1use crate::assemble_excerpts::assemble_excerpt_ranges;
2use anyhow::Result;
3use collections::HashMap;
4use futures::{FutureExt, StreamExt as _, channel::mpsc, future};
5use gpui::{App, AppContext, AsyncApp, Context, Entity, EntityId, EventEmitter, Task, WeakEntity};
6use language::{Anchor, Buffer, BufferSnapshot, OffsetRangeExt as _, Point, ToOffset as _};
7use project::{LocationLink, Project, ProjectPath};
8use smallvec::SmallVec;
9use std::{
10 collections::hash_map,
11 ops::Range,
12 path::Path,
13 sync::Arc,
14 time::{Duration, Instant},
15};
16use util::paths::PathStyle;
17use util::rel_path::RelPath;
18use util::{RangeExt as _, ResultExt};
19
20mod assemble_excerpts;
21#[cfg(test)]
22mod edit_prediction_context_tests;
23#[cfg(test)]
24mod fake_definition_lsp;
25
26pub use zeta_prompt::{RelatedExcerpt, RelatedFile};
27
28const IDENTIFIER_LINE_COUNT: u32 = 3;
29
30pub struct RelatedExcerptStore {
31 project: WeakEntity<Project>,
32 related_buffers: Vec<RelatedBuffer>,
33 cache: HashMap<Identifier, Arc<CacheEntry>>,
34 update_tx: mpsc::UnboundedSender<(Entity<Buffer>, Anchor)>,
35 identifier_line_count: u32,
36}
37
38struct RelatedBuffer {
39 buffer: Entity<Buffer>,
40 path: Arc<Path>,
41 anchor_ranges: Vec<Range<Anchor>>,
42 cached_file: Option<CachedRelatedFile>,
43}
44
45struct CachedRelatedFile {
46 excerpts: Vec<RelatedExcerpt>,
47 buffer_version: clock::Global,
48}
49
50pub enum RelatedExcerptStoreEvent {
51 StartedRefresh,
52 FinishedRefresh {
53 cache_hit_count: usize,
54 cache_miss_count: usize,
55 mean_definition_latency: Duration,
56 max_definition_latency: Duration,
57 },
58}
59
60#[derive(Clone, Debug, PartialEq, Eq, Hash)]
61struct Identifier {
62 pub name: String,
63 pub range: Range<Anchor>,
64}
65
66enum DefinitionTask {
67 CacheHit(Arc<CacheEntry>),
68 CacheMiss(Task<Result<Option<Vec<LocationLink>>>>),
69}
70
71#[derive(Debug)]
72struct CacheEntry {
73 definitions: SmallVec<[CachedDefinition; 1]>,
74}
75
76#[derive(Clone, Debug)]
77struct CachedDefinition {
78 path: ProjectPath,
79 buffer: Entity<Buffer>,
80 anchor_range: Range<Anchor>,
81}
82
83const DEBOUNCE_DURATION: Duration = Duration::from_millis(100);
84
85impl EventEmitter<RelatedExcerptStoreEvent> for RelatedExcerptStore {}
86
87impl RelatedExcerptStore {
88 pub fn new(project: &Entity<Project>, cx: &mut Context<Self>) -> Self {
89 let (update_tx, mut update_rx) = mpsc::unbounded::<(Entity<Buffer>, Anchor)>();
90 cx.spawn(async move |this, cx| {
91 let executor = cx.background_executor().clone();
92 while let Some((mut buffer, mut position)) = update_rx.next().await {
93 let mut timer = executor.timer(DEBOUNCE_DURATION).fuse();
94 loop {
95 futures::select_biased! {
96 next = update_rx.next() => {
97 if let Some((new_buffer, new_position)) = next {
98 buffer = new_buffer;
99 position = new_position;
100 timer = executor.timer(DEBOUNCE_DURATION).fuse();
101 } else {
102 return anyhow::Ok(());
103 }
104 }
105 _ = timer => break,
106 }
107 }
108
109 Self::fetch_excerpts(this.clone(), buffer, position, cx).await?;
110 }
111 anyhow::Ok(())
112 })
113 .detach_and_log_err(cx);
114
115 RelatedExcerptStore {
116 project: project.downgrade(),
117 update_tx,
118 related_buffers: Vec::new(),
119 cache: Default::default(),
120 identifier_line_count: IDENTIFIER_LINE_COUNT,
121 }
122 }
123
