In January 2021 we will introduce a 10 GB quota for project repositories. Higher limits for individual projects will be available on request. Please see https://doku.lrz.de/display/PUBLIC/GitLab for more information.

parse.py 95.7 KB
Newer Older
1
# parse.py - parser combinators for DHParser
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
#
# Copyright 2016  by Eckhart Arnold (arnold@badw.de)
#                 Bavarian Academy of Sciences an Humanities (badw.de)
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
# implied.  See the License for the specific language governing
# permissions and limitations under the License.
Eckhart Arnold's avatar
Eckhart Arnold committed
17

18 19

"""
20
Module ``parse`` contains the python classes and functions for
21 22
DHParser's packrat-parser. It's central class is the
``Grammar``-class, which is the base class for any concrete
23 24 25 26 27 28 29 30 31
Grammar. Grammar-objects are callable and parsing is done by
calling a Grammar-object with a source text as argument.

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

32

33
from collections import defaultdict
34
import copy
35

36
from DHParser.error import Error, linebreaks, line_col
37 38
from DHParser.log import is_logging, HistoryRecord
from DHParser.preprocess import BEGIN_TOKEN, END_TOKEN, RX_TOKEN_NAME
39
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
40 41
from DHParser.syntaxtree import Node, FrozenNode, RootNode, WHITESPACE_PTYPE, \
    TOKEN_PTYPE, ZOMBIE_TAG, ResultType
42 43
from DHParser.toolkit import sane_parser_name, escape_control_characters, get_config_value,\
    CONFIG_PRESET, re, typing, cython
Eckhart Arnold's avatar
Eckhart Arnold committed
44
from typing import Callable, cast, List, Tuple, Set, Dict, DefaultDict, Union, Optional, Any
45

46

47
__all__ = ('Parser',
48
           'UnknownParserError',
49
           'Grammar',
Eckhart Arnold's avatar
Eckhart Arnold committed
50
           'EMPTY_NODE',
51
           'PreprocessorToken',
52
           'Token',
53
           'DropToken',
54
           'RegExp',
55 56
           'RE',
           'TKN',
57
           'Whitespace',
58
           'DropWhitespace',
59
           'mixin_comment',
60
           'MetaParser',
eckhart's avatar
eckhart committed
61 62
           'UnaryParser',
           'NaryParser',
63 64 65 66 67 68 69 70 71
           'Synonym',
           'Option',
           'ZeroOrMore',
           'OneOrMore',
           'Series',
           'Alternative',
           'AllOf',
           'SomeOf',
           'Unordered',
eckhart's avatar
eckhart committed
72
           'Required',
73 74 75 76 77 78 79 80 81 82
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
           'last_value',
           'counterpart',
           'accumulate',
           'Capture',
           'Retrieve',
           'Pop',
83
           'Forward')
84 85


86 87 88 89 90
########################################################################
#
# Presets
#
########################################################################
91

92
CONFIG_PRESET['flatten_tree_while_parsing'] = True
93 94


95 96
########################################################################
#
97
# Parser base class
98 99 100 101 102 103 104 105 106
#
########################################################################


LEFT_RECURSION_DEPTH = 8  # type: int
# because of python's recursion depth limit, this value ought not to be
# set too high. PyPy allows higher values than CPython
MAX_DROPOUTS = 3  # type: int
# stop trying to recover parsing after so many errors
107
EMPTY_NODE = FrozenNode(':EMPTY__', '')
108 109


Eckhart Arnold's avatar
Eckhart Arnold committed
110 111 112
class ParserError(Exception):
    """
    A `ParserError` is thrown for those parser errors that allow the
113
    controlled re-entrance of the parsing process after the error occurred.
Eckhart Arnold's avatar
Eckhart Arnold committed
114
    If a reentry-rule has been configured for the parser where the error
115
    occurred, the parser guard can resume the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
116 117 118 119 120

    Currently, the only case when a `ParserError` is thrown (and not some
    different kind of error like `UnknownParserError`, is when a `Series`-
    detects a missing mandatory element.
    """
121
    def __init__(self, node: Node, rest: StringView, first_throw: bool):
Eckhart Arnold's avatar
Eckhart Arnold committed
122 123
        self.node = node  # type: Node
        self.rest = rest  # type: StringView
124
        self.first_throw = first_throw  # type: bool
Eckhart Arnold's avatar
Eckhart Arnold committed
125 126


127
ResumeList = List[Union[str, Any]]  # list of strings or regular expressiones
Eckhart Arnold's avatar
Eckhart Arnold committed
128 129


