parser.py 90.8 KB
Newer Older
1
"""parsers.py - parser combinators for 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
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46

Module ``parsers.py`` contains a number of classes that together
make up parser combinators for left-recursive grammers. For each
element of the extended Backus-Naur-Form as well as for a regular
expression token a class is defined. The set of classes can be used to
define a parser for (ambiguous) left-recursive grammers.


References and Acknowledgements:

Dominikus Herzberg: Objekt-orientierte Parser-Kombinatoren in Python,
Blog-Post, September, 18th 2008 on denkspuren. gedanken, ideen,
anregungen und links rund um informatik-themen, URL:
http://denkspuren.blogspot.de/2008/09/objekt-orientierte-parser-kombinatoren.html

Dominikus Herzberg: Eine einfache Grammatik für LaTeX, Blog-Post,
September, 18th 2008 on denkspuren. gedanken, ideen, anregungen und
links rund um informatik-themen, URL:
http://denkspuren.blogspot.de/2008/09/eine-einfache-grammatik-fr-latex.html

Dominikus Herzberg: Uniform Syntax, Blog-Post, February, 27th 2007 on
denkspuren. gedanken, ideen, anregungen und links rund um
informatik-themen, URL:
http://denkspuren.blogspot.de/2007/02/uniform-syntax.html

Richard A. Frost, Rahmatullah Hafiz and Paul Callaghan: Parser
Combinators for Ambiguous Left-Recursive Grammars, in: P. Hudak and
D.S. Warren (Eds.): PADL 2008, LNCS 4902, pp. 167–181, Springer-Verlag
Berlin Heidelberg 2008.

47
48
49
50
Elizabeth Scott and Adrian Johnstone, GLL Parsing,
in: Electronic Notes in Theoretical Computer Science 253 (2010) 177–189,
http://dotat.at/tmp/gll.pdf

Eckhart Arnold's avatar
Eckhart Arnold committed
51
52
Juancarlo Añez: grako, a PEG parser generator in Python,
https://bitbucket.org/apalala/grako
53
54
55

Vegard Øye: General Parser Combinators in Racket, 2012,
https://epsil.github.io/gll/
56
57
"""

58
import collections
59
import copy
60
import html
61
import os
Eckhart Arnold's avatar
Eckhart Arnold committed
62
63
from functools import partial

64
from DHParser.error import Error, is_error, has_errors, linebreaks, line_col
65
66
67
68
from DHParser.stringview import StringView, EMPTY_STRING_VIEW
from DHParser.syntaxtree import Node, TransformationFunc, ParserBase, WHITESPACE_PTYPE, \
    TOKEN_PTYPE, ZOMBIE_PARSER
from DHParser.toolkit import is_logging, log_dir, logfile_basename, escape_re, sane_parser_name, \
di68kap's avatar
di68kap committed
69
70
71
    load_if_file, re, typing
from typing import Any, Callable, cast, Dict, List, Set, Tuple, Union, Optional

Eckhart Arnold's avatar
Eckhart Arnold committed
72

73
__all__ = ('PreprocessorFunc',
74
           'HistoryRecord',
Eckhart Arnold's avatar
Eckhart Arnold committed
75
           'Parser',
76
           'Grammar',
77
78
79
           'RX_PREPROCESSOR_TOKEN',
           'BEGIN_TOKEN',
           'END_TOKEN',
Eckhart Arnold's avatar
Eckhart Arnold committed
80
           'make_token',
81
82
           'nil_preprocessor',
           'PreprocessorToken',
Eckhart Arnold's avatar
Eckhart Arnold committed
83
84
85
86
           'RegExp',
           'RE',
           'Token',
           'mixin_comment',
87
88
89
           # 'UnaryOperator',
           # 'NaryOperator',
           'Synonym',
90
           'Option',
Eckhart Arnold's avatar
Eckhart Arnold committed
91
92
           'ZeroOrMore',
           'OneOrMore',
93
           'Series',
Eckhart Arnold's avatar
Eckhart Arnold committed
94
           'Alternative',
95
96
           'AllOf',
           'SomeOf',
97
           'Unordered',
Eckhart Arnold's avatar
Eckhart Arnold committed
98
99
100
101
           'Lookahead',
           'NegativeLookahead',
           'Lookbehind',
           'NegativeLookbehind',
102
103
104
           'last_value',
           'counterpart',
           'accumulate',
Eckhart Arnold's avatar
Eckhart Arnold committed
105
106
107
108
           'Capture',
           'Retrieve',
           'Pop',
           'Forward',
109
           'Compiler',
110
           'compile_source')
Eckhart Arnold's avatar
Eckhart Arnold committed
111

112

113
114
115
116
117
118
119
########################################################################
#
# Grammar and parsing infrastructure
#
########################################################################


120
PreprocessorFunc = Union[Callable[[str], str], partial]
121
122


123
124
125
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
Eckhart Arnold's avatar
Eckhart Arnold committed
126
MAX_DROPOUTS = 3  # type: int
127
# stop trying to recover parsing after so many errors
128

129

130
131
132
class HistoryRecord:
    """
    Stores debugging information about one completed step in the
133
    parsing history.
134
135
136
137
138
139
140
141
142

