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 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 (
22 can_omit_invisible_parens,
26 from black.mode import Feature, Mode, Preview
27 from black.nodes import (
39 is_atom_with_invisible_parens,
45 is_one_sequence_between,
50 is_tuple_containing_walrus,
57 from black.numerics import normalize_numeric_literal
58 from black.strings import (
61 normalize_string_prefix,
62 normalize_string_quotes,
63 normalize_unicode_escape_sequences,
65 from black.trans import (
74 from blib2to3.pgen2 import token
75 from blib2to3.pytree import Leaf, Node
79 LN = Union[Leaf, Node]
82 class CannotSplit(CannotTransform):
83 """A readable split that fits the allotted line length is impossible."""
86 # This isn't a dataclass because @dataclass + Generic breaks mypyc.
87 # See also https://github.com/mypyc/mypyc/issues/827.
88 class LineGenerator(Visitor[Line]):
89 """Generates reformatted Line objects. Empty lines are not emitted.
91 Note: destroys the tree it's visiting by mutating prefixes of its leaves
92 in ways that will no longer stringify to valid Python code on the tree.
95 def __init__(self, mode: Mode, features: Collection[Feature]) -> None:
97 self.features = features
98 self.current_line: Line
101 def line(self, indent: int = 0) -> Iterator[Line]:
104 If the line is empty, only emit if it makes sense.
105 If the line is too long, split it first and then generate.
107 If any lines were generated, set up a new current_line.
109 if not self.current_line:
110 self.current_line.depth += indent
111 return # Line is empty, don't emit. Creating a new one unnecessary.
113 complete_line = self.current_line
114 self.current_line = Line(mode=self.mode, depth=complete_line.depth + indent)
117 def visit_default(self, node: LN) -> Iterator[Line]:
118 """Default `visit_*()` implementation. Recurses to children of `node`."""
119 if isinstance(node, Leaf):
120 any_open_brackets = self.current_line.bracket_tracker.any_open_brackets()
121 for comment in generate_comments(node):
122 if any_open_brackets:
123 # any comment within brackets is subject to splitting
124 self.current_line.append(comment)
125 elif comment.type == token.COMMENT:
126 # regular trailing comment
127 self.current_line.append(comment)
128 yield from self.line()
131 # regular standalone comment
132 yield from self.line()
134 self.current_line.append(comment)
135 yield from self.line()
137 normalize_prefix(node, inside_brackets=any_open_brackets)
138 if self.mode.string_normalization and node.type == token.STRING:
139 node.value = normalize_string_prefix(node.value)
140 node.value = normalize_string_quotes(node.value)
141 if node.type == token.NUMBER:
142 normalize_numeric_literal(node)
143 if node.type not in WHITESPACE:
144 self.current_line.append(node)
145 yield from super().visit_default(node)
147 def visit_test(self, node: Node) -> Iterator[Line]:
148 """Visit an `x if y else z` test"""
150 if Preview.parenthesize_conditional_expressions in self.mode:
151 already_parenthesized = (
152 node.prev_sibling and node.prev_sibling.type == token.LPAR
155 if not already_parenthesized:
156 lpar = Leaf(token.LPAR, "")
157 rpar = Leaf(token.RPAR, "")
158 node.insert_child(0, lpar)
159 node.append_child(rpar)
161 yield from self.visit_default(node)
163 def visit_INDENT(self, node: Leaf) -> Iterator[Line]:
164 """Increase indentation level, maybe yield a line."""
165 # In blib2to3 INDENT never holds comments.
166 yield from self.line(+1)
167 yield from self.visit_default(node)
169 def visit_DEDENT(self, node: Leaf) -> Iterator[Line]:
170 """Decrease indentation level, maybe yield a line."""
171 # The current line might still wait for trailing comments. At DEDENT time
172 # there won't be any (they would be prefixes on the preceding NEWLINE).
173 # Emit the line then.
174 yield from self.line()
176 # While DEDENT has no value, its prefix may contain standalone comments
177 # that belong to the current indentation level. Get 'em.
178 yield from self.visit_default(node)
180 # Finally, emit the dedent.
181 yield from self.line(-1)
184 self, node: Node, keywords: Set[str], parens: Set[str]
186 """Visit a statement.
188 This implementation is shared for `if`, `while`, `for`, `try`, `except`,
189 `def`, `with`, `class`, `assert`, and assignments.
191 The relevant Python language `keywords` for a given statement will be
192 NAME leaves within it. This methods puts those on a separate line.
194 `parens` holds a set of string leaf values immediately after which
195 invisible parens should be put.
197 normalize_invisible_parens(
198 node, parens_after=parens, mode=self.mode, features=self.features
200 for child in node.children:
201 if is_name_token(child) and child.value in keywords:
202 yield from self.line()
204 yield from self.visit(child)
206 def visit_dictsetmaker(self, node: Node) -> Iterator[Line]:
207 if Preview.wrap_long_dict_values_in_parens in self.mode:
208 for i, child in enumerate(node.children):
211 if node.children[i - 1].type == token.COLON:
212 if child.type == syms.atom and child.children[0].type == token.LPAR:
213 if maybe_make_parens_invisible_in_atom(
216 remove_brackets_around_comma=False,
218 wrap_in_parentheses(node, child, visible=False)
220 wrap_in_parentheses(node, child, visible=False)
221 yield from self.visit_default(node)
223 def visit_funcdef(self, node: Node) -> Iterator[Line]:
224 """Visit function definition."""
225 yield from self.line()
227 # Remove redundant brackets around return type annotation.
228 is_return_annotation = False
229 for child in node.children:
230 if child.type == token.RARROW:
231 is_return_annotation = True
232 elif is_return_annotation:
233 if child.type == syms.atom and child.children[0].type == token.LPAR:
234 if maybe_make_parens_invisible_in_atom(
237 remove_brackets_around_comma=False,
239 wrap_in_parentheses(node, child, visible=False)
241 wrap_in_parentheses(node, child, visible=False)
242 is_return_annotation = False
244 for child in node.children:
245 yield from self.visit(child)
247 def visit_match_case(self, node: Node) -> Iterator[Line]:
248 """Visit either a match or case statement."""
