2019-01-10 08:50:49 -06:00
|
|
|
use itertools::Itertools;
|
|
|
|
use ra_syntax::{
|
|
|
|
SourceFile, TextRange, TextUnit, AstNode, SyntaxNode,
|
2019-01-10 09:32:02 -06:00
|
|
|
SyntaxKind::{self, WHITESPACE, COMMA, R_CURLY, R_PAREN, R_BRACK},
|
2019-02-21 10:49:03 -06:00
|
|
|
algo::{find_covering_node, non_trivia_sibling},
|
2019-01-10 08:50:49 -06:00
|
|
|
ast,
|
2019-02-21 10:49:03 -06:00
|
|
|
Direction,
|
2019-01-10 08:50:49 -06:00
|
|
|
};
|
2019-02-08 11:58:27 -06:00
|
|
|
use ra_fmt::{
|
|
|
|
compute_ws, extract_trivial_expression
|
|
|
|
};
|
2019-03-21 15:42:22 -05:00
|
|
|
use ra_text_edit::{TextEdit, TextEditBuilder};
|
2019-01-10 08:50:49 -06:00
|
|
|
|
2019-03-22 11:10:20 -05:00
|
|
|
pub fn join_lines(file: &SourceFile, range: TextRange) -> TextEdit {
|
|
|
|
let range = if range.is_empty() {
|
2019-01-10 08:50:49 -06:00
|
|
|
let syntax = file.syntax();
|
2019-03-22 11:10:20 -05:00
|
|
|
let text = syntax.text().slice(range.start()..);
|
2019-01-10 08:50:49 -06:00
|
|
|
let pos = match text.find('\n') {
|
2019-03-21 15:42:22 -05:00
|
|
|
None => return TextEditBuilder::default().finish(),
|
2019-01-10 08:50:49 -06:00
|
|
|
Some(pos) => pos,
|
|
|
|
};
|
2019-03-22 11:10:20 -05:00
|
|
|
TextRange::offset_len(range.start() + pos, TextUnit::of_char('\n'))
|
2019-01-10 08:50:49 -06:00
|
|
|
} else {
|
2019-03-22 11:10:20 -05:00
|
|
|
range
|
2019-01-10 08:50:49 -06:00
|
|
|
};
|
|
|
|
|
|
|
|
let node = find_covering_node(file.syntax(), range);
|
|
|
|
let mut edit = TextEditBuilder::default();
|
|
|
|
for node in node.descendants() {
|
|
|
|
let text = match node.leaf_text() {
|
|
|
|
Some(text) => text,
|
|
|
|
None => continue,
|
|
|
|
};
|
|
|
|
let range = match range.intersection(&node.range()) {
|
|
|
|
Some(range) => range,
|
|
|
|
None => continue,
|
|
|
|
} - node.range().start();
|
|
|
|
for (pos, _) in text[range].bytes().enumerate().filter(|&(_, b)| b == b'\n') {
|
|
|
|
let pos: TextUnit = (pos as u32).into();
|
|
|
|
let off = node.range().start() + range.start() + pos;
|
|
|
|
if !edit.invalidates_offset(off) {
|
|
|
|
remove_newline(&mut edit, node, text.as_str(), off);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-03-21 15:42:22 -05:00
|
|
|
edit.finish()
|
2019-01-10 08:50:49 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
fn remove_newline(
|
|
|
|
edit: &mut TextEditBuilder,
|
|
|
|
node: &SyntaxNode,
|
|
|
|
node_text: &str,
|
|
|
|
offset: TextUnit,
|
|
|
|
) {
|
|
|
|
if node.kind() != WHITESPACE || node_text.bytes().filter(|&b| b == b'\n').count() != 1 {
|
|
|
|
// The node is either the first or the last in the file
|
|
|
|
let suff = &node_text[TextRange::from_to(
|
|
|
|
offset - node.range().start() + TextUnit::of_char('\n'),
|
|
|
|
TextUnit::of_str(node_text),
|
|
|
|
)];
|
|
|
|
let spaces = suff.bytes().take_while(|&b| b == b' ').count();
|
|
|
|
|
2019-02-08 05:49:43 -06:00
|
|
|
edit.replace(TextRange::offset_len(offset, ((spaces + 1) as u32).into()), " ".to_string());
|
2019-01-10 08:50:49 -06:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Special case that turns something like:
|
|
|
|
//
|
|
|
|
// ```
|
|
|