124 pub fn set_identifier_line_count(&mut self, count: u32) {
125 self.identifier_line_count = count;
126 }
127
128 pub fn refresh(&mut self, buffer: Entity<Buffer>, position: Anchor, _: &mut Context<Self>) {
129 self.update_tx.unbounded_send((buffer, position)).ok();
130 }
131
132 pub fn related_files(&mut self, cx: &App) -> Vec<RelatedFile> {
133 self.related_buffers
134 .iter_mut()
135 .map(|related| related.related_file(cx))
136 .collect()
137 }
138
139 pub fn related_files_with_buffers(
140 &mut self,
141 cx: &App,
142 ) -> impl Iterator<Item = (RelatedFile, Entity<Buffer>)> {
143 self.related_buffers
144 .iter_mut()
145 .map(|related| (related.related_file(cx), related.buffer.clone()))
146 }
147
148 pub fn set_related_files(&mut self, files: Vec<RelatedFile>, cx: &App) {
149 self.related_buffers = files
150 .into_iter()
151 .filter_map(|file| {
152 let project = self.project.upgrade()?;
153 let project = project.read(cx);
154 let worktree = project.worktrees(cx).find(|wt| {
155 let root_name = wt.read(cx).root_name().as_unix_str();
156 file.path
157 .components()
158 .next()
159 .is_some_and(|c| c.as_os_str() == root_name)
160 })?;
161 let worktree = worktree.read(cx);
162 let relative_path = file
163 .path
164 .strip_prefix(worktree.root_name().as_unix_str())
165 .ok()?;
166 let relative_path = RelPath::new(relative_path, PathStyle::Posix).ok()?;
167 let project_path = ProjectPath {
168 worktree_id: worktree.id(),
169 path: relative_path.into_owned().into(),
170 };
171 let buffer = project.get_open_buffer(&project_path, cx)?;
172 let snapshot = buffer.read(cx).snapshot();
173 let anchor_ranges = file
174 .excerpts
175 .iter()
176 .map(|excerpt| {
177 let start = snapshot.anchor_before(Point::new(excerpt.row_range.start, 0));
178 let end_row = excerpt.row_range.end;
179 let end_col = snapshot.line_len(end_row);
180 let end = snapshot.anchor_after(Point::new(end_row, end_col));
181 start..end
182 })
183 .collect();
184 Some(RelatedBuffer {
185 buffer,
186 path: file.path.clone(),
187 anchor_ranges,
188 cached_file: None,
189 })
190 })
191 .collect();
192 }
193
194 async fn fetch_excerpts(
195 this: WeakEntity<Self>,
196 buffer: Entity<Buffer>,
197 position: Anchor,
198 cx: &mut AsyncApp,
199 ) -> Result<()> {
200 let (project, snapshot, identifier_line_count) = this.read_with(cx, |this, cx| {
201 (
202 this.project.upgrade(),
203 buffer.read(cx).snapshot(),
204 this.identifier_line_count,
205 )
206 })?;
207 let Some(project) = project else {
208 return Ok(());
209 };
210
211 let file = snapshot.file().cloned();
212 if let Some(file) = &file {
213 log::debug!("retrieving_context buffer:{}", file.path().as_unix_str());
214 }
215
216 this.update(cx, |_, cx| {
217 cx.emit(RelatedExcerptStoreEvent::StartedRefresh);
218 })?;
219
220 let identifiers = cx
221 .background_spawn(async move {
222 identifiers_for_position(&snapshot, position, identifier_line_count)
223 })
224 .await;
225
226 let async_cx = cx.clone();
227 let start_time = Instant::now();
228 let futures = this.update(cx, |this, cx| {
229 identifiers
230 .into_iter()
231 .filter_map(|identifier| {
232 let task = if let Some(entry) = this.cache.get(&identifier) {
233 DefinitionTask::CacheHit(entry.clone())
234 } else {
235 DefinitionTask::CacheMiss(
236 this.project
237 .update(cx, |project, cx| {
