]> git.madduck.net Git - etc/vim.git/blob - src/black/linegen.py

madduck's git repository

Every one of the projects in this repository is available at the canonical URL git://git.madduck.net/madduck/pub/<projectpath> — see each project's metadata for the exact URL.

All patches and comments are welcome. Please squash your changes to logical commits before using git-format-patch and git-send-email to patches@git.madduck.net. If you'd read over the Git project's submission guidelines and adhered to them, I'd be especially grateful.

SSH access, as well as push access can be individually arranged.

If you use my repositories frequently, consider adding the following snippet to ~/.gitconfig and using the third clone URL listed for each project:

[url "git://git.madduck.net/madduck/"]
  insteadOf = madduck:

Bump sphinx from 5.0.1 to 5.0.2 in /docs (#3128)
[etc/vim.git] / src / black / linegen.py
1 """
2 Generating lines of code.
3 """
4 from functools import partial, wraps
5 import sys
6 from typing import Collection, Iterator, List, Optional, Set, Union, cast
7
8 from black.nodes import WHITESPACE, RARROW, STATEMENT, STANDALONE_COMMENT
9 from black.nodes import ASSIGNMENTS, OPENING_BRACKETS, CLOSING_BRACKETS
10 from black.nodes import Visitor, syms, is_arith_like, ensure_visible
11 from black.nodes import (
12     is_docstring,
13     is_empty_tuple,
14     is_one_tuple,
15     is_one_sequence_between,
16 )
17 from black.nodes import is_name_token, is_lpar_token, is_rpar_token
18 from black.nodes import is_walrus_assignment, is_yield, is_vararg, is_multiline_string
19 from black.nodes import is_stub_suite, is_stub_body, is_atom_with_invisible_parens
20 from black.nodes import wrap_in_parentheses
21 from black.brackets import max_delimiter_priority_in_atom
22 from black.brackets import DOT_PRIORITY, COMMA_PRIORITY
23 from black.lines import Line, line_to_string, is_line_short_enough
24 from black.lines import can_omit_invisible_parens, can_be_split, append_leaves
25 from black.comments import generate_comments, list_comments, FMT_OFF
26 from black.numerics import normalize_numeric_literal
27 from black.strings import get_string_prefix, fix_docstring
28 from black.strings import normalize_string_prefix, normalize_string_quotes
29 from black.trans import Transformer, CannotTransform, StringMerger, StringSplitter
30 from black.trans import StringParenWrapper, StringParenStripper, hug_power_op
31 from black.mode import Mode, Feature, Preview
32
33 from blib2to3.pytree import Node, Leaf
34 from blib2to3.pgen2 import token
35
36
37 # types
38 LeafID = int
39 LN = Union[Leaf, Node]
40
41
42 class CannotSplit(CannotTransform):
43     """A readable split that fits the allotted line length is impossible."""
44
45
46 # This isn't a dataclass because @dataclass + Generic breaks mypyc.
47 # See also https://github.com/mypyc/mypyc/issues/827.
48 class LineGenerator(Visitor[Line]):
49     """Generates reformatted Line objects.  Empty lines are not emitted.
50
51     Note: destroys the tree it's visiting by mutating prefixes of its leaves
52     in ways that will no longer stringify to valid Python code on the tree.
53     """
54
55     def __init__(self, mode: Mode) -> None:
56         self.mode = mode
57         self.current_line: Line
58         self.__post_init__()
59
60     def line(self, indent: int = 0) -> Iterator[Line]:
61         """Generate a line.
62
63         If the line is empty, only emit if it makes sense.
64         If the line is too long, split it first and then generate.
65
66         If any lines were generated, set up a new current_line.
67         """
68         if not self.current_line:
69             self.current_line.depth += indent
70             return  # Line is empty, don't emit. Creating a new one unnecessary.
71
72         complete_line = self.current_line
73         self.current_line = Line(mode=self.mode, depth=complete_line.depth + indent)
74         yield complete_line
75
76     def visit_default(self, node: LN) -> Iterator[Line]:
77         """Default `visit_*()` implementation. Recurses to children of `node`."""
78         if isinstance(node, Leaf):
79             any_open_brackets = self.current_line.bracket_tracker.any_open_brackets()
80             for comment in generate_comments(node, preview=self.mode.preview):
81                 if any_open_brackets:
82                     # any comment within brackets is subject to splitting
83                     self.current_line.append(comment)
84                 elif comment.type == token.COMMENT:
85                     # regular trailing comment
86                     self.current_line.append(comment)
87                     yield from self.line()
88
89                 else:
90                     # regular standalone comment
91                     yield from self.line()
92
93                     self.current_line.append(comment)
94                     yield from self.line()
95
96             normalize_prefix(node, inside_brackets=any_open_brackets)
97             if self.mode.string_normalization and node.type == token.STRING:
98                 node.value = normalize_string_prefix(node.value)
99                 node.value = normalize_string_quotes(node.value)
100             if node.type == token.NUMBER:
101                 normalize_numeric_literal(node)
102             if node.type not in WHITESPACE:
103                 self.current_line.append(node)
104         yield from super().visit_default(node)
105
106     def visit_INDENT(self, node: Leaf) -> Iterator[Line]:
107         """Increase indentation level, maybe yield a line."""
108         # In blib2to3 INDENT never holds comments.
109         yield from self.line(+1)
110         yield from self.visit_default(node)
111
112     def visit_DEDENT(self, node: Leaf) -> Iterator[Line]:
113         """Decrease indentation level, maybe yield a line."""
114         # The current line might still wait for trailing comments.  At DEDENT time
115         # there won't be any (they would be prefixes on the preceding NEWLINE).
116         # Emit the line then.
117         yield from self.line()
118
119         # While DEDENT has no value, its prefix may contain standalone comments
120         # that belong to the current indentation level.  Get 'em.
121         yield from self.visit_default(node)
122
123         # Finally, emit the dedent.
124         yield from self.line(-1)
125
126     def visit_stmt(
127         self, node: Node, keywords: Set[str], parens: Set[str]
128     ) -> Iterator[Line]:
129         """Visit a statement.