130
def reentry_point(rest: StringView, rules: ResumeList) -> int:
Eckhart Arnold's avatar
Eckhart Arnold committed
131 132 133 134 135 136 137 138 139
    """
    Finds the point where parsing should resume after a ParserError has been caught.
    Args:
        rest:  The rest of the parsed text or, in other words, the point where
                a ParserError was thrown.
        rules: A list of strings or regular expressions. The rest of the text is
                searched for each of these. The closest match is the point where
                parsing will be resumed.
    Returns:
140 141
        The integer index of the closest reentry point or -1 if no reentry-point
        was found.
Eckhart Arnold's avatar
Eckhart Arnold committed
142
    """
143
    upper_limit = len(rest) + 1
Eckhart Arnold's avatar
Eckhart Arnold committed
144 145 146 147
    i = upper_limit
    #find closest match
    for rule in rules:
        if isinstance(rule, str):
148 149
            k = rest.find(rule)
            i = min(k if k >= 0 else upper_limit, i)
Eckhart Arnold's avatar
Eckhart Arnold committed
150 151 152
        else:
            m = rest.search(rule)
            if m:
153
                i = min(rest.index(m.start()), i)
154
    # in case no rule matched return -1
155
    if i == upper_limit:
156 157
        i = -1
    return i
Eckhart Arnold's avatar
Eckhart Arnold committed
158 159


eckhart's avatar
eckhart committed
160 161 162 163
ApplyFunc = Callable[['Parser'], None]
FlagFunc = Callable[[ApplyFunc, Set[ApplyFunc]], bool]


164
class Parser:
165 166 167 168 169 170 171 172 173 174 175 176 177 178
    """
    (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
179
    1. *Named parsers* for which a name is set in field `parser.pname`.
180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217
       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:
        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.

218
        _grammar:  A reference to the Grammar object to which the parser
219 220 221
                is attached.
    """

222
    def __init__(self) -> None:
223
        # assert isinstance(name, str), str(name)
di68kap's avatar
di68kap committed
224
        self.pname = ''  # type: str
225
        self.tag_name = self.ptype  # type: str
226 227 228 229
        try:
            self._grammar = GRAMMAR_PLACEHOLDER  # type: Grammar
        except NameError:
            pass
230 231 232
        self.reset()

    def __deepcopy__(self, memo):
233
        """        Deepcopy method of the parser. Upon instantiation of a Grammar-
234 235 236 237 238
        object, parsers will be deep-copied to the Grammar object. If a
        derived parser-class changes the signature of the constructor,
        `__deepcopy__`-method must be replaced (i.e. overridden without
        calling the same method from the superclass) by the derived class.
        """
239
        duplicate = self.__class__()
di68kap's avatar
di68kap committed
240
        duplicate.pname = self.pname
241
        duplicate.tag_name = self.tag_name
242
        return duplicate
243

244
    def __repr__(self):
di68kap's avatar
di68kap committed
245
        return self.pname + self.ptype
246 247

    def __str__(self):
di68kap's avatar
di68kap committed
248
        return self.pname + (' = ' if self.pname else '') + repr(self)
249

250 251 252 253 254 255
    @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__

256 257 258
    @property
    def repr(self) -> str:
        """Returns the parser's name if it has a name and self.__repr___() otherwise."""
di68kap's avatar
di68kap committed
259
        return self.pname if self.pname else self.__repr__()
260

261 262 263 264
    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."""
265
        self.visited = dict()  # type: Dict[int, Tuple[Optional[Node], StringView]]
eckhart's avatar
eckhart committed
266
        self.recursion_counter = defaultdict(int)  # type: DefaultDict[int, int]
eckhart's avatar
eckhart committed
267
        self.cycle_detection = set()  # type: Set[ApplyFunc]
268

eckhart's avatar
eckhart committed
269
    @cython.locals(location=cython.int, gap=cython.int, i=cython.int)
270 271 272 273 274
    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
        done in the overridden method `_parse()`.
        """
275
        grammar = self._grammar
276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295
        location = grammar.document_length__ - len(text)

        try:
            if grammar.last_rb__loc__ >= location:
                grammar.rollback_to__(location)

            # if location has already been visited by the current parser,
            # return saved result
            if location in self.visited:
                # no history recording in case of memoized results
                return self.visited[location]

            # break left recursion at the maximum allowed depth
            if grammar.left_recursion_handling__:
                if self.recursion_counter[location] > LEFT_RECURSION_DEPTH:
                    grammar.recursion_locations__.add(location)
                    return None, text
                self.recursion_counter[location] += 1

            if grammar.history_tracking__:
296 297
                grammar.call_stack__.append(self.repr if self.tag_name in (':RegExp', ':Token')
                                            else self.tag_name)
298 299 300 301 302 303 304 305
                grammar.moving_forward__ = True

            try:
                # PARSER CALL: run _parse() method
                node, rest = self._parse(text)
            except ParserError as error:
                # does this play well with variable setting? add rollback clause here? tests needed...
                gap = len(text) - len(error.rest)
di68kap's avatar
di68kap committed
306
                rules = grammar.resume_rules__.get(self.pname, [])
307 308 309 310 311 312
                rest = error.rest[len(error.node):]
                i = reentry_point(rest, rules)
                if i >= 0 or self == grammar.start_parser__:
                    # apply reentry-rule or catch error at root-parser
                    if i < 0:
                        i = 1
313
                    nd = Node(ZOMBIE_TAG, rest[:i])
314
                    rest = rest[i:]
eckhart's avatar
eckhart committed
315
                    assert error.node.children or (not error.node.result)
316 317
                    if error.first_throw:
                        node = error.node
eckhart's avatar
eckhart committed
318
                        node.result = node.children + (nd,)
319
                    else:
di68kap's avatar
di68kap committed
320 321
                        # TODO: ggf. Fehlermeldung, die sagt, wo es weitergeht anfügen
                        #       dürfte allerdings erst an den nächsten(!) Knoten angehängt werden (wie?)
322
                        node = Node(self.tag_name, (Node(ZOMBIE_TAG, text[:gap]), error.node, nd))
323 324 325
                elif error.first_throw:
                    raise ParserError(error.node, error.rest, first_throw=False)
                else:
326
                    result = (Node(ZOMBIE_TAG, text[:gap]), error.node) if gap else error.node  # type: ResultType
327 328 329 330 331
                    if grammar.tree__.errors[-1].code == Error.MANDATORY_CONTINUATION_AT_EOF:  # EXPERIMENTAL!!
                        node = error.node
                    else:
                        raise ParserError(Node(self.tag_name, result), text, first_throw=False)