249 normalize_invisible_parens(
250 node, parens_after=set(), mode=self.mode, features=self.features
253 yield from self.line()
254 for child in node.children:
255 yield from self.visit(child)
257 def visit_suite(self, node: Node) -> Iterator[Line]:
259 if self.mode.is_pyi and is_stub_suite(node):
260 yield from self.visit(node.children[2])
262 yield from self.visit_default(node)
264 def visit_simple_stmt(self, node: Node) -> Iterator[Line]:
265 """Visit a statement without nested statements."""
266 prev_type: Optional[int] = None
267 for child in node.children:
268 if (prev_type is None or prev_type == token.SEMI) and is_arith_like(child):
269 wrap_in_parentheses(node, child, visible=False)
270 prev_type = child.type
272 is_suite_like = node.parent and node.parent.type in STATEMENT
274 if self.mode.is_pyi and is_stub_body(node):
275 yield from self.visit_default(node)
277 yield from self.line(+1)
278 yield from self.visit_default(node)
279 yield from self.line(-1)
285 or not is_stub_suite(node.parent)
287 yield from self.line()
288 yield from self.visit_default(node)
290 def visit_async_stmt(self, node: Node) -> Iterator[Line]:
291 """Visit `async def`, `async for`, `async with`."""
292 yield from self.line()
294 children = iter(node.children)
295 for child in children:
296 yield from self.visit(child)
298 if child.type == token.ASYNC or child.type == STANDALONE_COMMENT:
299 # STANDALONE_COMMENT happens when `# fmt: skip` is applied on the async
303 internal_stmt = next(children)
304 for child in internal_stmt.children:
305 yield from self.visit(child)
307 def visit_decorators(self, node: Node) -> Iterator[Line]:
308 """Visit decorators."""
309 for child in node.children:
310 yield from self.line()
311 yield from self.visit(child)
313 def visit_power(self, node: Node) -> Iterator[Line]:
314 for idx, leaf in enumerate(node.children[:-1]):
315 next_leaf = node.children[idx + 1]
317 if not isinstance(leaf, Leaf):
320 value = leaf.value.lower()
322 leaf.type == token.NUMBER
323 and next_leaf.type == syms.trailer
324 # Ensure that we are in an attribute trailer
325 and next_leaf.children[0].type == token.DOT
326 # It shouldn't wrap hexadecimal, binary and octal literals
327 and not value.startswith(("0x", "0b", "0o"))
328 # It shouldn't wrap complex literals
331 wrap_in_parentheses(node, leaf)
333 remove_await_parens(node)
335 yield from self.visit_default(node)
337 def visit_SEMI(self, leaf: Leaf) -> Iterator[Line]:
338 """Remove a semicolon and put the other statement on a separate line."""
339 yield from self.line()
341 def visit_ENDMARKER(self, leaf: Leaf) -> Iterator[Line]:
342 """End of file. Process outstanding comments and end with a newline."""
343 yield from self.visit_default(leaf)
344 yield from self.line()
346 def visit_STANDALONE_COMMENT(self, leaf: Leaf) -> Iterator[Line]:
347 if not self.current_line.bracket_tracker.any_open_brackets():
348 yield from self.line()
349 yield from self.visit_default(leaf)
351 def visit_factor(self, node: Node) -> Iterator[Line]:
352 """Force parentheses between a unary op and a binary power:
356 _operator, operand = node.children
358 operand.type == syms.power
359 and len(operand.children) == 3
360 and operand.children[1].type == token.DOUBLESTAR
362 lpar = Leaf(token.LPAR, "(")
363 rpar = Leaf(token.RPAR, ")")
364 index = operand.remove() or 0
365 node.insert_child(index, Node(syms.atom, [lpar, operand, rpar]))
366 yield from self.visit_default(node)
368 def visit_STRING(self, leaf: Leaf) -> Iterator[Line]:
369 if Preview.hex_codes_in_unicode_sequences in self.mode:
370 normalize_unicode_escape_sequences(leaf)
372 if is_docstring(leaf) and "\\\n" not in leaf.value:
373 # We're ignoring docstrings with backslash newline escapes because changing
374 # indentation of those changes the AST representation of the code.
375 if self.mode.string_normalization:
376 docstring = normalize_string_prefix(leaf.value)
377 # visit_default() does handle string normalization for us, but
378 # since this method acts differently depending on quote style (ex.
379 # see padding logic below), there's a possibility for unstable
380 # formatting as visit_default() is called *after*. To avoid a
381 # situation where this function formats a docstring differently on
382 # the second pass, normalize it early.
383 docstring = normalize_string_quotes(docstring)
385 docstring = leaf.value
386 prefix = get_string_prefix(docstring)
387 docstring = docstring[len(prefix) :] # Remove the prefix
388 quote_char = docstring[0]
389 # A natural way to remove the outer quotes is to do:
390 # docstring = docstring.strip(quote_char)
391 # but that breaks on """""x""" (which is '""x').
392 # So we actually need to remove the first character and the next two
393 # characters but only if they are the same as the first.
394 quote_len = 1 if docstring[1] != quote_char else 3
395 docstring = docstring[quote_len:-quote_len]
396 docstring_started_empty = not docstring
397 indent = " " * 4 * self.current_line.depth
399 if is_multiline_string(leaf):
400 docstring = fix_docstring(docstring, indent)
402 docstring = docstring.strip()
404 has_trailing_backslash = False
406 # Add some padding if the docstring starts / ends with a quote mark.