130
131         This implementation is shared for `if`, `while`, `for`, `try`, `except`,
132         `def`, `with`, `class`, `assert`, and assignments.
133
134         The relevant Python language `keywords` for a given statement will be
135         NAME leaves within it. This methods puts those on a separate line.
136
137         `parens` holds a set of string leaf values immediately after which
138         invisible parens should be put.
139         """
140         normalize_invisible_parens(node, parens_after=parens, preview=self.mode.preview)
141         for child in node.children:
142             if is_name_token(child) and child.value in keywords:
143                 yield from self.line()
144
145             yield from self.visit(child)
146
147     def visit_funcdef(self, node: Node) -> Iterator[Line]:
148         """Visit function definition."""
149         if Preview.annotation_parens not in self.mode:
150             yield from self.visit_stmt(node, keywords={"def"}, parens=set())
151         else:
152             yield from self.line()
153
154             # Remove redundant brackets around return type annotation.
155             is_return_annotation = False
156             for child in node.children:
157                 if child.type == token.RARROW:
158                     is_return_annotation = True
159                 elif is_return_annotation:
160                     if child.type == syms.atom and child.children[0].type == token.LPAR:
161                         if maybe_make_parens_invisible_in_atom(
162                             child,
163                             parent=node,
164                             remove_brackets_around_comma=False,
165                         ):
166                             wrap_in_parentheses(node, child, visible=False)
167                     else:
168                         wrap_in_parentheses(node, child, visible=False)
169                     is_return_annotation = False
170
171             for child in node.children:
172                 yield from self.visit(child)
173
174     def visit_match_case(self, node: Node) -> Iterator[Line]:
175         """Visit either a match or case statement."""
176         normalize_invisible_parens(node, parens_after=set(), preview=self.mode.preview)
177
178         yield from self.line()
179         for child in node.children:
180             yield from self.visit(child)
181
182     def visit_suite(self, node: Node) -> Iterator[Line]:
183         """Visit a suite."""
184         if self.mode.is_pyi and is_stub_suite(node):
185             yield from self.visit(node.children[2])
186         else:
187             yield from self.visit_default(node)
188
189     def visit_simple_stmt(self, node: Node) -> Iterator[Line]:
190         """Visit a statement without nested statements."""
191         prev_type: Optional[int] = None
192         for child in node.children:
193             if (prev_type is None or prev_type == token.SEMI) and is_arith_like(child):
194                 wrap_in_parentheses(node, child, visible=False)
195             prev_type = child.type
196
197         is_suite_like = node.parent and node.parent.type in STATEMENT
198         if is_suite_like:
199             if self.mode.is_pyi and is_stub_body(node):
200                 yield from self.visit_default(node)
201             else:
202                 yield from self.line(+1)
203                 yield from self.visit_default(node)
204                 yield from self.line(-1)
205
206         else:
207             if (
208                 not self.mode.is_pyi
209                 or not node.parent
210                 or not is_stub_suite(node.parent)
211             ):
212                 yield from self.line()
213             yield from self.visit_default(node)
214
215     def visit_async_stmt(self, node: Node) -> Iterator[Line]:
216         """Visit `async def`, `async for`, `async with`."""
217         yield from self.line()
218
219         children = iter(node.children)
220         for child in children:
221             yield from self.visit(child)
222
223             if child.type == token.ASYNC:
224                 break
225
226         internal_stmt = next(children)
227         for child in internal_stmt.children:
228             yield from self.visit(child)
229
230     def visit_decorators(self, node: Node) -> Iterator[Line]:
231         """Visit decorators."""
232         for child in node.children:
233             yield from self.line()
234             yield from self.visit(child)
235
236     def visit_power(self, node: Node) -> Iterator[Line]:
237         for idx, leaf in enumerate(node.children[:-1]):
238             next_leaf = node.children[idx + 1]
239
240             if not isinstance(leaf, Leaf):
241                 continue
242
243             value = leaf.value.lower()
244             if (
245                 leaf.type == token.NUMBER
246                 and next_leaf.type == syms.trailer
247                 # Ensure that we are in an attribute trailer
248                 and next_leaf.children[0].type == token.DOT
249                 # It shouldn't wrap hexadecimal, binary and octal literals
250                 and not value.startswith(("0x", "0b", "0o"))
251                 # It shouldn't wrap complex literals
252                 and "j" not in value
253             ):
254                 wrap_in_parentheses(node, leaf)
255
256         if Preview.remove_redundant_parens in self.mode:
257             remove_await_parens(node)
258
259         yield from self.visit_default(node)
260
261     def visit_SEMI(self, leaf: Leaf) -> Iterator[Line]:
262         """Remove a semicolon and put the other statement on a separate line."""
263         yield from self.line()
264
265     def visit_ENDMARKER(self, leaf: Leaf) -> Iterator[Line]:
266         """End of file. Process outstanding comments and end with a newline."""
267         yield from self.visit_default(leaf)
268         yield from self.line()
269
270     def visit_STANDALONE_COMMENT(self, leaf: Leaf) -> Iterator[Line]:
271         if not self.current_line.bracket_tracker.any_open_brackets():
272             yield from self.line()
273         yield from self.visit_default(leaf)
274
275     def visit_factor(self, node: Node) -> Iterator[Line]:
276         """Force parentheses between a unary op and a binary power:
277
278         -2 ** 8 -> -(2 ** 8)
279         """
280         _operator, operand = node.children
281         if (
282             operand.type == syms.power
283             and len(operand.children) == 3
284             and operand.children[1].type == token.DOUBLESTAR
285         ):
286             lpar = Leaf(token.LPAR, "(")
287             rpar = Leaf(token.RPAR, ")")
288             index = operand.remove() or 0
289             node.insert_child(index, Node(syms.atom, [lpar, operand, rpar]))
290         yield from self.visit_default(node)
291
292     def visit_STRING(self, leaf: Leaf) -> Iterator[Line]:
293         if is_docstring(leaf) and "\\\n" not in leaf.value:
294             # We're ignoring docstrings with backslash newline escapes because changing
295             # indentation of those changes the AST representation of the code.