332 333 334 335 336 337 338 339 340 341

            if grammar.left_recursion_handling__:
                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__:
                    if location in self.visited:
                        node, rest = self.visited[location]
342 343 344 345 346
                        if location != grammar.last_recursion_location__:
                            grammar.tree__.add_error(node, Error("Left recursion encountered. "
                                                                 "Refactor grammar to avoid slow parsing.",
                                                                 node.pos, Error.LEFT_RECURSION_WARING))
                            grammar.last_recursion_location__ = location
347 348 349 350 351 352
                    # 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
                    self.visited[location] = (None, rest)
            else:
353
                # assert node._pos < 0 or node == EMPTY_NODE
354
                node._pos = location
355 356
                # assert node._pos >= 0 or node == EMPTY_NODE, \
                #     str("%i < %i" % (grammar.document_length__, location))
357 358 359 360 361 362 363 364 365 366 367 368 369 370
                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
                    # TODO: need a unit-test concerning interference of variable manipulation and left recursion algorithm?
                    self.visited[location] = (node, rest)

            # Mind that memoized parser calls will not appear in the history record!
            # Does this make sense? Or should it be changed?
            if grammar.history_tracking__:
                # don't track returning parsers except in case an error has occurred
                # remaining = len(rest)
371
                if grammar.moving_forward__:
eckhart's avatar
eckhart committed
372
                    record = HistoryRecord(grammar.call_stack__, node or EMPTY_NODE, text,
373
                                           grammar.line_col__(text))
374
                    grammar.history__.append(record)
375 376 377 378 379 380 381
                elif node:
                    nid = id(node)  # type: int
                    if nid in grammar.tree__.error_nodes:
                        record = HistoryRecord(grammar.call_stack__, node or EMPTY_NODE, text,
                                               grammar.line_col__(text),
                                               grammar.tree__.error_nodes[nid])
                        grammar.history__.append(record)
382 383 384 385
                grammar.moving_forward__ = False
                grammar.call_stack__.pop()

        except RecursionError:
386
            node = Node(ZOMBIE_TAG, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
387 388 389 390 391 392
            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
393 394 395 396 397 398 399 400 401 402 403 404

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

eckhart's avatar
eckhart committed
405

406 407 408 409 410 411
    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

412
    @property
eckhart's avatar
eckhart committed
413
    def grammar(self) -> 'Grammar':
414 415 416 417 418
        try:
            if self._grammar != GRAMMAR_PLACEHOLDER:
                return self._grammar
            else:
                raise AssertionError('Grammar has not yet been set!')
419 420
        except (AttributeError, NameError):
            raise AttributeError('Parser placeholder does not have a grammar!')
421 422 423

    @grammar.setter
    def grammar(self, grammar: 'Grammar'):
424 425 426 427 428 429 430 431 432
        try:
            if self._grammar == GRAMMAR_PLACEHOLDER:
                self._grammar = grammar
                # self._grammar_assigned_notifier()
            elif self._grammar != grammar:
                  raise AssertionError("Parser has already been assigned"
                                       "to a different Grammar object!")
        except AttributeError:
            pass  # ignore setting of grammar attribute for placeholder parser
433 434
        except NameError:  # Cython: No access to GRAMMA_PLACEHOLDER, yet :-(
            self._grammar = grammar
435

eckhart's avatar
eckhart committed
436
    def _apply(self, func: ApplyFunc, flip: FlagFunc) -> bool:
437 438
        """
        Applies function `func(parser)` recursively to this parser and all
eckhart's avatar
eckhart committed
439 440 441 442 443 444 445 446 447
        descendant parsers, if any exist.

