

    def __str__(self):
di68kap's avatar
di68kap committed
364
        return self.pname + (' = ' if self.pname else '') + repr(self)
365

366
367
368
369
370
371
    @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__

372
373
374
375
376
377
378
379
    @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:
            self._symbol = self.grammar.associated_symbol(self).pname
        return self._symbol

380
381
382
    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
di68kap's avatar
di68kap committed
383
        return self.pname if self.pname else self.__repr__()
384

385
386
387
388
    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."""
389
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
390
        self.recursion_counter = defaultdict(int)  # type: DefaultDict[int, int]
391

eckhart's avatar
eckhart committed
392
    @cython.locals(location=cython.int, gap=cython.int, i=cython.int)
393
394
395
    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
396
397
        done in the overridden method `_parse()`. This wrapper-method can be thought of
        as a "parser guard", because it guards the parsing process.
398
        """
eckhart's avatar
eckhart committed
399
400
401
402
403
404
405
406
        # 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
407

408
        grammar = self._grammar
di68kap's avatar
di68kap committed
409
        location = grammar.document_length__ - text._len  # faster then len(text)?
410
411

        try:
412
413
            # rollback variable changing operation if parser backtracks
            # to a position before the variable changing operation occurred
414
            if grammar.last_rb__loc__ > location:
415
416
                grammar.rollback_to__(location)

417
            # if location has already been visited by the current parser, return saved result
418
419
            visited = self.visited  # using local variable for better performance
            if location in visited:
420
                # no history recording in case of memoized results
421
                return visited[location]
422
423

            # break left recursion at the maximum allowed depth
424
425
426
            left_recursion_depth__ = grammar.left_recursion_depth__
            if left_recursion_depth__:
                if self.recursion_counter[location] > left_recursion_depth__:
427
428
429
430
                    grammar.recursion_locations__.add(location)
                    return None, text
                self.recursion_counter[location] += 1

431
            # finally, the actual parser call!
432
            try:
Eckhart Arnold's avatar
Eckhart Arnold committed
433
                node, rest = self._parse_proxy(text)
434
            except ParserError as pe:
435
                # catching up with parsing after an error occurred
436
                gap = len(text) - len(pe.rest)
di68kap's avatar
di68kap committed
437
                rules = grammar.resume_rules__.get(self.pname, [])
438
                rest = pe.rest[len(pe.node):]
439
440
                i = reentry_point(rest, rules, grammar.comment_rx__,
                                  grammar.reentry_search_window__)
441
                if i >= 0 or self == grammar.start_parser__:
442
                    assert pe.node.children or (not pe.node.result)
443
                    # apply reentry-rule or catch error at root-parser
eckhart's avatar
eckhart committed
444
                    if i < 0:  i = 0
445
                    try:
446
                        zombie = pe.node.pick_child(ZOMBIE_TAG)  # type: Optional[Node]
447
                    except (KeyError, ValueError):
448
449
450
                        zombie = None
                    if zombie and not zombie.result:
                        zombie.result = rest[:i]
eckhart's avatar
eckhart committed
451
                        tail = tuple()  # type: ChildrenType
452
453
                    else:
                        nd = Node(ZOMBIE_TAG, rest[:i]).with_pos(location)
eckhart's avatar
eckhart committed
454
                        # nd.attr['err'] = pe.error.message
455
                        tail = (nd,)
456
                    rest = rest[i:]
457
458
                    if pe.first_throw:
                        node = pe.node
459
                        node.result = node.children + tail
460
                    else:
461
462
                        node = Node(
                            self.tag_name,
di68kap's avatar
di68kap committed
463
464
                            (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) + tail) \
                            .with_pos(location)
465
                elif pe.first_throw:
eckhart's avatar
eckhart committed
466
                    # TODO: Is this case still needed with module "trace"?
467
                    raise ParserError(pe.node, pe.rest, pe.error, first_throw=False)
468
                elif grammar.tree__.errors[-1].code == MANDATORY_CONTINUATION_AT_EOF:
eckhart's avatar
eckhart committed
469
470
                    # try to create tree as faithful as possible
                    node = Node(self.tag_name, pe.node).with_pos(location)
471
                else:
472
473
                    result = (Node(ZOMBIE_TAG, text[:gap]).with_pos(location), pe.node) if gap \
                        else pe.node  # type: ResultType
474
475
                    raise ParserError(Node(self.tag_name, result).with_pos(location),
                                      text, pe.error, first_throw=False)
476