296             docstring = normalize_string_prefix(leaf.value)
297             prefix = get_string_prefix(docstring)
298             docstring = docstring[len(prefix) :]  # Remove the prefix
299             quote_char = docstring[0]
300             # A natural way to remove the outer quotes is to do:
301             #   docstring = docstring.strip(quote_char)
302             # but that breaks on """""x""" (which is '""x').
303             # So we actually need to remove the first character and the next two
304             # characters but only if they are the same as the first.
305             quote_len = 1 if docstring[1] != quote_char else 3
306             docstring = docstring[quote_len:-quote_len]
307             docstring_started_empty = not docstring
308             indent = " " * 4 * self.current_line.depth
309
310             if is_multiline_string(leaf):
311                 docstring = fix_docstring(docstring, indent)
312             else:
313                 docstring = docstring.strip()
314
315             if docstring:
316                 # Add some padding if the docstring starts / ends with a quote mark.
317                 if docstring[0] == quote_char:
318                     docstring = " " + docstring
319                 if docstring[-1] == quote_char:
320                     docstring += " "
321                 if docstring[-1] == "\\":
322                     backslash_count = len(docstring) - len(docstring.rstrip("\\"))
323                     if backslash_count % 2:
324                         # Odd number of tailing backslashes, add some padding to
325                         # avoid escaping the closing string quote.
326                         docstring += " "
327             elif not docstring_started_empty:
328                 docstring = " "
329
330             # We could enforce triple quotes at this point.
331             quote = quote_char * quote_len
332
333             if Preview.long_docstring_quotes_on_newline in self.mode:
334                 # We need to find the length of the last line of the docstring
335                 # to find if we can add the closing quotes to the line without
336                 # exceeding the maximum line length.
337                 # If docstring is one line, then we need to add the length
338                 # of the indent, prefix, and starting quotes. Ending quote are
339                 # handled later
340                 lines = docstring.splitlines()
341                 last_line_length = len(lines[-1]) if docstring else 0
342
343                 if len(lines) == 1:
344                     last_line_length += len(indent) + len(prefix) + quote_len
345
346                 # If adding closing quotes would cause the last line to exceed
347                 # the maximum line length then put a line break before the
348                 # closing quotes
349                 if last_line_length + quote_len > self.mode.line_length:
350                     leaf.value = prefix + quote + docstring + "\n" + indent + quote
351                 else:
352                     leaf.value = prefix + quote + docstring + quote
353             else:
354                 leaf.value = prefix + quote + docstring + quote
355
356         yield from self.visit_default(leaf)
357
358     def __post_init__(self) -> None:
359         """You are in a twisty little maze of passages."""
360         self.current_line = Line(mode=self.mode)
361
362         v = self.visit_stmt
363         Ø: Set[str] = set()
364         self.visit_assert_stmt = partial(v, keywords={"assert"}, parens={"assert", ","})
365         self.visit_if_stmt = partial(
366             v, keywords={"if", "else", "elif"}, parens={"if", "elif"}
367         )
368         self.visit_while_stmt = partial(v, keywords={"while", "else"}, parens={"while"})
369         self.visit_for_stmt = partial(v, keywords={"for", "else"}, parens={"for", "in"})
370         self.visit_try_stmt = partial(
371             v, keywords={"try", "except", "else", "finally"}, parens=Ø
372         )
373         if self.mode.preview:
374             self.visit_except_clause = partial(
375                 v, keywords={"except"}, parens={"except"}
376             )
377             self.visit_with_stmt = partial(v, keywords={"with"}, parens={"with"})
378         else:
379             self.visit_except_clause = partial(v, keywords={"except"}, parens=Ø)
380             self.visit_with_stmt = partial(v, keywords={"with"}, parens=Ø)
381         self.visit_classdef = partial(v, keywords={"class"}, parens=Ø)
382         self.visit_expr_stmt = partial(v, keywords=Ø, parens=ASSIGNMENTS)
383         self.visit_return_stmt = partial(v, keywords={"return"}, parens={"return"})
384         self.visit_import_from = partial(v, keywords=Ø, parens={"import"})
385         self.visit_del_stmt = partial(v, keywords=Ø, parens={"del"})
386         self.visit_async_funcdef = self.visit_async_stmt
387         self.visit_decorated = self.visit_decorators
388
389         # PEP 634
390         self.visit_match_stmt = self.visit_match_case
391         self.visit_case_block = self.visit_match_case
392
393
394 def transform_line(
395     line: Line, mode: Mode, features: Collection[Feature] = ()
396 ) -> Iterator[Line]:
397     """Transform a `line`, potentially splitting it into many lines.
398
399     They should fit in the allotted `line_length` but might not be able to.
400
401     `features` are syntactical features that may be used in the output.
402     """
403     if line.is_comment:
404         yield line
405         return
406
407     line_str = line_to_string(line)
408
409     ll = mode.line_length
410     sn = mode.string_normalization
411     string_merge = StringMerger(ll, sn)
412     string_paren_strip = StringParenStripper(ll, sn)
413     string_split = StringSplitter(ll, sn)
414     string_paren_wrap = StringParenWrapper(ll, sn)
415
416     transformers: List[Transformer]
417     if (
418         not line.contains_uncollapsable_type_comments()
419         and not line.should_split_rhs
420         and not line.magic_trailing_comma
421         and (
422             is_line_short_enough(line, line_length=mode.line_length, line_str=line_str)
423             or line.contains_unsplittable_type_ignore()
424         )
425         and not (line.inside_brackets and line.contains_standalone_comments())
426     ):
427         # Only apply basic string preprocessing, since lines shouldn't be split here.
428         if Preview.string_processing in mode:
429             transformers = [string_merge, string_paren_strip]
430         else:
431             transformers = []
432     elif line.is_def:
433         transformers = [left_hand_split]
434     else:
435
436         def _rhs(
437             self: object, line: Line, features: Collection[Feature]
438         ) -> Iterator[Line]:
439             """Wraps calls to `right_hand_split`.
