1use std::{
2 cell::LazyCell,
3 collections::BTreeSet,
4 io::{BufRead, BufReader},
5 ops::Range,
6 path::{Path, PathBuf},
7 pin::pin,
8 sync::Arc,
9};
10
11use anyhow::Context;
12use collections::HashSet;
13use fs::Fs;
14use futures::{SinkExt, StreamExt, select_biased, stream::FuturesOrdered};
15use gpui::{App, AppContext, AsyncApp, Entity, Task};
16use language::{Buffer, BufferSnapshot};
17use parking_lot::Mutex;
18use postage::oneshot;
19use rpc::{AnyProtoClient, proto};
20use smol::{
21 channel::{Receiver, Sender, bounded, unbounded},
22 future::FutureExt,
23};
24
25use text::BufferId;
26use util::{ResultExt, maybe, paths::compare_rel_paths};
27use worktree::{Entry, ProjectEntryId, Snapshot, Worktree, WorktreeSettings};
28
29use crate::{
30 Project, ProjectItem, ProjectPath, RemotelyCreatedModels,
31 buffer_store::BufferStore,
32 search::{SearchQuery, SearchResult},
33 worktree_store::WorktreeStore,
34};
35
36pub struct Search {
37 buffer_store: Entity<BufferStore>,
38 worktree_store: Entity<WorktreeStore>,
39 limit: usize,
40 kind: SearchKind,
41}
42
43/// Represents search setup, before it is actually kicked off with Search::into_results
44enum SearchKind {
45 /// Search for candidates by inspecting file contents on file system, avoiding loading the buffer unless we know that a given file contains a match.
46 Local {
47 fs: Arc<dyn Fs>,
48 worktrees: Vec<Entity<Worktree>>,
49 },
50 /// Query remote host for candidates. As of writing, the host runs a local search in "buffers with matches only" mode.
51 Remote {
52 client: AnyProtoClient,
53 remote_id: u64,
54 models: Arc<Mutex<RemotelyCreatedModels>>,
55 },
56 /// Run search against a known set of candidates. Even when working with a remote host, this won't round-trip to host.
57 OpenBuffersOnly,
58}
59
60/// Represents results of project search and allows one to either obtain match positions OR
61/// just the handles to buffers that may match the search. Grabbing the handles is cheaper than obtaining full match positions, because in that case we'll look for
62/// at most one match in each file.
63#[must_use]
64pub struct SearchResultsHandle {
65 results: Receiver<SearchResult>,
66 matching_buffers: Receiver<Entity<Buffer>>,
67 trigger_search: Box<dyn FnOnce(&mut App) -> Task<()> + Send + Sync>,
68}
69
70impl SearchResultsHandle {
71 pub fn results(self, cx: &mut App) -> Receiver<SearchResult> {
72 (self.trigger_search)(cx).detach();
73 self.results
74 }
75 pub fn matching_buffers(self, cx: &mut App) -> Receiver<Entity<Buffer>> {
76 (self.trigger_search)(cx).detach();
77 self.matching_buffers
78 }
79}
80
81#[derive(Clone)]
82enum FindSearchCandidates {
83 Local {
84 fs: Arc<dyn Fs>,
85 /// Start off with all paths in project and filter them based on:
86 /// - Include filters
87 /// - Exclude filters
88 /// - Only open buffers
89 /// - Scan ignored files
90 /// Put another way: filter out files that can't match (without looking at file contents)
91 input_paths_rx: Receiver<InputPath>,
92 /// After that, if the buffer is not yet loaded, we'll figure out if it contains at least one match
93 /// based on disk contents of a buffer. This step is not performed for buffers we already have in memory.
94 confirm_contents_will_match_tx: Sender<MatchingEntry>,
95 confirm_contents_will_match_rx: Receiver<MatchingEntry>,
96 /// Of those that contain at least one match (or are already in memory), look for rest of matches (and figure out their ranges).