477
            if left_recursion_depth__:
478
479
480
481
482
483
                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__:
484
485
                    if location in visited:
                        node, rest = visited[location]
486
                        if location != grammar.last_recursion_location__:
487
488
489
                            grammar.tree__.add_error(
                                node, Error("Left recursion encountered. "
                                            "Refactor grammar to avoid slow parsing.",
490
                                            node.pos if node else location,
491
                                            LEFT_RECURSION_WARNING))
eckhart's avatar
eckhart committed
492
                            # error_id = id(node)
493
                            grammar.last_recursion_location__ = location
494
495
496
497
                    # 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
498
                    visited[location] = (None, rest)
499
            else:
500
                # assert node._pos < 0 or node == EMPTY_NODE
501
                # if node._pos != EMPTY_NODE:
502
                node._pos = location
503
504
                # assert node._pos >= 0 or node == EMPTY_NODE, \
                #     str("%i < %i" % (grammar.document_length__, location))
505
506
507
508
509
510
                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
511
512
                    # TODO: need a unit-test concerning interference of variable manipulation
                    #       and left recursion algorithm?
513
                    visited[location] = (node, rest)
514
515

        except RecursionError:
516
            node = Node(ZOMBIE_TAG, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
517
518
519
520
521
522
            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
523
524
525
526

    def __add__(self, other: 'Parser') -> 'Series':
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
di68kap's avatar
di68kap committed
527
528
        if isinstance(other, Series):
            return cast('Series', other).__radd__(self)
529
530
531
532
533
534
        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
535
536
        if isinstance(other, Alternative):
            return cast('Alternative', other).__ror__(self)
537
538
        return Alternative(self, other)

di68kap's avatar
di68kap committed
539
540
541
542
543
544
545
546
547
    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)

548
549
550
551
552
553
    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
554
555
556
557
558
559
560
    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

561
    def set_proxy(self, proxy: Optional[ParseFunc]):
562
563
564
        """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`.
565
566
        """
        if proxy is None:
Eckhart Arnold's avatar
Eckhart Arnold committed
567
            self._parse_proxy = self._parse
568
        else:
eckhart's avatar
eckhart committed
569
570
            if not isinstance(proxy, type(self._parse)):
                # assume that proxy is a function and bind it to self
571
572
                proxy = proxy.__get__(self, type(self))
            else:
eckhart's avatar
eckhart committed
573
                # if proxy is a method it must be a method of self
574
                assert proxy.__self__ == self
Eckhart Arnold's avatar
Eckhart Arnold committed
575
            self._parse_proxy = cast(ParseFunc, proxy)
576

577
    @property
eckhart's avatar
eckhart committed
578
    def grammar(self) -> 'Grammar':
579
580
581
582
583
        try:
            if self._grammar != GRAMMAR_PLACEHOLDER:
                return self._grammar
            else:
                raise AssertionError('Grammar has not yet been set!')
584
585
        except (AttributeError, NameError):
            raise AttributeError('Parser placeholder does not have a grammar!')
586
587
588

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
589
590
591
592
593
        try:
            if self._grammar == GRAMMAR_PLACEHOLDER:
                self._grammar = grammar
                # self._grammar_assigned_notifier()
            elif self._grammar != grammar:
eckhart's avatar
eckhart committed
594
595
                raise AssertionError("Parser has already been assigned"
                                     "to a different Grammar object!")
596
597
        except AttributeError:
            pass  # ignore setting of grammar attribute for placeholder parser
598
599
        except NameError:  # Cython: No access to GRAMMA_PLACEHOLDER, yet :-(
            self._grammar = grammar
600

eckhart's avatar
eckhart committed
601
    def sub_parsers(self) -> Tuple['Parser', ...]:
602
603
604
        """Returns the list of sub-parsers if there are any.
        Overridden by Unary, Nary and Forward.
        """
eckhart's avatar
eckhart committed
605
        return tuple()
606