440
441             The calls increasingly `omit` right-hand trailers (bracket pairs with
442             content), meaning the trailers get glued together to split on another
443             bracket pair instead.
444             """
445             for omit in generate_trailers_to_omit(line, mode.line_length):
446                 lines = list(
447                     right_hand_split(line, mode.line_length, features, omit=omit)
448                 )
449                 # Note: this check is only able to figure out if the first line of the
450                 # *current* transformation fits in the line length.  This is true only
451                 # for simple cases.  All others require running more transforms via
452                 # `transform_line()`.  This check doesn't know if those would succeed.
453                 if is_line_short_enough(lines[0], line_length=mode.line_length):
454                     yield from lines
455                     return
456
457             # All splits failed, best effort split with no omits.
458             # This mostly happens to multiline strings that are by definition
459             # reported as not fitting a single line, as well as lines that contain
460             # trailing commas (those have to be exploded).
461             yield from right_hand_split(
462                 line, line_length=mode.line_length, features=features
463             )
464
465         # HACK: nested functions (like _rhs) compiled by mypyc don't retain their
466         # __name__ attribute which is needed in `run_transformer` further down.
467         # Unfortunately a nested class breaks mypyc too. So a class must be created
468         # via type ... https://github.com/mypyc/mypyc/issues/884
469         rhs = type("rhs", (), {"__call__": _rhs})()
470
471         if Preview.string_processing in mode:
472             if line.inside_brackets:
473                 transformers = [
474                     string_merge,
475                     string_paren_strip,
476                     string_split,
477                     delimiter_split,
478                     standalone_comment_split,
479                     string_paren_wrap,
480                     rhs,
481                 ]
482             else:
483                 transformers = [
484                     string_merge,
485                     string_paren_strip,
486                     string_split,
487                     string_paren_wrap,
488                     rhs,
489                 ]
490         else:
491             if line.inside_brackets:
492                 transformers = [delimiter_split, standalone_comment_split, rhs]
493             else:
494                 transformers = [rhs]
495     # It's always safe to attempt hugging of power operations and pretty much every line
496     # could match.
497     transformers.append(hug_power_op)
498
499     for transform in transformers:
500         # We are accumulating lines in `result` because we might want to abort
501         # mission and return the original line in the end, or attempt a different
502         # split altogether.
503         try:
504             result = run_transformer(line, transform, mode, features, line_str=line_str)
505         except CannotTransform:
506             continue
507         else:
508             yield from result
509             break
510
511     else:
512         yield line
513
514
515 def left_hand_split(line: Line, _features: Collection[Feature] = ()) -> Iterator[Line]:
516     """Split line into many lines, starting with the first matching bracket pair.
517
518     Note: this usually looks weird, only use this for function definitions.
519     Prefer RHS otherwise.  This is why this function is not symmetrical with
520     :func:`right_hand_split` which also handles optional parentheses.
521     """
522     tail_leaves: List[Leaf] = []
523     body_leaves: List[Leaf] = []
524     head_leaves: List[Leaf] = []
525     current_leaves = head_leaves
526     matching_bracket: Optional[Leaf] = None
527     for leaf in line.leaves:
528         if (
529             current_leaves is body_leaves
530             and leaf.type in CLOSING_BRACKETS
531             and leaf.opening_bracket is matching_bracket
532             and isinstance(matching_bracket, Leaf)
533         ):
534             ensure_visible(leaf)
535             ensure_visible(matching_bracket)
536             current_leaves = tail_leaves if body_leaves else head_leaves
537         current_leaves.append(leaf)
538         if current_leaves is head_leaves:
539             if leaf.type in OPENING_BRACKETS:
540                 matching_bracket = leaf
541                 current_leaves = body_leaves
542     if not matching_bracket:
543         raise CannotSplit("No brackets found")
544
545     head = bracket_split_build_line(head_leaves, line, matching_bracket)
546     body = bracket_split_build_line(body_leaves, line, matching_bracket, is_body=True)
547     tail = bracket_split_build_line(tail_leaves, line, matching_bracket)
548     bracket_split_succeeded_or_raise(head, body, tail)
549     for result in (head, body, tail):
550         if result:
551             yield result
552
553
554 def right_hand_split(
555     line: Line,
556     line_length: int,
557     features: Collection[Feature] = (),
558     omit: Collection[LeafID] = (),
559 ) -> Iterator[Line]:
560     """Split line into many lines, starting with the last matching bracket pair.
561
562     If the split was by optional parentheses, attempt splitting without them, too.
563     `omit` is a collection of closing bracket IDs that shouldn't be considered for
564     this split.
565
566     Note: running this function modifies `bracket_depth` on the leaves of `line`.
567     """
568     tail_leaves: List[Leaf] = []
569     body_leaves: List[Leaf] = []
570     head_leaves: List[Leaf] = []
571     current_leaves = tail_leaves
572     opening_bracket: Optional[Leaf] = None
573     closing_bracket: Optional[Leaf] = None
574     for leaf in reversed(line.leaves):
575         if current_leaves is body_leaves:
576             if leaf is opening_bracket:
577                 current_leaves = head_leaves if body_leaves else tail_leaves
578         current_leaves.append(leaf)
579         if current_leaves is tail_leaves:
580             if leaf.type in CLOSING_BRACKETS and id(leaf) not in omit:
581                 opening_bracket = leaf.opening_bracket
582                 closing_bracket = leaf
583                 current_leaves = body_leaves
584     if not (opening_bracket and closing_bracket and head_leaves):
585         # If there is no opening or closing_bracket that means the split failed and
586         # all content is in the tail.  Otherwise, if `head_leaves` are empty, it means
587         # the matching `opening_bracket` wasn't available on `line` anymore.