97 /// But wait - first, we need to go back to the main thread to open a buffer (& create an entity for it).
98 get_buffer_for_full_scan_tx: Sender<ProjectPath>,
99 },
100 Remote,
101 OpenBuffersOnly,
102}
103
104impl Search {
105 pub fn local(
106 fs: Arc<dyn Fs>,
107 buffer_store: Entity<BufferStore>,
108 worktree_store: Entity<WorktreeStore>,
109 limit: usize,
110 cx: &mut App,
111 ) -> Self {
112 let worktrees = worktree_store.read(cx).visible_worktrees(cx).collect();
113 Self {
114 kind: SearchKind::Local { fs, worktrees },
115 buffer_store,
116 worktree_store,
117 limit,
118 }
119 }
120
121 pub(crate) fn remote(
122 buffer_store: Entity<BufferStore>,
123 worktree_store: Entity<WorktreeStore>,
124 limit: usize,
125 client_state: (AnyProtoClient, u64, Arc<Mutex<RemotelyCreatedModels>>),
126 ) -> Self {
127 Self {
128 kind: SearchKind::Remote {
129 client: client_state.0,
130 remote_id: client_state.1,
131 models: client_state.2,
132 },
133 buffer_store,
134 worktree_store,
135 limit,
136 }
137 }
138 pub(crate) fn open_buffers_only(
139 buffer_store: Entity<BufferStore>,
140 worktree_store: Entity<WorktreeStore>,
141 limit: usize,
142 ) -> Self {
143 Self {
144 kind: SearchKind::OpenBuffersOnly,
145 buffer_store,
146 worktree_store,
147 limit,
148 }
149 }
150
151 pub(crate) const MAX_SEARCH_RESULT_FILES: usize = 5_000;
152 pub(crate) const MAX_SEARCH_RESULT_RANGES: usize = 10_000;
153 /// Prepares a project search run. The resulting [`SearchResultsHandle`] has to be used to specify whether you're interested in matching buffers
154 /// or full search results.
155 pub fn into_handle(mut self, query: SearchQuery, cx: &mut App) -> SearchResultsHandle {
156 let mut open_buffers = HashSet::default();
157 let mut unnamed_buffers = Vec::new();
158 const MAX_CONCURRENT_BUFFER_OPENS: usize = 64;
159 let buffers = self.buffer_store.read(cx);
160 for handle in buffers.buffers() {
161 let buffer = handle.read(cx);
162 if !buffers.is_searchable(&buffer.remote_id()) {
163 continue;
164 } else if let Some(entry_id) = buffer.entry_id(cx) {
165 open_buffers.insert(entry_id);
166 } else {
167 self.limit = self.limit.saturating_sub(1);
168 unnamed_buffers.push(handle)
169 };
170 }
171 let executor = cx.background_executor().clone();
172 let (tx, rx) = unbounded();
173 let (grab_buffer_snapshot_tx, grab_buffer_snapshot_rx) = unbounded();
174 let matching_buffers = grab_buffer_snapshot_rx.clone();
175 let trigger_search = Box::new(move |cx: &mut App| {
176 cx.spawn(async move |cx| {
177 for buffer in unnamed_buffers {
178 _ = grab_buffer_snapshot_tx.send(buffer).await;
179 }
180
181 let (find_all_matches_tx, find_all_matches_rx) =
182 bounded(MAX_CONCURRENT_BUFFER_OPENS);
183 let query = Arc::new(query);
184 let (candidate_searcher, tasks) = match self.kind {
185 SearchKind::OpenBuffersOnly => {
186 let Ok(open_buffers) = cx.update(|cx| self.all_loaded_buffers(&query, cx))
187 else {
188 return;
189 };
190 let fill_requests = cx
191 .background_spawn(async move {
192 for buffer in open_buffers {
193 if let Err(_) = grab_buffer_snapshot_tx.send(buffer).await {
194 return;
195 }
196 }
197 })
198 .boxed_local();
199 (FindSearchCandidates::OpenBuffersOnly, vec![fill_requests])
200 }
201 SearchKind::Local {
202 fs,