|
// my_function({<|>
|
|
|
|
// <some-expr>
|
|
|
|
// })
|
|
|
|
// ```
|
|
|
|
//
|
|
|
|
// into `my_function(<some-expr>)`
|
|
|
|
if join_single_expr_block(edit, node).is_some() {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
// ditto for
|
|
|
|
//
|
|
|
|
// ```
|
|
|
|
// use foo::{<|>
|
|
|
|
// bar
|
|
|
|
// };
|
|
|
|
// ```
|
|
|
|
if join_single_use_tree(edit, node).is_some() {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// The node is between two other nodes
|
|
|
|
let prev = node.prev_sibling().unwrap();
|
|
|
|
let next = node.next_sibling().unwrap();
|
|
|
|
if is_trailing_comma(prev.kind(), next.kind()) {
|
|
|
|
// Removes: trailing comma, newline (incl. surrounding whitespace)
|
|
|
|
edit.delete(TextRange::from_to(prev.range().start(), node.range().end()));
|
|
|
|
} else if prev.kind() == COMMA && next.kind() == R_CURLY {
|
|
|
|
// Removes: comma, newline (incl. surrounding whitespace)
|
2019-02-08 05:49:43 -06:00
|
|
|
let space = if let Some(left) = prev.prev_sibling() { compute_ws(left, next) } else { " " };
|
2019-01-10 08:50:49 -06:00
|
|
|
edit.replace(
|
|
|
|
TextRange::from_to(prev.range().start(), node.range().end()),
|
|
|
|
space.to_string(),
|
|
|
|
);
|
|
|
|
} else if let (Some(_), Some(next)) = (ast::Comment::cast(prev), ast::Comment::cast(next)) {
|
|
|
|
// Removes: newline (incl. surrounding whitespace), start of the next comment
|
|
|
|
edit.delete(TextRange::from_to(
|
|
|
|
node.range().start(),
|
|
|
|
next.syntax().range().start() + TextUnit::of_str(next.prefix()),
|
|
|
|
));
|
|
|
|
} else {
|
|
|
|
// Remove newline but add a computed amount of whitespace characters
|
|
|
|
edit.replace(node.range(), compute_ws(prev, next).to_string());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-02-21 10:26:27 -06:00
|
|
|
fn has_comma_after(node: &SyntaxNode) -> bool {
|
2019-02-21 10:49:03 -06:00
|
|
|
match non_trivia_sibling(node, Direction::Next) {
|
|
|
|
Some(n) => n.kind() == COMMA,
|
2019-02-21 10:26:27 -06:00
|
|
|
_ => false,
|
2019-02-21 09:51:41 -06:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-01-10 08:50:49 -06:00
|
|
|
fn join_single_expr_block(edit: &mut TextEditBuilder, node: &SyntaxNode) -> Option<()> {
|
|
|
|
let block = ast::Block::cast(node.parent()?)?;
|
|
|
|
let block_expr = ast::BlockExpr::cast(block.syntax().parent()?)?;
|
2019-01-10 09:32:02 -06:00
|
|
|
let expr = extract_trivial_expression(block)?;
|
2019-02-21 09:51:41 -06:00
|
|
|
|
|
|
|
let block_range = block_expr.syntax().range();
|
2019-02-21 10:26:27 -06:00
|
|
|
let mut buf = expr.syntax().text().to_string();
|
2019-02-21 09:51:41 -06:00
|
|
|
|
|
|
|
// Match block needs to have a comma after the block
|
|
|
|
if let Some(match_arm) = block_expr.syntax().parent().and_then(ast::MatchArm::cast) {
|
2019-02-21 10:26:27 -06:00
|
|
|
if !has_comma_after(match_arm.syntax()) {
|
|
|
|
buf.push(',');
|
|
|
|
}
|
2019-02-21 09:51:41 -06:00
|
|
|
}
|
2019-02-21 10:26:27 -06:00
|
|
|
|
|
|
|
edit.replace(block_range, buf);
|
|
|
|
|
2019-01-10 08:50:49 -06:00
|
|
|
Some(())
|
|
|
|
}
|
|
|
|
|
|
|
|