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

        This is a protected function and should not called from outside
        class Parser or any of its descendants. The entry point for external
        calls is the method `apply()` without underscore!
448
        """
eckhart's avatar
eckhart committed
449
        if flip(func, self.cycle_detection):
450 451 452 453 454
            return False
        else:
            func(self)
            return True

eckhart's avatar
eckhart committed
455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484
    def apply(self, func: ApplyFunc):
        """
        Applies function `func(parser)` recursively to this parser and all
        descendant parsers, if any exist. Traversal is pre-order.
        """
        def positive_flip(f: ApplyFunc, flagged: Set[Callable]) -> bool:
            """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

        def negative_flip(f: ApplyFunc, flagged: Set[Callable]) -> bool:
            """Returns True, if function `f` has already been applied to this
            parser and sets the flag accordingly. Interprets `f in flagged == False`
            as meaning that `f` has already been applied."""
            if f not in flagged:
                return True
            else:
                flagged.remove(f)
                return False

        if func in self.cycle_detection:
            self._apply(func, negative_flip)
        else:
            self._apply(func, positive_flip)

485

486
PARSER_PLACEHOLDER = Parser()
487 488 489 490 491 492 493 494 495


########################################################################
#
# Grammar class, central administration of all parser of a grammar
#
########################################################################


496 497 498 499 500 501 502 503 504 505
def mixin_comment(whitespace: str, comment: str) -> str:
    """
    Returns a regular expression that merges comment and whitespace
    regexps. Thus comments cann occur whereever whitespace is allowed
    and will be skipped just as implicit whitespace.

    Note, that because this works on the level of regular expressions,
    nesting comments is not possible. It also makes it much harder to
    use directives inside comments (which isn't recommended, anyway).
    """
506 507 508
    if comment:
        return '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return whitespace
509 510


511
class UnknownParserError(KeyError):
eckhart's avatar
eckhart committed
512
    """UnknownParserError is raised if a Grammar object is called with a
513
    parser that does not exist or if in the course of parsing a parser
eckhart's avatar
eckhart committed
514
    is referred to that does not exist."""
515 516


517 518 519 520 521 522 523 524 525 526 527 528 529
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
530
    Example for direct instantiation of a grammar::
531

532
        >>> number = RE('\d+') + RE('\.') + RE('\d+') | RE('\d+')
533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557
        >>> 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.

    3. The parsers in the class do not necessarily need to be connected
       to one single root parser, which is helpful for testing and
       building up a parser successively of several components.

    As a consequence, though, it is highly recommended that a Grammar
    class should not define any other variables or methods with names
    that are legal parser names. A name ending with a double
    underscore '__' is *not* a legal parser name and can safely be
    used.

558
    Example::
559 560 561 562 563 564 565 566 567

        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+')
568 569 570
            factor = INTEGER | TKN("(") + expression + TKN(")")
            term = factor + ZeroOrMore((TKN("*") | TKN("/")) + factor)
            expression.set(term + ZeroOrMore((TKN("+") | TKN("-")) + term))
571 572 573 574 575
            root__ = expression

    Upon instantiation the parser objects are deep-copied to the
    Grammar object and assigned to object variables of the same name.
    Any parser that is directly assigned to a class variable is a
di68kap's avatar
di68kap committed
576
    'named' parser and its field `parser.pname` contains the variable
577 578
    name after instantiation of the Grammar class. All other parsers,
    i.e. parsers that are defined within a `named` parser, remain
di68kap's avatar
di68kap committed
579
    "anonymous parsers" where `parser.pname` is the empty string, unless
580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595
    a name has been passed explicitly upon instantiation.
    If one and the same parser is assigned to several class variables
    such as, for example the parser `expression` in the example above,
    the first name sticks.

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

    Grammar 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:

596
    Class Attributes:
597 598 599 600
        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.
601

Eckhart Arnold's avatar
Eckhart Arnold committed
602
        resume_rules__: A mapping of parser names to a list of regular expressions or search
603
                strings that act as rules to find the the reentry point if a ParserError
Eckhart Arnold's avatar
Eckhart Arnold committed
604 605
                was thrown during the execution of the parser with the respective name.

606 607 608 609 610
        parser_initializiation__:  Before the parser class (!) has been initialized,
                 which happens upon the first time it is instantiated (see
                 :func:_assign_parser_names()` for an explanation), this class
                 field contains a value other than "done". A value of "done" indicates
                 that the class has already been initialized.
611

eckhart's avatar
eckhart committed
612 613 614 615
        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`).

616
    Instance Attributes:
617 618
        all_parsers__:  A set of all parsers connected to this grammar object

eckhart's avatar
eckhart committed
619 620 621 622
        start_parser__:  During parsing, the parser with which the parsing process
                was started (see method `__call__`) or `None` if no parsing process
                is running.

623 624 625 626 627 628 629 630 631 632
        history_tracking__:  A flag indicating that the parsing history shall
                be tracked

        _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
633 634
        document_length__:  the length of the document.

635 636 637
        document_lbreaks__:  list of linebreaks within the document, starting
                with -1 and ending with EOF. This helps generating line
                and column number for history recording and will only be
638
                initialized if :attr:`history_tracking__` is true.
639

640 641 642 643 644
        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.