238 project.definitions(&buffer, identifier.range.start, cx)
239 })
240 .ok()?,
241 )
242 };
243
244 let cx = async_cx.clone();
245 let project = project.clone();
246 Some(async move {
247 match task {
248 DefinitionTask::CacheHit(cache_entry) => {
249 Some((identifier, cache_entry, None))
250 }
251 DefinitionTask::CacheMiss(task) => {
252 let locations = task.await.log_err()??;
253 let duration = start_time.elapsed();
254 Some(cx.update(|cx| {
255 (
256 identifier,
257 Arc::new(CacheEntry {
258 definitions: locations
259 .into_iter()
260 .filter_map(|location| {
261 process_definition(location, &project, cx)
262 })
263 .collect(),
264 }),
265 Some(duration),
266 )
267 }))
268 }
269 }
270 })
271 })
272 .collect::<Vec<_>>()
273 })?;
274
275 let mut cache_hit_count = 0;
276 let mut cache_miss_count = 0;
277 let mut mean_definition_latency = Duration::ZERO;
278 let mut max_definition_latency = Duration::ZERO;
279 let mut new_cache = HashMap::default();
280 new_cache.reserve(futures.len());
281 for (identifier, entry, duration) in future::join_all(futures).await.into_iter().flatten() {
282 new_cache.insert(identifier, entry);
283 if let Some(duration) = duration {
284 cache_miss_count += 1;
285 mean_definition_latency += duration;
286 max_definition_latency = max_definition_latency.max(duration);
287 } else {
288 cache_hit_count += 1;
289 }
290 }
291 mean_definition_latency /= cache_miss_count.max(1) as u32;
292
293 let (new_cache, related_buffers) = rebuild_related_files(&project, new_cache, cx).await?;
294
295 if let Some(file) = &file {
296 log::debug!(
297 "finished retrieving context buffer:{}, latency:{:?}",
298 file.path().as_unix_str(),
299 start_time.elapsed()
300 );
301 }
302
303 this.update(cx, |this, cx| {
304 this.cache = new_cache;
305 this.related_buffers = related_buffers;
306 cx.emit(RelatedExcerptStoreEvent::FinishedRefresh {
307 cache_hit_count,
308 cache_miss_count,
309 mean_definition_latency,
310 max_definition_latency,
311 });
312 })?;
313
314 anyhow::Ok(())
315 }
316}
317
318async fn rebuild_related_files(
319 project: &Entity<Project>,
320 mut new_entries: HashMap<Identifier, Arc<CacheEntry>>,
321 cx: &mut AsyncApp,
322) -> Result<(HashMap<Identifier, Arc<CacheEntry>>, Vec<RelatedBuffer>)> {
323 let mut snapshots = HashMap::default();
324 let mut worktree_root_names = HashMap::default();
325 for entry in new_entries.values() {
326 for definition in &entry.definitions {
327 if let hash_map::Entry::Vacant(e) = snapshots.entry(definition.buffer.entity_id()) {
328 definition
329 .buffer
330 .read_with(cx, |buffer, _| buffer.parsing_idle())
331 .await;
332 e.insert(
333 definition
334 .buffer
335 .read_with(cx, |buffer, _| buffer.snapshot()),
336 );
337 }
338 let worktree_id = definition.path.worktree_id;
339 if let hash_map::Entry::Vacant(e) =
340 worktree_root_names.entry(definition.path.worktree_id)
341 {
342 project.read_with(cx, |project, cx| {
343 if let Some(worktree) = project.worktree_for_id(worktree_id, cx) {
344 e.insert(worktree.read(cx).root_name().as_unix_str().to_string());
345 }
346 });
347 }
348 }
349 }
350
351 Ok(cx
352 .background_spawn(async move {
353 let mut ranges_by_buffer =
354 HashMap::<EntityId, (Entity<Buffer>, Vec<Range<Point>>)>::default();
355 let mut paths_by_buffer = HashMap::default();
356 for entry in new_entries.values_mut() {