fn join_single_use_tree(edit: &mut TextEditBuilder, node: &SyntaxNode) -> Option<()> {
|
|
|
|
let use_tree_list = ast::UseTreeList::cast(node.parent()?)?;
|
|
|
|
let (tree,) = use_tree_list.use_trees().collect_tuple()?;
|
2019-02-08 05:49:43 -06:00
|
|
|
edit.replace(use_tree_list.syntax().range(), tree.syntax().text().to_string());
|
2019-01-10 08:50:49 -06:00
|
|
|
Some(())
|
|
|
|
}
|
|
|
|
|
|
|
|
fn is_trailing_comma(left: SyntaxKind, right: SyntaxKind) -> bool {
|
|
|
|
match (left, right) {
|
|
|
|
(COMMA, R_PAREN) | (COMMA, R_BRACK) => true,
|
|
|
|
_ => false,
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
#[cfg(test)]
|
|
|
|
mod tests {
|
|
|
|
use crate::test_utils::{assert_eq_text, check_action, extract_range};
|
|
|
|
|
|
|
|
use super::*;
|
|
|
|
|
|
|
|
fn check_join_lines(before: &str, after: &str) {
|
|
|
|
check_action(before, after, |file, offset| {
|
|
|
|
let range = TextRange::offset_len(offset, 0.into());
|
|
|
|
let res = join_lines(file, range);
|
|
|
|
Some(res)
|
|
|
|
})
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_comma() {
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
<|>foo(1,
|
|
|
|
)
|
|
|
|
}
|
|
|
|
",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
<|>foo(1)
|
|
|
|
}
|
|
|
|
",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_lambda_block() {
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
pub fn reparse(&self, edit: &AtomTextEdit) -> File {
|
|
|
|
<|>self.incremental_reparse(edit).unwrap_or_else(|| {
|
|
|
|
self.full_reparse(edit)
|
|
|
|
})
|
|
|
|
}
|
|
|
|
",
|
|
|
|
r"
|
|
|
|
pub fn reparse(&self, edit: &AtomTextEdit) -> File {
|
|
|
|
<|>self.incremental_reparse(edit).unwrap_or_else(|| self.full_reparse(edit))
|
|
|
|
}
|
|
|
|
",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_block() {
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
foo(<|>{
|
|
|
|
92
|
|
|
|
})
|
|
|
|
}",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
foo(<|>92)
|
|
|
|
}",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
2019-02-21 05:06:21 -06:00
|
|
|
#[test]
|
|
|
|
fn join_lines_adds_comma_for_block_in_match_arm() {
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo(e: Result<U, V>) {
|
|
|
|
match e {
|
|
|
|
Ok(u) => <|>{
|
|
|
|
u.foo()
|
|
|
|
}
|
|
|
|
Err(v) => v,
|
|
|
|
}
|
|
|
|
}",
|
|
|
|
r"
|
|
|
|
fn foo(e: Result<U, V>) {
|
|
|
|
match e {
|
|
|
|
Ok(u) => <|>u.foo(),
|
|
|
|
Err(v) => v,
|
|
|
|
}
|
|
|
|
}",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
2019-02-21 09:51:41 -06:00
|
|
|
#[test]
|
|
|
|
fn join_lines_keeps_comma_for_block_in_match_arm() {
|
|
|
|
// We already have a comma
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo(e: Result<U, V>) {
|
|
|
|
match e {
|
|
|
|
Ok(u) => <|>{
|
|
|
|
u.foo()
|
|
|
|
},
|
|
|
|
Err(v) => v,
|
|
|
|
}
|
|
|
|
}",
|
|
|
|
r"
|
|
|
|
fn foo(e: Result<U, V>) {
|
|
|
|
match e {
|
|
|
|
Ok(u) => <|>u.foo(),
|
|
|
|
Err(v) => v,
|
|
|
|
}
|
|
|
|
}",
|
|
|
|
);
|
|
|
|
|
|
|
|
// comma with whitespace between brace and ,
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo(e: Result<U, V>) {
|
|
|
|
match e {
|
|
|
|
Ok(u) => <|>{
|
|
|
|
u.foo()
|
|
