transform.py

# transform.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.


"""
Module ``transform`` contains the functions for transforming the
concrete syntax tree (CST) into an abstract syntax tree (AST).

As these functions are very generic, they can in principle be
used for any kind of tree transformations, not necessarily only
for CST -> AST transformations.
"""


import collections.abc
import inspect
from functools import partial, singledispatch

from DHParser.error import Error, ErrorCode
from DHParser.syntaxtree import Node, WHITESPACE_PTYPE, TOKEN_PTYPE, PLACEHOLDER, RootNode, parse_sxpr, flatten_sxpr
from DHParser.toolkit import issubtype, isgenerictype, expand_table, smart_list, re, typing
from typing import AbstractSet, Any, ByteString, Callable, cast, Container, Dict, \
    Tuple, List, Sequence, Union, Text, Generic

__all__ = ('TransformationDict',
           'TransformationProc',
           'TransformationFunc',
           'ConditionFunc',
           'KeyFunc',
           'transformation_factory',
           'key_tag_name',
           'traverse',
           'always',
           'is_named',
           'update_attr',
           'replace_by_single_child',
           'reduce_single_child',
           'replace_or_reduce',
           'change_tag_name',
           'collapse',
           'collapse_if',
           'replace_content',
           'replace_content_by',
           'normalize_whitespace',
           'move_adjacent',
           'left_associative',
           'lean_left',
           'apply_if',
           'apply_unless',
           'traverse_locally',
           'is_anonymous',
           'is_insignificant_whitespace',
           'contains_only_whitespace',
           'is_any_kind_of_whitespace',
           'is_empty',
           'is_token',
           'is_one_of',
           'not_one_of',
           'matches_re',
           'has_content',
           'has_parent',
           'lstrip',
           'rstrip',
           'strip',
           'keep_children',
           'keep_children_if',
           'keep_tokens',
           'keep_nodes',
           'keep_content',
           'remove_children_if',
           'remove_nodes',
           'remove_content',
           'remove_first',
           'remove_last',
           'remove_whitespace',
           'remove_empty',
           'remove_anonymous_empty',
           'remove_anonymous_tokens',
           'remove_brackets',
           'remove_infix_operator',
           'remove_single_child',
           'remove_tokens',
           'flatten',
           'forbid',
           'require',
           'assert_content',
           'error_on',
           'assert_has_children',
           'peek')


TransformationProc = Callable[[List[Node]], None]
TransformationDict = Dict[str, Sequence[Callable]]
TransformationFunc = Union[Callable[[Node], Any], partial]
ProcessingTableType = Dict[str, Union[Sequence[Callable], TransformationDict]]
ConditionFunc = Callable  # Callable[[List[Node]], bool]
KeyFunc = Callable[[Node], str]
CriteriaType = Union[int, str, Callable]


def transformation_factory(t1=None, t2=None, t3=None, t4=None, t5=None):
    """
    Creates factory functions from transformation-functions that
    dispatch on the first parameter after the context parameter.

    Decorating a transformation-function that has more than merely the
    ``context``-parameter with ``transformation_factory`` creates a
    function with the same name, which returns a partial-function that
    takes just the context-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(AbstractSet[str])
        def remove_tokens(context, tokens):
            ...

    or, alternatively::

        @transformation_factory
        def remove_tokens(context, tokens: AbstractSet[str]):
            ...

    Example::

        trans_table = { 'expression': remove_tokens('+', '-') }

    instead of::

        trans_table = { 'expression': partial(remove_tokens, tokens={'+', '-'}) }

    Parameters:
        t1:  type of the second argument of the transformation function,
            only necessary if the transformation functions' parameter list
            does not have type annotations.
    """