407 if docstring[0] == quote_char:
408 docstring = " " + docstring
409 if docstring[-1] == quote_char:
411 if docstring[-1] == "\\":
412 backslash_count = len(docstring) - len(docstring.rstrip("\\"))
413 if backslash_count % 2:
414 # Odd number of tailing backslashes, add some padding to
415 # avoid escaping the closing string quote.
417 has_trailing_backslash = True
418 elif not docstring_started_empty:
421 # We could enforce triple quotes at this point.
422 quote = quote_char * quote_len
424 # It's invalid to put closing single-character quotes on a new line.
425 if self.mode and quote_len == 3:
426 # We need to find the length of the last line of the docstring
427 # to find if we can add the closing quotes to the line without
428 # exceeding the maximum line length.
429 # If docstring is one line, we don't put the closing quotes on a
430 # separate line because it looks ugly (#3320).
431 lines = docstring.splitlines()
432 last_line_length = len(lines[-1]) if docstring else 0
434 # If adding closing quotes would cause the last line to exceed
435 # the maximum line length then put a line break before the
439 and last_line_length + quote_len > self.mode.line_length
440 and len(indent) + quote_len <= self.mode.line_length
441 and not has_trailing_backslash
443 leaf.value = prefix + quote + docstring + "\n" + indent + quote
445 leaf.value = prefix + quote + docstring + quote
447 leaf.value = prefix + quote + docstring + quote
449 yield from self.visit_default(leaf)
451 def __post_init__(self) -> None:
452 """You are in a twisty little maze of passages."""
453 self.current_line = Line(mode=self.mode)
457 self.visit_assert_stmt = partial(v, keywords={"assert"}, parens={"assert", ","})
458 self.visit_if_stmt = partial(
459 v, keywords={"if", "else", "elif"}, parens={"if", "elif"}
461 self.visit_while_stmt = partial(v, keywords={"while", "else"}, parens={"while"})
462 self.visit_for_stmt = partial(v, keywords={"for", "else"}, parens={"for", "in"})
463 self.visit_try_stmt = partial(
464 v, keywords={"try", "except", "else", "finally"}, parens=Ø
466 self.visit_except_clause = partial(v, keywords={"except"}, parens={"except"})
467 self.visit_with_stmt = partial(v, keywords={"with"}, parens={"with"})
468 self.visit_classdef = partial(v, keywords={"class"}, parens=Ø)
469 self.visit_expr_stmt = partial(v, keywords=Ø, parens=ASSIGNMENTS)
470 self.visit_return_stmt = partial(v, keywords={"return"}, parens={"return"})
471 self.visit_import_from = partial(v, keywords=Ø, parens={"import"})
472 self.visit_del_stmt = partial(v, keywords=Ø, parens={"del"})
473 self.visit_async_funcdef = self.visit_async_stmt
474 self.visit_decorated = self.visit_decorators
477 self.visit_match_stmt = self.visit_match_case
478 self.visit_case_block = self.visit_match_case
482 line: Line, mode: Mode, features: Collection[Feature] = ()
484 """Transform a `line`, potentially splitting it into many lines.
486 They should fit in the allotted `line_length` but might not be able to.
488 `features` are syntactical features that may be used in the output.
494 line_str = line_to_string(line)
496 ll = mode.line_length
497 sn = mode.string_normalization
498 string_merge = StringMerger(ll, sn)
499 string_paren_strip = StringParenStripper(ll, sn)
500 string_split = StringSplitter(ll, sn)
501 string_paren_wrap = StringParenWrapper(ll, sn)
503 transformers: List[Transformer]
505 not line.contains_uncollapsable_type_comments()
506 and not line.should_split_rhs
507 and not line.magic_trailing_comma
509 is_line_short_enough(line, mode=mode, line_str=line_str)
510 or line.contains_unsplittable_type_ignore()
512 and not (line.inside_brackets and line.contains_standalone_comments())
514 # Only apply basic string preprocessing, since lines shouldn't be split here.
515 if Preview.string_processing in mode:
516 transformers = [string_merge, string_paren_strip]
520 transformers = [left_hand_split]
524 self: object, line: Line, features: Collection[Feature], mode: Mode
526 """Wraps calls to `right_hand_split`.
528 The calls increasingly `omit` right-hand trailers (bracket pairs with
529 content), meaning the trailers get glued together to split on another
530 bracket pair instead.
532 for omit in generate_trailers_to_omit(line, mode.line_length):
533 lines = list(right_hand_split(line, mode, features, omit=omit))
534 # Note: this check is only able to figure out if the first line of the
535 # *current* transformation fits in the line length. This is true only
536 # for simple cases. All others require running more transforms via
537 # `transform_line()`. This check doesn't know if those would succeed.
538 if is_line_short_enough(lines[0], mode=mode):
542 # All splits failed, best effort split with no omits.
543 # This mostly happens to multiline strings that are by definition
544 # reported as not fitting a single line, as well as lines that contain
545 # trailing commas (those have to be exploded).
546 yield from right_hand_split(line, mode, features=features)
548 # HACK: nested functions (like _rhs) compiled by mypyc don't retain their
549 # __name__ attribute which is needed in `run_transformer` further down.
550 # Unfortunately a nested class breaks mypyc too. So a class must be created
551 # via type ... https://github.com/mypyc/mypyc/issues/884
552 rhs = type("rhs", (), {"__call__": _rhs})()
554 if Preview.string_processing in mode:
555 if line.inside_brackets:
561 standalone_comment_split,
574 if line.inside_brackets:
575 transformers = [delimiter_split, standalone_comment_split, rhs]
578 # It's always safe to attempt hugging of power operations and pretty much every line
580 transformers.append(hug_power_op)
582 for transform in transformers:
583 # We are accumulating lines in `result` because we might want to abort
584 # mission and return the original line in the end, or attempt a different
587 result = run_transformer(line, transform, mode, features, line_str=line_str)
588 except CannotTransform:
598 class _BracketSplitComponent(Enum):
605 line: Line, _features: Collection[Feature], mode: Mode
607 """Split line into many lines, starting with the first matching bracket pair.