|
|
} ,
|
|
|
|
Err(v) => v,
|
|
|
|
}
|
|
|
|
}",
|
|
|
|
r"
|
|
|
|
fn foo(e: Result<U, V>) {
|
|
|
|
match e {
|
2019-02-21 10:26:27 -06:00
|
|
|
Ok(u) => <|>u.foo() ,
|
2019-02-21 09:51:41 -06:00
|
|
|
Err(v) => v,
|
|
|
|
}
|
|
|
|
}",
|
|
|
|
);
|
|
|
|
|
|
|
|
// comma with newline between brace and ,
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo(e: Result<U, V>) {
|
|
|
|
match e {
|
|
|
|
Ok(u) => <|>{
|
|
|
|
u.foo()
|
|
|
|
}
|
|
|
|
,
|
|
|
|
Err(v) => v,
|
|
|
|
}
|
|
|
|
}",
|
|
|
|
r"
|
|
|
|
fn foo(e: Result<U, V>) {
|
|
|
|
match e {
|
2019-02-21 10:26:27 -06:00
|
|
|
Ok(u) => <|>u.foo()
|
|
|
|
,
|
2019-02-21 09:51:41 -06:00
|
|
|
Err(v) => v,
|
|
|
|
}
|
|
|
|
}",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn join_lines_keeps_comma_with_single_arg_tuple() {
|
|
|
|
// A single arg tuple
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
let x = (<|>{
|
|
|
|
4
|
|
|
|
},);
|
|
|
|
}",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
let x = (<|>4,);
|
|
|
|
}",
|
|
|
|
);
|
|
|
|
|
|
|
|
// single arg tuple with whitespace between brace and comma
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
let x = (<|>{
|
|
|
|
4
|
|
|
|
} ,);
|
|
|
|
}",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
2019-02-21 10:26:27 -06:00
|
|
|
let x = (<|>4 ,);
|
2019-02-21 09:51:41 -06:00
|
|
|
}",
|
|
|
|
);
|
|
|
|
|
|
|
|
// single arg tuple with newline between brace and comma
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
let x = (<|>{
|
|
|
|
4
|
|
|
|
}
|
|
|
|
,);
|
|
|
|
}",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
2019-02-21 10:26:27 -06:00
|
|
|
let x = (<|>4
|
|
|
|
,);
|
2019-02-21 09:51:41 -06:00
|
|
|
}",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
2019-01-10 08:50:49 -06:00
|
|
|
#[test]
|
|
|
|
fn test_join_lines_use_items_left() {
|
|
|
|
// No space after the '{'
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
<|>use ra_syntax::{
|
|
|
|
TextUnit, TextRange,
|
|
|
|
};",
|
|
|
|
r"
|
|
|
|
<|>use ra_syntax::{TextUnit, TextRange,
|
|
|
|
};",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_use_items_right() {
|
|
|
|
// No space after the '}'
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
use ra_syntax::{
|
|
|
|
<|> TextUnit, TextRange
|
|
|
|
};",
|
|
|
|
r"
|
|
|
|
use ra_syntax::{
|
|
|
|
<|> TextUnit, TextRange};",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_use_items_right_comma() {
|
|
|
|
// No space after the '}'
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
use ra_syntax::{
|
|
|
|
<|> TextUnit, TextRange,
|
|
|
|
};",
|
|
|
|
r"
|
|
|
|
use ra_syntax::{
|
|
|
|
<|> TextUnit, TextRange};",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_use_tree() {
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
use ra_syntax::{
|
|
|
|
algo::<|>{
|
|
|
|
find_leaf_at_offset,
|
|
|
|
},
|
|
|
|
ast,
|
|
|
|
};",
|
|
|
|
r"
|
|
|
|
use ra_syntax::{
|
|
|
|
algo::<|>find_leaf_at_offset,
|
|
|
|
ast,
|
|
|
|
};",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_normal_comments() {
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
// Hello<|>
|
|
|
|
// world!