588         raise CannotSplit("No brackets found")
589
590     tail_leaves.reverse()
591     body_leaves.reverse()
592     head_leaves.reverse()
593     head = bracket_split_build_line(head_leaves, line, opening_bracket)
594     body = bracket_split_build_line(body_leaves, line, opening_bracket, is_body=True)
595     tail = bracket_split_build_line(tail_leaves, line, opening_bracket)
596     bracket_split_succeeded_or_raise(head, body, tail)
597     if (
598         Feature.FORCE_OPTIONAL_PARENTHESES not in features
599         # the opening bracket is an optional paren
600         and opening_bracket.type == token.LPAR
601         and not opening_bracket.value
602         # the closing bracket is an optional paren
603         and closing_bracket.type == token.RPAR
604         and not closing_bracket.value
605         # it's not an import (optional parens are the only thing we can split on
606         # in this case; attempting a split without them is a waste of time)
607         and not line.is_import
608         # there are no standalone comments in the body
609         and not body.contains_standalone_comments(0)
610         # and we can actually remove the parens
611         and can_omit_invisible_parens(body, line_length)
612     ):
613         omit = {id(closing_bracket), *omit}
614         try:
615             yield from right_hand_split(line, line_length, features=features, omit=omit)
616             return
617
618         except CannotSplit as e:
619             if not (
620                 can_be_split(body)
621                 or is_line_short_enough(body, line_length=line_length)
622             ):
623                 raise CannotSplit(
624                     "Splitting failed, body is still too long and can't be split."
625                 ) from e
626
627             elif head.contains_multiline_strings() or tail.contains_multiline_strings():
628                 raise CannotSplit(
629                     "The current optional pair of parentheses is bound to fail to"
630                     " satisfy the splitting algorithm because the head or the tail"
631                     " contains multiline strings which by definition never fit one"
632                     " line."
633                 ) from e
634
635     ensure_visible(opening_bracket)
636     ensure_visible(closing_bracket)
637     for result in (head, body, tail):
638         if result:
639             yield result
640
641
642 def bracket_split_succeeded_or_raise(head: Line, body: Line, tail: Line) -> None:
643     """Raise :exc:`CannotSplit` if the last left- or right-hand split failed.
644
645     Do nothing otherwise.
646
647     A left- or right-hand split is based on a pair of brackets. Content before
648     (and including) the opening bracket is left on one line, content inside the
649     brackets is put on a separate line, and finally content starting with and
650     following the closing bracket is put on a separate line.
651
652     Those are called `head`, `body`, and `tail`, respectively. If the split
653     produced the same line (all content in `head`) or ended up with an empty `body`
654     and the `tail` is just the closing bracket, then it's considered failed.
655     """
656     tail_len = len(str(tail).strip())
657     if not body:
658         if tail_len == 0:
659             raise CannotSplit("Splitting brackets produced the same line")
660
661         elif tail_len < 3:
662             raise CannotSplit(
663                 f"Splitting brackets on an empty body to save {tail_len} characters is"
664                 " not worth it"
665             )
666
667
668 def bracket_split_build_line(
669     leaves: List[Leaf], original: Line, opening_bracket: Leaf, *, is_body: bool = False
670 ) -> Line:
671     """Return a new line with given `leaves` and respective comments from `original`.
672
673     If `is_body` is True, the result line is one-indented inside brackets and as such
674     has its first leaf's prefix normalized and a trailing comma added when expected.
675     """
676     result = Line(mode=original.mode, depth=original.depth)
677     if is_body:
678         result.inside_brackets = True
679         result.depth += 1
680         if leaves:
681             # Since body is a new indent level, remove spurious leading whitespace.
682             normalize_prefix(leaves[0], inside_brackets=True)
683             # Ensure a trailing comma for imports and standalone function arguments, but
684             # be careful not to add one after any comments or within type annotations.
685             no_commas = (
686                 original.is_def
687                 and opening_bracket.value == "("
688                 and not any(leaf.type == token.COMMA for leaf in leaves)
689                 # In particular, don't add one within a parenthesized return annotation.
690                 # Unfortunately the indicator we're in a return annotation (RARROW) may
691                 # be defined directly in the parent node, the parent of the parent ...
692                 # and so on depending on how complex the return annotation is.
693                 # This isn't perfect and there's some false negatives but they are in
694                 # contexts were a comma is actually fine.
695                 and not any(
696                     node.prev_sibling.type == RARROW
697                     for node in (
698                         leaves[0].parent,
699                         getattr(leaves[0].parent, "parent", None),
700                     )
701                     if isinstance(node, Node) and isinstance(node.prev_sibling, Leaf)
702                 )
703             )
704
705             if original.is_import or no_commas:
706                 for i in range(len(leaves) - 1, -1, -1):
707                     if leaves[i].type == STANDALONE_COMMENT:
708                         continue
709
710                     if leaves[i].type != token.COMMA:
711                         new_comma = Leaf(token.COMMA, ",")
712                         leaves.insert(i + 1, new_comma)
713                     break
714
715     # Populate the line
716     for leaf in leaves:
717         result.append(leaf, preformatted=True)
718         for comment_after in original.comments_after(leaf):
719             result.append(comment_after, preformatted=True)
720     if is_body and should_split_line(result, opening_bracket):
721         result.should_split_rhs = True
722     return result
723
724
725 def dont_increase_indentation(split_func: Transformer) -> Transformer:
726     """Normalize prefix of the first leaf in every line returned by `split_func`.
727
728     This is a decorator over relevant split functions.
729     """
730
731     @wraps(split_func)
732     def split_wrapper(line: Line, features: Collection[Feature] = ()) -> Iterator[Line]:
733         for split_line in split_func(line, features):
734             normalize_prefix(split_line.leaves[0], inside_brackets=True)
735             yield split_line
736
737     return split_wrapper
738
739
740 @dont_increase_indentation
741 def delimiter_split(line: Line, features: Collection[Feature] = ()) -> Iterator[Line]:
742     """Split according to delimiters of the highest priority.
743
744     If the appropriate Features are given, the split will add trailing commas
745     also in function signatures and calls that contain `*` and `**`.
