Skip over folded regions when iterating over multibuffer chunks (#15646)

This commit weaves through new APIs for language::BufferChunks, multi_buffer::MultiBufferChunks and inlay_map::InlayChunks that allow seeking with an upper-bound. This allows us to omit doing syntax highligting and looking up diagnostics for folded ranges. This in turn directly improves performance of assistant panel with large contexts.

Release Notes:

- Fixed poor performance when editing in the assistant panel after
inserting large files using slash commands

---------

Co-authored-by: Max <max@zed.dev>
Co-authored-by: Max Brunsfeld <maxbrunsfeld@gmail.com>
This commit is contained in:
Piotr Osiewicz 2024-08-02 19:51:26 +02:00 committed by GitHub
parent 390815dd76
commit ad11d83724
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
9 changed files with 261 additions and 106 deletions

1
Cargo.lock generated
View file

@ -3599,6 +3599,7 @@ dependencies = [
"time",
"time_format",
"tree-sitter-html",
"tree-sitter-md",
"tree-sitter-rust",
"tree-sitter-typescript",
"ui",

View file

@ -93,6 +93,7 @@ settings = { workspace = true, features = ["test-support"] }
text = { workspace = true, features = ["test-support"] }
theme = { workspace = true, features = ["test-support"] }
tree-sitter-html.workspace = true
tree-sitter-md.workspace = true
tree-sitter-rust.workspace = true
tree-sitter-typescript.workspace = true
unindent.workspace = true

