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.
2 Generating lines of code.
5 from dataclasses import dataclass
6 from enum import Enum, auto
7 from functools import partial, wraps
8 from typing import Collection, Iterator, List, Optional, Set, Union, cast
10 from black.brackets import (
13 get_leaves_inside_matching_brackets,
14 max_delimiter_priority_in_atom,
16 from black.comments import FMT_OFF, generate_comments, list_comments
17 from black.lines import (
21 can_omit_invisible_parens,
25 from black.mode import Feature, Mode, Preview
26 from black.nodes import (
38 is_atom_with_invisible_parens,
44 is_one_sequence_between,
49 is_tuple_containing_walrus,
56 from black.numerics import normalize_numeric_literal
57 from black.strings import (
60 normalize_string_prefix,
61 normalize_string_quotes,
62 normalize_unicode_escape_sequences,
64 from black.trans import (
73 from blib2to3.pgen2 import token
74 from blib2to3.pytree import Leaf, Node
78 LN = Union[Leaf, Node]
81 class CannotSplit(CannotTransform):
82 """A readable split that fits the allotted line length is impossible."""
85 # This isn't a dataclass because @dataclass + Generic breaks mypyc.
86 # See also https://github.com/mypyc/mypyc/issues/827.
87 class LineGenerator(Visitor[Line]):
88 """Generates reformatted Line objects. Empty lines are not emitted.
90 Note: destroys the tree it's visiting by mutating prefixes of its leaves
91 in ways that will no longer stringify to valid Python code on the tree.
94 def __init__(self, mode: Mode, features: Collection[Feature]) -> None:
96 self.features = features
97 self.current_line: Line
100 def line(self, indent: int = 0) -> Iterator[Line]:
103 If the line is empty, only emit if it makes sense.
104 If the line is too long, split it first and then generate.
106 If any lines were generated, set up a new current_line.
108 if not self.current_line:
109 self.current_line.depth += indent
110 return # Line is empty, don't emit. Creating a new one unnecessary.
112 complete_line = self.current_line
113 self.current_line = Line(mode=self.mode, depth=complete_line.depth + indent)
116 def visit_default(self, node: LN) -> Iterator[Line]:
117 """Default `visit_*()` implementation. Recurses to children of `node`."""
118 if isinstance(node, Leaf):
119 any_open_brackets = self.current_line.bracket_tracker.any_open_brackets()
120 for comment in generate_comments(node):
121 if any_open_brackets:
122 # any comment within brackets is subject to splitting
123 self.current_line.append(comment)
124 elif comment.type == token.COMMENT:
125 # regular trailing comment
126 self.current_line.append(comment)
127 yield from self.line()
130 # regular standalone comment
131 yield from self.line()
133 self.current_line.append(comment)
134 yield from self.line()
136 normalize_prefix(node, inside_brackets=any_open_brackets)
137 if self.mode.string_normalization and node.type == token.STRING:
138 node.value = normalize_string_prefix(node.value)
139 node.value = normalize_string_quotes(node.value)
140 if node.type == token.NUMBER:
141 normalize_numeric_literal(node)
142 if node.type not in WHITESPACE:
143 self.current_line.append(node)
144 yield from super().visit_default(node)
146 def visit_test(self, node: Node) -> Iterator[Line]:
147 """Visit an `x if y else z` test"""
149 if Preview.parenthesize_conditional_expressions in self.mode:
150 already_parenthesized = (
151 node.prev_sibling and node.prev_sibling.type == token.LPAR
154 if not already_parenthesized:
155 lpar = Leaf(token.LPAR, "")
156 rpar = Leaf(token.RPAR, "")
157 node.insert_child(0, lpar)
158 node.append_child(rpar)
160 yield from self.visit_default(node)
162 def visit_INDENT(self, node: Leaf) -> Iterator[Line]:
163 """Increase indentation level, maybe yield a line."""
164 # In blib2to3 INDENT never holds comments.
165 yield from self.line(+1)
166 yield from self.visit_default(node)
168 def visit_DEDENT(self, node: Leaf) -> Iterator[Line]:
169 """Decrease indentation level, maybe yield a line."""
170 # The current line might still wait for trailing comments. At DEDENT time
171 # there won't be any (they would be prefixes on the preceding NEWLINE).
172 # Emit the line then.
173 yield from self.line()
175 # While DEDENT has no value, its prefix may contain standalone comments
176 # that belong to the current indentation level. Get 'em.
177 yield from self.visit_default(node)
179 # Finally, emit the dedent.
180 yield from self.line(-1)
183 self, node: Node, keywords: Set[str], parens: Set[str]
185 """Visit a statement.
187 This implementation is shared for `if`, `while`, `for`, `try`, `except`,
188 `def`, `with`, `class`, `assert`, and assignments.
190 The relevant Python language `keywords` for a given statement will be
191 NAME leaves within it. This methods puts those on a separate line.
193 `parens` holds a set of string leaf values immediately after which
194 invisible parens should be put.
196 normalize_invisible_parens(
197 node, parens_after=parens, mode=self.mode, features=self.features
199 for child in node.children:
200 if is_name_token(child) and child.value in keywords:
201 yield from self.line()
203 yield from self.visit(child)
205 def visit_dictsetmaker(self, node: Node) -> Iterator[Line]:
206 if Preview.wrap_long_dict_values_in_parens in self.mode:
207 for i, child in enumerate(node.children):
210 if node.children[i - 1].type == token.COLON:
211 if child.type == syms.atom and child.children[0].type == token.LPAR:
212 if maybe_make_parens_invisible_in_atom(
215 remove_brackets_around_comma=False,
217 wrap_in_parentheses(node, child, visible=False)
219 wrap_in_parentheses(node, child, visible=False)
220 yield from self.visit_default(node)
222 def visit_funcdef(self, node: Node) -> Iterator[Line]:
223 """Visit function definition."""
224 yield from self.line()
226 # Remove redundant brackets around return type annotation.
227 is_return_annotation = False
228 for child in node.children:
229 if child.type == token.RARROW:
230 is_return_annotation = True
231 elif is_return_annotation:
232 if child.type == syms.atom and child.children[0].type == token.LPAR:
233 if maybe_make_parens_invisible_in_atom(
236 remove_brackets_around_comma=False,
238 wrap_in_parentheses(node, child, visible=False)
240 wrap_in_parentheses(node, child, visible=False)
241 is_return_annotation = False
243 for child in node.children:
244 yield from self.visit(child)
246 def visit_match_case(self, node: Node) -> Iterator[Line]:
247 """Visit either a match or case statement."""
248 normalize_invisible_parens(
249 node, parens_after=set(), mode=self.mode, features=self.features
252 yield from self.line()
253 for child in node.children:
254 yield from self.visit(child)
256 def visit_suite(self, node: Node) -> Iterator[Line]:
258 if self.mode.is_pyi and is_stub_suite(node):
259 yield from self.visit(node.children[2])
261 yield from self.visit_default(node)
263 def visit_simple_stmt(self, node: Node) -> Iterator[Line]:
264 """Visit a statement without nested statements."""
265 prev_type: Optional[int] = None
266 for child in node.children:
267 if (prev_type is None or prev_type == token.SEMI) and is_arith_like(child):
268 wrap_in_parentheses(node, child, visible=False)
269 prev_type = child.type
271 is_suite_like = node.parent and node.parent.type in STATEMENT
273 if self.mode.is_pyi and is_stub_body(node):
274 yield from self.visit_default(node)
276 yield from self.line(+1)
277 yield from self.visit_default(node)
278 yield from self.line(-1)
284 or not is_stub_suite(node.parent)
286 yield from self.line()
287 yield from self.visit_default(node)
289 def visit_async_stmt(self, node: Node) -> Iterator[Line]:
290 """Visit `async def`, `async for`, `async with`."""
291 yield from self.line()
293 children = iter(node.children)
294 for child in children:
295 yield from self.visit(child)
297 if child.type == token.ASYNC or child.type == STANDALONE_COMMENT:
298 # STANDALONE_COMMENT happens when `# fmt: skip` is applied on the async
302 internal_stmt = next(children)
303 for child in internal_stmt.children:
304 yield from self.visit(child)
306 def visit_decorators(self, node: Node) -> Iterator[Line]:
307 """Visit decorators."""
308 for child in node.children:
309 yield from self.line()
310 yield from self.visit(child)
312 def visit_power(self, node: Node) -> Iterator[Line]:
313 for idx, leaf in enumerate(node.children[:-1]):
314 next_leaf = node.children[idx + 1]
316 if not isinstance(leaf, Leaf):
319 value = leaf.value.lower()
321 leaf.type == token.NUMBER
322 and next_leaf.type == syms.trailer
323 # Ensure that we are in an attribute trailer
324 and next_leaf.children[0].type == token.DOT
325 # It shouldn't wrap hexadecimal, binary and octal literals
326 and not value.startswith(("0x", "0b", "0o"))
327 # It shouldn't wrap complex literals
330 wrap_in_parentheses(node, leaf)
332 remove_await_parens(node)
334 yield from self.visit_default(node)
336 def visit_SEMI(self, leaf: Leaf) -> Iterator[Line]:
337 """Remove a semicolon and put the other statement on a separate line."""
338 yield from self.line()
340 def visit_ENDMARKER(self, leaf: Leaf) -> Iterator[Line]:
341 """End of file. Process outstanding comments and end with a newline."""
342 yield from self.visit_default(leaf)
343 yield from self.line()
345 def visit_STANDALONE_COMMENT(self, leaf: Leaf) -> Iterator[Line]:
346 if not self.current_line.bracket_tracker.any_open_brackets():
347 yield from self.line()
348 yield from self.visit_default(leaf)
350 def visit_factor(self, node: Node) -> Iterator[Line]:
351 """Force parentheses between a unary op and a binary power:
355 _operator, operand = node.children
357 operand.type == syms.power
358 and len(operand.children) == 3
359 and operand.children[1].type == token.DOUBLESTAR
361 lpar = Leaf(token.LPAR, "(")
362 rpar = Leaf(token.RPAR, ")")
363 index = operand.remove() or 0
364 node.insert_child(index, Node(syms.atom, [lpar, operand, rpar]))
365 yield from self.visit_default(node)
367 def visit_STRING(self, leaf: Leaf) -> Iterator[Line]:
368 if Preview.hex_codes_in_unicode_sequences in self.mode:
369 normalize_unicode_escape_sequences(leaf)
371 if is_docstring(leaf) and "\\\n" not in leaf.value:
372 # We're ignoring docstrings with backslash newline escapes because changing
373 # indentation of those changes the AST representation of the code.
374 if self.mode.string_normalization:
375 docstring = normalize_string_prefix(leaf.value)
376 # visit_default() does handle string normalization for us, but
377 # since this method acts differently depending on quote style (ex.
378 # see padding logic below), there's a possibility for unstable
379 # formatting as visit_default() is called *after*. To avoid a
380 # situation where this function formats a docstring differently on
381 # the second pass, normalize it early.
382 docstring = normalize_string_quotes(docstring)
384 docstring = leaf.value
385 prefix = get_string_prefix(docstring)
386 docstring = docstring[len(prefix) :] # Remove the prefix
387 quote_char = docstring[0]
388 # A natural way to remove the outer quotes is to do:
389 # docstring = docstring.strip(quote_char)
390 # but that breaks on """""x""" (which is '""x').
391 # So we actually need to remove the first character and the next two
392 # characters but only if they are the same as the first.
393 quote_len = 1 if docstring[1] != quote_char else 3
394 docstring = docstring[quote_len:-quote_len]
395 docstring_started_empty = not docstring
396 indent = " " * 4 * self.current_line.depth
398 if is_multiline_string(leaf):
399 docstring = fix_docstring(docstring, indent)
401 docstring = docstring.strip()
403 has_trailing_backslash = False
405 # Add some padding if the docstring starts / ends with a quote mark.
406 if docstring[0] == quote_char:
407 docstring = " " + docstring
408 if docstring[-1] == quote_char:
410 if docstring[-1] == "\\":
411 backslash_count = len(docstring) - len(docstring.rstrip("\\"))
412 if backslash_count % 2:
413 # Odd number of tailing backslashes, add some padding to
414 # avoid escaping the closing string quote.
416 has_trailing_backslash = True
417 elif not docstring_started_empty:
420 # We could enforce triple quotes at this point.
421 quote = quote_char * quote_len
423 # It's invalid to put closing single-character quotes on a new line.
424 if self.mode and quote_len == 3:
425 # We need to find the length of the last line of the docstring
426 # to find if we can add the closing quotes to the line without
427 # exceeding the maximum line length.
428 # If docstring is one line, we don't put the closing quotes on a
429 # separate line because it looks ugly (#3320).
430 lines = docstring.splitlines()
431 last_line_length = len(lines[-1]) if docstring else 0
433 # If adding closing quotes would cause the last line to exceed
434 # the maximum line length then put a line break before the
438 and last_line_length + quote_len > self.mode.line_length
439 and len(indent) + quote_len <= self.mode.line_length
440 and not has_trailing_backslash
442 leaf.value = prefix + quote + docstring + "\n" + indent + quote
444 leaf.value = prefix + quote + docstring + quote
446 leaf.value = prefix + quote + docstring + quote
448 yield from self.visit_default(leaf)
450 def __post_init__(self) -> None:
451 """You are in a twisty little maze of passages."""
452 self.current_line = Line(mode=self.mode)
456 self.visit_assert_stmt = partial(v, keywords={"assert"}, parens={"assert", ","})
457 self.visit_if_stmt = partial(
458 v, keywords={"if", "else", "elif"}, parens={"if", "elif"}
460 self.visit_while_stmt = partial(v, keywords={"while", "else"}, parens={"while"})
461 self.visit_for_stmt = partial(v, keywords={"for", "else"}, parens={"for", "in"})
462 self.visit_try_stmt = partial(
463 v, keywords={"try", "except", "else", "finally"}, parens=Ø
465 self.visit_except_clause = partial(v, keywords={"except"}, parens={"except"})
466 self.visit_with_stmt = partial(v, keywords={"with"}, parens={"with"})
467 self.visit_classdef = partial(v, keywords={"class"}, parens=Ø)
468 self.visit_expr_stmt = partial(v, keywords=Ø, parens=ASSIGNMENTS)
469 self.visit_return_stmt = partial(v, keywords={"return"}, parens={"return"})
470 self.visit_import_from = partial(v, keywords=Ø, parens={"import"})
471 self.visit_del_stmt = partial(v, keywords=Ø, parens={"del"})
472 self.visit_async_funcdef = self.visit_async_stmt
473 self.visit_decorated = self.visit_decorators
476 self.visit_match_stmt = self.visit_match_case
477 self.visit_case_block = self.visit_match_case
481 line: Line, mode: Mode, features: Collection[Feature] = ()
483 """Transform a `line`, potentially splitting it into many lines.
485 They should fit in the allotted `line_length` but might not be able to.
487 `features` are syntactical features that may be used in the output.
493 line_str = line_to_string(line)
495 ll = mode.line_length
496 sn = mode.string_normalization
497 string_merge = StringMerger(ll, sn)
498 string_paren_strip = StringParenStripper(ll, sn)
499 string_split = StringSplitter(ll, sn)
500 string_paren_wrap = StringParenWrapper(ll, sn)
502 transformers: List[Transformer]
504 not line.contains_uncollapsable_type_comments()
505 and not line.should_split_rhs
506 and not line.magic_trailing_comma
508 is_line_short_enough(line, line_length=mode.line_length, line_str=line_str)
509 or line.contains_unsplittable_type_ignore()
511 and not (line.inside_brackets and line.contains_standalone_comments())
513 # Only apply basic string preprocessing, since lines shouldn't be split here.
514 if Preview.string_processing in mode:
515 transformers = [string_merge, string_paren_strip]
519 transformers = [left_hand_split]
523 self: object, line: Line, features: Collection[Feature], mode: Mode
525 """Wraps calls to `right_hand_split`.
527 The calls increasingly `omit` right-hand trailers (bracket pairs with
528 content), meaning the trailers get glued together to split on another
529 bracket pair instead.
531 for omit in generate_trailers_to_omit(line, mode.line_length):
533 right_hand_split(line, mode.line_length, features, omit=omit)
535 # Note: this check is only able to figure out if the first line of the
536 # *current* transformation fits in the line length. This is true only
537 # for simple cases. All others require running more transforms via
538 # `transform_line()`. This check doesn't know if those would succeed.
539 if is_line_short_enough(lines[0], line_length=mode.line_length):
543 # All splits failed, best effort split with no omits.
544 # This mostly happens to multiline strings that are by definition
545 # reported as not fitting a single line, as well as lines that contain
546 # trailing commas (those have to be exploded).
547 yield from right_hand_split(
548 line, line_length=mode.line_length, features=features
551 # HACK: nested functions (like _rhs) compiled by mypyc don't retain their
552 # __name__ attribute which is needed in `run_transformer` further down.
553 # Unfortunately a nested class breaks mypyc too. So a class must be created
554 # via type ... https://github.com/mypyc/mypyc/issues/884
555 rhs = type("rhs", (), {"__call__": _rhs})()
557 if Preview.string_processing in mode:
558 if line.inside_brackets:
564 standalone_comment_split,
577 if line.inside_brackets:
578 transformers = [delimiter_split, standalone_comment_split, rhs]
581 # It's always safe to attempt hugging of power operations and pretty much every line
583 transformers.append(hug_power_op)
585 for transform in transformers:
586 # We are accumulating lines in `result` because we might want to abort
587 # mission and return the original line in the end, or attempt a different
590 result = run_transformer(line, transform, mode, features, line_str=line_str)
591 except CannotTransform:
601 class _BracketSplitComponent(Enum):
608 line: Line, _features: Collection[Feature], mode: Mode
610 """Split line into many lines, starting with the first matching bracket pair.
612 Note: this usually looks weird, only use this for function definitions.
613 Prefer RHS otherwise. This is why this function is not symmetrical with
614 :func:`right_hand_split` which also handles optional parentheses.
616 tail_leaves: List[Leaf] = []
617 body_leaves: List[Leaf] = []
618 head_leaves: List[Leaf] = []
619 current_leaves = head_leaves
620 matching_bracket: Optional[Leaf] = None
621 for leaf in line.leaves:
623 current_leaves is body_leaves
624 and leaf.type in CLOSING_BRACKETS
625 and leaf.opening_bracket is matching_bracket
626 and isinstance(matching_bracket, Leaf)
629 ensure_visible(matching_bracket)
630 current_leaves = tail_leaves if body_leaves else head_leaves
631 current_leaves.append(leaf)
632 if current_leaves is head_leaves:
633 if leaf.type in OPENING_BRACKETS:
634 matching_bracket = leaf
635 current_leaves = body_leaves
636 if not matching_bracket:
637 raise CannotSplit("No brackets found")
639 head = bracket_split_build_line(
640 head_leaves, line, matching_bracket, component=_BracketSplitComponent.head
642 body = bracket_split_build_line(
643 body_leaves, line, matching_bracket, component=_BracketSplitComponent.body
645 tail = bracket_split_build_line(
646 tail_leaves, line, matching_bracket, component=_BracketSplitComponent.tail
648 bracket_split_succeeded_or_raise(head, body, tail)
649 for result in (head, body, tail):
656 """Intermediate split result from a right hand split."""
661 opening_bracket: Leaf
662 closing_bracket: Leaf
665 def right_hand_split(
668 features: Collection[Feature] = (),
669 omit: Collection[LeafID] = (),
671 """Split line into many lines, starting with the last matching bracket pair.
673 If the split was by optional parentheses, attempt splitting without them, too.
674 `omit` is a collection of closing bracket IDs that shouldn't be considered for
677 Note: running this function modifies `bracket_depth` on the leaves of `line`.
679 rhs_result = _first_right_hand_split(line, omit=omit)
680 yield from _maybe_split_omitting_optional_parens(
681 rhs_result, line, line_length, features=features, omit=omit
685 def _first_right_hand_split(
687 omit: Collection[LeafID] = (),
689 """Split the line into head, body, tail starting with the last bracket pair.
691 Note: this function should not have side effects. It's relied upon by
692 _maybe_split_omitting_optional_parens to get an opinion whether to prefer
693 splitting on the right side of an assignment statement.
695 tail_leaves: List[Leaf] = []
696 body_leaves: List[Leaf] = []
697 head_leaves: List[Leaf] = []
698 current_leaves = tail_leaves
699 opening_bracket: Optional[Leaf] = None
700 closing_bracket: Optional[Leaf] = None
701 for leaf in reversed(line.leaves):
702 if current_leaves is body_leaves:
703 if leaf is opening_bracket:
704 current_leaves = head_leaves if body_leaves else tail_leaves
705 current_leaves.append(leaf)
706 if current_leaves is tail_leaves:
707 if leaf.type in CLOSING_BRACKETS and id(leaf) not in omit:
708 opening_bracket = leaf.opening_bracket
709 closing_bracket = leaf
710 current_leaves = body_leaves
711 if not (opening_bracket and closing_bracket and head_leaves):
712 # If there is no opening or closing_bracket that means the split failed and
713 # all content is in the tail. Otherwise, if `head_leaves` are empty, it means
714 # the matching `opening_bracket` wasn't available on `line` anymore.
715 raise CannotSplit("No brackets found")
717 tail_leaves.reverse()
718 body_leaves.reverse()
719 head_leaves.reverse()
720 head = bracket_split_build_line(
721 head_leaves, line, opening_bracket, component=_BracketSplitComponent.head
723 body = bracket_split_build_line(
724 body_leaves, line, opening_bracket, component=_BracketSplitComponent.body
726 tail = bracket_split_build_line(
727 tail_leaves, line, opening_bracket, component=_BracketSplitComponent.tail
729 bracket_split_succeeded_or_raise(head, body, tail)
730 return _RHSResult(head, body, tail, opening_bracket, closing_bracket)
733 def _maybe_split_omitting_optional_parens(
737 features: Collection[Feature] = (),
738 omit: Collection[LeafID] = (),
741 Feature.FORCE_OPTIONAL_PARENTHESES not in features
742 # the opening bracket is an optional paren
743 and rhs.opening_bracket.type == token.LPAR
744 and not rhs.opening_bracket.value
745 # the closing bracket is an optional paren
746 and rhs.closing_bracket.type == token.RPAR
747 and not rhs.closing_bracket.value
748 # it's not an import (optional parens are the only thing we can split on
749 # in this case; attempting a split without them is a waste of time)
750 and not line.is_import
751 # there are no standalone comments in the body
752 and not rhs.body.contains_standalone_comments(0)
753 # and we can actually remove the parens
754 and can_omit_invisible_parens(rhs.body, line_length)
756 omit = {id(rhs.closing_bracket), *omit}
758 # The _RHSResult Omitting Optional Parens.
759 rhs_oop = _first_right_hand_split(line, omit=omit)
761 Preview.prefer_splitting_right_hand_side_of_assignments in line.mode
762 # the split is right after `=`
763 and len(rhs.head.leaves) >= 2
764 and rhs.head.leaves[-2].type == token.EQUAL
765 # the left side of assignment contains brackets
766 and any(leaf.type in BRACKETS for leaf in rhs.head.leaves[:-1])
767 # the left side of assignment is short enough (the -1 is for the ending
769 and is_line_short_enough(rhs.head, line_length=line_length - 1)
770 # the left side of assignment won't explode further because of magic
772 and rhs.head.magic_trailing_comma is None
773 # the split by omitting optional parens isn't preferred by some other
775 and not _prefer_split_rhs_oop(rhs_oop, line_length=line_length)
777 yield from _maybe_split_omitting_optional_parens(
778 rhs_oop, line, line_length, features=features, omit=omit
782 except CannotSplit as e:
784 can_be_split(rhs.body)
785 or is_line_short_enough(rhs.body, line_length=line_length)
788 "Splitting failed, body is still too long and can't be split."
792 rhs.head.contains_multiline_strings()
793 or rhs.tail.contains_multiline_strings()
796 "The current optional pair of parentheses is bound to fail to"
797 " satisfy the splitting algorithm because the head or the tail"
798 " contains multiline strings which by definition never fit one"
802 ensure_visible(rhs.opening_bracket)
803 ensure_visible(rhs.closing_bracket)
804 for result in (rhs.head, rhs.body, rhs.tail):
809 def _prefer_split_rhs_oop(rhs_oop: _RHSResult, line_length: int) -> bool:
811 Returns whether we should prefer the result from a split omitting optional parens.
813 has_closing_bracket_after_assign = False
814 for leaf in reversed(rhs_oop.head.leaves):
815 if leaf.type == token.EQUAL:
817 if leaf.type in CLOSING_BRACKETS:
818 has_closing_bracket_after_assign = True
821 # contains matching brackets after the `=` (done by checking there is a
823 has_closing_bracket_after_assign
825 # the split is actually from inside the optional parens (done by checking
826 # the first line still contains the `=`)
827 any(leaf.type == token.EQUAL for leaf in rhs_oop.head.leaves)
828 # the first line is short enough
829 and is_line_short_enough(rhs_oop.head, line_length=line_length)
831 # contains unsplittable type ignore
832 or rhs_oop.head.contains_unsplittable_type_ignore()
833 or rhs_oop.body.contains_unsplittable_type_ignore()
834 or rhs_oop.tail.contains_unsplittable_type_ignore()
838 def bracket_split_succeeded_or_raise(head: Line, body: Line, tail: Line) -> None:
839 """Raise :exc:`CannotSplit` if the last left- or right-hand split failed.
841 Do nothing otherwise.
843 A left- or right-hand split is based on a pair of brackets. Content before
844 (and including) the opening bracket is left on one line, content inside the
845 brackets is put on a separate line, and finally content starting with and
846 following the closing bracket is put on a separate line.
848 Those are called `head`, `body`, and `tail`, respectively. If the split
849 produced the same line (all content in `head`) or ended up with an empty `body`
850 and the `tail` is just the closing bracket, then it's considered failed.
852 tail_len = len(str(tail).strip())
855 raise CannotSplit("Splitting brackets produced the same line")
859 f"Splitting brackets on an empty body to save {tail_len} characters is"
864 def bracket_split_build_line(
867 opening_bracket: Leaf,
869 component: _BracketSplitComponent,
871 """Return a new line with given `leaves` and respective comments from `original`.
873 If it's the head component, brackets will be tracked so trailing commas are
876 If it's the body component, the result line is one-indented inside brackets and as
877 such has its first leaf's prefix normalized and a trailing comma added when
880 result = Line(mode=original.mode, depth=original.depth)
881 if component is _BracketSplitComponent.body:
882 result.inside_brackets = True
885 # Since body is a new indent level, remove spurious leading whitespace.
886 normalize_prefix(leaves[0], inside_brackets=True)
887 # Ensure a trailing comma for imports and standalone function arguments, but
888 # be careful not to add one after any comments or within type annotations.
891 and opening_bracket.value == "("
892 and not any(leaf.type == token.COMMA for leaf in leaves)
893 # In particular, don't add one within a parenthesized return annotation.
894 # Unfortunately the indicator we're in a return annotation (RARROW) may
895 # be defined directly in the parent node, the parent of the parent ...
896 # and so on depending on how complex the return annotation is.
897 # This isn't perfect and there's some false negatives but they are in
898 # contexts were a comma is actually fine.
900 node.prev_sibling.type == RARROW
903 getattr(leaves[0].parent, "parent", None),
905 if isinstance(node, Node) and isinstance(node.prev_sibling, Leaf)
909 if original.is_import or no_commas:
910 for i in range(len(leaves) - 1, -1, -1):
911 if leaves[i].type == STANDALONE_COMMENT:
914 if leaves[i].type != token.COMMA:
915 new_comma = Leaf(token.COMMA, ",")
916 leaves.insert(i + 1, new_comma)
919 leaves_to_track: Set[LeafID] = set()
920 if component is _BracketSplitComponent.head:
921 leaves_to_track = get_leaves_inside_matching_brackets(leaves)
927 track_bracket=id(leaf) in leaves_to_track,
929 for comment_after in original.comments_after(leaf):
930 result.append(comment_after, preformatted=True)
931 if component is _BracketSplitComponent.body and should_split_line(
932 result, opening_bracket
934 result.should_split_rhs = True
938 def dont_increase_indentation(split_func: Transformer) -> Transformer:
939 """Normalize prefix of the first leaf in every line returned by `split_func`.
941 This is a decorator over relevant split functions.
946 line: Line, features: Collection[Feature], mode: Mode
948 for split_line in split_func(line, features, mode):
949 normalize_prefix(split_line.leaves[0], inside_brackets=True)
955 def _get_last_non_comment_leaf(line: Line) -> Optional[int]:
956 for leaf_idx in range(len(line.leaves) - 1, 0, -1):
957 if line.leaves[leaf_idx].type != STANDALONE_COMMENT:
962 def _safe_add_trailing_comma(safe: bool, delimiter_priority: int, line: Line) -> Line:
965 and delimiter_priority == COMMA_PRIORITY
966 and line.leaves[-1].type != token.COMMA
967 and line.leaves[-1].type != STANDALONE_COMMENT
969 new_comma = Leaf(token.COMMA, ",")
970 line.append(new_comma)
974 @dont_increase_indentation
976 line: Line, features: Collection[Feature], mode: Mode
978 """Split according to delimiters of the highest priority.
980 If the appropriate Features are given, the split will add trailing commas
981 also in function signatures and calls that contain `*` and `**`.
984 last_leaf = line.leaves[-1]
986 raise CannotSplit("Line empty") from None
988 bt = line.bracket_tracker
990 delimiter_priority = bt.max_delimiter_priority(exclude={id(last_leaf)})
992 raise CannotSplit("No delimiters found") from None
994 if delimiter_priority == DOT_PRIORITY:
995 if bt.delimiter_count_with_priority(delimiter_priority) == 1:
996 raise CannotSplit("Splitting a single attribute from its owner looks wrong")
999 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1001 lowest_depth = sys.maxsize
1002 trailing_comma_safe = True
1004 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1005 """Append `leaf` to current line or to new line if appending impossible."""
1006 nonlocal current_line
1008 current_line.append_safe(leaf, preformatted=True)
1012 current_line = Line(
1013 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1015 current_line.append(leaf)
1017 last_non_comment_leaf = _get_last_non_comment_leaf(line)
1018 for leaf_idx, leaf in enumerate(line.leaves):
1019 yield from append_to_line(leaf)
1021 for comment_after in line.comments_after(leaf):
1022 yield from append_to_line(comment_after)
1024 lowest_depth = min(lowest_depth, leaf.bracket_depth)
1025 if leaf.bracket_depth == lowest_depth:
1026 if is_vararg(leaf, within={syms.typedargslist}):
1027 trailing_comma_safe = (
1028 trailing_comma_safe and Feature.TRAILING_COMMA_IN_DEF in features
1030 elif is_vararg(leaf, within={syms.arglist, syms.argument}):
1031 trailing_comma_safe = (
1032 trailing_comma_safe and Feature.TRAILING_COMMA_IN_CALL in features
1036 Preview.add_trailing_comma_consistently in mode
1037 and last_leaf.type == STANDALONE_COMMENT
1038 and leaf_idx == last_non_comment_leaf
1040 current_line = _safe_add_trailing_comma(
1041 trailing_comma_safe, delimiter_priority, current_line
1044 leaf_priority = bt.delimiters.get(id(leaf))
1045 if leaf_priority == delimiter_priority:
1048 current_line = Line(
1049 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1052 current_line = _safe_add_trailing_comma(
1053 trailing_comma_safe, delimiter_priority, current_line
1058 @dont_increase_indentation
1059 def standalone_comment_split(
1060 line: Line, features: Collection[Feature], mode: Mode
1061 ) -> Iterator[Line]:
1062 """Split standalone comments from the rest of the line."""
1063 if not line.contains_standalone_comments(0):
1064 raise CannotSplit("Line does not have any standalone comments")
1066 current_line = Line(
1067 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1070 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1071 """Append `leaf` to current line or to new line if appending impossible."""
1072 nonlocal current_line
1074 current_line.append_safe(leaf, preformatted=True)
1078 current_line = Line(
1079 line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1081 current_line.append(leaf)
1083 for leaf in line.leaves:
1084 yield from append_to_line(leaf)
1086 for comment_after in line.comments_after(leaf):
1087 yield from append_to_line(comment_after)
1093 def normalize_prefix(leaf: Leaf, *, inside_brackets: bool) -> None:
1094 """Leave existing extra newlines if not `inside_brackets`. Remove everything
1097 Note: don't use backslashes for formatting or you'll lose your voting rights.
1099 if not inside_brackets:
1100 spl = leaf.prefix.split("#")
1101 if "\\" not in spl[0]:
1102 nl_count = spl[-1].count("\n")
1105 leaf.prefix = "\n" * nl_count
1111 def normalize_invisible_parens(
1112 node: Node, parens_after: Set[str], *, mode: Mode, features: Collection[Feature]
1114 """Make existing optional parentheses invisible or create new ones.
1116 `parens_after` is a set of string leaf values immediately after which parens
1119 Standardizes on visible parentheses for single-element tuples, and keeps
1120 existing visible parentheses for other tuples and generator expressions.
1122 for pc in list_comments(node.prefix, is_endmarker=False):
1123 if pc.value in FMT_OFF:
1124 # This `node` has a prefix with `# fmt: off`, don't mess with parens.
1127 # The multiple context managers grammar has a different pattern, thus this is
1128 # separate from the for-loop below. This possibly wraps them in invisible parens,
1129 # and later will be removed in remove_with_parens when needed.
1130 if node.type == syms.with_stmt:
1131 _maybe_wrap_cms_in_parens(node, mode, features)
1134 for index, child in enumerate(list(node.children)):
1135 # Fixes a bug where invisible parens are not properly stripped from
1136 # assignment statements that contain type annotations.
1137 if isinstance(child, Node) and child.type == syms.annassign:
1138 normalize_invisible_parens(
1139 child, parens_after=parens_after, mode=mode, features=features
1142 # Add parentheses around long tuple unpacking in assignments.
1145 and isinstance(child, Node)
1146 and child.type == syms.testlist_star_expr
1152 child.type == syms.atom
1153 and node.type == syms.for_stmt
1154 and isinstance(child.prev_sibling, Leaf)
1155 and child.prev_sibling.type == token.NAME
1156 and child.prev_sibling.value == "for"
1158 if maybe_make_parens_invisible_in_atom(
1161 remove_brackets_around_comma=True,
1163 wrap_in_parentheses(node, child, visible=False)
1164 elif isinstance(child, Node) and node.type == syms.with_stmt:
1165 remove_with_parens(child, node)
1166 elif child.type == syms.atom:
1167 if maybe_make_parens_invisible_in_atom(
1171 wrap_in_parentheses(node, child, visible=False)
1172 elif is_one_tuple(child):
1173 wrap_in_parentheses(node, child, visible=True)
1174 elif node.type == syms.import_from:
1175 _normalize_import_from(node, child, index)
1179 and child.type == token.STAR
1180 and node.type == syms.except_clause
1182 # In except* (PEP 654), the star is actually part of
1183 # of the keyword. So we need to skip the insertion of
1184 # invisible parentheses to work more precisely.
1187 elif not (isinstance(child, Leaf) and is_multiline_string(child)):
1188 wrap_in_parentheses(node, child, visible=False)
1190 comma_check = child.type == token.COMMA
1192 check_lpar = isinstance(child, Leaf) and (
1193 child.value in parens_after or comma_check
1197 def _normalize_import_from(parent: Node, child: LN, index: int) -> None:
1198 # "import from" nodes store parentheses directly as part of
1200 if is_lpar_token(child):
1201 assert is_rpar_token(parent.children[-1])
1202 # make parentheses invisible
1204 parent.children[-1].value = ""
1205 elif child.type != token.STAR:
1206 # insert invisible parentheses
1207 parent.insert_child(index, Leaf(token.LPAR, ""))
1208 parent.append_child(Leaf(token.RPAR, ""))
1211 def remove_await_parens(node: Node) -> None:
1212 if node.children[0].type == token.AWAIT and len(node.children) > 1:
1214 node.children[1].type == syms.atom
1215 and node.children[1].children[0].type == token.LPAR
1217 if maybe_make_parens_invisible_in_atom(
1220 remove_brackets_around_comma=True,
1222 wrap_in_parentheses(node, node.children[1], visible=False)
1224 # Since await is an expression we shouldn't remove
1225 # brackets in cases where this would change
1226 # the AST due to operator precedence.
1227 # Therefore we only aim to remove brackets around
1228 # power nodes that aren't also await expressions themselves.
1229 # https://peps.python.org/pep-0492/#updated-operator-precedence-table
1230 # N.B. We've still removed any redundant nested brackets though :)
1231 opening_bracket = cast(Leaf, node.children[1].children[0])
1232 closing_bracket = cast(Leaf, node.children[1].children[-1])
1233 bracket_contents = node.children[1].children[1]
1234 if isinstance(bracket_contents, Node):
1235 if bracket_contents.type != syms.power:
1236 ensure_visible(opening_bracket)
1237 ensure_visible(closing_bracket)
1239 bracket_contents.type == syms.power
1240 and bracket_contents.children[0].type == token.AWAIT
1242 ensure_visible(opening_bracket)
1243 ensure_visible(closing_bracket)
1244 # If we are in a nested await then recurse down.
1245 remove_await_parens(bracket_contents)
1248 def _maybe_wrap_cms_in_parens(
1249 node: Node, mode: Mode, features: Collection[Feature]
1251 """When enabled and safe, wrap the multiple context managers in invisible parens.
1253 It is only safe when `features` contain Feature.PARENTHESIZED_CONTEXT_MANAGERS.
1256 Feature.PARENTHESIZED_CONTEXT_MANAGERS not in features
1257 or Preview.wrap_multiple_context_managers_in_parens not in mode
1258 or len(node.children) <= 2
1259 # If it's an atom, it's already wrapped in parens.
1260 or node.children[1].type == syms.atom
1263 colon_index: Optional[int] = None
1264 for i in range(2, len(node.children)):
1265 if node.children[i].type == token.COLON:
1268 if colon_index is not None:
1269 lpar = Leaf(token.LPAR, "")
1270 rpar = Leaf(token.RPAR, "")
1271 context_managers = node.children[1:colon_index]
1272 for child in context_managers:
1274 # After wrapping, the with_stmt will look like this:
1280 # ... <-- context_managers
1286 syms.atom, [lpar, Node(syms.testlist_gexp, context_managers), rpar]
1288 node.insert_child(1, new_child)
1291 def remove_with_parens(node: Node, parent: Node) -> None:
1292 """Recursively hide optional parens in `with` statements."""
1293 # Removing all unnecessary parentheses in with statements in one pass is a tad
1294 # complex as different variations of bracketed statements result in pretty
1295 # different parse trees:
1297 # with (open("file")) as f: # this is an asexpr_test
1300 # with (open("file") as f): # this is an atom containing an
1303 # with (open("file")) as f, (open("file")) as f: # this is asexpr_test, COMMA,
1306 # with (open("file") as f, open("file") as f): # an atom containing a
1307 # ... # testlist_gexp which then
1308 # # contains multiple asexpr_test(s)
1309 if node.type == syms.atom:
1310 if maybe_make_parens_invisible_in_atom(
1313 remove_brackets_around_comma=True,
1315 wrap_in_parentheses(parent, node, visible=False)
1316 if isinstance(node.children[1], Node):
1317 remove_with_parens(node.children[1], node)
1318 elif node.type == syms.testlist_gexp:
1319 for child in node.children:
1320 if isinstance(child, Node):
1321 remove_with_parens(child, node)
1322 elif node.type == syms.asexpr_test and not any(
1323 leaf.type == token.COLONEQUAL for leaf in node.leaves()
1325 if maybe_make_parens_invisible_in_atom(
1328 remove_brackets_around_comma=True,
1330 wrap_in_parentheses(node, node.children[0], visible=False)
1333 def maybe_make_parens_invisible_in_atom(
1336 remove_brackets_around_comma: bool = False,
1338 """If it's safe, make the parens in the atom `node` invisible, recursively.
1339 Additionally, remove repeated, adjacent invisible parens from the atom `node`
1340 as they are redundant.
1342 Returns whether the node should itself be wrapped in invisible parentheses.
1345 node.type != syms.atom
1346 or is_empty_tuple(node)
1347 or is_one_tuple(node)
1348 or (is_yield(node) and parent.type != syms.expr_stmt)
1350 # This condition tries to prevent removing non-optional brackets
1351 # around a tuple, however, can be a bit overzealous so we provide
1352 # and option to skip this check for `for` and `with` statements.
1353 not remove_brackets_around_comma
1354 and max_delimiter_priority_in_atom(node) >= COMMA_PRIORITY
1356 or is_tuple_containing_walrus(node)
1360 if is_walrus_assignment(node):
1369 # these ones aren't useful to end users, but they do please fuzzers
1376 first = node.children[0]
1377 last = node.children[-1]
1378 if is_lpar_token(first) and is_rpar_token(last):
1379 middle = node.children[1]
1380 # make parentheses invisible
1383 maybe_make_parens_invisible_in_atom(
1386 remove_brackets_around_comma=remove_brackets_around_comma,
1389 if is_atom_with_invisible_parens(middle):
1390 # Strip the invisible parens from `middle` by replacing
1391 # it with the child in-between the invisible parens
1392 middle.replace(middle.children[1])
1399 def should_split_line(line: Line, opening_bracket: Leaf) -> bool:
1400 """Should `line` be immediately split with `delimiter_split()` after RHS?"""
1402 if not (opening_bracket.parent and opening_bracket.value in "[{("):
1405 # We're essentially checking if the body is delimited by commas and there's more
1406 # than one of them (we're excluding the trailing comma and if the delimiter priority
1407 # is still commas, that means there's more).
1409 trailing_comma = False
1411 last_leaf = line.leaves[-1]
1412 if last_leaf.type == token.COMMA:
1413 trailing_comma = True
1414 exclude.add(id(last_leaf))
1415 max_priority = line.bracket_tracker.max_delimiter_priority(exclude=exclude)
1416 except (IndexError, ValueError):
1419 return max_priority == COMMA_PRIORITY and (
1420 (line.mode.magic_trailing_comma and trailing_comma)
1421 # always explode imports
1422 or opening_bracket.parent.type in {syms.atom, syms.import_from}
1426 def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[LeafID]]:
1427 """Generate sets of closing bracket IDs that should be omitted in a RHS.
1429 Brackets can be omitted if the entire trailer up to and including
1430 a preceding closing bracket fits in one line.
1432 Yielded sets are cumulative (contain results of previous yields, too). First
1433 set is empty, unless the line should explode, in which case bracket pairs until
1434 the one that needs to explode are omitted.
1437 omit: Set[LeafID] = set()
1438 if not line.magic_trailing_comma:
1441 length = 4 * line.depth
1442 opening_bracket: Optional[Leaf] = None
1443 closing_bracket: Optional[Leaf] = None
1444 inner_brackets: Set[LeafID] = set()
1445 for index, leaf, leaf_length in line.enumerate_with_length(reversed=True):
1446 length += leaf_length
1447 if length > line_length:
1450 has_inline_comment = leaf_length > len(leaf.value) + len(leaf.prefix)
1451 if leaf.type == STANDALONE_COMMENT or has_inline_comment:
1455 if leaf is opening_bracket:
1456 opening_bracket = None
1457 elif leaf.type in CLOSING_BRACKETS:
1458 prev = line.leaves[index - 1] if index > 0 else None
1461 and prev.type == token.COMMA
1462 and leaf.opening_bracket is not None
1463 and not is_one_sequence_between(
1464 leaf.opening_bracket, leaf, line.leaves
1467 # Never omit bracket pairs with trailing commas.
1468 # We need to explode on those.
1471 inner_brackets.add(id(leaf))
1472 elif leaf.type in CLOSING_BRACKETS:
1473 prev = line.leaves[index - 1] if index > 0 else None
1474 if prev and prev.type in OPENING_BRACKETS:
1475 # Empty brackets would fail a split so treat them as "inner"
1476 # brackets (e.g. only add them to the `omit` set if another
1477 # pair of brackets was good enough.
1478 inner_brackets.add(id(leaf))
1482 omit.add(id(closing_bracket))
1483 omit.update(inner_brackets)
1484 inner_brackets.clear()
1489 and prev.type == token.COMMA
1490 and leaf.opening_bracket is not None
1491 and not is_one_sequence_between(leaf.opening_bracket, leaf, line.leaves)
1493 # Never omit bracket pairs with trailing commas.
1494 # We need to explode on those.
1498 opening_bracket = leaf.opening_bracket
1499 closing_bracket = leaf
1502 def run_transformer(
1504 transform: Transformer,
1506 features: Collection[Feature],
1511 line_str = line_to_string(line)
1512 result: List[Line] = []
1513 for transformed_line in transform(line, features, mode):
1514 if str(transformed_line).strip("\n") == line_str:
1515 raise CannotTransform("Line transformer returned an unchanged result")
1517 result.extend(transform_line(transformed_line, mode=mode, features=features))
1519 features_set = set(features)
1521 Feature.FORCE_OPTIONAL_PARENTHESES in features_set
1522 or transform.__class__.__name__ != "rhs"
1523 or not line.bracket_tracker.invisible
1524 or any(bracket.value for bracket in line.bracket_tracker.invisible)
1525 or line.contains_multiline_strings()
1526 or result[0].contains_uncollapsable_type_comments()
1527 or result[0].contains_unsplittable_type_ignore()
1528 or is_line_short_enough(result[0], line_length=mode.line_length)
1529 # If any leaves have no parents (which _can_ occur since
1530 # `transform(line)` potentially destroys the line's underlying node
1531 # structure), then we can't proceed. Doing so would cause the below
1532 # call to `append_leaves()` to fail.
1533 or any(leaf.parent is None for leaf in line.leaves)
1537 line_copy = line.clone()
1538 append_leaves(line_copy, line, line.leaves)
1539 features_fop = features_set | {Feature.FORCE_OPTIONAL_PARENTHESES}
1540 second_opinion = run_transformer(
1541 line_copy, transform, mode, features_fop, line_str=line_str
1544 is_line_short_enough(ln, line_length=mode.line_length) for ln in second_opinion
1546 result = second_opinion