746     """
747     try:
748         last_leaf = line.leaves[-1]
749     except IndexError:
750         raise CannotSplit("Line empty") from None
751
752     bt = line.bracket_tracker
753     try:
754         delimiter_priority = bt.max_delimiter_priority(exclude={id(last_leaf)})
755     except ValueError:
756         raise CannotSplit("No delimiters found") from None
757
758     if delimiter_priority == DOT_PRIORITY:
759         if bt.delimiter_count_with_priority(delimiter_priority) == 1:
760             raise CannotSplit("Splitting a single attribute from its owner looks wrong")
761
762     current_line = Line(
763         mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
764     )
765     lowest_depth = sys.maxsize
766     trailing_comma_safe = True
767
768     def append_to_line(leaf: Leaf) -> Iterator[Line]:
769         """Append `leaf` to current line or to new line if appending impossible."""
770         nonlocal current_line
771         try:
772             current_line.append_safe(leaf, preformatted=True)
773         except ValueError:
774             yield current_line
775
776             current_line = Line(
777                 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
778             )
779             current_line.append(leaf)
780
781     for leaf in line.leaves:
782         yield from append_to_line(leaf)
783
784         for comment_after in line.comments_after(leaf):
785             yield from append_to_line(comment_after)
786
787         lowest_depth = min(lowest_depth, leaf.bracket_depth)
788         if leaf.bracket_depth == lowest_depth:
789             if is_vararg(leaf, within={syms.typedargslist}):
790                 trailing_comma_safe = (
791                     trailing_comma_safe and Feature.TRAILING_COMMA_IN_DEF in features
792                 )
793             elif is_vararg(leaf, within={syms.arglist, syms.argument}):
794                 trailing_comma_safe = (
795                     trailing_comma_safe and Feature.TRAILING_COMMA_IN_CALL in features
796                 )
797
798         leaf_priority = bt.delimiters.get(id(leaf))
799         if leaf_priority == delimiter_priority:
800             yield current_line
801
802             current_line = Line(
803                 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
804             )
805     if current_line:
806         if (
807             trailing_comma_safe
808             and delimiter_priority == COMMA_PRIORITY
809             and current_line.leaves[-1].type != token.COMMA
810             and current_line.leaves[-1].type != STANDALONE_COMMENT
811         ):
812             new_comma = Leaf(token.COMMA, ",")
813             current_line.append(new_comma)
814         yield current_line
815
816
817 @dont_increase_indentation
818 def standalone_comment_split(
819     line: Line, features: Collection[Feature] = ()
820 ) -> Iterator[Line]:
821     """Split standalone comments from the rest of the line."""
822     if not line.contains_standalone_comments(0):
823         raise CannotSplit("Line does not have any standalone comments")
824
825     current_line = Line(
826         mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
827     )
828
829     def append_to_line(leaf: Leaf) -> Iterator[Line]:
830         """Append `leaf` to current line or to new line if appending impossible."""
831         nonlocal current_line
832         try:
833             current_line.append_safe(leaf, preformatted=True)
834         except ValueError:
835             yield current_line
836
837             current_line = Line(
838                 line.mode, depth=line.depth, inside_brackets=line.inside_brackets
839             )
840             current_line.append(leaf)
841
842     for leaf in line.leaves:
843         yield from append_to_line(leaf)
844
845         for comment_after in line.comments_after(leaf):
846             yield from append_to_line(comment_after)
847
848     if current_line:
849         yield current_line
850
851
852 def normalize_prefix(leaf: Leaf, *, inside_brackets: bool) -> None:
853     """Leave existing extra newlines if not `inside_brackets`. Remove everything
854     else.
855
856     Note: don't use backslashes for formatting or you'll lose your voting rights.
857     """
858     if not inside_brackets:
859         spl = leaf.prefix.split("#")
860         if "\\" not in spl[0]:
861             nl_count = spl[-1].count("\n")
862             if len(spl) > 1:
863                 nl_count -= 1
864             leaf.prefix = "\n" * nl_count
865             return
866
867     leaf.prefix = ""
868
869
870 def normalize_invisible_parens(
871     node: Node, parens_after: Set[str], *, preview: bool
872 ) -> None:
873     """Make existing optional parentheses invisible or create new ones.
874
875     `parens_after` is a set of string leaf values immediately after which parens
876     should be put.
877
878     Standardizes on visible parentheses for single-element tuples, and keeps
879     existing visible parentheses for other tuples and generator expressions.
880     """
881     for pc in list_comments(node.prefix, is_endmarker=False, preview=preview):
882         if pc.value in FMT_OFF:
883             # This `node` has a prefix with `# fmt: off`, don't mess with parens.
884             return
885     check_lpar = False
886     for index, child in enumerate(list(node.children)):
887         # Fixes a bug where invisible parens are not properly stripped from
888         # assignment statements that contain type annotations.
889         if isinstance(child, Node) and child.type == syms.annassign:
890             normalize_invisible_parens(
891                 child, parens_after=parens_after, preview=preview
892             )
893
894         # Add parentheses around long tuple unpacking in assignments.
895         if (
896             index == 0
897             and isinstance(child, Node)
898             and child.type == syms.testlist_star_expr
899         ):
900             check_lpar = True
901
902         if check_lpar:
903             if (
904                 preview
905                 and child.type == syms.atom
906                 and node.type == syms.for_stmt
907                 and isinstance(child.prev_sibling, Leaf)
908                 and child.prev_sibling.type == token.NAME
909                 and child.prev_sibling.value == "for"
910             ):
911                 if maybe_make_parens_invisible_in_atom(
912                     child,
913                     parent=node,
914                     remove_brackets_around_comma=True,
915                 ):
916                     wrap_in_parentheses(node, child, visible=False)
917             elif preview and isinstance(child, Node) and node.type == syms.with_stmt:
918                 remove_with_parens(child, node)
919             elif child.type == syms.atom:
920                 if maybe_make_parens_invisible_in_atom(
921                     child,
922                     parent=node,
923                 ):
924                     wrap_in_parentheses(node, child, visible=False)
925             elif is_one_tuple(child):
926                 wrap_in_parentheses(node, child, visible=True)
927             elif node.type == syms.import_from:
928                 # "import from" nodes store parentheses directly as part of
929                 # the statement
930                 if is_lpar_token(child):
931                     assert is_rpar_token(node.children[-1])
932                     # make parentheses invisible
933                     child.value = ""
934                     node.children[-1].value = ""
935                 elif child.type != token.STAR:
936                     # insert invisible parentheses
937                     node.insert_child(index, Leaf(token.LPAR, ""))
938                     node.append_child(Leaf(token.RPAR, ""))
939                 break
940             elif (
941                 index == 1
942                 and child.type == token.STAR
943                 and node.type == syms.except_clause
944             ):
945                 # In except* (PEP 654), the star is actually part of
946                 # of the keyword. So we need to skip the insertion of
947                 # invisible parentheses to work more precisely.
