transform.py 20 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
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
"""transformation.py - transformation functions for converting the
                       concrete into the abstract syntax tree

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.
"""

import inspect
from functools import partial, reduce, singledispatch

from DHParser.syntaxtree import WHITESPACE_PTYPE, TOKEN_PTYPE, MockParser, Node

try:
    import regex as re
except ImportError:
    import re
try:
    from typing import AbstractSet, Any, ByteString, Callable, cast, Container, Dict, \
        Iterator, List, NamedTuple, Sequence, Union, Text, Tuple
except ImportError:
    from .typing34 import AbstractSet, Any, ByteString, Callable, cast, Container, Dict, \
        Iterator, List, NamedTuple, Sequence, Union, Text, Tuple

from DHParser.toolkit import expand_table, smart_list

__all__ = ('transformation_factory',
           'key_parser_name',
           'key_tag_name',
           'traverse',
42
           'is_named',
43
44
           'replace_by_single_child',
           'reduce_single_child',
45
           'replace_or_reduce',
46
47
           'replace_parser',
           'collapse',
48
           'merge_children',
49
50
           'replace_content',
           'apply_if',
51
           'is_anonymous',
52
53
54
55
           'is_whitespace',
           'is_empty',
           'is_expendable',
           'is_token',
56
           'is_one_of',
57
           'has_content',
58
           'remove_children_if',
59
60
61
62
63
64
65
66
           'remove_parser',
           'remove_content',
           'remove_first',
           'remove_last',
           'remove_whitespace',
           'remove_empty',
           'remove_expendables',
           'remove_brackets',
67
68
           'remove_infix_operator',
           'remove_single_child',
69
           'remove_tokens',
70
           'keep_children',
71
72
73
           'flatten',
           'forbid',
           'require',
74
75
76
77
           'assert_content',
           'assert_condition',
           'assert_has_children',
           'TRUE_CONDITION')
78
79


80
81
82
83
84
TransformationProc = Callable[[List[Node]], None]
ConditionFunc = Callable  # Callable[[List[Node]], bool]
KeyFunc = Callable[[Node], str]


85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
def transformation_factory(t=None):
    """Creates factory functions from transformation-functions that
    dispatch on the first parameter after the node parameter.

    Decorating a transformation-function that has more than merely the
    ``node``-parameter with ``transformation_factory`` creates a
    function with the same name, which returns a partial-function that
    takes just the node-parameter.

    Additionally, there is some some syntactic sugar for
    transformation-functions that receive a collection as their second
    parameter and do not have any further parameters. In this case a
    list of parameters passed to the factory function will be converted
    into a collection.

    Main benefit is readability of processing tables.

    Usage:
        @transformation_factory(AbtractSet[str])
104
        def remove_tokens(context, tokens):
105
106
107
            ...
      or, alternatively:
        @transformation_factory
108
        def remove_tokens(context, tokens: AbstractSet[str]):
109
110
111
112
113
114
            ...

    Example:
        trans_table = { 'expression': remove_tokens('+', '-') }
      instead of:
        trans_table = { 'expression': partial(remove_tokens, tokens={'+', '-'}) }
115
116
117
118
119

    Parameters:
        t:  type of the second argument of the transformation function,
            only necessary if the transformation functions' parameter list
            does not have type annotations.
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
    """

    def decorator(f):
        sig = inspect.signature(f)
        params = list(sig.parameters.values())[1:]
        if len(params) == 0:
            return f  # '@transformer' not needed w/o free parameters
        assert t or params[0].annotation != params[0].empty, \
            "No type information on second parameter found! Please, use type " \
            "annotation or provide the type information via transfomer-decorator."
        p1type = t or params[0].annotation
        f = singledispatch(f)
        if len(params) == 1 and issubclass(p1type, Container) and not issubclass(p1type, Text) \
                and not issubclass(p1type, ByteString):
            def gen_special(*args):
                c = set(args) if issubclass(p1type, AbstractSet) else \
                    list(args) if issubclass(p1type, Sequence) else args
                d = {params[0].name: c}
                return partial(f, **d)