203 ref mut worktrees,
204 } => {
205 let (get_buffer_for_full_scan_tx, get_buffer_for_full_scan_rx) =
206 unbounded();
207 let (confirm_contents_will_match_tx, confirm_contents_will_match_rx) =
208 bounded(64);
209 let (sorted_search_results_tx, sorted_search_results_rx) = unbounded();
210
211 let (input_paths_tx, input_paths_rx) = unbounded();
212 let tasks = vec![
213 cx.spawn(Self::provide_search_paths(
214 std::mem::take(worktrees),
215 query.clone(),
216 input_paths_tx,
217 sorted_search_results_tx,
218 ))
219 .boxed_local(),
220 Self::open_buffers(
221 &self.buffer_store,
222 get_buffer_for_full_scan_rx,
223 grab_buffer_snapshot_tx,
224 cx.clone(),
225 )
226 .boxed_local(),
227 cx.background_spawn(Self::maintain_sorted_search_results(
228 sorted_search_results_rx,
229 get_buffer_for_full_scan_tx.clone(),
230 self.limit,
231 ))
232 .boxed_local(),
233 ];
234 (
235 FindSearchCandidates::Local {
236 fs,
237 get_buffer_for_full_scan_tx,
238 confirm_contents_will_match_tx,
239 confirm_contents_will_match_rx,
240 input_paths_rx,
241 },
242 tasks,
243 )
244 }
245 SearchKind::Remote {
246 client,
247 remote_id,
248 models,
249 } => {
250 let request = client.request(proto::FindSearchCandidates {
251 project_id: remote_id,
252 query: Some(query.to_proto()),
253 limit: self.limit as _,
254 });
255 let Ok(guard) = cx.update(|cx| {
256 Project::retain_remotely_created_models_impl(
257 &models,
258 &self.buffer_store,
259 &self.worktree_store,
260 cx,
261 )
262 }) else {
263 return;
264 };
265 let buffer_store = self.buffer_store.downgrade();
266 let issue_remote_buffers_request = cx
267 .spawn(async move |cx| {
268 let _ = maybe!(async move {
269 let response = request.await?;
270 log::error!(
271 "Received {} match candidates for a project search",
272 response.buffer_ids.len()
273 );
274 for buffer_id in response.buffer_ids {
275 let buffer_id = BufferId::new(buffer_id)?;
276 let buffer = buffer_store
277 .update(cx, |buffer_store, cx| {
278 buffer_store.wait_for_remote_buffer(buffer_id, cx)
279 })?
280 .await?;
281 let _ = grab_buffer_snapshot_tx.send(buffer).await;
282 }
283
284 drop(guard);
285 anyhow::Ok(())
286 })
287 .await
288 .log_err();
289 })
290 .boxed_local();
291 (
292 FindSearchCandidates::Remote,
293 vec![issue_remote_buffers_request],
294 )
295 }
296 };
297
298 let should_find_all_matches = !tx.is_closed();
299
300 let worker_pool = executor.scoped(|scope| {
301 let num_cpus = executor.num_cpus();
302
303 assert!(num_cpus > 0);
304 for _ in 0..executor.num_cpus() - 1 {
305 let worker = Worker {
306 query: &query,
307 open_buffers: &open_buffers,
308 candidates: candidate_searcher.clone(),
309 find_all_matches_rx: find_all_matches_rx.clone(),
310 };
311 scope.spawn(worker.run());
312 }
313
314 drop(find_all_matches_rx);
315 drop(candidate_searcher);
316 });
317
318 let (sorted_matches_tx, sorted_matches_rx) = unbounded();
319 // The caller of `into_handle` decides whether they're interested in all matches (files that matched + all matching ranges) or
320 // just the files. *They are using the same stream as the guts of the project search do*.
321 // This means that we cannot grab values off of that stream unless it's strictly needed for making a progress in project search.