948                 continue
949
950             elif not (isinstance(child, Leaf) and is_multiline_string(child)):
951                 wrap_in_parentheses(node, child, visible=False)
952
953         comma_check = child.type == token.COMMA if preview else False
954
955         check_lpar = isinstance(child, Leaf) and (
956             child.value in parens_after or comma_check
957         )
958
959
960 def remove_await_parens(node: Node) -> None:
961     if node.children[0].type == token.AWAIT and len(node.children) > 1:
962         if (
963             node.children[1].type == syms.atom
964             and node.children[1].children[0].type == token.LPAR
965         ):
966             if maybe_make_parens_invisible_in_atom(
967                 node.children[1],
968                 parent=node,
969                 remove_brackets_around_comma=True,
970             ):
971                 wrap_in_parentheses(node, node.children[1], visible=False)
972
973             # Since await is an expression we shouldn't remove
974             # brackets in cases where this would change
975             # the AST due to operator precedence.
976             # Therefore we only aim to remove brackets around
977             # power nodes that aren't also await expressions themselves.
978             # https://peps.python.org/pep-0492/#updated-operator-precedence-table
979             # N.B. We've still removed any redundant nested brackets though :)
980             opening_bracket = cast(Leaf, node.children[1].children[0])
981             closing_bracket = cast(Leaf, node.children[1].children[-1])
982             bracket_contents = cast(Node, node.children[1].children[1])
983             if bracket_contents.type != syms.power:
984                 ensure_visible(opening_bracket)
985                 ensure_visible(closing_bracket)
986             elif (
987                 bracket_contents.type == syms.power
988                 and bracket_contents.children[0].type == token.AWAIT
989             ):
990                 ensure_visible(opening_bracket)
991                 ensure_visible(closing_bracket)
992                 # If we are in a nested await then recurse down.
993                 remove_await_parens(bracket_contents)
994
995
996 def remove_with_parens(node: Node, parent: Node) -> None:
997     """Recursively hide optional parens in `with` statements."""
998     # Removing all unnecessary parentheses in with statements in one pass is a tad
999     # complex as different variations of bracketed statements result in pretty
1000     # different parse trees:
1001     #
1002     # with (open("file")) as f:                       # this is an asexpr_test
1003     #     ...
1004     #
1005     # with (open("file") as f):                       # this is an atom containing an
1006     #     ...                                         # asexpr_test
1007     #
1008     # with (open("file")) as f, (open("file")) as f:  # this is asexpr_test, COMMA,
1009     #     ...                                         # asexpr_test
1010     #
1011     # with (open("file") as f, open("file") as f):    # an atom containing a
1012     #     ...                                         # testlist_gexp which then
1013     #                                                 # contains multiple asexpr_test(s)
1014     if node.type == syms.atom:
1015         if maybe_make_parens_invisible_in_atom(
1016             node,
1017             parent=parent,
1018             remove_brackets_around_comma=True,
1019         ):
1020             wrap_in_parentheses(parent, node, visible=False)
1021         if isinstance(node.children[1], Node):
1022             remove_with_parens(node.children[1], node)
1023     elif node.type == syms.testlist_gexp:
1024         for child in node.children:
1025             if isinstance(child, Node):
1026                 remove_with_parens(child, node)
1027     elif node.type == syms.asexpr_test and not any(
1028         leaf.type == token.COLONEQUAL for leaf in node.leaves()
1029     ):
1030         if maybe_make_parens_invisible_in_atom(
1031             node.children[0],
1032             parent=node,
1033             remove_brackets_around_comma=True,
1034         ):
1035             wrap_in_parentheses(node, node.children[0], visible=False)
1036
1037
1038 def maybe_make_parens_invisible_in_atom(
1039     node: LN,
1040     parent: LN,
1041     remove_brackets_around_comma: bool = False,
1042 ) -> bool:
1043     """If it's safe, make the parens in the atom `node` invisible, recursively.
1044     Additionally, remove repeated, adjacent invisible parens from the atom `node`
1045     as they are redundant.
1046
1047     Returns whether the node should itself be wrapped in invisible parentheses.
1048     """
1049     if (
1050         node.type != syms.atom
1051         or is_empty_tuple(node)
1052         or is_one_tuple(node)
1053         or (is_yield(node) and parent.type != syms.expr_stmt)
1054         or (
1055             # This condition tries to prevent removing non-optional brackets
1056             # around a tuple, however, can be a bit overzealous so we provide
1057             # and option to skip this check for `for` and `with` statements.
1058             not remove_brackets_around_comma
1059             and max_delimiter_priority_in_atom(node) >= COMMA_PRIORITY
1060         )
1061     ):
1062         return False
1063
1064     if is_walrus_assignment(node):
1065         if parent.type in [
1066             syms.annassign,
1067             syms.expr_stmt,
1068             syms.assert_stmt,
1069             syms.return_stmt,
1070             # these ones aren't useful to end users, but they do please fuzzers
1071             syms.for_stmt,
1072             syms.del_stmt,
1073         ]:
1074             return False
1075
1076     first = node.children[0]
1077     last = node.children[-1]
1078     if is_lpar_token(first) and is_rpar_token(last):
1079         middle = node.children[1]
1080         # make parentheses invisible
1081         first.value = ""
1082         last.value = ""
1083         maybe_make_parens_invisible_in_atom(
1084             middle,
1085             parent=parent,
1086             remove_brackets_around_comma=remove_brackets_around_comma,
1087         )
1088
1089         if is_atom_with_invisible_parens(middle):
1090             # Strip the invisible parens from `middle` by replacing
1091             # it with the child in-between the invisible parens
1092             middle.replace(middle.children[1])
1093
1094         return False
1095
1096     return True
1097
1098
1099 def should_split_line(line: Line, opening_bracket: Leaf) -> bool:
1100     """Should `line` be immediately split with `delimiter_split()` after RHS?"""