    def type_guard(t):
        """Raises an error if type `t` is a generic type or could be mistaken
        for the type of the canonical first parameter "List[Node] of
        transformation functions. Returns `t`."""
        # if isinstance(t, GenericMeta):
        #     raise TypeError("Generic Type %s not permitted\n in transformation_factory "
        #                     "decorator. Use the equivalent non-generic type instead!"
        #                     % str(t))
        if isinstance(t, str):          # ensure compatibility with python versions
            t = eval(t.replace('unicode', 'str'))  # with alternative type handling.
        if isgenerictype(t):
            raise TypeError("Generic Type %s not permitted\n in transformation_factory "
                            "decorator. Use the equivalent non-generic type instead!"
                            % str(t))
        if issubtype(List[Node], t):
            raise TypeError("Sequence type %s not permitted\nin transformation_factory "
                            "decorator, because it could be mistaken for a base class "
                            "of List[Node]\nwhich is the type of the canonical first "
                            "argument of transformation functions. Try 'tuple' instead!"
                            % str(t))
        return t

    def decorator(f):
        nonlocal t1
        sig = inspect.signature(f)
        params = list(sig.parameters.values())[1:]
        if len(params) == 0:
            return f  # '@transformer' not needed w/o free parameters
        assert t1 or params[0].annotation != params[0].empty, \
            "No type information on second parameter found! Please, use type " \
            "annotation or provide the type information via transformer-decorator."
        f = singledispatch(f)
        p1type = params[0].annotation
        if t1 is None:
            t1 = type_guard(p1type)
        elif issubtype(p1type, type_guard(t1)):
            try:
                if len(params) == 1 and issubtype(p1type, Container) \
                        and not (issubtype(p1type, Text) or issubtype(p1type, ByteString)):
                    def gen_special(*args):
                        c = set(args) if issubtype(p1type, AbstractSet) else \
                            tuple(args) if issubtype(p1type, Sequence) else args
                        d = {params[0].name: c}
                        return partial(f, **d)
                    f.register(type_guard(p1type.__args__[0]), gen_special)
            except AttributeError:
                pass  # Union Type does not allow subclassing, but is not needed here
        else:
            raise TypeError("Annotated type %s is not a subclass of decorated type %s !"
                            % (str(p1type), str(t1)))

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

        for t in (t1, t2, t3, t4, t5):
            if t:
                f.register(type_guard(t), gen_partial)
            else:
                break
        return f

    if isinstance(t1, 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.
        func = t1
        t1 = None
        return decorator(func)
    else:
        return decorator


def key_tag_name(node: Node) -> str:
    """
    Returns the tag name of the node as key of selecting transformations
    from the transformation table.
    """
    return node.tag_name


def traverse(root_node: Node,
             processing_table: ProcessingTableType,
             key_func: KeyFunc = key_tag_name) -> None:
    """
    Traverses the syntax tree starting with the given ``node`` depth
    first and applies the sequences of callback-functions registered
    in the ``processing_table``-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
            :func:`expand_table` or :func:`EBNFCompiler.EBNFTransTable`
        key_func (function): A mapping key_func(node) -> keystr. The default
            key_func yields node.tag_name.

    Example::

        table = { "term": [replace_by_single_child, flatten],
                  "factor, flowmarker, retrieveop": replace_by_single_child }
        traverse(node, table)

    """

    # Is this optimazation really needed?
    if '__cache__' in processing_table:
        # assume that processing table has already been expanded
        table = processing_table               # type: ProcessingTableType
        cache = cast(TransformationDict, processing_table['__cache__'])  # type: TransformationDict
    else:
        # normalize processing_table entries by turning single values
        # into lists with a single value
        table = {name: cast(Sequence[Callable], smart_list(call))
                 for name, call in list(processing_table.items())}
        table = expand_table(table)
        # substitute key for insiginificant whitespace
        assert '+' not in table, 'Symbol "+" in processing table is obsolete, use "<" instead'
        if '~' in table:
            if ':Whitespace' in table:
                raise AssertionError(
                    '"~" is a synonym for ":Whitespace" in the processing table. '
                    'To avoid confusion, choose either of the two, but do not use '
                    'both at the same time!')
            whitespace_transformation = table['~']
            del table['~']
            table[':Whitespace'] = whitespace_transformation
        # cache expanded table
        cache = cast(TransformationDict,
                     table.setdefault('__cache__', cast(TransformationDict, dict())))
        # change processing table in place, so its already expanded and cache filled next time
        processing_table.clear()
        processing_table.update(table)