|
|
|
|
}
|
|
|
|
",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
// Hello<|> world!
|
|
|
|
}
|
|
|
|
",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_doc_comments() {
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
/// Hello<|>
|
|
|
|
/// world!
|
|
|
|
}
|
|
|
|
",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
/// Hello<|> world!
|
|
|
|
}
|
|
|
|
",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_mod_comments() {
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
//! Hello<|>
|
|
|
|
//! world!
|
|
|
|
}
|
|
|
|
",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
//! Hello<|> world!
|
|
|
|
}
|
|
|
|
",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_multiline_comments_1() {
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
// Hello<|>
|
|
|
|
/* world! */
|
|
|
|
}
|
|
|
|
",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
// Hello<|> world! */
|
|
|
|
}
|
|
|
|
",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_multiline_comments_2() {
|
|
|
|
check_join_lines(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
// The<|>
|
|
|
|
/* quick
|
|
|
|
brown
|
|
|
|
fox! */
|
|
|
|
}
|
|
|
|
",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
// The<|> quick
|
|
|
|
brown
|
|
|
|
fox! */
|
|
|
|
}
|
|
|
|
",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
fn check_join_lines_sel(before: &str, after: &str) {
|
|
|
|
let (sel, before) = extract_range(before);
|
|
|
|
let file = SourceFile::parse(&before);
|
|
|
|
let result = join_lines(&file, sel);
|
2019-03-22 11:10:20 -05:00
|
|
|
let actual = result.apply(&before);
|
2019-01-10 08:50:49 -06:00
|
|
|
assert_eq_text!(after, &actual);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_selection_fn_args() {
|
|
|
|
check_join_lines_sel(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
<|>foo(1,
|
|
|
|
2,
|
|
|
|
3,
|
|
|
|
<|>)
|
|
|
|
}
|
|
|
|
",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
foo(1, 2, 3)
|
|
|
|
}
|
|
|
|
",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_selection_struct() {
|
|
|
|
check_join_lines_sel(
|
|
|
|
r"
|
|
|
|
struct Foo <|>{
|
|
|
|
f: u32,
|
|
|
|
}<|>
|
|
|
|
",
|
|
|
|
r"
|
|
|
|
struct Foo { f: u32 }
|
|
|
|
",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_selection_dot_chain() {
|
|
|
|
check_join_lines_sel(
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
join(<|>type_params.type_params()
|
|
|
|
.filter_map(|it| it.name())
|
|
|
|
.map(|it| it.text())<|>)
|
|
|
|
}",
|
|
|
|
r"
|
|
|
|
fn foo() {
|
|
|
|
join(type_params.type_params().filter_map(|it| it.name()).map(|it| it.text()))
|
|
|
|
}",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn test_join_lines_selection_lambda_block_body() {
|
|
|
|
check_join_lines_sel(
|
|
|
|
r"
|
|
|
|
pub fn handle_find_matching_brace() {
|
|
|
|
params.offsets
|
|
|
|
.map(|offset| <|>{
|
|
|
|
world.analysis().matching_brace(&file, offset).unwrap_or(offset)
|
|
|
|
}<|>)
|
|
|
|
.collect();
|
|
|
|
}",
|
|
|
|
r"
|
|
|
|
pub fn handle_find_matching_brace() {
|
|
|
|
params.offsets
|
|
|
|
.map(|offset| world.analysis().matching_brace(&file, offset).unwrap_or(offset))
|
|
|
|
.collect();
|
|
|
|
}",
|
|
|
|
);
|
|
|
|
}
|
|
|
|
}
|