    A parsing step is "completed" when the last one of a nested
    sequence of parser-calls returns. The call stack including
    the last parser call will be frozen in the ``HistoryRecord``-
    object. In addition a reference to the generated leaf node
    (if any) will be stored and the result status of the last
    parser call, which ist either MATCH, FAIL (i.e. no match)
    or ERROR.
    """
143
    __slots__ = ('call_stack', 'node', 'text', 'line_col')
144
145
146
147

    MATCH = "MATCH"
    ERROR = "ERROR"
    FAIL = "FAIL"
148
    Snapshot = collections.namedtuple('Snapshot', ['line', 'column', 'stack', 'status', 'text'])
di68kap's avatar
di68kap committed
149
150
    COLGROUP = '<colgroup>\n<col style="width:2%"/><col style="width:2%"/><col style="width:75"/>' \
               '<col style="width:6%"/><col style="width:15%"/>\n</colgroup>\n'
eckhart's avatar
eckhart committed
151
152
    HTML_LEAD_IN  = (
        '<html>\n<head>\n<meta charset="utf-8"/>\n<style>\n'
di68kap's avatar
di68kap committed
153
154
155
156
        'td.line, td.column {font-family:monospace;color:darkgrey}\n'
        'td.stack{font-family:monospace}\n'
        'td.status{font-family:monospace;font-weight:bold}\n'
        'td.text{font-family:monospace;color:darkblue}\n'
eckhart's avatar
eckhart committed
157
        'table{border-spacing: 0px; border: thin solid darkgrey; width:100%}\n'
di68kap's avatar
di68kap committed
158
159
160
161
162
        'td{border-right: thin solid grey; border-bottom: thin solid grey}\n'
        'span.delimiter{color:grey;}\nspan.match{color:darkgreen}\n'
        'span.fail{color:darkgrey}\nspan.error{color:red}\n'
        'span.matchstack{font-weight:bold;color:darkred}'
        '\n</style>\n</head>\n<body>\n<table>\n' + COLGROUP)
eckhart's avatar
eckhart committed
163
    HTML_LEAD_OUT = '\n</table>\n</body>\n</html>\n'
164

165
    def __init__(self, call_stack: List['Parser'], node: Node, text: StringView) -> None:
166
        # copy call stack, dropping uninformative Forward-Parsers
Eckhart Arnold's avatar
Eckhart Arnold committed
167
        self.call_stack = [p for p in call_stack if p.ptype != ":Forward"]  # type: List['Parser']
168
        self.node = node                # type: Node
169
        self.text = text                # type: StringView
170
171
        self.line_col = (1, 1)          # type: Tuple[int, int]
        if call_stack:
172
173
174
            grammar = call_stack[-1].grammar
            document = grammar.document__
            lbreaks = grammar.document_lbreaks__
175
            self.line_col = line_col(lbreaks, len(document) - len(text))
176
177

    def __str__(self):
178
179
180
        return '%4i, %2i:  %s;  %s;  "%s"' % self.as_tuple()

    def as_tuple(self) -> Snapshot:
eckhart's avatar
eckhart committed
181
182
        return self.Snapshot(self.line_col[0], self.line_col[1],
                             self.stack, self.status, self.excerpt)
183

eckhart's avatar
eckhart committed
184
    def as_html_tr(self) -> str:
eckhart's avatar
eckhart committed
185
186
187
188
189
190
        stack = html.escape(self.stack).replace(
            '-&gt;', '<span class="delimiter">&shy;-&gt;</span>')
        status = html.escape(self.status)
        excerpt = html.escape(self.excerpt)
        if status == self.MATCH:
            status = '<span class="match">' + status + '</span>'
di68kap's avatar
di68kap committed
191
192
193
194
195
196
197
198
199
200
201
202
203
            i = stack.rfind('-&gt;')
            chr = stack[i+12:i+13]
            while not chr.isidentifier() and i >= 0:
                i = stack.rfind('-&gt;', 0, i)
                chr = stack[i+12:i+13]
            if i >= 0:
                i += 12
                k = stack.find('<', i)
                if k < 0:
                    stack = stack[:i] + '<span class="matchstack">' + stack[i:]
                else:
                    stack = stack[:i] + '<span class="matchstack">' + stack[i:k] \
                            + '</span>' + stack[k:]
eckhart's avatar
eckhart committed
204
205
206
207
208
209
        elif status == self.FAIL:
            status = '<span class="fail">' + status + '</span>'
        else:
            stack += '<br/>\n' + status
            status = '<span class="error">ERROR</span>'
        tpl = self.Snapshot(str(self.line_col[0]), str(self.line_col[1]), stack, status, excerpt)
di68kap's avatar
di68kap committed
210
        # return ''.join(['<tr>'] + [('<td>%s</td>' % item) for item in tpl] + ['</tr>'])
eckhart's avatar
eckhart committed
211
        return ''.join(['<tr>'] + [('<td class="%s">%s</td>' % (cls, item))
eckhart's avatar
eckhart committed
212
                                   for cls, item in zip(tpl._fields, tpl)] + ['</tr>'])
213

214
    def err_msg(self) -> str:
215
216
        return self.ERROR + ": " + "; ".join(
            str(e) for e in (self.node._errors if self.node._errors else
217
                             self.node.collect_errors()[:2]))
218