    def traverse_recursive(context):
        nonlocal cache
        node = context[-1]
        if node.children:
            context.append(PLACEHOLDER)
            for child in node.children:
                context[-1] = child
                traverse_recursive(context)  # depth first
            context.pop()

        key = key_func(node)
        try:
            sequence = cache[key]
        except KeyError:
            sequence = table.get('<', []) \
                + table.get(key, table.get('*', [])) \
                + table.get('>', [])
            cache[key] = sequence

        for call in sequence:
            call(context)

    traverse_recursive([root_node])
    # assert processing_table['__cache__']


#######################################################################
#
# meta transformations, i.e. transformations that call other
# transformations
#
#######################################################################


@transformation_factory(dict)
def traverse_locally(context: List[Node],
                     processing_table: Dict,              # actually: ProcessingTableType
                     key_func: Callable = key_tag_name):  # actually: KeyFunc
    """
    Transforms the syntax tree starting from the last node in the context
    according to the given processing table. The purpose of this function is
    to apply certain transformations locally, i.e. only for those nodes that
    have the last node in the context as their parent node.
    """
    traverse(context[-1], processing_table, key_func)


@transformation_factory(collections.abc.Callable)
def apply_if(context: List[Node], transformation: Callable, condition: Callable):
    """
    Applies a transformation only if a certain condition is met.
    """
    if condition(context):
        transformation(context)


@transformation_factory(collections.abc.Callable)
def apply_unless(context: List[Node], transformation: Callable, condition: Callable):
    """
    Applies a transformation if a certain condition is *not* met.
    """
    if not condition(context):
        transformation(context)


#######################################################################
#
# conditionals that determine whether the context (or the last node in
# the context for that matter) fulfill a specific condition.
# ---------------------------------------------------------------------
#
# The context of a node is understood as a list of all parent nodes
# leading up to and including the node itself. If represented as list,
# the last element of the list is the node itself.
#
#######################################################################


def always(context: List[Node]) -> bool:
    """Always returns True, no matter that the state of the context."""
    return True


def is_single_child(context: List[Node]) -> bool:
    """Returns ``True`` if the current node does not have any siblings."""
    return len(context[-2].children) == 1


def is_named(context: List[Node]) -> bool:
    """Returns ``True`` if the current node's parser is a named parser."""
    return not context[-1].is_anonymous()


def is_anonymous(context: List[Node]) -> bool:
    """Returns ``True`` if the current node's parser is an anonymous parser."""
    return context[-1].is_anonymous()


def is_insignificant_whitespace(context: List[Node]) -> bool:
    """Returns ``True`` for whitespace and comments defined with the
    ``@comment``-directive."""
    return context[-1].tag_name == WHITESPACE_PTYPE


RX_WHITESPACE = re.compile(r'\s+')


def contains_only_whitespace(context: List[Node]) -> bool:
    """Returns ``True`` for nodes that contain only whitespace regardless
    of the tag_name, i.e. nodes the content of which matches the regular
    expression /\s*/, including empty nodes. Note, that this is not true
    for anonymous whitespace nodes that contain comments."""
    content = context[-1].content
    return bool(not content or RX_WHITESPACE.match(content))


def is_any_kind_of_whitespace(context: List[Node]) -> bool:
    """Returns ``True`` for nodes that either contain only whitespace or
    are insignificant whitespace nodes, i.e. nodes with the ``tag_name``
    ``PTYPE_WHITESPACE``, including those that contain comment-text."""
    node = context[-1]
    return node.tag_name == WHITESPACE_PTYPE or RX_WHITESPACE.match(node.content)


def is_empty(context: List[Node]) -> bool:
    """Returns ``True`` if the current node's content is empty."""
    return not context[-1].result


# DEPRECATED, because name is too ambiguous
# def is_expendable(context: List[Node]) -> bool:
#     """Returns ``True`` if the current node either is a node containing
#     whitespace or an empty node."""
#     return is_empty(context) or is_insignificant_whitespace(context)