609 Note: this usually looks weird, only use this for function definitions.
610 Prefer RHS otherwise. This is why this function is not symmetrical with
611 :func:`right_hand_split` which also handles optional parentheses.
613 tail_leaves: List[Leaf] = []
614 body_leaves: List[Leaf] = []
615 head_leaves: List[Leaf] = []
616 current_leaves = head_leaves
617 matching_bracket: Optional[Leaf] = None
618 for leaf in line.leaves:
620 current_leaves is body_leaves
621 and leaf.type in CLOSING_BRACKETS
622 and leaf.opening_bracket is matching_bracket
623 and isinstance(matching_bracket, Leaf)
626 ensure_visible(matching_bracket)
627 current_leaves = tail_leaves if body_leaves else head_leaves
628 current_leaves.append(leaf)
629 if current_leaves is head_leaves:
630 if leaf.type in OPENING_BRACKETS:
631 matching_bracket = leaf
632 current_leaves = body_leaves
633 if not matching_bracket:
634 raise CannotSplit("No brackets found")
636 head = bracket_split_build_line(
637 head_leaves, line, matching_bracket, component=_BracketSplitComponent.head
639 body = bracket_split_build_line(
640 body_leaves, line, matching_bracket, component=_BracketSplitComponent.body
642 tail = bracket_split_build_line(
643 tail_leaves, line, matching_bracket, component=_BracketSplitComponent.tail
645 bracket_split_succeeded_or_raise(head, body, tail)
646 for result in (head, body, tail):
651 def right_hand_split(
654 features: Collection[Feature] = (),
655 omit: Collection[LeafID] = (),
657 """Split line into many lines, starting with the last matching bracket pair.
659 If the split was by optional parentheses, attempt splitting without them, too.
660 `omit` is a collection of closing bracket IDs that shouldn't be considered for
663 Note: running this function modifies `bracket_depth` on the leaves of `line`.
665 rhs_result = _first_right_hand_split(line, omit=omit)
666 yield from _maybe_split_omitting_optional_parens(
667 rhs_result, line, mode, features=features, omit=omit
671 def _first_right_hand_split(
673 omit: Collection[LeafID] = (),
675 """Split the line into head, body, tail starting with the last bracket pair.
677 Note: this function should not have side effects. It's relied upon by
678 _maybe_split_omitting_optional_parens to get an opinion whether to prefer
679 splitting on the right side of an assignment statement.
681 tail_leaves: List[Leaf] = []
682 body_leaves: List[Leaf] = []
683 head_leaves: List[Leaf] = []
684 current_leaves = tail_leaves
685 opening_bracket: Optional[Leaf] = None
686 closing_bracket: Optional[Leaf] = None
687 for leaf in reversed(line.leaves):
688 if current_leaves is body_leaves:
689 if leaf is opening_bracket:
690 current_leaves = head_leaves if body_leaves else tail_leaves
691 current_leaves.append(leaf)
692 if current_leaves is tail_leaves:
693 if leaf.type in CLOSING_BRACKETS and id(leaf) not in omit:
694 opening_bracket = leaf.opening_bracket
695 closing_bracket = leaf
696 current_leaves = body_leaves
697 if not (opening_bracket and closing_bracket and head_leaves):
698 # If there is no opening or closing_bracket that means the split failed and
699 # all content is in the tail. Otherwise, if `head_leaves` are empty, it means
700 # the matching `opening_bracket` wasn't available on `line` anymore.
701 raise CannotSplit("No brackets found")
703 tail_leaves.reverse()
704 body_leaves.reverse()
705 head_leaves.reverse()
706 head = bracket_split_build_line(
707 head_leaves, line, opening_bracket, component=_BracketSplitComponent.head
709 body = bracket_split_build_line(
710 body_leaves, line, opening_bracket, component=_BracketSplitComponent.body
712 tail = bracket_split_build_line(
713 tail_leaves, line, opening_bracket, component=_BracketSplitComponent.tail
715 bracket_split_succeeded_or_raise(head, body, tail)
716 return RHSResult(head, body, tail, opening_bracket, closing_bracket)
719 def _maybe_split_omitting_optional_parens(
723 features: Collection[Feature] = (),
724 omit: Collection[LeafID] = (),
727 Feature.FORCE_OPTIONAL_PARENTHESES not in features
728 # the opening bracket is an optional paren
729 and rhs.opening_bracket.type == token.LPAR
730 and not rhs.opening_bracket.value
731 # the closing bracket is an optional paren
732 and rhs.closing_bracket.type == token.RPAR
733 and not rhs.closing_bracket.value
734 # it's not an import (optional parens are the only thing we can split on
735 # in this case; attempting a split without them is a waste of time)
736 and not line.is_import
737 # there are no standalone comments in the body
738 and not rhs.body.contains_standalone_comments(0)
739 # and we can actually remove the parens
740 and can_omit_invisible_parens(rhs, mode.line_length)
742 omit = {id(rhs.closing_bracket), *omit}
744 # The RHSResult Omitting Optional Parens.
745 rhs_oop = _first_right_hand_split(line, omit=omit)
747 Preview.prefer_splitting_right_hand_side_of_assignments in line.mode
748 # the split is right after `=`
749 and len(rhs.head.leaves) >= 2
750 and rhs.head.leaves[-2].type == token.EQUAL
751 # the left side of assignment contains brackets
752 and any(leaf.type in BRACKETS for leaf in rhs.head.leaves[:-1])
753 # the left side of assignment is short enough (the -1 is for the ending
755 and is_line_short_enough(
756 rhs.head, mode=replace(mode, line_length=mode.line_length - 1)
758 # the left side of assignment won't explode further because of magic
760 and rhs.head.magic_trailing_comma is None
761 # the split by omitting optional parens isn't preferred by some other
763 and not _prefer_split_rhs_oop(rhs_oop, mode)
765 yield from _maybe_split_omitting_optional_parens(
766 rhs_oop, line, mode, features=features, omit=omit
770 except CannotSplit as e:
772 can_be_split(rhs.body) or is_line_short_enough(rhs.body, mode=mode)
775 "Splitting failed, body is still too long and can't be split."