219
    @property
220
    def stack(self) -> str:
221
        return "->".join((p.repr if p.ptype == ':RegExp' else p.name or p.ptype)
222
                         for p in self.call_stack)
223
224

    @property
225
    def status(self) -> str:
226
        return self.FAIL if self.node is None else \
227
228
            ('"%s"' % self.err_msg()) if self.node.error_flag else self.MATCH  # has_errors(self.node._errors)

eckhart's avatar
eckhart committed
229
230
231
    @property
    def excerpt(self):
        length = len(self.node) if self.node else len(self.text)
di68kap's avatar
di68kap committed
232
        excerpt = str(self.node)[:min(length, 20)] if self.node else self.text[:20]
eckhart's avatar
eckhart committed
233
234
235
236
237
        excerpt = excerpt.replace('\n', '\\n')
        if length > 20:
            excerpt += '...'
        return excerpt

238
239
240
241
    # @property
    # def extent(self) -> slice:
    #     return (slice(-self.remaining - len(self.node), -self.remaining) if self.node
    #             else slice(-self.remaining, None))
242
243

    @property
244
245
    def remaining(self) -> int:
        return len(self.text) - (len(self.node) if self.node else 0)
246

247
    @staticmethod
Eckhart Arnold's avatar
Eckhart Arnold committed
248
    def last_match(history: List['HistoryRecord']) -> Union['HistoryRecord', None]:
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
        """
        Returns the last match from the parsing-history.
        Args:
            history:  the parsing-history as a list of HistoryRecord objects

        Returns:
            the history record of the last match or none if either history is
            empty or no parser could match
        """
        for record in reversed(history):
            if record.status == HistoryRecord.MATCH:
                return record
        return None

    @staticmethod
Eckhart Arnold's avatar
Eckhart Arnold committed
264
    def most_advanced_match(history: List['HistoryRecord']) -> Union['HistoryRecord', None]:
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
        """
        Returns the closest-to-the-end-match from the parsing-history.
        Args:
            history:  the parsing-history as a list of HistoryRecord objects

        Returns:
            the history record of the closest-to-the-end-match or none if either history is
            empty or no parser could match
        """
        remaining = -1
        result = None
        for record in history:
            if (record.status == HistoryRecord.MATCH and
                    (record.remaining < remaining or remaining < 0)):
                result = record
                remaining = record.remaining
        return result


284
def add_parser_guard(parser_func):
285
286
    """
    Add a wrapper function to a parser functions (i.e. Parser.__call__ method)
287
    that takes care of memoizing, left recursion and, optionally, tracing
288
289
    (aka "history tracking") of parser calls. Returns the wrapped call.
    """
290
    def guarded_call(parser: 'Parser', text: StringView) -> Tuple[Optional[Node], StringView]:
291
        try:
292
            location = len(text)    # mind that location is always the distance to the end
293
294
            grammar = parser.grammar  # grammar may be 'None' for unconnected parsers!

295
296
            if grammar.last_rb__loc__ <= location:
                grammar.rollback_to__(location)
297

298
299
300
            # if location has already been visited by the current parser,
            # return saved result
            if location in parser.visited:
di68kap's avatar
di68kap committed
301
                return parser.visited[location]
302

Eckhart Arnold's avatar
Eckhart Arnold committed
303
304
            if grammar.history_tracking__:
                grammar.call_stack__.append(parser)
305
                grammar.moving_forward__ = True
Eckhart Arnold's avatar
Eckhart Arnold committed
306

307
308
309
310
311
312
            # break left recursion at the maximum allowed depth
            if grammar.left_recursion_handling__:
                if parser.recursion_counter.setdefault(location, 0) > LEFT_RECURSION_DEPTH:
                    grammar.recursion_locations__.add(location)
                    return None, text
                parser.recursion_counter[location] += 1
313
314
315
316

            # run original __call__ method
            node, rest = parser_func(parser, text)

317
318
319
            if grammar.left_recursion_handling__:
                parser.recursion_counter[location] -= 1

320
            if node is None:
321
                # retrieve an earlier match result (from left recursion) if it exists
322
323
                if location in grammar.recursion_locations__:
                    if location in parser.visited:
di68kap's avatar
di68kap committed
324
                        node, rest = parser.visited[location]
325
                        # TODO: maybe add a warning about occurrence of left-recursion here?
326
327
328
                    # 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__:
329
                    # otherwise also cache None-results
di68kap's avatar
di68kap committed
330
                    parser.visited[location] = (None, rest)
Eckhart Arnold's avatar
Eckhart Arnold committed
331
332
            elif (grammar.last_rb__loc__ > location
                  and (grammar.memoization__ or location in grammar.recursion_locations__)):
333
334
335
336
                # - 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
di68kap's avatar
di68kap committed
337
                parser.visited[location] = (node, rest)
338

339
            if grammar.history_tracking__:
340
                # don't track returning parsers except in case an error has occurred
341
                remaining = len(rest)
342
                if grammar.moving_forward__ or (node and node.error_flag):  # node._errors
343
                    record = HistoryRecord(grammar.call_stack__, node, text)
344
                    grammar.history__.append(record)
345
                    # print(record.stack, record.status, rest[:20].replace('\n', '|'))
346
                grammar.moving_forward__ = False
347
                grammar.call_stack__.pop()
348