@transformation_factory(collections.abc.Set)
def is_token(context: List[Node], tokens: AbstractSet[str] = frozenset()) -> bool:
    """
    Checks whether the last node in the context has `ptype == TOKEN_PTYPE`
    and it's content matches one of the given tokens. Leading and trailing
    whitespace-tokens will be ignored. In case an empty set of tokens is passed,
    any token is a match.
    """
    node = context[-1]
    return node.tag_name == TOKEN_PTYPE and (not tokens or node.content in tokens)


@transformation_factory(collections.abc.Set)
def is_one_of(context: List[Node], tag_name_set: AbstractSet[str]) -> bool:
    """Returns true, if the node's tag_name is one of the given tag names."""
    return context[-1].tag_name in tag_name_set


@transformation_factory(collections.abc.Set)
def not_one_of(context: List[Node], tag_name_set: AbstractSet[str]) -> bool:
    """Returns true, if the node's tag_name is not one of the given tag names."""
    return context[-1].tag_name not in tag_name_set


@transformation_factory(collections.abc.Set)
def matches_re(context: List[Node], patterns: AbstractSet[str]) -> bool:
    """
    Returns true, if the node's tag_name matches one of the regular
    expressions in `patterns`. For example, ':.*' matches all anonymous nodes.
    """
    tn = context[-1].tag_name
    for pattern in patterns:
        if re.match(pattern, tn):
            return True
    return False


@transformation_factory
def has_content(context: List[Node], regexp: str) -> bool:
    """
    Checks a node's content against a regular expression.

    In contrast to ``re.match`` the regular expression must match the complete
    string and not just the beginning of the string to succeed!
    """
    if not regexp.endswith('$'):
        regexp += "$"
    return bool(re.match(regexp, context[-1].content))


@transformation_factory(collections.abc.Set)
def has_parent(context: List[Node], tag_name_set: AbstractSet[str]) -> bool:
    """
    Checks whether a node with one of the given tag names appears somewhere
     in the context before the last node in the context.
     """
    for i in range(2, len(context) + 1):
        if context[-i].tag_name in tag_name_set:
            return True
    return False


#######################################################################
#
# utility functions (private)
#
#######################################################################


def update_attr(node: Node, child: Node):
    """
    Adds all attributes from `child` to `node`.This is needed, in order
    to keep the attributes if the child node is going to be eliminated.
    """
    if hasattr(child, '_xml_attr'):
        for k, v in child.attr:
            if k in node.attr and v != node.attr[k]:
                raise ValueError('Conflicting attribute values %s and %s for key %s '
                                 'when reducing %s to %s ! Tree transformation stopped.'
                                 % (v, node.attr[k], k, str(child), str(node)))
            node.attr[k] = v


def swap_attributes(node: Node, other: Node):
    """
    Exchanges the attributes between node and other. This might be
    needed when rearanging trees.
    """
    NA = node.attr_active()
    OA = other.attr_active()
    if NA or OA:
        save = node._xml_attr if NA else None
        if OA:
            node._xml_attr = other._xml_attr
        elif NA:
            node._xml_attr = None
        if NA:
            other._xml_attr = node._xml_attr
        elif OA:
            other._xml_attr = None


def _replace_by(node: Node, child: Node):
    """
    Replaces node's contents by child's content including the tag name.
    """
    if node.is_anonymous() or not child.is_anonymous():
        node.tag_name = child.tag_name
        # name, ptype = (node.tag_name.split(':') + [''])[:2]
        # child.parser = MockParser(name, ptype)
        # parser names must not be overwritten, else: child.parser.name = node.parser.name
    node.result = child.result
    update_attr(node, child)


def _reduce_child(node: Node, child: Node):
    """
    Sets node's results to the child's result, keeping node's tag_name.
    """
    node.result = child.result
    update_attr(node, child)


#######################################################################
#
# 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)
#
#######################################################################