View file

@ -11,7 +11,7 @@ use std::{
ops::{Add, AddAssign, Deref, DerefMut, Range, Sub},
sync::Arc,
};
use sum_tree::{Bias, Cursor, FilterCursor, SumTree};
use sum_tree::{Bias, Cursor, FilterCursor, SumTree, Summary};
use util::post_inc;
#[derive(Clone)]
@ -277,6 +277,17 @@ impl FoldMap {
"transform tree does not match inlay snapshot's length"
);
let mut prev_transform_isomorphic = false;
for transform in self.snapshot.transforms.iter() {
if !transform.is_fold() && prev_transform_isomorphic {
panic!(
"found adjacent isomorphic transforms: {:?}",
self.snapshot.transforms.items(&())
);
}
prev_transform_isomorphic = !transform.is_fold();
}
let mut folds = self.snapshot.folds.iter().peekable();
while let Some(fold) = folds.next() {
if let Some(next_fold) = folds.peek() {
@ -303,11 +314,24 @@ impl FoldMap {
} else {
let mut inlay_edits_iter = inlay_edits.iter().cloned().peekable();
let mut new_transforms = SumTree::new();
let mut new_transforms = SumTree::<Transform>::new();
let mut cursor = self.snapshot.transforms.cursor::<InlayOffset>();
cursor.seek(&InlayOffset(0), Bias::Right, &());
while let Some(mut edit) = inlay_edits_iter.next() {
if let Some(item) = cursor.item() {
if !item.is_fold() {
new_transforms.update_last(
|transform| {
if !transform.is_fold() {
transform.summary.add_summary(&item.summary, &());
cursor.next(&());
}
},
&(),
);
}
}
new_transforms.append(cursor.slice(&edit.old.start, Bias::Left, &()), &());
edit.new.start -= edit.old.start - *cursor.start();
edit.old.start = *cursor.start();
@ -392,16 +416,7 @@ impl FoldMap {
if fold_range.start.0 > sum.input.len {
let text_summary = inlay_snapshot
.text_summary_for_range(InlayOffset(sum.input.len)..fold_range.start);
new_transforms.push(
Transform {
summary: TransformSummary {
output: text_summary.clone(),
input: text_summary,
},
placeholder: None,
},
&(),
);
push_isomorphic(&mut new_transforms, text_summary);
}
if fold_range.end > fold_range.start {
@ -438,32 +453,14 @@ impl FoldMap {
if sum.input.len < edit.new.end.0 {
let text_summary = inlay_snapshot
.text_summary_for_range(InlayOffset(sum.input.len)..edit.new.end);
new_transforms.push(
Transform {
summary: TransformSummary {
output: text_summary.clone(),
input: text_summary,
},
placeholder: None,
},
&(),
);
push_isomorphic(&mut new_transforms, text_summary);
}
}
new_transforms.append(cursor.suffix(&()), &());
if new_transforms.is_empty() {
let text_summary = inlay_snapshot.text_summary();
new_transforms.push(
Transform {
summary: TransformSummary {
output: text_summary.clone(),
input: text_summary,
},
placeholder: None,
},
&(),
);
push_isomorphic(&mut new_transforms, text_summary);
}
drop(cursor);
@ -715,17 +712,25 @@ impl FoldSnapshot {
highlights: Highlights<'a>,
) -> FoldChunks<'a> {
let mut transform_cursor = self.transforms.cursor::<(FoldOffset, InlayOffset)>();
transform_cursor.seek(&range.start, Bias::Right, &());
let inlay_end = {
transform_cursor.seek(&range.end, Bias::Right, &());
let overshoot = range.end.0 - transform_cursor.start().0 .0;
let inlay_start = {
let overshoot = range.start.0 - transform_cursor.start().0 .0;
transform_cursor.start().1 + InlayOffset(overshoot)
};
let inlay_start = {
transform_cursor.seek(&range.start, Bias::Right, &());
let overshoot = range.start.0 - transform_cursor.start().0 .0;
let transform_end = transform_cursor.end(&());
let inlay_end = if transform_cursor
.item()
.map_or(true, |transform| transform.is_fold())
{
inlay_start
} else if range.end < transform_end.0 {
let overshoot = range.end.0 - transform_cursor.start().0 .0;
transform_cursor.start().1 + InlayOffset(overshoot)
} else {
transform_end.1
};
FoldChunks {
@ -737,8 +742,8 @@ impl FoldSnapshot {
),
inlay_chunk: None,
inlay_offset: inlay_start,
output_offset: range.start.0,
max_output_offset: range.end.0,
output_offset: range.start,
max_output_offset: range.end,
}
}
@ -783,6 +788,32 @@ impl FoldSnapshot {
}
}
fn push_isomorphic(transforms: &mut SumTree<Transform>, summary: TextSummary) {
let mut did_merge = false;
transforms.update_last(
|last| {
if !last.is_fold() {
last.summary.input += summary.clone();
last.summary.output += summary.clone();
did_merge = true;
}
},
&(),
);
if !did_merge {
transforms.push(
Transform {
summary: TransformSummary {
input: summary.clone(),
output: summary,
},
placeholder: None,
},
&(),
)
}
}
fn intersecting_folds<'a, T>(
inlay_snapshot: &'a InlaySnapshot,
folds: &'a SumTree<Fold>,
@ -1079,8 +1110,8 @@ pub struct FoldChunks<'a> {
inlay_chunks: InlayChunks<'a>,
inlay_chunk: Option<(InlayOffset, Chunk<'a>)>,
inlay_offset: InlayOffset,
output_offset: usize,
max_output_offset: usize,
output_offset: FoldOffset,
max_output_offset: FoldOffset,
}
impl<'a> Iterator for FoldChunks<'a> {
@ -1098,7 +1129,6 @@ impl<'a> Iterator for FoldChunks<'a> {
if let Some(placeholder) = transform.placeholder.as_ref() {
self.inlay_chunk.take();
self.inlay_offset += InlayOffset(transform.summary.input.len);
self.inlay_chunks.seek(self.inlay_offset);
while self.inlay_offset >= self.transform_cursor.end(&()).1
&& self.transform_cursor.item().is_some()
@ -1106,7 +1136,7 @@ impl<'a> Iterator for FoldChunks<'a> {
self.transform_cursor.next(&());
}
self.output_offset += placeholder.text.len();
self.output_offset.0 += placeholder.text.len();
return Some(Chunk {
text: placeholder.text,
renderer: Some(placeholder.renderer.clone()),
@ -1114,6 +1144,23 @@ impl<'a> Iterator for FoldChunks<'a> {
});
}
// When we reach a non-fold region, seek the underlying text
// chunk iterator to the next unfolded range.
if self.inlay_offset == self.transform_cursor.start().1
&& self.inlay_chunks.offset() != self.inlay_offset
{
let transform_start = self.transform_cursor.start();
let transform_end = self.transform_cursor.end(&());
let inlay_end = if self.max_output_offset < transform_end.0 {
let overshoot = self.max_output_offset.0 - transform_start.0 .0;
transform_start.1 + InlayOffset(overshoot)
} else {
transform_end.1
};
self.inlay_chunks.seek(self.inlay_offset..inlay_end);
}
// Retrieve a chunk from the current location in the buffer.
if self.inlay_chunk.is_none() {
let chunk_offset = self.inlay_chunks.offset();
@ -1136,7 +1183,7 @@ impl<'a> Iterator for FoldChunks<'a> {
}
self.inlay_offset = chunk_end;
self.output_offset += chunk.text.len();
self.output_offset.0 += chunk.text.len();
return Some(chunk);
}