607
    def _apply(self, func: ApplyFunc, context: List['Parser'], flip: FlagFunc) -> bool:
608
609
        """
        Applies function `func(parser)` recursively to this parser and all
610
611
        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
612
613
614
615
616
617
618
619

        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!
620
        """
621
622
        if not flip(func, self.cycle_detection):
            if func(context + [self]):
623
624
625
                return True
            else:
                for parser in self.sub_parsers():
626
                    if parser._apply(func, context + [self], flip):
627
628
629
                        return True
                return False
        return False
630

631
    def apply(self, func: ApplyFunc) -> bool:
eckhart's avatar
eckhart committed
632
633
        """
        Applies function `func(parser)` recursively to this parser and all
634
635
636
637
638
639
640
641
642
643
644
645
646
647
        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
648
        """
Eckhart Arnold's avatar
Eckhart Arnold committed
649
        def positive_flip(f: Callable[['Parser'], None], flagged: Set[Callable]) -> bool:
eckhart's avatar
eckhart committed
650
651
652
653
654
655
656
657
658
            """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
659
        def negative_flip(f: Callable[['Parser'], None], flagged: Set[Callable]) -> bool:
eckhart's avatar
eckhart committed
660
661
662
663
664
665
666
667
668
669
            """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:
670
            return self._apply(func, [], negative_flip)
eckhart's avatar
eckhart committed
671
        else:
672
            return self._apply(func, [], positive_flip)
673

674
    def static_error(self, msg: str, code: ErrorCode) -> 'AnalysisError':
675
        return (self.symbol, self, Error(msg, 0, code))
676
677

    def static_analysis(self) -> Optional[List['AnalysisError']]:
678
679
680
        """Analyses the parser for logical errors after the grammar has been
        instantiated."""
        return None
eckhart's avatar
eckhart committed
681

682

683
684
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
685
686
687
688
689
690
    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
691
692
def Drop(parser: Parser) -> Parser:
    """Returns the parser with the `parser.drop_content`-property set to `True`."""
Eckhart Arnold's avatar
Eckhart Arnold committed
693
    assert parser.anonymous, "Parser must be anonymous to be allowed to drop ist content."
eckhart's avatar
eckhart committed
694
695
696
    if isinstance(parser, Forward):
        cast(Forward, parser).parser.drop_content = True
    parser.drop_content = True
Eckhart Arnold's avatar
Eckhart Arnold committed
697
    return parser
eckhart's avatar
eckhart committed
698
699


700
PARSER_PLACEHOLDER = Parser()
701
702


eckhart's avatar
eckhart committed
703
704
705
706
707
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"


708
709
710
711
712
713
########################################################################
#
# Grammar class, central administration of all parser of a grammar
#
########################################################################

714
715
def mixin_comment(whitespace: str, comment: str) -> str:
    """
716
    Returns a regular expression pattern that merges comment and whitespace
717
    regexps. Thus comments can occur wherever whitespace is allowed
718
719
720
721
722
723
    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).
    """
724
    if comment:
eckhart's avatar
eckhart committed
725
726
        whitespace = '(?:' + whitespace + ')'
        comment = '(?:' + comment + ')'
727
728
        return '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return whitespace
729
730


di68kap's avatar
di68kap committed
731
def mixin_nonempty(whitespace: str) -> str:
Eckhart Arnold's avatar
Eckhart Arnold committed
732
    r"""
733
734
735
    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
736
737
    If `whitespace`  does not match the empty string '', anyway,
    then it will be returned unaltered.
738
739
740
741
742
743
744
745
746

    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
747
748
749
750
751
    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".
752
753
754
755
756
757
758
759
760
761

    :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


762
class UnknownParserError(KeyError):
eckhart's avatar
eckhart committed
763
    """UnknownParserError is raised if a Grammar object is called with a
764
    parser that does not exist or if in the course of parsing a parser
eckhart's avatar
eckhart committed
765
    is referred to that does not exist."""
766
767


