All patches and comments are welcome. Please squash your changes to logical
commits before using git-format-patch and git-send-email to
patches@git.madduck.net.
If you'd read over the Git project's submission guidelines and adhered to them,
I'd be especially grateful.
2 Generating lines of code.
5 from dataclasses import dataclass
6 from enum import Enum, auto
7 from functools import partial, wraps
8 from typing import Collection, Iterator, List, Optional, Set, Union, cast
10 from black.brackets import (
13 get_leaves_inside_matching_brackets,
14 max_delimiter_priority_in_atom,
16 from black.comments import FMT_OFF, generate_comments, list_comments
17 from black.lines import (
21 can_omit_invisible_parens,
25 from black.mode import Feature, Mode, Preview
26 from black.nodes import (
38 is_atom_with_invisible_parens,
44 is_one_sequence_between,
49 is_tuple_containing_walrus,
56 from black.numerics import normalize_numeric_literal
57 from black.strings import (
60 normalize_string_prefix,
61 normalize_string_quotes,
62 normalize_unicode_escape_sequences,
64 from black.trans import (
73 from blib2to3.pgen2 import token
74 from blib2to3.pytree import Leaf, Node
78 LN = Union[Leaf, Node]
81 class CannotSplit(CannotTransform):
82 """A readable split that fits the allotted line length is impossible."""
85 # This isn't a dataclass because @dataclass + Generic breaks mypyc.
86 # See also https://github.com/mypyc/mypyc/issues/827.
87 class LineGenerator(Visitor[Line]):
88 """Generates reformatted Line objects. Empty lines are not emitted.
90 Note: destroys the tree it's visiting by mutating prefixes of its leaves
91 in ways that will no longer stringify to valid Python code on the tree.
94 def __init__(self, mode: Mode, features: Collection[Feature]) -> None:
96 self.features = features
97 self.current_line: Line
100 def line(self, indent: int = 0) -> Iterator[Line]:
103 If the line is empty, only emit if it makes sense.
104 If the line is too long, split it first and then generate.
106 If any lines were generated, set up a new current_line.
108 if not self.current_line:
109 self.current_line.depth += indent
110 return # Line is empty, don't emit. Creating a new one unnecessary.
112 complete_line = self.current_line
113 self.current_line = Line(mode=self.mode, depth=complete_line.depth + indent)
116 def visit_default(self, node: LN) -> Iterator[Line]:
117 """Default `visit_*()` implementation. Recurses to children of `node`."""
118 if isinstance(node, Leaf):
119 any_open_brackets = self.current_line.bracket_tracker.any_open_brackets()
120 for comment in generate_comments(node, preview=self.mode.preview):
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 if Preview.annotation_parens not in self.mode:
225 yield from self.visit_stmt(node, keywords={"def"}, parens=set())
227 yield from self.line()
229 # Remove redundant brackets around return type annotation.
230 is_return_annotation = False
231 for child in node.children:
232 if child.type == token.RARROW:
233 is_return_annotation = True
234 elif is_return_annotation:
235 if child.type == syms.atom and child.children[0].type == token.LPAR:
236 if maybe_make_parens_invisible_in_atom(
239 remove_brackets_around_comma=False,
241 wrap_in_parentheses(node, child, visible=False)
243 wrap_in_parentheses(node, child, visible=False)
244 is_return_annotation = False
246 for child in node.children:
247 yield from self.visit(child)
249 def visit_match_case(self, node: Node) -> Iterator[Line]:
250 """Visit either a match or case statement."""
251 normalize_invisible_parens(
252 node, parens_after=set(), mode=self.mode, features=self.features
255 yield from self.line()
256 for child in node.children:
257 yield from self.visit(child)
259 def visit_suite(self, node: Node) -> Iterator[Line]:
261 if self.mode.is_pyi and is_stub_suite(node):
262 yield from self.visit(node.children[2])
264 yield from self.visit_default(node)
266 def visit_simple_stmt(self, node: Node) -> Iterator[Line]:
267 """Visit a statement without nested statements."""
268 prev_type: Optional[int] = None
269 for child in node.children:
270 if (prev_type is None or prev_type == token.SEMI) and is_arith_like(child):
271 wrap_in_parentheses(node, child, visible=False)
272 prev_type = child.type
274 is_suite_like = node.parent and node.parent.type in STATEMENT
276 if self.mode.is_pyi and is_stub_body(node):
277 yield from self.visit_default(node)
279 yield from self.line(+1)
280 yield from self.visit_default(node)
281 yield from self.line(-1)
287 or not is_stub_suite(node.parent)
289 yield from self.line()
290 yield from self.visit_default(node)
292 def visit_async_stmt(self, node: Node) -> Iterator[Line]:
293 """Visit `async def`, `async for`, `async with`."""
294 yield from self.line()
296 children = iter(node.children)
297 for child in children:
298 yield from self.visit(child)
300 if child.type == token.ASYNC or child.type == STANDALONE_COMMENT:
301 # STANDALONE_COMMENT happens when `# fmt: skip` is applied on the async
305 internal_stmt = next(children)
306 for child in internal_stmt.children:
307 yield from self.visit(child)
309 def visit_decorators(self, node: Node) -> Iterator[Line]:
310 """Visit decorators."""
311 for child in node.children:
312 yield from self.line()
313 yield from self.visit(child)
315 def visit_power(self, node: Node) -> Iterator[Line]:
316 for idx, leaf in enumerate(node.children[:-1]):
317 next_leaf = node.children[idx + 1]
319 if not isinstance(leaf, Leaf):
322 value = leaf.value.lower()
324 leaf.type == token.NUMBER
325 and next_leaf.type == syms.trailer
326 # Ensure that we are in an attribute trailer
327 and next_leaf.children[0].type == token.DOT
328 # It shouldn't wrap hexadecimal, binary and octal literals
329 and not value.startswith(("0x", "0b", "0o"))
330 # It shouldn't wrap complex literals
333 wrap_in_parentheses(node, leaf)
335 if Preview.remove_redundant_parens in self.mode:
336 remove_await_parens(node)
338 yield from self.visit_default(node)
340 def visit_SEMI(self, leaf: Leaf) -> Iterator[Line]:
341 """Remove a semicolon and put the other statement on a separate line."""
342 yield from self.line()
344 def visit_ENDMARKER(self, leaf: Leaf) -> Iterator[Line]:
345 """End of file. Process outstanding comments and end with a newline."""
346 yield from self.visit_default(leaf)
347 yield from self.line()
349 def visit_STANDALONE_COMMENT(self, leaf: Leaf) -> Iterator[Line]:
350 if not self.current_line.bracket_tracker.any_open_brackets():
351 yield from self.line()
352 yield from self.visit_default(leaf)
354 def visit_factor(self, node: Node) -> Iterator[Line]:
355 """Force parentheses between a unary op and a binary power:
359 _operator, operand = node.children
361 operand.type == syms.power
362 and len(operand.children) == 3
363 and operand.children[1].type == token.DOUBLESTAR
365 lpar = Leaf(token.LPAR, "(")
366 rpar = Leaf(token.RPAR, ")")
367 index = operand.remove() or 0
368 node.insert_child(index, Node(syms.atom, [lpar, operand, rpar]))
369 yield from self.visit_default(node)
371 def visit_STRING(self, leaf: Leaf) -> Iterator[Line]:
372 if Preview.hex_codes_in_unicode_sequences in self.mode:
373 normalize_unicode_escape_sequences(leaf)
375 if is_docstring(leaf) and "\\\n" not in leaf.value:
376 # We're ignoring docstrings with backslash newline escapes because changing
377 # indentation of those changes the AST representation of the code.
378 if Preview.normalize_docstring_quotes_and_prefixes_properly in self.mode:
379 # There was a bug where --skip-string-normalization wouldn't stop us
380 # from normalizing docstring prefixes. To maintain stability, we can
381 # only address this buggy behaviour while the preview style is enabled.
382 if self.mode.string_normalization:
383 docstring = normalize_string_prefix(leaf.value)
384 # visit_default() does handle string normalization for us, but
385 # since this method acts differently depending on quote style (ex.
386 # see padding logic below), there's a possibility for unstable
387 # formatting as visit_default() is called *after*. To avoid a
388 # situation where this function formats a docstring differently on
389 # the second pass, normalize it early.
390 docstring = normalize_string_quotes(docstring)
392 docstring = leaf.value
394 # ... otherwise, we'll keep the buggy behaviour >.<
395 docstring = normalize_string_prefix(leaf.value)
396 prefix = get_string_prefix(docstring)
397 docstring = docstring[len(prefix) :] # Remove the prefix
398 quote_char = docstring[0]
399 # A natural way to remove the outer quotes is to do:
400 # docstring = docstring.strip(quote_char)
401 # but that breaks on """""x""" (which is '""x').
402 # So we actually need to remove the first character and the next two
403 # characters but only if they are the same as the first.
404 quote_len = 1 if docstring[1] != quote_char else 3
405 docstring = docstring[quote_len:-quote_len]
406 docstring_started_empty = not docstring
407 indent = " " * 4 * self.current_line.depth
409 if is_multiline_string(leaf):
410 docstring = fix_docstring(docstring, indent)
412 docstring = docstring.strip()
414 has_trailing_backslash = False
416 # Add some padding if the docstring starts / ends with a quote mark.
417 if docstring[0] == quote_char:
418 docstring = " " + docstring
419 if docstring[-1] == quote_char:
421 if docstring[-1] == "\\":
422 backslash_count = len(docstring) - len(docstring.rstrip("\\"))
423 if backslash_count % 2:
424 # Odd number of tailing backslashes, add some padding to
425 # avoid escaping the closing string quote.
427 has_trailing_backslash = True
428 elif not docstring_started_empty:
431 # We could enforce triple quotes at this point.
432 quote = quote_char * quote_len
434 # It's invalid to put closing single-character quotes on a new line.
435 if Preview.long_docstring_quotes_on_newline in self.mode and quote_len == 3:
436 # We need to find the length of the last line of the docstring
437 # to find if we can add the closing quotes to the line without
438 # exceeding the maximum line length.
439 # If docstring is one line, we don't put the closing quotes on a
440 # separate line because it looks ugly (#3320).
441 lines = docstring.splitlines()
442 last_line_length = len(lines[-1]) if docstring else 0
444 # If adding closing quotes would cause the last line to exceed
445 # the maximum line length then put a line break before the
449 and last_line_length + quote_len > self.mode.line_length
450 and len(indent) + quote_len <= self.mode.line_length
451 and not has_trailing_backslash
453 leaf.value = prefix + quote + docstring + "\n" + indent + quote
455 leaf.value = prefix + quote + docstring + quote
457 leaf.value = prefix + quote + docstring + quote
459 yield from self.visit_default(leaf)
461 def __post_init__(self) -> None:
462 """You are in a twisty little maze of passages."""