View file

@ -225,14 +225,16 @@ pub struct InlayChunks<'a> {
}
impl<'a> InlayChunks<'a> {
pub fn seek(&mut self, offset: InlayOffset) {
self.transforms.seek(&offset, Bias::Right, &());
pub fn seek(&mut self, new_range: Range<InlayOffset>) {
self.transforms.seek(&new_range.start, Bias::Right, &());
let buffer_offset = self.snapshot.to_buffer_offset(offset);
self.buffer_chunks.seek(buffer_offset);
let buffer_range = self.snapshot.to_buffer_offset(new_range.start)
..self.snapshot.to_buffer_offset(new_range.end);
self.buffer_chunks.seek(buffer_range);
self.inlay_chunks = None;
self.buffer_chunk = None;
self.output_offset = offset;
self.output_offset = new_range.start;
self.max_output_offset = new_range.end;
}
pub fn offset(&self) -> InlayOffset {

View file

@ -3668,6 +3668,63 @@ fn test_duplicate_line(cx: &mut TestAppContext) {
);
});
}
#[gpui::test]
async fn test_fold_perf(cx: &mut TestAppContext) {
use std::fmt::Write;
init_test(cx, |_| {});
let mut view = EditorTestContext::new(cx).await;
let language_registry = view.language_registry();
let language_name = Arc::from("Markdown");
let md_language = Arc::new(
Language::new(
LanguageConfig {
name: Arc::clone(&language_name),
matcher: LanguageMatcher {
path_suffixes: vec!["md".to_string()],
..Default::default()
},
..Default::default()
},
Some(tree_sitter_md::language()),
)
.with_highlights_query(
r#"
"#,
)
.unwrap(),
);
language_registry.add(md_language.clone());
let mut text = String::default();
writeln!(&mut text, "start").unwrap();
writeln!(&mut text, "```").unwrap();
const LINE_COUNT: u32 = 10000;
for i in 0..LINE_COUNT {
writeln!(&mut text, "{i}").unwrap();
}
writeln!(&mut text, "```").unwrap();
writeln!(&mut text, "end").unwrap();
view.update_buffer(|buffer, cx| {
buffer.set_language(Some(md_language), cx);
});
let t0 = Instant::now();
_ = view.update_editor(|view, cx| {
eprintln!("Text length: {}", text.len());
view.set_text(text, cx);
eprintln!(">>");
view.fold_ranges(
vec![(
Point::new(1, 0)..Point::new(LINE_COUNT + 2, 3),
FoldPlaceholder::test(),
)],
false,
cx,
);
});
eprintln!("{:?}", t0.elapsed());
eprintln!("<<");
}
#[gpui::test]
fn test_move_line_up_down(cx: &mut TestAppContext) {