349
        except RecursionError:
350
            node = Node(None, str(text[:min(10, max(1, text.find("\n")))]) + " ...")
351
352
            node.add_error("maximum recursion depth of parser reached; "
                           "potentially due to too many errors!")
353
            rest = EMPTY_STRING_VIEW
354
355
356
357
358
359

        return node, rest

    return guarded_call


360
class Parser(ParserBase):
361
362
363
364
    """
    (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.
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381

    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:

    1. *Named parsers* for which a name is set in field parser.name.
       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.
Eckhart Arnold's avatar
Eckhart Arnold committed
382
383

    Parser objects are callable and parsing is done by calling a parser
Eckhart Arnold's avatar
Eckhart Arnold committed
384
385
386
387
388
389
    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
390
    `ZeroOrMore` or `Option`). If `i > 0` then the parser has "moved
Eckhart Arnold's avatar
Eckhart Arnold committed
391
392
393
394
395
396
    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.
397

398
    Attributes and Properties:
399
        visited:  Mapping of places this parser has already been to
di68kap's avatar
di68kap committed
400
                during the current parsing process onto the results the
401
402
403
404
405
406
407
408
                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.

409
        cycle_detection:  The apply()-method uses this variable to make
410
411
412
                sure that one and the same function will not be applied
                (recursively) a second time, if it has already been
                applied to this parser.
413
414
415

        grammar:  A reference to the Grammar object to which the parser
                is attached.
416
417
    """

418
419
    ApplyFunc = Callable[['Parser'], None]

420
421
    def __init__(self, name: str = '') -> None:
        # assert isinstance(name, str), str(name)
Eckhart Arnold's avatar
Eckhart Arnold committed
422
        super(Parser, self).__init__(name)
423
        self._grammar = None  # type: 'Grammar'
424
425
        self.reset()

426
427
        # add "aspect oriented" wrapper around parser calls
        # for memoizing, left recursion and tracing
Eckhart Arnold's avatar
Eckhart Arnold committed
428

429
430
431
432
433
434
435
        guarded_parser_call = add_parser_guard(self.__class__.__call__)
        # The following check is necessary for classes that don't override
        # the __call__() method, because in these cases the non-overridden
        # __call__()-method would be substituted a second time!
        if self.__class__.__call__.__code__ != guarded_parser_call.__code__:
            self.__class__.__call__ = guarded_parser_call

Eckhart Arnold's avatar
Eckhart Arnold committed
436
    def __deepcopy__(self, memo):
Eckhart Arnold's avatar
Eckhart Arnold committed
437
438
439
440
441
442
        """Deepcopy method of the parser. Upon instantiation of a Grammar-
        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.
        """
443
444
        return self.__class__(self.name)

445
    def reset(self):
Eckhart Arnold's avatar
Eckhart Arnold committed
446
447
448
        """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."""
449
        self.visited = dict()            # type: Dict[int, Tuple[Optional[Node], StringView]]
450
        self.recursion_counter = dict()  # type: Dict[int, int]
451
        self.cycle_detection = set()     # type: Set[Callable]
452

453
    def __call__(self, text: StringView) -> Tuple[Optional[Node], StringView]:
Eckhart Arnold's avatar
Eckhart Arnold committed
454
455
456
        """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."""
457
458
        return None, text  # default behaviour: don't match

459
    def __add__(self, other: 'Parser') -> 'Series':
460
461
        """The + operator generates a series-parser that applies two
        parsers in sequence."""
462
        return Series(self, other)
463

464
    def __or__(self, other: 'Parser') -> 'Alternative':
465
466
467
        """The | operator generates an alternative parser that applies
        the first parser and, if that does not match, the second parser.
        """
468
469
        return Alternative(self, other)

470
    @property
471
    def grammar(self) -> 'Grammar':
472
473
474
        return self._grammar

    @grammar.setter
475
    def grammar(self, grammar: 'Grammar'):
Eckhart Arnold's avatar
Eckhart Arnold committed
476
477
478
479
480
481
        if self._grammar is None:
            self._grammar = grammar
            self._grammar_assigned_notifier()
        else:
            assert self._grammar == grammar, \
                "Parser has already been assigned to a different Grammar object!"
482
483

    def _grammar_assigned_notifier(self):
484
485
        """A function that notifies the parser object that it has been
        assigned to a grammar."""
486
487
        pass

488
    def apply(self, func: ApplyFunc) -> bool:
489
490
        """
        Applies function `func(parser)` recursively to this parser and all
Eckhart Arnold's avatar
Eckhart Arnold committed
491
        descendant parsers if any exist. The same function can never
492
        be applied twice between calls of the ``reset()``-method!
Eckhart Arnold's avatar
Eckhart Arnold committed
493
494
        Returns `True`, if function has been applied, `False` if function
        had been applied earlier already and thus has not been applied again.
495
496
497
498
        """
        if func in self.cycle_detection:
            return False
        else:
Eckhart Arnold's avatar
Eckhart Arnold committed
499
500
            assert not self.visited, "No calls to Parser.apply() during or " \
                                     "after ongoing parsing process. (Call Parser.reset() first.)"
501
502
503
504
505
            self.cycle_detection.add(func)
            func(self)
            return True


506
def mixin_comment(whitespace: str, comment: str) -> str:
507
508
    """
    Returns a regular expression that merges comment and whitespace
509
510
511
512
513
514
515
516
517
518
519
    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).
    """
    wspc = '(?:' + whitespace + '(?:' + comment + whitespace + ')*)'
    return wspc


520
class Grammar:
eckhart's avatar
eckhart committed
521
    r"""
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
    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.

