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]:
284 if self.mode.is_pyi and is_stub_suite(node):
285 yield from self.visit(node.children[2])
287 yield from self.visit_default(node)
289 def visit_simple_stmt(self, node: Node) -> Iterator[Line]:
290 """Visit a statement without nested statements."""
291 prev_type: Optional[int] = None
292 for child in node.children:
293 if (prev_type is None or prev_type == token.SEMI) and is_arith_like(child):
294 wrap_in_parentheses(node, child, visible=False)
295 prev_type = child.type
297 is_suite_like = node.parent and node.parent.type in STATEMENT
299 if self.mode.is_pyi and is_stub_body(node):
300 yield from self.visit_default(node)
302 yield from self.line(+1)
303 yield from self.visit_default(node)
304 yield from self.line(-1)
310 or not is_stub_suite(node.parent)
312 yield from self.line()
313 yield from self.visit_default(node)
315 def visit_async_stmt(self, node: Node) -> Iterator[Line]:
316 """Visit `async def`, `async for`, `async with`."""
317 yield from self.line()
319 children = iter(node.children)
320 for child in children:
321 yield from self.visit(child)
323 if child.type == token.ASYNC or child.type == STANDALONE_COMMENT:
324 # STANDALONE_COMMENT happens when `# fmt: skip` is applied on the async
328 internal_stmt = next(children)
329 if Preview.improved_async_statements_handling in self.mode:
330 yield from self.visit(internal_stmt)
332 for child in internal_stmt.children:
333 yield from self.visit(child)
335 def visit_decorators(self, node: Node) -> Iterator[Line]:
336 """Visit decorators."""
337 for child in node.children:
338 yield from self.line()
339 yield from self.visit(child)
341 def visit_power(self, node: Node) -> Iterator[Line]:
342 for idx, leaf in enumerate(node.children[:-1]):
343 next_leaf = node.children[idx + 1]
345 if not isinstance(leaf, Leaf):
348 value = leaf.value.lower()
350 leaf.type == token.NUMBER
351 and next_leaf.type == syms.trailer
352 # Ensure that we are in an attribute trailer
353 and next_leaf.children[0].type == token.DOT
354 # It shouldn't wrap hexadecimal, binary and octal literals
355 and not value.startswith(("0x", "0b", "0o"))
356 # It shouldn't wrap complex literals
359 wrap_in_parentheses(node, leaf)
361 remove_await_parens(node)
363 yield from self.visit_default(node)
365 def visit_SEMI(self, leaf: Leaf) -> Iterator[Line]:
366 """Remove a semicolon and put the other statement on a separate line."""
367 yield from self.line()
369 def visit_ENDMARKER(self, leaf: Leaf) -> Iterator[Line]:
370 """End of file. Process outstanding comments and end with a newline."""
371 yield from self.visit_default(leaf)
372 yield from self.line()
374 def visit_STANDALONE_COMMENT(self, leaf: Leaf) -> Iterator[Line]:
375 if not self.current_line.bracket_tracker.any_open_brackets():
376 yield from self.line()
377 yield from self.visit_default(leaf)
379 def visit_factor(self, node: Node) -> Iterator[Line]:
380 """Force parentheses between a unary op and a binary power:
384 _operator, operand = node.children
386 operand.type == syms.power
387 and len(operand.children) == 3
388 and operand.children[1].type == token.DOUBLESTAR
390 lpar = Leaf(token.LPAR, "(")
391 rpar = Leaf(token.RPAR, ")")
392 index = operand.remove() or 0
393 node.insert_child(index, Node(syms.atom, [lpar, operand, rpar]))
394 yield from self.visit_default(node)
396 def visit_STRING(self, leaf: Leaf) -> Iterator[Line]:
397 if Preview.hex_codes_in_unicode_sequences in self.mode:
398 normalize_unicode_escape_sequences(leaf)
400 if is_docstring(leaf) and "\\\n" not in leaf.value:
401 # We're ignoring docstrings with backslash newline escapes because changing
402 # indentation of those changes the AST representation of the code.
403 if self.mode.string_normalization:
404 docstring = normalize_string_prefix(leaf.value)
405 # visit_default() does handle string normalization for us, but
406 # since this method acts differently depending on quote style (ex.
407 # see padding logic below), there's a possibility for unstable
408 # formatting as visit_default() is called *after*. To avoid a
409 # situation where this function formats a docstring differently on
410 # the second pass, normalize it early.
411 docstring = normalize_string_quotes(docstring)
413 docstring = leaf.value
414 prefix = get_string_prefix(docstring)
415 docstring = docstring[len(prefix) :] # Remove the prefix
416 quote_char = docstring[0]
417 # A natural way to remove the outer quotes is to do:
418 # docstring = docstring.strip(quote_char)
419 # but that breaks on """""x""" (which is '""x').
420 # So we actually need to remove the first character and the next two
421 # characters but only if they are the same as the first.
422 quote_len = 1 if docstring[1] != quote_char else 3
423 docstring = docstring[quote_len:-quote_len]
424 docstring_started_empty = not docstring
425 indent = " " * 4 * self.current_line.depth
427 if is_multiline_string(leaf):
428 docstring = fix_docstring(docstring, indent)
430 docstring = docstring.strip()
432 has_trailing_backslash = False
434 # Add some padding if the docstring starts / ends with a quote mark.