# DEPRECATED
# def flatten_anonymous_nodes(context: List[Node]):
#     """
#     Flattens non-recursively all anonymous non-leaf children by adding
#     their result to the result of the parent node. Empty anonymous children
#     will be dropped altogether. If the parent node (i.e. `context[-1]) is
#     anonymous itself and has only one child node, it will be replaced by
#     its single child node.
#     """
#     node = context[-1]
#     if node.children:
#         new_result = []  # type: List[Node]
#         for child in node.children:
#             if child.is_anonymous():
#                 if child.children:
#                     new_result.extend(child.children)
#                     update_attr(node, child)
#                 elif child.result:
#                     new_result.append(child)
#             else:
#                 new_result.append(child)
#         if len(new_result) == 1:
#             child = new_result[0]
#             if node.is_anonymous():
#                 node.tag_name = child.tag_name
#                 node.result = child.result
#                 update_attr(node, child)
#                 return
#             elif child.is_anonymous():
#                 node.result = child.result
#                 update_attr(node, child)
#                 return
#         node.result = tuple(new_result)


def replace_by_single_child(context: List[Node]):
    """
    Removes single branch node, replacing it by its immediate descendant.
    Replacement only takes place, if the last node in the context has
    exactly one child.
    """
    node = context[-1]
    if len(node.children) == 1:
        _replace_by(node, node.children[0])


def reduce_single_child(context: List[Node]):
    """
    Reduces a single branch node by transferring the result of its
    immediate descendant to this node, but keeping this node's parser entry.
    Reduction only takes place if the last node in the context has
    exactly one child.
    """
    node = context[-1]
    if len(node.children) == 1:
        _reduce_child(node, node.children[0])


@transformation_factory(collections.abc.Callable)
def replace_or_reduce(context: List[Node], condition: Callable = is_named):
    """
    Replaces node by a single child, if condition is met on child,
    otherwise (i.e. if the child is anonymous) reduces the child.
    """
    node = context[-1]
    if len(node.children) == 1:
        child = node.children[0]
        if condition(context):
            _replace_by(node, child)
        else:
            _reduce_child(node, child)


@transformation_factory(str)
def change_tag_name(context: List[Node], tag_name: str, restriction: Callable = always):
    """
    Changes the tag name of a node.

    Parameters:
        restriction: A function of the context that returns False in cases
                where the tag name shall not be exchanged
        context: the context where the parser shall be replaced
        tag_name: The new tag name.
    """
    if restriction(context):
        node = context[-1]
        node.tag_name = tag_name


@transformation_factory(collections.abc.Callable)
def flatten(context: List[Node], condition: Callable = is_anonymous, recursive: bool = True):
    """
    Flattens all children, that fulfill the given ``condition``
    (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)
    """

    node = context[-1]
    if node.children:
        new_result = []     # type: List[Node]
        context.append(PLACEHOLDER)
        for child in node.children:
            context[-1] = child
            if child.children and condition(context):
                if recursive:
                    flatten(context, condition, recursive)
                new_result.extend(child.children)
                update_attr(node, child)
            else:
                new_result.append(child)
        context.pop()
        node.result = tuple(new_result)


def collapse(context: List[Node]):
    """
    Collapses all sub-nodes of a node by replacing them with the
    string representation of the node. USE WITH CARE!
    """
    node = context[-1]
    node.result = node.content


@transformation_factory(collections.abc.Callable)
def collapse_if(context: List[Node], condition: Callable, target_tag: str):
    """
    (Recursively) merges the content of all adjacent child nodes that
    fulfil the given `condition` into a single leaf node with parser
    `target_tag`. Nodes that do not fulfil the condition will be preserved.