779 rhs.head.contains_multiline_strings()
780 or rhs.tail.contains_multiline_strings()
783 "The current optional pair of parentheses is bound to fail to"
784 " satisfy the splitting algorithm because the head or the tail"
785 " contains multiline strings which by definition never fit one"
789 ensure_visible(rhs.opening_bracket)
790 ensure_visible(rhs.closing_bracket)
791 for result in (rhs.head, rhs.body, rhs.tail):
796 def _prefer_split_rhs_oop(rhs_oop: RHSResult, mode: Mode) -> bool:
798 Returns whether we should prefer the result from a split omitting optional parens.
800 has_closing_bracket_after_assign = False
801 for leaf in reversed(rhs_oop.head.leaves):
802 if leaf.type == token.EQUAL:
804 if leaf.type in CLOSING_BRACKETS:
805 has_closing_bracket_after_assign = True
808 # contains matching brackets after the `=` (done by checking there is a
810 has_closing_bracket_after_assign
812 # the split is actually from inside the optional parens (done by checking
813 # the first line still contains the `=`)
814 any(leaf.type == token.EQUAL for leaf in rhs_oop.head.leaves)
815 # the first line is short enough
816 and is_line_short_enough(rhs_oop.head, mode=mode)
818 # contains unsplittable type ignore
819 or rhs_oop.head.contains_unsplittable_type_ignore()
820 or rhs_oop.body.contains_unsplittable_type_ignore()
821 or rhs_oop.tail.contains_unsplittable_type_ignore()
825 def bracket_split_succeeded_or_raise(head: Line, body: Line, tail: Line) -> None:
826 """Raise :exc:`CannotSplit` if the last left- or right-hand split failed.
828 Do nothing otherwise.
830 A left- or right-hand split is based on a pair of brackets. Content before
831 (and including) the opening bracket is left on one line, content inside the
832 brackets is put on a separate line, and finally content starting with and
833 following the closing bracket is put on a separate line.
835 Those are called `head`, `body`, and `tail`, respectively. If the split
836 produced the same line (all content in `head`) or ended up with an empty `body`
837 and the `tail` is just the closing bracket, then it's considered failed.
839 tail_len = len(str(tail).strip())
842 raise CannotSplit("Splitting brackets produced the same line")
846 f"Splitting brackets on an empty body to save {tail_len} characters is"
851 def bracket_split_build_line(
854 opening_bracket: Leaf,
856 component: _BracketSplitComponent,
858 """Return a new line with given `leaves` and respective comments from `original`.
860 If it's the head component, brackets will be tracked so trailing commas are
863 If it's the body component, the result line is one-indented inside brackets and as
864 such has its first leaf's prefix normalized and a trailing comma added when
867 result = Line(mode=original.mode, depth=original.depth)
868 if component is _BracketSplitComponent.body:
869 result.inside_brackets = True
872 # Since body is a new indent level, remove spurious leading whitespace.
873 normalize_prefix(leaves[0], inside_brackets=True)
874 # Ensure a trailing comma for imports and standalone function arguments, but
875 # be careful not to add one after any comments or within type annotations.
878 and opening_bracket.value == "("
879 and not any(leaf.type == token.COMMA for leaf in leaves)
880 # In particular, don't add one within a parenthesized return annotation.
881 # Unfortunately the indicator we're in a return annotation (RARROW) may
882 # be defined directly in the parent node, the parent of the parent ...
883 # and so on depending on how complex the return annotation is.
884 # This isn't perfect and there's some false negatives but they are in
885 # contexts were a comma is actually fine.
887 node.prev_sibling.type == RARROW
890 getattr(leaves[0].parent, "parent", None),
892 if isinstance(node, Node) and isinstance(node.prev_sibling, Leaf)
896 if original.is_import or no_commas:
897 for i in range(len(leaves) - 1, -1, -1):
898 if leaves[i].type == STANDALONE_COMMENT:
901 if leaves[i].type != token.COMMA:
902 new_comma = Leaf(token.COMMA, ",")
903 leaves.insert(i + 1, new_comma)
906 leaves_to_track: Set[LeafID] = set()
907 if component is _BracketSplitComponent.head:
908 leaves_to_track = get_leaves_inside_matching_brackets(leaves)
914 track_bracket=id(leaf) in leaves_to_track,
916 for comment_after in original.comments_after(leaf):
917 result.append(comment_after, preformatted=True)
918 if component is _BracketSplitComponent.body and should_split_line(
919 result, opening_bracket
921 result.should_split_rhs = True
925 def dont_increase_indentation(split_func: Transformer) -> Transformer:
926 """Normalize prefix of the first leaf in every line returned by `split_func`.
928 This is a decorator over relevant split functions.
933 line: Line, features: Collection[Feature], mode: Mode
935 for split_line in split_func(line, features, mode):
936 normalize_prefix(split_line.leaves[0], inside_brackets=True)
942 def _get_last_non_comment_leaf(line: Line) -> Optional[int]:
943 for leaf_idx in range(len(line.leaves) - 1, 0, -1):
944 if line.leaves[leaf_idx].type != STANDALONE_COMMENT:
949 def _safe_add_trailing_comma(safe: bool, delimiter_priority: int, line: Line) -> Line:
952 and delimiter_priority == COMMA_PRIORITY
953 and line.leaves[-1].type != token.COMMA
954 and line.leaves[-1].type != STANDALONE_COMMENT
956 new_comma = Leaf(token.COMMA, ",")
957 line.append(new_comma)
961 @dont_increase_indentation
963 line: Line, features: Collection[Feature], mode: Mode
965 """Split according to delimiters of the highest priority.