    Example for direct instantian of a grammar:

        >>> number = RE('\d+') + RE('\.') + RE('\d+') | RE('\d+')
        >>> number_parser = Grammar(number)
Eckhart Arnold's avatar
Eckhart Arnold committed
537
        >>> number_parser("3.1416").content()
538
539
        '3.1416'

Eckhart Arnold's avatar
Eckhart Arnold committed
540
    Collecting the parsers that define a grammar in a descendant class of
541
542
543
544
545
546
547
    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
Eckhart Arnold's avatar
Eckhart Arnold committed
548
549
       assign them to class variables, which results in better
       readability of the Python code.
550

Eckhart Arnold's avatar
Eckhart Arnold committed
551
552
553
    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.
554
555
556
557
558
559

    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.
560
561

    Example:
562

563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
        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+')
            factor = INTEGER | Token("(") + expression + Token(")")
            term = factor + ZeroOrMore((Token("*") | Token("/")) + factor)
            expression.set(term + ZeroOrMore((Token("+") | Token("-")) + term))
            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
    'named' parser and its field `parser.name` contains the variable
    name after instantiation of the Grammar class. All other parsers,
581
582
583
    i.e. parsers that are defined within a `named` parser, remain
    "anonymous parsers" where `parser.name` is the empty string, unless
    a name has been passed explicitly upon instantiation.
584
585
586
587
588
589
590
591
592
593
594
    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.

595
    Grammar classes contain a few special class fields for implicit
596
597
    whitespace and comments that should be overwritten, if the defaults
    (no comments, horizontal right aligned whitespace) don't fit:
598
599
    COMMENT__:  regular expression string for matching comments
    WSP__:   regular expression for whitespace and comments
Eckhart Arnold's avatar
Eckhart Arnold committed
600

601
602
    wspL__:  regular expression string for left aligned whitespace,
             which either equals WSP__ or is empty.
Eckhart Arnold's avatar
Eckhart Arnold committed
603

604
605
    wspR__:  regular expression string for right aligned whitespace,
             which either equals WSP__ or is empty.
Eckhart Arnold's avatar
Eckhart Arnold committed
606

607
608
609
610
    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.
Eckhart Arnold's avatar
Eckhart Arnold committed
611

612
613
614
615
616
617
618
619
620
    parser_initializiation__:  Before the parser class (!) has been initialized,
             which happens upon the first time it is instantiated (see doctring for
             method `_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.

    Attributes:
        all_parsers__:  A set of all parsers connected to this grammar object

Eckhart Arnold's avatar
Eckhart Arnold committed
621
        history_tracking__:  A flag indicating that the parsing history shall
622
623
624
625
626
627
628
                be tracked

        wsp_left_parser__:  A parser for the default left-adjacent-whitespace
                or the zombie-parser (see `syntaxtree.ZOMBIE_PARSER`) if the
                default is empty. The default whitespace will be used by parsers
                `Token` and, if no other parsers are passed to its constructor,
                by parser `RE'.
Eckhart Arnold's avatar
Eckhart Arnold committed
629

630
631
632
633
634
635
636
637
638
639
640
        wsp_right_parser__: The same for the default right-adjacent-whitespace.
                Both wsp_left_parser__ and wsp_right_parser__ merely serve the
                purpose to avoid having to specify the default whitespace
                explicitly every time an `RE`-parser-object is created.

        _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 Arnold's avatar
Eckhart Arnold committed
641

642
643
644
645
646
        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
                initialized if `history_tracking__` is true.

647
648
        _reversed__:  the same text in reverse order - needed by the `Lookbehind'-
                parsers.
Eckhart Arnold's avatar
Eckhart Arnold committed
649

650
651
652
        variables__:  A mapping for variable names to a stack of their respective
                string values - needed by the `Capture`-, `Retrieve`- and `Pop`-
                parsers.
Eckhart Arnold's avatar
Eckhart Arnold committed
653

654
655
656
657
658
        rollback__:  A list of tuples (location, rollback-function) that are
                deposited by the `Capture`- and `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.
Eckhart Arnold's avatar
Eckhart Arnold committed
659

660
661
662
663
664
665
666
667
        last_rb__loc__:  The last, i.e. most advanced location in the text
                where a variable changing operation occurred. If the parser
                backtracks to a location at or before `last_rb__loc__` (which,
                since locations are counted from the reverse, means:
                `location >= last_rb__loc__`) then a rollback of all variable
                changing operations is necessary that occurred after the
                location to which the parser backtracks. This is done by
                calling method `.rollback_to__(location)`.
Eckhart Arnold's avatar
Eckhart Arnold committed
668