322 //
323 // Grabbing buffer snapshots is only necessary when we're looking for all matches. If the caller decided that they're not interested
324 // in all matches, running that task unconditionally would hinder caller's ability to observe all matching file paths.
325 let buffer_snapshots = if should_find_all_matches {
326 Some(
327 Self::grab_buffer_snapshots(
328 grab_buffer_snapshot_rx,
329 find_all_matches_tx,
330 sorted_matches_tx,
331 cx.clone(),
332 )
333 .boxed_local(),
334 )
335 } else {
336 drop(find_all_matches_tx);
337
338 None
339 };
340 let ensure_matches_are_reported_in_order = if should_find_all_matches {
341 Some(
342 Self::ensure_matched_ranges_are_reported_in_order(sorted_matches_rx, tx)
343 .boxed_local(),
344 )
345 } else {
346 drop(tx);
347 None
348 };
349
350 futures::future::join_all(
351 [worker_pool.boxed_local()]
352 .into_iter()
353 .chain(buffer_snapshots)
354 .chain(ensure_matches_are_reported_in_order)
355 .chain(tasks),
356 )
357 .await;
358 })
359 });
360
361 SearchResultsHandle {
362 results: rx,
363 matching_buffers,
364 trigger_search,
365 }
366 }
367
368 fn provide_search_paths(
369 worktrees: Vec<Entity<Worktree>>,
370 query: Arc<SearchQuery>,
371 tx: Sender<InputPath>,
372 results: Sender<oneshot::Receiver<ProjectPath>>,
373 ) -> impl AsyncFnOnce(&mut AsyncApp) {
374 async move |cx| {
375 _ = maybe!(async move {
376 let gitignored_tracker = PathInclusionMatcher::new(query.clone());
377 for worktree in worktrees {
378 let (mut snapshot, worktree_settings) = worktree
379 .read_with(cx, |this, _| {
380 Some((this.snapshot(), this.as_local()?.settings()))
381 })?
382 .context("The worktree is not local")?;
383 if query.include_ignored() {
384 // Pre-fetch all of the ignored directories as they're going to be searched.
385 let mut entries_to_refresh = vec![];
386
387 for entry in snapshot.entries(query.include_ignored(), 0) {
388 if gitignored_tracker.should_scan_gitignored_dir(
389 entry,
390 &snapshot,
391 &worktree_settings,
392 ) {
393 entries_to_refresh.push(entry.path.clone());
394 }
395 }
396 let barrier = worktree.update(cx, |this, _| {
397 let local = this.as_local_mut()?;
398 let barrier = entries_to_refresh
399 .into_iter()
400 .map(|path| local.add_path_prefix_to_scan(path).into_future())
401 .collect::<Vec<_>>();
402 Some(barrier)
403 })?;
404 if let Some(barriers) = barrier {
405 futures::future::join_all(barriers).await;
406 }
407 snapshot = worktree.read_with(cx, |this, _| this.snapshot())?;
408 }
409 cx.background_executor()
410 .scoped(|scope| {
411 scope.spawn(async {
412 for entry in snapshot.files(query.include_ignored(), 0) {
413 let (should_scan_tx, should_scan_rx) = oneshot::channel();
414
415 let Ok(_) = tx
416 .send(InputPath {
417 entry: entry.clone(),
418 snapshot: snapshot.clone(),
419 should_scan_tx,
420 })
421 .await
422 else {
423 return;
424 };
425 if results.send(should_scan_rx).await.is_err() {
426 return;
427 };
428 }
429 })
430 })
431 .await;
432 }
433 anyhow::Ok(())
434 })
435 .await;
436 }
437 }
438
439 async fn maintain_sorted_search_results(
440 rx: Receiver<oneshot::Receiver<ProjectPath>>,
441 paths_for_full_scan: Sender<ProjectPath>,
442 limit: usize,
443 ) {
444 let mut rx = pin!(rx);
445 let mut matched = 0;
446 while let Some(mut next_path_result) = rx.next().await {
447 let Some(successful_path) = next_path_result.next().await else {
448 // This file did not produce a match, hence skip it.