435 if docstring[0] == quote_char:
436 docstring = " " + docstring
437 if docstring[-1] == quote_char:
439 if docstring[-1] == "\\":
440 backslash_count = len(docstring) - len(docstring.rstrip("\\"))
441 if backslash_count % 2:
442 # Odd number of tailing backslashes, add some padding to
443 # avoid escaping the closing string quote.
445 has_trailing_backslash = True
446 elif not docstring_started_empty:
449 # We could enforce triple quotes at this point.
450 quote = quote_char * quote_len
452 # It's invalid to put closing single-character quotes on a new line.
453 if self.mode and quote_len == 3:
454 # We need to find the length of the last line of the docstring
455 # to find if we can add the closing quotes to the line without
456 # exceeding the maximum line length.
457 # If docstring is one line, we don't put the closing quotes on a
458 # separate line because it looks ugly (#3320).
459 lines = docstring.splitlines()
460 last_line_length = len(lines[-1]) if docstring else 0
462 # If adding closing quotes would cause the last line to exceed
463 # the maximum line length then put a line break before the
467 and last_line_length + quote_len > self.mode.line_length
468 and len(indent) + quote_len <= self.mode.line_length
469 and not has_trailing_backslash
471 leaf.value = prefix + quote + docstring + "\n" + indent + quote
473 leaf.value = prefix + quote + docstring + quote
475 leaf.value = prefix + quote + docstring + quote
477 yield from self.visit_default(leaf)
479 def __post_init__(self) -> None:
480 """You are in a twisty little maze of passages."""
481 self.current_line = Line(mode=self.mode)
485 self.visit_assert_stmt = partial(v, keywords={"assert"}, parens={"assert", ","})
486 self.visit_if_stmt = partial(
487 v, keywords={"if", "else", "elif"}, parens={"if", "elif"}
489 self.visit_while_stmt = partial(v, keywords={"while", "else"}, parens={"while"})
490 self.visit_for_stmt = partial(v, keywords={"for", "else"}, parens={"for", "in"})
491 self.visit_try_stmt = partial(
492 v, keywords={"try", "except", "else", "finally"}, parens=Ø
494 self.visit_except_clause = partial(v, keywords={"except"}, parens={"except"})
495 self.visit_with_stmt = partial(v, keywords={"with"}, parens={"with"})
496 self.visit_classdef = partial(v, keywords={"class"}, parens=Ø)
497 self.visit_expr_stmt = partial(v, keywords=Ø, parens=ASSIGNMENTS)
498 self.visit_return_stmt = partial(v, keywords={"return"}, parens={"return"})
499 self.visit_import_from = partial(v, keywords=Ø, parens={"import"})
500 self.visit_del_stmt = partial(v, keywords=Ø, parens={"del"})
501 self.visit_async_funcdef = self.visit_async_stmt
502 self.visit_decorated = self.visit_decorators
505 self.visit_match_stmt = self.visit_match_case
506 self.visit_case_block = self.visit_match_case
510 line: Line, mode: Mode, features: Collection[Feature] = ()
512 """Transform a `line`, potentially splitting it into many lines.
514 They should fit in the allotted `line_length` but might not be able to.
516 `features` are syntactical features that may be used in the output.
522 line_str = line_to_string(line)
524 ll = mode.line_length
525 sn = mode.string_normalization
526 string_merge = StringMerger(ll, sn)
527 string_paren_strip = StringParenStripper(ll, sn)
528 string_split = StringSplitter(ll, sn)
529 string_paren_wrap = StringParenWrapper(ll, sn)
531 transformers: List[Transformer]
533 not line.contains_uncollapsable_type_comments()
534 and not line.should_split_rhs
535 and not line.magic_trailing_comma
537 is_line_short_enough(line, mode=mode, line_str=line_str)
538 or line.contains_unsplittable_type_ignore()
540 and not (line.inside_brackets and line.contains_standalone_comments())
542 # Only apply basic string preprocessing, since lines shouldn't be split here.
543 if Preview.string_processing in mode:
544 transformers = [string_merge, string_paren_strip]
548 transformers = [left_hand_split]
552 self: object, line: Line, features: Collection[Feature], mode: Mode
554 """Wraps calls to `right_hand_split`.
556 The calls increasingly `omit` right-hand trailers (bracket pairs with
557 content), meaning the trailers get glued together to split on another
558 bracket pair instead.
560 for omit in generate_trailers_to_omit(line, mode.line_length):
561 lines = list(right_hand_split(line, mode, features, omit=omit))
562 # Note: this check is only able to figure out if the first line of the
563 # *current* transformation fits in the line length. This is true only
564 # for simple cases. All others require running more transforms via
565 # `transform_line()`. This check doesn't know if those would succeed.
566 if is_line_short_enough(lines[0], mode=mode):
570 # All splits failed, best effort split with no omits.
571 # This mostly happens to multiline strings that are by definition
572 # reported as not fitting a single line, as well as lines that contain
573 # trailing commas (those have to be exploded).