View file

@ -449,6 +449,7 @@ struct BufferChunkHighlights<'a> {
/// An iterator that yields chunks of a buffer's text, along with their
/// syntax highlights and diagnostic status.
pub struct BufferChunks<'a> {
buffer_snapshot: Option<&'a BufferSnapshot>,
range: Range<usize>,
chunks: text::Chunks<'a>,
diagnostic_endpoints: Peekable<vec::IntoIter<DiagnosticEndpoint>>,
@ -2475,6 +2476,17 @@ impl BufferSnapshot {
None
}
fn get_highlights(&self, range: Range<usize>) -> (SyntaxMapCaptures, Vec<HighlightMap>) {
let captures = self.syntax.captures(range, &self.text, |grammar| {
grammar.highlights_query.as_ref()
});
let highlight_maps = captures
.grammars()
.into_iter()
.map(|grammar| grammar.highlight_map())
.collect();
(captures, highlight_maps)
}
/// Iterates over chunks of text in the given range of the buffer. Text is chunked
/// in an arbitrary way due to being stored in a [`Rope`](text::Rope). The text is also
/// returned in chunks where each chunk has a single syntax highlighting style and
@ -2483,36 +2495,11 @@ impl BufferSnapshot {
let range = range.start.to_offset(self)..range.end.to_offset(self);
let mut syntax = None;
let mut diagnostic_endpoints = Vec::new();
if language_aware {
let captures = self.syntax.captures(range.clone(), &self.text, |grammar| {
grammar.highlights_query.as_ref()
});
let highlight_maps = captures
.grammars()
.into_iter()
.map(|grammar| grammar.highlight_map())
.collect();
syntax = Some((captures, highlight_maps));
for entry in self.diagnostics_in_range::<_, usize>(range.clone(), false) {
diagnostic_endpoints.push(DiagnosticEndpoint {
offset: entry.range.start,
is_start: true,
severity: entry.diagnostic.severity,
is_unnecessary: entry.diagnostic.is_unnecessary,
});
diagnostic_endpoints.push(DiagnosticEndpoint {
offset: entry.range.end,
is_start: false,
severity: entry.diagnostic.severity,
is_unnecessary: entry.diagnostic.is_unnecessary,
});
}
diagnostic_endpoints
.sort_unstable_by_key(|endpoint| (endpoint.offset, !endpoint.is_start));
syntax = Some(self.get_highlights(range.clone()));
}
BufferChunks::new(self.text.as_rope(), range, syntax, diagnostic_endpoints)
BufferChunks::new(self.text.as_rope(), range, syntax, Some(self))
}
/// Invokes the given callback for each line of text in the given range of the buffer.
@ -2936,7 +2923,7 @@ impl BufferSnapshot {
}
let mut offset = buffer_range.start;
chunks.seek(offset);
chunks.seek(buffer_range.clone());
for mut chunk in chunks.by_ref() {
if chunk.text.len() > buffer_range.end - offset {
chunk.text = &chunk.text[0..(buffer_range.end - offset)];
@ -3731,7 +3718,7 @@ impl<'a> BufferChunks<'a> {
text: &'a Rope,
range: Range<usize>,
syntax: Option<(SyntaxMapCaptures<'a>, Vec<HighlightMap>)>,
diagnostic_endpoints: Vec<DiagnosticEndpoint>,
buffer_snapshot: Option<&'a BufferSnapshot>,
) -> Self {
let mut highlights = None;
if let Some((captures, highlight_maps)) = syntax {
@ -3743,11 +3730,12 @@ impl<'a> BufferChunks<'a> {
})
}
let diagnostic_endpoints = diagnostic_endpoints.into_iter().peekable();
let diagnostic_endpoints = Vec::new().into_iter().peekable();
let chunks = text.chunks_in_range(range.clone());
BufferChunks {
let mut this = BufferChunks {
range,
buffer_snapshot,
chunks,
diagnostic_endpoints,
error_depth: 0,
@ -3756,30 +3744,72 @@ impl<'a> BufferChunks<'a> {
hint_depth: 0,
unnecessary_depth: 0,
highlights,
}
};
this.initialize_diagnostic_endpoints();
this
}
/// Seeks to the given byte offset in the buffer.
pub fn seek(&mut self, offset: usize) {
self.range.start = offset;
self.chunks.seek(self.range.start);
pub fn seek(&mut self, range: Range<usize>) {
let old_range = std::mem::replace(&mut self.range, range.clone());
self.chunks.set_range(self.range.clone());
if let Some(highlights) = self.highlights.as_mut() {
highlights
.stack
.retain(|(end_offset, _)| *end_offset > offset);
if let Some(capture) = &highlights.next_capture {
if offset >= capture.node.start_byte() {
let next_capture_end = capture.node.end_byte();
if offset < next_capture_end {
highlights.stack.push((
next_capture_end,
highlights.highlight_maps[capture.grammar_index].get(capture.index),
));
if old_range.start >= self.range.start && old_range.end <= self.range.end {
// Reuse existing highlights stack, as the new range is a subrange of the old one.
highlights
.stack
.retain(|(end_offset, _)| *end_offset > range.start);
if let Some(capture) = &highlights.next_capture {
if range.start >= capture.node.start_byte() {
let next_capture_end = capture.node.end_byte();
if range.start < next_capture_end {
highlights.stack.push((
next_capture_end,
highlights.highlight_maps[capture.grammar_index].get(capture.index),
));
}
highlights.next_capture.take();
}
highlights.next_capture.take();
}
} else if let Some(snapshot) = self.buffer_snapshot {
let (captures, highlight_maps) = snapshot.get_highlights(self.range.clone());
*highlights = BufferChunkHighlights {
captures,
next_capture: None,
stack: Default::default(),
highlight_maps,
};
} else {
// We cannot obtain new highlights for a language-aware buffer iterator, as we don't have a buffer snapshot.
// Seeking such BufferChunks is not supported.
debug_assert!(false, "Attempted to seek on a language-aware buffer iterator without associated buffer snapshot");
}
highlights.captures.set_byte_range(self.range.clone());
self.initialize_diagnostic_endpoints();
}
}
fn initialize_diagnostic_endpoints(&mut self) {
if let Some(buffer) = self.buffer_snapshot {
let mut diagnostic_endpoints = Vec::new();
for entry in buffer.diagnostics_in_range::<_, usize>(self.range.clone(), false) {
diagnostic_endpoints.push(DiagnosticEndpoint {
offset: entry.range.start,
is_start: true,
severity: entry.diagnostic.severity,
is_unnecessary: entry.diagnostic.is_unnecessary,
});
diagnostic_endpoints.push(DiagnosticEndpoint {
offset: entry.range.end,
is_start: false,
severity: entry.diagnostic.severity,
is_unnecessary: entry.diagnostic.is_unnecessary,
});
}
diagnostic_endpoints
.sort_unstable_by_key(|endpoint| (endpoint.offset, !endpoint.is_start));
self.diagnostic_endpoints = diagnostic_endpoints.into_iter().peekable();
}
}

