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 enum import Enum, auto
6 from functools import partial, wraps
7 from typing import Collection, Iterator, List, Optional, Set, Union, cast
9 from black.brackets import (
12 get_leaves_inside_matching_brackets,
13 max_delimiter_priority_in_atom,
15 from black.comments import FMT_OFF, generate_comments, list_comments
16 from black.lines import (
20 can_omit_invisible_parens,
24 from black.mode import Feature, Mode, Preview
25 from black.nodes import (
36 is_atom_with_invisible_parens,
42 is_one_sequence_between,
53 from black.numerics import normalize_numeric_literal
54 from black.strings import (
57 normalize_string_prefix,
58 normalize_string_quotes,
60 from black.trans import (
69 from blib2to3.pgen2 import token
70 from blib2to3.pytree import Leaf, Node
74 LN = Union[Leaf, Node]
77 class CannotSplit(CannotTransform):
78 """A readable split that fits the allotted line length is impossible."""
81 # This isn't a dataclass because @dataclass + Generic breaks mypyc.
82 # See also https://github.com/mypyc/mypyc/issues/827.
83 class LineGenerator(Visitor[Line]):
84 """Generates reformatted Line objects. Empty lines are not emitted.
86 Note: destroys the tree it's visiting by mutating prefixes of its leaves
87 in ways that will no longer stringify to valid Python code on the tree.
90 def __init__(self, mode: Mode) -> None:
92 self.current_line: Line
95 def line(self, indent: int = 0) -> Iterator[Line]:
98 If the line is empty, only emit if it makes sense.
99 If the line is too long, split it first and then generate.
101 If any lines were generated, set up a new current_line.
103 if not self.current_line:
104 self.current_line.depth += indent
105 return # Line is empty, don't emit. Creating a new one unnecessary.
107 complete_line = self.current_line
108 self.current_line = Line(mode=self.mode, depth=complete_line.depth + indent)
111 def visit_default(self, node: LN) -> Iterator[Line]:
112 """Default `visit_*()` implementation. Recurses to children of `node`."""
113 if isinstance(node, Leaf):
114 any_open_brackets = self.current_line.bracket_tracker.any_open_brackets()
115 for comment in generate_comments(node, preview=self.mode.preview):
116 if any_open_brackets:
117 # any comment within brackets is subject to splitting
118 self.current_line.append(comment)
119 elif comment.type == token.COMMENT:
120 # regular trailing comment
121 self.current_line.append(comment)
122 yield from self.line()
125 # regular standalone comment
126 yield from self.line()
128 self.current_line.append(comment)
129 yield from self.line()
131 normalize_prefix(node, inside_brackets=any_open_brackets)
132 if self.mode.string_normalization and node.type == token.STRING:
133 node.value = normalize_string_prefix(node.value)
134 node.value = normalize_string_quotes(node.value)
135 if node.type == token.NUMBER:
136 normalize_numeric_literal(node)
137 if node.type not in WHITESPACE:
138 self.current_line.append(node)
139 yield from super().visit_default(node)
141 def visit_INDENT(self, node: Leaf) -> Iterator[Line]:
142 """Increase indentation level, maybe yield a line."""
143 # In blib2to3 INDENT never holds comments.
144 yield from self.line(+1)
145 yield from self.visit_default(node)
147 def visit_DEDENT(self, node: Leaf) -> Iterator[Line]:
148 """Decrease indentation level, maybe yield a line."""
149 # The current line might still wait for trailing comments. At DEDENT time
150 # there won't be any (they would be prefixes on the preceding NEWLINE).
151 # Emit the line then.
152 yield from self.line()
154 # While DEDENT has no value, its prefix may contain standalone comments
155 # that belong to the current indentation level. Get 'em.
156 yield from self.visit_default(node)
158 # Finally, emit the dedent.
159 yield from self.line(-1)
162 self, node: Node, keywords: Set[str], parens: Set[str]
164 """Visit a statement.
166 This implementation is shared for `if`, `while`, `for`, `try`, `except`,
167 `def`, `with`, `class`, `assert`, and assignments.
169 The relevant Python language `keywords` for a given statement will be
170 NAME leaves within it. This methods puts those on a separate line.
172 `parens` holds a set of string leaf values immediately after which
173 invisible parens should be put.
175 normalize_invisible_parens(node, parens_after=parens, preview=self.mode.preview)
176 for child in node.children:
177 if is_name_token(child) and child.value in keywords:
178 yield from self.line()
180 yield from self.visit(child)
182 def visit_funcdef(self, node: Node) -> Iterator[Line]:
183 """Visit function definition."""
184 if Preview.annotation_parens not in self.mode:
185 yield from self.visit_stmt(node, keywords={"def"}, parens=set())
187 yield from self.line()
189 # Remove redundant brackets around return type annotation.
190 is_return_annotation = False
191 for child in node.children:
192 if child.type == token.RARROW:
193 is_return_annotation = True
194 elif is_return_annotation:
195 if child.type == syms.atom and child.children[0].type == token.LPAR:
196 if maybe_make_parens_invisible_in_atom(
199 remove_brackets_around_comma=False,
201 wrap_in_parentheses(node, child, visible=False)
203 wrap_in_parentheses(node, child, visible=False)
204 is_return_annotation = False
206 for child in node.children:
207 yield from self.visit(child)
209 def visit_match_case(self, node: Node) -> Iterator[Line]:
210 """Visit either a match or case statement."""