967 If the appropriate Features are given, the split will add trailing commas
968 also in function signatures and calls that contain `*` and `**`.
971 last_leaf = line.leaves[-1]
973 raise CannotSplit("Line empty") from None
975 bt = line.bracket_tracker
977 delimiter_priority = bt.max_delimiter_priority(exclude={id(last_leaf)})
979 raise CannotSplit("No delimiters found") from None
981 if delimiter_priority == DOT_PRIORITY:
982 if bt.delimiter_count_with_priority(delimiter_priority) == 1:
983 raise CannotSplit("Splitting a single attribute from its owner looks wrong")
986 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
988 lowest_depth = sys.maxsize
989 trailing_comma_safe = True
991 def append_to_line(leaf: Leaf) -> Iterator[Line]:
992 """Append `leaf` to current line or to new line if appending impossible."""
993 nonlocal current_line
995 current_line.append_safe(leaf, preformatted=True)
1000 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1002 current_line.append(leaf)
1004 last_non_comment_leaf = _get_last_non_comment_leaf(line)
1005 for leaf_idx, leaf in enumerate(line.leaves):
1006 yield from append_to_line(leaf)
1008 for comment_after in line.comments_after(leaf):
1009 yield from append_to_line(comment_after)
1011 lowest_depth = min(lowest_depth, leaf.bracket_depth)
1012 if leaf.bracket_depth == lowest_depth:
1013 if is_vararg(leaf, within={syms.typedargslist}):
1014 trailing_comma_safe = (
1015 trailing_comma_safe and Feature.TRAILING_COMMA_IN_DEF in features
1017 elif is_vararg(leaf, within={syms.arglist, syms.argument}):
1018 trailing_comma_safe = (
1019 trailing_comma_safe and Feature.TRAILING_COMMA_IN_CALL in features
1023 Preview.add_trailing_comma_consistently in mode
1024 and last_leaf.type == STANDALONE_COMMENT
1025 and leaf_idx == last_non_comment_leaf
1027 current_line = _safe_add_trailing_comma(
1028 trailing_comma_safe, delimiter_priority, current_line
1031 leaf_priority = bt.delimiters.get(id(leaf))
1032 if leaf_priority == delimiter_priority:
1035 current_line = Line(
1036 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1039 current_line = _safe_add_trailing_comma(
1040 trailing_comma_safe, delimiter_priority, current_line
1045 @dont_increase_indentation
1046 def standalone_comment_split(
1047 line: Line, features: Collection[Feature], mode: Mode
1048 ) -> Iterator[Line]:
1049 """Split standalone comments from the rest of the line."""
1050 if not line.contains_standalone_comments(0):
1051 raise CannotSplit("Line does not have any standalone comments")
1053 current_line = Line(
1054 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1057 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1058 """Append `leaf` to current line or to new line if appending impossible."""
1059 nonlocal current_line
1061 current_line.append_safe(leaf, preformatted=True)
1065 current_line = Line(
1066 line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1068 current_line.append(leaf)
1070 for leaf in line.leaves:
1071 yield from append_to_line(leaf)
1073 for comment_after in line.comments_after(leaf):
1074 yield from append_to_line(comment_after)
1080 def normalize_prefix(leaf: Leaf, *, inside_brackets: bool) -> None:
1081 """Leave existing extra newlines if not `inside_brackets`. Remove everything
1084 Note: don't use backslashes for formatting or you'll lose your voting rights.
1086 if not inside_brackets:
1087 spl = leaf.prefix.split("#")
1088 if "\\" not in spl[0]:
1089 nl_count = spl[-1].count("\n")
1092 leaf.prefix = "\n" * nl_count
1098 def normalize_invisible_parens(
1099 node: Node, parens_after: Set[str], *, mode: Mode, features: Collection[Feature]
1101 """Make existing optional parentheses invisible or create new ones.
1103 `parens_after` is a set of string leaf values immediately after which parens
1106 Standardizes on visible parentheses for single-element tuples, and keeps
1107 existing visible parentheses for other tuples and generator expressions.
1109 for pc in list_comments(node.prefix, is_endmarker=False):
1110 if pc.value in FMT_OFF:
1111 # This `node` has a prefix with `# fmt: off`, don't mess with parens.
1114 # The multiple context managers grammar has a different pattern, thus this is
1115 # separate from the for-loop below. This possibly wraps them in invisible parens,
1116 # and later will be removed in remove_with_parens when needed.
1117 if node.type == syms.with_stmt:
1118 _maybe_wrap_cms_in_parens(node, mode, features)
1121 for index, child in enumerate(list(node.children)):
1122 # Fixes a bug where invisible parens are not properly stripped from
1123 # assignment statements that contain type annotations.
1124 if isinstance(child, Node) and child.type == syms.annassign:
1125 normalize_invisible_parens(
1126 child, parens_after=parens_after, mode=mode, features=features
1129 # Add parentheses around long tuple unpacking in assignments.
1132 and isinstance(child, Node)
1133 and child.type == syms.testlist_star_expr
1139 child.type == syms.atom
1140 and node.type == syms.for_stmt
1141 and isinstance(child.prev_sibling, Leaf)
1142 and child.prev_sibling.type == token.NAME
1143 and child.prev_sibling.value == "for"
1145 if maybe_make_parens_invisible_in_atom(
1148 remove_brackets_around_comma=True,
1150 wrap_in_parentheses(node, child, visible=False)
1151 elif isinstance(child, Node) and node.type == syms.with_stmt:
1152 remove_with_parens(child, node)
1153 elif child.type == syms.atom:
1154 if maybe_make_parens_invisible_in_atom(
1158 wrap_in_parentheses(node, child, visible=False)
1159 elif is_one_tuple(child):
1160 wrap_in_parentheses(node, child, visible=True)
1161 elif node.type == syms.import_from:
1162 _normalize_import_from(node, child, index)
1166 and child.type == token.STAR
1167 and node.type == syms.except_clause
1169 # In except* (PEP 654), the star is actually part of
1170 # of the keyword. So we need to skip the insertion of
1171 # invisible parentheses to work more precisely.
