Files
ruff/crates/ruff_python_formatter/src/statement/stmt_assign.rs
Dylan 880513a013 Respect fmt: skip for multiple statements on same logical line (#22119)
This PR adjusts the logic for skipping formatting so that a `fmt: skip`
can affect multiple statements if they lie on the same line.

Specifically, a `fmt: skip` comment will now suppress all the statements
in the suite in which it appears whose range intersects the line
containing the skip directive. For example:

```python
x=[
'1'
];x=2 # fmt: skip
```

remains unchanged after formatting.

(Note that compound statements are somewhat special and were handled in
a previous PR - see #20633).


Closes #17331 and #11430.

Simplest to review commit by commit - the key diffs of interest are the
commit introducing the core logic, and the diff between the snapshots
introduced in the last commit (compared to the second commit).

# Implementation

On `main` we format a suite of statements by iterating through them. If
we meet a statement with a leading or trailing (own-line)`fmt: off`
comment, then we suppress formatting until we meet a `fmt: on` comment.
Otherwise we format the statement using its own formatting rule.

How are `fmt: skip` comments handled then? They are handled internally
to the formatting of each statement. Specifically, calling `.fmt` on a
statement node will first check to see if there is a trailing,
end-of-line `fmt: skip` (or `fmt: off`/`yapf: off`), and if so then
write the node with suppressed formatting.

In this PR we move the responsibility for handling `fmt: skip` into the
formatting logic of the suite itself. This is done as follows:

- Before beginning to format the suite, we do a pass through the
statements and collect the data of ranges with skipped formatting. More
specifically, we create a map with key given by the _first_ skipped
statement in a block and value a pair consisting of the _last_ skipped
statement and the _range_ to write verbatim.
- We iterate as before, but if we meet a statement that is a key in the
map constructed above, we pause to write the associated range verbatim.
We then advance the iterator to the last statement in the block and
proceed as before.

## Addendum on range formatting

We also had to make some changes to range formatting in order to support
this new behavior. For example, we want to make sure that

```python
<RANGE_START>x=1<RANGE_END>;x=2 # fmt: skip
```

formats verbatim, rather than becoming 

```python
x = 1;x=2 # fmt: skip
```

Recall that range formatting proceeds in two steps:
1. Find the smallest enclosing node containing the range AND that has
enough info to format the range (so it may be larger than you think,
e.g. a docstring has enclosing node given by the suite, not the string
itself.)
2. Carve out the formatted range from the result of formatting that
enclosing node.

We had to modify (1), since the suite knows how to format skipped nodes,
but nodes may not "know" they are skipped. To do this we altered the
`visit_body` bethod of the `FindEnclosingNode` visitor: now we iterate
through the statements and check for skipped ranges intersecting the
format range. If we find them, we return without descending. The result
is to consider the statement containing the suite as the enclosing node
in this case.
2026-01-10 14:56:58 +00:00

1386 lines
57 KiB
Rust

use ruff_formatter::{FormatError, RemoveSoftLinesBuffer, format_args, write};
use ruff_python_ast::{
AnyNodeRef, Expr, ExprAttribute, ExprCall, FString, Operator, StmtAssign, StringLike, TString,
TypeParams,
};
use crate::builders::parenthesize_if_expands;
use crate::comments::{
Comments, LeadingDanglingTrailingComments, SourceComment, trailing_comments,
};
use crate::context::{NodeLevel, WithNodeLevel};
use crate::expression::expr_lambda::ExprLambdaLayout;
use crate::expression::parentheses::{
NeedsParentheses, OptionalParentheses, Parentheses, Parenthesize, is_expression_parenthesized,
optional_parentheses,
};
use crate::expression::{
can_omit_optional_parentheses, has_own_parentheses, has_parentheses,
maybe_parenthesize_expression,
};
use crate::other::interpolated_string::InterpolatedStringLayout;
use crate::prelude::*;
use crate::preview::is_parenthesize_lambda_bodies_enabled;
use crate::statement::trailing_semicolon;
use crate::string::StringLikeExtensions;
use crate::string::implicit::{
FormatImplicitConcatenatedStringExpanded, FormatImplicitConcatenatedStringFlat,
ImplicitConcatenatedLayout,
};
#[derive(Default)]
pub struct FormatStmtAssign;
impl FormatNodeRule<StmtAssign> for FormatStmtAssign {
fn fmt_fields(&self, item: &StmtAssign, f: &mut PyFormatter) -> FormatResult<()> {
let StmtAssign {
range: _,
node_index: _,
targets,
value,
} = item;
let (first, rest) = targets.split_first().ok_or(FormatError::syntax_error(
"Expected at least on assignment target",
))?;
// The first target is special because it never gets parenthesized nor does the formatter remove parentheses if unnecessary.
let format_first = FormatTargetWithEqualOperator {
target: first,
preserve_parentheses: true,
};
// Avoid parenthesizing the value if the last target before the assigned value expands.
if let Some((last, head)) = rest.split_last() {
format_first.fmt(f)?;
for target in head {
FormatTargetWithEqualOperator {
target,
preserve_parentheses: false,
}
.fmt(f)?;
}
FormatStatementsLastExpression::RightToLeft {
before_operator: AnyBeforeOperator::Expression(last),
operator: AnyAssignmentOperator::Assign,
value,
statement: item.into(),
}
.fmt(f)?;
}
// Avoid parenthesizing the value for single-target assignments where the
// target has its own parentheses (list, dict, tuple, ...) and the target expands.
else if has_target_own_parentheses(first, f.context())
&& !is_expression_parenthesized(
first.into(),
f.context().comments().ranges(),
f.context().source(),
)
{
FormatStatementsLastExpression::RightToLeft {
before_operator: AnyBeforeOperator::Expression(first),
operator: AnyAssignmentOperator::Assign,
value,
statement: item.into(),
}
.fmt(f)?;
}
// For single targets that have no split points, parenthesize the value only
// if it makes it fit. Otherwise omit the parentheses.
else {
format_first.fmt(f)?;
FormatStatementsLastExpression::left_to_right(value, item).fmt(f)?;
}
if f.options().source_type().is_ipynb()
&& f.context().node_level().is_last_top_level_statement()
&& trailing_semicolon(item.into(), f.context().source()).is_some()
&& matches!(targets.as_slice(), [Expr::Name(_)])
{
token(";").fmt(f)?;
}
Ok(())
}
}
/// Formats a single target with the equal operator.
struct FormatTargetWithEqualOperator<'a> {
target: &'a Expr,
/// Whether parentheses should be preserved as in the source or if the target
/// should only be parenthesized if necessary (because of comments or because it doesn't fit).
preserve_parentheses: bool,
}
impl Format<PyFormatContext<'_>> for FormatTargetWithEqualOperator<'_> {
fn fmt(&self, f: &mut Formatter<PyFormatContext<'_>>) -> FormatResult<()> {
// Preserve parentheses for the first target or around targets with leading or trailing comments.
if self.preserve_parentheses
|| f.context().comments().has_leading(self.target)
|| f.context().comments().has_trailing(self.target)
{
self.target.format().fmt(f)?;
} else if should_parenthesize_target(self.target, f.context()) {
parenthesize_if_expands(&self.target.format().with_options(Parentheses::Never))
.fmt(f)?;
} else {
self.target
.format()
.with_options(Parentheses::Never)
.fmt(f)?;
}
write!(f, [space(), token("="), space()])
}
}
/// Formats the last expression in statements that start with a keyword (like `return`) or after an operator (assignments).
///
/// The implementation avoids parenthesizing unsplittable values (like `None`, `True`, `False`, Names, a subset of strings)
/// if the value won't fit even when parenthesized.
///
/// ## Trailing comments
/// Trailing comments are inlined inside the `value`'s parentheses rather than formatted at the end
/// of the statement for unsplittable values if the `value` gets parenthesized.
///
/// Inlining the trailing comments prevent situations where the parenthesized value
/// still exceeds the configured line width, but parenthesizing helps to make the trailing comment fit.
/// Instead, it only parenthesizes `value` if it makes both the `value` and the trailing comment fit.
/// See [PR 8431](https://github.com/astral-sh/ruff/pull/8431) for more details.
///
/// The implementation formats the statement's and value's trailing end of line comments:
/// * after the expression if the expression needs no parentheses (necessary or the `expand_parent` makes the group never fit).
/// * inside the parentheses if the expression exceeds the line-width.
///
/// ```python
/// a = loooooooooooooooooooooooooooong # with_comment
/// b = (
/// short # with_comment
/// )
/// ```
///
/// Which gets formatted to:
///
/// ```python
/// # formatted
/// a = (
/// loooooooooooooooooooooooooooong # with comment
/// )
/// b = short # with comment
/// ```
///
/// The long name gets parenthesized because it exceeds the configured line width and the trailing comment of the
/// statement gets formatted inside (instead of outside) the parentheses.
///
/// No parentheses are added for `short` because it fits into the configured line length, regardless of whether
/// the comment exceeds the line width or not.
///
/// This logic isn't implemented in `place_comment` by associating trailing statement comments to the expression because
/// doing so breaks the suite empty lines formatting that relies on trailing comments to be stored on the statement.
#[derive(Debug)]
pub(super) enum FormatStatementsLastExpression<'a> {
/// Prefers to split what's left of `value` before splitting the value.
///
/// ```python
/// aaaaaaa[bbbbbbbb] = some_long_value
/// ```
///
/// This layout splits `aaaaaaa[bbbbbbbb]` first assuming the whole statements exceeds the line width, resulting in
///
/// ```python
/// aaaaaaa[
/// bbbbbbbb
/// ] = some_long_value
/// ```
///
/// This layout is preferred over [`Self::RightToLeft`] if the left is unsplittable (single
/// keyword like `return` or a Name) because it has better performance characteristics.
LeftToRight {
/// The right side of an assignment or the value returned in a return statement.
value: &'a Expr,
/// The parent statement that encloses the `value` expression.
statement: AnyNodeRef<'a>,
},
/// Prefers parenthesizing the value before splitting the left side. Specific to assignments.
///
/// Formats what's left of `value` together with the assignment operator and the assigned `value`.
/// This layout prefers parenthesizing the value over parenthesizing the left (target or type annotation):
///
/// ```python
/// aaaaaaa[bbbbbbbb] = some_long_value
/// ```
///
/// gets formatted to...
///
/// ```python
/// aaaaaaa[bbbbbbbb] = (
/// some_long_value
/// )
/// ```
///
/// ... regardless whether the value will fit or not.
///
/// The left only gets parenthesized if the left exceeds the configured line width on its own or
/// is forced to split because of a magical trailing comma or contains comments:
///
/// ```python
/// aaaaaaa[bbbbbbbb_exceeds_the_line_width] = some_long_value
/// ```
///
/// gets formatted to
/// ```python
/// aaaaaaa[
/// bbbbbbbb_exceeds_the_line_width
/// ] = some_long_value
/// ```
///
/// The layout avoids parenthesizing the value when the left splits to avoid
/// unnecessary parentheses. Adding the parentheses, as shown in the below example, reduces readability.
///
/// ```python
/// aaaaaaa[
/// bbbbbbbb_exceeds_the_line_width
/// ] = (
/// some_long_value
/// )
///
/// ## Non-fluent Call Expressions
/// Non-fluent call expressions in the `value` position are only parenthesized if the opening parentheses
/// exceeds the configured line length. The layout prefers splitting after the opening parentheses
/// if the `callee` expression and the opening parentheses fit.
/// fits on the line.
RightToLeft {
/// The expression that comes before the assignment operator. This is either
/// the last target, or the type annotation of an annotated assignment.
before_operator: AnyBeforeOperator<'a>,
/// The assignment operator. Either `Assign` (`=`) or the operator used by the augmented assignment statement.
operator: AnyAssignmentOperator,
/// The assigned `value`.
value: &'a Expr,
/// The assignment statement.
statement: AnyNodeRef<'a>,
},
}
impl<'a> FormatStatementsLastExpression<'a> {
pub(super) fn left_to_right<S: Into<AnyNodeRef<'a>>>(value: &'a Expr, statement: S) -> Self {
Self::LeftToRight {
value,
statement: statement.into(),
}
}
}
impl Format<PyFormatContext<'_>> for FormatStatementsLastExpression<'_> {
fn fmt(&self, f: &mut Formatter<PyFormatContext<'_>>) -> FormatResult<()> {
match self {
FormatStatementsLastExpression::LeftToRight { value, statement } => {
let can_inline_comment = should_inline_comments(value, *statement, f.context());
let string_like = StringLike::try_from(*value).ok();
let format_interpolated_string = string_like
.and_then(|string| format_interpolated_string_assignment(string, f.context()));
let format_implicit_flat = string_like.and_then(|string| {
FormatImplicitConcatenatedStringFlat::new(string, f.context())
});
if !can_inline_comment
&& format_implicit_flat.is_none()
&& format_interpolated_string.is_none()
{
return maybe_parenthesize_value(value, *statement).fmt(f);
}
let comments = f.context().comments().clone();
let expression_comments = comments.leading_dangling_trailing(*value);
if let Some(inline_comments) = OptionalParenthesesInlinedComments::new(
&expression_comments,
*statement,
&comments,
) {
let group_id = f.group_id("optional_parentheses");
// Special case for implicit concatenated strings in assignment value positions.
// The special handling is necessary to prevent an instability where an assignment has
// a trailing own line comment and the implicit concatenated string fits on the line,
// but only if the comment doesn't get inlined.
//
// ```python
// ____aaa = (
// "aaaaaaaaaaaaaaaaaaaaa" "aaaaaaaaaaaaaaaaaabbbbbbbbbbbbbbbbbbbbbbbbbbbvvvvvvvvvvvvvvv"
// ) # c
// ```
//
// Without the special handling, this would get formatted to:
// ```python
// ____aaa = (
// "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaabbbbbbbbbbbbbbbbbbbbbbbbbbbvvvvvvvvvvvvvvv"
// ) # c
// ```
//
// However, this now gets reformatted again because Ruff now takes the `BestFit` layout for the string
// because the value is no longer an implicit concatenated string.
// ```python
// ____aaa = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaabbbbbbbbbbbbbbbbbbbbbbbbbbbvvvvvvvvvvvvvvv" # c
// ```
//
// The special handling here ensures that the implicit concatenated string only gets
// joined **if** it fits with the trailing comment inlined. Otherwise, keep the multiline
// formatting.
if let Some(flat) = format_implicit_flat {
inline_comments.mark_formatted();
let string = flat.string();
let flat = format_with(|f| {
if string.is_interpolated_string() {
let mut buffer = RemoveSoftLinesBuffer::new(&mut *f);
write!(buffer, [flat])
} else {
flat.fmt(f)
}
})
.memoized();
// F-string or T-string containing an expression with a magic trailing comma, a comment, or a
// multiline debug expression should never be joined. Use the default layout.
// ```python
// aaaa = f"abcd{[
// 1,
// 2,
// ]}" "more"
// ```
if string.is_interpolated_string() && flat.inspect(f)?.will_break() {
inline_comments.mark_unformatted();
return write!(
f,
[maybe_parenthesize_expression(
value,
*statement,
Parenthesize::IfBreaks,
)]
);
}
let expanded = format_with(|f| {
let f =
&mut WithNodeLevel::new(NodeLevel::Expression(Some(group_id)), f);
write!(
f,
[FormatImplicitConcatenatedStringExpanded::new(
string,
ImplicitConcatenatedLayout::MaybeFlat
)]
)
});
// Join the implicit concatenated string if it fits on a single line
// ```python
// a = "testmorelong" # comment
// ```
let single_line = format_with(|f| write!(f, [flat, inline_comments]));
// Parenthesize the string but join the implicit concatenated string and inline the comment.
// ```python
// a = (
// "testmorelong" # comment
// )
// ```
let joined_parenthesized = format_with(|f| {
group(&format_args![
token("("),
soft_block_indent(&format_args![flat, inline_comments]),
token(")"),
])
.with_id(Some(group_id))
.should_expand(true)
.fmt(f)
});
// Keep the implicit concatenated string multiline and don't inline the comment.
// ```python
// a = (
// "test"
// "more"
// "long"
// ) # comment
// ```
let implicit_expanded = format_with(|f| {
group(&format_args![
token("("),
block_indent(&expanded),
token(")"),
inline_comments,
])
.with_id(Some(group_id))
.should_expand(true)
.fmt(f)
});
// We can't use `optional_parentheses` here because the `inline_comments` contains
// a `expand_parent` which results in an instability because the next format
// collapses the parentheses.
// We can't use `parenthesize_if_expands` because it defaults to
// the *flat* layout when the expanded layout doesn't fit.
best_fitting![single_line, joined_parenthesized, implicit_expanded]
.with_mode(BestFittingMode::AllLines)
.fmt(f)?;
} else if let Some(format_interpolated_string) = format_interpolated_string {
inline_comments.mark_formatted();
let interpolated_string_flat = format_with(|f| {
let mut buffer = RemoveSoftLinesBuffer::new(&mut *f);
write!(buffer, [format_interpolated_string])
})
.memoized();
// F/T-String containing an interpolation with a magic trailing comma, a comment, or a
// multiline debug interpolation should never be joined. Use the default layout.
// ```python
// aaaa = f"aaaa {[
// 1, 2,
// ]} bbbb"
// ```
if interpolated_string_flat.inspect(f)?.will_break() {
inline_comments.mark_unformatted();
return write!(
f,
[maybe_parenthesize_expression(
value,
*statement,
Parenthesize::IfBreaks,
)]
);
}
// Considering the following example:
// ```python
// aaaaaaaaaaaaaaaaaa = f"testeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee{
// expression}moreeeeeeeeeeeeeeeee"
// ```
// Flatten the f/t-string.
// ```python
// aaaaaaaaaaaaaaaaaa = f"testeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee{expression}moreeeeeeeeeeeeeeeee"
// ```
let single_line =
format_with(|f| write!(f, [interpolated_string_flat, inline_comments]));
// Parenthesize the t-string and flatten the t-string.
// ```python
// aaaaaaaaaaaaaaaaaa = (
// t"testeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee{expression}moreeeeeeeeeeeeeeeee"
// )
// ```
let joined_parenthesized = format_with(|f| {
group(&format_args![
token("("),
soft_block_indent(&format_args![
interpolated_string_flat,
inline_comments
]),
token(")"),
])
.with_id(Some(group_id))
.should_expand(true)
.fmt(f)
});
// Avoid flattening or parenthesizing the f/t-string, keep the original
// f/t-string formatting.
// ```python
// aaaaaaaaaaaaaaaaaa = t"testeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee{
// expression
// }moreeeeeeeeeeeeeeeee"
// ```
let format_interpolated_string = format_with(|f| {
write!(f, [format_interpolated_string, inline_comments])
});
best_fitting![
single_line,
joined_parenthesized,
format_interpolated_string
]
.with_mode(BestFittingMode::AllLines)
.fmt(f)?;
} else {
best_fit_parenthesize(&format_once(|f| {
inline_comments.mark_formatted();
value.format().with_options(Parentheses::Never).fmt(f)?;
if !inline_comments.is_empty() {
// If the expressions exceeds the line width, format the comments in the parentheses
if_group_breaks(&inline_comments).fmt(f)?;
}
Ok(())
}))
.with_group_id(Some(group_id))
.fmt(f)?;
if !inline_comments.is_empty() {
// If the line fits into the line width, format the comments after the parenthesized expression
if_group_fits_on_line(&inline_comments)
.with_group_id(Some(group_id))
.fmt(f)?;
}
}
Ok(())
} else {
// Preserve the parentheses if the expression has any leading or trailing comments,
// to avoid syntax errors, similar to `maybe_parenthesize_expression`.
value.format().with_options(Parentheses::Always).fmt(f)
}
}
FormatStatementsLastExpression::RightToLeft {
before_operator,
operator,
value,
statement,
} => {
let should_inline_comments = should_inline_comments(value, *statement, f.context());
let string_like = StringLike::try_from(*value).ok();
let format_interpolated_string = string_like
.and_then(|string| format_interpolated_string_assignment(string, f.context()));
let format_implicit_flat = string_like.and_then(|string| {
FormatImplicitConcatenatedStringFlat::new(string, f.context())
});
// Use the normal `maybe_parenthesize_layout` for splittable `value`s.
if !should_inline_comments
&& !should_non_inlineable_use_best_fit(value, *statement, f.context())
&& format_implicit_flat.is_none()
&& format_interpolated_string.is_none()
{
return write!(
f,
[
before_operator,
space(),
operator,
space(),
maybe_parenthesize_value(value, *statement)
]
);
}
let comments = f.context().comments().clone();
let expression_comments = comments.leading_dangling_trailing(*value);
// Don't inline comments for attribute and call expressions for black compatibility
let inline_comments = if should_inline_comments
|| format_implicit_flat.is_some()
|| format_interpolated_string.is_some()
{
OptionalParenthesesInlinedComments::new(
&expression_comments,
*statement,
&comments,
)
} else if expression_comments.has_leading()
|| expression_comments.has_trailing_own_line()
{
None
} else {
Some(OptionalParenthesesInlinedComments::default())
};
let Some(inline_comments) = inline_comments else {
// Preserve the parentheses if the expression has any leading or trailing own line comments
// same as `maybe_parenthesize_expression`
return write!(
f,
[
before_operator,
space(),
operator,
space(),
value.format().with_options(Parentheses::Always)
]
);
};
// Prevent inline comments to be formatted as part of the expression.
inline_comments.mark_formatted();
let last_target = before_operator.memoized();
let last_target_breaks = last_target.inspect(f)?.will_break();
// Don't parenthesize the `value` if it is known that the target will break.
// This is mainly a performance optimisation that avoids unnecessary memoization
// and using the costly `BestFitting` layout if it is already known that only the last variant
// can ever fit because the left breaks.
if format_implicit_flat.is_none()
&& format_interpolated_string.is_none()
&& last_target_breaks
{
return write!(
f,
[
last_target,
space(),
operator,
space(),
value.format().with_options(Parentheses::Never),
inline_comments
]
);
}
let format_value = format_with(|f| {
if let Some(format_implicit_flat) = format_implicit_flat.as_ref() {
if format_implicit_flat.string().is_interpolated_string() {
// Remove any soft line breaks emitted by the f-string formatting.
// This is important when formatting f-strings as part of an assignment right side
// because `best_fit_parenthesize` will otherwise still try to break inner
// groups if wrapped in a `group(..).should_expand(true)`
let mut buffer = RemoveSoftLinesBuffer::new(&mut *f);
write!(buffer, [format_implicit_flat])
} else {
format_implicit_flat.fmt(f)
}
} else if let Some(format_interpolated_string) =
format_interpolated_string.as_ref()
{
// Similar to above, remove any soft line breaks emitted by the f-string
// formatting.
let mut buffer = RemoveSoftLinesBuffer::new(&mut *f);
write!(buffer, [format_interpolated_string])
} else {
value.format().with_options(Parentheses::Never).fmt(f)
}
})
.memoized();
// Tries to fit the `left` and the `value` on a single line:
// ```python
// a = b = c
// ```
let single_line = format_with(|f| {
write!(
f,
[
last_target,
space(),
operator,
space(),
format_value,
inline_comments
]
)
});
// Don't break the last assignment target but parenthesize the value to see if it fits (break right first).
//
// ```python
// a["bbbbb"] = (
// c
// )
// ```
let flat_target_parenthesize_value = format_with(|f| {
write!(
f,
[
last_target,
space(),
operator,
space(),
token("("),
group(&soft_block_indent(&format_args![
format_value,
inline_comments
]))
.should_expand(true),
token(")")
]
)
});
// Fall back to parenthesizing (or splitting) the last target part if we can't make the value
// fit. Don't parenthesize the value to avoid unnecessary parentheses.
//
// ```python
// a[
// "bbbbb"
// ] = c
// ```
let split_target_flat_value = format_with(|f| {
write!(
f,
[
group(&last_target).should_expand(true),
space(),
operator,
space(),
format_value,
inline_comments
]
)
});
// For call expressions, prefer breaking after the call expression's opening parentheses
// over parenthesizing the entire call expression.
// For subscripts, try breaking the subscript first
// For attribute chains that contain any parenthesized value: Try expanding the parenthesized value first.
if value.is_call_expr() || value.is_subscript_expr() || value.is_attribute_expr() {
best_fitting![
single_line,
// Avoid parenthesizing the call expression if the `(` fit on the line
format_args![
last_target,
space(),
operator,
space(),
group(&format_value).should_expand(true),
],
flat_target_parenthesize_value,
split_target_flat_value
]
.fmt(f)
} else if let Some(format_implicit_flat) = &format_implicit_flat {
// F-String containing an expression with a magic trailing comma, a comment, or a
// multiline debug expression should never be joined. Use the default layout.
//
// ```python
// aaaa = f"abcd{[
// 1,
// 2,
// ]}" "more"
// ```
if format_implicit_flat.string().is_interpolated_string()
&& format_value.inspect(f)?.will_break()
{
inline_comments.mark_unformatted();
return write!(
f,
[
before_operator,
space(),
operator,
space(),
maybe_parenthesize_expression(
value,
*statement,
Parenthesize::IfBreaks
)
]
);
}
let group_id = f.group_id("optional_parentheses");
let format_expanded = format_with(|f| {
let f = &mut WithNodeLevel::new(NodeLevel::Expression(Some(group_id)), f);
FormatImplicitConcatenatedStringExpanded::new(
StringLike::try_from(*value).unwrap(),
ImplicitConcatenatedLayout::MaybeFlat,
)
.fmt(f)
})
.memoized();
// Keep the target flat, parenthesize the value, and keep it multiline.
//
// ```python
// Literal[ "a", "b"] = (
// "looooooooooooooooooooooooooooooong"
// "string"
// ) # comment
// ```
let flat_target_value_parenthesized_multiline = format_with(|f| {
write!(
f,
[
last_target,
space(),
operator,
space(),
token("("),
group(&soft_block_indent(&format_expanded))
.with_id(Some(group_id))
.should_expand(true),
token(")"),
inline_comments
]
)
});
// Expand the parent and parenthesize the joined string with the inlined comment.
//
// ```python
// Literal[
// "a",
// "b",
// ] = (
// "not that long string" # comment
// )
// ```
let split_target_value_parenthesized_flat = format_with(|f| {
write!(
f,
[
group(&last_target).should_expand(true),
space(),
operator,
space(),
token("("),
group(&soft_block_indent(&format_args![
format_value,
inline_comments
]))
.should_expand(true),
token(")")
]
)
});
// The most expanded variant: Expand both the target and the string.
//
// ```python
// Literal[
// "a",
// "b",
// ] = (
// "looooooooooooooooooooooooooooooong"
// "string"
// ) # comment
// ```
let split_target_value_parenthesized_multiline = format_with(|f| {
write!(
f,
[
group(&last_target).should_expand(true),
space(),
operator,
space(),
token("("),
group(&soft_block_indent(&format_expanded))
.with_id(Some(group_id))
.should_expand(true),
token(")"),
inline_comments
]
)
});
// This is only a perf optimisation. No point in trying all the "flat-target"
// variants if we know that the last target must break.
if last_target_breaks {
best_fitting![
split_target_flat_value,
split_target_value_parenthesized_flat,
split_target_value_parenthesized_multiline,
]
.with_mode(BestFittingMode::AllLines)
.fmt(f)
} else {
best_fitting![
single_line,
flat_target_parenthesize_value,
flat_target_value_parenthesized_multiline,
split_target_flat_value,
split_target_value_parenthesized_flat,
split_target_value_parenthesized_multiline,
]
.with_mode(BestFittingMode::AllLines)
.fmt(f)
}
} else if let Some(format_interpolated_string) = &format_interpolated_string {
// F/T-String containing an interpolation with a magic trailing comma, a comment, or a
// multiline debug expression should never be joined. Use the default layout.
//
// ```python
// aaaa, bbbb = t"aaaa {[
// 1, 2,
// ]} bbbb"
// ```
if format_value.inspect(f)?.will_break() {
inline_comments.mark_unformatted();
return write!(
f,
[
before_operator,
space(),
operator,
space(),
maybe_parenthesize_expression(
value,
*statement,
Parenthesize::IfBreaks
)
]
);
}
let format_interpolated_string =
format_with(|f| write!(f, [format_interpolated_string, inline_comments]))
.memoized();
// Considering the following initial source:
//
// ```python
// aaaaaaaaaaaa["bbbbbbbbbbbbbbbb"] = (
// t"aaaaaaaaaaaaaaaaaaa {
// aaaaaaaaa + bbbbbbbbbbb + cccccccccccccc} ddddddddddddddddddd"
// )
// ```
//
// Keep the target flat, and use the regular f/t-string formatting.
//
// ```python
// aaaaaaaaaaaa["bbbbbbbbbbbbbbbb"] = t"aaaaaaaaaaaaaaaaaaa {
// aaaaaaaaa + bbbbbbbbbbb + cccccccccccccc
// } ddddddddddddddddddd"
// ```
let flat_target_regular_interpolated_string = format_with(|f| {
write!(
f,
[
last_target,
space(),
operator,
space(),
format_interpolated_string
]
)
});
// Expand the parent and parenthesize the flattened f/t-string.
//
// ```python
// aaaaaaaaaaaa[
// "bbbbbbbbbbbbbbbb"
// ] = (
// t"aaaaaaaaaaaaaaaaaaa {aaaaaaaaa + bbbbbbbbbbb + cccccccccccccc} ddddddddddddddddddd"
// )
// ```
let split_target_value_parenthesized_flat = format_with(|f| {
write!(
f,
[
group(&last_target).should_expand(true),
space(),
operator,
space(),
token("("),
group(&soft_block_indent(&format_args![
format_value,
inline_comments
]))
.should_expand(true),
token(")")
]
)
});
// Expand the parent, and use the regular f/t-string formatting.
//
// ```python
// aaaaaaaaaaaa[
// "bbbbbbbbbbbbbbbb"
// ] = t"aaaaaaaaaaaaaaaaaaa {
// aaaaaaaaa + bbbbbbbbbbb + cccccccccccccc
// } ddddddddddddddddddd"
// ```
let split_target_regular_interpolated_string = format_with(|f| {
write!(
f,
[
group(&last_target).should_expand(true),
space(),
operator,
space(),
format_interpolated_string,
]
)
});
// This is only a perf optimisation. No point in trying all the "flat-target"
// variants if we know that the last target must break.
if last_target_breaks {
best_fitting![
split_target_flat_value,
split_target_value_parenthesized_flat,
split_target_regular_interpolated_string,
]
.with_mode(BestFittingMode::AllLines)
.fmt(f)
} else {
best_fitting![
single_line,
flat_target_parenthesize_value,
flat_target_regular_interpolated_string,
split_target_flat_value,
split_target_value_parenthesized_flat,
split_target_regular_interpolated_string,
]
.with_mode(BestFittingMode::AllLines)
.fmt(f)
}
} else {
best_fitting![
single_line,
flat_target_parenthesize_value,
split_target_flat_value
]
.fmt(f)
}
}
}
}
}
#[derive(Debug, Copy, Clone)]
enum InterpolatedString<'a> {
FString(&'a FString),
TString(&'a TString),
}
impl Format<PyFormatContext<'_>> for InterpolatedString<'_> {
fn fmt(&self, f: &mut PyFormatter) -> FormatResult<()> {
match self {
InterpolatedString::FString(string) => string.format().fmt(f),
InterpolatedString::TString(string) => string.format().fmt(f),
}
}
}
/// Formats an f/t-string that is at the value position of an assignment statement.
///
/// For legibility, we discuss only the case of f-strings below, but the
/// same comments apply to t-strings.
///
/// This is just a wrapper around [`FormatFString`](crate::other::f_string::FormatFString) while
/// considering a special case when the f-string is at an assignment statement's value position.
/// This is necessary to prevent an instability where an f-string contains a multiline expression
/// and the f-string fits on the line, but only when it's surrounded by parentheses.
///
/// ```python
/// aaaaaaaaaaaaaaaaaa = f"testeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee{
/// expression}moreeeeeeeeeeeeeeeee"
/// ```
///
/// Without the special handling, this would get formatted to:
/// ```python
/// aaaaaaaaaaaaaaaaaa = f"testeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee{
/// expression
/// }moreeeeeeeeeeeeeeeee"
/// ```
///
/// However, if the parentheses already existed in the source like:
/// ```python
/// aaaaaaaaaaaaaaaaaa = (
/// f"testeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee{expression}moreeeeeeeeeeeeeeeee"
/// )
/// ```
///
/// Then, it would remain unformatted because it fits on the line. This means that even in the
/// first example, the f-string should be formatted by surrounding it with parentheses.
///
/// One might ask why not just use the `BestFit` layout in this case. Consider the following
/// example in which the f-string doesn't fit on the line even when surrounded by parentheses:
/// ```python
/// xxxxxxx = f"{
/// {'aaaaaaaaaaaaaaaaaaaaaaaaa', 'bbbbbbbbbbbbbbbbbbbbbbbbbbb', 'cccccccccccccccccccccccccc'}
/// }"
/// ```
///
/// The `BestFit` layout will format this as:
/// ```python
/// xxxxxxx = (
/// f"{
/// {
/// 'aaaaaaaaaaaaaaaaaaaaaaaaa',
/// 'bbbbbbbbbbbbbbbbbbbbbbbbbbb',
/// 'cccccccccccccccccccccccccc',
/// }
/// }"
/// )
/// ```
///
/// The reason for this is because (a) f-string already has a multiline expression thus it tries to
/// break the expression and (b) the `BestFit` layout doesn't considers the layout where the
/// multiline f-string isn't surrounded by parentheses.
fn format_interpolated_string_assignment<'a>(
string: StringLike<'a>,
context: &PyFormatContext,
) -> Option<InterpolatedString<'a>> {
let (interpolated_string, elements) = match string {
StringLike::TString(expr) => {
let t_string = expr.as_single_part_tstring()?;
(InterpolatedString::TString(t_string), &t_string.elements)
}
StringLike::FString(expr) => {
let f_string = expr.as_single_part_fstring()?;
(InterpolatedString::FString(f_string), &f_string.elements)
}
_ => {
return None;
}
};
// If the f/t-string is flat, there are no breakpoints from which it can be made multiline.
// This is the case when the f/t-string has no expressions or if it does then the expressions
// are flat (no newlines).
if InterpolatedStringLayout::from_interpolated_string_elements(elements, context.source())
.is_flat()
{
return None;
}
// This checks whether the f/t-string is multi-line and it can *never* be flattened. Thus,
// it's useless to try the flattened layout.
if string.is_multiline(context) {
return None;
}
Some(interpolated_string)
}
#[derive(Debug, Default)]
struct OptionalParenthesesInlinedComments<'a> {
expression: &'a [SourceComment],
statement: &'a [SourceComment],
}
impl<'a> OptionalParenthesesInlinedComments<'a> {
fn new(
expression_comments: &LeadingDanglingTrailingComments<'a>,
statement: AnyNodeRef<'a>,
comments: &'a Comments<'a>,
) -> Option<Self> {
if expression_comments.has_leading() || expression_comments.has_trailing_own_line() {
return None;
}
let statement_trailing_comments = comments.trailing(statement);
let after_end_of_line = statement_trailing_comments
.partition_point(|comment| comment.line_position().is_end_of_line());
let (stmt_inline_comments, _) = statement_trailing_comments.split_at(after_end_of_line);
let after_end_of_line = expression_comments
.trailing
.partition_point(|comment| comment.line_position().is_end_of_line());
let (expression_inline_comments, trailing_own_line_comments) =
expression_comments.trailing.split_at(after_end_of_line);
debug_assert!(
trailing_own_line_comments.is_empty(),
"The method should have returned early if the expression has trailing own line comments"
);
Some(OptionalParenthesesInlinedComments {
expression: expression_inline_comments,
statement: stmt_inline_comments,
})
}
fn is_empty(&self) -> bool {
self.expression.is_empty() && self.statement.is_empty()
}
fn iter_comments(&self) -> impl Iterator<Item = &'a SourceComment> {
self.expression.iter().chain(self.statement)
}
fn mark_formatted(&self) {
for comment in self.expression {
comment.mark_formatted();
}
}
fn mark_unformatted(&self) {
for comment in self.expression {
comment.mark_unformatted();
}
}
}
impl Format<PyFormatContext<'_>> for OptionalParenthesesInlinedComments<'_> {
fn fmt(&self, f: &mut Formatter<PyFormatContext<'_>>) -> FormatResult<()> {
for comment in self.iter_comments() {
comment.mark_unformatted();
}
write!(
f,
[
trailing_comments(self.expression),
trailing_comments(self.statement)
]
)
}
}
#[derive(Copy, Clone, Debug)]
pub(super) enum AnyAssignmentOperator {
Assign,
AugAssign(Operator),
}
impl Format<PyFormatContext<'_>> for AnyAssignmentOperator {
fn fmt(&self, f: &mut Formatter<PyFormatContext<'_>>) -> FormatResult<()> {
match self {
AnyAssignmentOperator::Assign => token("=").fmt(f),
AnyAssignmentOperator::AugAssign(operator) => {
write!(f, [operator.format(), token("=")])
}
}
}
}
#[derive(Copy, Clone, Debug)]
pub(super) enum AnyBeforeOperator<'a> {
Expression(&'a Expr),
TypeParams(&'a TypeParams),
}
impl Format<PyFormatContext<'_>> for AnyBeforeOperator<'_> {
fn fmt(&self, f: &mut Formatter<PyFormatContext<'_>>) -> FormatResult<()> {
match self {
AnyBeforeOperator::Expression(expression) => {
// Preserve parentheses around targets with comments.
if f.context().comments().has_leading(*expression)
|| f.context().comments().has_trailing(*expression)
{
expression
.format()
.with_options(Parentheses::Preserve)
.fmt(f)
}
// Never parenthesize targets that come with their own parentheses, e.g. don't parenthesize lists or dictionary literals.
else if should_parenthesize_target(expression, f.context()) {
if can_omit_optional_parentheses(expression, f.context()) {
optional_parentheses(&expression.format().with_options(Parentheses::Never))
.fmt(f)
} else {
parenthesize_if_expands(
&expression.format().with_options(Parentheses::Never),
)
.fmt(f)
}
} else {
expression.format().with_options(Parentheses::Never).fmt(f)
}
}
// Never parenthesize type params
AnyBeforeOperator::TypeParams(type_params) => type_params.format().fmt(f),
}
}
}
/// Returns `true` for unsplittable expressions for which comments should be inlined.
fn should_inline_comments(
expression: &Expr,
parent: AnyNodeRef,
context: &PyFormatContext,
) -> bool {
match expression {
Expr::Name(_) | Expr::NoneLiteral(_) | Expr::NumberLiteral(_) | Expr::BooleanLiteral(_) => {
true
}
Expr::StringLiteral(string) => {
string.needs_parentheses(parent, context) == OptionalParentheses::BestFit
}
Expr::BytesLiteral(bytes) => {
bytes.needs_parentheses(parent, context) == OptionalParentheses::BestFit
}
Expr::FString(fstring) => {
fstring.needs_parentheses(parent, context) == OptionalParentheses::BestFit
}
Expr::TString(tstring) => {
tstring.needs_parentheses(parent, context) == OptionalParentheses::BestFit
}
_ => false,
}
}
/// Tests whether an expression for which comments shouldn't be inlined should use the best fit layout
fn should_non_inlineable_use_best_fit(
expr: &Expr,
parent: AnyNodeRef,
context: &PyFormatContext,
) -> bool {
match expr {
Expr::Attribute(attribute) => {
attribute.needs_parentheses(parent, context) == OptionalParentheses::BestFit
}
Expr::Call(call) => call.needs_parentheses(parent, context) == OptionalParentheses::BestFit,
Expr::Subscript(subscript) => {
subscript.needs_parentheses(parent, context) == OptionalParentheses::BestFit
}
_ => false,
}
}
/// Returns `true` for targets that have their own set of parentheses when they split,
/// in which case we want to avoid parenthesizing the assigned value.
pub(super) fn has_target_own_parentheses(target: &Expr, context: &PyFormatContext) -> bool {
matches!(target, Expr::Tuple(_)) || has_own_parentheses(target, context).is_some()
}
pub(super) fn should_parenthesize_target(target: &Expr, context: &PyFormatContext) -> bool {
!(has_target_own_parentheses(target, context)
|| is_attribute_with_parenthesized_value(target, context))
}
fn is_attribute_with_parenthesized_value(target: &Expr, context: &PyFormatContext) -> bool {
match target {
Expr::Attribute(ExprAttribute { value, .. }) => {
has_parentheses(value.as_ref(), context).is_some()
|| is_attribute_with_parenthesized_value(value, context)
}
Expr::Subscript(_) => true,
Expr::Call(ExprCall { arguments, .. }) => !arguments.is_empty(),
_ => false,
}
}
/// Like [`maybe_parenthesize_expression`] but with special handling for lambdas in preview.
fn maybe_parenthesize_value<'a>(
expression: &'a Expr,
parent: AnyNodeRef<'a>,
) -> MaybeParenthesizeValue<'a> {
MaybeParenthesizeValue { expression, parent }
}
struct MaybeParenthesizeValue<'a> {
expression: &'a Expr,
parent: AnyNodeRef<'a>,
}
impl Format<PyFormatContext<'_>> for MaybeParenthesizeValue<'_> {
fn fmt(&self, f: &mut PyFormatter) -> FormatResult<()> {
let MaybeParenthesizeValue { expression, parent } = self;
if is_parenthesize_lambda_bodies_enabled(f.context())
&& let Expr::Lambda(lambda) = expression
&& !f.context().comments().has_leading(lambda)
{
parenthesize_if_expands(&lambda.format().with_options(ExprLambdaLayout::Assignment))
.fmt(f)
} else {
maybe_parenthesize_expression(expression, *parent, Parenthesize::IfBreaks).fmt(f)
}
}
}