645
        _reversed__:  the same text in reverse order - needed by the `Lookbehind`-
646 647 648
                parsers.

        variables__:  A mapping for variable names to a stack of their respective
649 650
                string values - needed by the :class:`Capture`-, :class:`Retrieve`-
                and :class:`Pop`-parsers.
651 652

        rollback__:  A list of tuples (location, rollback-function) that are
653 654 655 656
                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.
657 658 659

        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
660 661
                backtracks to a location at or before last_rb__loc__ (i.e.
                location <= last_rb__loc__) then a rollback of all variable
662 663
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
664
                calling method :func:`rollback_to__(location)`.
665

666 667
        call_stack__:  A stack of the tag names of all parsers that have been called.
                This is required for recording the parser history (for debugging)
668 669 670 671 672 673 674 675 676 677
                and, eventually, i.e. one day in the future, for tracing through
                the parsing process.

        history__:  A list of parser-call-stacks. A parser-call-stack is
                appended to the list each time a parser either matches, fails
                or if a parser-error occurs.

        moving_forward__: This flag indicates that the parsing process is currently
                moving forward . It is needed to reduce noise in history recording
                and should not be considered as having a valid value if history
678 679
                recording is turned off! (See :func:`add_parser_guard` and its local
                function :func:`guarded_call`)
680 681 682 683

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

        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)

        left_recursion_handling__:  Turns left recursion handling on or off.
                If turned off, a recursion error will result in case of left
                recursion.
698 699 700 701 702 703

        flatten_tree__:  If True (default), anonymous nodes will be flattened
                during parsing already. This greatly reduces the concrete syntax
                tree and simplifies and speeds up abstract syntax tree generation.
                The initial value will be read from the config variable
                'flatten_tree_while_parsing' upon class instantiation.
704
    """
eckhart's avatar
eckhart committed
705
    python_src__ = ''  # type: str
706
    root__ = PARSER_PLACEHOLDER  # type: Parser
707
    # root__ must be overwritten with the root-parser by grammar subclass
eckhart's avatar
eckhart committed
708
    parser_initialization__ = ["pending"]  # type: List[str]
eckhart's avatar
eckhart committed
709
    resume_rules__ = dict()  # type: Dict[str, ResumeList]
710
    # some default values
711 712
    # COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    # WSP_RE__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
713 714 715 716 717


    @classmethod
    def _assign_parser_names__(cls):
        """
di68kap's avatar
di68kap committed
718
        Initializes the `parser.pname` fields of those
719
        Parser objects that are directly assigned to a class field with
720
        the field's name, e.g.::
721

722 723
            class Grammar(Grammar):
                ...
724
                symbol = RE('(?!\\d)\\w+')
725

di68kap's avatar
di68kap committed
726
        After the call of this method symbol.pname == "symbol" holds.
di68kap's avatar
di68kap committed
727 728
        Parser names starting or ending with a double underscore like
        ``root__`` will be ignored. See :func:`sane_parser_name()`
729 730 731 732 733 734 735 736 737 738

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

        Attention: If there exists more than one reference to the same
        parser, only the first one will be chosen for python versions
        greater or equal 3.6.  For python version <= 3.5 an arbitrarily
        selected reference will be chosen. See PEP 520
        (www.python.org/dev/peps/pep-0520/) for an explanation of why.
        """
eckhart's avatar
eckhart committed
739
        if cls.parser_initialization__[0] != "done":
740 741 742
            cdict = cls.__dict__
            for entry, parser in cdict.items():
                if isinstance(parser, Parser) and sane_parser_name(entry):
di68kap's avatar
di68kap committed
743
                    if isinstance(parser, Forward):
di68kap's avatar
di68kap committed
744 745 746 747
                        if not cast(Forward, parser).parser.pname:
                            cast(Forward, parser).parser.pname = entry
                    else:   # if not parser.pname:
                        parser.pname = entry
eckhart's avatar
eckhart committed
748
            cls.parser_initialization__[0] = "done"
749 750 751


    def __init__(self, root: Parser = None) -> None:
752 753
        self.all_parsers__ = set()             # type: Set[Parser]
        self.start_parser__ = None             # type: Optional[Parser]
754 755 756 757
        self._dirty_flag__ = False             # type: bool
        self.history_tracking__ = False        # type: bool
        self.memoization__ = True              # type: bool
        self.left_recursion_handling__ = True  # type: bool
758
        self.flatten_tree__ = get_config_value('flatten_tree_while_parsing')  # type: bool
759 760 761 762 763 764 765 766 767
        self._reset__()

        # prepare parsers in the class, first
        self._assign_parser_names__()

        # then deep-copy the parser tree from class to instance;
        # parsers not connected to the root object will be copied later
        # on demand (see Grammar.__getitem__()). Usually, the need to
        # do so only arises during testing.
eckhart's avatar
eckhart committed
768 769 770 771
        self.root_parser__ = copy.deepcopy(root) if root else copy.deepcopy(self.__class__.root__)
        self.root_parser__.apply(self._add_parser__)
        assert 'root_parser__' in self.__dict__
        assert self.root_parser__ == self.__dict__['root_parser__']
772 773 774 775 776 777


    def __getitem__(self, key):
        try:
            return self.__dict__[key]
        except KeyError:
eckhart's avatar
eckhart committed
778
            parser_template = getattr(self.__class__, key, None)
779 780 781 782
            if parser_template:
                # add parser to grammar object on the fly...
                parser = copy.deepcopy(parser_template)
                parser.apply(self._add_parser__)
eckhart's avatar
eckhart committed
783
                assert self[key] == parser
784
                return self[key]
785
            raise UnknownParserError('Unknown parser "%s" !' % key)
786