1174 elif not (isinstance(child, Leaf) and is_multiline_string(child)):
1175 wrap_in_parentheses(node, child, visible=False)
1177 comma_check = child.type == token.COMMA
1179 check_lpar = isinstance(child, Leaf) and (
1180 child.value in parens_after or comma_check
1184 def _normalize_import_from(parent: Node, child: LN, index: int) -> None:
1185 # "import from" nodes store parentheses directly as part of
1187 if is_lpar_token(child):
1188 assert is_rpar_token(parent.children[-1])
1189 # make parentheses invisible
1191 parent.children[-1].value = ""
1192 elif child.type != token.STAR:
1193 # insert invisible parentheses
1194 parent.insert_child(index, Leaf(token.LPAR, ""))
1195 parent.append_child(Leaf(token.RPAR, ""))
1198 def remove_await_parens(node: Node) -> None:
1199 if node.children[0].type == token.AWAIT and len(node.children) > 1:
1201 node.children[1].type == syms.atom
1202 and node.children[1].children[0].type == token.LPAR
1204 if maybe_make_parens_invisible_in_atom(
1207 remove_brackets_around_comma=True,
1209 wrap_in_parentheses(node, node.children[1], visible=False)
1211 # Since await is an expression we shouldn't remove
1212 # brackets in cases where this would change
1213 # the AST due to operator precedence.
1214 # Therefore we only aim to remove brackets around
1215 # power nodes that aren't also await expressions themselves.
1216 # https://peps.python.org/pep-0492/#updated-operator-precedence-table
1217 # N.B. We've still removed any redundant nested brackets though :)
1218 opening_bracket = cast(Leaf, node.children[1].children[0])
1219 closing_bracket = cast(Leaf, node.children[1].children[-1])
1220 bracket_contents = node.children[1].children[1]
1221 if isinstance(bracket_contents, Node):
1222 if bracket_contents.type != syms.power:
1223 ensure_visible(opening_bracket)
1224 ensure_visible(closing_bracket)
1226 bracket_contents.type == syms.power
1227 and bracket_contents.children[0].type == token.AWAIT
1229 ensure_visible(opening_bracket)
1230 ensure_visible(closing_bracket)
1231 # If we are in a nested await then recurse down.
1232 remove_await_parens(bracket_contents)
1235 def _maybe_wrap_cms_in_parens(
1236 node: Node, mode: Mode, features: Collection[Feature]
1238 """When enabled and safe, wrap the multiple context managers in invisible parens.
1240 It is only safe when `features` contain Feature.PARENTHESIZED_CONTEXT_MANAGERS.
1243 Feature.PARENTHESIZED_CONTEXT_MANAGERS not in features
1244 or Preview.wrap_multiple_context_managers_in_parens not in mode
1245 or len(node.children) <= 2
1246 # If it's an atom, it's already wrapped in parens.
1247 or node.children[1].type == syms.atom
1250 colon_index: Optional[int] = None
1251 for i in range(2, len(node.children)):
1252 if node.children[i].type == token.COLON:
1255 if colon_index is not None:
1256 lpar = Leaf(token.LPAR, "")
1257 rpar = Leaf(token.RPAR, "")
1258 context_managers = node.children[1:colon_index]
1259 for child in context_managers:
1261 # After wrapping, the with_stmt will look like this:
1267 # ... <-- context_managers
1273 syms.atom, [lpar, Node(syms.testlist_gexp, context_managers), rpar]
1275 node.insert_child(1, new_child)
1278 def remove_with_parens(node: Node, parent: Node) -> None:
1279 """Recursively hide optional parens in `with` statements."""
1280 # Removing all unnecessary parentheses in with statements in one pass is a tad
1281 # complex as different variations of bracketed statements result in pretty
1282 # different parse trees:
1284 # with (open("file")) as f: # this is an asexpr_test
1287 # with (open("file") as f): # this is an atom containing an
1290 # with (open("file")) as f, (open("file")) as f: # this is asexpr_test, COMMA,
1293 # with (open("file") as f, open("file") as f): # an atom containing a
1294 # ... # testlist_gexp which then
1295 # # contains multiple asexpr_test(s)
1296 if node.type == syms.atom:
1297 if maybe_make_parens_invisible_in_atom(
1300 remove_brackets_around_comma=True,
1302 wrap_in_parentheses(parent, node, visible=False)
1303 if isinstance(node.children[1], Node):
1304 remove_with_parens(node.children[1], node)
1305 elif node.type == syms.testlist_gexp:
1306 for child in node.children:
1307 if isinstance(child, Node):
1308 remove_with_parens(child, node)
1309 elif node.type == syms.asexpr_test and not any(
1310 leaf.type == token.COLONEQUAL for leaf in node.leaves()
1312 if maybe_make_parens_invisible_in_atom(
1315 remove_brackets_around_comma=True,
1317 wrap_in_parentheses(node, node.children[0], visible=False)
1320 def maybe_make_parens_invisible_in_atom(
1323 remove_brackets_around_comma: bool = False,
1325 """If it's safe, make the parens in the atom `node` invisible, recursively.
1326 Additionally, remove repeated, adjacent invisible parens from the atom `node`
1327 as they are redundant.
1329 Returns whether the node should itself be wrapped in invisible parentheses.
1332 node.type != syms.atom
1333 or is_empty_tuple(node)
1334 or is_one_tuple(node)
1335 or (is_yield(node) and parent.type != syms.expr_stmt)
1337 # This condition tries to prevent removing non-optional brackets
1338 # around a tuple, however, can be a bit overzealous so we provide
1339 # and option to skip this check for `for` and `with` statements.