574 yield from right_hand_split(line, mode, features=features)
576 # HACK: nested functions (like _rhs) compiled by mypyc don't retain their
577 # __name__ attribute which is needed in `run_transformer` further down.
578 # Unfortunately a nested class breaks mypyc too. So a class must be created
579 # via type ... https://github.com/mypyc/mypyc/issues/884
580 rhs = type("rhs", (), {"__call__": _rhs})()
582 if Preview.string_processing in mode:
583 if line.inside_brackets:
589 standalone_comment_split,
602 if line.inside_brackets:
603 transformers = [delimiter_split, standalone_comment_split, rhs]
606 # It's always safe to attempt hugging of power operations and pretty much every line
608 transformers.append(hug_power_op)
610 for transform in transformers:
611 # We are accumulating lines in `result` because we might want to abort
612 # mission and return the original line in the end, or attempt a different
615 result = run_transformer(line, transform, mode, features, line_str=line_str)
616 except CannotTransform:
626 class _BracketSplitComponent(Enum):
633 line: Line, _features: Collection[Feature], mode: Mode
635 """Split line into many lines, starting with the first matching bracket pair.
637 Note: this usually looks weird, only use this for function definitions.
638 Prefer RHS otherwise. This is why this function is not symmetrical with
639 :func:`right_hand_split` which also handles optional parentheses.
641 tail_leaves: List[Leaf] = []
642 body_leaves: List[Leaf] = []
643 head_leaves: List[Leaf] = []
644 current_leaves = head_leaves
645 matching_bracket: Optional[Leaf] = None
646 for leaf in line.leaves:
648 current_leaves is body_leaves
649 and leaf.type in CLOSING_BRACKETS
650 and leaf.opening_bracket is matching_bracket
651 and isinstance(matching_bracket, Leaf)
654 ensure_visible(matching_bracket)
655 current_leaves = tail_leaves if body_leaves else head_leaves
656 current_leaves.append(leaf)
657 if current_leaves is head_leaves:
658 if leaf.type in OPENING_BRACKETS:
659 matching_bracket = leaf
660 current_leaves = body_leaves
661 if not matching_bracket:
662 raise CannotSplit("No brackets found")
664 head = bracket_split_build_line(
665 head_leaves, line, matching_bracket, component=_BracketSplitComponent.head
667 body = bracket_split_build_line(
668 body_leaves, line, matching_bracket, component=_BracketSplitComponent.body
670 tail = bracket_split_build_line(
671 tail_leaves, line, matching_bracket, component=_BracketSplitComponent.tail
673 bracket_split_succeeded_or_raise(head, body, tail)
674 for result in (head, body, tail):
679 def right_hand_split(
682 features: Collection[Feature] = (),
683 omit: Collection[LeafID] = (),
685 """Split line into many lines, starting with the last matching bracket pair.
687 If the split was by optional parentheses, attempt splitting without them, too.
688 `omit` is a collection of closing bracket IDs that shouldn't be considered for
691 Note: running this function modifies `bracket_depth` on the leaves of `line`.
693 rhs_result = _first_right_hand_split(line, omit=omit)
694 yield from _maybe_split_omitting_optional_parens(
695 rhs_result, line, mode, features=features, omit=omit
699 def _first_right_hand_split(
701 omit: Collection[LeafID] = (),
703 """Split the line into head, body, tail starting with the last bracket pair.
705 Note: this function should not have side effects. It's relied upon by
706 _maybe_split_omitting_optional_parens to get an opinion whether to prefer
707 splitting on the right side of an assignment statement.
709 tail_leaves: List[Leaf] = []
710 body_leaves: List[Leaf] = []
711 head_leaves: List[Leaf] = []
712 current_leaves = tail_leaves
713 opening_bracket: Optional[Leaf] = None
714 closing_bracket: Optional[Leaf] = None
715 for leaf in reversed(line.leaves):
716 if current_leaves is body_leaves:
717 if leaf is opening_bracket:
718 current_leaves = head_leaves if body_leaves else tail_leaves
719 current_leaves.append(leaf)
720 if current_leaves is tail_leaves:
721 if leaf.type in CLOSING_BRACKETS and id(leaf) not in omit:
722 opening_bracket = leaf.opening_bracket
723 closing_bracket = leaf
724 current_leaves = body_leaves
725 if not (opening_bracket and closing_bracket and head_leaves):
726 # If there is no opening or closing_bracket that means the split failed and
727 # all content is in the tail. Otherwise, if `head_leaves` are empty, it means
728 # the matching `opening_bracket` wasn't available on `line` anymore.
729 raise CannotSplit("No brackets found")
731 tail_leaves.reverse()
732 body_leaves.reverse()
733 head_leaves.reverse()
734 head = bracket_split_build_line(
735 head_leaves, line, opening_bracket, component=_BracketSplitComponent.head
737 body = bracket_split_build_line(
738 body_leaves, line, opening_bracket, component=_BracketSplitComponent.body
740 tail = bracket_split_build_line(
741 tail_leaves, line, opening_bracket, component=_BracketSplitComponent.tail
743 bracket_split_succeeded_or_raise(head, body, tail)
744 return RHSResult(head, body, tail, opening_bracket, closing_bracket)
747 def _maybe_split_omitting_optional_parens(
751 features: Collection[Feature] = (),
752 omit: Collection[LeafID] = (),
755 Feature.FORCE_OPTIONAL_PARENTHESES not in features
756 # the opening bracket is an optional paren
757 and rhs.opening_bracket.type == token.LPAR
758 and not rhs.opening_bracket.value
759 # the closing bracket is an optional paren
760 and rhs.closing_bracket.type == token.RPAR
761 and not rhs.closing_bracket.value
762 # it's not an import (optional parens are the only thing we can split on
763 # in this case; attempting a split without them is a waste of time)
764 and not line.is_import
765 # there are no standalone comments in the body
766 and not rhs.body.contains_standalone_comments(0)
767 # and we can actually remove the parens
768 and can_omit_invisible_parens(rhs, mode.line_length)
770 omit = {id(rhs.closing_bracket), *omit}
772 # The RHSResult Omitting Optional Parens.
773 rhs_oop = _first_right_hand_split(line, omit=omit)
775 Preview.prefer_splitting_right_hand_side_of_assignments in line.mode
776 # the split is right after `=`
777 and len(rhs.head.leaves) >= 2
778 and rhs.head.leaves[-2].type == token.EQUAL
779 # the left side of assignment contains brackets
780 and any(leaf.type in BRACKETS for leaf in rhs.head.leaves[:-1])
781 # the left side of assignment is short enough (the -1 is for the ending
783 and is_line_short_enough(
784 rhs.head, mode=replace(mode, line_length=mode.line_length - 1)
786 # the left side of assignment won't explode further because of magic
788 and rhs.head.magic_trailing_comma is None
789 # the split by omitting optional parens isn't preferred by some other
791 and not _prefer_split_rhs_oop(rhs_oop, mode)
793 yield from _maybe_split_omitting_optional_parens(
794 rhs_oop, line, mode, features=features, omit=omit
798 except CannotSplit as e:
800 can_be_split(rhs.body) or is_line_short_enough(rhs.body, mode=mode)
803 "Splitting failed, body is still too long and can't be split."
807 rhs.head.contains_multiline_strings()
808 or rhs.tail.contains_multiline_strings()
811 "The current optional pair of parentheses is bound to fail to"
812 " satisfy the splitting algorithm because the head or the tail"
813 " contains multiline strings which by definition never fit one"
817 ensure_visible(rhs.opening_bracket)
818 ensure_visible(rhs.closing_bracket)
819 for result in (rhs.head, rhs.body, rhs.tail):
824 def _prefer_split_rhs_oop(rhs_oop: RHSResult, mode: Mode) -> bool:
826 Returns whether we should prefer the result from a split omitting optional parens.
828 has_closing_bracket_after_assign = False
829 for leaf in reversed(rhs_oop.head.leaves):
830 if leaf.type == token.EQUAL:
832 if leaf.type in CLOSING_BRACKETS:
833 has_closing_bracket_after_assign = True
836 # contains matching brackets after the `=` (done by checking there is a
838 has_closing_bracket_after_assign
840 # the split is actually from inside the optional parens (done by checking
841 # the first line still contains the `=`)
842 any(leaf.type == token.EQUAL for leaf in rhs_oop.head.leaves)
843 # the first line is short enough
844 and is_line_short_enough(rhs_oop.head, mode=mode)
846 # contains unsplittable type ignore
847 or rhs_oop.head.contains_unsplittable_type_ignore()
848 or rhs_oop.body.contains_unsplittable_type_ignore()
849 or rhs_oop.tail.contains_unsplittable_type_ignore()
853 def bracket_split_succeeded_or_raise(head: Line, body: Line, tail: Line) -> None:
854 """Raise :exc:`CannotSplit` if the last left- or right-hand split failed.
856 Do nothing otherwise.
858 A left- or right-hand split is based on a pair of brackets. Content before
859 (and including) the opening bracket is left on one line, content inside the
860 brackets is put on a separate line, and finally content starting with and
861 following the closing bracket is put on a separate line.
863 Those are called `head`, `body`, and `tail`, respectively. If the split
864 produced the same line (all content in `head`) or ended up with an empty `body`
865 and the `tail` is just the closing bracket, then it's considered failed.
867 tail_len = len(str(tail).strip())
870 raise CannotSplit("Splitting brackets produced the same line")
874 f"Splitting brackets on an empty body to save {tail_len} characters is"
879 def bracket_split_build_line(
882 opening_bracket: Leaf,
884 component: _BracketSplitComponent,
886 """Return a new line with given `leaves` and respective comments from `original`.
888 If it's the head component, brackets will be tracked so trailing commas are
891 If it's the body component, the result line is one-indented inside brackets and as
892 such has its first leaf's prefix normalized and a trailing comma added when
895 result = Line(mode=original.mode, depth=original.depth)
896 if component is _BracketSplitComponent.body:
897 result.inside_brackets = True
900 # Since body is a new indent level, remove spurious leading whitespace.
901 normalize_prefix(leaves[0], inside_brackets=True)
902 # Ensure a trailing comma for imports and standalone function arguments, but
903 # be careful not to add one after any comments or within type annotations.
906 and opening_bracket.value == "("
907 and not any(leaf.type == token.COMMA for leaf in leaves)
908 # In particular, don't add one within a parenthesized return annotation.
909 # Unfortunately the indicator we're in a return annotation (RARROW) may
910 # be defined directly in the parent node, the parent of the parent ...
911 # and so on depending on how complex the return annotation is.
912 # This isn't perfect and there's some false negatives but they are in
913 # contexts were a comma is actually fine.
915 node.prev_sibling.type == RARROW
918 getattr(leaves[0].parent, "parent", None),
920 if isinstance(node, Node) and isinstance(node.prev_sibling, Leaf)
922 # Except the false negatives above for PEP 604 unions where we
923 # can't add the comma.
926 and leaves[0].parent.next_sibling
927 and leaves[0].parent.next_sibling.type == token.VBAR
931 if original.is_import or no_commas:
932 for i in range(len(leaves) - 1, -1, -1):
933 if leaves[i].type == STANDALONE_COMMENT:
936 if leaves[i].type != token.COMMA:
937 new_comma = Leaf(token.COMMA, ",")
938 leaves.insert(i + 1, new_comma)
941 leaves_to_track: Set[LeafID] = set()
942 if component is _BracketSplitComponent.head:
943 leaves_to_track = get_leaves_inside_matching_brackets(leaves)
949 track_bracket=id(leaf) in leaves_to_track,
951 for comment_after in original.comments_after(leaf):
952 result.append(comment_after, preformatted=True)
953 if component is _BracketSplitComponent.body and should_split_line(
954 result, opening_bracket
956 result.should_split_rhs = True
960 def dont_increase_indentation(split_func: Transformer) -> Transformer:
961 """Normalize prefix of the first leaf in every line returned by `split_func`.
963 This is a decorator over relevant split functions.
968 line: Line, features: Collection[Feature], mode: Mode
970 for split_line in split_func(line, features, mode):
971 normalize_prefix(split_line.leaves[0], inside_brackets=True)
977 def _get_last_non_comment_leaf(line: Line) -> Optional[int]:
978 for leaf_idx in range(len(line.leaves) - 1, 0, -1):
979 if line.leaves[leaf_idx].type != STANDALONE_COMMENT:
984 def _safe_add_trailing_comma(safe: bool, delimiter_priority: int, line: Line) -> Line:
987 and delimiter_priority == COMMA_PRIORITY
988 and line.leaves[-1].type != token.COMMA
989 and line.leaves[-1].type != STANDALONE_COMMENT
991 new_comma = Leaf(token.COMMA, ",")
992 line.append(new_comma)
996 @dont_increase_indentation
998 line: Line, features: Collection[Feature], mode: Mode
1000 """Split according to delimiters of the highest priority.
1002 If the appropriate Features are given, the split will add trailing commas
1003 also in function signatures and calls that contain `*` and `**`.
1006 last_leaf = line.leaves[-1]
1008 raise CannotSplit("Line empty") from None
1010 bt = line.bracket_tracker
1012 delimiter_priority = bt.max_delimiter_priority(exclude={id(last_leaf)})
1014 raise CannotSplit("No delimiters found") from None
1016 if delimiter_priority == DOT_PRIORITY:
1017 if bt.delimiter_count_with_priority(delimiter_priority) == 1:
1018 raise CannotSplit("Splitting a single attribute from its owner looks wrong")
1020 current_line = Line(
1021 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1023 lowest_depth = sys.maxsize
1024 trailing_comma_safe = True
1026 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1027 """Append `leaf` to current line or to new line if appending impossible."""
1028 nonlocal current_line
1030 current_line.append_safe(leaf, preformatted=True)
1034 current_line = Line(
1035 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1037 current_line.append(leaf)
1039 last_non_comment_leaf = _get_last_non_comment_leaf(line)
1040 for leaf_idx, leaf in enumerate(line.leaves):
1041 yield from append_to_line(leaf)
1043 for comment_after in line.comments_after(leaf):
1044 yield from append_to_line(comment_after)
1046 lowest_depth = min(lowest_depth, leaf.bracket_depth)
1047 if leaf.bracket_depth == lowest_depth:
1048 if is_vararg(leaf, within={syms.typedargslist}):
1049 trailing_comma_safe = (
1050 trailing_comma_safe and Feature.TRAILING_COMMA_IN_DEF in features
1052 elif is_vararg(leaf, within={syms.arglist, syms.argument}):
1053 trailing_comma_safe = (
1054 trailing_comma_safe and Feature.TRAILING_COMMA_IN_CALL in features
1058 Preview.add_trailing_comma_consistently in mode
1059 and last_leaf.type == STANDALONE_COMMENT
1060 and leaf_idx == last_non_comment_leaf
1062 current_line = _safe_add_trailing_comma(
1063 trailing_comma_safe, delimiter_priority, current_line
1066 leaf_priority = bt.delimiters.get(id(leaf))
1067 if leaf_priority == delimiter_priority:
1070 current_line = Line(
1071 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1074 current_line = _safe_add_trailing_comma(
1075 trailing_comma_safe, delimiter_priority, current_line
1080 @dont_increase_indentation
1081 def standalone_comment_split(
1082 line: Line, features: Collection[Feature], mode: Mode
1083 ) -> Iterator[Line]:
1084 """Split standalone comments from the rest of the line."""
