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_async_stmt_or_funcdef,
40 is_atom_with_invisible_parens,
46 is_one_sequence_between,
51 is_tuple_containing_walrus,
58 from black.numerics import normalize_numeric_literal
59 from black.strings import (
62 normalize_string_prefix,
63 normalize_string_quotes,
64 normalize_unicode_escape_sequences,
66 from black.trans import (
75 from blib2to3.pgen2 import token
76 from blib2to3.pytree import Leaf, Node
80 LN = Union[Leaf, Node]
83 class CannotSplit(CannotTransform):
84 """A readable split that fits the allotted line length is impossible."""
87 # This isn't a dataclass because @dataclass + Generic breaks mypyc.
88 # See also https://github.com/mypyc/mypyc/issues/827.
89 class LineGenerator(Visitor[Line]):
90 """Generates reformatted Line objects. Empty lines are not emitted.
92 Note: destroys the tree it's visiting by mutating prefixes of its leaves
93 in ways that will no longer stringify to valid Python code on the tree.
96 def __init__(self, mode: Mode, features: Collection[Feature]) -> None:
98 self.features = features
99 self.current_line: Line
102 def line(self, indent: int = 0) -> Iterator[Line]:
105 If the line is empty, only emit if it makes sense.
106 If the line is too long, split it first and then generate.
108 If any lines were generated, set up a new current_line.
110 if not self.current_line:
111 self.current_line.depth += indent
112 return # Line is empty, don't emit. Creating a new one unnecessary.
115 Preview.improved_async_statements_handling in self.mode
116 and len(self.current_line.leaves) == 1
117 and is_async_stmt_or_funcdef(self.current_line.leaves[0])
119 # Special case for async def/for/with statements. `visit_async_stmt`
120 # adds an `ASYNC` leaf then visits the child def/for/with statement
121 # nodes. Line yields from those nodes shouldn't treat the former
122 # `ASYNC` leaf as a complete line.
125 complete_line = self.current_line
126 self.current_line = Line(mode=self.mode, depth=complete_line.depth + indent)
129 def visit_default(self, node: LN) -> Iterator[Line]:
130 """Default `visit_*()` implementation. Recurses to children of `node`."""
131 if isinstance(node, Leaf):
132 any_open_brackets = self.current_line.bracket_tracker.any_open_brackets()
133 for comment in generate_comments(node):
134 if any_open_brackets:
135 # any comment within brackets is subject to splitting
136 self.current_line.append(comment)
137 elif comment.type == token.COMMENT:
138 # regular trailing comment
139 self.current_line.append(comment)
140 yield from self.line()
143 # regular standalone comment
144 yield from self.line()
146 self.current_line.append(comment)
147 yield from self.line()
149 normalize_prefix(node, inside_brackets=any_open_brackets)
150 if self.mode.string_normalization and node.type == token.STRING:
151 node.value = normalize_string_prefix(node.value)
152 node.value = normalize_string_quotes(node.value)
153 if node.type == token.NUMBER:
154 normalize_numeric_literal(node)
155 if node.type not in WHITESPACE:
156 self.current_line.append(node)
157 yield from super().visit_default(node)
159 def visit_test(self, node: Node) -> Iterator[Line]:
160 """Visit an `x if y else z` test"""
162 if Preview.parenthesize_conditional_expressions in self.mode:
163 already_parenthesized = (
164 node.prev_sibling and node.prev_sibling.type == token.LPAR
167 if not already_parenthesized:
168 lpar = Leaf(token.LPAR, "")
169 rpar = Leaf(token.RPAR, "")
170 node.insert_child(0, lpar)
171 node.append_child(rpar)
173 yield from self.visit_default(node)
175 def visit_INDENT(self, node: Leaf) -> Iterator[Line]:
176 """Increase indentation level, maybe yield a line."""
177 # In blib2to3 INDENT never holds comments.
178 yield from self.line(+1)
179 yield from self.visit_default(node)
181 def visit_DEDENT(self, node: Leaf) -> Iterator[Line]:
182 """Decrease indentation level, maybe yield a line."""
183 # The current line might still wait for trailing comments. At DEDENT time
184 # there won't be any (they would be prefixes on the preceding NEWLINE).
185 # Emit the line then.
186 yield from self.line()
188 # While DEDENT has no value, its prefix may contain standalone comments
189 # that belong to the current indentation level. Get 'em.
190 yield from self.visit_default(node)
192 # Finally, emit the dedent.
193 yield from self.line(-1)
196 self, node: Node, keywords: Set[str], parens: Set[str]
198 """Visit a statement.
200 This implementation is shared for `if`, `while`, `for`, `try`, `except`,
201 `def`, `with`, `class`, `assert`, and assignments.
203 The relevant Python language `keywords` for a given statement will be
204 NAME leaves within it. This methods puts those on a separate line.
206 `parens` holds a set of string leaf values immediately after which
207 invisible parens should be put.
209 normalize_invisible_parens(
210 node, parens_after=parens, mode=self.mode, features=self.features
212 for child in node.children:
213 if is_name_token(child) and child.value in keywords:
214 yield from self.line()
216 yield from self.visit(child)
218 def visit_typeparams(self, node: Node) -> Iterator[Line]:
219 yield from self.visit_default(node)
220 node.children[0].prefix = ""
222 def visit_typevartuple(self, node: Node) -> Iterator[Line]:
223 yield from self.visit_default(node)
224 node.children[1].prefix = ""
226 def visit_paramspec(self, node: Node) -> Iterator[Line]:
227 yield from self.visit_default(node)
228 node.children[1].prefix = ""
230 def visit_dictsetmaker(self, node: Node) -> Iterator[Line]:
231 if Preview.wrap_long_dict_values_in_parens in self.mode:
232 for i, child in enumerate(node.children):
235 if node.children[i - 1].type == token.COLON:
236 if child.type == syms.atom and child.children[0].type == token.LPAR:
237 if maybe_make_parens_invisible_in_atom(
240 remove_brackets_around_comma=False,
242 wrap_in_parentheses(node, child, visible=False)
244 wrap_in_parentheses(node, child, visible=False)
245 yield from self.visit_default(node)
247 def visit_funcdef(self, node: Node) -> Iterator[Line]:
248 """Visit function definition."""