787 788
    def __contains__(self, key):
        return key in self.__dict__ or hasattr(self, key)
789 790

    def _reset__(self):
791
        self.tree__ = RootNode()              # type: RootNode
792 793
        self.document__ = EMPTY_STRING_VIEW   # type: StringView
        self._reversed__ = EMPTY_STRING_VIEW  # type: StringView
eckhart's avatar
eckhart committed
794
        self.document_length__ = 0            # type: int
795 796
        self.document_lbreaks__ = []          # type: List[int]
        # variables stored and recalled by Capture and Retrieve parsers
eckhart's avatar
eckhart committed
797
        self.variables__ = defaultdict(lambda: [])  # type: DefaultDict[str, List[str]]
798 799 800
        self.rollback__ = []                  # type: List[Tuple[int, Callable]]
        self.last_rb__loc__ = -1              # type: int
        # support for call stack tracing
801
        self.call_stack__ = []                # type: List[str]
802 803 804 805 806
        # snapshots of call stacks
        self.history__ = []                   # type: List[HistoryRecord]
        # also needed for call stack tracing
        self.moving_forward__ = False         # type: bool
        self.recursion_locations__ = set()    # type: Set[int]
807
        self.last_recursion_location__ = -1   # type: int
808 809 810 811 812 813 814 815 816 817


    @property
    def reversed__(self) -> StringView:
        """
        Returns a reversed version of the currently parsed document. As
        about the only case where this is needed is the Lookbehind-parser,
        this is done lazily.
        """
        if not self._reversed__:
818
            self._reversed__ = StringView(self.document__.get_text()[::-1])
819 820 821 822 823 824 825 826
        return self._reversed__


    def _add_parser__(self, parser: Parser) -> None:
        """
        Adds the particular copy of the parser object to this
        particular instance of Grammar.
        """
di68kap's avatar
di68kap committed
827
        if parser.pname:
828
            # prevent overwriting instance variables or parsers of a different class
di68kap's avatar
di68kap committed
829 830
            assert parser.pname not in self.__dict__ or \
                   isinstance(self.__dict__[parser.pname], parser.__class__), \
831
                ('Cannot add parser "%s" because a field with the same name '
di68kap's avatar
di68kap committed
832
                 'already exists in grammar object: %s!'
di68kap's avatar
di68kap committed
833 834 835
                 % (parser.pname, str(self.__dict__[parser.pname])))
            setattr(self, parser.pname, parser)
        parser.tag_name = parser.pname or parser.ptype
836 837 838 839
        self.all_parsers__.add(parser)
        parser.grammar = self


eckhart's avatar
eckhart committed
840 841
    def __call__(self,
                 document: str,
eckhart's avatar
eckhart committed
842
                 start_parser: Union[str, Parser] = "root_parser__",
eckhart's avatar
eckhart committed
843
                 track_history: bool = False) -> RootNode:
844 845 846 847 848
        """
        Parses a document with with parser-combinators.

        Args:
            document (str): The source text to be parsed.
di68kap's avatar
di68kap committed
849 850
            start_parser (str or Parser): The name of the parser with which
                to start. This is useful for testing particular parsers
851
                (i.e. particular parts of the EBNF-Grammar.)
di68kap's avatar
di68kap committed
852 853 854 855 856
            track_history (bool): If true, the parsing history will be
                recorded in self.history__. If logging is turned on (i.e.
                DHParser.log.is_logging() returns true), the parsing history
                will always be recorded, even if `False` is passed to
                the `track_history` parameter.
857
        Returns:
di68kap's avatar
di68kap committed
858
            Node: The root node to the parse tree.
859
        """
860

eckhart's avatar
eckhart committed
861
        def tail_pos(predecessors: Union[List[Node], Tuple[Node, ...], None]) -> int:
eckhart's avatar
eckhart committed
862 863
            """Adds the position after the last node in the list of
            predecessors to the node."""
eckhart's avatar
eckhart committed
864
            return predecessors[-1].pos + len(predecessors[-1]) if predecessors else 0
eckhart's avatar
eckhart committed
865