            f.register(p1type.__args__[0], gen_special)

        def gen_partial(*args, **kwargs):
            d = {p.name: arg for p, arg in zip(params, args)}
            d.update(kwargs)
            return partial(f, **d)

        f.register(p1type, gen_partial)
        return f

    if isinstance(t, type(lambda: 1)):
        # Provide for the case that transformation_factory has been
        # written as plain decorator and not as a function call that
        # returns the decorator proper.
154
        func = t
155
156
157
158
159
160
        t = None
        return decorator(func)
    else:
        return decorator


161
def key_parser_name(node: Node) -> str:
162
163
164
    return node.parser.name


165
def key_tag_name(node: Node) -> str:
166
167
168
    return node.tag_name


169
170
171
172
173
def traverse(root_node: Node,
             processing_table: Dict[str, List[Callable]],
             key_func: KeyFunc=key_tag_name) -> None:
    """
    Traverses the snytax tree starting with the given ``node`` depth
174
175
176
177
178
179
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
    first and applies the sequences of callback-functions registered
    in the ``calltable``-dictionary.

    The most important use case is the transformation of a concrete
    syntax tree into an abstract tree (AST). But it is also imaginable
    to employ tree-traversal for the semantic analysis of the AST.

    In order to assign sequences of callback-functions to nodes, a
    dictionary ("processing table") is used. The keys usually represent
    tag names, but any other key function is possible. There exist
    three special keys:
        '+': always called (before any other processing function)
        '*': called for those nodes for which no (other) processing
             function appears in the table
        '~': always called (after any other processing function)

    Args:
        root_node (Node): The root-node of the syntax tree to be traversed
        processing_table (dict): node key -> sequence of functions that
            will be applied to matching nodes in order. This dictionary
            is interpreted as a ``compact_table``. See
            ``toolkit.expand_table`` or ``EBNFCompiler.EBNFTransTable``
        key_func (function): A mapping key_func(node) -> keystr. The default
            key_func yields node.parser.name.

    Example:
        table = { "term": [replace_by_single_child, flatten],
            "factor, flowmarker, retrieveop": replace_by_single_child }
        traverse(node, table)
    """
    # normalize processing_table entries by turning single values into lists
    # with a single value
    table = {name: smart_list(call) for name, call in list(processing_table.items())}
    table = expand_table(table)
    cache = {}  # type: Dict[str, List[Callable]]

210
211
    def traverse_recursive(context):
        node = context[-1]
212
213
        if node.children:
            for child in node.result:
214
215
                context.append(child)
                traverse_recursive(context)  # depth first
216
                node.error_flag |= child.error_flag  # propagate error flag
217
                context.pop()
218
219
220
221
222
223
224
225
226
227
228
229
230
231

        key = key_func(node)
        sequence = cache.get(key, None)
        if sequence is None:
            sequence = table.get('+', []) + \
                       table.get(key, table.get('*', [])) + \
                       table.get('~', [])
            # '+' always called (before any other processing function)
            # '*' called for those nodes for which no (other) processing function
            #     appears in the table
            # '~' always called (after any other processing function)
            cache[key] = sequence

        for call in sequence:
232
            call(context)
233

234
    traverse_recursive([root_node])
235
236
237
238
239
240
241
242
243
244
245
246
247


# ------------------------------------------------
#
# rearranging transformations:
#     - tree may be rearranged (e.g.flattened)
#     - nodes that are not leaves may be dropped
#     - order is preserved
#     - leave content is preserved (though not necessarily the leaves themselves)
#
# ------------------------------------------------


