transform.py 21.4 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
"""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

23
from DHParser.syntaxtree import Node, WHITESPACE_PTYPE, TOKEN_PTYPE, MockParser
24

25
26
27
28
from DHParser.toolkit import expand_table, smart_list, re, typing

from typing import AbstractSet, Any, ByteString, Callable, cast, Container, Dict, \
    Iterator, List, NamedTuple, Sequence, Union, Text, Tuple
29

30
31
32
33
34
__all__ = ('TransformationDict',
           'TransformationProc',
           'ConditionFunc',
           'KeyFunc',
           'transformation_factory',
35
36
37
           'key_parser_name',
           'key_tag_name',
           'traverse',
38
           'is_named',
39
40
           'replace_by_single_child',
           'reduce_single_child',
41
           'replace_or_reduce',
42
43
           'replace_parser',
           'collapse',
44
           'merge_children',
45
46
           'replace_content',
           'apply_if',
47
           'is_anonymous',
48
49
50
51
           'is_whitespace',
           'is_empty',
           'is_expendable',
           'is_token',
52
           'is_one_of',
53
           'has_content',
54
           'remove_children_if',
eckhart's avatar
eckhart committed
55
           'remove_children',
56
57
58
59
60
61
62
63
           'remove_parser',
           'remove_content',
           'remove_first',
           'remove_last',
           'remove_whitespace',
           'remove_empty',
           'remove_expendables',
           'remove_brackets',
64
65
           'remove_infix_operator',
           'remove_single_child',
66
           'remove_tokens',
67
           'keep_children',
68
69
70
           'flatten',
           'forbid',
           'require',
71
72
73
74
           'assert_content',
           'assert_condition',
           'assert_has_children',
           'TRUE_CONDITION')
75
76


77
TransformationProc = Callable[[List[Node]], None]
Eckhart Arnold's avatar
Eckhart Arnold committed
78
79
TransformationDict = Dict[str, Sequence[Callable]]
ProcessingTableType = Dict[str, Union[Sequence[Callable], TransformationDict]]
80
81
82
83
ConditionFunc = Callable  # Callable[[List[Node]], bool]
KeyFunc = Callable[[Node], str]


84
85
def transformation_factory(t=None):
    """Creates factory functions from transformation-functions that
86
    dispatch on the first parameter after the context parameter.
87
88
89
90

    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
91
    takes just the context-parameter.
92
93
94
95
96
97
98
99
100
101
102

    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])
103
        def remove_tokens(context, tokens):
104
105
106
            ...
      or, alternatively:
        @transformation_factory
107
        def remove_tokens(context, tokens: AbstractSet[str]):
108
109
110
111
112
113
            ...

    Example:
        trans_table = { 'expression': remove_tokens('+', '-') }
      instead of:
        trans_table = { 'expression': partial(remove_tokens, tokens={'+', '-'}) }
114
115
116
117
118

    Parameters:
        t:  type of the second argument of the transformation function,
            only necessary if the transformation functions' parameter list
            does not have type annotations.
119
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
    """

    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.
153
        func = t
154
155
156
157
158
159
        t = None
        return decorator(func)
    else:
        return decorator


160
def key_parser_name(node: Node) -> str:
161
162
163
    return node.parser.name


164
def key_tag_name(node: Node) -> str:
165
166
167
    return node.tag_name


168
def traverse(root_node: Node,
Eckhart Arnold's avatar
Eckhart Arnold committed
169
             processing_table: ProcessingTableType,
170
171
172
             key_func: KeyFunc=key_tag_name) -> None:
    """
    Traverses the snytax tree starting with the given ``node`` depth
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
    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
193
194
            is interpreted as a `compact_table`. See
            `toolkit.expand_table` or ``EBNFCompiler.EBNFTransTable`
195
196
197
198
199
200
201
202
        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)
    """
203
204
205
206
207
208
    # Is this optimazation really needed?
    if '__cache__' in processing_table:
        # assume that processing table has already been expanded
        table = processing_table
        cache = processing_table['__cache__']
    else:
209
210
        # normalize processing_table entries by turning single values
        # into lists with a single value
211
212
        table = {name: cast(Sequence[Callable], smart_list(call))
                 for name, call in list(processing_table.items())}
213
        table = expand_table(table)
Eckhart Arnold's avatar
Eckhart Arnold committed
214
        cache = table.setdefault('__cache__', cast(TransformationDict, dict()))
215
216
        # change processing table in place, so its already expanded and cache filled next time
        processing_table.clear()
217
218
219
220
221
222
        processing_table.update(table)

    # assert '__cache__' in processing_table
    # # Code without optimization
    # table = {name: smart_list(call) for name, call in list(processing_table.items())}
    # table = expand_table(table)
Eckhart Arnold's avatar
Eckhart Arnold committed
223
    # cache = {}  # type: Dict[str, List[Callable]]
224