866 867 868 869 870 871 872
        # assert isinstance(document, str), type(document)
        if self._dirty_flag__:
            self._reset__()
            for parser in self.all_parsers__:
                parser.reset()
        else:
            self._dirty_flag__ = True
di68kap's avatar
di68kap committed
873 874 875 876
        self.history_tracking__ = track_history or is_logging()
        # safe tracking state, because history_tracking__ might be set to false, later,
        # but original tracking state is needed for additional error information.
        track_history = self.history_tracking__
877
        self.document__ = StringView(document)
eckhart's avatar
eckhart committed
878
        self.document_length__ = len(self.document__)
879
        self.document_lbreaks__ = linebreaks(document) if self.history_tracking__ else []
Eckhart Arnold's avatar
Eckhart Arnold committed
880
        self.last_rb__loc__ = -1  # rollback location
881
        parser = self[start_parser] if isinstance(start_parser, str) else start_parser
eckhart's avatar
eckhart committed
882
        self.start_parser__ = parser
883 884 885 886 887 888 889 890
        assert parser.grammar == self, "Cannot run parsers from a different grammar object!" \
                                       " %s vs. %s" % (str(self), str(parser.grammar))
        result = None  # type: Optional[Node]
        stitches = []  # type: List[Node]
        rest = self.document__
        if not rest:
            result, _ = parser(rest)
            if result is None:
Eckhart Arnold's avatar
Eckhart Arnold committed
891
                result = Node(ZOMBIE_TAG, '').with_pos(0)
eckhart's avatar
eckhart committed
892
                self.tree__.new_error(result,
di68kap's avatar
di68kap committed
893 894
                                      'Parser "%s" did not match empty document.' % str(parser),
                                      Error.PARSER_DID_NOT_MATCH)
895 896 897 898 899 900
        while rest and len(stitches) < MAX_DROPOUTS:
            result, rest = parser(rest)
            if rest:
                fwd = rest.find("\n") + 1 or len(rest)
                skip, rest = rest[:fwd], rest[fwd:]
                if result is None:
di68kap's avatar
di68kap committed
901 902 903 904 905 906
                    err_info = '' if not track_history else \
                               '\n    Most advanced: %s\n    Last match:    %s;' % \
                               (str(HistoryRecord.most_advanced_match(self.history__)),
                                str(HistoryRecord.last_match(self.history__)))
                    # Check if a Lookahead-Parser did match. Needed for testing, because
                    # in a test case this is not necessarily an error.
eckhart's avatar
eckhart committed
907
                    last_record = self.history__[-2] if len(self.history__) > 1 else None  # type: Optional[HistoryRecord]
908

eckhart's avatar
eckhart committed
909
                    if last_record and parser != self.root_parser__ \
di68kap's avatar
di68kap committed
910
                            and last_record.status == HistoryRecord.MATCH \
eckhart's avatar
eckhart committed
911 912
                            and last_record.node.pos \
                            + len(last_record.node) >= len(self.document__) \
913 914 915
                            and any(tn in self and isinstance(self[tn], Lookahead)
                                    or tn[0] == ':' and issubclass(eval(tn[1:]), Lookahead)
                                    for tn in last_record.call_stack):
di68kap's avatar
di68kap committed
916 917 918 919 920
                        error_msg = 'Parser did not match except for lookahead! ' + err_info
                        error_code = Error.PARSER_LOOKAHEAD_MATCH_ONLY
                    else:
                        error_msg = 'Parser did not match!' + err_info
                        error_code = Error.PARSER_DID_NOT_MATCH
921 922
                else:
                    stitches.append(result)
eckhart's avatar
eckhart committed
923 924 925 926 927 928
                    error_msg = "Parser stopped before end" \
                        + (("! trying to recover"
                            + (" but stopping history recording at this point."
                               if self.history_tracking__ else "..."))
                            if len(stitches) < MAX_DROPOUTS
                            else " too often! Terminating parser.")
di68kap's avatar
di68kap committed
929
                    error_code = Error.PARSER_STOPPED_BEFORE_END
Eckhart Arnold's avatar
Eckhart Arnold committed
930
                stitches.append(Node(ZOMBIE_TAG, skip).with_pos(tail_pos(stitches)))
di68kap's avatar
di68kap committed
931
                self.tree__.new_error(stitches[-1], error_msg, error_code)
932
                if self.history_tracking__:
eckhart's avatar
eckhart committed
933 934 935 936 937 938
                    # # some parsers may have matched and left history records with nodes != None.
                    # # Because these are not connected to the stitched root node, their pos-
                    # # properties will not be initialized by setting the root node's pos property
                    # # to zero. Therefore, their pos properties need to be initialized here
                    # for record in self.history__:
                    #     if record.node and record.node._pos < 0:
Eckhart Arnold's avatar
Eckhart Arnold committed
939
                    #         record.node.with_pos(0)
940 941
                    record = HistoryRecord(self.call_stack__.copy(), stitches[-1], rest,
                                           self.line_col__(rest))
942 943 944 945 946
                    self.history__.append(record)
                    # stop history tracking when parser returned too early
                    self.history_tracking__ = False
        if stitches:
            if rest:
947
                stitches.append(Node(ZOMBIE_TAG, rest))
948
            #try:
Eckhart Arnold's avatar
Eckhart Arnold committed
949
            result = Node(ZOMBIE_TAG, tuple(stitches)).with_pos(0)
950 951
            # except AssertionError as error:
            #     # some debugging output
952
            #     print(Node(ZOMBIE_TAG, tuple(stitches)).as_sxpr())
953
            #     raise error
954
        if any(self.variables__.values()):
di68kap's avatar
di68kap committed
955
            error_msg = "Capture-retrieve-stack not empty after end of parsing: " \
956
                + str(self.variables__)
di68kap's avatar
di68kap committed
957
            error_code = Error.CAPTURE_STACK_NOT_EMPTY
958 959 960 961 962
            if result:
                if result.children:
                    # add another child node at the end to ensure that the position
                    # of the error will be the end of the text. Otherwise, the error
                    # message above ("...after end of parsing") would appear illogical.
Eckhart Arnold's avatar
Eckhart Arnold committed
963
                    error_node = Node(ZOMBIE_TAG, '').with_pos(tail_pos(result.children))
di68kap's avatar
di68kap committed
964
                    self.tree__.new_error(error_node, error_msg, error_code)
eckhart's avatar
eckhart committed
965
                    result.result = result.children + (error_node,)
966
                else:
di68kap's avatar
di68kap committed
967
                    self.tree__.new_error(result, error_msg, error_code)
eckhart's avatar
eckhart committed
968
        # result.pos = 0  # calculate all positions
969
        # result.errors(self.document__)
eckhart's avatar
eckhart committed
970 971
        if result:
            self.tree__.swallow(result)
eckhart's avatar
eckhart committed
972
        self.start_parser__ = None
973
        return self.tree__
974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991