1085 if not line.contains_standalone_comments(0):
1086 raise CannotSplit("Line does not have any standalone comments")
1088 current_line = Line(
1089 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1092 def append_to_line(leaf: Leaf) -> Iterator[Line]:
1093 """Append `leaf` to current line or to new line if appending impossible."""
1094 nonlocal current_line
1096 current_line.append_safe(leaf, preformatted=True)
1100 current_line = Line(
1101 line.mode, depth=line.depth, inside_brackets=line.inside_brackets
1103 current_line.append(leaf)
1105 for leaf in line.leaves:
1106 yield from append_to_line(leaf)
1108 for comment_after in line.comments_after(leaf):
1109 yield from append_to_line(comment_after)
1115 def normalize_prefix(leaf: Leaf, *, inside_brackets: bool) -> None:
1116 """Leave existing extra newlines if not `inside_brackets`. Remove everything
1119 Note: don't use backslashes for formatting or you'll lose your voting rights.
1121 if not inside_brackets:
1122 spl = leaf.prefix.split("#")
1123 if "\\" not in spl[0]:
1124 nl_count = spl[-1].count("\n")
1127 leaf.prefix = "\n" * nl_count
1133 def normalize_invisible_parens(
1134 node: Node, parens_after: Set[str], *, mode: Mode, features: Collection[Feature]
1136 """Make existing optional parentheses invisible or create new ones.
1138 `parens_after` is a set of string leaf values immediately after which parens
1141 Standardizes on visible parentheses for single-element tuples, and keeps
1142 existing visible parentheses for other tuples and generator expressions.
1144 for pc in list_comments(node.prefix, is_endmarker=False):
1145 if pc.value in FMT_OFF:
1146 # This `node` has a prefix with `# fmt: off`, don't mess with parens.
1149 # The multiple context managers grammar has a different pattern, thus this is
1150 # separate from the for-loop below. This possibly wraps them in invisible parens,
1151 # and later will be removed in remove_with_parens when needed.
1152 if node.type == syms.with_stmt:
1153 _maybe_wrap_cms_in_parens(node, mode, features)
1156 for index, child in enumerate(list(node.children)):
1157 # Fixes a bug where invisible parens are not properly stripped from
1158 # assignment statements that contain type annotations.
1159 if isinstance(child, Node) and child.type == syms.annassign:
1160 normalize_invisible_parens(
1161 child, parens_after=parens_after, mode=mode, features=features
1164 # Add parentheses around long tuple unpacking in assignments.
1167 and isinstance(child, Node)
1168 and child.type == syms.testlist_star_expr
1174 child.type == syms.atom
1175 and node.type == syms.for_stmt
1176 and isinstance(child.prev_sibling, Leaf)
1177 and child.prev_sibling.type == token.NAME
1178 and child.prev_sibling.value == "for"
1180 if maybe_make_parens_invisible_in_atom(
1183 remove_brackets_around_comma=True,
1185 wrap_in_parentheses(node, child, visible=False)
1186 elif isinstance(child, Node) and node.type == syms.with_stmt:
1187 remove_with_parens(child, node)
1188 elif child.type == syms.atom:
1189 if maybe_make_parens_invisible_in_atom(
1193 wrap_in_parentheses(node, child, visible=False)
1194 elif is_one_tuple(child):
1195 wrap_in_parentheses(node, child, visible=True)
1196 elif node.type == syms.import_from:
1197 _normalize_import_from(node, child, index)
1201 and child.type == token.STAR
1202 and node.type == syms.except_clause
1204 # In except* (PEP 654), the star is actually part of
1205 # of the keyword. So we need to skip the insertion of
1206 # invisible parentheses to work more precisely.
1209 elif not (isinstance(child, Leaf) and is_multiline_string(child)):
1210 wrap_in_parentheses(node, child, visible=False)
1212 comma_check = child.type == token.COMMA
1214 check_lpar = isinstance(child, Leaf) and (
1215 child.value in parens_after or comma_check
1219 def _normalize_import_from(parent: Node, child: LN, index: int) -> None:
1220 # "import from" nodes store parentheses directly as part of
1222 if is_lpar_token(child):
1223 assert is_rpar_token(parent.children[-1])
1224 # make parentheses invisible
1226 parent.children[-1].value = ""
1227 elif child.type != token.STAR:
1228 # insert invisible parentheses
1229 parent.insert_child(index, Leaf(token.LPAR, ""))
1230 parent.append_child(Leaf(token.RPAR, ""))
1233 def remove_await_parens(node: Node) -> None:
1234 if node.children[0].type == token.AWAIT and len(node.children) > 1:
1236 node.children[1].type == syms.atom
1237 and node.children[1].children[0].type == token.LPAR
1239 if maybe_make_parens_invisible_in_atom(
1242 remove_brackets_around_comma=True,
1244 wrap_in_parentheses(node, node.children[1], visible=False)
1246 # Since await is an expression we shouldn't remove
1247 # brackets in cases where this would change
1248 # the AST due to operator precedence.
1249 # Therefore we only aim to remove brackets around
1250 # power nodes that aren't also await expressions themselves.
1251 # https://peps.python.org/pep-0492/#updated-operator-precedence-table
1252 # N.B. We've still removed any redundant nested brackets though :)
1253 opening_bracket = cast(Leaf, node.children[1].children[0])
1254 closing_bracket = cast(Leaf, node.children[1].children[-1])
1255 bracket_contents = node.children[1].children[1]
1256 if isinstance(bracket_contents, Node):
1257 if bracket_contents.type != syms.power:
1258 ensure_visible(opening_bracket)
1259 ensure_visible(closing_bracket)
1261 bracket_contents.type == syms.power
1262 and bracket_contents.children[0].type == token.AWAIT
1264 ensure_visible(opening_bracket)
1265 ensure_visible(closing_bracket)
1266 # If we are in a nested await then recurse down.
1267 remove_await_parens(bracket_contents)
1270 def _maybe_wrap_cms_in_parens(
1271 node: Node, mode: Mode, features: Collection[Feature]
1273 """When enabled and safe, wrap the multiple context managers in invisible parens.
1275 It is only safe when `features` contain Feature.PARENTHESIZED_CONTEXT_MANAGERS.
1278 Feature.PARENTHESIZED_CONTEXT_MANAGERS not in features
1279 or Preview.wrap_multiple_context_managers_in_parens not in mode
1280 or len(node.children) <= 2
1281 # If it's an atom, it's already wrapped in parens.
1282 or node.children[1].type == syms.atom
1285 colon_index: Optional[int] = None
1286 for i in range(2, len(node.children)):
1287 if node.children[i].type == token.COLON:
1290 if colon_index is not None:
1291 lpar = Leaf(token.LPAR, "")
1292 rpar = Leaf(token.RPAR, "")
1293 context_managers = node.children[1:colon_index]
1294 for child in context_managers:
1296 # After wrapping, the with_stmt will look like this:
1302 # ... <-- context_managers
1308 syms.atom, [lpar, Node(syms.testlist_gexp, context_managers), rpar]
1310 node.insert_child(1, new_child)
1313 def remove_with_parens(node: Node, parent: Node) -> None:
1314 """Recursively hide optional parens in `with` statements."""
1315 # Removing all unnecessary parentheses in with statements in one pass is a tad
1316 # complex as different variations of bracketed statements result in pretty
1317 # different parse trees:
1319 # with (open("file")) as f: # this is an asexpr_test
1322 # with (open("file") as f): # this is an atom containing an
1325 # with (open("file")) as f, (open("file")) as f: # this is asexpr_test, COMMA,
1328 # with (open("file") as f, open("file") as f): # an atom containing a
1329 # ... # testlist_gexp which then
1330 # # contains multiple asexpr_test(s)
1331 if node.type == syms.atom:
1332 if maybe_make_parens_invisible_in_atom(
1335 remove_brackets_around_comma=True,
1337 wrap_in_parentheses(parent, node, visible=False)
1338 if isinstance(node.children[1], Node):
1339 remove_with_parens(node.children[1], node)
1340 elif node.type == syms.testlist_gexp:
1341 for child in node.children:
1342 if isinstance(child, Node):
1343 remove_with_parens(child, node)
1344 elif node.type == syms.asexpr_test and not any(
1345 leaf.type == token.COLONEQUAL for leaf in node.leaves()
1347 if maybe_make_parens_invisible_in_atom(
1350 remove_brackets_around_comma=True,
1352 wrap_in_parentheses(node, node.children[0], visible=False)
1355 def maybe_make_parens_invisible_in_atom(
1358 remove_brackets_around_comma: bool = False,
1360 """If it's safe, make the parens in the atom `node` invisible, recursively.
1361 Additionally, remove repeated, adjacent invisible parens from the atom `node`
1362 as they are redundant.
1364 Returns whether the node should itself be wrapped in invisible parentheses.
1367 node.type != syms.atom
1368 or is_empty_tuple(node)
1369 or is_one_tuple(node)
1370 or (is_yield(node) and parent.type != syms.expr_stmt)
1372 # This condition tries to prevent removing non-optional brackets
1373 # around a tuple, however, can be a bit overzealous so we provide
1374 # and option to skip this check for `for` and `with` statements.
1375 not remove_brackets_around_comma
1376 and max_delimiter_priority_in_atom(node) >= COMMA_PRIORITY
1378 or is_tuple_containing_walrus(node)
1382 if is_walrus_assignment(node):
1391 # these ones aren't useful to end users, but they do please fuzzers
1398 first = node.children[0]
1399 last = node.children[-1]
1400 if is_lpar_token(first) and is_rpar_token(last):
1401 middle = node.children[1]
1402 # make parentheses invisible
1404 # If the prefix of `middle` includes a type comment with
1405 # ignore annotation, then we do not remove the parentheses
1406 not is_type_ignore_comment_string(middle.prefix.strip())
1410 maybe_make_parens_invisible_in_atom(
1413 remove_brackets_around_comma=remove_brackets_around_comma,
1416 if is_atom_with_invisible_parens(middle):
1417 # Strip the invisible parens from `middle` by replacing
1418 # it with the child in-between the invisible parens
1419 middle.replace(middle.children[1])
1426 def should_split_line(line: Line, opening_bracket: Leaf) -> bool:
1427 """Should `line` be immediately split with `delimiter_split()` after RHS?"""
1429 if not (opening_bracket.parent and opening_bracket.value in "[{("):
1432 # We're essentially checking if the body is delimited by commas and there's more
1433 # than one of them (we're excluding the trailing comma and if the delimiter priority
1434 # is still commas, that means there's more).
1436 trailing_comma = False
1438 last_leaf = line.leaves[-1]
1439 if last_leaf.type == token.COMMA:
1440 trailing_comma = True
1441 exclude.add(id(last_leaf))
1442 max_priority = line.bracket_tracker.max_delimiter_priority(exclude=exclude)
1443 except (IndexError, ValueError):
1446 return max_priority == COMMA_PRIORITY and (
1447 (line.mode.magic_trailing_comma and trailing_comma)
1448 # always explode imports
1449 or opening_bracket.parent.type in {syms.atom, syms.import_from}
1453 def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[LeafID]]:
1454 """Generate sets of closing bracket IDs that should be omitted in a RHS.
1456 Brackets can be omitted if the entire trailer up to and including
1457 a preceding closing bracket fits in one line.
1459 Yielded sets are cumulative (contain results of previous yields, too). First
1460 set is empty, unless the line should explode, in which case bracket pairs until
1461 the one that needs to explode are omitted.
1464 omit: Set[LeafID] = set()
1465 if not line.magic_trailing_comma:
1468 length = 4 * line.depth
1469 opening_bracket: Optional[Leaf] = None
1470 closing_bracket: Optional[Leaf] = None
1471 inner_brackets: Set[LeafID] = set()
1472 for index, leaf, leaf_length in line.enumerate_with_length(reversed=True):
1473 length += leaf_length
1474 if length > line_length:
1477 has_inline_comment = leaf_length > len(leaf.value) + len(leaf.prefix)
1478 if leaf.type == STANDALONE_COMMENT or has_inline_comment:
1482 if leaf is opening_bracket:
1483 opening_bracket = None
1484 elif leaf.type in CLOSING_BRACKETS:
1485 prev = line.leaves[index - 1] if index > 0 else None
1488 and prev.type == token.COMMA
1489 and leaf.opening_bracket is not None
1490 and not is_one_sequence_between(
1491 leaf.opening_bracket, leaf, line.leaves
1494 # Never omit bracket pairs with trailing commas.
1495 # We need to explode on those.
1498 inner_brackets.add(id(leaf))
1499 elif leaf.type in CLOSING_BRACKETS:
1500 prev = line.leaves[index - 1] if index > 0 else None
1501 if prev and prev.type in OPENING_BRACKETS:
1502 # Empty brackets would fail a split so treat them as "inner"
1503 # brackets (e.g. only add them to the `omit` set if another
1504 # pair of brackets was good enough.
1505 inner_brackets.add(id(leaf))
1509 omit.add(id(closing_bracket))
1510 omit.update(inner_brackets)
1511 inner_brackets.clear()
1516 and prev.type == token.COMMA
1517 and leaf.opening_bracket is not None
1518 and not is_one_sequence_between(leaf.opening_bracket, leaf, line.leaves)
1520 # Never omit bracket pairs with trailing commas.
1521 # We need to explode on those.
1525 opening_bracket = leaf.opening_bracket
1526 closing_bracket = leaf
1529 def run_transformer(
1531 transform: Transformer,
1533 features: Collection[Feature],
1538 line_str = line_to_string(line)
1539 result: List[Line] = []
1540 for transformed_line in transform(line, features, mode):
1541 if str(transformed_line).strip("\n") == line_str:
1542 raise CannotTransform("Line transformer returned an unchanged result")
1544 result.extend(transform_line(transformed_line, mode=mode, features=features))
1546 features_set = set(features)
1548 Feature.FORCE_OPTIONAL_PARENTHESES in features_set
1549 or transform.__class__.__name__ != "rhs"
1550 or not line.bracket_tracker.invisible
1551 or any(bracket.value for bracket in line.bracket_tracker.invisible)
1552 or line.contains_multiline_strings()
1553 or result[0].contains_uncollapsable_type_comments()
1554 or result[0].contains_unsplittable_type_ignore()
1555 or is_line_short_enough(result[0], mode=mode)
1556 # If any leaves have no parents (which _can_ occur since
1557 # `transform(line)` potentially destroys the line's underlying node
1558 # structure), then we can't proceed. Doing so would cause the below
1559 # call to `append_leaves()` to fail.
1560 or any(leaf.parent is None for leaf in line.leaves)
1564 line_copy = line.clone()
1565 append_leaves(line_copy, line, line.leaves)
1566 features_fop = features_set | {Feature.FORCE_OPTIONAL_PARENTHESES}
1567 second_opinion = run_transformer(
1568 line_copy, transform, mode, features_fop, line_str=line_str
1570 if all(is_line_short_enough(ln, mode=mode) for ln in second_opinion):
1571 result = second_opinion