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,
52 is_type_ignore_comment_string,
59 from black.numerics import normalize_numeric_literal
60 from black.strings import (
63 normalize_string_prefix,
64 normalize_string_quotes,
65 normalize_unicode_escape_sequences,
67 from black.trans import (
76 from blib2to3.pgen2 import token
77 from blib2to3.pytree import Leaf, Node
81 LN = Union[Leaf, Node]
84 class CannotSplit(CannotTransform):
85 """A readable split that fits the allotted line length is impossible."""
88 # This isn't a dataclass because @dataclass + Generic breaks mypyc.
89 # See also https://github.com/mypyc/mypyc/issues/827.
90 class LineGenerator(Visitor[Line]):
91 """Generates reformatted Line objects. Empty lines are not emitted.
93 Note: destroys the tree it's visiting by mutating prefixes of its leaves
94 in ways that will no longer stringify to valid Python code on the tree.
97 def __init__(self, mode: Mode, features: Collection[Feature]) -> None:
99 self.features = features
100 self.current_line: Line
103 def line(self, indent: int = 0) -> Iterator[Line]:
106 If the line is empty, only emit if it makes sense.
107 If the line is too long, split it first and then generate.
109 If any lines were generated, set up a new current_line.
111 if not self.current_line:
112 self.current_line.depth += indent
113 return # Line is empty, don't emit. Creating a new one unnecessary.
116 Preview.improved_async_statements_handling in self.mode
117 and len(self.current_line.leaves) == 1
118 and is_async_stmt_or_funcdef(self.current_line.leaves[0])
120 # Special case for async def/for/with statements. `visit_async_stmt`
121 # adds an `ASYNC` leaf then visits the child def/for/with statement
122 # nodes. Line yields from those nodes shouldn't treat the former
123 # `ASYNC` leaf as a complete line.
126 complete_line = self.current_line
127 self.current_line = Line(mode=self.mode, depth=complete_line.depth + indent)
130 def visit_default(self, node: LN) -> Iterator[Line]:
131 """Default `visit_*()` implementation. Recurses to children of `node`."""
132 if isinstance(node, Leaf):
133 any_open_brackets = self.current_line.bracket_tracker.any_open_brackets()
134 for comment in generate_comments(node):
135 if any_open_brackets:
136 # any comment within brackets is subject to splitting
137 self.current_line.append(comment)
138 elif comment.type == token.COMMENT:
139 # regular trailing comment
140 self.current_line.append(comment)
141 yield from self.line()
144 # regular standalone comment
145 yield from self.line()
147 self.current_line.append(comment)
148 yield from self.line()
150 normalize_prefix(node, inside_brackets=any_open_brackets)
151 if self.mode.string_normalization and node.type == token.STRING:
152 node.value = normalize_string_prefix(node.value)
153 node.value = normalize_string_quotes(node.value)
154 if node.type == token.NUMBER:
155 normalize_numeric_literal(node)
156 if node.type not in WHITESPACE:
157 self.current_line.append(node)
158 yield from super().visit_default(node)
160 def visit_test(self, node: Node) -> Iterator[Line]:
161 """Visit an `x if y else z` test"""
163 if Preview.parenthesize_conditional_expressions in self.mode:
164 already_parenthesized = (
165 node.prev_sibling and node.prev_sibling.type == token.LPAR
168 if not already_parenthesized:
169 lpar = Leaf(token.LPAR, "")
170 rpar = Leaf(token.RPAR, "")
171 node.insert_child(0, lpar)
172 node.append_child(rpar)
174 yield from self.visit_default(node)
176 def visit_INDENT(self, node: Leaf) -> Iterator[Line]:
177 """Increase indentation level, maybe yield a line."""
178 # In blib2to3 INDENT never holds comments.
179 yield from self.line(+1)
180 yield from self.visit_default(node)
182 def visit_DEDENT(self, node: Leaf) -> Iterator[Line]:
183 """Decrease indentation level, maybe yield a line."""
184 # The current line might still wait for trailing comments. At DEDENT time
185 # there won't be any (they would be prefixes on the preceding NEWLINE).
186 # Emit the line then.
187 yield from self.line()
189 # While DEDENT has no value, its prefix may contain standalone comments
190 # that belong to the current indentation level. Get 'em.
191 yield from self.visit_default(node)
193 # Finally, emit the dedent.
194 yield from self.line(-1)
197 self, node: Node, keywords: Set[str], parens: Set[str]
199 """Visit a statement.
201 This implementation is shared for `if`, `while`, `for`, `try`, `except`,
202 `def`, `with`, `class`, `assert`, and assignments.
204 The relevant Python language `keywords` for a given statement will be
205 NAME leaves within it. This methods puts those on a separate line.
207 `parens` holds a set of string leaf values immediately after which
208 invisible parens should be put.
210 normalize_invisible_parens(
211 node, parens_after=parens, mode=self.mode, features=self.features
213 for child in node.children:
214 if is_name_token(child) and child.value in keywords:
215 yield from self.line()
217 yield from self.visit(child)
219 def visit_typeparams(self, node: Node) -> Iterator[Line]:
220 yield from self.visit_default(node)
221 node.children[0].prefix = ""
223 def visit_typevartuple(self, node: Node) -> Iterator[Line]:
224 yield from self.visit_default(node)
225 node.children[1].prefix = ""
227 def visit_paramspec(self, node: Node) -> Iterator[Line]:
228 yield from self.visit_default(node)
229 node.children[1].prefix = ""
231 def visit_dictsetmaker(self, node: Node) -> Iterator[Line]:
232 if Preview.wrap_long_dict_values_in_parens in self.mode:
233 for i, child in enumerate(node.children):
236 if node.children[i - 1].type == token.COLON:
237 if child.type == syms.atom and child.children[0].type == token.LPAR:
238 if maybe_make_parens_invisible_in_atom(
241 remove_brackets_around_comma=False,
243 wrap_in_parentheses(node, child, visible=False)
245 wrap_in_parentheses(node, child, visible=False)
246 yield from self.visit_default(node)
248 def visit_funcdef(self, node: Node) -> Iterator[Line]:
249 """Visit function definition."""