463 self.current_line = Line(mode=self.mode)
467 self.visit_assert_stmt = partial(v, keywords={"assert"}, parens={"assert", ","})
468 self.visit_if_stmt = partial(
469 v, keywords={"if", "else", "elif"}, parens={"if", "elif"}
471 self.visit_while_stmt = partial(v, keywords={"while", "else"}, parens={"while"})
472 self.visit_for_stmt = partial(v, keywords={"for", "else"}, parens={"for", "in"})
473 self.visit_try_stmt = partial(
474 v, keywords={"try", "except", "else", "finally"}, parens=Ø
476 if self.mode.preview:
477 self.visit_except_clause = partial(
478 v, keywords={"except"}, parens={"except"}
480 self.visit_with_stmt = partial(v, keywords={"with"}, parens={"with"})
482 self.visit_except_clause = partial(v, keywords={"except"}, parens=Ø)
483 self.visit_with_stmt = partial(v, keywords={"with"}, parens=Ø)
484 self.visit_classdef = partial(v, keywords={"class"}, parens=Ø)
485 self.visit_expr_stmt = partial(v, keywords=Ø, parens=ASSIGNMENTS)
486 self.visit_return_stmt = partial(v, keywords={"return"}, parens={"return"})
487 self.visit_import_from = partial(v, keywords=Ø, parens={"import"})
488 self.visit_del_stmt = partial(v, keywords=Ø, parens={"del"})
489 self.visit_async_funcdef = self.visit_async_stmt
490 self.visit_decorated = self.visit_decorators
493 self.visit_match_stmt = self.visit_match_case
494 self.visit_case_block = self.visit_match_case
498 line: Line, mode: Mode, features: Collection[Feature] = ()
500 """Transform a `line`, potentially splitting it into many lines.
502 They should fit in the allotted `line_length` but might not be able to.
504 `features` are syntactical features that may be used in the output.
510 line_str = line_to_string(line)
512 ll = mode.line_length
513 sn = mode.string_normalization
514 string_merge = StringMerger(ll, sn)
515 string_paren_strip = StringParenStripper(ll, sn)
516 string_split = StringSplitter(ll, sn)
517 string_paren_wrap = StringParenWrapper(ll, sn)
519 transformers: List[Transformer]
521 not line.contains_uncollapsable_type_comments()
522 and not line.should_split_rhs
523 and not line.magic_trailing_comma
525 is_line_short_enough(line, line_length=mode.line_length, line_str=line_str)
526 or line.contains_unsplittable_type_ignore()
528 and not (line.inside_brackets and line.contains_standalone_comments())
530 # Only apply basic string preprocessing, since lines shouldn't be split here.
531 if Preview.string_processing in mode:
532 transformers = [string_merge, string_paren_strip]
536 transformers = [left_hand_split]
540 self: object, line: Line, features: Collection[Feature]
542 """Wraps calls to `right_hand_split`.
544 The calls increasingly `omit` right-hand trailers (bracket pairs with
545 content), meaning the trailers get glued together to split on another
546 bracket pair instead.
548 for omit in generate_trailers_to_omit(line, mode.line_length):
550 right_hand_split(line, mode.line_length, features, omit=omit)
552 # Note: this check is only able to figure out if the first line of the
553 # *current* transformation fits in the line length. This is true only
554 # for simple cases. All others require running more transforms via
555 # `transform_line()`. This check doesn't know if those would succeed.
556 if is_line_short_enough(lines[0], line_length=mode.line_length):
560 # All splits failed, best effort split with no omits.
561 # This mostly happens to multiline strings that are by definition
562 # reported as not fitting a single line, as well as lines that contain
563 # trailing commas (those have to be exploded).
564 yield from right_hand_split(
565 line, line_length=mode.line_length, features=features
568 # HACK: nested functions (like _rhs) compiled by mypyc don't retain their
569 # __name__ attribute which is needed in `run_transformer` further down.
570 # Unfortunately a nested class breaks mypyc too. So a class must be created
571 # via type ... https://github.com/mypyc/mypyc/issues/884
572 rhs = type("rhs", (), {"__call__": _rhs})()
574 if Preview.string_processing in mode:
575 if line.inside_brackets:
581 standalone_comment_split,
594 if line.inside_brackets:
595 transformers = [delimiter_split, standalone_comment_split, rhs]
598 # It's always safe to attempt hugging of power operations and pretty much every line
600 transformers.append(hug_power_op)
602 for transform in transformers:
603 # We are accumulating lines in `result` because we might want to abort
604 # mission and return the original line in the end, or attempt a different
607 result = run_transformer(line, transform, mode, features, line_str=line_str)
608 except CannotTransform:
618 class _BracketSplitComponent(Enum):
624 def left_hand_split(line: Line, _features: Collection[Feature] = ()) -> Iterator[Line]:
625 """Split line into many lines, starting with the first matching bracket pair.
627 Note: this usually looks weird, only use this for function definitions.
628 Prefer RHS otherwise. This is why this function is not symmetrical with
629 :func:`right_hand_split` which also handles optional parentheses.
631 tail_leaves: List[Leaf] = []
632 body_leaves: List[Leaf] = []
633 head_leaves: List[Leaf] = []
634 current_leaves = head_leaves
635 matching_bracket: Optional[Leaf] = None
636 for leaf in line.leaves:
638 current_leaves is body_leaves
639 and leaf.type in CLOSING_BRACKETS
640 and leaf.opening_bracket is matching_bracket
641 and isinstance(matching_bracket, Leaf)
644 ensure_visible(matching_bracket)
645 current_leaves = tail_leaves if body_leaves else head_leaves
646 current_leaves.append(leaf)
647 if current_leaves is head_leaves:
648 if leaf.type in OPENING_BRACKETS:
649 matching_bracket = leaf
650 current_leaves = body_leaves
651 if not matching_bracket:
652 raise CannotSplit("No brackets found")
654 head = bracket_split_build_line(
655 head_leaves, line, matching_bracket, component=_BracketSplitComponent.head
657 body = bracket_split_build_line(
658 body_leaves, line, matching_bracket, component=_BracketSplitComponent.body
660 tail = bracket_split_build_line(
661 tail_leaves, line, matching_bracket, component=_BracketSplitComponent.tail
663 bracket_split_succeeded_or_raise(head, body, tail)
664 for result in (head, body, tail):
671 """Intermediate split result from a right hand split."""
676 opening_bracket: Leaf
677 closing_bracket: Leaf
680 def right_hand_split(
683 features: Collection[Feature] = (),
684 omit: Collection[LeafID] = (),
686 """Split line into many lines, starting with the last matching bracket pair.
688 If the split was by optional parentheses, attempt splitting without them, too.
689 `omit` is a collection of closing bracket IDs that shouldn't be considered for
692 Note: running this function modifies `bracket_depth` on the leaves of `line`.
694 rhs_result = _first_right_hand_split(line, omit=omit)
695 yield from _maybe_split_omitting_optional_parens(
696 rhs_result, line, line_length, features=features, omit=omit
700 def _first_right_hand_split(
702 omit: Collection[LeafID] = (),
704 """Split the line into head, body, tail starting with the last bracket pair.
706 Note: this function should not have side effects. It's relied upon by
707 _maybe_split_omitting_optional_parens to get an opinion whether to prefer
708 splitting on the right side of an assignment statement.
710 tail_leaves: List[Leaf] = []
711 body_leaves: List[Leaf] = []
712 head_leaves: List[Leaf] = []
713 current_leaves = tail_leaves
714 opening_bracket: Optional[Leaf] = None
715 closing_bracket: Optional[Leaf] = None
716 for leaf in reversed(line.leaves):
717 if current_leaves is body_leaves:
718 if leaf is opening_bracket:
719 current_leaves = head_leaves if body_leaves else tail_leaves
720 current_leaves.append(leaf)
721 if current_leaves is tail_leaves:
722 if leaf.type in CLOSING_BRACKETS and id(leaf) not in omit:
723 opening_bracket = leaf.opening_bracket
724 closing_bracket = leaf
725 current_leaves = body_leaves
726 if not (opening_bracket and closing_bracket and head_leaves):
727 # If there is no opening or closing_bracket that means the split failed and
728 # all content is in the tail. Otherwise, if `head_leaves` are empty, it means
729 # the matching `opening_bracket` wasn't available on `line` anymore.
730 raise CannotSplit("No brackets found")
732 tail_leaves.reverse()
733 body_leaves.reverse()
734 head_leaves.reverse()
735 head = bracket_split_build_line(
736 head_leaves, line, opening_bracket, component=_BracketSplitComponent.head
738 body = bracket_split_build_line(
739 body_leaves, line, opening_bracket, component=_BracketSplitComponent.body
741 tail = bracket_split_build_line(
742 tail_leaves, line, opening_bracket, component=_BracketSplitComponent.tail
744 bracket_split_succeeded_or_raise(head, body, tail)
745 return _RHSResult(head, body, tail, opening_bracket, closing_bracket)
748 def _maybe_split_omitting_optional_parens(
752 features: Collection[Feature] = (),
753 omit: Collection[LeafID] = (),
756 Feature.FORCE_OPTIONAL_PARENTHESES not in features
757 # the opening bracket is an optional paren
758 and rhs.opening_bracket.type == token.LPAR
759 and not rhs.opening_bracket.value
760 # the closing bracket is an optional paren
761 and rhs.closing_bracket.type == token.RPAR
762 and not rhs.closing_bracket.value
763 # it's not an import (optional parens are the only thing we can split on
764 # in this case; attempting a split without them is a waste of time)
765 and not line.is_import
766 # there are no standalone comments in the body
767 and not rhs.body.contains_standalone_comments(0)
768 # and we can actually remove the parens
769 and can_omit_invisible_parens(rhs.body, line_length)
771 omit = {id(rhs.closing_bracket), *omit}
773 # The _RHSResult Omitting Optional Parens.
774 rhs_oop = _first_right_hand_split(line, omit=omit)
776 Preview.prefer_splitting_right_hand_side_of_assignments in line.mode
777 # the split is right after `=`
778 and len(rhs.head.leaves) >= 2
779 and rhs.head.leaves[-2].type == token.EQUAL
780 # the left side of assignement contains brackets
781 and any(leaf.type in BRACKETS for leaf in rhs.head.leaves[:-1])
782 # the left side of assignment is short enough (the -1 is for the ending
784 and is_line_short_enough(rhs.head, line_length=line_length - 1)
785 # the left side of assignment won't explode further because of magic
787 and rhs.head.magic_trailing_comma is None
788 # the split by omitting optional parens isn't preferred by some other
790 and not _prefer_split_rhs_oop(rhs_oop, line_length=line_length)
792 yield from _maybe_split_omitting_optional_parens(
793 rhs_oop, line, line_length, features=features, omit=omit
797 except CannotSplit as e:
799 can_be_split(rhs.body)
800 or is_line_short_enough(rhs.body, line_length=line_length)
803 "Splitting failed, body is still too long and can't be split."
807 rhs.head.contains_multiline_strings()
808 or rhs.tail.contains_multiline_strings()
811 "The current optional pair of parentheses is bound to fail to"
812 " satisfy the splitting algorithm because the head or the tail"
813 " contains multiline strings which by definition never fit one"
817 ensure_visible(rhs.opening_bracket)
818 ensure_visible(rhs.closing_bracket)
819 for result in (rhs.head, rhs.body, rhs.tail):
824 def _prefer_split_rhs_oop(rhs_oop: _RHSResult, line_length: int) -> bool:
826 Returns whether we should prefer the result from a split omitting optional parens.
828 has_closing_bracket_after_assign = False
829 for leaf in reversed(rhs_oop.head.leaves):
830 if leaf.type == token.EQUAL:
832 if leaf.type in CLOSING_BRACKETS:
833 has_closing_bracket_after_assign = True
836 # contains matching brackets after the `=` (done by checking there is a
838 has_closing_bracket_after_assign
840 # the split is actually from inside the optional parens (done by checking
841 # the first line still contains the `=`)
842 any(leaf.type == token.EQUAL for leaf in rhs_oop.head.leaves)
843 # the first line is short enough
844 and is_line_short_enough(rhs_oop.head, line_length=line_length)
846 # contains unsplittable type ignore
847 or rhs_oop.head.contains_unsplittable_type_ignore()
848 or rhs_oop.body.contains_unsplittable_type_ignore()
849 or rhs_oop.tail.contains_unsplittable_type_ignore()
853 def bracket_split_succeeded_or_raise(head: Line, body: Line, tail: Line) -> None:
854 """Raise :exc:`CannotSplit` if the last left- or right-hand split failed.
856 Do nothing otherwise.
858 A left- or right-hand split is based on a pair of brackets. Content before
859 (and including) the opening bracket is left on one line, content inside the
860 brackets is put on a separate line, and finally content starting with and
861 following the closing bracket is put on a separate line.
863 Those are called `head`, `body`, and `tail`, respectively. If the split
864 produced the same line (all content in `head`) or ended up with an empty `body`
865 and the `tail` is just the closing bracket, then it's considered failed.
867 tail_len = len(str(tail).strip())
870 raise CannotSplit("Splitting brackets produced the same line")
874 f"Splitting brackets on an empty body to save {tail_len} characters is"
879 def bracket_split_build_line(
882 opening_bracket: Leaf,
884 component: _BracketSplitComponent,
886 """Return a new line with given `leaves` and respective comments from `original`.
888 If it's the head component, brackets will be tracked so trailing commas are
891 If it's the body component, the result line is one-indented inside brackets and as
892 such has its first leaf's prefix normalized and a trailing comma added when
895 result = Line(mode=original.mode, depth=original.depth)
896 if component is _BracketSplitComponent.body:
897 result.inside_brackets = True
900 # Since body is a new indent level, remove spurious leading whitespace.
901 normalize_prefix(leaves[0], inside_brackets=True)
902 # Ensure a trailing comma for imports and standalone function arguments, but
903 # be careful not to add one after any comments or within type annotations.
906 and opening_bracket.value == "("
907 and not any(leaf.type == token.COMMA for leaf in leaves)
908 # In particular, don't add one within a parenthesized return annotation.
909 # Unfortunately the indicator we're in a return annotation (RARROW) may
910 # be defined directly in the parent node, the parent of the parent ...
911 # and so on depending on how complex the return annotation is.
912 # This isn't perfect and there's some false negatives but they are in
913 # contexts were a comma is actually fine.
915 node.prev_sibling.type == RARROW
918 getattr(leaves[0].parent, "parent", None),
920 if isinstance(node, Node) and isinstance(node.prev_sibling, Leaf)
924 if original.is_import or no_commas:
925 for i in range(len(leaves) - 1, -1, -1):
926 if leaves[i].type == STANDALONE_COMMENT:
929 if leaves[i].type != token.COMMA:
930 new_comma = Leaf(token.COMMA, ",")
931 leaves.insert(i + 1, new_comma)
934 leaves_to_track: Set[LeafID] = set()
936 Preview.handle_trailing_commas_in_head in original.mode
937 and component is _BracketSplitComponent.head
939 leaves_to_track = get_leaves_inside_matching_brackets(leaves)
945 track_bracket=id(leaf) in leaves_to_track,
947 for comment_after in original.comments_after(leaf):
948 result.append(comment_after, preformatted=True)
949 if component is _BracketSplitComponent.body and should_split_line(
950 result, opening_bracket
952 result.should_split_rhs = True
956 def dont_increase_indentation(split_func: Transformer) -> Transformer:
957 """Normalize prefix of the first leaf in every line returned by `split_func`.
959 This is a decorator over relevant split functions.
963 def split_wrapper(line: Line, features: Collection[Feature] = ()) -> Iterator[Line]:
964 for split_line in split_func(line, features):
965 normalize_prefix(split_line.leaves[0], inside_brackets=True)
971 @dont_increase_indentation
972 def delimiter_split(line: Line, features: Collection[Feature] = ()) -> Iterator[Line]:
973 """Split according to delimiters of the highest priority.
975 If the appropriate Features are given, the split will add trailing commas
976 also in function signatures and calls that contain `*` and `**`.
979 last_leaf = line.leaves[-1]
981 raise CannotSplit("Line empty") from None
983 bt = line.bracket_tracker
985 delimiter_priority = bt.max_delimiter_priority(exclude={id(last_leaf)})
987 raise CannotSplit("No delimiters found") from None
989 if delimiter_priority == DOT_PRIORITY:
990 if bt.delimiter_count_with_priority(delimiter_priority) == 1:
991 raise CannotSplit("Splitting a single attribute from its owner looks wrong")
994 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
996 lowest_depth = sys.maxsize
997 trailing_comma_safe = True
999 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1000 """Append `leaf` to current line or to new line if appending impossible."""
1001 nonlocal current_line
1003 current_line.append_safe(leaf, preformatted=True)
1007 current_line = Line(
1008 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1010 current_line.append(leaf)
1012 for leaf in line.leaves:
1013 yield from append_to_line(leaf)
1015 for comment_after in line.comments_after(leaf):
1016 yield from append_to_line(comment_after)
1018 lowest_depth = min(lowest_depth, leaf.bracket_depth)
1019 if leaf.bracket_depth == lowest_depth:
1020 if is_vararg(leaf, within={syms.typedargslist}):
1021 trailing_comma_safe = (
1022 trailing_comma_safe and Feature.TRAILING_COMMA_IN_DEF in features
1024 elif is_vararg(leaf, within={syms.arglist, syms.argument}):
1025 trailing_comma_safe = (
1026 trailing_comma_safe and Feature.TRAILING_COMMA_IN_CALL in features
1029 leaf_priority = bt.delimiters.get(id(leaf))
1030 if leaf_priority == delimiter_priority:
1033 current_line = Line(
1034 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1039 and delimiter_priority == COMMA_PRIORITY
1040 and current_line.leaves[-1].type != token.COMMA
1041 and current_line.leaves[-1].type != STANDALONE_COMMENT
1043 new_comma = Leaf(token.COMMA, ",")
1044 current_line.append(new_comma)
1048 @dont_increase_indentation
1049 def standalone_comment_split(
1050 line: Line, features: Collection[Feature] = ()
1051 ) -> Iterator[Line]:
1052 """Split standalone comments from the rest of the line."""
1053 if not line.contains_standalone_comments(0):
1054 raise CannotSplit("Line does not have any standalone comments")
1056 current_line = Line(
1057 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1060 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1061 """Append `leaf` to current line or to new line if appending impossible."""
1062 nonlocal current_line
1064 current_line.append_safe(leaf, preformatted=True)
1068 current_line = Line(
1069 line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1071 current_line.append(leaf)
1073 for leaf in line.leaves:
1074 yield from append_to_line(leaf)
1076 for comment_after in line.comments_after(leaf):
1077 yield from append_to_line(comment_after)
1083 def normalize_prefix(leaf: Leaf, *, inside_brackets: bool) -> None:
1084 """Leave existing extra newlines if not `inside_brackets`. Remove everything
1087 Note: don't use backslashes for formatting or you'll lose your voting rights.
1089 if not inside_brackets:
1090 spl = leaf.prefix.split("#")
1091 if "\\" not in spl[0]:
1092 nl_count = spl[-1].count("\n")
1095 leaf.prefix = "\n" * nl_count
1101 def normalize_invisible_parens(
1102 node: Node, parens_after: Set[str], *, mode: Mode, features: Collection[Feature]
1104 """Make existing optional parentheses invisible or create new ones.
1106 `parens_after` is a set of string leaf values immediately after which parens
1109 Standardizes on visible parentheses for single-element tuples, and keeps
1110 existing visible parentheses for other tuples and generator expressions.
1112 for pc in list_comments(node.prefix, is_endmarker=False, preview=mode.preview):
1113 if pc.value in FMT_OFF:
1114 # This `node` has a prefix with `# fmt: off`, don't mess with parens.
1117 # The multiple context managers grammar has a different pattern, thus this is
1118 # separate from the for-loop below. This possibly wraps them in invisible parens,
1119 # and later will be removed in remove_with_parens when needed.
1120 if node.type == syms.with_stmt:
1121 _maybe_wrap_cms_in_parens(node, mode, features)
1124 for index, child in enumerate(list(node.children)):
1125 # Fixes a bug where invisible parens are not properly stripped from
1126 # assignment statements that contain type annotations.
1127 if isinstance(child, Node) and child.type == syms.annassign:
1128 normalize_invisible_parens(
1129 child, parens_after=parens_after, mode=mode, features=features
1132 # Add parentheses around long tuple unpacking in assignments.
1135 and isinstance(child, Node)
1136 and child.type == syms.testlist_star_expr
1143 and child.type == syms.atom
1144 and node.type == syms.for_stmt
1145 and isinstance(child.prev_sibling, Leaf)
1146 and child.prev_sibling.type == token.NAME
1147 and child.prev_sibling.value == "for"
1149 if maybe_make_parens_invisible_in_atom(
1152 remove_brackets_around_comma=True,
1154 wrap_in_parentheses(node, child, visible=False)
1156 mode.preview and isinstance(child, Node) and node.type == syms.with_stmt
1158 remove_with_parens(child, node)
1159 elif child.type == syms.atom:
1160 if maybe_make_parens_invisible_in_atom(
1164 wrap_in_parentheses(node, child, visible=False)
1165 elif is_one_tuple(child):
1166 wrap_in_parentheses(node, child, visible=True)
1167 elif node.type == syms.import_from:
1168 _normalize_import_from(node, child, index)
1172 and child.type == token.STAR
1173 and node.type == syms.except_clause
1175 # In except* (PEP 654), the star is actually part of
1176 # of the keyword. So we need to skip the insertion of
1177 # invisible parentheses to work more precisely.
1180 elif not (isinstance(child, Leaf) and is_multiline_string(child)):
1181 wrap_in_parentheses(node, child, visible=False)
1183 comma_check = child.type == token.COMMA if mode.preview else False
1185 check_lpar = isinstance(child, Leaf) and (
1186 child.value in parens_after or comma_check
1190 def _normalize_import_from(parent: Node, child: LN, index: int) -> None:
1191 # "import from" nodes store parentheses directly as part of
1193 if is_lpar_token(child):
1194 assert is_rpar_token(parent.children[-1])
1195 # make parentheses invisible
1197 parent.children[-1].value = ""
1198 elif child.type != token.STAR:
1199 # insert invisible parentheses
1200 parent.insert_child(index, Leaf(token.LPAR, ""))
1201 parent.append_child(Leaf(token.RPAR, ""))
1204 def remove_await_parens(node: Node) -> None:
1205 if node.children[0].type == token.AWAIT and len(node.children) > 1:
1207 node.children[1].type == syms.atom
1208 and node.children[1].children[0].type == token.LPAR
1210 if maybe_make_parens_invisible_in_atom(
1213 remove_brackets_around_comma=True,
1215 wrap_in_parentheses(node, node.children[1], visible=False)
1217 # Since await is an expression we shouldn't remove
1218 # brackets in cases where this would change
1219 # the AST due to operator precedence.
1220 # Therefore we only aim to remove brackets around
1221 # power nodes that aren't also await expressions themselves.
1222 # https://peps.python.org/pep-0492/#updated-operator-precedence-table
1223 # N.B. We've still removed any redundant nested brackets though :)
1224 opening_bracket = cast(Leaf, node.children[1].children[0])
1225 closing_bracket = cast(Leaf, node.children[1].children[-1])
1226 bracket_contents = node.children[1].children[1]
1227 if isinstance(bracket_contents, Node):
1228 if bracket_contents.type != syms.power:
1229 ensure_visible(opening_bracket)
1230 ensure_visible(closing_bracket)
1232 bracket_contents.type == syms.power
1233 and bracket_contents.children[0].type == token.AWAIT
1235 ensure_visible(opening_bracket)
1236 ensure_visible(closing_bracket)
1237 # If we are in a nested await then recurse down.
1238 remove_await_parens(bracket_contents)
1241 def _maybe_wrap_cms_in_parens(
1242 node: Node, mode: Mode, features: Collection[Feature]
1244 """When enabled and safe, wrap the multiple context managers in invisible parens.
1246 It is only safe when `features` contain Feature.PARENTHESIZED_CONTEXT_MANAGERS.
1249 Feature.PARENTHESIZED_CONTEXT_MANAGERS not in features
1250 or Preview.wrap_multiple_context_managers_in_parens not in mode
1251 or len(node.children) <= 2
1252 # If it's an atom, it's already wrapped in parens.
1253 or node.children[1].type == syms.atom
1256 colon_index: Optional[int] = None
1257 for i in range(2, len(node.children)):
1258 if node.children[i].type == token.COLON:
1261 if colon_index is not None:
1262 lpar = Leaf(token.LPAR, "")
1263 rpar = Leaf(token.RPAR, "")
1264 context_managers = node.children[1:colon_index]
1265 for child in context_managers:
1267 # After wrapping, the with_stmt will look like this:
1273 # ... <-- context_managers
1279 syms.atom, [lpar, Node(syms.testlist_gexp, context_managers), rpar]
1281 node.insert_child(1, new_child)
1284 def remove_with_parens(node: Node, parent: Node) -> None:
1285 """Recursively hide optional parens in `with` statements."""
1286 # Removing all unnecessary parentheses in with statements in one pass is a tad
1287 # complex as different variations of bracketed statements result in pretty
1288 # different parse trees:
1290 # with (open("file")) as f: # this is an asexpr_test
1293 # with (open("file") as f): # this is an atom containing an
1296 # with (open("file")) as f, (open("file")) as f: # this is asexpr_test, COMMA,
1299 # with (open("file") as f, open("file") as f): # an atom containing a
1300 # ... # testlist_gexp which then
1301 # # contains multiple asexpr_test(s)
1302 if node.type == syms.atom:
1303 if maybe_make_parens_invisible_in_atom(
1306 remove_brackets_around_comma=True,
1308 wrap_in_parentheses(parent, node, visible=False)
1309 if isinstance(node.children[1], Node):
1310 remove_with_parens(node.children[1], node)
1311 elif node.type == syms.testlist_gexp:
1312 for child in node.children:
1313 if isinstance(child, Node):
1314 remove_with_parens(child, node)
1315 elif node.type == syms.asexpr_test and not any(
1316 leaf.type == token.COLONEQUAL for leaf in node.leaves()
1318 if maybe_make_parens_invisible_in_atom(
1321 remove_brackets_around_comma=True,
1323 wrap_in_parentheses(node, node.children[0], visible=False)
1326 def maybe_make_parens_invisible_in_atom(
1329 remove_brackets_around_comma: bool = False,
1331 """If it's safe, make the parens in the atom `node` invisible, recursively.
1332 Additionally, remove repeated, adjacent invisible parens from the atom `node`
1333 as they are redundant.
1335 Returns whether the node should itself be wrapped in invisible parentheses.
1338 node.type != syms.atom
1339 or is_empty_tuple(node)
1340 or is_one_tuple(node)
1341 or (is_yield(node) and parent.type != syms.expr_stmt)
1343 # This condition tries to prevent removing non-optional brackets
1344 # around a tuple, however, can be a bit overzealous so we provide
1345 # and option to skip this check for `for` and `with` statements.
1346 not remove_brackets_around_comma
1347 and max_delimiter_priority_in_atom(node) >= COMMA_PRIORITY
1349 or is_tuple_containing_walrus(node)
1353 if is_walrus_assignment(node):
1362 # these ones aren't useful to end users, but they do please fuzzers
1369 first = node.children[0]
1370 last = node.children[-1]
1371 if is_lpar_token(first) and is_rpar_token(last):
1372 middle = node.children[1]
1373 # make parentheses invisible
1376 maybe_make_parens_invisible_in_atom(
1379 remove_brackets_around_comma=remove_brackets_around_comma,
1382 if is_atom_with_invisible_parens(middle):
1383 # Strip the invisible parens from `middle` by replacing
1384 # it with the child in-between the invisible parens
1385 middle.replace(middle.children[1])
1392 def should_split_line(line: Line, opening_bracket: Leaf) -> bool:
1393 """Should `line` be immediately split with `delimiter_split()` after RHS?"""
1395 if not (opening_bracket.parent and opening_bracket.value in "[{("):
1398 # We're essentially checking if the body is delimited by commas and there's more
1399 # than one of them (we're excluding the trailing comma and if the delimiter priority
1400 # is still commas, that means there's more).
1402 trailing_comma = False
1404 last_leaf = line.leaves[-1]
1405 if last_leaf.type == token.COMMA:
1406 trailing_comma = True
1407 exclude.add(id(last_leaf))
1408 max_priority = line.bracket_tracker.max_delimiter_priority(exclude=exclude)
1409 except (IndexError, ValueError):
1412 return max_priority == COMMA_PRIORITY and (
1413 (line.mode.magic_trailing_comma and trailing_comma)
1414 # always explode imports
1415 or opening_bracket.parent.type in {syms.atom, syms.import_from}
1419 def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[LeafID]]:
1420 """Generate sets of closing bracket IDs that should be omitted in a RHS.
1422 Brackets can be omitted if the entire trailer up to and including
1423 a preceding closing bracket fits in one line.
1425 Yielded sets are cumulative (contain results of previous yields, too). First
1426 set is empty, unless the line should explode, in which case bracket pairs until
1427 the one that needs to explode are omitted.
1430 omit: Set[LeafID] = set()
1431 if not line.magic_trailing_comma:
1434 length = 4 * line.depth
1435 opening_bracket: Optional[Leaf] = None
1436 closing_bracket: Optional[Leaf] = None
1437 inner_brackets: Set[LeafID] = set()
1438 for index, leaf, leaf_length in line.enumerate_with_length(reversed=True):
1439 length += leaf_length
1440 if length > line_length:
1443 has_inline_comment = leaf_length > len(leaf.value) + len(leaf.prefix)
1444 if leaf.type == STANDALONE_COMMENT or has_inline_comment:
1448 if leaf is opening_bracket:
1449 opening_bracket = None
1450 elif leaf.type in CLOSING_BRACKETS:
1451 prev = line.leaves[index - 1] if index > 0 else None
1454 and prev.type == token.COMMA
1455 and leaf.opening_bracket is not None
1456 and not is_one_sequence_between(
1457 leaf.opening_bracket, leaf, line.leaves
1460 # Never omit bracket pairs with trailing commas.
1461 # We need to explode on those.
1464 inner_brackets.add(id(leaf))
1465 elif leaf.type in CLOSING_BRACKETS:
1466 prev = line.leaves[index - 1] if index > 0 else None
1467 if prev and prev.type in OPENING_BRACKETS:
1468 # Empty brackets would fail a split so treat them as "inner"
1469 # brackets (e.g. only add them to the `omit` set if another
1470 # pair of brackets was good enough.
1471 inner_brackets.add(id(leaf))
1475 omit.add(id(closing_bracket))
1476 omit.update(inner_brackets)
1477 inner_brackets.clear()
1482 and prev.type == token.COMMA
1483 and leaf.opening_bracket is not None
1484 and not is_one_sequence_between(leaf.opening_bracket, leaf, line.leaves)
1486 # Never omit bracket pairs with trailing commas.
1487 # We need to explode on those.
1491 opening_bracket = leaf.opening_bracket
1492 closing_bracket = leaf
1495 def run_transformer(
1497 transform: Transformer,
1499 features: Collection[Feature],
1504 line_str = line_to_string(line)
1505 result: List[Line] = []
1506 for transformed_line in transform(line, features):
1507 if str(transformed_line).strip("\n") == line_str:
1508 raise CannotTransform("Line transformer returned an unchanged result")
1510 result.extend(transform_line(transformed_line, mode=mode, features=features))
1512 features_set = set(features)
1514 Feature.FORCE_OPTIONAL_PARENTHESES in features_set
1515 or transform.__class__.__name__ != "rhs"
1516 or not line.bracket_tracker.invisible
1517 or any(bracket.value for bracket in line.bracket_tracker.invisible)
1518 or line.contains_multiline_strings()
1519 or result[0].contains_uncollapsable_type_comments()
1520 or result[0].contains_unsplittable_type_ignore()
1521 or is_line_short_enough(result[0], line_length=mode.line_length)
1522 # If any leaves have no parents (which _can_ occur since
1523 # `transform(line)` potentially destroys the line's underlying node
1524 # structure), then we can't proceed. Doing so would cause the below
1525 # call to `append_leaves()` to fail.
1526 or any(leaf.parent is None for leaf in line.leaves)
1530 line_copy = line.clone()
1531 append_leaves(line_copy, line, line.leaves)
1532 features_fop = features_set | {Feature.FORCE_OPTIONAL_PARENTHESES}
1533 second_opinion = run_transformer(
1534 line_copy, transform, mode, features_fop, line_str=line_str
1537 is_line_short_enough(ln, line_length=mode.line_length) for ln in second_opinion
1539 result = second_opinion