669
670
671
672
        call_stack__:  A stack of all parsers that have been called. This
                is required for recording the parser history (for debugging)
                and, eventually, i.e. one day in the future, for tracing through
                the parsing process.
Eckhart Arnold's avatar
Eckhart Arnold committed
673

674
675
676
        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.
Eckhart Arnold's avatar
Eckhart Arnold committed
677

678
679
680
681
682
        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
                recording is turned off! (See `add_parser_guard` and its local
                function `guarded_call`)
Eckhart Arnold's avatar
Eckhart Arnold committed
683

684
685
686
687
688
        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 `add_parser_guard` and its
                local function `guarded_call`)
Eckhart Arnold's avatar
Eckhart Arnold committed
689

690
691
692
        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
693
694
695
696
697
                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
    root__ = ZOMBIE_PARSER  # type: ParserBase
700
    # root__ must be overwritten with the root-parser by grammar subclass
Eckhart Arnold's avatar
Eckhart Arnold committed
701
    parser_initialization__ = "pending"  # type: str
702
    # some default values
Eckhart Arnold's avatar
Eckhart Arnold committed
703
704
705
706
    COMMENT__ = r''  # type: str  # r'#.*(?:\n|$)'
    WSP__ = mixin_comment(whitespace=r'[\t ]*', comment=COMMENT__)  # type: str
    wspL__ = ''     # type: str
    wspR__ = WSP__  # type: str
707

708

709
    @classmethod
710
    def _assign_parser_names__(cls):
711
712
        """
        Initializes the `parser.name` fields of those
713
714
        Parser objects that are directly assigned to a class field with
        the field's name, e.g.
715
            class Grammar(Grammar):
716
717
718
                ...
                symbol = RE('(?!\\d)\\w+')
        After the call of this method symbol.name == "symbol"
719
720
721
722
        holds. Names assigned via the ``name``-parameter of the
        constructor will not be overwritten. Parser names starting or
        ending with a double underscore like ``root__`` will be
        ignored. See ``toolkit.sane_parser_name()``
723
724
725
726

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

727
        Attention: If there exists more than one reference to the same
eckhart's avatar
eckhart committed
728
        parser, only the first one will be chosen for python versions
729
730
        greater or equal 3.6.  For python version <= 3.5 an arbitrarily
        selected reference will be chosen. See PEP 520
eckhart's avatar
eckhart committed
731
        (www.python.org/dev/peps/pep-0520/) for an explanation of why.
732
        """
733
734
735
736
737
        if cls.parser_initialization__ != "done":
            cdict = cls.__dict__
            for entry, parser in cdict.items():
                if isinstance(parser, Parser) and sane_parser_name(entry):
                    if not parser.name:
738
                        parser._name = entry
739
                    if isinstance(parser, Forward) and (not parser.parser._name):
740
                        parser.parser._name = entry
741
            cls.parser_initialization__ = "done"
742

743

eckhart's avatar
eckhart committed
744
    def __init__(self, root: Parser = None) -> None:
Eckhart Arnold's avatar
Eckhart Arnold committed
745
746
        # if not hasattr(self.__class__, 'parser_initialization__'):
        #     self.__class__.parser_initialization__ = "pending"
747
748
749
750
        # if not hasattr(self.__class__, 'wspL__'):
        #     self.wspL__ = ''
        # if not hasattr(self.__class__, 'wspR__'):
        #     self.wspR__ = ''
751
        self.all_parsers__ = set()             # type: Set[ParserBase]
752
753
754
        self._dirty_flag__ = False             # type: bool
        self.history_tracking__ = False        # type: bool
        self.memoization__ = True              # type: bool
755
        self.left_recursion_handling__ = True  # type: bool
756
        self._reset__()
757

Eckhart Arnold's avatar
Eckhart Arnold committed
758
        # prepare parsers in the class, first
759
        self._assign_parser_names__()
760

Eckhart Arnold's avatar
Eckhart Arnold committed
761
762
763
764
        # 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 Arnold's avatar
Eckhart Arnold committed
765
        self.root__ = copy.deepcopy(root) if root else copy.deepcopy(self.__class__.root__)
766

767
        if self.wspL__:
Eckhart Arnold's avatar
Eckhart Arnold committed
768
            self.wsp_left_parser__ = Whitespace(self.wspL__)  # type: ParserBase
769
            self.wsp_left_parser__.grammar = self
770
            self.all_parsers__.add(self.wsp_left_parser__)  # don't you forget about me...
771
772
773
        else:
            self.wsp_left_parser__ = ZOMBIE_PARSER
        if self.wspR__:
Eckhart Arnold's avatar
Eckhart Arnold committed
774
            self.wsp_right_parser__ = Whitespace(self.wspR__)  # type: ParserBase
775
            self.wsp_right_parser__.grammar = self
776
            self.all_parsers__.add(self.wsp_right_parser__)  # don't you forget about me...
777
778
        else:
            self.wsp_right_parser__ = ZOMBIE_PARSER
779
        self.root__.apply(self._add_parser__)
780