249 yield from self.line()
251 # Remove redundant brackets around return type annotation.
252 is_return_annotation = False
253 for child in node.children:
254 if child.type == token.RARROW:
255 is_return_annotation = True
256 elif is_return_annotation:
257 if child.type == syms.atom and child.children[0].type == token.LPAR:
258 if maybe_make_parens_invisible_in_atom(
261 remove_brackets_around_comma=False,
263 wrap_in_parentheses(node, child, visible=False)
265 wrap_in_parentheses(node, child, visible=False)
266 is_return_annotation = False
268 for child in node.children:
269 yield from self.visit(child)
271 def visit_match_case(self, node: Node) -> Iterator[Line]:
272 """Visit either a match or case statement."""
273 normalize_invisible_parens(
274 node, parens_after=set(), mode=self.mode, features=self.features
277 yield from self.line()
278 for child in node.children:
279 yield from self.visit(child)
281 def visit_suite(self, node: Node) -> Iterator[Line]:
283 if self.mode.is_pyi and is_stub_suite(node):
284 yield from self.visit(node.children[2])
286 yield from self.visit_default(node)
288 def visit_simple_stmt(self, node: Node) -> Iterator[Line]:
289 """Visit a statement without nested statements."""
290 prev_type: Optional[int] = None
291 for child in node.children:
292 if (prev_type is None or prev_type == token.SEMI) and is_arith_like(child):
293 wrap_in_parentheses(node, child, visible=False)
294 prev_type = child.type
296 is_suite_like = node.parent and node.parent.type in STATEMENT
298 if self.mode.is_pyi and is_stub_body(node):
299 yield from self.visit_default(node)
301 yield from self.line(+1)
302 yield from self.visit_default(node)
303 yield from self.line(-1)
309 or not is_stub_suite(node.parent)
311 yield from self.line()
312 yield from self.visit_default(node)
314 def visit_async_stmt(self, node: Node) -> Iterator[Line]:
315 """Visit `async def`, `async for`, `async with`."""
316 yield from self.line()
318 children = iter(node.children)
319 for child in children:
320 yield from self.visit(child)
322 if child.type == token.ASYNC or child.type == STANDALONE_COMMENT:
323 # STANDALONE_COMMENT happens when `# fmt: skip` is applied on the async
327 internal_stmt = next(children)
328 if Preview.improved_async_statements_handling in self.mode:
329 yield from self.visit(internal_stmt)
331 for child in internal_stmt.children:
332 yield from self.visit(child)
334 def visit_decorators(self, node: Node) -> Iterator[Line]:
335 """Visit decorators."""
336 for child in node.children:
337 yield from self.line()
338 yield from self.visit(child)
340 def visit_power(self, node: Node) -> Iterator[Line]:
341 for idx, leaf in enumerate(node.children[:-1]):
342 next_leaf = node.children[idx + 1]
344 if not isinstance(leaf, Leaf):
347 value = leaf.value.lower()
349 leaf.type == token.NUMBER
350 and next_leaf.type == syms.trailer
351 # Ensure that we are in an attribute trailer
352 and next_leaf.children[0].type == token.DOT
353 # It shouldn't wrap hexadecimal, binary and octal literals
354 and not value.startswith(("0x", "0b", "0o"))
355 # It shouldn't wrap complex literals
358 wrap_in_parentheses(node, leaf)
360 remove_await_parens(node)
362 yield from self.visit_default(node)
364 def visit_SEMI(self, leaf: Leaf) -> Iterator[Line]:
365 """Remove a semicolon and put the other statement on a separate line."""
366 yield from self.line()
368 def visit_ENDMARKER(self, leaf: Leaf) -> Iterator[Line]:
369 """End of file. Process outstanding comments and end with a newline."""
370 yield from self.visit_default(leaf)
371 yield from self.line()
373 def visit_STANDALONE_COMMENT(self, leaf: Leaf) -> Iterator[Line]:
374 if not self.current_line.bracket_tracker.any_open_brackets():
375 yield from self.line()
376 yield from self.visit_default(leaf)
378 def visit_factor(self, node: Node) -> Iterator[Line]:
379 """Force parentheses between a unary op and a binary power:
383 _operator, operand = node.children
385 operand.type == syms.power
386 and len(operand.children) == 3
387 and operand.children[1].type == token.DOUBLESTAR
389 lpar = Leaf(token.LPAR, "(")
390 rpar = Leaf(token.RPAR, ")")
391 index = operand.remove() or 0
392 node.insert_child(index, Node(syms.atom, [lpar, operand, rpar]))
393 yield from self.visit_default(node)
395 def visit_STRING(self, leaf: Leaf) -> Iterator[Line]:
396 if Preview.hex_codes_in_unicode_sequences in self.mode:
397 normalize_unicode_escape_sequences(leaf)
399 if is_docstring(leaf) and "\\\n" not in leaf.value:
400 # We're ignoring docstrings with backslash newline escapes because changing
401 # indentation of those changes the AST representation of the code.
402 if self.mode.string_normalization:
403 docstring = normalize_string_prefix(leaf.value)
404 # visit_default() does handle string normalization for us, but
405 # since this method acts differently depending on quote style (ex.
406 # see padding logic below), there's a possibility for unstable
407 # formatting as visit_default() is called *after*. To avoid a
408 # situation where this function formats a docstring differently on
409 # the second pass, normalize it early.
410 docstring = normalize_string_quotes(docstring)
412 docstring = leaf.value
413 prefix = get_string_prefix(docstring)
414 docstring = docstring[len(prefix) :] # Remove the prefix
415 quote_char = docstring[0]
416 # A natural way to remove the outer quotes is to do:
417 # docstring = docstring.strip(quote_char)
418 # but that breaks on """""x""" (which is '""x').
419 # So we actually need to remove the first character and the next two
420 # characters but only if they are the same as the first.
421 quote_len = 1 if docstring[1] != quote_char else 3
422 docstring = docstring[quote_len:-quote_len]
423 docstring_started_empty = not docstring
424 indent = " " * 4 * self.current_line.depth
426 if is_multiline_string(leaf):
427 docstring = fix_docstring(docstring, indent)
429 docstring = docstring.strip()
431 has_trailing_backslash = False
433 # Add some padding if the docstring starts / ends with a quote mark.
434 if docstring[0] == quote_char:
435 docstring = " " + docstring
436 if docstring[-1] == quote_char:
438 if docstring[-1] == "\\":
439 backslash_count = len(docstring) - len(docstring.rstrip("\\"))
440 if backslash_count % 2:
441 # Odd number of tailing backslashes, add some padding to
442 # avoid escaping the closing string quote.
444 has_trailing_backslash = True
445 elif not docstring_started_empty:
448 # We could enforce triple quotes at this point.
449 quote = quote_char * quote_len
451 # It's invalid to put closing single-character quotes on a new line.
452 if self.mode and quote_len == 3:
453 # We need to find the length of the last line of the docstring
454 # to find if we can add the closing quotes to the line without
455 # exceeding the maximum line length.
456 # If docstring is one line, we don't put the closing quotes on a
457 # separate line because it looks ugly (#3320).
458 lines = docstring.splitlines()
459 last_line_length = len(lines[-1]) if docstring else 0
461 # If adding closing quotes would cause the last line to exceed
462 # the maximum line length then put a line break before the
466 and last_line_length + quote_len > self.mode.line_length
467 and len(indent) + quote_len <= self.mode.line_length
468 and not has_trailing_backslash
470 leaf.value = prefix + quote + docstring + "\n" + indent + quote
472 leaf.value = prefix + quote + docstring + quote
474 leaf.value = prefix + quote + docstring + quote
476 yield from self.visit_default(leaf)
478 def __post_init__(self) -> None:
479 """You are in a twisty little maze of passages."""
480 self.current_line = Line(mode=self.mode)
484 self.visit_assert_stmt = partial(v, keywords={"assert"}, parens={"assert", ","})
485 self.visit_if_stmt = partial(
486 v, keywords={"if", "else", "elif"}, parens={"if", "elif"}
488 self.visit_while_stmt = partial(v, keywords={"while", "else"}, parens={"while"})
489 self.visit_for_stmt = partial(v, keywords={"for", "else"}, parens={"for", "in"})
490 self.visit_try_stmt = partial(
491 v, keywords={"try", "except", "else", "finally"}, parens=Ø
493 self.visit_except_clause = partial(v, keywords={"except"}, parens={"except"})
494 self.visit_with_stmt = partial(v, keywords={"with"}, parens={"with"})
495 self.visit_classdef = partial(v, keywords={"class"}, parens=Ø)
496 self.visit_expr_stmt = partial(v, keywords=Ø, parens=ASSIGNMENTS)
497 self.visit_return_stmt = partial(v, keywords={"return"}, parens={"return"})
498 self.visit_import_from = partial(v, keywords=Ø, parens={"import"})
499 self.visit_del_stmt = partial(v, keywords=Ø, parens={"del"})
500 self.visit_async_funcdef = self.visit_async_stmt
501 self.visit_decorated = self.visit_decorators
504 self.visit_match_stmt = self.visit_match_case
505 self.visit_case_block = self.visit_match_case
509 line: Line, mode: Mode, features: Collection[Feature] = ()
511 """Transform a `line`, potentially splitting it into many lines.
513 They should fit in the allotted `line_length` but might not be able to.
515 `features` are syntactical features that may be used in the output.
521 line_str = line_to_string(line)
523 ll = mode.line_length
524 sn = mode.string_normalization
525 string_merge = StringMerger(ll, sn)
526 string_paren_strip = StringParenStripper(ll, sn)
527 string_split = StringSplitter(ll, sn)
528 string_paren_wrap = StringParenWrapper(ll, sn)
530 transformers: List[Transformer]
532 not line.contains_uncollapsable_type_comments()
533 and not line.should_split_rhs
534 and not line.magic_trailing_comma
536 is_line_short_enough(line, mode=mode, line_str=line_str)
537 or line.contains_unsplittable_type_ignore()
539 and not (line.inside_brackets and line.contains_standalone_comments())
541 # Only apply basic string preprocessing, since lines shouldn't be split here.
542 if Preview.string_processing in mode:
543 transformers = [string_merge, string_paren_strip]
547 transformers = [left_hand_split]
551 self: object, line: Line, features: Collection[Feature], mode: Mode
553 """Wraps calls to `right_hand_split`.
555 The calls increasingly `omit` right-hand trailers (bracket pairs with
556 content), meaning the trailers get glued together to split on another
557 bracket pair instead.
559 for omit in generate_trailers_to_omit(line, mode.line_length):
560 lines = list(right_hand_split(line, mode, features, omit=omit))
561 # Note: this check is only able to figure out if the first line of the
562 # *current* transformation fits in the line length. This is true only
563 # for simple cases. All others require running more transforms via
564 # `transform_line()`. This check doesn't know if those would succeed.
565 if is_line_short_enough(lines[0], mode=mode):
569 # All splits failed, best effort split with no omits.
570 # This mostly happens to multiline strings that are by definition
571 # reported as not fitting a single line, as well as lines that contain
572 # trailing commas (those have to be exploded).
573 yield from right_hand_split(line, mode, features=features)
575 # HACK: nested functions (like _rhs) compiled by mypyc don't retain their
576 # __name__ attribute which is needed in `run_transformer` further down.