    >>> sxpr = '(place (abbreviation "p.") (page "26") (superscript "b") (mark ",") (page "18"))'
    >>> tree = parse_sxpr(sxpr)
    >>> collapse_if([tree], not_one_of({'superscript', 'subscript'}), 'text')
    >>> print(flatten_sxpr(tree.as_sxpr()))
    (place (text "p.26") (superscript "b") (text ",18"))

    See `test_transform.TestComplexTransformations` for examples.
    """

    assert isinstance(target_tag, str)  # TODO: Delete this when safe

    node = context[-1]
    package = []  # type: List[Node]
    result = []  # type: List[Node]

    def close_package():
        nonlocal package
        if package:
            s = "".join(nd.content for nd in package)
            result.append(Node(target_tag, s))
            package = []

    for child in node.children:
        if condition([child]):
            if child.children:
                collapse_if([child], condition, target_tag)
                for c in child.children:
                    if condition([c]):
                        package.append(c)
                    else:
                        close_package()
                        result.append(c)
                close_package()
            else:
                package.append(child)
        else:
            close_package()
            result.append(child)
    close_package()
    node.result = tuple(result)


@transformation_factory(collections.abc.Callable)
def replace_content(context: List[Node], func: Callable):  # Callable[[Node], ResultType]
    """
    Replaces the content of the node. ``func`` takes the node's result
    as an argument an returns the mapped result.
    """
    node = context[-1]
    node.result = func(node.result)


@transformation_factory  # (str)
def replace_content_by(context: List[Node], content: str):  # Callable[[Node], ResultType]
    """
    Replaces the content of the node with the given text content.
    """
    node = context[-1]
    node.result = content


def normalize_whitespace(context):
    """
    Normalizes Whitespace inside a leaf node, i.e. any sequence of
    whitespaces, tabs and linefeeds will be replaced by a single
    whitespace. Empty (i.e. zero-length) Whitespace remains empty,
    however.
    """
    node = context[-1]
    assert not node.children
    if is_insignificant_whitespace(context):
        if node.result:
            node.result = ' '
    else:
        node.result = re.sub(r'\s+', ' ', node.result)


def merge_whitespace(context):
    """
    Merges adjacent whitespace. UNTESTED!
    """
    node = context[-1]
    children = node.children
    new_result = []
    i = 0
    L = len(children)
    while i < L:
        if children[i].tag_name == WHITESPACE_PTYPE:
            k = i
            while i < L and children[k].tag_name == WHITESPACE_PTYPE:
                i += 1
            if i > k:
                children[k].result = sum(children[n].result for n in range(k, i + 1))
            new_result.append(children[k])
        i += 1
    node.result = tuple(new_result)


@transformation_factory(collections.abc.Callable)
def move_adjacent(context: List[Node], condition: Callable = is_insignificant_whitespace):
    """
    Moves adjacent nodes that fulfill the given condition to the parent node.
    """
    def join_results(a: Node, b: Node, c: Node) -> bool:
        """Joins the results of node `a` and `b` and write them to the result
        of `c` type-safely, if possible. Return True, if join was possible
        and done, False otherwise."""
        if a.children and b.children:
            c.result = cast(Tuple[Node, ...], a.result) + cast(Tuple[Node, ...], b.result)
            return True
        elif not a.children and not b.children:
            c.result = cast(str, a.result) + cast(str, b.result)
            return True
        return False


    node = context[-1]
    if len(context) <= 1 or not node.children:
        return
    parent = context[-2]
    children = node.children
    if condition([children[0]]):
        before = (children[0],)   # type: Tuple[Node, ...]
        children = children[1:]
    else:
        before = ()
    if children and condition([children[-1]]):
        after = (children[-1],)   # type: Tuple[Node, ...]
        children = children[:-1]
    else:
        after = tuple()

    if before or after:
        node.result = children
        for i, child in enumerate(parent.children):
            if id(child) == id(node):
                break

        # merge adjacent whitespace
        prevN = parent.children[i - 1] if i > 0 else None
        nextN = parent.children[i + 1] if i < len(parent.children) - 1 else None
        if before and prevN and condition([prevN]):
            # prevN.result = prevN.result + before[0].result
            # before = ()
            if join_results(prevN, before[0], prevN):
                before = ()
        if after and nextN and condition([nextN]):
            # nextN.result = after[0].result + nextN.result
            # after = ()
            if join_results(after[0], nextN, nextN):
                after = ()

        parent.result = parent.children[:i] + before + (node,) + after + parent.children[i+1:]


def left_associative(context: List[Node]):
    """
    Rearranges a flat node with infix operators into a left associative tree.
    """
    node = context[-1]
    if len(node.children) >= 3:
        assert (len(node.children) + 1) % 2 == 0
        rest = list(node.result)
        left, rest = rest[0], rest[1:]
        while rest:
            infix, right, rest = rest[0], rest[1], rest[2:]
            assert not infix.children
            assert infix.tag_name[0:1] != ":"
            left = Node(infix.tag_name, (left, right))
        node.result = left