248
def TRUE_CONDITION(context: List[Node]) -> bool:
249
250
251
    return True


252
def replace_child(node: Node):
253
254
255
256
257
258
259
260
    assert len(node.children) == 1
    if not node.children[0].parser.name:
        node.children[0].parser.name = node.parser.name
    node.parser = node.children[0].parser
    node._errors.extend(node.children[0]._errors)
    node.result = node.result[0].result


261
def reduce_child(node: Node):
262
263
264
265
266
    assert len(node.children) == 1
    node._errors.extend(node.children[0]._errors)
    node.result = node.result[0].result


267
@transformation_factory(Callable)
268
269
270
def replace_by_single_child(context: List[Node], condition: Callable=TRUE_CONDITION):
    """
    Remove single branch node, replacing it by its immediate descendant
271
    if and only if the condision on the descendant is true.
272
273
274
    (In case the descendant's name is empty (i.e. anonymous) the
    name of this node's parser is kept.)
    """
275
    node = context[-1]
276
277
278
279
280
    if len(node.children) == 1:
        context.append(node.children[0])
        if  condition(context):
            replace_child(node)
        context.pop()
281
282


283
@transformation_factory(Callable)
284
285
286
def reduce_single_child(context: List[Node], condition: Callable=TRUE_CONDITION):
    """
    Reduce a single branch node, by transferring the result of its
287
    immediate descendant to this node, but keeping this node's parser entry.
288
289
    If the condition evaluates to false on the descendant, it will not
    be reduced.
290
    """
291
    node = context[-1]
292
293
294
295
296
    if len(node.children) == 1:
        context.append(node.children[0])
        if condition(context):
            reduce_child(node)
        context.pop()
297
298


299
300
301
302
303
304
def is_named(context: List[Node]) -> bool:
    return context[-1].parser.name


def is_anonymous(context: List[Node]) -> bool:
    return not context[-1].parser.name
305
306
307


@transformation_factory(Callable)
308
309
310
def replace_or_reduce(context: List[Node], condition: Callable=is_named):
    """
    Replaces node by a single child, if condition is met on child,
311
312
    otherwise (i.e. if the child is anonymous) reduces the child.
    """
313
    node = context[-1]
314
315
316
317
318
319
320
    if len(node.children) == 1:
        context.append(node.children[0])
        if condition(context):
            replace_child(node)
        else:
            reduce_child(node)
        context.pop()
321
322
323


@transformation_factory
324
325
326
def replace_parser(context: List[Node], name: str):
    """
    Replaces the parser of a Node with a mock parser with the given
327
328
329
330
331
332
    name.

    Parameters:
        name(str): "NAME:PTYPE" of the surogate. The ptype is optional
        node(Node): The node where the parser shall be replaced
    """
333
    node = context[-1]
334
335
336
337
338
    name, ptype = (name.split(':') + [''])[:2]
    node.parser = MockParser(name, ptype)


@transformation_factory(Callable)
339
340
341
def flatten(context: List[Node], condition: Callable=is_anonymous, recursive: bool=True):
    """
    Flattens all children, that fulfil the given `condition`
342
343
344
345
346
347
348
349
350
351
352
353
    (default: all unnamed children). Flattening means that wherever a
    node has child nodes, the child nodes are inserted in place of the
    node.
     If the parameter `recursive` is `True` the same will recursively be
    done with the child-nodes, first. In other words, all leaves of
    this node and its child nodes are collected in-order as direct
    children of this node.
     Applying flatten recursively will result in these kinds of
    structural transformation:
        (1 (+ 2) (+ 3)     ->   (1 + 2 + 3)
        (1 (+ (2 + (3))))  ->   (1 + 2 + 3)
    """
354
    node = context[-1]
355
356
357
    if node.children:
        new_result = []
        for child in node.children:
358
359
            context.append(child)
            if child.children and condition(context):
360
                if recursive:
361
                    flatten(context, condition, recursive)
362
363
364
                new_result.extend(child.children)
            else:
                new_result.append(child)
365
            context.pop()
366
367
368
        node.result = tuple(new_result)