    def push_rollback__(self, location, func):
        """
        Adds a rollback function that either removes or re-adds
        values on the variable stack (`self.variables`) that have been
        added (or removed) by Capture or Pop Parsers, the results of
        which have been dismissed.
        """
        self.rollback__.append((location, func))
        self.last_rb__loc__ = location


    def rollback_to__(self, location):
        """
        Rolls back the variable stacks (`self.variables`) to its
        state at an earlier location in the parsed document.
        """
Eckhart Arnold's avatar
Eckhart Arnold committed
992
        while self.rollback__ and self.rollback__[-1][0] >= location:
993 994 995 996 997 998 999 1000 1001 1002
            _, rollback_func = self.rollback__.pop()
            # assert not loc > self.last_rb__loc__, \
            #     "Rollback confusion: line %i, col %i < line %i, col %i" % \
            #     (*line_col(self.document__, len(self.document__) - loc),
            #      *line_col(self.document__, len(self.document__) - self.last_rb__loc__))
            rollback_func()
        self.last_rb__loc__ == self.rollback__[-1][0] if self.rollback__ \
            else (len(self.document__) + 1)


1003 1004 1005 1006 1007 1008 1009 1010 1011
    def line_col__(self, text):
        """
        Returns the line and column where text is located in the document.
        Does not check, whether text is actually a substring of the currently
        parsed document.
        """
        return line_col(self.document_lbreaks__, self.document_length__ - len(text))


1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033
def dsl_error_msg(parser: Parser, error_str: str) -> str:
    """
    Returns an error message for errors in the parser configuration,
    e.g. errors that result in infinite loops.

    Args:
        parser (Parser):  The parser where the error was noticed. Note
            that this is not necessarily the parser that caused the
            error but only where the error became apparent.
        error_str (str):  A short string describing the error.
    Returns:
        str: An error message including the call stack if history
        tacking has been turned in the grammar object.
    """
    msg = ["DSL parser specification error:", error_str, 'Caught by parser "%s".' % str(parser)]
    if parser.grammar.history__:
        msg.extend(["\nCall stack:", parser.grammar.history__[-1].stack])
    else:
        msg.extend(["\nEnable history tracking in Grammar object to display call stack."])
    return " ".join(msg)


1034
GRAMMAR_PLACEHOLDER = Grammar()
eckhart's avatar
eckhart committed
1035 1036


1037 1038
########################################################################
#
1039
# _Token and Regular Expression parser classes (i.e. leaf classes)
1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056
#
########################################################################


class PreprocessorToken(Parser):
    """
    Parses tokens that have been inserted by a preprocessor.

    Preprocessors can generate Tokens with the ``make_token``-function.
    These tokens start and end with magic characters that can only be
    matched by the PreprocessorToken Parser. Such tokens can be used to
    insert BEGIN - END delimiters at the beginning or ending of a
    quoted block, for example.
    """

    def __init__(self, token: str) -> None:
        assert token and token.isupper()
1057
        assert RX_TOKEN_NAME.match(token)
1058
        super(PreprocessorToken, self).__init__()
di68kap's avatar
di68kap committed
1059
        self.pname = token
1060

1061
    def __deepcopy__(self, memo):
di68kap's avatar
di68kap committed
1062
        duplicate = self.__class__(self.pname)
1063
        # duplicate.pname = self.pname  # will be written by the constructor, anyway
1064
        duplicate.tag_name = self.tag_name
1065 1066
        return duplicate

1067
    def _parse(self, text: StringView) -> Tuple[Optional[Node], StringView]:
1068 1069 1070
        if text[0:1] == BEGIN_TOKEN:
            end = text.find(END_TOKEN, 1)
            if end < 0:
1071
                node = Node(self.tag_name, '')
eckhart's avatar
eckhart committed
1072 1073
                self.grammar.tree__.new_error(
                    node,
1074 1075
                    'END_TOKEN delimiter missing from preprocessor token. '
                    '(Most likely due to a preprocessor bug!)')  # type: Node
1076
                return node, text[1:]
1077
            elif end == 0:
1078
                node = Node(self.tag_name, '')
eckhart's avatar
eckhart committed
1079 1080
                self.grammar.tree__.new_error(
                    node,
1081 1082
                    'Preprocessor-token cannot have zero length. '
                    '(Most likely due to a preprocessor bug!)')
1083
                return node, text[2:]
1084
            elif text.find(BEGIN_TOKEN, 1, end) >= 0: