fmt: off..on suppression comments (#6477)

This commit is contained in:
Micha Reiser 2023-08-14 17:57:36 +02:00 committed by GitHub
parent 278a4f6e14
commit 09c8b17661
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
34 changed files with 1883 additions and 978 deletions

View file

@ -1,12 +1,11 @@
use crate::{verbatim_text, FormatNodeRule, PyFormatter};
use ruff_formatter::{write, Buffer, FormatResult};
use ruff_python_ast::StmtIpyEscapeCommand;
use crate::prelude::*;
use ruff_python_ast::{Ranged, StmtIpyEscapeCommand};
#[derive(Default)]
pub struct FormatStmtIpyEscapeCommand;
impl FormatNodeRule<StmtIpyEscapeCommand> for FormatStmtIpyEscapeCommand {
fn fmt_fields(&self, item: &StmtIpyEscapeCommand, f: &mut PyFormatter) -> FormatResult<()> {
write!(f, [verbatim_text(item)])
source_text_slice(item.range(), ContainsNewlines::No).fmt(f)
}
}

View file

@ -1,14 +1,19 @@
use crate::comments::{leading_comments, trailing_comments};
use ruff_formatter::{write, FormatOwnedWithRule, FormatRefWithRule, FormatRuleWithOptions};
use ruff_python_ast::helpers::is_compound_statement;
use ruff_python_ast::{self as ast, Ranged, Stmt, Suite};
use ruff_python_ast::{Constant, ExprConstant};
use ruff_python_ast::node::AnyNodeRef;
use ruff_python_ast::{self as ast, Expr, ExprConstant, Ranged, Stmt, Suite};
use ruff_python_trivia::{lines_after_ignoring_trivia, lines_before};
use ruff_text_size::TextRange;
use crate::comments::{leading_comments, trailing_comments};
use crate::context::{NodeLevel, WithNodeLevel};
use crate::expression::expr_constant::ExprConstantLayout;
use crate::expression::string::StringLayout;
use crate::prelude::*;
use crate::verbatim::{
write_suppressed_statements_starting_with_leading_comment,
write_suppressed_statements_starting_with_trailing_comment,
};
/// Level at which the [`Suite`] appears in the source code.
#[derive(Copy, Clone, Debug)]
@ -51,196 +56,231 @@ impl FormatRule<Suite, PyFormatContext<'_>> for FormatSuite {
let comments = f.context().comments().clone();
let source = f.context().source();
let mut iter = statements.iter();
let Some(first) = iter.next() else {
return Ok(());
};
let mut f = WithNodeLevel::new(node_level, f);
write!(
f,
[format_with(|f| {
let mut iter = statements.iter();
let Some(first) = iter.next() else {
return Ok(());
};
// Format the first statement in the body, which often has special formatting rules.
let mut last = first;
match self.kind {
SuiteKind::Other => {
if is_class_or_function_definition(first) && !comments.has_leading_comments(first) {
// Add an empty line for any nested functions or classes defined within
// non-function or class compound statements, e.g., this is stable formatting:
// ```python
// if True:
//
// def test():
// ...
// ```
write!(f, [empty_line()])?;
}
write!(f, [first.format()])?;
}
SuiteKind::Function => {
if let Some(constant) = get_docstring(first) {
write!(
f,
[
// We format the expression, but the statement carries the comments
leading_comments(comments.leading_comments(first)),
constant
.format()
.with_options(ExprConstantLayout::String(StringLayout::DocString)),
trailing_comments(comments.trailing_comments(first)),
]
)?;
// Format the first statement in the body, which often has special formatting rules.
let first = match self.kind {
SuiteKind::Other => {
if is_class_or_function_definition(first)
&& !comments.has_leading_comments(first)
{
// Add an empty line for any nested functions or classes defined within
// non-function or class compound statements, e.g., this is stable formatting:
// ```python
// if True:
//
// def test():
// ...
// ```
empty_line().fmt(f)?;
}
SuiteChildStatement::Other(first)
}
SuiteKind::Function => {
if let Some(docstring) = DocstringStmt::try_from_statement(first) {
SuiteChildStatement::Docstring(docstring)
} else {
SuiteChildStatement::Other(first)
}
}
SuiteKind::Class => {
if let Some(docstring) = DocstringStmt::try_from_statement(first) {
if !comments.has_leading_comments(first)
&& lines_before(first.start(), source) > 1
{
// Allow up to one empty line before a class docstring, e.g., this is
// stable formatting:
// ```python
// class Test:
//
// """Docstring"""
// ```
empty_line().fmt(f)?;
}
SuiteChildStatement::Docstring(docstring)
} else {
SuiteChildStatement::Other(first)
}
}
SuiteKind::TopLevel => SuiteChildStatement::Other(first),
};
let (mut preceding, mut after_class_docstring) = if comments
.leading_comments(first)
.iter()
.any(|comment| comment.is_suppression_off_comment(source))
{
(
write_suppressed_statements_starting_with_leading_comment(
first, &mut iter, f,
)?,
false,
)
} else if comments
.trailing_comments(first)
.iter()
.any(|comment| comment.is_suppression_off_comment(source))
{
(
write_suppressed_statements_starting_with_trailing_comment(
first, &mut iter, f,
)?,
false,
)
} else {
write!(f, [first.format()])?;
}
}
SuiteKind::Class => {
if let Some(constant) = get_docstring(first) {
if !comments.has_leading_comments(first)
&& lines_before(first.start(), source) > 1
first.fmt(f)?;
(
first.statement(),
matches!(first, SuiteChildStatement::Docstring(_))
&& matches!(self.kind, SuiteKind::Class),
)
};
while let Some(following) = iter.next() {
if is_class_or_function_definition(preceding)
|| is_class_or_function_definition(following)
{
// Allow up to one empty line before a class docstring
match self.kind {
SuiteKind::TopLevel => {
write!(f, [empty_line(), empty_line()])?;
}
SuiteKind::Function | SuiteKind::Class | SuiteKind::Other => {
empty_line().fmt(f)?;
}
}
} else if is_import_definition(preceding) && !is_import_definition(following) {
empty_line().fmt(f)?;
} else if is_compound_statement(preceding) {
// Handles the case where a body has trailing comments. The issue is that RustPython does not include
// the comments in the range of the suite. This means, the body ends right after the last statement in the body.
// ```python
// def test():
// ...
// # The body of `test` ends right after `...` and before this comment
//
// # leading comment
//
//
// a = 10
// ```
// Using `lines_after` for the node doesn't work because it would count the lines after the `...`
// which is 0 instead of 1, the number of lines between the trailing comment and
// the leading comment. This is why the suite handling counts the lines before the
// start of the next statement or before the first leading comments for compound statements.
let start = if let Some(first_leading) =
comments.leading_comments(following).first()
{
first_leading.slice().start()
} else {
following.start()
};
match lines_before(start, source) {
0 | 1 => hard_line_break().fmt(f)?,
2 => empty_line().fmt(f)?,
3.. => match self.kind {
SuiteKind::TopLevel => write!(f, [empty_line(), empty_line()])?,
SuiteKind::Function | SuiteKind::Class | SuiteKind::Other => {
empty_line().fmt(f)?;
}
},
}
} else if after_class_docstring {
// Enforce an empty line after a class docstring, e.g., these are both stable
// formatting:
// ```python
// class Test:
// """Docstring"""
//
// ...
//
//
// class Test:
//
// """Docstring"""
//
// ...
// ```
write!(f, [empty_line()])?;
}
write!(
f,
[
// We format the expression, but the statement carries the comments
leading_comments(comments.leading_comments(first)),
constant
.format()
.with_options(ExprConstantLayout::String(StringLayout::DocString)),
trailing_comments(comments.trailing_comments(first)),
]
)?;
// Enforce an empty line after a class docstring
// ```python
// class Test:
// """Docstring"""
//
// ...
//
//
// class Test:
//
// """Docstring"""
//
// ...
// ```
// Unlike black, we add the newline also after single quoted docstrings
if let Some(second) = iter.next() {
// Format the subsequent statement immediately. This rule takes precedence
// over the rules in the loop below (and most of them won't apply anyway,
// e.g., we know the first statement isn't an import).
write!(f, [empty_line(), second.format()])?;
last = second;
}
} else {
// No docstring, use normal formatting
write!(f, [first.format()])?;
}
}
SuiteKind::TopLevel => {
write!(f, [first.format()])?;
}
}
for statement in iter {
if is_class_or_function_definition(last) || is_class_or_function_definition(statement) {
match self.kind {
SuiteKind::TopLevel => {
write!(f, [empty_line(), empty_line(), statement.format()])?;
}
SuiteKind::Function | SuiteKind::Class | SuiteKind::Other => {
write!(f, [empty_line(), statement.format()])?;
}
}
} else if is_import_definition(last) && !is_import_definition(statement) {
write!(f, [empty_line(), statement.format()])?;
} else if is_compound_statement(last) {
// Handles the case where a body has trailing comments. The issue is that RustPython does not include
// the comments in the range of the suite. This means, the body ends right after the last statement in the body.
// ```python
// def test():
// ...
// # The body of `test` ends right after `...` and before this comment
//
// # leading comment
//
//
// a = 10
// ```
// Using `lines_after` for the node doesn't work because it would count the lines after the `...`
// which is 0 instead of 1, the number of lines between the trailing comment and
// the leading comment. This is why the suite handling counts the lines before the
// start of the next statement or before the first leading comments for compound statements.
let start =
if let Some(first_leading) = comments.leading_comments(statement).first() {
first_leading.slice().start()
empty_line().fmt(f)?;
after_class_docstring = false;
} else {
statement.start()
};
// Insert the appropriate number of empty lines based on the node level, e.g.:
// * [`NodeLevel::Module`]: Up to two empty lines
// * [`NodeLevel::CompoundStatement`]: Up to one empty line
// * [`NodeLevel::Expression`]: No empty lines
match lines_before(start, source) {
0 | 1 => write!(f, [hard_line_break()])?,
2 => write!(f, [empty_line()])?,
3.. => match self.kind {
SuiteKind::TopLevel => write!(f, [empty_line(), empty_line()])?,
SuiteKind::Function | SuiteKind::Class | SuiteKind::Other => {
write!(f, [empty_line()])?;
let count_lines = |offset| {
// It's necessary to skip any trailing line comment because RustPython doesn't include trailing comments
// in the node's range
// ```python
// a # The range of `a` ends right before this comment
//
// b
// ```
//
// Simply using `lines_after` doesn't work if a statement has a trailing comment because
// it then counts the lines between the statement and the trailing comment, which is
// always 0. This is why it skips any trailing trivia (trivia that's on the same line)
// and counts the lines after.
lines_after_ignoring_trivia(offset, source)
};
match node_level {
NodeLevel::TopLevel => match count_lines(preceding.end()) {
0 | 1 => hard_line_break().fmt(f)?,
2 => empty_line().fmt(f)?,
_ => write!(f, [empty_line(), empty_line()])?,
},
NodeLevel::CompoundStatement => match count_lines(preceding.end()) {
0 | 1 => hard_line_break().fmt(f)?,
_ => empty_line().fmt(f)?,
},
NodeLevel::Expression(_) | NodeLevel::ParenthesizedExpression => {
hard_line_break().fmt(f)?;
}
}
},
}
}
write!(f, [statement.format()])?;
} else {
// Insert the appropriate number of empty lines based on the node level, e.g.:
// * [`NodeLevel::Module`]: Up to two empty lines
// * [`NodeLevel::CompoundStatement`]: Up to one empty line
// * [`NodeLevel::Expression`]: No empty lines
let count_lines = |offset| {
// It's necessary to skip any trailing line comment because RustPython doesn't include trailing comments
// in the node's range
// ```python
// a # The range of `a` ends right before this comment
//
// b
// ```
//
// Simply using `lines_after` doesn't work if a statement has a trailing comment because
// it then counts the lines between the statement and the trailing comment, which is
// always 0. This is why it skips any trailing trivia (trivia that's on the same line)
// and counts the lines after.
lines_after_ignoring_trivia(offset, source)
};
match node_level {
NodeLevel::TopLevel => match count_lines(last.end()) {
0 | 1 => write!(f, [hard_line_break()])?,
2 => write!(f, [empty_line()])?,
_ => write!(f, [empty_line(), empty_line()])?,
},
NodeLevel::CompoundStatement => match count_lines(last.end()) {
0 | 1 => write!(f, [hard_line_break()])?,
_ => write!(f, [empty_line()])?,
},
NodeLevel::Expression(_) | NodeLevel::ParenthesizedExpression => {
write!(f, [hard_line_break()])?;
if comments
.leading_comments(following)
.iter()
.any(|comment| comment.is_suppression_off_comment(source))
{
preceding = write_suppressed_statements_starting_with_leading_comment(
SuiteChildStatement::Other(following),
&mut iter,
f,
)?;
} else if comments
.trailing_comments(following)
.iter()
.any(|comment| comment.is_suppression_off_comment(source))
{
preceding = write_suppressed_statements_starting_with_trailing_comment(
SuiteChildStatement::Other(following),
&mut iter,
f,
)?;
} else {
following.format().fmt(f)?;
preceding = following;
}
}
write!(f, [statement.format()])?;
}
last = statement;
}
Ok(())
Ok(())
})]
)
}
}
@ -254,23 +294,6 @@ const fn is_import_definition(stmt: &Stmt) -> bool {
matches!(stmt, Stmt::Import(_) | Stmt::ImportFrom(_))
}
/// Checks if the statement is a simple string that can be formatted as a docstring
fn get_docstring(stmt: &Stmt) -> Option<&ExprConstant> {
let stmt_expr = stmt.as_expr_stmt()?;
let expr_constant = stmt_expr.value.as_constant_expr()?;
if matches!(
expr_constant.value,
Constant::Str(ast::StringConstant {
implicit_concatenated: false,
..
})
) {
Some(expr_constant)
} else {
None
}
}
impl FormatRuleWithOptions<Suite, PyFormatContext<'_>> for FormatSuite {
type Options = SuiteKind;
@ -296,6 +319,93 @@ impl<'ast> IntoFormat<PyFormatContext<'ast>> for Suite {
}
}
/// A statement representing a docstring.
#[derive(Copy, Clone)]
pub(crate) struct DocstringStmt<'a>(&'a Stmt);
impl<'a> DocstringStmt<'a> {
/// Checks if the statement is a simple string that can be formatted as a docstring
fn try_from_statement(stmt: &'a Stmt) -> Option<DocstringStmt<'a>> {
let Stmt::Expr(ast::StmtExpr { value, .. }) = stmt else {
return None;
};
if let Expr::Constant(ExprConstant { value, .. }) = value.as_ref() {
if !value.is_implicit_concatenated() {
return Some(DocstringStmt(stmt));
}
}
None
}
}
impl Format<PyFormatContext<'_>> for DocstringStmt<'_> {
fn fmt(&self, f: &mut Formatter<PyFormatContext<'_>>) -> FormatResult<()> {
// SAFETY: Safe because `DocStringStmt` guarantees that it only ever wraps a `ExprStmt` containing a `ConstantExpr`.
let constant = self
.0
.as_expr_stmt()
.unwrap()
.value
.as_constant_expr()
.unwrap();
let comments = f.context().comments().clone();
// We format the expression, but the statement carries the comments
write!(
f,
[
leading_comments(comments.leading_comments(self.0)),
constant
.format()
.with_options(ExprConstantLayout::String(StringLayout::DocString)),
trailing_comments(comments.trailing_comments(self.0)),
]
)
}
}
/// A Child of a suite.
#[derive(Copy, Clone)]
pub(crate) enum SuiteChildStatement<'a> {
/// A docstring documenting a class or function definition.
Docstring(DocstringStmt<'a>),
/// Any other statement.
Other(&'a Stmt),
}
impl<'a> SuiteChildStatement<'a> {
pub(crate) const fn statement(self) -> &'a Stmt {
match self {
SuiteChildStatement::Docstring(docstring) => docstring.0,
SuiteChildStatement::Other(statement) => statement,
}
}
}
impl Ranged for SuiteChildStatement<'_> {
fn range(&self) -> TextRange {
self.statement().range()
}
}
impl<'a> From<SuiteChildStatement<'a>> for AnyNodeRef<'a> {
fn from(value: SuiteChildStatement<'a>) -> Self {
value.statement().into()
}
}
impl Format<PyFormatContext<'_>> for SuiteChildStatement<'_> {
fn fmt(&self, f: &mut Formatter<PyFormatContext<'_>>) -> FormatResult<()> {
match self {
SuiteChildStatement::Docstring(docstring) => docstring.fmt(f),
SuiteChildStatement::Other(statement) => statement.format().fmt(f),
}
}
}
#[cfg(test)]
mod tests {
use ruff_formatter::format;