369
370
371
def collapse(context: List[Node]):
    """
    Collapses all sub-nodes of a node by replacing them with the
372
    string representation of the node.
373
    """
374
    node = context[-1]
375
376
377
378
    node.result = str(node)


@transformation_factory
379
380
381
382
383
def merge_children(context: List[Node], tag_names: List[str]):
    """
    Joins all children next to each other and with particular tag-
    names into a single child node with a mock-parser with the name of
    the first tag-name in the list.
384
    """
385
    node = context
386
    result = []
387
    name, ptype = ('', tag_names[0]) if tag_names[0][:1] == ':' else (tag_names[0], '')
388
    if node.children:
389
        i = 0
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
        L = len(node.children)
        while i < L:
            while i < L and not node.children[i].tag_name in tag_names:
                result.append(node.children[i])
                i += 1
            k = i + 1
            while (k < L and node.children[k].tag_name in tag_names
                   and bool(node.children[i].children) == bool(node.children[k].children)):
                k += 1
            if i < L:
                result.append(Node(MockParser(name, ptype),
                                   reduce(lambda a, b: a + b, (node.result for node in node.children[i:k]))))
            i = k
        node.result = tuple(result)


# ------------------------------------------------
#
# destructive transformations:
#     - tree may be rearranged (flattened),
#     - order is preserved
#     - but (irrelevant) leaves may be dropped
#     - errors of dropped leaves will be lost
#
# ------------------------------------------------


@transformation_factory
418
def replace_content(context: List[Node], func: Callable):  # Callable[[Node], ResultType]
419
420
421
    """Replaces the content of the node. ``func`` takes the node
    as an argument an returns the mapped result.
    """
422
    node = context[-1]
423
424
425
    node.result = func(node.result)


426
def is_whitespace(context: List[Node]) -> bool:
427
428
    """Removes whitespace and comments defined with the
    ``@comment``-directive."""
429
    return context[-1].parser.ptype == WHITESPACE_PTYPE
430
431


432
433
def is_empty(context: List[Node]) -> bool:
    return not context[-1].result
434
435


436
437
def is_expendable(context: List[Node]) -> bool:
    return is_empty(context) or is_whitespace(context)
438
439


440
441
def is_token(context: List[Node], tokens: AbstractSet[str] = frozenset()) -> bool:
    node = context[-1]
442
443
444
    return node.parser.ptype == TOKEN_PTYPE and (not tokens or node.result in tokens)


445
def is_one_of(context: List[Node], tag_name_set: AbstractSet[str]) -> bool:
446
447
    """Returns true, if the node's tag_name is on of the
    given tag names."""
448
    return context[-1].tag_name in tag_name_set
449
450


451
def has_content(context: List[Node], regexp: str) -> bool:
452
    """Checks a node's content against a regular expression."""
453
    return bool(re.match(regexp, str(context[-1])))
454
455


456
@transformation_factory(Callable)
457
458
def apply_if(context: List[Node], transformation: Callable, condition: Callable):
    """Applies a transformation only if a certain condition is met."""
459
    node = context[-1]
460
    if condition(node):
461
        transformation(context)
462
463


464
@transformation_factory(slice)
465
def keep_children(context: List[Node], section: slice = slice(None)):
466
    """Keeps only child-nodes which fall into a slice of the result field."""
467
    node = context[-1]
468
    if node.children:
469
        node.result = node.children[section]
470
471
472