449 continue;
450 };
451 if paths_for_full_scan.send(successful_path).await.is_err() {
452 return;
453 };
454 matched += 1;
455 if matched >= limit {
456 break;
457 }
458 }
459 }
460
461 /// Background workers cannot open buffers by themselves, hence main thread will do it on their behalf.
462 async fn open_buffers(
463 buffer_store: &Entity<BufferStore>,
464 rx: Receiver<ProjectPath>,
465 find_all_matches_tx: Sender<Entity<Buffer>>,
466 mut cx: AsyncApp,
467 ) {
468 let mut rx = pin!(rx.ready_chunks(64));
469 _ = maybe!(async move {
470 while let Some(requested_paths) = rx.next().await {
471 let mut buffers = buffer_store.update(&mut cx, |this, cx| {
472 requested_paths
473 .into_iter()
474 .map(|path| this.open_buffer(path, cx))
475 .collect::<FuturesOrdered<_>>()
476 })?;
477
478 while let Some(buffer) = buffers.next().await {
479 if let Some(buffer) = buffer.log_err() {
480 find_all_matches_tx.send(buffer).await?;
481 }
482 }
483 }
484 Result::<_, anyhow::Error>::Ok(())
485 })
486 .await;
487 }
488
489 async fn grab_buffer_snapshots(
490 rx: Receiver<Entity<Buffer>>,
491 find_all_matches_tx: Sender<(
492 Entity<Buffer>,
493 BufferSnapshot,
494 oneshot::Sender<(Entity<Buffer>, Vec<Range<language::Anchor>>)>,
495 )>,
496 results: Sender<oneshot::Receiver<(Entity<Buffer>, Vec<Range<language::Anchor>>)>>,
497 mut cx: AsyncApp,
498 ) {
499 _ = maybe!(async move {
500 while let Ok(buffer) = rx.recv().await {
501 let snapshot = buffer.read_with(&mut cx, |this, _| this.snapshot())?;
502 let (tx, rx) = oneshot::channel();
503 find_all_matches_tx.send((buffer, snapshot, tx)).await?;
504 results.send(rx).await?;
505 }
506 debug_assert!(rx.is_empty());
507 Result::<_, anyhow::Error>::Ok(())
508 })
509 .await;
510 }
511
512 async fn ensure_matched_ranges_are_reported_in_order(
513 rx: Receiver<oneshot::Receiver<(Entity<Buffer>, Vec<Range<language::Anchor>>)>>,
514 tx: Sender<SearchResult>,
515 ) {
516 use postage::stream::Stream;
517 _ = maybe!(async move {
518 let mut matched_buffers = 0;
519 let mut matches = 0;
520 while let Ok(mut next_buffer_matches) = rx.recv().await {
521 let Some((buffer, ranges)) = next_buffer_matches.recv().await else {
522 continue;
523 };
524
525 if matched_buffers > Search::MAX_SEARCH_RESULT_FILES
526 || matches > Search::MAX_SEARCH_RESULT_RANGES
527 {
528 _ = tx.send(SearchResult::LimitReached).await;
529 break;
530 }
531 matched_buffers += 1;
532 matches += ranges.len();
533
534 _ = tx.send(SearchResult::Buffer { buffer, ranges }).await?;
535 }
536 anyhow::Ok(())
537 })
538 .await;
539 }
540
541 fn all_loaded_buffers(&self, search_query: &SearchQuery, cx: &App) -> Vec<Entity<Buffer>> {
542 let worktree_store = self.worktree_store.read(cx);
543 let mut buffers = search_query
544 .buffers()
545 .into_iter()
546 .flatten()
547 .filter(|buffer| {
548 let b = buffer.read(cx);
549 if let Some(file) = b.file() {
550 if !search_query.match_path(file.path().as_std_path()) {
551 return false;
552 }
553 if !search_query.include_ignored()
554 && let Some(entry) = b
555 .entry_id(cx)
556 .and_then(|entry_id| worktree_store.entry_for_id(entry_id, cx))
557 && entry.is_ignored