768
769
AnalysisError = Tuple[str, Parser, Error]      # pname, parser, error
# TODO: replace with a named tuple?
770
771
772
773
774
775


class GrammarError(Exception):
    """GrammarError will be raised if static analysis reveals errors
    in the grammar.
    """
776
    def __init__(self, static_analysis_result: List[AnalysisError]):
777
778
779
780
781
782
783
784
785
786
        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))


787
788
789
790
791
792
793
794
795
796
797
798
799
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
800
    Example for direct instantiation of a grammar::
801

802
        >>> number = RE(r'\d+') + RE(r'\.') + RE(r'\d+') | RE(r'\d+')
803
804
805
806
807
808
809
810
811
812
813
814
815
816
        >>> 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.
817
       See classmethod `Grammar._assign_parser_names__()`
818
819

    3. The parsers in the class do not necessarily need to be connected
820
821
       to one single root parser, which is helpful for testing and when
       building up a parser gradually from several components.
822
823
824
825
826
827
828

    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.

829
    Example::
830
831
832
833
834
835
836
837
838

        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+')
839
840
841
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
842
843
844
845
            root__ = expression

    Upon instantiation the parser objects are deep-copied to the
    Grammar object and assigned to object variables of the same name.
846
847
848
849
850
    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.
851
    If one and the same parser is assigned to several class variables
852
    such as, for example, the parser `expression` in the example above,
853
    which is also assigned to `root__`, the first name sticks.
854
855
856
857
858
859
860
861
862
863
864
865

    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:

866
    Class Attributes:
867
868
869
870
        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.
871

872
873
874
        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
875

876
877
878
879
880
        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
881
882
                parsers that are not directly assigned to a class field.

eckhart's avatar
eckhart committed
883
        parser_initialization__:  Before the grammar class (!) has been initialized,
884
                 which happens upon the first time it is instantiated (see
885
                 `:func:_assign_parser_names()` for an explanation), this class
886
887
                 field contains a value other than "done". A value of "done" indicates
                 that the class has already been initialized.
888

889
890
891
892
893
894
895
        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.
896

897
898
899
        static_analysis_errors__: A list of errors and warnings that were found in the
                static analysis

eckhart's avatar
eckhart committed
900
901
902
903
        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`).

904
    Instance Attributes:
905
906
        all_parsers__:  A set of all parsers connected to this grammar object

907
908
        comment_rx__:  The compiled regular expression for comments. If no
                comments have been defined, it defaults to RX_NEVER_MATCH
909
910
                This instance-attribute will only be defined if a class-attribute
                with the same name does not already exist!
911

eckhart's avatar
eckhart committed
912
913
914
915
        start_parser__:  During parsing, the parser with which the parsing process
                was started (see method `__call__`) or `None` if no parsing process
                is running.

916
917
918
919
920
921
922
        _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
923
924
        document_length__:  the length of the document.

925
926
        document_lbreaks__ (property):  list of linebreaks within the document,
                starting with -1 and ending with EOF. This helps generating line
927
                and column number for history recording and will only be
928
                initialized if :attr:`history_tracking__` is true.
929

930
931
932
933
934
        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.

935
        _reversed__:  the same text in reverse order - needed by the `Lookbehind`-
936
937
938
                parsers.

        variables__:  A mapping for variable names to a stack of their respective
939
940
                string values - needed by the :class:`Capture`-, :class:`Retrieve`-
                and :class:`Pop`-parsers.
941
942

        rollback__:  A list of tuples (location, rollback-function) that are
943
944
945
946
                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.
947
948
949

        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
950
951
                backtracks to a location at or before last_rb__loc__ (i.e.
                location <= last_rb__loc__) then a rollback of all variable
952
953
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
954
                calling method :func:`rollback_to__(location)`.
955

956
957
958
959
960
961
962
963
964
965
966
967
968
969
        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)

970
971
972
973
974
975
976
977
978
979
980
        # 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
                parser, this pointer will be overwritten with an instance variable
                that serves the same function.

        static_analysis_errors__: A pointer to the class attribute of the same name.
                (See the description above.) If the class is instantiated with a
                parser, this pointer will be overwritten with an instance variable
                that serves the same function.
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000

        # 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.