1101
1102     if not (opening_bracket.parent and opening_bracket.value in "[{("):
1103         return False
1104
1105     # We're essentially checking if the body is delimited by commas and there's more
1106     # than one of them (we're excluding the trailing comma and if the delimiter priority
1107     # is still commas, that means there's more).
1108     exclude = set()
1109     trailing_comma = False
1110     try:
1111         last_leaf = line.leaves[-1]
1112         if last_leaf.type == token.COMMA:
1113             trailing_comma = True
1114             exclude.add(id(last_leaf))
1115         max_priority = line.bracket_tracker.max_delimiter_priority(exclude=exclude)
1116     except (IndexError, ValueError):
1117         return False
1118
1119     return max_priority == COMMA_PRIORITY and (
1120         (line.mode.magic_trailing_comma and trailing_comma)
1121         # always explode imports
1122         or opening_bracket.parent.type in {syms.atom, syms.import_from}
1123     )
1124
1125
1126 def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[LeafID]]:
1127     """Generate sets of closing bracket IDs that should be omitted in a RHS.
1128
1129     Brackets can be omitted if the entire trailer up to and including
1130     a preceding closing bracket fits in one line.
1131
1132     Yielded sets are cumulative (contain results of previous yields, too).  First
1133     set is empty, unless the line should explode, in which case bracket pairs until
1134     the one that needs to explode are omitted.
1135     """
1136
1137     omit: Set[LeafID] = set()
1138     if not line.magic_trailing_comma:
1139         yield omit
1140
1141     length = 4 * line.depth
1142     opening_bracket: Optional[Leaf] = None
1143     closing_bracket: Optional[Leaf] = None
1144     inner_brackets: Set[LeafID] = set()
1145     for index, leaf, leaf_length in line.enumerate_with_length(reversed=True):
1146         length += leaf_length
1147         if length > line_length:
1148             break
1149
1150         has_inline_comment = leaf_length > len(leaf.value) + len(leaf.prefix)
1151         if leaf.type == STANDALONE_COMMENT or has_inline_comment:
1152             break
1153
1154         if opening_bracket:
1155             if leaf is opening_bracket:
1156                 opening_bracket = None
1157             elif leaf.type in CLOSING_BRACKETS:
1158                 prev = line.leaves[index - 1] if index > 0 else None
1159                 if (
1160                     prev
1161                     and prev.type == token.COMMA
1162                     and leaf.opening_bracket is not None
1163                     and not is_one_sequence_between(
1164                         leaf.opening_bracket, leaf, line.leaves
1165                     )
1166                 ):
1167                     # Never omit bracket pairs with trailing commas.
1168                     # We need to explode on those.
1169                     break
1170
1171                 inner_brackets.add(id(leaf))
1172         elif leaf.type in CLOSING_BRACKETS:
1173             prev = line.leaves[index - 1] if index > 0 else None
1174             if prev and prev.type in OPENING_BRACKETS:
1175                 # Empty brackets would fail a split so treat them as "inner"
1176                 # brackets (e.g. only add them to the `omit` set if another
1177                 # pair of brackets was good enough.
1178                 inner_brackets.add(id(leaf))
1179                 continue
1180
1181             if closing_bracket:
1182                 omit.add(id(closing_bracket))
1183                 omit.update(inner_brackets)
1184                 inner_brackets.clear()
1185                 yield omit
1186
1187             if (
1188                 prev
1189                 and prev.type == token.COMMA
1190                 and leaf.opening_bracket is not None
1191                 and not is_one_sequence_between(leaf.opening_bracket, leaf, line.leaves)
1192             ):
1193                 # Never omit bracket pairs with trailing commas.
1194                 # We need to explode on those.
1195                 break
1196
1197             if leaf.value:
1198                 opening_bracket = leaf.opening_bracket
1199                 closing_bracket = leaf
1200
1201
1202 def run_transformer(
1203     line: Line,
1204     transform: Transformer,
1205     mode: Mode,
1206     features: Collection[Feature],
1207     *,
1208     line_str: str = "",
1209 ) -> List[Line]:
1210     if not line_str:
1211         line_str = line_to_string(line)
1212     result: List[Line] = []
1213     for transformed_line in transform(line, features):
1214         if str(transformed_line).strip("\n") == line_str:
1215             raise CannotTransform("Line transformer returned an unchanged result")
1216
1217         result.extend(transform_line(transformed_line, mode=mode, features=features))
1218
1219     if (
1220         transform.__class__.__name__ != "rhs"
1221         or not line.bracket_tracker.invisible
1222         or any(bracket.value for bracket in line.bracket_tracker.invisible)
1223         or line.contains_multiline_strings()
1224         or result[0].contains_uncollapsable_type_comments()
1225         or result[0].contains_unsplittable_type_ignore()
1226         or is_line_short_enough(result[0], line_length=mode.line_length)
1227         # If any leaves have no parents (which _can_ occur since
1228         # `transform(line)` potentially destroys the line's underlying node
1229         # structure), then we can't proceed. Doing so would cause the below
1230         # call to `append_leaves()` to fail.
1231         or any(leaf.parent is None for leaf in line.leaves)
1232     ):
1233         return result
1234
1235     line_copy = line.clone()
1236     append_leaves(line_copy, line, line.leaves)
1237     features_fop = set(features) | {Feature.FORCE_OPTIONAL_PARENTHESES}
1238     second_opinion = run_transformer(
1239         line_copy, transform, mode, features_fop, line_str=line_str
1240     )
1241     if all(
1242         is_line_short_enough(ln, line_length=mode.line_length) for ln in second_opinion
1243     ):
1244         result = second_opinion
1245     return result