781

782
    def __getitem__(self, key):
783
784
785
        try:
            return self.__dict__[key]
        except KeyError:
786
787
            parser_template = getattr(self, key, None)
            if parser_template:
Eckhart Arnold's avatar
Eckhart Arnold committed
788
                # add parser to grammar object on the fly...
789
                parser = copy.deepcopy(parser_template)
790
                parser.apply(self._add_parser__)
791
                # assert self[key] == parser
Eckhart Arnold's avatar
Eckhart Arnold committed
792
                return self[key]
793
            raise KeyError('Unknown parser "%s" !' % key)
794

795

796
    def _reset__(self):
797
798
        self.document__ = EMPTY_STRING_VIEW   # type: StringView
        self._reversed__ = EMPTY_STRING_VIEW  # type: StringView
799
        self.document_lbreaks__ = []          # type: List[int]
800
        # variables stored and recalled by Capture and Retrieve parsers
801
802
803
        self.variables__ = dict()             # type: Dict[str, List[str]]
        self.rollback__ = []                  # type: List[Tuple[int, Callable]]
        self.last_rb__loc__ = -1              # type: int
804
        # support for call stack tracing
805
        self.call_stack__ = []                # type: List[Parser]
806
        # snapshots of call stacks
807
        self.history__ = []                   # type: List[HistoryRecord]
808
        # also needed for call stack tracing
809
810
        self.moving_forward__ = False         # type: bool
        self.recursion_locations__ = set()    # type: Set[int]
811

812

Eckhart Arnold's avatar
Eckhart Arnold committed
813
    @property
Eckhart Arnold's avatar
Eckhart Arnold committed
814
    def reversed__(self) -> StringView:
eckhart's avatar
eckhart committed
815
816
817
818
819
        """
        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.
        """
Eckhart Arnold's avatar
Eckhart Arnold committed
820
        if not self._reversed__:
821
            self._reversed__ = StringView(self.document__.text[::-1])
Eckhart Arnold's avatar
Eckhart Arnold committed
822
823
824
        return self._reversed__


825
    def _add_parser__(self, parser: Parser) -> None:
826
827
        """
        Adds the particular copy of the parser object to this
828
        particular instance of Grammar.
829
        """
830
        if parser.name:
Eckhart Arnold's avatar
Eckhart Arnold committed
831
832
833
            # prevent overwriting instance variables or parsers of a different class
            assert parser.name not in self.__dict__ or \
                   isinstance(self.__dict__[parser.name], parser.__class__), \
834
835
                ('Cannot add parser "%s" because a field with the same name '
                 'already exists in grammar object!' % parser.name)
836
            setattr(self, parser.name, parser)
837
        self.all_parsers__.add(parser)
838
839
        parser.grammar = self

840

Eckhart Arnold's avatar
Eckhart Arnold committed
841
    def __call__(self, document: str, start_parser="root__") -> Node:
842
843
        """
        Parses a document with with parser-combinators.
844
845
846

        Args:
            document (str): The source text to be parsed.
847
848
849
            start_parser (str): The name of the parser with which to
                start. This is useful for testing particular parsers
                (i.e. particular parts of the EBNF-Grammar.)
850
851
852
        Returns:
            Node: The root node ot the parse tree.
        """
853
        # assert isinstance(document, str), type(document)
854
855
        if self.root__ is None:
            raise NotImplementedError()
856
        if self._dirty_flag__:
857
            self._reset__()
858
            for parser in self.all_parsers__:
859
860
                parser.reset()
        else:
861
            self._dirty_flag__ = True
862
        self.history_tracking__ = is_logging()
863
        self.document__ = StringView(document)
864
        self.document_lbreaks__ = linebreaks(document) if self.history_tracking__ else []
865
        self.last_rb__loc__ = len(self.document__) + 1  # rollback location
Eckhart Arnold's avatar
Eckhart Arnold committed
866
        parser = self[start_parser] if isinstance(start_parser, str) else start_parser
867
868
        assert parser.grammar == self, "Cannot run parsers from a different grammar object!" \
                                       " %s vs. %s" % (str(self), str(parser.grammar))
869
        result = None  # type: Optional[Node]
870
        stitches = []  # type: List[Node]
871
        rest = self.document__
872
        if not rest:
eckhart's avatar
eckhart committed
873
            result, _ = parser(rest)
di68kap's avatar
di68kap committed
874
875
876
            if result is None:
                result = Node(None, '')
                result.add_error('Parser "%s" did not match empty document.' % str(parser))
877
878
879
880
881
882
        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:
883
884
885
886
                    error_msg = 'Parser did not match! Invalid source file?' \
                                '\n    Most advanced: %s\n    Last match:    %s;' % \
                                (str(HistoryRecord.most_advanced_match(self.history__)),
                                 str(HistoryRecord.last_match(self.history__)))
887
888
889
                else:
                    stitches.append(result)
                    error_msg = "Parser stopped before end" + \
890
891
                                (("! trying to recover" +
                                  (" but stopping history recording at this point."
892
                                   if self.history_tracking__ else "..."))
893
894
895
896
                                 if len(stitches) < MAX_DROPOUTS
                                 else " too often! Terminating parser.")
                stitches.append(Node(None, skip))
                stitches[-1].add_error(error_msg)
897
                if self.history_tracking__:
898
                    # some parsers may have matched and left history records with nodes != None.
899
                    # Because these are not connected to the stitched root node, their pos-
900
901
                    # properties will not be initialized by setting the root node's pos property
                    # to zero. Therefore, their pos properties need to be initialized here
902
                    for record in self.history__:
903
904
                        if record.node and record.node._pos < 0:
                            record.node.pos = 0
905
                    record = HistoryRecord(self.call_stack__.copy(), stitches[-1], rest)
906
                    self.history__.append(record)
907
908
                    # stop history tracking when parser returned too early
                    self.history_tracking__ = False
909
910
911
912
        if stitches:
            if rest:
                stitches.append(Node(None, rest))
            result = Node(None, tuple(stitches))
913
        if any(self.variables__.values()):
Eckhart Arnold's avatar
Eckhart Arnold committed
914
915
            error_str = "Capture-retrieve-stack not empty after end of parsing: " + \
                            str(self.variables__)
916
917
918
919
920
921
922
923
924
925
            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.
                    error_node = Node(ZOMBIE_PARSER, '')
                    error_node.add_error(error_str)
                    result.result = result.children + (error_node,)
                else:
                    result.add_error(error_str)
926
        result.pos = 0  # calculate all positions
927
        # result.collect_errors(self.document__)
928
929
        return result