@transformation_factory(Callable)
473
def remove_children_if(context: List[Node], condition: Callable, section: slice = slice(None)):
474
    """Removes all nodes from a slice of the result field if the function
475
    `condition(child_node)` evaluates to `True`."""
476
    node = context[-1]
477
    if node.children:
478
479
480
        c = node.children
        N = len(c)
        rng = range(*section.indices(N))
481
482
483
484
485
486
487
488
489
490
        node.result = tuple(c[i] for i in range(N)
                            if not i in rng or not condition(context + [c[i]]))
        # selection = []
        # for i in range(N):
        #     context.append(c[i])
        #     if not i in rng or not condition(context):
        #         selection.append(c[i])
        #     context.pop()
        # if len(selection) != c:
        #     node.result = tuple(selection)
491
492
493
494
495


remove_whitespace = remove_children_if(is_whitespace)  # partial(remove_children_if, condition=is_whitespace)
remove_empty = remove_children_if(is_empty)
remove_expendables = remove_children_if(is_expendable)  # partial(remove_children_if, condition=is_expendable)
496
497
498
499
500
remove_first = apply_if(keep_children(slice(1, None)), lambda nd: len(nd.children) > 1)
remove_last = apply_if(keep_children(slice(None, -1)), lambda nd: len(nd.children) > 1)
remove_brackets = apply_if(keep_children(slice(1, -1)), lambda nd: len(nd.children) >= 2)
remove_infix_operator = keep_children(slice(0, None, 2))
remove_single_child = apply_if(keep_children(slice(0)), lambda nd: len(nd.children) == 1)
501
502
503


@transformation_factory
504
def remove_tokens(context: List[Node], tokens: AbstractSet[str] = frozenset()):
505
506
507
    """Reomoves any among a particular set of tokens from the immediate
    descendants of a node. If ``tokens`` is the empty set, all tokens
    are removed."""
508
    remove_children_if(context, partial(is_token, tokens=tokens))
509
510
511


@transformation_factory
512
def remove_parser(context: List[Node], tag_names: AbstractSet[str]):
Eckhart Arnold's avatar
Eckhart Arnold committed
513
    """Removes children by tag name."""
514
    remove_children_if(context, partial(is_one_of, tag_name_set=tag_names))
515
516
517


@transformation_factory
518
def remove_content(context: List[Node], regexp: str):
519
    """Removes children depending on their string value."""
520
    remove_children_if(context, partial(has_content, regexp=regexp))
521
522
523
524


########################################################################
#
525
# AST semantic validation functions (EXPERIMENTAL!!!)
526
527
528
#
########################################################################

529
@transformation_factory(Callable)
530
def assert_condition(context: List[Node], condition: Callable, error_msg: str = '') -> bool:
531
    """Checks for `condition`; adds an error message if condition is not met."""
532
    node = context[-1]
533
    if not condition(context):
534
535
536
537
538
539
540
541
542
543
544
545
546
        if error_msg:
            node.add_error(error_msg % node.tag_name if error_msg.find("%s") > 0 else error_msg)
        else:
            cond_name = condition.__name__ if hasattr(condition, '__name__') \
                        else condition.__class__.__name__ if hasattr(condition, '__class__') \
                        else '<unknown>'
            node.add_error("transform.assert_condition: Failed to meet condition " + cond_name)


assert_has_children = assert_condition(lambda nd: nd.children, 'Element "%s" has no children')


@transformation_factory
547
def assert_content(context: List[Node], regexp: str):
548
    node = context[-1]
549
    if not has_content(context, regexp):
550
551
552
        node.add_error('Element "%s" violates %s on %s' %
                       (node.parser.name, str(regexp), str(node)))

553
554

@transformation_factory
555
def require(context: List[Node], child_tags: AbstractSet[str]):
556
    node = context[-1]
557
558
559
560
561
562
563
    for child in node.children:
        if child.tag_name not in child_tags:
            node.add_error('Element "%s" is not allowed inside "%s".' %
                           (child.parser.name, node.parser.name))


@transformation_factory
564
def forbid(context: List[Node], child_tags: AbstractSet[str]):
565
    node = context[-1]
566
567
568
569
    for child in node.children:
        if child.tag_name in child_tags:
            node.add_error('Element "%s" cannot be nested inside "%s".' %
                           (child.parser.name, node.parser.name))