250 yield from self.line()
252 # Remove redundant brackets around return type annotation.
253 is_return_annotation = False
254 for child in node.children:
255 if child.type == token.RARROW:
256 is_return_annotation = True
257 elif is_return_annotation:
258 if child.type == syms.atom and child.children[0].type == token.LPAR:
259 if maybe_make_parens_invisible_in_atom(
262 remove_brackets_around_comma=False,
264 wrap_in_parentheses(node, child, visible=False)
266 wrap_in_parentheses(node, child, visible=False)
267 is_return_annotation = False
269 for child in node.children:
270 yield from self.visit(child)
272 def visit_match_case(self, node: Node) -> Iterator[Line]:
273 """Visit either a match or case statement."""
274 normalize_invisible_parens(
275 node, parens_after=set(), mode=self.mode, features=self.features
278 yield from self.line()
279 for child in node.children:
280 yield from self.visit(child)
282 def visit_suite(self, node: Node) -> Iterator[Line]:
285 self.mode.is_pyi or Preview.dummy_implementations in self.mode
286 ) and is_stub_suite(node):
287 yield from self.visit(node.children[2])
289 yield from self.visit_default(node)
291 def visit_simple_stmt(self, node: Node) -> Iterator[Line]:
292 """Visit a statement without nested statements."""
293 prev_type: Optional[int] = None
294 for child in node.children:
295 if (prev_type is None or prev_type == token.SEMI) and is_arith_like(child):
296 wrap_in_parentheses(node, child, visible=False)
297 prev_type = child.type
299 is_suite_like = node.parent and node.parent.type in STATEMENT
302 self.mode.is_pyi or Preview.dummy_implementations in self.mode
303 ) and is_stub_body(node):
304 yield from self.visit_default(node)
306 yield from self.line(+1)
307 yield from self.visit_default(node)
308 yield from self.line(-1)
312 not (self.mode.is_pyi or Preview.dummy_implementations in self.mode)
314 or not is_stub_suite(node.parent)
316 yield from self.line()
317 yield from self.visit_default(node)
319 def visit_async_stmt(self, node: Node) -> Iterator[Line]:
320 """Visit `async def`, `async for`, `async with`."""
321 yield from self.line()
323 children = iter(node.children)
324 for child in children:
325 yield from self.visit(child)
327 if child.type == token.ASYNC or child.type == STANDALONE_COMMENT:
328 # STANDALONE_COMMENT happens when `# fmt: skip` is applied on the async
332 internal_stmt = next(children)
333 if Preview.improved_async_statements_handling in self.mode:
334 yield from self.visit(internal_stmt)
336 for child in internal_stmt.children:
337 yield from self.visit(child)
339 def visit_decorators(self, node: Node) -> Iterator[Line]:
340 """Visit decorators."""
341 for child in node.children:
342 yield from self.line()
343 yield from self.visit(child)
345 def visit_power(self, node: Node) -> Iterator[Line]:
346 for idx, leaf in enumerate(node.children[:-1]):
347 next_leaf = node.children[idx + 1]
349 if not isinstance(leaf, Leaf):
352 value = leaf.value.lower()
354 leaf.type == token.NUMBER
355 and next_leaf.type == syms.trailer
356 # Ensure that we are in an attribute trailer
357 and next_leaf.children[0].type == token.DOT
358 # It shouldn't wrap hexadecimal, binary and octal literals
359 and not value.startswith(("0x", "0b", "0o"))
360 # It shouldn't wrap complex literals
363 wrap_in_parentheses(node, leaf)
365 remove_await_parens(node)
367 yield from self.visit_default(node)
369 def visit_SEMI(self, leaf: Leaf) -> Iterator[Line]:
370 """Remove a semicolon and put the other statement on a separate line."""
371 yield from self.line()
373 def visit_ENDMARKER(self, leaf: Leaf) -> Iterator[Line]:
374 """End of file. Process outstanding comments and end with a newline."""
375 yield from self.visit_default(leaf)
376 yield from self.line()
378 def visit_STANDALONE_COMMENT(self, leaf: Leaf) -> Iterator[Line]:
379 if not self.current_line.bracket_tracker.any_open_brackets():
380 yield from self.line()
381 yield from self.visit_default(leaf)
383 def visit_factor(self, node: Node) -> Iterator[Line]:
384 """Force parentheses between a unary op and a binary power:
388 _operator, operand = node.children
390 operand.type == syms.power
391 and len(operand.children) == 3
392 and operand.children[1].type == token.DOUBLESTAR
394 lpar = Leaf(token.LPAR, "(")
395 rpar = Leaf(token.RPAR, ")")
396 index = operand.remove() or 0
397 node.insert_child(index, Node(syms.atom, [lpar, operand, rpar]))
398 yield from self.visit_default(node)
400 def visit_STRING(self, leaf: Leaf) -> Iterator[Line]:
401 if Preview.hex_codes_in_unicode_sequences in self.mode:
402 normalize_unicode_escape_sequences(leaf)
404 if is_docstring(leaf) and "\\\n" not in leaf.value:
405 # We're ignoring docstrings with backslash newline escapes because changing
406 # indentation of those changes the AST representation of the code.
407 if self.mode.string_normalization:
408 docstring = normalize_string_prefix(leaf.value)
409 # visit_default() does handle string normalization for us, but
410 # since this method acts differently depending on quote style (ex.
411 # see padding logic below), there's a possibility for unstable
412 # formatting as visit_default() is called *after*. To avoid a
413 # situation where this function formats a docstring differently on
414 # the second pass, normalize it early.
415 docstring = normalize_string_quotes(docstring)
417 docstring = leaf.value
418 prefix = get_string_prefix(docstring)
419 docstring = docstring[len(prefix) :] # Remove the prefix
420 quote_char = docstring[0]
421 # A natural way to remove the outer quotes is to do:
422 # docstring = docstring.strip(quote_char)
423 # but that breaks on """""x""" (which is '""x').
424 # So we actually need to remove the first character and the next two
425 # characters but only if they are the same as the first.
426 quote_len = 1 if docstring[1] != quote_char else 3
427 docstring = docstring[quote_len:-quote_len]
428 docstring_started_empty = not docstring
429 indent = " " * 4 * self.current_line.depth
431 if is_multiline_string(leaf):
432 docstring = fix_docstring(docstring, indent)
434 docstring = docstring.strip()
436 has_trailing_backslash = False
438 # Add some padding if the docstring starts / ends with a quote mark.
439 if docstring[0] == quote_char:
440 docstring = " " + docstring
441 if docstring[-1] == quote_char:
443 if docstring[-1] == "\\":
444 backslash_count = len(docstring) - len(docstring.rstrip("\\"))
445 if backslash_count % 2:
446 # Odd number of tailing backslashes, add some padding to
447 # avoid escaping the closing string quote.
449 has_trailing_backslash = True
450 elif not docstring_started_empty:
453 # We could enforce triple quotes at this point.
454 quote = quote_char * quote_len
456 # It's invalid to put closing single-character quotes on a new line.
457 if self.mode and quote_len == 3:
458 # We need to find the length of the last line of the docstring
459 # to find if we can add the closing quotes to the line without
460 # exceeding the maximum line length.
461 # If docstring is one line, we don't put the closing quotes on a
462 # separate line because it looks ugly (#3320).
463 lines = docstring.splitlines()
464 last_line_length = len(lines[-1]) if docstring else 0
466 # If adding closing quotes would cause the last line to exceed
467 # the maximum line length then put a line break before the
471 and last_line_length + quote_len > self.mode.line_length
472 and len(indent) + quote_len <= self.mode.line_length
473 and not has_trailing_backslash
475 leaf.value = prefix + quote + docstring + "\n" + indent + quote
477 leaf.value = prefix + quote + docstring + quote
479 leaf.value = prefix + quote + docstring + quote
481 yield from self.visit_default(leaf)
483 def __post_init__(self) -> None:
484 """You are in a twisty little maze of passages."""
485 self.current_line = Line(mode=self.mode)
489 self.visit_assert_stmt = partial(v, keywords={"assert"}, parens={"assert", ","})
490 self.visit_if_stmt = partial(
491 v, keywords={"if", "else", "elif"}, parens={"if", "elif"}
493 self.visit_while_stmt = partial(v, keywords={"while", "else"}, parens={"while"})
494 self.visit_for_stmt = partial(v, keywords={"for", "else"}, parens={"for", "in"})
495 self.visit_try_stmt = partial(
496 v, keywords={"try", "except", "else", "finally"}, parens=Ø
498 self.visit_except_clause = partial(v, keywords={"except"}, parens={"except"})
499 self.visit_with_stmt = partial(v, keywords={"with"}, parens={"with"})
500 self.visit_classdef = partial(v, keywords={"class"}, parens=Ø)
501 self.visit_expr_stmt = partial(v, keywords=Ø, parens=ASSIGNMENTS)
502 self.visit_return_stmt = partial(v, keywords={"return"}, parens={"return"})
503 self.visit_import_from = partial(v, keywords=Ø, parens={"import"})
504 self.visit_del_stmt = partial(v, keywords=Ø, parens={"del"})
505 self.visit_async_funcdef = self.visit_async_stmt
506 self.visit_decorated = self.visit_decorators
509 self.visit_match_stmt = self.visit_match_case
510 self.visit_case_block = self.visit_match_case
514 line: Line, mode: Mode, features: Collection[Feature] = ()
516 """Transform a `line`, potentially splitting it into many lines.
518 They should fit in the allotted `line_length` but might not be able to.
520 `features` are syntactical features that may be used in the output.
526 line_str = line_to_string(line)
528 ll = mode.line_length
529 sn = mode.string_normalization
530 string_merge = StringMerger(ll, sn)
531 string_paren_strip = StringParenStripper(ll, sn)
532 string_split = StringSplitter(ll, sn)
533 string_paren_wrap = StringParenWrapper(ll, sn)
535 transformers: List[Transformer]
537 not line.contains_uncollapsable_type_comments()
538 and not line.should_split_rhs
539 and not line.magic_trailing_comma
541 is_line_short_enough(line, mode=mode, line_str=line_str)
542 or line.contains_unsplittable_type_ignore()
544 and not (line.inside_brackets and line.contains_standalone_comments())
546 # Only apply basic string preprocessing, since lines shouldn't be split here.
547 if Preview.string_processing in mode:
548 transformers = [string_merge, string_paren_strip]
552 transformers = [left_hand_split]
556 self: object, line: Line, features: Collection[Feature], mode: Mode
558 """Wraps calls to `right_hand_split`.
560 The calls increasingly `omit` right-hand trailers (bracket pairs with
561 content), meaning the trailers get glued together to split on another
562 bracket pair instead.
564 for omit in generate_trailers_to_omit(line, mode.line_length):
565 lines = list(right_hand_split(line, mode, features, omit=omit))
566 # Note: this check is only able to figure out if the first line of the
567 # *current* transformation fits in the line length. This is true only
568 # for simple cases. All others require running more transforms via
569 # `transform_line()`. This check doesn't know if those would succeed.
570 if is_line_short_enough(lines[0], mode=mode):
574 # All splits failed, best effort split with no omits.
575 # This mostly happens to multiline strings that are by definition
576 # reported as not fitting a single line, as well as lines that contain
577 # trailing commas (those have to be exploded).
578 yield from right_hand_split(line, mode, features=features)
580 # HACK: nested functions (like _rhs) compiled by mypyc don't retain their
581 # __name__ attribute which is needed in `run_transformer` further down.
582 # Unfortunately a nested class breaks mypyc too. So a class must be created
583 # via type ... https://github.com/mypyc/mypyc/issues/884
584 rhs = type("rhs", (), {"__call__": _rhs})()
586 if Preview.string_processing in mode:
587 if line.inside_brackets:
593 standalone_comment_split,
606 if line.inside_brackets:
607 transformers = [delimiter_split, standalone_comment_split, rhs]
610 # It's always safe to attempt hugging of power operations and pretty much every line
612 transformers.append(hug_power_op)
614 for transform in transformers:
615 # We are accumulating lines in `result` because we might want to abort
616 # mission and return the original line in the end, or attempt a different
619 result = run_transformer(line, transform, mode, features, line_str=line_str)
620 except CannotTransform:
630 class _BracketSplitComponent(Enum):
637 line: Line, _features: Collection[Feature], mode: Mode
639 """Split line into many lines, starting with the first matching bracket pair.
641 Note: this usually looks weird, only use this for function definitions.
642 Prefer RHS otherwise. This is why this function is not symmetrical with
643 :func:`right_hand_split` which also handles optional parentheses.
645 tail_leaves: List[Leaf] = []
646 body_leaves: List[Leaf] = []
647 head_leaves: List[Leaf] = []
648 current_leaves = head_leaves
649 matching_bracket: Optional[Leaf] = None
650 for leaf in line.leaves:
652 current_leaves is body_leaves
653 and leaf.type in CLOSING_BRACKETS
654 and leaf.opening_bracket is matching_bracket
655 and isinstance(matching_bracket, Leaf)
658 ensure_visible(matching_bracket)
659 current_leaves = tail_leaves if body_leaves else head_leaves
660 current_leaves.append(leaf)
661 if current_leaves is head_leaves:
662 if leaf.type in OPENING_BRACKETS:
663 matching_bracket = leaf
664 current_leaves = body_leaves
665 if not matching_bracket:
666 raise CannotSplit("No brackets found")
668 head = bracket_split_build_line(
669 head_leaves, line, matching_bracket, component=_BracketSplitComponent.head
671 body = bracket_split_build_line(
672 body_leaves, line, matching_bracket, component=_BracketSplitComponent.body
674 tail = bracket_split_build_line(
675 tail_leaves, line, matching_bracket, component=_BracketSplitComponent.tail
677 bracket_split_succeeded_or_raise(head, body, tail)
678 for result in (head, body, tail):
683 def right_hand_split(
686 features: Collection[Feature] = (),
687 omit: Collection[LeafID] = (),
689 """Split line into many lines, starting with the last matching bracket pair.
691 If the split was by optional parentheses, attempt splitting without them, too.
692 `omit` is a collection of closing bracket IDs that shouldn't be considered for
695 Note: running this function modifies `bracket_depth` on the leaves of `line`.
697 rhs_result = _first_right_hand_split(line, omit=omit)
698 yield from _maybe_split_omitting_optional_parens(
699 rhs_result, line, mode, features=features, omit=omit
703 def _first_right_hand_split(
705 omit: Collection[LeafID] = (),
707 """Split the line into head, body, tail starting with the last bracket pair.
709 Note: this function should not have side effects. It's relied upon by
710 _maybe_split_omitting_optional_parens to get an opinion whether to prefer
711 splitting on the right side of an assignment statement.
713 tail_leaves: List[Leaf] = []
714 body_leaves: List[Leaf] = []
715 head_leaves: List[Leaf] = []
716 current_leaves = tail_leaves
717 opening_bracket: Optional[Leaf] = None
718 closing_bracket: Optional[Leaf] = None
719 for leaf in reversed(line.leaves):
720 if current_leaves is body_leaves:
721 if leaf is opening_bracket:
722 current_leaves = head_leaves if body_leaves else tail_leaves
723 current_leaves.append(leaf)
724 if current_leaves is tail_leaves:
725 if leaf.type in CLOSING_BRACKETS and id(leaf) not in omit:
726 opening_bracket = leaf.opening_bracket
727 closing_bracket = leaf
728 current_leaves = body_leaves
729 if not (opening_bracket and closing_bracket and head_leaves):
730 # If there is no opening or closing_bracket that means the split failed and
731 # all content is in the tail. Otherwise, if `head_leaves` are empty, it means
732 # the matching `opening_bracket` wasn't available on `line` anymore.
733 raise CannotSplit("No brackets found")
735 tail_leaves.reverse()
736 body_leaves.reverse()
737 head_leaves.reverse()
738 head = bracket_split_build_line(
739 head_leaves, line, opening_bracket, component=_BracketSplitComponent.head
741 body = bracket_split_build_line(
742 body_leaves, line, opening_bracket, component=_BracketSplitComponent.body
744 tail = bracket_split_build_line(
745 tail_leaves, line, opening_bracket, component=_BracketSplitComponent.tail
747 bracket_split_succeeded_or_raise(head, body, tail)
748 return RHSResult(head, body, tail, opening_bracket, closing_bracket)
751 def _maybe_split_omitting_optional_parens(
755 features: Collection[Feature] = (),
756 omit: Collection[LeafID] = (),
759 Feature.FORCE_OPTIONAL_PARENTHESES not in features
760 # the opening bracket is an optional paren
761 and rhs.opening_bracket.type == token.LPAR
762 and not rhs.opening_bracket.value
763 # the closing bracket is an optional paren
764 and rhs.closing_bracket.type == token.RPAR
765 and not rhs.closing_bracket.value
766 # it's not an import (optional parens are the only thing we can split on
767 # in this case; attempting a split without them is a waste of time)
768 and not line.is_import
769 # there are no standalone comments in the body
770 and not rhs.body.contains_standalone_comments(0)
771 # and we can actually remove the parens
772 and can_omit_invisible_parens(rhs, mode.line_length)
774 omit = {id(rhs.closing_bracket), *omit}
776 # The RHSResult Omitting Optional Parens.
777 rhs_oop = _first_right_hand_split(line, omit=omit)
779 Preview.prefer_splitting_right_hand_side_of_assignments in line.mode
780 # the split is right after `=`
781 and len(rhs.head.leaves) >= 2
782 and rhs.head.leaves[-2].type == token.EQUAL
783 # the left side of assignment contains brackets
784 and any(leaf.type in BRACKETS for leaf in rhs.head.leaves[:-1])
785 # the left side of assignment is short enough (the -1 is for the ending
787 and is_line_short_enough(
788 rhs.head, mode=replace(mode, line_length=mode.line_length - 1)
790 # the left side of assignment won't explode further because of magic
792 and rhs.head.magic_trailing_comma is None
793 # the split by omitting optional parens isn't preferred by some other
795 and not _prefer_split_rhs_oop(rhs_oop, mode)
797 yield from _maybe_split_omitting_optional_parens(
798 rhs_oop, line, mode, features=features, omit=omit
802 except CannotSplit as e:
804 can_be_split(rhs.body) or is_line_short_enough(rhs.body, mode=mode)
807 "Splitting failed, body is still too long and can't be split."
811 rhs.head.contains_multiline_strings()
812 or rhs.tail.contains_multiline_strings()
815 "The current optional pair of parentheses is bound to fail to"
816 " satisfy the splitting algorithm because the head or the tail"
817 " contains multiline strings which by definition never fit one"
821 ensure_visible(rhs.opening_bracket)
822 ensure_visible(rhs.closing_bracket)
823 for result in (rhs.head, rhs.body, rhs.tail):
828 def _prefer_split_rhs_oop(rhs_oop: RHSResult, mode: Mode) -> bool:
830 Returns whether we should prefer the result from a split omitting optional parens.
832 has_closing_bracket_after_assign = False
833 for leaf in reversed(rhs_oop.head.leaves):
834 if leaf.type == token.EQUAL:
836 if leaf.type in CLOSING_BRACKETS:
837 has_closing_bracket_after_assign = True
840 # contains matching brackets after the `=` (done by checking there is a
842 has_closing_bracket_after_assign
844 # the split is actually from inside the optional parens (done by checking
845 # the first line still contains the `=`)
846 any(leaf.type == token.EQUAL for leaf in rhs_oop.head.leaves)
847 # the first line is short enough
848 and is_line_short_enough(rhs_oop.head, mode=mode)
850 # contains unsplittable type ignore
851 or rhs_oop.head.contains_unsplittable_type_ignore()
852 or rhs_oop.body.contains_unsplittable_type_ignore()
853 or rhs_oop.tail.contains_unsplittable_type_ignore()
857 def bracket_split_succeeded_or_raise(head: Line, body: Line, tail: Line) -> None:
858 """Raise :exc:`CannotSplit` if the last left- or right-hand split failed.
860 Do nothing otherwise.
862 A left- or right-hand split is based on a pair of brackets. Content before
863 (and including) the opening bracket is left on one line, content inside the
864 brackets is put on a separate line, and finally content starting with and
865 following the closing bracket is put on a separate line.
867 Those are called `head`, `body`, and `tail`, respectively. If the split
868 produced the same line (all content in `head`) or ended up with an empty `body`
869 and the `tail` is just the closing bracket, then it's considered failed.
871 tail_len = len(str(tail).strip())
874 raise CannotSplit("Splitting brackets produced the same line")
878 f"Splitting brackets on an empty body to save {tail_len} characters is"
883 def bracket_split_build_line(
886 opening_bracket: Leaf,
888 component: _BracketSplitComponent,
890 """Return a new line with given `leaves` and respective comments from `original`.
892 If it's the head component, brackets will be tracked so trailing commas are
895 If it's the body component, the result line is one-indented inside brackets and as
896 such has its first leaf's prefix normalized and a trailing comma added when
899 result = Line(mode=original.mode, depth=original.depth)
900 if component is _BracketSplitComponent.body:
901 result.inside_brackets = True
904 # Since body is a new indent level, remove spurious leading whitespace.
905 normalize_prefix(leaves[0], inside_brackets=True)
906 # Ensure a trailing comma for imports and standalone function arguments, but
907 # be careful not to add one after any comments or within type annotations.
910 and opening_bracket.value == "("
911 and not any(leaf.type == token.COMMA for leaf in leaves)
912 # In particular, don't add one within a parenthesized return annotation.
913 # Unfortunately the indicator we're in a return annotation (RARROW) may
914 # be defined directly in the parent node, the parent of the parent ...
915 # and so on depending on how complex the return annotation is.
916 # This isn't perfect and there's some false negatives but they are in
917 # contexts were a comma is actually fine.
919 node.prev_sibling.type == RARROW
922 getattr(leaves[0].parent, "parent", None),
924 if isinstance(node, Node) and isinstance(node.prev_sibling, Leaf)
926 # Except the false negatives above for PEP 604 unions where we
927 # can't add the comma.
930 and leaves[0].parent.next_sibling
931 and leaves[0].parent.next_sibling.type == token.VBAR
935 if original.is_import or no_commas:
936 for i in range(len(leaves) - 1, -1, -1):
937 if leaves[i].type == STANDALONE_COMMENT:
940 if leaves[i].type != token.COMMA:
941 new_comma = Leaf(token.COMMA, ",")
942 leaves.insert(i + 1, new_comma)
945 leaves_to_track: Set[LeafID] = set()
946 if component is _BracketSplitComponent.head:
947 leaves_to_track = get_leaves_inside_matching_brackets(leaves)
953 track_bracket=id(leaf) in leaves_to_track,
955 for comment_after in original.comments_after(leaf):
956 result.append(comment_after, preformatted=True)
957 if component is _BracketSplitComponent.body and should_split_line(
958 result, opening_bracket
960 result.should_split_rhs = True
964 def dont_increase_indentation(split_func: Transformer) -> Transformer:
965 """Normalize prefix of the first leaf in every line returned by `split_func`.
967 This is a decorator over relevant split functions.
972 line: Line, features: Collection[Feature], mode: Mode
974 for split_line in split_func(line, features, mode):
975 normalize_prefix(split_line.leaves[0], inside_brackets=True)
981 def _get_last_non_comment_leaf(line: Line) -> Optional[int]:
982 for leaf_idx in range(len(line.leaves) - 1, 0, -1):
983 if line.leaves[leaf_idx].type != STANDALONE_COMMENT:
988 def _safe_add_trailing_comma(safe: bool, delimiter_priority: int, line: Line) -> Line:
991 and delimiter_priority == COMMA_PRIORITY
992 and line.leaves[-1].type != token.COMMA
993 and line.leaves[-1].type != STANDALONE_COMMENT
995 new_comma = Leaf(token.COMMA, ",")
996 line.append(new_comma)
1000 @dont_increase_indentation
1001 def delimiter_split(
1002 line: Line, features: Collection[Feature], mode: Mode
1003 ) -> Iterator[Line]:
1004 """Split according to delimiters of the highest priority.
