ASYNC_KEYWORDS = 7
ASSIGNMENT_EXPRESSIONS = 8
POS_ONLY_ARGUMENTS = 9
+ FORCE_OPTIONAL_PARENTHESES = 50
VERSION_TO_FEATURES: Dict[TargetVersion, Set[Feature]] = {
previous: Optional[Leaf] = None
_for_loop_depths: List[int] = field(default_factory=list)
_lambda_argument_depths: List[int] = field(default_factory=list)
+ invisible: List[Leaf] = field(default_factory=list)
def mark(self, leaf: Leaf) -> None:
"""Mark `leaf` with bracket-related metadata. Keep track of delimiters.
self.depth -= 1
opening_bracket = self.bracket_match.pop((self.depth, leaf.type))
leaf.opening_bracket = opening_bracket
+ if not leaf.value:
+ self.invisible.append(leaf)
leaf.bracket_depth = self.depth
if self.depth == 0:
delim = is_split_before_delimiter(leaf, self.previous)
if leaf.type in OPENING_BRACKETS:
self.bracket_match[self.depth, BRACKET[leaf.type]] = leaf
self.depth += 1
+ if not leaf.value:
+ self.invisible.append(leaf)
self.previous = leaf
self.maybe_increment_lambda_arguments(leaf)
self.maybe_increment_for_loop_variable(leaf)
)
if self.inside_brackets or not preformatted:
self.bracket_tracker.mark(leaf)
- self.maybe_remove_trailing_comma(leaf)
+ if self.maybe_should_explode(leaf):
+ self.should_explode = True
if not self.append_comment(leaf):
self.leaves.append(leaf)
Leaf(token.DOT, ".") for _ in range(3)
]
- @property
- def is_collection_with_optional_trailing_comma(self) -> bool:
- """Is this line a collection literal with a trailing comma that's optional?
-
- Note that the trailing comma in a 1-tuple is not optional.
- """
- if not self.leaves or len(self.leaves) < 4:
- return False
-
- # Look for and address a trailing colon.
- if self.leaves[-1].type == token.COLON:
- closer = self.leaves[-2]
- close_index = -2
- else:
- closer = self.leaves[-1]
- close_index = -1
- if closer.type not in CLOSING_BRACKETS or self.inside_brackets:
- return False
-
- if closer.type == token.RPAR:
- # Tuples require an extra check, because if there's only
- # one element in the tuple removing the comma unmakes the
- # tuple.
- #
- # We also check for parens before looking for the trailing
- # comma because in some cases (eg assigning a dict
- # literal) the literal gets wrapped in temporary parens
- # during parsing. This case is covered by the
- # collections.py test data.
- opener = closer.opening_bracket
- for _open_index, leaf in enumerate(self.leaves):
- if leaf is opener:
- break
-
- else:
- # Couldn't find the matching opening paren, play it safe.
- return False
-
- commas = 0
- comma_depth = self.leaves[close_index - 1].bracket_depth
- for leaf in self.leaves[_open_index + 1 : close_index]:
- if leaf.bracket_depth == comma_depth and leaf.type == token.COMMA:
- commas += 1
- if commas > 1:
- # We haven't looked yet for the trailing comma because
- # we might also have caught noop parens.
- return self.leaves[close_index - 1].type == token.COMMA
-
- elif commas == 1:
- return False # it's either a one-tuple or didn't have a trailing comma
-
- if self.leaves[close_index - 1].type in CLOSING_BRACKETS:
- close_index -= 1
- closer = self.leaves[close_index]
- if closer.type == token.RPAR:
- # TODO: this is a gut feeling. Will we ever see this?
- return False
-
- if self.leaves[close_index - 1].type != token.COMMA:
- return False
-
- return True
-
@property
def is_def(self) -> bool:
"""Is this a function definition? (Also returns True for async defs.)"""
def contains_multiline_strings(self) -> bool:
return any(is_multiline_string(leaf) for leaf in self.leaves)
- def maybe_remove_trailing_comma(self, closing: Leaf) -> bool:
- """Remove trailing comma if there is one and it's safe."""
- if not (self.leaves and self.leaves[-1].type == token.COMMA):
- return False
-
- # We remove trailing commas only in the case of importing a
- # single name from a module.
+ def maybe_should_explode(self, closing: Leaf) -> bool:
+ """Return True if this line should explode (always be split), that is when:
+ - there's a pre-existing trailing comma here; and
+ - it's not a one-tuple.
+ """
if not (
- self.leaves
- and self.is_import
- and len(self.leaves) > 4
+ closing.type in CLOSING_BRACKETS
+ and self.leaves
and self.leaves[-1].type == token.COMMA
- and closing.type in CLOSING_BRACKETS
- and self.leaves[-4].type == token.NAME
- and (
- # regular `from foo import bar,`
- self.leaves[-4].value == "import"
- # `from foo import (bar as baz,)
- or (
- len(self.leaves) > 6
- and self.leaves[-6].value == "import"
- and self.leaves[-3].value == "as"
- )
- # `from foo import bar as baz,`
- or (
- len(self.leaves) > 5
- and self.leaves[-5].value == "import"
- and self.leaves[-3].value == "as"
- )
- )
- and closing.type == token.RPAR
+ and not self.leaves[-1].was_checked # pre-existing
):
return False
- self.remove_trailing_comma()
- return True
+ if closing.type in {token.RBRACE, token.RSQB}:
+ return True
+
+ if self.is_import:
+ return True
+
+ if not is_one_tuple_between(closing.opening_bracket, closing, self.leaves):
+ return True
+
+ return False
def append_comment(self, comment: Leaf) -> bool:
"""Add an inline or standalone comment to the line."""
yield from self.visit_default(node)
def visit_STRING(self, leaf: Leaf) -> Iterator[Line]:
- # Check if it's a docstring
- if prev_siblings_are(
- leaf.parent, [None, token.NEWLINE, token.INDENT, syms.simple_stmt]
- ) and is_multiline_string(leaf):
- prefix = " " * self.current_line.depth
- docstring = fix_docstring(leaf.value[3:-3], prefix)
- leaf.value = leaf.value[0:3] + docstring + leaf.value[-3:]
+ if is_docstring(leaf) and "\\\n" not in leaf.value:
+ # We're ignoring docstrings with backslash newline escapes because changing
+ # indentation of those changes the AST representation of the code.
+ prefix = get_string_prefix(leaf.value)
+ lead_len = len(prefix) + 3
+ tail_len = -3
+ indent = " " * 4 * self.current_line.depth
+ docstring = fix_docstring(leaf.value[lead_len:tail_len], indent)
+ if docstring:
+ if leaf.value[lead_len - 1] == docstring[0]:
+ docstring = " " + docstring
+ if leaf.value[tail_len + 1] == docstring[-1]:
+ docstring = docstring + " "
+ leaf.value = leaf.value[0:lead_len] + docstring + leaf.value[tail_len:]
normalize_string_quotes(leaf)
yield from self.visit_default(leaf)
if (
not line.contains_uncollapsable_type_comments()
and not line.should_explode
- and not line.is_collection_with_optional_trailing_comma
and (
is_line_short_enough(line, line_length=mode.line_length, line_str=line_str)
or line.contains_unsplittable_type_ignore()
)
- and not (line.contains_standalone_comments() and line.inside_brackets)
+ and not (line.inside_brackets and line.contains_standalone_comments())
):
# Only apply basic string preprocessing, since lines shouldn't be split here.
if mode.experimental_string_processing:
else:
def rhs(line: Line, features: Collection[Feature]) -> Iterator[Line]:
+ """Wraps calls to `right_hand_split`.
+
+ The calls increasingly `omit` right-hand trailers (bracket pairs with
+ content), meaning the trailers get glued together to split on another
+ bracket pair instead.
+ """
for omit in generate_trailers_to_omit(line, mode.line_length):
lines = list(
right_hand_split(line, mode.line_length, features, omit=omit)
)
+ # Note: this check is only able to figure out if the first line of the
+ # *current* transformation fits in the line length. This is true only
+ # for simple cases. All others require running more transforms via
+ # `transform_line()`. This check doesn't know if those would succeed.
if is_line_short_enough(lines[0], line_length=mode.line_length):
yield from lines
return
# All splits failed, best effort split with no omits.
# This mostly happens to multiline strings that are by definition
- # reported as not fitting a single line.
- # line_length=1 here was historically a bug that somehow became a feature.
- # See #762 and #781 for the full story.
- yield from right_hand_split(line, line_length=1, features=features)
+ # reported as not fitting a single line, as well as lines that contain
+ # pre-existing trailing commas (those have to be exploded).
+ yield from right_hand_split(
+ line, line_length=mode.line_length, features=features
+ )
if mode.experimental_string_processing:
if line.inside_brackets:
# We are accumulating lines in `result` because we might want to abort
# mission and return the original line in the end, or attempt a different
# split altogether.
- result: List[Line] = []
try:
- for transformed_line in transform(line, features):
- if str(transformed_line).strip("\n") == line_str:
- raise CannotTransform(
- "Line transformer returned an unchanged result"
- )
-
- result.extend(
- transform_line(transformed_line, mode=mode, features=features)
- )
+ result = run_transformer(line, transform, mode, features, line_str=line_str)
except CannotTransform:
continue
else:
line_length: int
normalize_strings: bool
+ __name__ = "StringTransformer"
@abstractmethod
def do_match(self, line: Line) -> TMatchResult:
)
new_line = line.clone()
- new_line.comments = line.comments
+ new_line.comments = line.comments.copy()
append_leaves(new_line, line, LL)
new_string_leaf = new_line.leaves[string_idx]
new_line = line.clone()
new_line.comments = line.comments.copy()
-
append_leaves(new_line, line, LL[: string_idx - 1])
string_leaf = Leaf(token.STRING, LL[string_idx].value)
tail = bracket_split_build_line(tail_leaves, line, opening_bracket)
bracket_split_succeeded_or_raise(head, body, tail)
if (
- # the body shouldn't be exploded
- not body.should_explode
+ Feature.FORCE_OPTIONAL_PARENTHESES not in features
# the opening bracket is an optional paren
and opening_bracket.type == token.LPAR
and not opening_bracket.value
# there are no standalone comments in the body
and not body.contains_standalone_comments(0)
# and we can actually remove the parens
- and can_omit_invisible_parens(body, line_length)
+ and can_omit_invisible_parens(body, line_length, omit_on_explode=omit)
):
omit = {id(closing_bracket), *omit}
try:
continue
if leaves[i].type != token.COMMA:
- leaves.insert(i + 1, Leaf(token.COMMA, ","))
+ new_comma = Leaf(token.COMMA, ",")
+ new_comma.was_checked = True
+ leaves.insert(i + 1, new_comma)
break
# Populate the line
result.append(leaf, preformatted=True)
for comment_after in original.comments_after(leaf):
result.append(comment_after, preformatted=True)
- if is_body:
- result.should_explode = should_explode(result, opening_bracket)
+ if is_body and should_split_body_explode(result, opening_bracket):
+ result.should_explode = True
return result
and current_line.leaves[-1].type != token.COMMA
and current_line.leaves[-1].type != STANDALONE_COMMENT
):
- current_line.append(Leaf(token.COMMA, ","))
+ new_comma = Leaf(token.COMMA, ",")
+ new_comma.was_checked = True
+ current_line.append(new_comma)
yield current_line
leaf.value = ")"
-def should_explode(line: Line, opening_bracket: Leaf) -> bool:
- """Should `line` immediately be split with `delimiter_split()` after RHS?"""
+def should_split_body_explode(line: Line, opening_bracket: Leaf) -> bool:
+ """Should `line` be immediately split with `delimiter_split()` after RHS?"""
- if not (
- opening_bracket.parent
- and opening_bracket.parent.type in {syms.atom, syms.import_from}
- and opening_bracket.value in "[{("
- ):
+ if not (opening_bracket.parent and opening_bracket.value in "[{("):
return False
+ # We're essentially checking if the body is delimited by commas and there's more
+ # than one of them (we're excluding the trailing comma and if the delimiter priority
+ # is still commas, that means there's more).
+ exclude = set()
+ pre_existing_trailing_comma = False
try:
last_leaf = line.leaves[-1]
- exclude = {id(last_leaf)} if last_leaf.type == token.COMMA else set()
+ if last_leaf.type == token.COMMA:
+ pre_existing_trailing_comma = not last_leaf.was_checked
+ exclude.add(id(last_leaf))
max_priority = line.bracket_tracker.max_delimiter_priority(exclude=exclude)
except (IndexError, ValueError):
return False
- return max_priority == COMMA_PRIORITY
+ return max_priority == COMMA_PRIORITY and (
+ # always explode imports
+ opening_bracket.parent.type in {syms.atom, syms.import_from}
+ or pre_existing_trailing_comma
+ )
+
+
+def is_one_tuple_between(opening: Leaf, closing: Leaf, leaves: List[Leaf]) -> bool:
+ """Return True if content between `opening` and `closing` looks like a one-tuple."""
+ if opening.type != token.LPAR and closing.type != token.RPAR:
+ return False
+
+ depth = closing.bracket_depth + 1
+ for _opening_index, leaf in enumerate(leaves):
+ if leaf is opening:
+ break
+
+ else:
+ raise LookupError("Opening paren not found in `leaves`")
+
+ commas = 0
+ _opening_index += 1
+ for leaf in leaves[_opening_index:]:
+ if leaf is closing:
+ break
+
+ bracket_depth = leaf.bracket_depth
+ if bracket_depth == depth and leaf.type == token.COMMA:
+ commas += 1
+ if leaf.parent and leaf.parent.type in {
+ syms.arglist,
+ syms.typedargslist,
+ }:
+ commas += 1
+ break
+
+ return commas < 2
def get_features_used(node: Node) -> Set[Feature]:
a preceding closing bracket fits in one line.
Yielded sets are cumulative (contain results of previous yields, too). First
- set is empty.
+ set is empty, unless the line should explode, in which case bracket pairs until
+ the one that needs to explode are omitted.
"""
omit: Set[LeafID] = set()
- yield omit
+ if not line.should_explode:
+ yield omit
length = 4 * line.depth
opening_bracket: Optional[Leaf] = None
if leaf is opening_bracket:
opening_bracket = None
elif leaf.type in CLOSING_BRACKETS:
+ prev = line.leaves[index - 1] if index > 0 else None
+ if (
+ line.should_explode
+ and prev
+ and prev.type == token.COMMA
+ and not prev.was_checked
+ and not is_one_tuple_between(
+ leaf.opening_bracket, leaf, line.leaves
+ )
+ ):
+ # Never omit bracket pairs with pre-existing trailing commas.
+ # We need to explode on those.
+ break
+
inner_brackets.add(id(leaf))
elif leaf.type in CLOSING_BRACKETS:
- if index > 0 and line.leaves[index - 1].type in OPENING_BRACKETS:
+ prev = line.leaves[index - 1] if index > 0 else None
+ if prev and prev.type in OPENING_BRACKETS:
# Empty brackets would fail a split so treat them as "inner"
# brackets (e.g. only add them to the `omit` set if another
# pair of brackets was good enough.
inner_brackets.clear()
yield omit
+ if (
+ line.should_explode
+ and prev
+ and prev.type == token.COMMA
+ and not prev.was_checked
+ and not is_one_tuple_between(leaf.opening_bracket, leaf, line.leaves)
+ ):
+ # Never omit bracket pairs with pre-existing trailing commas.
+ # We need to explode on those.
+ break
+
if leaf.value:
opening_bracket = leaf.opening_bracket
closing_bracket = leaf
return True
-def can_omit_invisible_parens(line: Line, line_length: int) -> bool:
+def can_omit_invisible_parens(
+ line: Line,
+ line_length: int,
+ omit_on_explode: Collection[LeafID] = (),
+) -> bool:
"""Does `line` have a shape safe to reformat without optional parens around it?
Returns True for only a subset of potentially nice looking formattings but
assert len(line.leaves) >= 2, "Stranded delimiter"
- first = line.leaves[0]
- second = line.leaves[1]
- penultimate = line.leaves[-2]
- last = line.leaves[-1]
-
# With a single delimiter, omit if the expression starts or ends with
# a bracket.
+ first = line.leaves[0]
+ second = line.leaves[1]
if first.type in OPENING_BRACKETS and second.type not in CLOSING_BRACKETS:
- remainder = False
- length = 4 * line.depth
- for _index, leaf, leaf_length in enumerate_with_length(line):
- if leaf.type in CLOSING_BRACKETS and leaf.opening_bracket is first:
- remainder = True
- if remainder:
- length += leaf_length
- if length > line_length:
- break
-
- if leaf.type in OPENING_BRACKETS:
- # There are brackets we can further split on.
- remainder = False
-
- else:
- # checked the entire string and line length wasn't exceeded
- if len(line.leaves) == _index + 1:
- return True
+ if _can_omit_opening_paren(line, first=first, line_length=line_length):
+ return True
# Note: we are not returning False here because a line might have *both*
# a leading opening bracket and a trailing closing bracket. If the
# opening bracket doesn't match our rule, maybe the closing will.
+ penultimate = line.leaves[-2]
+ last = line.leaves[-1]
+ if line.should_explode:
+ try:
+ penultimate, last = last_two_except(line.leaves, omit=omit_on_explode)
+ except LookupError:
+ # Turns out we'd omit everything. We cannot skip the optional parentheses.
+ return False
+
if (
last.type == token.RPAR
or last.type == token.RBRACE
# unnecessary.
return True
- length = 4 * line.depth
- seen_other_brackets = False
- for _index, leaf, leaf_length in enumerate_with_length(line):
+ if (
+ line.should_explode
+ and penultimate.type == token.COMMA
+ and not penultimate.was_checked
+ ):
+ # The rightmost non-omitted bracket pair is the one we want to explode on.
+ return True
+
+ if _can_omit_closing_paren(line, last=last, line_length=line_length):
+ return True
+
+ return False
+
+
+def _can_omit_opening_paren(line: Line, *, first: Leaf, line_length: int) -> bool:
+ """See `can_omit_invisible_parens`."""
+ remainder = False
+ length = 4 * line.depth
+ _index = -1
+ for _index, leaf, leaf_length in enumerate_with_length(line):
+ if leaf.type in CLOSING_BRACKETS and leaf.opening_bracket is first:
+ remainder = True
+ if remainder:
length += leaf_length
- if leaf is last.opening_bracket:
- if seen_other_brackets or length <= line_length:
- return True
+ if length > line_length:
+ break
- elif leaf.type in OPENING_BRACKETS:
+ if leaf.type in OPENING_BRACKETS:
# There are brackets we can further split on.
- seen_other_brackets = True
+ remainder = False
+
+ else:
+ # checked the entire string and line length wasn't exceeded
+ if len(line.leaves) == _index + 1:
+ return True
return False
+def _can_omit_closing_paren(line: Line, *, last: Leaf, line_length: int) -> bool:
+ """See `can_omit_invisible_parens`."""
+ length = 4 * line.depth
+ seen_other_brackets = False
+ for _index, leaf, leaf_length in enumerate_with_length(line):
+ length += leaf_length
+ if leaf is last.opening_bracket:
+ if seen_other_brackets or length <= line_length:
+ return True
+
+ elif leaf.type in OPENING_BRACKETS:
+ # There are brackets we can further split on.
+ seen_other_brackets = True
+
+ return False
+
+
+def last_two_except(leaves: List[Leaf], omit: Collection[LeafID]) -> Tuple[Leaf, Leaf]:
+ """Return (penultimate, last) leaves skipping brackets in `omit` and contents."""
+ stop_after = None
+ last = None
+ for leaf in reversed(leaves):
+ if stop_after:
+ if leaf is stop_after:
+ stop_after = None
+ continue
+
+ if last:
+ return leaf, last
+
+ if id(leaf) in omit:
+ stop_after = leaf.opening_bracket
+ else:
+ last = leaf
+ else:
+ raise LookupError("Last two leaves were also skipped")
+
+
+def run_transformer(
+ line: Line,
+ transform: Transformer,
+ mode: Mode,
+ features: Collection[Feature],
+ *,
+ line_str: str = "",
+) -> List[Line]:
+ if not line_str:
+ line_str = line_to_string(line)
+ result: List[Line] = []
+ for transformed_line in transform(line, features):
+ if str(transformed_line).strip("\n") == line_str:
+ raise CannotTransform("Line transformer returned an unchanged result")
+
+ result.extend(transform_line(transformed_line, mode=mode, features=features))
+
+ if not (
+ transform.__name__ == "rhs"
+ and line.bracket_tracker.invisible
+ and not any(bracket.value for bracket in line.bracket_tracker.invisible)
+ and not line.contains_multiline_strings()
+ and not result[0].contains_uncollapsable_type_comments()
+ and not result[0].contains_unsplittable_type_ignore()
+ and not is_line_short_enough(result[0], line_length=mode.line_length)
+ ):
+ return result
+
+ line_copy = line.clone()
+ append_leaves(line_copy, line, line.leaves)
+ features_fop = set(features) | {Feature.FORCE_OPTIONAL_PARENTHESES}
+ second_opinion = run_transformer(
+ line_copy, transform, mode, features_fop, line_str=line_str
+ )
+ if all(
+ is_line_short_enough(ln, line_length=mode.line_length) for ln in second_opinion
+ ):
+ result = second_opinion
+ return result
+
+
def get_cache_file(mode: Mode) -> Path:
return CACHE_DIR / f"cache.{mode.get_cache_key()}.pickle"
main()
+def is_docstring(leaf: Leaf) -> bool:
+ if not is_multiline_string(leaf):
+ # For the purposes of docstring re-indentation, we don't need to do anything
+ # with single-line docstrings.
+ return False
+
+ if prev_siblings_are(
+ leaf.parent, [None, token.NEWLINE, token.INDENT, syms.simple_stmt]
+ ):
+ return True
+
+ # Multiline docstring on the same line as the `def`.
+ if prev_siblings_are(leaf.parent, [syms.parameters, token.COLON, syms.simple_stmt]):
+ # `syms.parameters` is only used in funcdefs and async_funcdefs in the Python
+ # grammar. We're safe to return True without further checks.
+ return True
+
+ return False
+
+
def fix_docstring(docstring: str, prefix: str) -> str:
# https://www.python.org/dev/peps/pep-0257/#handling-docstring-indentation
if not docstring:
trimmed.append(prefix + stripped_line)
else:
trimmed.append("")
- # Return a single string:
return "\n".join(trimmed)
projects / etc / vim.git / blobdiff