View file

@ -1341,7 +1341,7 @@ impl Language {
});
let highlight_maps = vec![grammar.highlight_map()];
let mut offset = 0;
for chunk in BufferChunks::new(text, range, Some((captures, highlight_maps)), vec![]) {
for chunk in BufferChunks::new(text, range, Some((captures, highlight_maps)), None) {
let end_offset = offset + chunk.text.len();
if let Some(highlight_id) = chunk.syntax_highlight_id {
if !highlight_id.is_default() {

View file

@ -2425,7 +2425,7 @@ impl MultiBufferSnapshot {
excerpt_chunks: None,
language_aware,
};
chunks.seek(range.start);
chunks.seek(range);
chunks
}
@ -4164,10 +4164,19 @@ impl Excerpt {
}
}
fn seek_chunks(&self, excerpt_chunks: &mut ExcerptChunks, offset: usize) {
fn seek_chunks(&self, excerpt_chunks: &mut ExcerptChunks, range: Range<usize>) {
let content_start = self.range.context.start.to_offset(&self.buffer);
let chunks_start = content_start + offset;
excerpt_chunks.content_chunks.seek(chunks_start);
let chunks_start = content_start + range.start;
let chunks_end = content_start + cmp::min(range.end, self.text_summary.len);
excerpt_chunks.content_chunks.seek(chunks_start..chunks_end);
excerpt_chunks.footer_height = if self.has_trailing_newline
&& range.start <= self.text_summary.len
&& range.end > self.text_summary.len
{
1
} else {
0
};
}
fn bytes_in_range(&self, range: Range<usize>) -> ExcerptBytes {
@ -4504,9 +4513,9 @@ impl<'a> MultiBufferChunks<'a> {
self.range.start
}
pub fn seek(&mut self, offset: usize) {
self.range.start = offset;
self.excerpts.seek(&offset, Bias::Right, &());
pub fn seek(&mut self, new_range: Range<usize>) {
self.range = new_range.clone();
self.excerpts.seek(&new_range.start, Bias::Right, &());
if let Some(excerpt) = self.excerpts.item() {
let excerpt_start = self.excerpts.start();
if let Some(excerpt_chunks) = self
@ -4514,7 +4523,10 @@ impl<'a> MultiBufferChunks<'a> {
.as_mut()
.filter(|chunks| excerpt.id == chunks.excerpt_id)
{
excerpt.seek_chunks(excerpt_chunks, self.range.start - excerpt_start);
excerpt.seek_chunks(
excerpt_chunks,
self.range.start - excerpt_start..self.range.end - excerpt_start,
);
} else {
self.excerpt_chunks = Some(excerpt.chunks_in_range(
self.range.start - excerpt_start..self.range.end - excerpt_start,

View file

@ -615,6 +615,11 @@ impl<'a> Chunks<'a> {
self.offset = offset;
}
pub fn set_range(&mut self, range: Range<usize>) {
self.range = range.clone();
self.seek(range.start);
}
/// Moves this cursor to the start of the next line in the rope.
///
/// This method advances the cursor to the beginning of the next line.