1006 If the appropriate Features are given, the split will add trailing commas
1007 also in function signatures and calls that contain `*` and `**`.
1010 last_leaf = line.leaves[-1]
1012 raise CannotSplit("Line empty") from None
1014 bt = line.bracket_tracker
1016 delimiter_priority = bt.max_delimiter_priority(exclude={id(last_leaf)})
1018 raise CannotSplit("No delimiters found") from None
1020 if delimiter_priority == DOT_PRIORITY:
1021 if bt.delimiter_count_with_priority(delimiter_priority) == 1:
1022 raise CannotSplit("Splitting a single attribute from its owner looks wrong")
1024 current_line = Line(
1025 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1027 lowest_depth = sys.maxsize
1028 trailing_comma_safe = True
1030 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1031 """Append `leaf` to current line or to new line if appending impossible."""
1032 nonlocal current_line
1034 current_line.append_safe(leaf, preformatted=True)
1038 current_line = Line(
1039 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1041 current_line.append(leaf)
1043 last_non_comment_leaf = _get_last_non_comment_leaf(line)
1044 for leaf_idx, leaf in enumerate(line.leaves):
1045 yield from append_to_line(leaf)
1047 for comment_after in line.comments_after(leaf):
1048 yield from append_to_line(comment_after)
1050 lowest_depth = min(lowest_depth, leaf.bracket_depth)
1051 if leaf.bracket_depth == lowest_depth:
1052 if is_vararg(leaf, within={syms.typedargslist}):
1053 trailing_comma_safe = (
1054 trailing_comma_safe and Feature.TRAILING_COMMA_IN_DEF in features
1056 elif is_vararg(leaf, within={syms.arglist, syms.argument}):
1057 trailing_comma_safe = (
1058 trailing_comma_safe and Feature.TRAILING_COMMA_IN_CALL in features
1062 Preview.add_trailing_comma_consistently in mode
1063 and last_leaf.type == STANDALONE_COMMENT
1064 and leaf_idx == last_non_comment_leaf
1066 current_line = _safe_add_trailing_comma(
1067 trailing_comma_safe, delimiter_priority, current_line
1070 leaf_priority = bt.delimiters.get(id(leaf))
1071 if leaf_priority == delimiter_priority:
1074 current_line = Line(
1075 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1078 current_line = _safe_add_trailing_comma(
1079 trailing_comma_safe, delimiter_priority, current_line
1084 @dont_increase_indentation
1085 def standalone_comment_split(
1086 line: Line, features: Collection[Feature], mode: Mode
1087 ) -> Iterator[Line]:
1088 """Split standalone comments from the rest of the line."""
1089 if not line.contains_standalone_comments(0):
1090 raise CannotSplit("Line does not have any standalone comments")
1092 current_line = Line(
1093 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1096 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1097 """Append `leaf` to current line or to new line if appending impossible."""
1098 nonlocal current_line
1100 current_line.append_safe(leaf, preformatted=True)
1104 current_line = Line(
1105 line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1107 current_line.append(leaf)
1109 for leaf in line.leaves:
1110 yield from append_to_line(leaf)
1112 for comment_after in line.comments_after(leaf):
1113 yield from append_to_line(comment_after)
1119 def normalize_prefix(leaf: Leaf, *, inside_brackets: bool) -> None:
1120 """Leave existing extra newlines if not `inside_brackets`. Remove everything
1123 Note: don't use backslashes for formatting or you'll lose your voting rights.
1125 if not inside_brackets:
1126 spl = leaf.prefix.split("#")
1127 if "\\" not in spl[0]:
1128 nl_count = spl[-1].count("\n")
1131 leaf.prefix = "\n" * nl_count
1137 def normalize_invisible_parens(
1138 node: Node, parens_after: Set[str], *, mode: Mode, features: Collection[Feature]
1140 """Make existing optional parentheses invisible or create new ones.
1142 `parens_after` is a set of string leaf values immediately after which parens
1145 Standardizes on visible parentheses for single-element tuples, and keeps
1146 existing visible parentheses for other tuples and generator expressions.
1148 for pc in list_comments(node.prefix, is_endmarker=False):
1149 if pc.value in FMT_OFF:
1150 # This `node` has a prefix with `# fmt: off`, don't mess with parens.
1153 # The multiple context managers grammar has a different pattern, thus this is
1154 # separate from the for-loop below. This possibly wraps them in invisible parens,
1155 # and later will be removed in remove_with_parens when needed.
1156 if node.type == syms.with_stmt:
1157 _maybe_wrap_cms_in_parens(node, mode, features)
1160 for index, child in enumerate(list(node.children)):
1161 # Fixes a bug where invisible parens are not properly stripped from
1162 # assignment statements that contain type annotations.
1163 if isinstance(child, Node) and child.type == syms.annassign:
1164 normalize_invisible_parens(
1165 child, parens_after=parens_after, mode=mode, features=features
1168 # Add parentheses around long tuple unpacking in assignments.
1171 and isinstance(child, Node)
1172 and child.type == syms.testlist_star_expr
1178 child.type == syms.atom
1179 and node.type == syms.for_stmt
1180 and isinstance(child.prev_sibling, Leaf)
1181 and child.prev_sibling.type == token.NAME
1182 and child.prev_sibling.value == "for"
1184 if maybe_make_parens_invisible_in_atom(
1187 remove_brackets_around_comma=True,
1189 wrap_in_parentheses(node, child, visible=False)
1190 elif isinstance(child, Node) and node.type == syms.with_stmt:
1191 remove_with_parens(child, node)
1192 elif child.type == syms.atom:
1193 if maybe_make_parens_invisible_in_atom(
1197 wrap_in_parentheses(node, child, visible=False)
1198 elif is_one_tuple(child):
1199 wrap_in_parentheses(node, child, visible=True)
1200 elif node.type == syms.import_from:
1201 _normalize_import_from(node, child, index)
1205 and child.type == token.STAR
1206 and node.type == syms.except_clause
1208 # In except* (PEP 654), the star is actually part of
1209 # of the keyword. So we need to skip the insertion of
1210 # invisible parentheses to work more precisely.
1213 elif not (isinstance(child, Leaf) and is_multiline_string(child)):
1214 wrap_in_parentheses(node, child, visible=False)
1216 comma_check = child.type == token.COMMA
1218 check_lpar = isinstance(child, Leaf) and (
1219 child.value in parens_after or comma_check
1223 def _normalize_import_from(parent: Node, child: LN, index: int) -> None:
1224 # "import from" nodes store parentheses directly as part of
1226 if is_lpar_token(child):
1227 assert is_rpar_token(parent.children[-1])
1228 # make parentheses invisible
1230 parent.children[-1].value = ""
1231 elif child.type != token.STAR:
1232 # insert invisible parentheses
1233 parent.insert_child(index, Leaf(token.LPAR, ""))
1234 parent.append_child(Leaf(token.RPAR, ""))
1237 def remove_await_parens(node: Node) -> None:
1238 if node.children[0].type == token.AWAIT and len(node.children) > 1:
1240 node.children[1].type == syms.atom
1241 and node.children[1].children[0].type == token.LPAR
1243 if maybe_make_parens_invisible_in_atom(
1246 remove_brackets_around_comma=True,
1248 wrap_in_parentheses(node, node.children[1], visible=False)
1250 # Since await is an expression we shouldn't remove
1251 # brackets in cases where this would change
1252 # the AST due to operator precedence.
1253 # Therefore we only aim to remove brackets around
1254 # power nodes that aren't also await expressions themselves.
1255 # https://peps.python.org/pep-0492/#updated-operator-precedence-table
1256 # N.B. We've still removed any redundant nested brackets though :)
1257 opening_bracket = cast(Leaf, node.children[1].children[0])
1258 closing_bracket = cast(Leaf, node.children[1].children[-1])
1259 bracket_contents = node.children[1].children[1]
1260 if isinstance(bracket_contents, Node):
1261 if bracket_contents.type != syms.power:
1262 ensure_visible(opening_bracket)
1263 ensure_visible(closing_bracket)
1265 bracket_contents.type == syms.power
1266 and bracket_contents.children[0].type == token.AWAIT
1268 ensure_visible(opening_bracket)
1269 ensure_visible(closing_bracket)
1270 # If we are in a nested await then recurse down.
1271 remove_await_parens(bracket_contents)
1274 def _maybe_wrap_cms_in_parens(
1275 node: Node, mode: Mode, features: Collection[Feature]
1277 """When enabled and safe, wrap the multiple context managers in invisible parens.
1279 It is only safe when `features` contain Feature.PARENTHESIZED_CONTEXT_MANAGERS.
1282 Feature.PARENTHESIZED_CONTEXT_MANAGERS not in features
1283 or Preview.wrap_multiple_context_managers_in_parens not in mode
1284 or len(node.children) <= 2
1285 # If it's an atom, it's already wrapped in parens.
1286 or node.children[1].type == syms.atom
1289 colon_index: Optional[int] = None
1290 for i in range(2, len(node.children)):
1291 if node.children[i].type == token.COLON:
1294 if colon_index is not None:
1295 lpar = Leaf(token.LPAR, "")
1296 rpar = Leaf(token.RPAR, "")
1297 context_managers = node.children[1:colon_index]
1298 for child in context_managers:
1300 # After wrapping, the with_stmt will look like this:
1306 # ... <-- context_managers
1312 syms.atom, [lpar, Node(syms.testlist_gexp, context_managers), rpar]
1314 node.insert_child(1, new_child)
1317 def remove_with_parens(node: Node, parent: Node) -> None:
1318 """Recursively hide optional parens in `with` statements."""
1319 # Removing all unnecessary parentheses in with statements in one pass is a tad
1320 # complex as different variations of bracketed statements result in pretty
1321 # different parse trees:
1323 # with (open("file")) as f: # this is an asexpr_test
1326 # with (open("file") as f): # this is an atom containing an
1329 # with (open("file")) as f, (open("file")) as f: # this is asexpr_test, COMMA,
1332 # with (open("file") as f, open("file") as f): # an atom containing a
1333 # ... # testlist_gexp which then
1334 # # contains multiple asexpr_test(s)
1335 if node.type == syms.atom:
1336 if maybe_make_parens_invisible_in_atom(
1339 remove_brackets_around_comma=True,
1341 wrap_in_parentheses(parent, node, visible=False)
1342 if isinstance(node.children[1], Node):
1343 remove_with_parens(node.children[1], node)
1344 elif node.type == syms.testlist_gexp:
1345 for child in node.children:
1346 if isinstance(child, Node):
1347 remove_with_parens(child, node)
1348 elif node.type == syms.asexpr_test and not any(
1349 leaf.type == token.COLONEQUAL for leaf in node.leaves()
1351 if maybe_make_parens_invisible_in_atom(
1354 remove_brackets_around_comma=True,
1356 wrap_in_parentheses(node, node.children[0], visible=False)
1359 def maybe_make_parens_invisible_in_atom(
1362 remove_brackets_around_comma: bool = False,
1364 """If it's safe, make the parens in the atom `node` invisible, recursively.
1365 Additionally, remove repeated, adjacent invisible parens from the atom `node`
1366 as they are redundant.
1368 Returns whether the node should itself be wrapped in invisible parentheses.
1371 node.type != syms.atom
1372 or is_empty_tuple(node)
1373 or is_one_tuple(node)
1374 or (is_yield(node) and parent.type != syms.expr_stmt)
1376 # This condition tries to prevent removing non-optional brackets
1377 # around a tuple, however, can be a bit overzealous so we provide
1378 # and option to skip this check for `for` and `with` statements.
1379 not remove_brackets_around_comma
1380 and max_delimiter_priority_in_atom(node) >= COMMA_PRIORITY
1382 or is_tuple_containing_walrus(node)
1386 if is_walrus_assignment(node):
1395 # these ones aren't useful to end users, but they do please fuzzers
1402 first = node.children[0]
1403 last = node.children[-1]
1404 if is_lpar_token(first) and is_rpar_token(last):
1405 middle = node.children[1]
1406 # make parentheses invisible
1408 # If the prefix of `middle` includes a type comment with
1409 # ignore annotation, then we do not remove the parentheses
1410 not is_type_ignore_comment_string(middle.prefix.strip())
1414 maybe_make_parens_invisible_in_atom(
1417 remove_brackets_around_comma=remove_brackets_around_comma,
1420 if is_atom_with_invisible_parens(middle):
1421 # Strip the invisible parens from `middle` by replacing
1422 # it with the child in-between the invisible parens
1423 middle.replace(middle.children[1])
1430 def should_split_line(line: Line, opening_bracket: Leaf) -> bool:
1431 """Should `line` be immediately split with `delimiter_split()` after RHS?"""
1433 if not (opening_bracket.parent and opening_bracket.value in "[{("):
1436 # We're essentially checking if the body is delimited by commas and there's more
1437 # than one of them (we're excluding the trailing comma and if the delimiter priority
1438 # is still commas, that means there's more).
1440 trailing_comma = False
1442 last_leaf = line.leaves[-1]
1443 if last_leaf.type == token.COMMA:
1444 trailing_comma = True
1445 exclude.add(id(last_leaf))
1446 max_priority = line.bracket_tracker.max_delimiter_priority(exclude=exclude)
1447 except (IndexError, ValueError):
1450 return max_priority == COMMA_PRIORITY and (
1451 (line.mode.magic_trailing_comma and trailing_comma)
1452 # always explode imports
1453 or opening_bracket.parent.type in {syms.atom, syms.import_from}
1457 def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[LeafID]]:
1458 """Generate sets of closing bracket IDs that should be omitted in a RHS.
1460 Brackets can be omitted if the entire trailer up to and including
1461 a preceding closing bracket fits in one line.
1463 Yielded sets are cumulative (contain results of previous yields, too). First
1464 set is empty, unless the line should explode, in which case bracket pairs until
1465 the one that needs to explode are omitted.
1468 omit: Set[LeafID] = set()
1469 if not line.magic_trailing_comma:
1472 length = 4 * line.depth
1473 opening_bracket: Optional[Leaf] = None
1474 closing_bracket: Optional[Leaf] = None
1475 inner_brackets: Set[LeafID] = set()
1476 for index, leaf, leaf_length in line.enumerate_with_length(reversed=True):
1477 length += leaf_length
1478 if length > line_length:
1481 has_inline_comment = leaf_length > len(leaf.value) + len(leaf.prefix)
1482 if leaf.type == STANDALONE_COMMENT or has_inline_comment:
1486 if leaf is opening_bracket:
1487 opening_bracket = None
1488 elif leaf.type in CLOSING_BRACKETS:
1489 prev = line.leaves[index - 1] if index > 0 else None
1492 and prev.type == token.COMMA
1493 and leaf.opening_bracket is not None
1494 and not is_one_sequence_between(
1495 leaf.opening_bracket, leaf, line.leaves
1498 # Never omit bracket pairs with trailing commas.
1499 # We need to explode on those.
1502 inner_brackets.add(id(leaf))
1503 elif leaf.type in CLOSING_BRACKETS:
1504 prev = line.leaves[index - 1] if index > 0 else None
1505 if prev and prev.type in OPENING_BRACKETS:
1506 # Empty brackets would fail a split so treat them as "inner"
1507 # brackets (e.g. only add them to the `omit` set if another
1508 # pair of brackets was good enough.
1509 inner_brackets.add(id(leaf))
1513 omit.add(id(closing_bracket))
1514 omit.update(inner_brackets)
1515 inner_brackets.clear()
1520 and prev.type == token.COMMA
1521 and leaf.opening_bracket is not None
1522 and not is_one_sequence_between(leaf.opening_bracket, leaf, line.leaves)
1524 # Never omit bracket pairs with trailing commas.
1525 # We need to explode on those.
1529 opening_bracket = leaf.opening_bracket
1530 closing_bracket = leaf
1533 def run_transformer(
1535 transform: Transformer,
1537 features: Collection[Feature],
1542 line_str = line_to_string(line)
1543 result: List[Line] = []
1544 for transformed_line in transform(line, features, mode):
1545 if str(transformed_line).strip("\n") == line_str:
1546 raise CannotTransform("Line transformer returned an unchanged result")
1548 result.extend(transform_line(transformed_line, mode=mode, features=features))
1550 features_set = set(features)
1552 Feature.FORCE_OPTIONAL_PARENTHESES in features_set
1553 or transform.__class__.__name__ != "rhs"
1554 or not line.bracket_tracker.invisible
1555 or any(bracket.value for bracket in line.bracket_tracker.invisible)
1556 or line.contains_multiline_strings()
1557 or result[0].contains_uncollapsable_type_comments()
1558 or result[0].contains_unsplittable_type_ignore()
1559 or is_line_short_enough(result[0], mode=mode)
1560 # If any leaves have no parents (which _can_ occur since
1561 # `transform(line)` potentially destroys the line's underlying node
1562 # structure), then we can't proceed. Doing so would cause the below
1563 # call to `append_leaves()` to fail.
1564 or any(leaf.parent is None for leaf in line.leaves)
1568 line_copy = line.clone()
1569 append_leaves(line_copy, line, line.leaves)
1570 features_fop = features_set | {Feature.FORCE_OPTIONAL_PARENTHESES}
1571 second_opinion = run_transformer(
1572 line_copy, transform, mode, features_fop, line_str=line_str
1574 if all(is_line_short_enough(ln, mode=mode) for ln in second_opinion):
1575 result = second_opinion