930

931
    def push_rollback__(self, location, func):
932
933
        """
        Adds a rollback function that either removes or re-adds
934
935
936
937
        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.
        """
938
939
940
        self.rollback__.append((location, func))
        self.last_rb__loc__ = location

941

942
    def rollback_to__(self, location):
943
944
        """
        Rolls back the variable stacks (`self.variables`) to its
945
946
        state at an earlier location in the parsed document.
        """
947
        while self.rollback__ and self.rollback__[-1][0] <= location:
948
            _, rollback_func = self.rollback__.pop()
949
950
951
952
            # 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__))
953
954
955
956
            rollback_func()
        self.last_rb__loc__ == self.rollback__[-1][0] if self.rollback__ \
            else (len(self.document__) + 1)

957

di68kap's avatar
di68kap committed
958
    def log_parsing_history__(self, log_file_name: str = '', html: bool=True) -> None:
959
960
        """
        Writes a log of the parsing history of the most recently parsed
eckhart's avatar
eckhart committed
961
        document.
962
963
        """
        def write_log(history, log_name):
eckhart's avatar
eckhart committed
964
965
            htm = '.html' if html else ''
            path = os.path.join(log_dir(), log_name + "_parser.log" + htm)
Eckhart Arnold's avatar
Eckhart Arnold committed
966
967
968
            if os.path.exists(path):
                os.remove(path)
                print('WARNING: Log-file "%s" already existed and was deleted.' % path)
969
970
            if history:
                with open(path, "w", encoding="utf-8") as f:
eckhart's avatar
eckhart committed
971
972
973
974
975
976
                    if html:
                        f.write(HistoryRecord.HTML_LEAD_IN)
                        f.write("\n".join(history))
                        f.write(HistoryRecord.HTML_LEAD_OUT)
                    else:
                        f.write("\n".join(history))
977

di68kap's avatar
di68kap committed
978
979
980
981
982
983
984
985
        def append_line(log, line):
            """Appends a line to a list of HTML table rows. Starts a new
            table every 100 rows to allow browser to speed up rendering.
            Does this really work...?"""
            log.append(line)
            if html and len(log) % 100 == 0:
                log.append('\n</table>\n<table>\n' + HistoryRecord.COLGROUP)

986
987
988
989
990
991
        if is_logging():
            assert self.history__, \
                "Parser did not yet run or logging was turned off when running parser!"
            if not log_file_name:
                name = self.__class__.__name__
                log_file_name = name[:-7] if name.lower().endswith('grammar') else name
di68kap's avatar
di68kap committed
992
993
            elif log_file_name.lower().endswith('.log'):
                log_file_name = log_file_name[:-4]
994
995
            full_history, match_history, errors_only = [], [], []
            for record in self.history__:
eckhart's avatar
eckhart committed
996
                line = record.as_html_tr() if html else str(record)
di68kap's avatar
di68kap committed
997
                append_line(full_history, line)
998
                if record.node and record.node.parser.ptype != WHITESPACE_PTYPE:
di68kap's avatar
di68kap committed
999
                    append_line(match_history, line)
1000
                    if record.node.error_flag:
di68kap's avatar
di68kap committed
1001
                        append_line(errors_only, line)
1002
1003
1004
            write_log(full_history, log_file_name + '_full')
            write_log(match_history, log_file_name + '_match')
            write_log(errors_only, log_file_name + '_errors')
1005
1006


Eckhart Arnold's avatar
Eckhart Arnold committed
1007
def dsl_error_msg(parser: Parser, error_str: str) -> str:
1008
1009
    """
    Returns an error message for errors in the parser configuration,
1010
1011
1012
    e.g. errors that result in infinite loops.

    Args:
Eckhart Arnold's avatar
Eckhart Arnold committed
1013
        parser (Parser):  The parser where the error was noticed. Note
1014
            that this is not necessarily the parser that caused the
1015
            error but only where the error became apparent.
Eckhart Arnold's avatar