1340 not remove_brackets_around_comma
1341 and max_delimiter_priority_in_atom(node) >= COMMA_PRIORITY
1343 or is_tuple_containing_walrus(node)
1347 if is_walrus_assignment(node):
1356 # these ones aren't useful to end users, but they do please fuzzers
1363 first = node.children[0]
1364 last = node.children[-1]
1365 if is_lpar_token(first) and is_rpar_token(last):
1366 middle = node.children[1]
1367 # make parentheses invisible
1370 maybe_make_parens_invisible_in_atom(
1373 remove_brackets_around_comma=remove_brackets_around_comma,
1376 if is_atom_with_invisible_parens(middle):
1377 # Strip the invisible parens from `middle` by replacing
1378 # it with the child in-between the invisible parens
1379 middle.replace(middle.children[1])
1386 def should_split_line(line: Line, opening_bracket: Leaf) -> bool:
1387 """Should `line` be immediately split with `delimiter_split()` after RHS?"""
1389 if not (opening_bracket.parent and opening_bracket.value in "[{("):
1392 # We're essentially checking if the body is delimited by commas and there's more
1393 # than one of them (we're excluding the trailing comma and if the delimiter priority
1394 # is still commas, that means there's more).
1396 trailing_comma = False
1398 last_leaf = line.leaves[-1]
1399 if last_leaf.type == token.COMMA:
1400 trailing_comma = True
1401 exclude.add(id(last_leaf))
1402 max_priority = line.bracket_tracker.max_delimiter_priority(exclude=exclude)
1403 except (IndexError, ValueError):
1406 return max_priority == COMMA_PRIORITY and (
1407 (line.mode.magic_trailing_comma and trailing_comma)
1408 # always explode imports
1409 or opening_bracket.parent.type in {syms.atom, syms.import_from}
1413 def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[LeafID]]:
1414 """Generate sets of closing bracket IDs that should be omitted in a RHS.
1416 Brackets can be omitted if the entire trailer up to and including
1417 a preceding closing bracket fits in one line.
1419 Yielded sets are cumulative (contain results of previous yields, too). First
1420 set is empty, unless the line should explode, in which case bracket pairs until
1421 the one that needs to explode are omitted.
1424 omit: Set[LeafID] = set()
1425 if not line.magic_trailing_comma:
1428 length = 4 * line.depth
1429 opening_bracket: Optional[Leaf] = None
1430 closing_bracket: Optional[Leaf] = None
1431 inner_brackets: Set[LeafID] = set()
1432 for index, leaf, leaf_length in line.enumerate_with_length(reversed=True):
1433 length += leaf_length
1434 if length > line_length:
1437 has_inline_comment = leaf_length > len(leaf.value) + len(leaf.prefix)
1438 if leaf.type == STANDALONE_COMMENT or has_inline_comment:
1442 if leaf is opening_bracket:
1443 opening_bracket = None
1444 elif leaf.type in CLOSING_BRACKETS:
1445 prev = line.leaves[index - 1] if index > 0 else None
1448 and prev.type == token.COMMA
1449 and leaf.opening_bracket is not None
1450 and not is_one_sequence_between(
1451 leaf.opening_bracket, leaf, line.leaves
1454 # Never omit bracket pairs with trailing commas.
1455 # We need to explode on those.
1458 inner_brackets.add(id(leaf))
1459 elif leaf.type in CLOSING_BRACKETS:
1460 prev = line.leaves[index - 1] if index > 0 else None
1461 if prev and prev.type in OPENING_BRACKETS:
1462 # Empty brackets would fail a split so treat them as "inner"
1463 # brackets (e.g. only add them to the `omit` set if another
1464 # pair of brackets was good enough.
1465 inner_brackets.add(id(leaf))
1469 omit.add(id(closing_bracket))
1470 omit.update(inner_brackets)
1471 inner_brackets.clear()
1476 and prev.type == token.COMMA
1477 and leaf.opening_bracket is not None
1478 and not is_one_sequence_between(leaf.opening_bracket, leaf, line.leaves)
1480 # Never omit bracket pairs with trailing commas.
1481 # We need to explode on those.
1485 opening_bracket = leaf.opening_bracket
1486 closing_bracket = leaf
1489 def run_transformer(
1491 transform: Transformer,
1493 features: Collection[Feature],
1498 line_str = line_to_string(line)
1499 result: List[Line] = []
1500 for transformed_line in transform(line, features, mode):
1501 if str(transformed_line).strip("\n") == line_str:
1502 raise CannotTransform("Line transformer returned an unchanged result")
1504 result.extend(transform_line(transformed_line, mode=mode, features=features))
1506 features_set = set(features)
1508 Feature.FORCE_OPTIONAL_PARENTHESES in features_set
1509 or transform.__class__.__name__ != "rhs"
1510 or not line.bracket_tracker.invisible
1511 or any(bracket.value for bracket in line.bracket_tracker.invisible)
1512 or line.contains_multiline_strings()
1513 or result[0].contains_uncollapsable_type_comments()
1514 or result[0].contains_unsplittable_type_ignore()
1515 or is_line_short_enough(result[0], mode=mode)
1516 # If any leaves have no parents (which _can_ occur since
1517 # `transform(line)` potentially destroys the line's underlying node
1518 # structure), then we can't proceed. Doing so would cause the below
1519 # call to `append_leaves()` to fail.
1520 or any(leaf.parent is None for leaf in line.leaves)
1524 line_copy = line.clone()
1525 append_leaves(line_copy, line, line.leaves)
1526 features_fop = features_set | {Feature.FORCE_OPTIONAL_PARENTHESES}
1527 second_opinion = run_transformer(
1528 line_copy, transform, mode, features_fop, line_str=line_str
1530 if all(is_line_short_enough(ln, mode=mode) for ln in second_opinion):
1531 result = second_opinion