225
226
    def traverse_recursive(context):
        node = context[-1]
227
228
        if node.children:
            for child in node.result:
229
230
                context.append(child)
                traverse_recursive(context)  # depth first
231
                node.error_flag = max(node.error_flag, child.error_flag)  # propagate error flag
232
                context.pop()
233
234

        key = key_func(node)
235
236
237
        try:
            sequence = cache[key]
        except KeyError:
238
239
240
241
242
243
244
245
246
247
            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:
248
            call(context)
249

250
    traverse_recursive([root_node])
251
252
    # assert processing_table['__cache__']

253
254
255
256
257
258
259
260
261
262
263
264

# ------------------------------------------------
#
# 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)
#
# ------------------------------------------------


265
def TRUE_CONDITION(context: List[Node]) -> bool:
266
267
268
    return True


269
def replace_child(node: Node):
270
    assert len(node.children) == 1
271
272
273
274
275
276
277
    child = node.children[0]
    if not child.parser.name:
        child.parser = MockParser(node.parser.name, child.parser.ptype)
        # parser names must not be overwritten, else: child.parser.name = node.parser.name
    node.parser = child.parser
    node._errors.extend(child._errors)
    node.result = child.result
278
279


280
def reduce_child(node: Node):
281
282
    assert len(node.children) == 1
    node._errors.extend(node.children[0]._errors)
Eckhart Arnold's avatar
Eckhart Arnold committed
283
    node.result = node.children[0].result
284
285


286
@transformation_factory(Callable)
287
288
289
def replace_by_single_child(context: List[Node], condition: Callable=TRUE_CONDITION):
    """
    Remove single branch node, replacing it by its immediate descendant
290
    if and only if the condision on the descendant is true.
291
292
293
    (In case the descendant's name is empty (i.e. anonymous) the
    name of this node's parser is kept.)
    """
294
    node = context[-1]
295
296
    if len(node.children) == 1:
        context.append(node.children[0])
297
        if condition(context):
298
299
            replace_child(node)
        context.pop()
300
301


302
@transformation_factory(Callable)
303
304
305
def reduce_single_child(context: List[Node], condition: Callable=TRUE_CONDITION):
    """
    Reduce a single branch node, by transferring the result of its
306
    immediate descendant to this node, but keeping this node's parser entry.
307
308
    If the condition evaluates to false on the descendant, it will not
    be reduced.
309
    """
310
    node = context[-1]
311
312
313
314
315
    if len(node.children) == 1:
        context.append(node.children[0])
        if condition(context):
            reduce_child(node)
        context.pop()
316
317


318
def is_named(context: List[Node]) -> bool:
Eckhart Arnold's avatar
Eckhart Arnold committed
319
    return bool(context[-1].parser.name)
320
321
322
323


def is_anonymous(context: List[Node]) -> bool:
    return not context[-1].parser.name
324
325
326


@transformation_factory(Callable)
327
328
329
def replace_or_reduce(context: List[Node], condition: Callable=is_named):
    """
    Replaces node by a single child, if condition is met on child,
330
331
    otherwise (i.e. if the child is anonymous) reduces the child.
    """
332
    node = context[-1]
333
334
335
336
337
338
339
    if len(node.children) == 1:
        context.append(node.children[0])
        if condition(context):
            replace_child(node)
        else:
            reduce_child(node)
        context.pop()
340
341
342


@transformation_factory
343
344
345
def replace_parser(context: List[Node], name: str):
    """
    Replaces the parser of a Node with a mock parser with the given
346
347
348
    name.

    Parameters:
349
350
        context: the context where the parser shall be replaced
        name: "NAME:PTYPE" of the surogate. The ptype is optional
351
    """
352
    node = context[-1]
353
354
355
356
357
    name, ptype = (name.split(':') + [''])[:2]
    node.parser = MockParser(name, ptype)


@transformation_factory(Callable)
358
359
360
def flatten(context: List[Node], condition: Callable=is_anonymous, recursive: bool=True):
    """
    Flattens all children, that fulfil the given `condition`
361
362
363
364
365
366
367
368
369
370
371
372
    (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)
    """
373
    node = context[-1]
374
    if node.children:
Eckhart Arnold's avatar
Eckhart Arnold committed
375
        new_result = []     # type: List[Node]
376
        for child in node.children:
377
378
            context.append(child)
            if child.children and condition(context):
379
                if recursive:
380
                    flatten(context, condition, recursive)
381
382
383
                new_result.extend(child.children)
            else:
                new_result.append(child)
384
            context.pop()
385
386
387
        node.result = tuple(new_result)


388
389
390
def collapse(context: List[Node]):
    """
    Collapses all sub-nodes of a node by replacing them with the
391
    string representation of the node.
392
    """
393
    node = context[-1]
394
395
396
397
    node.result = str(node)