577 # Unfortunately a nested class breaks mypyc too. So a class must be created
578 # via type ... https://github.com/mypyc/mypyc/issues/884
579 rhs = type("rhs", (), {"__call__": _rhs})()
581 if Preview.string_processing in mode:
582 if line.inside_brackets:
588 standalone_comment_split,
601 if line.inside_brackets:
602 transformers = [delimiter_split, standalone_comment_split, rhs]
605 # It's always safe to attempt hugging of power operations and pretty much every line
607 transformers.append(hug_power_op)
609 for transform in transformers:
610 # We are accumulating lines in `result` because we might want to abort
611 # mission and return the original line in the end, or attempt a different
614 result = run_transformer(line, transform, mode, features, line_str=line_str)
615 except CannotTransform:
625 class _BracketSplitComponent(Enum):
632 line: Line, _features: Collection[Feature], mode: Mode
634 """Split line into many lines, starting with the first matching bracket pair.
636 Note: this usually looks weird, only use this for function definitions.
637 Prefer RHS otherwise. This is why this function is not symmetrical with
638 :func:`right_hand_split` which also handles optional parentheses.
640 tail_leaves: List[Leaf] = []
641 body_leaves: List[Leaf] = []
642 head_leaves: List[Leaf] = []
643 current_leaves = head_leaves
644 matching_bracket: Optional[Leaf] = None
645 for leaf in line.leaves:
647 current_leaves is body_leaves
648 and leaf.type in CLOSING_BRACKETS
649 and leaf.opening_bracket is matching_bracket
650 and isinstance(matching_bracket, Leaf)
653 ensure_visible(matching_bracket)
654 current_leaves = tail_leaves if body_leaves else head_leaves
655 current_leaves.append(leaf)
656 if current_leaves is head_leaves:
657 if leaf.type in OPENING_BRACKETS:
658 matching_bracket = leaf
659 current_leaves = body_leaves
660 if not matching_bracket:
661 raise CannotSplit("No brackets found")
663 head = bracket_split_build_line(
664 head_leaves, line, matching_bracket, component=_BracketSplitComponent.head
666 body = bracket_split_build_line(
667 body_leaves, line, matching_bracket, component=_BracketSplitComponent.body
669 tail = bracket_split_build_line(
670 tail_leaves, line, matching_bracket, component=_BracketSplitComponent.tail
672 bracket_split_succeeded_or_raise(head, body, tail)
673 for result in (head, body, tail):
678 def right_hand_split(
681 features: Collection[Feature] = (),
682 omit: Collection[LeafID] = (),
684 """Split line into many lines, starting with the last matching bracket pair.
686 If the split was by optional parentheses, attempt splitting without them, too.
687 `omit` is a collection of closing bracket IDs that shouldn't be considered for
690 Note: running this function modifies `bracket_depth` on the leaves of `line`.
692 rhs_result = _first_right_hand_split(line, omit=omit)
693 yield from _maybe_split_omitting_optional_parens(
694 rhs_result, line, mode, features=features, omit=omit
698 def _first_right_hand_split(
700 omit: Collection[LeafID] = (),
702 """Split the line into head, body, tail starting with the last bracket pair.
704 Note: this function should not have side effects. It's relied upon by
705 _maybe_split_omitting_optional_parens to get an opinion whether to prefer
706 splitting on the right side of an assignment statement.
708 tail_leaves: List[Leaf] = []
709 body_leaves: List[Leaf] = []
710 head_leaves: List[Leaf] = []
711 current_leaves = tail_leaves
712 opening_bracket: Optional[Leaf] = None
713 closing_bracket: Optional[Leaf] = None
714 for leaf in reversed(line.leaves):
715 if current_leaves is body_leaves:
716 if leaf is opening_bracket:
717 current_leaves = head_leaves if body_leaves else tail_leaves
718 current_leaves.append(leaf)
719 if current_leaves is tail_leaves:
720 if leaf.type in CLOSING_BRACKETS and id(leaf) not in omit:
721 opening_bracket = leaf.opening_bracket
722 closing_bracket = leaf
723 current_leaves = body_leaves
724 if not (opening_bracket and closing_bracket and head_leaves):
725 # If there is no opening or closing_bracket that means the split failed and
726 # all content is in the tail. Otherwise, if `head_leaves` are empty, it means
727 # the matching `opening_bracket` wasn't available on `line` anymore.
728 raise CannotSplit("No brackets found")
730 tail_leaves.reverse()
731 body_leaves.reverse()
732 head_leaves.reverse()
733 head = bracket_split_build_line(
734 head_leaves, line, opening_bracket, component=_BracketSplitComponent.head
736 body = bracket_split_build_line(
737 body_leaves, line, opening_bracket, component=_BracketSplitComponent.body
739 tail = bracket_split_build_line(
740 tail_leaves, line, opening_bracket, component=_BracketSplitComponent.tail
742 bracket_split_succeeded_or_raise(head, body, tail)
743 return RHSResult(head, body, tail, opening_bracket, closing_bracket)
746 def _maybe_split_omitting_optional_parens(
750 features: Collection[Feature] = (),
751 omit: Collection[LeafID] = (),
754 Feature.FORCE_OPTIONAL_PARENTHESES not in features
755 # the opening bracket is an optional paren
756 and rhs.opening_bracket.type == token.LPAR
757 and not rhs.opening_bracket.value
758 # the closing bracket is an optional paren
759 and rhs.closing_bracket.type == token.RPAR
760 and not rhs.closing_bracket.value
761 # it's not an import (optional parens are the only thing we can split on
762 # in this case; attempting a split without them is a waste of time)
763 and not line.is_import
764 # there are no standalone comments in the body
765 and not rhs.body.contains_standalone_comments(0)
766 # and we can actually remove the parens
767 and can_omit_invisible_parens(rhs, mode.line_length)
769 omit = {id(rhs.closing_bracket), *omit}
771 # The RHSResult Omitting Optional Parens.
772 rhs_oop = _first_right_hand_split(line, omit=omit)
774 Preview.prefer_splitting_right_hand_side_of_assignments in line.mode
775 # the split is right after `=`
776 and len(rhs.head.leaves) >= 2
777 and rhs.head.leaves[-2].type == token.EQUAL
778 # the left side of assignment contains brackets
779 and any(leaf.type in BRACKETS for leaf in rhs.head.leaves[:-1])
780 # the left side of assignment is short enough (the -1 is for the ending
782 and is_line_short_enough(
783 rhs.head, mode=replace(mode, line_length=mode.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, mode)
792 yield from _maybe_split_omitting_optional_parens(
793 rhs_oop, line, mode, features=features, omit=omit
797 except CannotSplit as e:
799 can_be_split(rhs.body) or is_line_short_enough(rhs.body, mode=mode)
802 "Splitting failed, body is still too long and can't be split."
806 rhs.head.contains_multiline_strings()
807 or rhs.tail.contains_multiline_strings()
810 "The current optional pair of parentheses is bound to fail to"
811 " satisfy the splitting algorithm because the head or the tail"
812 " contains multiline strings which by definition never fit one"
816 ensure_visible(rhs.opening_bracket)
817 ensure_visible(rhs.closing_bracket)
818 for result in (rhs.head, rhs.body, rhs.tail):
823 def _prefer_split_rhs_oop(rhs_oop: RHSResult, mode: Mode) -> bool:
825 Returns whether we should prefer the result from a split omitting optional parens.
827 has_closing_bracket_after_assign = False
828 for leaf in reversed(rhs_oop.head.leaves):
829 if leaf.type == token.EQUAL:
831 if leaf.type in CLOSING_BRACKETS:
832 has_closing_bracket_after_assign = True
835 # contains matching brackets after the `=` (done by checking there is a
837 has_closing_bracket_after_assign
839 # the split is actually from inside the optional parens (done by checking
840 # the first line still contains the `=`)
841 any(leaf.type == token.EQUAL for leaf in rhs_oop.head.leaves)
842 # the first line is short enough
843 and is_line_short_enough(rhs_oop.head, mode=mode)
845 # contains unsplittable type ignore
846 or rhs_oop.head.contains_unsplittable_type_ignore()
847 or rhs_oop.body.contains_unsplittable_type_ignore()
848 or rhs_oop.tail.contains_unsplittable_type_ignore()
852 def bracket_split_succeeded_or_raise(head: Line, body: Line, tail: Line) -> None:
853 """Raise :exc:`CannotSplit` if the last left- or right-hand split failed.
855 Do nothing otherwise.
857 A left- or right-hand split is based on a pair of brackets. Content before
858 (and including) the opening bracket is left on one line, content inside the
859 brackets is put on a separate line, and finally content starting with and
860 following the closing bracket is put on a separate line.
862 Those are called `head`, `body`, and `tail`, respectively. If the split
863 produced the same line (all content in `head`) or ended up with an empty `body`
864 and the `tail` is just the closing bracket, then it's considered failed.
866 tail_len = len(str(tail).strip())
869 raise CannotSplit("Splitting brackets produced the same line")
873 f"Splitting brackets on an empty body to save {tail_len} characters is"
878 def bracket_split_build_line(
881 opening_bracket: Leaf,
883 component: _BracketSplitComponent,
885 """Return a new line with given `leaves` and respective comments from `original`.
887 If it's the head component, brackets will be tracked so trailing commas are
890 If it's the body component, the result line is one-indented inside brackets and as
891 such has its first leaf's prefix normalized and a trailing comma added when
894 result = Line(mode=original.mode, depth=original.depth)
895 if component is _BracketSplitComponent.body:
896 result.inside_brackets = True
899 # Since body is a new indent level, remove spurious leading whitespace.
900 normalize_prefix(leaves[0], inside_brackets=True)
901 # Ensure a trailing comma for imports and standalone function arguments, but
902 # be careful not to add one after any comments or within type annotations.
905 and opening_bracket.value == "("
906 and not any(leaf.type == token.COMMA for leaf in leaves)
907 # In particular, don't add one within a parenthesized return annotation.
908 # Unfortunately the indicator we're in a return annotation (RARROW) may
909 # be defined directly in the parent node, the parent of the parent ...
910 # and so on depending on how complex the return annotation is.
911 # This isn't perfect and there's some false negatives but they are in
912 # contexts were a comma is actually fine.
914 node.prev_sibling.type == RARROW
917 getattr(leaves[0].parent, "parent", None),
919 if isinstance(node, Node) and isinstance(node.prev_sibling, Leaf)
923 if original.is_import or no_commas:
924 for i in range(len(leaves) - 1, -1, -1):
925 if leaves[i].type == STANDALONE_COMMENT:
928 if leaves[i].type != token.COMMA:
929 new_comma = Leaf(token.COMMA, ",")
930 leaves.insert(i + 1, new_comma)
933 leaves_to_track: Set[LeafID] = set()
934 if component is _BracketSplitComponent.head:
935 leaves_to_track = get_leaves_inside_matching_brackets(leaves)
941 track_bracket=id(leaf) in leaves_to_track,
943 for comment_after in original.comments_after(leaf):
944 result.append(comment_after, preformatted=True)
945 if component is _BracketSplitComponent.body and should_split_line(
946 result, opening_bracket
948 result.should_split_rhs = True
952 def dont_increase_indentation(split_func: Transformer) -> Transformer:
953 """Normalize prefix of the first leaf in every line returned by `split_func`.
955 This is a decorator over relevant split functions.
960 line: Line, features: Collection[Feature], mode: Mode
962 for split_line in split_func(line, features, mode):
963 normalize_prefix(split_line.leaves[0], inside_brackets=True)
969 def _get_last_non_comment_leaf(line: Line) -> Optional[int]:
970 for leaf_idx in range(len(line.leaves) - 1, 0, -1):
971 if line.leaves[leaf_idx].type != STANDALONE_COMMENT:
976 def _safe_add_trailing_comma(safe: bool, delimiter_priority: int, line: Line) -> Line:
979 and delimiter_priority == COMMA_PRIORITY
980 and line.leaves[-1].type != token.COMMA
981 and line.leaves[-1].type != STANDALONE_COMMENT
983 new_comma = Leaf(token.COMMA, ",")
984 line.append(new_comma)
988 @dont_increase_indentation
990 line: Line, features: Collection[Feature], mode: Mode
992 """Split according to delimiters of the highest priority.
994 If the appropriate Features are given, the split will add trailing commas
995 also in function signatures and calls that contain `*` and `**`.
998 last_leaf = line.leaves[-1]
1000 raise CannotSplit("Line empty") from None
1002 bt = line.bracket_tracker
1004 delimiter_priority = bt.max_delimiter_priority(exclude={id(last_leaf)})
1006 raise CannotSplit("No delimiters found") from None
1008 if delimiter_priority == DOT_PRIORITY:
1009 if bt.delimiter_count_with_priority(delimiter_priority) == 1:
1010 raise CannotSplit("Splitting a single attribute from its owner looks wrong")
1012 current_line = Line(
1013 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1015 lowest_depth = sys.maxsize
1016 trailing_comma_safe = True
1018 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1019 """Append `leaf` to current line or to new line if appending impossible."""
1020 nonlocal current_line
1022 current_line.append_safe(leaf, preformatted=True)
1026 current_line = Line(
1027 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1029 current_line.append(leaf)
1031 last_non_comment_leaf = _get_last_non_comment_leaf(line)
1032 for leaf_idx, leaf in enumerate(line.leaves):
1033 yield from append_to_line(leaf)
1035 for comment_after in line.comments_after(leaf):
1036 yield from append_to_line(comment_after)
1038 lowest_depth = min(lowest_depth, leaf.bracket_depth)
1039 if leaf.bracket_depth == lowest_depth:
1040 if is_vararg(leaf, within={syms.typedargslist}):
1041 trailing_comma_safe = (
1042 trailing_comma_safe and Feature.TRAILING_COMMA_IN_DEF in features
1044 elif is_vararg(leaf, within={syms.arglist, syms.argument}):
1045 trailing_comma_safe = (
1046 trailing_comma_safe and Feature.TRAILING_COMMA_IN_CALL in features
1050 Preview.add_trailing_comma_consistently in mode
1051 and last_leaf.type == STANDALONE_COMMENT
1052 and leaf_idx == last_non_comment_leaf
1054 current_line = _safe_add_trailing_comma(
1055 trailing_comma_safe, delimiter_priority, current_line
1058 leaf_priority = bt.delimiters.get(id(leaf))
1059 if leaf_priority == delimiter_priority:
1062 current_line = Line(
1063 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1066 current_line = _safe_add_trailing_comma(
1067 trailing_comma_safe, delimiter_priority, current_line
1072 @dont_increase_indentation
1073 def standalone_comment_split(
1074 line: Line, features: Collection[Feature], mode: Mode
1075 ) -> Iterator[Line]:
1076 """Split standalone comments from the rest of the line."""
1077 if not line.contains_standalone_comments(0):
1078 raise CannotSplit("Line does not have any standalone comments")
1080 current_line = Line(
1081 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1084 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1085 """Append `leaf` to current line or to new line if appending impossible."""
1086 nonlocal current_line
1088 current_line.append_safe(leaf, preformatted=True)
1092 current_line = Line(
1093 line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1095 current_line.append(leaf)
1097 for leaf in line.leaves:
1098 yield from append_to_line(leaf)
1100 for comment_after in line.comments_after(leaf):
1101 yield from append_to_line(comment_after)
1107 def normalize_prefix(leaf: Leaf, *, inside_brackets: bool) -> None:
1108 """Leave existing extra newlines if not `inside_brackets`. Remove everything
1111 Note: don't use backslashes for formatting or you'll lose your voting rights.
1113 if not inside_brackets:
1114 spl = leaf.prefix.split("#")
1115 if "\\" not in spl[0]:
1116 nl_count = spl[-1].count("\n")
1119 leaf.prefix = "\n" * nl_count
1125 def normalize_invisible_parens(
1126 node: Node, parens_after: Set[str], *, mode: Mode, features: Collection[Feature]
1128 """Make existing optional parentheses invisible or create new ones.
1130 `parens_after` is a set of string leaf values immediately after which parens
1133 Standardizes on visible parentheses for single-element tuples, and keeps
1134 existing visible parentheses for other tuples and generator expressions.
1136 for pc in list_comments(node.prefix, is_endmarker=False):
1137 if pc.value in FMT_OFF:
1138 # This `node` has a prefix with `# fmt: off`, don't mess with parens.
1141 # The multiple context managers grammar has a different pattern, thus this is
1142 # separate from the for-loop below. This possibly wraps them in invisible parens,
1143 # and later will be removed in remove_with_parens when needed.
1144 if node.type == syms.with_stmt:
1145 _maybe_wrap_cms_in_parens(node, mode, features)
1148 for index, child in enumerate(list(node.children)):
1149 # Fixes a bug where invisible parens are not properly stripped from
1150 # assignment statements that contain type annotations.
1151 if isinstance(child, Node) and child.type == syms.annassign:
1152 normalize_invisible_parens(
1153 child, parens_after=parens_after, mode=mode, features=features
1156 # Add parentheses around long tuple unpacking in assignments.
1159 and isinstance(child, Node)
1160 and child.type == syms.testlist_star_expr
1166 child.type == syms.atom
1167 and node.type == syms.for_stmt
1168 and isinstance(child.prev_sibling, Leaf)
1169 and child.prev_sibling.type == token.NAME
1170 and child.prev_sibling.value == "for"
1172 if maybe_make_parens_invisible_in_atom(
1175 remove_brackets_around_comma=True,
1177 wrap_in_parentheses(node, child, visible=False)
1178 elif isinstance(child, Node) and node.type == syms.with_stmt:
1179 remove_with_parens(child, node)
1180 elif child.type == syms.atom:
1181 if maybe_make_parens_invisible_in_atom(
1185 wrap_in_parentheses(node, child, visible=False)
1186 elif is_one_tuple(child):
1187 wrap_in_parentheses(node, child, visible=True)
1188 elif node.type == syms.import_from:
1189 _normalize_import_from(node, child, index)
1193 and child.type == token.STAR
1194 and node.type == syms.except_clause
1196 # In except* (PEP 654), the star is actually part of
1197 # of the keyword. So we need to skip the insertion of
1198 # invisible parentheses to work more precisely.
1201 elif not (isinstance(child, Leaf) and is_multiline_string(child)):
1202 wrap_in_parentheses(node, child, visible=False)
1204 comma_check = child.type == token.COMMA
1206 check_lpar = isinstance(child, Leaf) and (
1207 child.value in parens_after or comma_check
1211 def _normalize_import_from(parent: Node, child: LN, index: int) -> None:
1212 # "import from" nodes store parentheses directly as part of
1214 if is_lpar_token(child):
1215 assert is_rpar_token(parent.children[-1])
1216 # make parentheses invisible
1218 parent.children[-1].value = ""
1219 elif child.type != token.STAR:
1220 # insert invisible parentheses
1221 parent.insert_child(index, Leaf(token.LPAR, ""))
1222 parent.append_child(Leaf(token.RPAR, ""))
1225 def remove_await_parens(node: Node) -> None:
1226 if node.children[0].type == token.AWAIT and len(node.children) > 1:
1228 node.children[1].type == syms.atom
1229 and node.children[1].children[0].type == token.LPAR
1231 if maybe_make_parens_invisible_in_atom(
1234 remove_brackets_around_comma=True,
1236 wrap_in_parentheses(node, node.children[1], visible=False)
1238 # Since await is an expression we shouldn't remove
1239 # brackets in cases where this would change
1240 # the AST due to operator precedence.
1241 # Therefore we only aim to remove brackets around
1242 # power nodes that aren't also await expressions themselves.
1243 # https://peps.python.org/pep-0492/#updated-operator-precedence-table
1244 # N.B. We've still removed any redundant nested brackets though :)
1245 opening_bracket = cast(Leaf, node.children[1].children[0])
1246 closing_bracket = cast(Leaf, node.children[1].children[-1])
1247 bracket_contents = node.children[1].children[1]
1248 if isinstance(bracket_contents, Node):
1249 if bracket_contents.type != syms.power:
1250 ensure_visible(opening_bracket)
1251 ensure_visible(closing_bracket)
1253 bracket_contents.type == syms.power
1254 and bracket_contents.children[0].type == token.AWAIT
1256 ensure_visible(opening_bracket)
1257 ensure_visible(closing_bracket)
1258 # If we are in a nested await then recurse down.
1259 remove_await_parens(bracket_contents)
1262 def _maybe_wrap_cms_in_parens(
1263 node: Node, mode: Mode, features: Collection[Feature]
1265 """When enabled and safe, wrap the multiple context managers in invisible parens.
1267 It is only safe when `features` contain Feature.PARENTHESIZED_CONTEXT_MANAGERS.
1270 Feature.PARENTHESIZED_CONTEXT_MANAGERS not in features
1271 or Preview.wrap_multiple_context_managers_in_parens not in mode
1272 or len(node.children) <= 2
1273 # If it's an atom, it's already wrapped in parens.
1274 or node.children[1].type == syms.atom
1277 colon_index: Optional[int] = None
1278 for i in range(2, len(node.children)):
1279 if node.children[i].type == token.COLON:
1282 if colon_index is not None:
1283 lpar = Leaf(token.LPAR, "")
1284 rpar = Leaf(token.RPAR, "")
1285 context_managers = node.children[1:colon_index]
1286 for child in context_managers:
1288 # After wrapping, the with_stmt will look like this:
1294 # ... <-- context_managers
1300 syms.atom, [lpar, Node(syms.testlist_gexp, context_managers), rpar]
1302 node.insert_child(1, new_child)
1305 def remove_with_parens(node: Node, parent: Node) -> None:
1306 """Recursively hide optional parens in `with` statements."""
1307 # Removing all unnecessary parentheses in with statements in one pass is a tad
1308 # complex as different variations of bracketed statements result in pretty
1309 # different parse trees:
1311 # with (open("file")) as f: # this is an asexpr_test
1314 # with (open("file") as f): # this is an atom containing an
1317 # with (open("file")) as f, (open("file")) as f: # this is asexpr_test, COMMA,
1320 # with (open("file") as f, open("file") as f): # an atom containing a
1321 # ... # testlist_gexp which then
1322 # # contains multiple asexpr_test(s)
1323 if node.type == syms.atom:
1324 if maybe_make_parens_invisible_in_atom(
1327 remove_brackets_around_comma=True,
1329 wrap_in_parentheses(parent, node, visible=False)
1330 if isinstance(node.children[1], Node):
1331 remove_with_parens(node.children[1], node)
1332 elif node.type == syms.testlist_gexp:
1333 for child in node.children:
1334 if isinstance(child, Node):
1335 remove_with_parens(child, node)
1336 elif node.type == syms.asexpr_test and not any(
1337 leaf.type == token.COLONEQUAL for leaf in node.leaves()
1339 if maybe_make_parens_invisible_in_atom(
1342 remove_brackets_around_comma=True,
1344 wrap_in_parentheses(node, node.children[0], visible=False)
1347 def maybe_make_parens_invisible_in_atom(
1350 remove_brackets_around_comma: bool = False,
1352 """If it's safe, make the parens in the atom `node` invisible, recursively.
1353 Additionally, remove repeated, adjacent invisible parens from the atom `node`
1354 as they are redundant.
1356 Returns whether the node should itself be wrapped in invisible parentheses.
1359 node.type != syms.atom
1360 or is_empty_tuple(node)
1361 or is_one_tuple(node)
1362 or (is_yield(node) and parent.type != syms.expr_stmt)
1364 # This condition tries to prevent removing non-optional brackets
1365 # around a tuple, however, can be a bit overzealous so we provide
1366 # and option to skip this check for `for` and `with` statements.
1367 not remove_brackets_around_comma
1368 and max_delimiter_priority_in_atom(node) >= COMMA_PRIORITY
1370 or is_tuple_containing_walrus(node)
1374 if is_walrus_assignment(node):
1383 # these ones aren't useful to end users, but they do please fuzzers
1390 first = node.children[0]
1391 last = node.children[-1]
1392 if is_lpar_token(first) and is_rpar_token(last):
1393 middle = node.children[1]
1394 # make parentheses invisible
1397 maybe_make_parens_invisible_in_atom(
1400 remove_brackets_around_comma=remove_brackets_around_comma,
1403 if is_atom_with_invisible_parens(middle):
1404 # Strip the invisible parens from `middle` by replacing
1405 # it with the child in-between the invisible parens
1406 middle.replace(middle.children[1])
1413 def should_split_line(line: Line, opening_bracket: Leaf) -> bool:
1414 """Should `line` be immediately split with `delimiter_split()` after RHS?"""
1416 if not (opening_bracket.parent and opening_bracket.value in "[{("):
1419 # We're essentially checking if the body is delimited by commas and there's more
1420 # than one of them (we're excluding the trailing comma and if the delimiter priority
1421 # is still commas, that means there's more).
1423 trailing_comma = False
1425 last_leaf = line.leaves[-1]
1426 if last_leaf.type == token.COMMA:
1427 trailing_comma = True
1428 exclude.add(id(last_leaf))
1429 max_priority = line.bracket_tracker.max_delimiter_priority(exclude=exclude)
1430 except (IndexError, ValueError):
1433 return max_priority == COMMA_PRIORITY and (
1434 (line.mode.magic_trailing_comma and trailing_comma)
1435 # always explode imports
1436 or opening_bracket.parent.type in {syms.atom, syms.import_from}
1440 def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[LeafID]]:
1441 """Generate sets of closing bracket IDs that should be omitted in a RHS.
1443 Brackets can be omitted if the entire trailer up to and including
1444 a preceding closing bracket fits in one line.
1446 Yielded sets are cumulative (contain results of previous yields, too). First
1447 set is empty, unless the line should explode, in which case bracket pairs until
1448 the one that needs to explode are omitted.
1451 omit: Set[LeafID] = set()
1452 if not line.magic_trailing_comma:
1455 length = 4 * line.depth
1456 opening_bracket: Optional[Leaf] = None
1457 closing_bracket: Optional[Leaf] = None
1458 inner_brackets: Set[LeafID] = set()
1459 for index, leaf, leaf_length in line.enumerate_with_length(reversed=True):
1460 length += leaf_length
1461 if length > line_length:
1464 has_inline_comment = leaf_length > len(leaf.value) + len(leaf.prefix)
1465 if leaf.type == STANDALONE_COMMENT or has_inline_comment:
1469 if leaf is opening_bracket:
1470 opening_bracket = None
1471 elif leaf.type in CLOSING_BRACKETS:
1472 prev = line.leaves[index - 1] if index > 0 else None
1475 and prev.type == token.COMMA
1476 and leaf.opening_bracket is not None
1477 and not is_one_sequence_between(
1478 leaf.opening_bracket, leaf, line.leaves
1481 # Never omit bracket pairs with trailing commas.
1482 # We need to explode on those.
1485 inner_brackets.add(id(leaf))
1486 elif leaf.type in CLOSING_BRACKETS:
1487 prev = line.leaves[index - 1] if index > 0 else None
1488 if prev and prev.type in OPENING_BRACKETS:
1489 # Empty brackets would fail a split so treat them as "inner"
1490 # brackets (e.g. only add them to the `omit` set if another
1491 # pair of brackets was good enough.
1492 inner_brackets.add(id(leaf))
1496 omit.add(id(closing_bracket))
1497 omit.update(inner_brackets)
1498 inner_brackets.clear()
1503 and prev.type == token.COMMA
1504 and leaf.opening_bracket is not None
1505 and not is_one_sequence_between(leaf.opening_bracket, leaf, line.leaves)
1507 # Never omit bracket pairs with trailing commas.
1508 # We need to explode on those.
1512 opening_bracket = leaf.opening_bracket
1513 closing_bracket = leaf
1516 def run_transformer(
1518 transform: Transformer,
1520 features: Collection[Feature],
1525 line_str = line_to_string(line)
1526 result: List[Line] = []
1527 for transformed_line in transform(line, features, mode):
1528 if str(transformed_line).strip("\n") == line_str:
1529 raise CannotTransform("Line transformer returned an unchanged result")
1531 result.extend(transform_line(transformed_line, mode=mode, features=features))
1533 features_set = set(features)
1535 Feature.FORCE_OPTIONAL_PARENTHESES in features_set
1536 or transform.__class__.__name__ != "rhs"
1537 or not line.bracket_tracker.invisible
1538 or any(bracket.value for bracket in line.bracket_tracker.invisible)
1539 or line.contains_multiline_strings()
1540 or result[0].contains_uncollapsable_type_comments()
1541 or result[0].contains_unsplittable_type_ignore()
1542 or is_line_short_enough(result[0], mode=mode)
1543 # If any leaves have no parents (which _can_ occur since
1544 # `transform(line)` potentially destroys the line's underlying node
1545 # structure), then we can't proceed. Doing so would cause the below
1546 # call to `append_leaves()` to fail.
1547 or any(leaf.parent is None for leaf in line.leaves)
1551 line_copy = line.clone()
1552 append_leaves(line_copy, line, line.leaves)
1553 features_fop = features_set | {Feature.FORCE_OPTIONAL_PARENTHESES}
1554 second_opinion = run_transformer(
1555 line_copy, transform, mode, features_fop, line_str=line_str
1557 if all(is_line_short_enough(ln, mode=mode) for ln in second_opinion):
1558 result = second_opinion