357 for definition in &entry.definitions {
358 let Some(snapshot) = snapshots.get(&definition.buffer.entity_id()) else {
359 continue;
360 };
361 paths_by_buffer.insert(definition.buffer.entity_id(), definition.path.clone());
362
363 ranges_by_buffer
364 .entry(definition.buffer.entity_id())
365 .or_insert_with(|| (definition.buffer.clone(), Vec::new()))
366 .1
367 .push(definition.anchor_range.to_point(snapshot));
368 }
369 }
370
371 let mut related_buffers: Vec<RelatedBuffer> = ranges_by_buffer
372 .into_iter()
373 .filter_map(|(entity_id, (buffer, ranges))| {
374 let snapshot = snapshots.get(&entity_id)?;
375 let project_path = paths_by_buffer.get(&entity_id)?;
376 let row_ranges = assemble_excerpt_ranges(snapshot, ranges);
377 let root_name = worktree_root_names.get(&project_path.worktree_id)?;
378
379 let path: Arc<Path> = Path::new(&format!(
380 "{}/{}",
381 root_name,
382 project_path.path.as_unix_str()
383 ))
384 .into();
385
386 let anchor_ranges = row_ranges
387 .into_iter()
388 .map(|row_range| {
389 let start = snapshot.anchor_before(Point::new(row_range.start, 0));
390 let end_col = snapshot.line_len(row_range.end);
391 let end = snapshot.anchor_after(Point::new(row_range.end, end_col));
392 start..end
393 })
394 .collect();
395
396 let mut related_buffer = RelatedBuffer {
397 buffer,
398 path,
399 anchor_ranges,
400 cached_file: None,
401 };
402 related_buffer.fill_cache(snapshot);
403 Some(related_buffer)
404 })
405 .collect();
406
407 related_buffers.sort_by_key(|related| related.path.clone());
408
409 (new_entries, related_buffers)
410 })
411 .await)
412}
413
414impl RelatedBuffer {
415 fn related_file(&mut self, cx: &App) -> RelatedFile {
416 let buffer = self.buffer.read(cx);
417 let path = self.path.clone();
418 let cached = if let Some(cached) = &self.cached_file
419 && buffer.version() == cached.buffer_version
420 {
421 cached
422 } else {
423 self.fill_cache(buffer)
424 };
425 let related_file = RelatedFile {
426 path,
427 excerpts: cached.excerpts.clone(),
428 max_row: buffer.max_point().row,
429 in_open_source_repo: false,
430 };
431 return related_file;
432 }
433
434 fn fill_cache(&mut self, buffer: &text::BufferSnapshot) -> &CachedRelatedFile {
435 let excerpts = self
436 .anchor_ranges
437 .iter()
438 .map(|range| {
439 let start = range.start.to_point(buffer);
440 let end = range.end.to_point(buffer);
441 RelatedExcerpt {
442 row_range: start.row..end.row,
443 text: buffer.text_for_range(start..end).collect::<String>().into(),
444 }
445 })
446 .collect::<Vec<_>>();
447 self.cached_file = Some(CachedRelatedFile {
448 excerpts: excerpts,
449 buffer_version: buffer.version().clone(),
450 });
451 self.cached_file.as_ref().unwrap()
452 }
453}
454
455use language::ToPoint as _;
456
457const MAX_TARGET_LEN: usize = 128;
458
459fn process_definition(
460 location: LocationLink,
461 project: &Entity<Project>,
462 cx: &mut App,
463) -> Option<CachedDefinition> {
464 let buffer = location.target.buffer.read(cx);
465 let anchor_range = location.target.range;
466 let file = buffer.file()?;
467 let worktree = project.read(cx).worktree_for_id(file.worktree_id(cx), cx)?;
468 if worktree.read(cx).is_single_file() {
469 return None;
470 }
471
472 // If the target range is large, it likely means we requested the definition of an entire module.