211 normalize_invisible_parens(node, parens_after=set(), preview=self.mode.preview)
213 yield from self.line()
214 for child in node.children:
215 yield from self.visit(child)
217 def visit_suite(self, node: Node) -> Iterator[Line]:
219 if self.mode.is_pyi and is_stub_suite(node):
220 yield from self.visit(node.children[2])
222 yield from self.visit_default(node)
224 def visit_simple_stmt(self, node: Node) -> Iterator[Line]:
225 """Visit a statement without nested statements."""
226 prev_type: Optional[int] = None
227 for child in node.children:
228 if (prev_type is None or prev_type == token.SEMI) and is_arith_like(child):
229 wrap_in_parentheses(node, child, visible=False)
230 prev_type = child.type
232 is_suite_like = node.parent and node.parent.type in STATEMENT
234 if self.mode.is_pyi and is_stub_body(node):
235 yield from self.visit_default(node)
237 yield from self.line(+1)
238 yield from self.visit_default(node)
239 yield from self.line(-1)
245 or not is_stub_suite(node.parent)
247 yield from self.line()
248 yield from self.visit_default(node)
250 def visit_async_stmt(self, node: Node) -> Iterator[Line]:
251 """Visit `async def`, `async for`, `async with`."""
252 yield from self.line()
254 children = iter(node.children)
255 for child in children:
256 yield from self.visit(child)
258 if child.type == token.ASYNC or child.type == STANDALONE_COMMENT:
259 # STANDALONE_COMMENT happens when `# fmt: skip` is applied on the async
263 internal_stmt = next(children)
264 for child in internal_stmt.children:
265 yield from self.visit(child)
267 def visit_decorators(self, node: Node) -> Iterator[Line]:
268 """Visit decorators."""
269 for child in node.children:
270 yield from self.line()
271 yield from self.visit(child)
273 def visit_power(self, node: Node) -> Iterator[Line]:
274 for idx, leaf in enumerate(node.children[:-1]):
275 next_leaf = node.children[idx + 1]
277 if not isinstance(leaf, Leaf):
280 value = leaf.value.lower()
282 leaf.type == token.NUMBER
283 and next_leaf.type == syms.trailer
284 # Ensure that we are in an attribute trailer
285 and next_leaf.children[0].type == token.DOT
286 # It shouldn't wrap hexadecimal, binary and octal literals
287 and not value.startswith(("0x", "0b", "0o"))
288 # It shouldn't wrap complex literals
291 wrap_in_parentheses(node, leaf)
293 if Preview.remove_redundant_parens in self.mode:
294 remove_await_parens(node)
296 yield from self.visit_default(node)
298 def visit_SEMI(self, leaf: Leaf) -> Iterator[Line]:
299 """Remove a semicolon and put the other statement on a separate line."""
300 yield from self.line()
302 def visit_ENDMARKER(self, leaf: Leaf) -> Iterator[Line]:
303 """End of file. Process outstanding comments and end with a newline."""
304 yield from self.visit_default(leaf)
305 yield from self.line()
307 def visit_STANDALONE_COMMENT(self, leaf: Leaf) -> Iterator[Line]:
308 if not self.current_line.bracket_tracker.any_open_brackets():
309 yield from self.line()
310 yield from self.visit_default(leaf)
312 def visit_factor(self, node: Node) -> Iterator[Line]:
313 """Force parentheses between a unary op and a binary power:
317 _operator, operand = node.children
319 operand.type == syms.power
320 and len(operand.children) == 3
321 and operand.children[1].type == token.DOUBLESTAR
323 lpar = Leaf(token.LPAR, "(")
324 rpar = Leaf(token.RPAR, ")")
325 index = operand.remove() or 0
326 node.insert_child(index, Node(syms.atom, [lpar, operand, rpar]))
327 yield from self.visit_default(node)
329 def visit_STRING(self, leaf: Leaf) -> Iterator[Line]:
330 if is_docstring(leaf) and "\\\n" not in leaf.value:
331 # We're ignoring docstrings with backslash newline escapes because changing
332 # indentation of those changes the AST representation of the code.
333 if Preview.normalize_docstring_quotes_and_prefixes_properly in self.mode:
334 # There was a bug where --skip-string-normalization wouldn't stop us
335 # from normalizing docstring prefixes. To maintain stability, we can
336 # only address this buggy behaviour while the preview style is enabled.
337 if self.mode.string_normalization:
338 docstring = normalize_string_prefix(leaf.value)
339 # visit_default() does handle string normalization for us, but
340 # since this method acts differently depending on quote style (ex.
341 # see padding logic below), there's a possibility for unstable
342 # formatting as visit_default() is called *after*. To avoid a
343 # situation where this function formats a docstring differently on
344 # the second pass, normalize it early.
345 docstring = normalize_string_quotes(docstring)
347 docstring = leaf.value
349 # ... otherwise, we'll keep the buggy behaviour >.<
350 docstring = normalize_string_prefix(leaf.value)
351 prefix = get_string_prefix(docstring)
352 docstring = docstring[len(prefix) :] # Remove the prefix
353 quote_char = docstring[0]
354 # A natural way to remove the outer quotes is to do:
355 # docstring = docstring.strip(quote_char)
356 # but that breaks on """""x""" (which is '""x').
357 # So we actually need to remove the first character and the next two
358 # characters but only if they are the same as the first.
359 quote_len = 1 if docstring[1] != quote_char else 3
360 docstring = docstring[quote_len:-quote_len]
361 docstring_started_empty = not docstring
362 indent = " " * 4 * self.current_line.depth
364 if is_multiline_string(leaf):
365 docstring = fix_docstring(docstring, indent)
367 docstring = docstring.strip()
370 # Add some padding if the docstring starts / ends with a quote mark.
371 if docstring[0] == quote_char:
372 docstring = " " + docstring
373 if docstring[-1] == quote_char:
375 if docstring[-1] == "\\":
376 backslash_count = len(docstring) - len(docstring.rstrip("\\"))
377 if backslash_count % 2:
378 # Odd number of tailing backslashes, add some padding to
379 # avoid escaping the closing string quote.
381 elif not docstring_started_empty:
384 # We could enforce triple quotes at this point.
385 quote = quote_char * quote_len
387 # It's invalid to put closing single-character quotes on a new line.
388 if Preview.long_docstring_quotes_on_newline in self.mode and quote_len == 3:
389 # We need to find the length of the last line of the docstring
390 # to find if we can add the closing quotes to the line without
391 # exceeding the maximum line length.
392 # If docstring is one line, then we need to add the length
393 # of the indent, prefix, and starting quotes. Ending quotes are
395 lines = docstring.splitlines()
396 last_line_length = len(lines[-1]) if docstring else 0
399 last_line_length += len(indent) + len(prefix) + quote_len
401 # If adding closing quotes would cause the last line to exceed
402 # the maximum line length then put a line break before the
404 if last_line_length + quote_len > self.mode.line_length:
405 leaf.value = prefix + quote + docstring + "\n" + indent + quote
407 leaf.value = prefix + quote + docstring + quote
409 leaf.value = prefix + quote + docstring + quote
411 yield from self.visit_default(leaf)
413 def __post_init__(self) -> None:
414 """You are in a twisty little maze of passages."""
415 self.current_line = Line(mode=self.mode)
419 self.visit_assert_stmt = partial(v, keywords={"assert"}, parens={"assert", ","})
420 self.visit_if_stmt = partial(
421 v, keywords={"if", "else", "elif"}, parens={"if", "elif"}
423 self.visit_while_stmt = partial(v, keywords={"while", "else"}, parens={"while"})
424 self.visit_for_stmt = partial(v, keywords={"for", "else"}, parens={"for", "in"})
425 self.visit_try_stmt = partial(
426 v, keywords={"try", "except", "else", "finally"}, parens=Ø
428 if self.mode.preview:
429 self.visit_except_clause = partial(
430 v, keywords={"except"}, parens={"except"}
432 self.visit_with_stmt = partial(v, keywords={"with"}, parens={"with"})
434 self.visit_except_clause = partial(v, keywords={"except"}, parens=Ø)
435 self.visit_with_stmt = partial(v, keywords={"with"}, parens=Ø)
436 self.visit_classdef = partial(v, keywords={"class"}, parens=Ø)
437 self.visit_expr_stmt = partial(v, keywords=Ø, parens=ASSIGNMENTS)
438 self.visit_return_stmt = partial(v, keywords={"return"}, parens={"return"})
439 self.visit_import_from = partial(v, keywords=Ø, parens={"import"})
440 self.visit_del_stmt = partial(v, keywords=Ø, parens={"del"})
441 self.visit_async_funcdef = self.visit_async_stmt
442 self.visit_decorated = self.visit_decorators
445 self.visit_match_stmt = self.visit_match_case
446 self.visit_case_block = self.visit_match_case
450 line: Line, mode: Mode, features: Collection[Feature] = ()
452 """Transform a `line`, potentially splitting it into many lines.
454 They should fit in the allotted `line_length` but might not be able to.
456 `features` are syntactical features that may be used in the output.
462 line_str = line_to_string(line)
464 ll = mode.line_length
465 sn = mode.string_normalization
466 string_merge = StringMerger(ll, sn)
467 string_paren_strip = StringParenStripper(ll, sn)
468 string_split = StringSplitter(ll, sn)
469 string_paren_wrap = StringParenWrapper(ll, sn)
471 transformers: List[Transformer]
473 not line.contains_uncollapsable_type_comments()
474 and not line.should_split_rhs
475 and not line.magic_trailing_comma
477 is_line_short_enough(line, line_length=mode.line_length, line_str=line_str)
478 or line.contains_unsplittable_type_ignore()
480 and not (line.inside_brackets and line.contains_standalone_comments())
482 # Only apply basic string preprocessing, since lines shouldn't be split here.
483 if Preview.string_processing in mode:
484 transformers = [string_merge, string_paren_strip]
488 transformers = [left_hand_split]
492 self: object, line: Line, features: Collection[Feature]
494 """Wraps calls to `right_hand_split`.
496 The calls increasingly `omit` right-hand trailers (bracket pairs with
497 content), meaning the trailers get glued together to split on another
498 bracket pair instead.
500 for omit in generate_trailers_to_omit(line, mode.line_length):
502 right_hand_split(line, mode.line_length, features, omit=omit)
504 # Note: this check is only able to figure out if the first line of the
505 # *current* transformation fits in the line length. This is true only
506 # for simple cases. All others require running more transforms via
507 # `transform_line()`. This check doesn't know if those would succeed.
508 if is_line_short_enough(lines[0], line_length=mode.line_length):
512 # All splits failed, best effort split with no omits.
513 # This mostly happens to multiline strings that are by definition
514 # reported as not fitting a single line, as well as lines that contain
515 # trailing commas (those have to be exploded).
516 yield from right_hand_split(
517 line, line_length=mode.line_length, features=features
520 # HACK: nested functions (like _rhs) compiled by mypyc don't retain their
521 # __name__ attribute which is needed in `run_transformer` further down.
522 # Unfortunately a nested class breaks mypyc too. So a class must be created
523 # via type ... https://github.com/mypyc/mypyc/issues/884
524 rhs = type("rhs", (), {"__call__": _rhs})()
526 if Preview.string_processing in mode:
527 if line.inside_brackets:
533 standalone_comment_split,
546 if line.inside_brackets:
547 transformers = [delimiter_split, standalone_comment_split, rhs]
550 # It's always safe to attempt hugging of power operations and pretty much every line
552 transformers.append(hug_power_op)
554 for transform in transformers:
555 # We are accumulating lines in `result` because we might want to abort
556 # mission and return the original line in the end, or attempt a different
559 result = run_transformer(line, transform, mode, features, line_str=line_str)
560 except CannotTransform:
570 class _BracketSplitComponent(Enum):
576 def left_hand_split(line: Line, _features: Collection[Feature] = ()) -> Iterator[Line]:
577 """Split line into many lines, starting with the first matching bracket pair.
579 Note: this usually looks weird, only use this for function definitions.
580 Prefer RHS otherwise. This is why this function is not symmetrical with
581 :func:`right_hand_split` which also handles optional parentheses.
583 tail_leaves: List[Leaf] = []
584 body_leaves: List[Leaf] = []
585 head_leaves: List[Leaf] = []
586 current_leaves = head_leaves
587 matching_bracket: Optional[Leaf] = None
588 for leaf in line.leaves:
590 current_leaves is body_leaves
591 and leaf.type in CLOSING_BRACKETS
592 and leaf.opening_bracket is matching_bracket
593 and isinstance(matching_bracket, Leaf)
596 ensure_visible(matching_bracket)
597 current_leaves = tail_leaves if body_leaves else head_leaves
598 current_leaves.append(leaf)
599 if current_leaves is head_leaves:
600 if leaf.type in OPENING_BRACKETS:
601 matching_bracket = leaf
602 current_leaves = body_leaves
603 if not matching_bracket:
604 raise CannotSplit("No brackets found")
606 head = bracket_split_build_line(
607 head_leaves, line, matching_bracket, component=_BracketSplitComponent.head
609 body = bracket_split_build_line(
610 body_leaves, line, matching_bracket, component=_BracketSplitComponent.body
612 tail = bracket_split_build_line(
613 tail_leaves, line, matching_bracket, component=_BracketSplitComponent.tail
615 bracket_split_succeeded_or_raise(head, body, tail)
616 for result in (head, body, tail):
621 def right_hand_split(
624 features: Collection[Feature] = (),
625 omit: Collection[LeafID] = (),
627 """Split line into many lines, starting with the last matching bracket pair.
629 If the split was by optional parentheses, attempt splitting without them, too.
630 `omit` is a collection of closing bracket IDs that shouldn't be considered for
633 Note: running this function modifies `bracket_depth` on the leaves of `line`.
635 tail_leaves: List[Leaf] = []
636 body_leaves: List[Leaf] = []
637 head_leaves: List[Leaf] = []
638 current_leaves = tail_leaves
639 opening_bracket: Optional[Leaf] = None
640 closing_bracket: Optional[Leaf] = None
641 for leaf in reversed(line.leaves):
642 if current_leaves is body_leaves:
643 if leaf is opening_bracket:
644 current_leaves = head_leaves if body_leaves else tail_leaves
645 current_leaves.append(leaf)
646 if current_leaves is tail_leaves:
647 if leaf.type in CLOSING_BRACKETS and id(leaf) not in omit:
648 opening_bracket = leaf.opening_bracket
649 closing_bracket = leaf
650 current_leaves = body_leaves
651 if not (opening_bracket and closing_bracket and head_leaves):
652 # If there is no opening or closing_bracket that means the split failed and
653 # all content is in the tail. Otherwise, if `head_leaves` are empty, it means
654 # the matching `opening_bracket` wasn't available on `line` anymore.
655 raise CannotSplit("No brackets found")
657 tail_leaves.reverse()
658 body_leaves.reverse()
659 head_leaves.reverse()
660 head = bracket_split_build_line(
661 head_leaves, line, opening_bracket, component=_BracketSplitComponent.head
663 body = bracket_split_build_line(
664 body_leaves, line, opening_bracket, component=_BracketSplitComponent.body
666 tail = bracket_split_build_line(
667 tail_leaves, line, opening_bracket, component=_BracketSplitComponent.tail
669 bracket_split_succeeded_or_raise(head, body, tail)
671 Feature.FORCE_OPTIONAL_PARENTHESES not in features
672 # the opening bracket is an optional paren
673 and opening_bracket.type == token.LPAR
674 and not opening_bracket.value
675 # the closing bracket is an optional paren
676 and closing_bracket.type == token.RPAR
677 and not closing_bracket.value
678 # it's not an import (optional parens are the only thing we can split on
679 # in this case; attempting a split without them is a waste of time)
680 and not line.is_import
681 # there are no standalone comments in the body
682 and not body.contains_standalone_comments(0)
683 # and we can actually remove the parens
684 and can_omit_invisible_parens(body, line_length)
686 omit = {id(closing_bracket), *omit}
688 yield from right_hand_split(line, line_length, features=features, omit=omit)
691 except CannotSplit as e:
694 or is_line_short_enough(body, line_length=line_length)
697 "Splitting failed, body is still too long and can't be split."
700 elif head.contains_multiline_strings() or tail.contains_multiline_strings():
702 "The current optional pair of parentheses is bound to fail to"
703 " satisfy the splitting algorithm because the head or the tail"
704 " contains multiline strings which by definition never fit one"
708 ensure_visible(opening_bracket)
709 ensure_visible(closing_bracket)
710 for result in (head, body, tail):
715 def bracket_split_succeeded_or_raise(head: Line, body: Line, tail: Line) -> None:
716 """Raise :exc:`CannotSplit` if the last left- or right-hand split failed.
718 Do nothing otherwise.
720 A left- or right-hand split is based on a pair of brackets. Content before
721 (and including) the opening bracket is left on one line, content inside the
722 brackets is put on a separate line, and finally content starting with and
723 following the closing bracket is put on a separate line.
725 Those are called `head`, `body`, and `tail`, respectively. If the split
726 produced the same line (all content in `head`) or ended up with an empty `body`
727 and the `tail` is just the closing bracket, then it's considered failed.
729 tail_len = len(str(tail).strip())
732 raise CannotSplit("Splitting brackets produced the same line")
736 f"Splitting brackets on an empty body to save {tail_len} characters is"
741 def bracket_split_build_line(
744 opening_bracket: Leaf,
746 component: _BracketSplitComponent,
748 """Return a new line with given `leaves` and respective comments from `original`.
750 If it's the head component, brackets will be tracked so trailing commas are
753 If it's the body component, the result line is one-indented inside brackets and as
754 such has its first leaf's prefix normalized and a trailing comma added when
757 result = Line(mode=original.mode, depth=original.depth)
758 if component is _BracketSplitComponent.body:
759 result.inside_brackets = True
762 # Since body is a new indent level, remove spurious leading whitespace.
763 normalize_prefix(leaves[0], inside_brackets=True)
764 # Ensure a trailing comma for imports and standalone function arguments, but
765 # be careful not to add one after any comments or within type annotations.
768 and opening_bracket.value == "("
769 and not any(leaf.type == token.COMMA for leaf in leaves)
770 # In particular, don't add one within a parenthesized return annotation.
771 # Unfortunately the indicator we're in a return annotation (RARROW) may
772 # be defined directly in the parent node, the parent of the parent ...
773 # and so on depending on how complex the return annotation is.
774 # This isn't perfect and there's some false negatives but they are in
775 # contexts were a comma is actually fine.
777 node.prev_sibling.type == RARROW
780 getattr(leaves[0].parent, "parent", None),
782 if isinstance(node, Node) and isinstance(node.prev_sibling, Leaf)
786 if original.is_import or no_commas:
787 for i in range(len(leaves) - 1, -1, -1):
788 if leaves[i].type == STANDALONE_COMMENT:
791 if leaves[i].type != token.COMMA:
792 new_comma = Leaf(token.COMMA, ",")
793 leaves.insert(i + 1, new_comma)
796 leaves_to_track: Set[LeafID] = set()
798 Preview.handle_trailing_commas_in_head in original.mode
799 and component is _BracketSplitComponent.head
801 leaves_to_track = get_leaves_inside_matching_brackets(leaves)
807 track_bracket=id(leaf) in leaves_to_track,
809 for comment_after in original.comments_after(leaf):
810 result.append(comment_after, preformatted=True)
811 if component is _BracketSplitComponent.body and should_split_line(
812 result, opening_bracket
814 result.should_split_rhs = True
818 def dont_increase_indentation(split_func: Transformer) -> Transformer:
819 """Normalize prefix of the first leaf in every line returned by `split_func`.
821 This is a decorator over relevant split functions.
825 def split_wrapper(line: Line, features: Collection[Feature] = ()) -> Iterator[Line]:
826 for split_line in split_func(line, features):
827 normalize_prefix(split_line.leaves[0], inside_brackets=True)
833 @dont_increase_indentation
834 def delimiter_split(line: Line, features: Collection[Feature] = ()) -> Iterator[Line]:
835 """Split according to delimiters of the highest priority.
837 If the appropriate Features are given, the split will add trailing commas
838 also in function signatures and calls that contain `*` and `**`.
841 last_leaf = line.leaves[-1]
843 raise CannotSplit("Line empty") from None
845 bt = line.bracket_tracker
847 delimiter_priority = bt.max_delimiter_priority(exclude={id(last_leaf)})
849 raise CannotSplit("No delimiters found") from None
851 if delimiter_priority == DOT_PRIORITY:
852 if bt.delimiter_count_with_priority(delimiter_priority) == 1:
853 raise CannotSplit("Splitting a single attribute from its owner looks wrong")
856 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
858 lowest_depth = sys.maxsize
859 trailing_comma_safe = True
861 def append_to_line(leaf: Leaf) -> Iterator[Line]:
862 """Append `leaf` to current line or to new line if appending impossible."""
863 nonlocal current_line
865 current_line.append_safe(leaf, preformatted=True)
870 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
872 current_line.append(leaf)
874 for leaf in line.leaves:
875 yield from append_to_line(leaf)
877 for comment_after in line.comments_after(leaf):
878 yield from append_to_line(comment_after)
880 lowest_depth = min(lowest_depth, leaf.bracket_depth)
881 if leaf.bracket_depth == lowest_depth:
882 if is_vararg(leaf, within={syms.typedargslist}):
883 trailing_comma_safe = (
884 trailing_comma_safe and Feature.TRAILING_COMMA_IN_DEF in features
886 elif is_vararg(leaf, within={syms.arglist, syms.argument}):
887 trailing_comma_safe = (
888 trailing_comma_safe and Feature.TRAILING_COMMA_IN_CALL in features
891 leaf_priority = bt.delimiters.get(id(leaf))
892 if leaf_priority == delimiter_priority:
896 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
901 and delimiter_priority == COMMA_PRIORITY
902 and current_line.leaves[-1].type != token.COMMA
903 and current_line.leaves[-1].type != STANDALONE_COMMENT
905 new_comma = Leaf(token.COMMA, ",")
906 current_line.append(new_comma)
910 @dont_increase_indentation
911 def standalone_comment_split(
912 line: Line, features: Collection[Feature] = ()
914 """Split standalone comments from the rest of the line."""
915 if not line.contains_standalone_comments(0):
916 raise CannotSplit("Line does not have any standalone comments")
919 mode=line.mode, depth=line.depth, inside_brackets=line.inside_brackets
922 def append_to_line(leaf: Leaf) -> Iterator[Line]:
923 """Append `leaf` to current line or to new line if appending impossible."""
924 nonlocal current_line
926 current_line.append_safe(leaf, preformatted=True)
931 line.mode, depth=line.depth, inside_brackets=line.inside_brackets
933 current_line.append(leaf)
935 for leaf in line.leaves:
936 yield from append_to_line(leaf)
938 for comment_after in line.comments_after(leaf):
939 yield from append_to_line(comment_after)
945 def normalize_prefix(leaf: Leaf, *, inside_brackets: bool) -> None:
946 """Leave existing extra newlines if not `inside_brackets`. Remove everything
949 Note: don't use backslashes for formatting or you'll lose your voting rights.
951 if not inside_brackets:
952 spl = leaf.prefix.split("#")
953 if "\\" not in spl[0]:
954 nl_count = spl[-1].count("\n")
957 leaf.prefix = "\n" * nl_count
963 def normalize_invisible_parens(
964 node: Node, parens_after: Set[str], *, preview: bool
966 """Make existing optional parentheses invisible or create new ones.
968 `parens_after` is a set of string leaf values immediately after which parens
971 Standardizes on visible parentheses for single-element tuples, and keeps
972 existing visible parentheses for other tuples and generator expressions.
974 for pc in list_comments(node.prefix, is_endmarker=False, preview=preview):
975 if pc.value in FMT_OFF:
976 # This `node` has a prefix with `# fmt: off`, don't mess with parens.
979 for index, child in enumerate(list(node.children)):
980 # Fixes a bug where invisible parens are not properly stripped from
981 # assignment statements that contain type annotations.
982 if isinstance(child, Node) and child.type == syms.annassign:
983 normalize_invisible_parens(
984 child, parens_after=parens_after, preview=preview
987 # Add parentheses around long tuple unpacking in assignments.
990 and isinstance(child, Node)
991 and child.type == syms.testlist_star_expr
998 and child.type == syms.atom
999 and node.type == syms.for_stmt
1000 and isinstance(child.prev_sibling, Leaf)
1001 and child.prev_sibling.type == token.NAME
1002 and child.prev_sibling.value == "for"
1004 if maybe_make_parens_invisible_in_atom(
1007 remove_brackets_around_comma=True,
1009 wrap_in_parentheses(node, child, visible=False)
1010 elif preview and isinstance(child, Node) and node.type == syms.with_stmt:
1011 remove_with_parens(child, node)
1012 elif child.type == syms.atom:
1013 if maybe_make_parens_invisible_in_atom(
1017 wrap_in_parentheses(node, child, visible=False)
1018 elif is_one_tuple(child):
1019 wrap_in_parentheses(node, child, visible=True)
1020 elif node.type == syms.import_from:
1021 # "import from" nodes store parentheses directly as part of
1023 if is_lpar_token(child):
1024 assert is_rpar_token(node.children[-1])
1025 # make parentheses invisible
1027 node.children[-1].value = ""
1028 elif child.type != token.STAR:
1029 # insert invisible parentheses
1030 node.insert_child(index, Leaf(token.LPAR, ""))
1031 node.append_child(Leaf(token.RPAR, ""))
1035 and child.type == token.STAR
1036 and node.type == syms.except_clause
1038 # In except* (PEP 654), the star is actually part of
1039 # of the keyword. So we need to skip the insertion of
1040 # invisible parentheses to work more precisely.
1043 elif not (isinstance(child, Leaf) and is_multiline_string(child)):
1044 wrap_in_parentheses(node, child, visible=False)
1046 comma_check = child.type == token.COMMA if preview else False
1048 check_lpar = isinstance(child, Leaf) and (
1049 child.value in parens_after or comma_check
1053 def remove_await_parens(node: Node) -> None:
1054 if node.children[0].type == token.AWAIT and len(node.children) > 1:
1056 node.children[1].type == syms.atom
1057 and node.children[1].children[0].type == token.LPAR
1059 if maybe_make_parens_invisible_in_atom(
1062 remove_brackets_around_comma=True,
1064 wrap_in_parentheses(node, node.children[1], visible=False)
1066 # Since await is an expression we shouldn't remove
1067 # brackets in cases where this would change
1068 # the AST due to operator precedence.
1069 # Therefore we only aim to remove brackets around
1070 # power nodes that aren't also await expressions themselves.
1071 # https://peps.python.org/pep-0492/#updated-operator-precedence-table
1072 # N.B. We've still removed any redundant nested brackets though :)
1073 opening_bracket = cast(Leaf, node.children[1].children[0])
1074 closing_bracket = cast(Leaf, node.children[1].children[-1])
1075 bracket_contents = cast(Node, node.children[1].children[1])
1076 if bracket_contents.type != syms.power:
1077 ensure_visible(opening_bracket)
1078 ensure_visible(closing_bracket)
1080 bracket_contents.type == syms.power
1081 and bracket_contents.children[0].type == token.AWAIT
1083 ensure_visible(opening_bracket)
1084 ensure_visible(closing_bracket)
1085 # If we are in a nested await then recurse down.
1086 remove_await_parens(bracket_contents)
1089 def remove_with_parens(node: Node, parent: Node) -> None:
1090 """Recursively hide optional parens in `with` statements."""
1091 # Removing all unnecessary parentheses in with statements in one pass is a tad
1092 # complex as different variations of bracketed statements result in pretty
1093 # different parse trees:
1095 # with (open("file")) as f: # this is an asexpr_test
1098 # with (open("file") as f): # this is an atom containing an
1101 # with (open("file")) as f, (open("file")) as f: # this is asexpr_test, COMMA,
1104 # with (open("file") as f, open("file") as f): # an atom containing a
1105 # ... # testlist_gexp which then
1106 # # contains multiple asexpr_test(s)
1107 if node.type == syms.atom:
1108 if maybe_make_parens_invisible_in_atom(
1111 remove_brackets_around_comma=True,
1113 wrap_in_parentheses(parent, node, visible=False)
1114 if isinstance(node.children[1], Node):
1115 remove_with_parens(node.children[1], node)
1116 elif node.type == syms.testlist_gexp:
1117 for child in node.children:
1118 if isinstance(child, Node):
1119 remove_with_parens(child, node)
1120 elif node.type == syms.asexpr_test and not any(
1121 leaf.type == token.COLONEQUAL for leaf in node.leaves()
1123 if maybe_make_parens_invisible_in_atom(
1126 remove_brackets_around_comma=True,
1128 wrap_in_parentheses(node, node.children[0], visible=False)
1131 def maybe_make_parens_invisible_in_atom(
1134 remove_brackets_around_comma: bool = False,
1136 """If it's safe, make the parens in the atom `node` invisible, recursively.
1137 Additionally, remove repeated, adjacent invisible parens from the atom `node`
1138 as they are redundant.
1140 Returns whether the node should itself be wrapped in invisible parentheses.
1143 node.type != syms.atom
1144 or is_empty_tuple(node)
1145 or is_one_tuple(node)
1146 or (is_yield(node) and parent.type != syms.expr_stmt)
1148 # This condition tries to prevent removing non-optional brackets
1149 # around a tuple, however, can be a bit overzealous so we provide
1150 # and option to skip this check for `for` and `with` statements.
1151 not remove_brackets_around_comma
1152 and max_delimiter_priority_in_atom(node) >= COMMA_PRIORITY
1157 if is_walrus_assignment(node):
1163 # these ones aren't useful to end users, but they do please fuzzers
1169 first = node.children[0]
1170 last = node.children[-1]
1171 if is_lpar_token(first) and is_rpar_token(last):
1172 middle = node.children[1]
1173 # make parentheses invisible
1176 maybe_make_parens_invisible_in_atom(
1179 remove_brackets_around_comma=remove_brackets_around_comma,
1182 if is_atom_with_invisible_parens(middle):
1183 # Strip the invisible parens from `middle` by replacing
1184 # it with the child in-between the invisible parens
1185 middle.replace(middle.children[1])
1192 def should_split_line(line: Line, opening_bracket: Leaf) -> bool:
1193 """Should `line` be immediately split with `delimiter_split()` after RHS?"""
1195 if not (opening_bracket.parent and opening_bracket.value in "[{("):
1198 # We're essentially checking if the body is delimited by commas and there's more
1199 # than one of them (we're excluding the trailing comma and if the delimiter priority
1200 # is still commas, that means there's more).
1202 trailing_comma = False
1204 last_leaf = line.leaves[-1]
1205 if last_leaf.type == token.COMMA:
1206 trailing_comma = True
1207 exclude.add(id(last_leaf))
1208 max_priority = line.bracket_tracker.max_delimiter_priority(exclude=exclude)
1209 except (IndexError, ValueError):
1212 return max_priority == COMMA_PRIORITY and (
1213 (line.mode.magic_trailing_comma and trailing_comma)
1214 # always explode imports
1215 or opening_bracket.parent.type in {syms.atom, syms.import_from}
1219 def generate_trailers_to_omit(line: Line, line_length: int) -> Iterator[Set[LeafID]]:
1220 """Generate sets of closing bracket IDs that should be omitted in a RHS.
1222 Brackets can be omitted if the entire trailer up to and including
1223 a preceding closing bracket fits in one line.
1225 Yielded sets are cumulative (contain results of previous yields, too). First
1226 set is empty, unless the line should explode, in which case bracket pairs until
1227 the one that needs to explode are omitted.
1230 omit: Set[LeafID] = set()
1231 if not line.magic_trailing_comma:
1234 length = 4 * line.depth
1235 opening_bracket: Optional[Leaf] = None
1236 closing_bracket: Optional[Leaf] = None
1237 inner_brackets: Set[LeafID] = set()
1238 for index, leaf, leaf_length in line.enumerate_with_length(reversed=True):
1239 length += leaf_length
1240 if length > line_length:
1243 has_inline_comment = leaf_length > len(leaf.value) + len(leaf.prefix)
1244 if leaf.type == STANDALONE_COMMENT or has_inline_comment:
1248 if leaf is opening_bracket:
1249 opening_bracket = None
1250 elif leaf.type in CLOSING_BRACKETS:
1251 prev = line.leaves[index - 1] if index > 0 else None
1254 and prev.type == token.COMMA
1255 and leaf.opening_bracket is not None
1256 and not is_one_sequence_between(
1257 leaf.opening_bracket, leaf, line.leaves
1260 # Never omit bracket pairs with trailing commas.
1261 # We need to explode on those.
1264 inner_brackets.add(id(leaf))
1265 elif leaf.type in CLOSING_BRACKETS:
1266 prev = line.leaves[index - 1] if index > 0 else None
1267 if prev and prev.type in OPENING_BRACKETS:
1268 # Empty brackets would fail a split so treat them as "inner"
1269 # brackets (e.g. only add them to the `omit` set if another
1270 # pair of brackets was good enough.
1271 inner_brackets.add(id(leaf))
1275 omit.add(id(closing_bracket))
1276 omit.update(inner_brackets)
1277 inner_brackets.clear()
1282 and prev.type == token.COMMA
1283 and leaf.opening_bracket is not None
1284 and not is_one_sequence_between(leaf.opening_bracket, leaf, line.leaves)
1286 # Never omit bracket pairs with trailing commas.
1287 # We need to explode on those.
1291 opening_bracket = leaf.opening_bracket
1292 closing_bracket = leaf
1295 def run_transformer(
1297 transform: Transformer,
1299 features: Collection[Feature],
1304 line_str = line_to_string(line)
1305 result: List[Line] = []
1306 for transformed_line in transform(line, features):
1307 if str(transformed_line).strip("\n") == line_str:
1308 raise CannotTransform("Line transformer returned an unchanged result")
1310 result.extend(transform_line(transformed_line, mode=mode, features=features))
1313 transform.__class__.__name__ != "rhs"
1314 or not line.bracket_tracker.invisible
1315 or any(bracket.value for bracket in line.bracket_tracker.invisible)
1316 or line.contains_multiline_strings()
1317 or result[0].contains_uncollapsable_type_comments()
1318 or result[0].contains_unsplittable_type_ignore()
1319 or is_line_short_enough(result[0], line_length=mode.line_length)
1320 # If any leaves have no parents (which _can_ occur since
1321 # `transform(line)` potentially destroys the line's underlying node
1322 # structure), then we can't proceed. Doing so would cause the below
1323 # call to `append_leaves()` to fail.
1324 or any(leaf.parent is None for leaf in line.leaves)
1328 line_copy = line.clone()
1329 append_leaves(line_copy, line, line.leaves)
1330 features_fop = set(features) | {Feature.FORCE_OPTIONAL_PARENTHESES}
1331 second_opinion = run_transformer(
1332 line_copy, transform, mode, features_fop, line_str=line_str
1335 is_line_short_enough(ln, line_length=mode.line_length) for ln in second_opinion
1337 result = second_opinion