@transformation_factory(collections.abc.Set)
def lean_left(context: List[Node], operators: AbstractSet[str]):
    """
    Turns a right leaning tree into a left leaning tree:
    (op1 a (op2 b c))  ->  (op2 (op1 a b) c)
    If a left-associative operator is parsed with a right-recursive
    parser, `lean_left' can be used to rearrange the tree structure
    so that it properly reflects the order of association.

    ATTENTION: This transformation function moves forward recursively,
    so grouping nodes must not be eliminated during traversal! This
    must be done in a second pass.
    """
    node = context[-1]
    assert node.children and len(node.children) == 2
    assert node.tag_name in operators
    right = node.children[1]
    if right.tag_name in operators:
        assert right.children and len(right.children) == 2
        a, b, c = node.children[0], right.children[0], right.children[1]
        op1 = node.tag_name
        op2 = right.tag_name
        right.result = (a, b)
        right.tag_name = op1
        node.result = (right, c)
        node.tag_name = op2
        swap_attributes(node, right)
        # continue recursively on the left branch
        lean_left([right], operators)


# @transformation_factory(collections.abc.Set)
# def left_associative_tree(context: List[Node], operators: AbstracSet[str]):
#     """
#     Rearranges a right associative tree into a left associative tree.
#     ``operators`` is a list of tag names of nodes that shall be rearranged.
#     Other nodes will be lelft untouched.
#     """
#     node = context[-1]
#     assert node.tag_name in operators
#     right = node.children[1]
#     while right.tag_name in operators:
#         node.result = (node.children[0], right.children[0])
#         right.result = (node, right.children[1])
#         node = right
#         right = node.children[1]
#     parent = context[-2]
#     result = list(parent.result)
#     for i in range(len(result)):
#         if result[i] == contexnt[-1]:
#             result[i] = node
#             parent.result = tuple(result)
#             break
#     else:
#         assert False, "???"


#######################################################################
#
# destructive transformations:
#
# - leaves may be dropped (e.g. if deemed irrelevant)
# - errors of dropped leaves may be be lost
# - no promise that order will be preserved
#
#######################################################################


@transformation_factory(collections.abc.Callable)
def lstrip(context: List[Node], condition: Callable = contains_only_whitespace):
    """Recursively removes all leading child-nodes that fulfill a given condition."""
    node = context[-1]
    i = 1
    while i > 0 and node.children:
        lstrip(context + [node.children[0]], condition)
        i, L = 0, len(node.children)
        while i < L and condition(context + [node.children[i]]):
            i += 1
        if i > 0:
            node.result = node.children[i:]


@transformation_factory(collections.abc.Callable)
def rstrip(context: List[Node], condition: Callable = contains_only_whitespace):
    """Recursively removes all leading nodes that fulfill a given condition."""
    node = context[-1]
    i, L = 0, len(node.children)
    while i < L and node.children:
        rstrip(context + [node.children[-1]], condition)
        L = len(node.children)
        i = L
        while i > 0 and condition(context + [node.children[i - 1]]):
            i -= 1
        if i < L:
            node.result = node.children[:i]


@transformation_factory(collections.abc.Callable)
def strip(context: List[Node], condition: Callable = contains_only_whitespace):
    """Removes leading and trailing child-nodes that fulfill a given condition."""
    lstrip(context, condition)
    rstrip(context, condition)


@transformation_factory  # (slice)
def keep_children(context: List[Node], section: slice = slice(None)):
    """Keeps only child-nodes which fall into a slice of the result field."""
    node = context[-1]
    if node.children: