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, replace
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, mode=mode, 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):
532 lines = list(right_hand_split(line, mode, features, omit=omit))
533 # Note: this check is only able to figure out if the first line of the
534 # *current* transformation fits in the line length. This is true only
535 # for simple cases. All others require running more transforms via
536 # `transform_line()`. This check doesn't know if those would succeed.
537 if is_line_short_enough(lines[0], mode=mode):
541 # All splits failed, best effort split with no omits.
542 # This mostly happens to multiline strings that are by definition
543 # reported as not fitting a single line, as well as lines that contain
544 # trailing commas (those have to be exploded).
545 yield from right_hand_split(line, mode, features=features)
547 # HACK: nested functions (like _rhs) compiled by mypyc don't retain their
548 # __name__ attribute which is needed in `run_transformer` further down.
549 # Unfortunately a nested class breaks mypyc too. So a class must be created
550 # via type ... https://github.com/mypyc/mypyc/issues/884
551 rhs = type("rhs", (), {"__call__": _rhs})()
553 if Preview.string_processing in mode:
554 if line.inside_brackets:
560 standalone_comment_split,
573 if line.inside_brackets:
574 transformers = [delimiter_split, standalone_comment_split, rhs]
577 # It's always safe to attempt hugging of power operations and pretty much every line
579 transformers.append(hug_power_op)
581 for transform in transformers:
582 # We are accumulating lines in `result` because we might want to abort
583 # mission and return the original line in the end, or attempt a different
586 result = run_transformer(line, transform, mode, features, line_str=line_str)
587 except CannotTransform:
597 class _BracketSplitComponent(Enum):
604 line: Line, _features: Collection[Feature], mode: Mode
606 """Split line into many lines, starting with the first matching bracket pair.
608 Note: this usually looks weird, only use this for function definitions.
609 Prefer RHS otherwise. This is why this function is not symmetrical with
610 :func:`right_hand_split` which also handles optional parentheses.
612 tail_leaves: List[Leaf] = []
613 body_leaves: List[Leaf] = []
614 head_leaves: List[Leaf] = []
615 current_leaves = head_leaves
616 matching_bracket: Optional[Leaf] = None
617 for leaf in line.leaves:
619 current_leaves is body_leaves
620 and leaf.type in CLOSING_BRACKETS
621 and leaf.opening_bracket is matching_bracket
622 and isinstance(matching_bracket, Leaf)
625 ensure_visible(matching_bracket)
626 current_leaves = tail_leaves if body_leaves else head_leaves
627 current_leaves.append(leaf)
628 if current_leaves is head_leaves:
629 if leaf.type in OPENING_BRACKETS:
630 matching_bracket = leaf
631 current_leaves = body_leaves
632 if not matching_bracket:
633 raise CannotSplit("No brackets found")
635 head = bracket_split_build_line(
636 head_leaves, line, matching_bracket, component=_BracketSplitComponent.head
638 body = bracket_split_build_line(
639 body_leaves, line, matching_bracket, component=_BracketSplitComponent.body
641 tail = bracket_split_build_line(
642 tail_leaves, line, matching_bracket, component=_BracketSplitComponent.tail
644 bracket_split_succeeded_or_raise(head, body, tail)
645 for result in (head, body, tail):
652 """Intermediate split result from a right hand split."""
657 opening_bracket: Leaf
658 closing_bracket: Leaf
661 def right_hand_split(
664 features: Collection[Feature] = (),
665 omit: Collection[LeafID] = (),
667 """Split line into many lines, starting with the last matching bracket pair.
669 If the split was by optional parentheses, attempt splitting without them, too.
670 `omit` is a collection of closing bracket IDs that shouldn't be considered for
673 Note: running this function modifies `bracket_depth` on the leaves of `line`.
675 rhs_result = _first_right_hand_split(line, omit=omit)
676 yield from _maybe_split_omitting_optional_parens(
677 rhs_result, line, mode, features=features, omit=omit
681 def _first_right_hand_split(
683 omit: Collection[LeafID] = (),
685 """Split the line into head, body, tail starting with the last bracket pair.
687 Note: this function should not have side effects. It's relied upon by
688 _maybe_split_omitting_optional_parens to get an opinion whether to prefer
689 splitting on the right side of an assignment statement.
691 tail_leaves: List[Leaf] = []
692 body_leaves: List[Leaf] = []
693 head_leaves: List[Leaf] = []
694 current_leaves = tail_leaves
695 opening_bracket: Optional[Leaf] = None
696 closing_bracket: Optional[Leaf] = None
697 for leaf in reversed(line.leaves):
698 if current_leaves is body_leaves:
699 if leaf is opening_bracket:
700 current_leaves = head_leaves if body_leaves else tail_leaves
701 current_leaves.append(leaf)
702 if current_leaves is tail_leaves:
703 if leaf.type in CLOSING_BRACKETS and id(leaf) not in omit:
704 opening_bracket = leaf.opening_bracket
705 closing_bracket = leaf
706 current_leaves = body_leaves
707 if not (opening_bracket and closing_bracket and head_leaves):
708 # If there is no opening or closing_bracket that means the split failed and
709 # all content is in the tail. Otherwise, if `head_leaves` are empty, it means
710 # the matching `opening_bracket` wasn't available on `line` anymore.
711 raise CannotSplit("No brackets found")
713 tail_leaves.reverse()
714 body_leaves.reverse()
715 head_leaves.reverse()
716 head = bracket_split_build_line(
717 head_leaves, line, opening_bracket, component=_BracketSplitComponent.head
719 body = bracket_split_build_line(
720 body_leaves, line, opening_bracket, component=_BracketSplitComponent.body
722 tail = bracket_split_build_line(
723 tail_leaves, line, opening_bracket, component=_BracketSplitComponent.tail
725 bracket_split_succeeded_or_raise(head, body, tail)
726 return _RHSResult(head, body, tail, opening_bracket, closing_bracket)
729 def _maybe_split_omitting_optional_parens(
733 features: Collection[Feature] = (),
734 omit: Collection[LeafID] = (),
737 Feature.FORCE_OPTIONAL_PARENTHESES not in features
738 # the opening bracket is an optional paren
739 and rhs.opening_bracket.type == token.LPAR
740 and not rhs.opening_bracket.value
741 # the closing bracket is an optional paren
742 and rhs.closing_bracket.type == token.RPAR
743 and not rhs.closing_bracket.value
744 # it's not an import (optional parens are the only thing we can split on
745 # in this case; attempting a split without them is a waste of time)
746 and not line.is_import
747 # there are no standalone comments in the body
748 and not rhs.body.contains_standalone_comments(0)
749 # and we can actually remove the parens
750 and can_omit_invisible_parens(rhs.body, mode.line_length)
752 omit = {id(rhs.closing_bracket), *omit}
754 # The _RHSResult Omitting Optional Parens.
755 rhs_oop = _first_right_hand_split(line, omit=omit)
757 Preview.prefer_splitting_right_hand_side_of_assignments in line.mode
758 # the split is right after `=`
759 and len(rhs.head.leaves) >= 2
760 and rhs.head.leaves[-2].type == token.EQUAL
761 # the left side of assignment contains brackets
762 and any(leaf.type in BRACKETS for leaf in rhs.head.leaves[:-1])
763 # the left side of assignment is short enough (the -1 is for the ending
765 and is_line_short_enough(
766 rhs.head, mode=replace(mode, line_length=mode.line_length - 1)
768 # the left side of assignment won't explode further because of magic
770 and rhs.head.magic_trailing_comma is None
771 # the split by omitting optional parens isn't preferred by some other
773 and not _prefer_split_rhs_oop(rhs_oop, mode)
775 yield from _maybe_split_omitting_optional_parens(
776 rhs_oop, line, mode, features=features, omit=omit
780 except CannotSplit as e:
782 can_be_split(rhs.body) or is_line_short_enough(rhs.body, mode=mode)
785 "Splitting failed, body is still too long and can't be split."
789 rhs.head.contains_multiline_strings()
790 or rhs.tail.contains_multiline_strings()
793 "The current optional pair of parentheses is bound to fail to"
794 " satisfy the splitting algorithm because the head or the tail"
795 " contains multiline strings which by definition never fit one"
799 ensure_visible(rhs.opening_bracket)
800 ensure_visible(rhs.closing_bracket)
801 for result in (rhs.head, rhs.body, rhs.tail):
806 def _prefer_split_rhs_oop(rhs_oop: _RHSResult, mode: Mode) -> bool:
808 Returns whether we should prefer the result from a split omitting optional parens.
810 has_closing_bracket_after_assign = False
811 for leaf in reversed(rhs_oop.head.leaves):
812 if leaf.type == token.EQUAL:
814 if leaf.type in CLOSING_BRACKETS:
815 has_closing_bracket_after_assign = True
818 # contains matching brackets after the `=` (done by checking there is a
820 has_closing_bracket_after_assign
822 # the split is actually from inside the optional parens (done by checking
823 # the first line still contains the `=`)
824 any(leaf.type == token.EQUAL for leaf in rhs_oop.head.leaves)
825 # the first line is short enough
826 and is_line_short_enough(rhs_oop.head, mode=mode)
828 # contains unsplittable type ignore
829 or rhs_oop.head.contains_unsplittable_type_ignore()
830 or rhs_oop.body.contains_unsplittable_type_ignore()
831 or rhs_oop.tail.contains_unsplittable_type_ignore()
835 def bracket_split_succeeded_or_raise(head: Line, body: Line, tail: Line) -> None:
836 """Raise :exc:`CannotSplit` if the last left- or right-hand split failed.
838 Do nothing otherwise.
840 A left- or right-hand split is based on a pair of brackets. Content before
841 (and including) the opening bracket is left on one line, content inside the
842 brackets is put on a separate line, and finally content starting with and
843 following the closing bracket is put on a separate line.
845 Those are called `head`, `body`, and `tail`, respectively. If the split
846 produced the same line (all content in `head`) or ended up with an empty `body`
847 and the `tail` is just the closing bracket, then it's considered failed.
849 tail_len = len(str(tail).strip())
852 raise CannotSplit("Splitting brackets produced the same line")
856 f"Splitting brackets on an empty body to save {tail_len} characters is"
861 def bracket_split_build_line(
864 opening_bracket: Leaf,
866 component: _BracketSplitComponent,
868 """Return a new line with given `leaves` and respective comments from `original`.
870 If it's the head component, brackets will be tracked so trailing commas are
873 If it's the body component, the result line is one-indented inside brackets and as
874 such has its first leaf's prefix normalized and a trailing comma added when
877 result = Line(mode=original.mode, depth=original.depth)
878 if component is _BracketSplitComponent.body:
879 result.inside_brackets = True
882 # Since body is a new indent level, remove spurious leading whitespace.
883 normalize_prefix(leaves[0], inside_brackets=True)
884 # Ensure a trailing comma for imports and standalone function arguments, but
885 # be careful not to add one after any comments or within type annotations.
888 and opening_bracket.value == "("
889 and not any(leaf.type == token.COMMA for leaf in leaves)
890 # In particular, don't add one within a parenthesized return annotation.
891 # Unfortunately the indicator we're in a return annotation (RARROW) may
892 # be defined directly in the parent node, the parent of the parent ...
893 # and so on depending on how complex the return annotation is.
894 # This isn't perfect and there's some false negatives but they are in
895 # contexts were a comma is actually fine.
897 node.prev_sibling.type == RARROW
900 getattr(leaves[0].parent, "parent", None),
902 if isinstance(node, Node) and isinstance(node.prev_sibling, Leaf)
906 if original.is_import or no_commas:
907 for i in range(len(leaves) - 1, -1, -1):
908 if leaves[i].type == STANDALONE_COMMENT:
911 if leaves[i].type != token.COMMA:
912 new_comma = Leaf(token.COMMA, ",")
913 leaves.insert(i + 1, new_comma)
916 leaves_to_track: Set[LeafID] = set()
917 if component is _BracketSplitComponent.head:
918 leaves_to_track = get_leaves_inside_matching_brackets(leaves)
924 track_bracket=id(leaf) in leaves_to_track,
926 for comment_after in original.comments_after(leaf):
927 result.append(comment_after, preformatted=True)
928 if component is _BracketSplitComponent.body and should_split_line(
929 result, opening_bracket
931 result.should_split_rhs = True
935 def dont_increase_indentation(split_func: Transformer) -> Transformer:
936 """Normalize prefix of the first leaf in every line returned by `split_func`.
938 This is a decorator over relevant split functions.
943 line: Line, features: Collection[Feature], mode: Mode
945 for split_line in split_func(line, features, mode):
946 normalize_prefix(split_line.leaves[0], inside_brackets=True)
952 def _get_last_non_comment_leaf(line: Line) -> Optional[int]:
953 for leaf_idx in range(len(line.leaves) - 1, 0, -1):
954 if line.leaves[leaf_idx].type != STANDALONE_COMMENT:
959 def _safe_add_trailing_comma(safe: bool, delimiter_priority: int, line: Line) -> Line:
962 and delimiter_priority == COMMA_PRIORITY
963 and line.leaves[-1].type != token.COMMA
964 and line.leaves[-1].type != STANDALONE_COMMENT
966 new_comma = Leaf(token.COMMA, ",")
967 line.append(new_comma)
971 @dont_increase_indentation
973 line: Line, features: Collection[Feature], mode: Mode
975 """Split according to delimiters of the highest priority.
977 If the appropriate Features are given, the split will add trailing commas
978 also in function signatures and calls that contain `*` and `**`.
981 last_leaf = line.leaves[-1]
983 raise CannotSplit("Line empty") from None
985 bt = line.bracket_tracker
987 delimiter_priority = bt.max_delimiter_priority(exclude={id(last_leaf)})
989 raise CannotSplit("No delimiters found") from None
991 if delimiter_priority == DOT_PRIORITY:
992 if bt.delimiter_count_with_priority(delimiter_priority) == 1:
993 raise CannotSplit("Splitting a single attribute from its owner looks wrong")
996 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
998 lowest_depth = sys.maxsize
999 trailing_comma_safe = True
1001 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1002 """Append `leaf` to current line or to new line if appending impossible."""
1003 nonlocal current_line
1005 current_line.append_safe(leaf, preformatted=True)
1009 current_line = Line(
1010 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1012 current_line.append(leaf)
1014 last_non_comment_leaf = _get_last_non_comment_leaf(line)
1015 for leaf_idx, leaf in enumerate(line.leaves):
1016 yield from append_to_line(leaf)
1018 for comment_after in line.comments_after(leaf):
1019 yield from append_to_line(comment_after)
1021 lowest_depth = min(lowest_depth, leaf.bracket_depth)
1022 if leaf.bracket_depth == lowest_depth:
1023 if is_vararg(leaf, within={syms.typedargslist}):
1024 trailing_comma_safe = (
1025 trailing_comma_safe and Feature.TRAILING_COMMA_IN_DEF in features
1027 elif is_vararg(leaf, within={syms.arglist, syms.argument}):
1028 trailing_comma_safe = (
1029 trailing_comma_safe and Feature.TRAILING_COMMA_IN_CALL in features
1033 Preview.add_trailing_comma_consistently in mode
1034 and last_leaf.type == STANDALONE_COMMENT
1035 and leaf_idx == last_non_comment_leaf
1037 current_line = _safe_add_trailing_comma(
1038 trailing_comma_safe, delimiter_priority, current_line
1041 leaf_priority = bt.delimiters.get(id(leaf))
1042 if leaf_priority == delimiter_priority:
1045 current_line = Line(
1046 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1049 current_line = _safe_add_trailing_comma(
1050 trailing_comma_safe, delimiter_priority, current_line
1055 @dont_increase_indentation
1056 def standalone_comment_split(
1057 line: Line, features: Collection[Feature], mode: Mode
1058 ) -> Iterator[Line]:
1059 """Split standalone comments from the rest of the line."""
1060 if not line.contains_standalone_comments(0):
1061 raise CannotSplit("Line does not have any standalone comments")
1063 current_line = Line(
1064 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1067 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1068 """Append `leaf` to current line or to new line if appending impossible."""
1069 nonlocal current_line
1071 current_line.append_safe(leaf, preformatted=True)
1075 current_line = Line(
1076 line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1078 current_line.append(leaf)
1080 for leaf in line.leaves:
1081 yield from append_to_line(leaf)
1083 for comment_after in line.comments_after(leaf):
1084 yield from append_to_line(comment_after)
1090 def normalize_prefix(leaf: Leaf, *, inside_brackets: bool) -> None:
1091 """Leave existing extra newlines if not `inside_brackets`. Remove everything
1094 Note: don't use backslashes for formatting or you'll lose your voting rights.
1096 if not inside_brackets:
1097 spl = leaf.prefix.split("#")
1098 if "\\" not in spl[0]:
1099 nl_count = spl[-1].count("\n")
1102 leaf.prefix = "\n" * nl_count
1108 def normalize_invisible_parens(
1109 node: Node, parens_after: Set[str], *, mode: Mode, features: Collection[Feature]
1111 """Make existing optional parentheses invisible or create new ones.
1113 `parens_after` is a set of string leaf values immediately after which parens
1116 Standardizes on visible parentheses for single-element tuples, and keeps
1117 existing visible parentheses for other tuples and generator expressions.
1119 for pc in list_comments(node.prefix, is_endmarker=False):
1120 if pc.value in FMT_OFF:
1121 # This `node` has a prefix with `# fmt: off`, don't mess with parens.
1124 # The multiple context managers grammar has a different pattern, thus this is
1125 # separate from the for-loop below. This possibly wraps them in invisible parens,
1126 # and later will be removed in remove_with_parens when needed.
1127 if node.type == syms.with_stmt:
1128 _maybe_wrap_cms_in_parens(node, mode, features)
1131 for index, child in enumerate(list(node.children)):
1132 # Fixes a bug where invisible parens are not properly stripped from
1133 # assignment statements that contain type annotations.
1134 if isinstance(child, Node) and child.type == syms.annassign:
1135 normalize_invisible_parens(
1136 child, parens_after=parens_after, mode=mode, features=features
1139 # Add parentheses around long tuple unpacking in assignments.
1142 and isinstance(child, Node)
1143 and child.type == syms.testlist_star_expr
1149 child.type == syms.atom
1150 and node.type == syms.for_stmt
1151 and isinstance(child.prev_sibling, Leaf)
1152 and child.prev_sibling.type == token.NAME
1153 and child.prev_sibling.value == "for"
1155 if maybe_make_parens_invisible_in_atom(
1158 remove_brackets_around_comma=True,
1160 wrap_in_parentheses(node, child, visible=False)
1161 elif isinstance(child, Node) and node.type == syms.with_stmt:
1162 remove_with_parens(child, node)
1163 elif child.type == syms.atom:
1164 if maybe_make_parens_invisible_in_atom(
1168 wrap_in_parentheses(node, child, visible=False)
1169 elif is_one_tuple(child):
1170 wrap_in_parentheses(node, child, visible=True)
1171 elif node.type == syms.import_from:
1172 _normalize_import_from(node, child, index)
1176 and child.type == token.STAR
1177 and node.type == syms.except_clause
1179 # In except* (PEP 654), the star is actually part of
1180 # of the keyword. So we need to skip the insertion of
1181 # invisible parentheses to work more precisely.
1184 elif not (isinstance(child, Leaf) and is_multiline_string(child)):
1185 wrap_in_parentheses(node, child, visible=False)
1187 comma_check = child.type == token.COMMA
1189 check_lpar = isinstance(child, Leaf) and (
1190 child.value in parens_after or comma_check
1194 def _normalize_import_from(parent: Node, child: LN, index: int) -> None:
1195 # "import from" nodes store parentheses directly as part of
1197 if is_lpar_token(child):
1198 assert is_rpar_token(parent.children[-1])
1199 # make parentheses invisible
1201 parent.children[-1].value = ""
1202 elif child.type != token.STAR:
1203 # insert invisible parentheses
1204 parent.insert_child(index, Leaf(token.LPAR, ""))
1205 parent.append_child(Leaf(token.RPAR, ""))
1208 def remove_await_parens(node: Node) -> None:
1209 if node.children[0].type == token.AWAIT and len(node.children) > 1:
1211 node.children[1].type == syms.atom
1212 and node.children[1].children[0].type == token.LPAR
1214 if maybe_make_parens_invisible_in_atom(
1217 remove_brackets_around_comma=True,
1219 wrap_in_parentheses(node, node.children[1], visible=False)
1221 # Since await is an expression we shouldn't remove
1222 # brackets in cases where this would change
1223 # the AST due to operator precedence.
1224 # Therefore we only aim to remove brackets around
1225 # power nodes that aren't also await expressions themselves.
1226 # https://peps.python.org/pep-0492/#updated-operator-precedence-table
1227 # N.B. We've still removed any redundant nested brackets though :)
1228 opening_bracket = cast(Leaf, node.children[1].children[0])
1229 closing_bracket = cast(Leaf, node.children[1].children[-1])
1230 bracket_contents = node.children[1].children[1]
1231 if isinstance(bracket_contents, Node):
1232 if bracket_contents.type != syms.power:
1233 ensure_visible(opening_bracket)
1234 ensure_visible(closing_bracket)
1236 bracket_contents.type == syms.power
1237 and bracket_contents.children[0].type == token.AWAIT
1239 ensure_visible(opening_bracket)
1240 ensure_visible(closing_bracket)
1241 # If we are in a nested await then recurse down.
1242 remove_await_parens(bracket_contents)
1245 def _maybe_wrap_cms_in_parens(
1246 node: Node, mode: Mode, features: Collection[Feature]
1248 """When enabled and safe, wrap the multiple context managers in invisible parens.
1250 It is only safe when `features` contain Feature.PARENTHESIZED_CONTEXT_MANAGERS.
1253 Feature.PARENTHESIZED_CONTEXT_MANAGERS not in features
1254 or Preview.wrap_multiple_context_managers_in_parens not in mode
1255 or len(node.children) <= 2
1256 # If it's an atom, it's already wrapped in parens.
1257 or node.children[1].type == syms.atom
1260 colon_index: Optional[int] = None
1261 for i in range(2, len(node.children)):
1262 if node.children[i].type == token.COLON:
1265 if colon_index is not None:
1266 lpar = Leaf(token.LPAR, "")
1267 rpar = Leaf(token.RPAR, "")
1268 context_managers = node.children[1:colon_index]
1269 for child in context_managers:
1271 # After wrapping, the with_stmt will look like this:
1277 # ... <-- context_managers
1283 syms.atom, [lpar, Node(syms.testlist_gexp, context_managers), rpar]
1285 node.insert_child(1, new_child)
1288 def remove_with_parens(node: Node, parent: Node) -> None:
1289 """Recursively hide optional parens in `with` statements."""
1290 # Removing all unnecessary parentheses in with statements in one pass is a tad
1291 # complex as different variations of bracketed statements result in pretty
1292 # different parse trees:
1294 # with (open("file")) as f: # this is an asexpr_test
1297 # with (open("file") as f): # this is an atom containing an
1300 # with (open("file")) as f, (open("file")) as f: # this is asexpr_test, COMMA,
1303 # with (open("file") as f, open("file") as f): # an atom containing a
1304 # ... # testlist_gexp which then
1305 # # contains multiple asexpr_test(s)
1306 if node.type == syms.atom:
1307 if maybe_make_parens_invisible_in_atom(
1310 remove_brackets_around_comma=True,
1312 wrap_in_parentheses(parent, node, visible=False)
1313 if isinstance(node.children[1], Node):
1314 remove_with_parens(node.children[1], node)
1315 elif node.type == syms.testlist_gexp:
1316 for child in node.children:
1317 if isinstance(child, Node):
1318 remove_with_parens(child, node)
1319 elif node.type == syms.asexpr_test and not any(
1320 leaf.type == token.COLONEQUAL for leaf in node.leaves()
1322 if maybe_make_parens_invisible_in_atom(
1325 remove_brackets_around_comma=True,
1327 wrap_in_parentheses(node, node.children[0], visible=False)
1330 def maybe_make_parens_invisible_in_atom(
1333 remove_brackets_around_comma: bool = False,
1335 """If it's safe, make the parens in the atom `node` invisible, recursively.
1336 Additionally, remove repeated, adjacent invisible parens from the atom `node`
1337 as they are redundant.
1339 Returns whether the node should itself be wrapped in invisible parentheses.
1342 node.type != syms.atom
1343 or is_empty_tuple(node)
1344 or is_one_tuple(node)
1345 or (is_yield(node) and parent.type != syms.expr_stmt)
1347 # This condition tries to prevent removing non-optional brackets
1348 # around a tuple, however, can be a bit overzealous so we provide
1349 # and option to skip this check for `for` and `with` statements.
1350 not remove_brackets_around_comma
1351 and max_delimiter_priority_in_atom(node) >= COMMA_PRIORITY
1353 or is_tuple_containing_walrus(node)
1357 if is_walrus_assignment(node):
1366 # these ones aren't useful to end users, but they do please fuzzers
1373 first = node.children[0]
1374 last = node.children[-1]
1375 if is_lpar_token(first) and is_rpar_token(last):
1376 middle = node.children[1]
1377 # make parentheses invisible
1380 maybe_make_parens_invisible_in_atom(
1383 remove_brackets_around_comma=remove_brackets_around_comma,
1386 if is_atom_with_invisible_parens(middle):
1387 # Strip the invisible parens from `middle` by replacing
1388 # it with the child in-between the invisible parens
1389 middle.replace(middle.children[1])
1396 def should_split_line(line: Line, opening_bracket: Leaf) -> bool:
1397 """Should `line` be immediately split with `delimiter_split()` after RHS?"""
1399 if not (opening_bracket.parent and opening_bracket.value in "[{("):
1402 # We're essentially checking if the body is delimited by commas and there's more
1403 # than one of them (we're excluding the trailing comma and if the delimiter priority
1404 # is still commas, that means there's more).
1406 trailing_comma = False
1408 last_leaf = line.leaves[-1]
1409 if last_leaf.type == token.COMMA:
1410 trailing_comma = True
1411 exclude.add(id(last_leaf))
1412 max_priority = line.bracket_tracker.max_delimiter_priority(exclude=exclude)
1413 except (IndexError, ValueError):
1416 return max_priority == COMMA_PRIORITY and (
1417 (line.mode.magic_trailing_comma and trailing_comma)
1418 # always explode imports
1419 or opening_bracket.parent.type in {syms.atom, syms.import_from}
1423 def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[LeafID]]:
1424 """Generate sets of closing bracket IDs that should be omitted in a RHS.
1426 Brackets can be omitted if the entire trailer up to and including
1427 a preceding closing bracket fits in one line.
1429 Yielded sets are cumulative (contain results of previous yields, too). First
1430 set is empty, unless the line should explode, in which case bracket pairs until
1431 the one that needs to explode are omitted.
1434 omit: Set[LeafID] = set()
1435 if not line.magic_trailing_comma:
1438 length = 4 * line.depth
1439 opening_bracket: Optional[Leaf] = None
1440 closing_bracket: Optional[Leaf] = None
1441 inner_brackets: Set[LeafID] = set()
1442 for index, leaf, leaf_length in line.enumerate_with_length(reversed=True):
1443 length += leaf_length
1444 if length > line_length:
1447 has_inline_comment = leaf_length > len(leaf.value) + len(leaf.prefix)
1448 if leaf.type == STANDALONE_COMMENT or has_inline_comment:
1452 if leaf is opening_bracket:
1453 opening_bracket = None
1454 elif leaf.type in CLOSING_BRACKETS:
1455 prev = line.leaves[index - 1] if index > 0 else None
1458 and prev.type == token.COMMA
1459 and leaf.opening_bracket is not None
1460 and not is_one_sequence_between(
1461 leaf.opening_bracket, leaf, line.leaves
1464 # Never omit bracket pairs with trailing commas.
1465 # We need to explode on those.
1468 inner_brackets.add(id(leaf))
1469 elif leaf.type in CLOSING_BRACKETS:
1470 prev = line.leaves[index - 1] if index > 0 else None
1471 if prev and prev.type in OPENING_BRACKETS:
1472 # Empty brackets would fail a split so treat them as "inner"
1473 # brackets (e.g. only add them to the `omit` set if another
1474 # pair of brackets was good enough.
1475 inner_brackets.add(id(leaf))
1479 omit.add(id(closing_bracket))
1480 omit.update(inner_brackets)
1481 inner_brackets.clear()
1486 and prev.type == token.COMMA
1487 and leaf.opening_bracket is not None
1488 and not is_one_sequence_between(leaf.opening_bracket, leaf, line.leaves)
1490 # Never omit bracket pairs with trailing commas.
1491 # We need to explode on those.
1495 opening_bracket = leaf.opening_bracket
1496 closing_bracket = leaf
1499 def run_transformer(
1501 transform: Transformer,
1503 features: Collection[Feature],
1508 line_str = line_to_string(line)
1509 result: List[Line] = []
1510 for transformed_line in transform(line, features, mode):
1511 if str(transformed_line).strip("\n") == line_str:
1512 raise CannotTransform("Line transformer returned an unchanged result")
1514 result.extend(transform_line(transformed_line, mode=mode, features=features))
1516 features_set = set(features)
1518 Feature.FORCE_OPTIONAL_PARENTHESES in features_set
1519 or transform.__class__.__name__ != "rhs"
1520 or not line.bracket_tracker.invisible
1521 or any(bracket.value for bracket in line.bracket_tracker.invisible)
1522 or line.contains_multiline_strings()
1523 or result[0].contains_uncollapsable_type_comments()
1524 or result[0].contains_unsplittable_type_ignore()
1525 or is_line_short_enough(result[0], mode=mode)
1526 # If any leaves have no parents (which _can_ occur since
1527 # `transform(line)` potentially destroys the line's underlying node
1528 # structure), then we can't proceed. Doing so would cause the below
1529 # call to `append_leaves()` to fail.
1530 or any(leaf.parent is None for leaf in line.leaves)
1534 line_copy = line.clone()
1535 append_leaves(line_copy, line, line.leaves)
1536 features_fop = features_set | {Feature.FORCE_OPTIONAL_PARENTHESES}
1537 second_opinion = run_transformer(
1538 line_copy, transform, mode, features_fop, line_str=line_str
1540 if all(is_line_short_enough(ln, mode=mode) for ln in second_opinion):
1541 result = second_opinion