473 // For individual definitions, the target range should be small as it only covers the symbol.
474 let buffer = location.target.buffer.read(cx);
475 let target_len = anchor_range.to_offset(&buffer).len();
476 if target_len > MAX_TARGET_LEN {
477 return None;
478 }
479
480 Some(CachedDefinition {
481 path: ProjectPath {
482 worktree_id: file.worktree_id(cx),
483 path: file.path().clone(),
484 },
485 buffer: location.target.buffer,
486 anchor_range,
487 })
488}
489
490/// Gets all of the identifiers that are present in the given line, and its containing
491/// outline items.
492fn identifiers_for_position(
493 buffer: &BufferSnapshot,
494 position: Anchor,
495 identifier_line_count: u32,
496) -> Vec<Identifier> {
497 let offset = position.to_offset(buffer);
498 let point = buffer.offset_to_point(offset);
499
500 // Search for identifiers on lines adjacent to the cursor.
501 let start = Point::new(point.row.saturating_sub(identifier_line_count), 0);
502 let end = Point::new(point.row + identifier_line_count + 1, 0).min(buffer.max_point());
503 let line_range = start..end;
504 let mut ranges = vec![line_range.to_offset(&buffer)];
505
506 // Search for identifiers mentioned in headers/signatures of containing outline items.
507 let outline_items = buffer.outline_items_as_offsets_containing(offset..offset, false, None);
508 for item in outline_items {
509 if let Some(body_range) = item.body_range(&buffer) {
510 ranges.push(item.range.start..body_range.start.to_offset(&buffer));
511 } else {
512 ranges.push(item.range.clone());
513 }
514 }
515
516 ranges.sort_by(|a, b| a.start.cmp(&b.start).then(b.end.cmp(&a.end)));
517 ranges.dedup_by(|a, b| {
518 if a.start <= b.end {
519 b.start = b.start.min(a.start);
520 b.end = b.end.max(a.end);
521 true
522 } else {
523 false
524 }
525 });
526
527 let mut identifiers = Vec::new();
528 let outer_range =
529 ranges.first().map_or(0, |r| r.start)..ranges.last().map_or(buffer.len(), |r| r.end);
530
531 let mut captures = buffer
532 .syntax
533 .captures(outer_range.clone(), &buffer.text, |grammar| {
534 grammar
535 .highlights_config
536 .as_ref()
537 .map(|config| &config.query)
538 });
539
540 for range in ranges {
541 captures.set_byte_range(range.start..outer_range.end);
542
543 let mut last_range = None;
544 while let Some(capture) = captures.peek() {
545 let node_range = capture.node.byte_range();
546 if node_range.start > range.end {
547 break;
548 }
549 let config = captures.grammars()[capture.grammar_index]
550 .highlights_config
551 .as_ref();
552
553 if let Some(config) = config
554 && config.identifier_capture_indices.contains(&capture.index)
555 && range.contains_inclusive(&node_range)
556 && Some(&node_range) != last_range.as_ref()
557 {
558 let name = buffer.text_for_range(node_range.clone()).collect();
559 identifiers.push(Identifier {
560 range: buffer.anchor_after(node_range.start)
561 ..buffer.anchor_before(node_range.end),
562 name,
563 });
564 last_range = Some(node_range);
565 }
566
567 captures.advance();
568 }
569 }
570
571 identifiers
572}