@transformation_factory
398
399
400
401
402
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.
403
    """
Eckhart Arnold's avatar
Eckhart Arnold committed
404
    node = context[-1]
405
    result = []
406
    name, ptype = ('', tag_names[0]) if tag_names[0][:1] == ':' else (tag_names[0], '')
407
    if node.children:
408
        i = 0
409
410
411
412
413
414
415
416
417
418
419
        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),
Eckhart Arnold's avatar
Eckhart Arnold committed
420
421
                                   reduce(lambda a, b: a + b,
                                          (node.children for node in node.children[i:k]))))
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
            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
438
def replace_content(context: List[Node], func: Callable):  # Callable[[Node], ResultType]
439
440
441
    """Replaces the content of the node. ``func`` takes the node
    as an argument an returns the mapped result.
    """
442
    node = context[-1]
443
444
445
    node.result = func(node.result)


446
def is_whitespace(context: List[Node]) -> bool:
447
448
    """Removes whitespace and comments defined with the
    ``@comment``-directive."""
449
    return context[-1].parser.ptype == WHITESPACE_PTYPE
450
451


452
453
def is_empty(context: List[Node]) -> bool:
    return not context[-1].result
454
455


456
457
def is_expendable(context: List[Node]) -> bool:
    return is_empty(context) or is_whitespace(context)
458
459


460
461
def is_token(context: List[Node], tokens: AbstractSet[str] = frozenset()) -> bool:
    node = context[-1]
462
463
464
    return node.parser.ptype == TOKEN_PTYPE and (not tokens or node.result in tokens)


465
def is_one_of(context: List[Node], tag_name_set: AbstractSet[str]) -> bool:
466
467
    """Returns true, if the node's tag_name is on of the
    given tag names."""
468
    return context[-1].tag_name in tag_name_set
469
470


471
def has_content(context: List[Node], regexp: str) -> bool:
472
    """Checks a node's content against a regular expression."""
473
    return bool(re.match(regexp, str(context[-1])))
474
475


476
@transformation_factory(Callable)
477
478
def apply_if(context: List[Node], transformation: Callable, condition: Callable):
    """Applies a transformation only if a certain condition is met."""
479
    node = context[-1]
480
    if condition(node):
481
        transformation(context)
482
483


484
@transformation_factory(slice)
485
def keep_children(context: List[Node], section: slice = slice(None)):
486
    """Keeps only child-nodes which fall into a slice of the result field."""
487
    node = context[-1]
488
    if node.children:
489
        node.result = node.children[section]
490
491
492


@transformation_factory(Callable)
493
def remove_children_if(context: List[Node], condition: Callable):  # , section: slice = slice(None)):
494
495
496
497
498
499
500
    """Removes all children for which `condition()` returns `True`."""
    node = context[-1]
    if node.children:
        node.result = tuple(c for c in node.children if not condition(context + [c]))


@transformation_factory(Callable)
eckhart's avatar
eckhart committed
501
502
503
def remove_children(context: List[Node],
                    condition: Callable = TRUE_CONDITION,
                    section: slice = slice(None)):
504
    """Removes all nodes from a slice of the result field if the function
505
    `condition(child_node)` evaluates to `True`."""
506
    node = context[-1]
507
    if node.children:
508
509
510
        c = node.children
        N = len(c)
        rng = range(*section.indices(N))
511
        node.result = tuple(c[i] for i in range(N)
512
                            if i not in rng or not condition(context + [c[i]]))
513
514
515
516
517
518
519
520
        # 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)
521
522
523
524
525


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)
526
527
528
529
530
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)
531
532
533


@transformation_factory
534
def remove_tokens(context: List[Node], tokens: AbstractSet[str] = frozenset()):
535
536
537
    """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."""
538
    remove_children_if(context, partial(is_token, tokens=tokens))
539
540
541


@transformation_factory
542
def remove_parser(context: List[Node], tag_names: AbstractSet[str]):
Eckhart Arnold's avatar
Eckhart Arnold committed
543
    """Removes children by tag name."""
544
    remove_children_if(context, partial(is_one_of, tag_name_set=tag_names))
545
546
547


@transformation_factory
548
def remove_content(context: List[Node], regexp: str):
549
    """Removes children depending on their string value."""
550
    remove_children_if(context, partial(has_content, regexp=regexp))
551
552
553
554


########################################################################
#
555
# AST semantic validation functions (EXPERIMENTAL!!!)
556
557
558
#
########################################################################

559
@transformation_factory(Callable)
Eckhart Arnold's avatar
Eckhart Arnold committed
560
def assert_condition(context: List[Node], condition: Callable, error_msg: str = ''):
561
    """Checks for `condition`; adds an error message if condition is not met."""
562
    node = context[-1]
563
    if not condition(context):
564
565
566
567
568
569
570
571
572
573
574
575
576
        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
577
def assert_content(context: List[Node], regexp: str):
578
    node = context[-1]
579
    if not has_content(context, regexp):
580
581
582
        node.add_error('Element "%s" violates %s on %s' %
                       (node.parser.name, str(regexp), str(node)))

583
584

@transformation_factory
585
def require(context: List[Node], child_tags: AbstractSet[str]):
586
    node = context[-1]
587
588
589
590
591
592
593
    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
594
def forbid(context: List[Node], child_tags: AbstractSet[str]):
595
    node = context[-1]
596
597
598
599
    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))