558 {
559 return false;
560 }
561 }
562 true
563 })
564 .cloned()
565 .collect::<Vec<_>>();
566 buffers.sort_by(|a, b| {
567 let a = a.read(cx);
568 let b = b.read(cx);
569 match (a.file(), b.file()) {
570 (None, None) => a.remote_id().cmp(&b.remote_id()),
571 (None, Some(_)) => std::cmp::Ordering::Less,
572 (Some(_), None) => std::cmp::Ordering::Greater,
573 (Some(a), Some(b)) => compare_rel_paths((a.path(), true), (b.path(), true)),
574 }
575 });
576
577 buffers
578 }
579}
580
581struct Worker<'search> {
582 query: &'search SearchQuery,
583 open_buffers: &'search HashSet<ProjectEntryId>,
584 candidates: FindSearchCandidates,
585 /// Ok, we're back in background: run full scan & find all matches in a given buffer snapshot.
586 /// Then, when you're done, share them via the channel you were given.
587 find_all_matches_rx: Receiver<(
588 Entity<Buffer>,
589 BufferSnapshot,
590 oneshot::Sender<(Entity<Buffer>, Vec<Range<language::Anchor>>)>,
591 )>,
592}
593
594impl Worker<'_> {
595 async fn run(self) {
596 let (
597 input_paths_rx,
598 confirm_contents_will_match_rx,
599 mut confirm_contents_will_match_tx,
600 mut get_buffer_for_full_scan_tx,
601 fs,
602 ) = match self.candidates {
603 FindSearchCandidates::Local {
604 fs,
605 input_paths_rx,
606 confirm_contents_will_match_rx,
607 confirm_contents_will_match_tx,
608 get_buffer_for_full_scan_tx,
609 } => (
610 input_paths_rx,
611 confirm_contents_will_match_rx,
612 confirm_contents_will_match_tx,
613 get_buffer_for_full_scan_tx,
614 Some(fs),
615 ),
616 FindSearchCandidates::Remote | FindSearchCandidates::OpenBuffersOnly => (
617 unbounded().1,
618 unbounded().1,
619 unbounded().0,
620 unbounded().0,
621 None,
622 ),
623 };
624 // WorkerA: grabs a request for "find all matches in file/a" <- takes 5 minutes
625 // right after: WorkerB: grabs a request for "find all matches in file/b" <- takes 5 seconds
626 let mut find_all_matches = pin!(self.find_all_matches_rx.fuse());
627 let mut find_first_match = pin!(confirm_contents_will_match_rx.fuse());
628 let mut scan_path = pin!(input_paths_rx.fuse());
629
630 loop {
631 let handler = RequestHandler {
632 query: self.query,
633 open_entries: &self.open_buffers,
634 fs: fs.as_deref(),
635 confirm_contents_will_match_tx: &confirm_contents_will_match_tx,
636 get_buffer_for_full_scan_tx: &get_buffer_for_full_scan_tx,
637 };
638 // Whenever we notice that some step of a pipeline is closed, we don't want to close subsequent
639 // steps straight away. Another worker might be about to produce a value that will
640 // be pushed there, thus we'll replace current worker's pipe with a dummy one.
641 // That way, we'll only ever close a next-stage channel when ALL workers do so.
642 select_biased! {
643 find_all_matches = find_all_matches.next() => {
644 let Some(matches) = find_all_matches else {
645 continue;
646 };
647 handler.handle_find_all_matches(matches).await;
648 },
649 find_first_match = find_first_match.next() => {
650 if let Some(buffer_with_at_least_one_match) = find_first_match {
651 handler.handle_find_first_match(buffer_with_at_least_one_match).await;
652 } else {
653 get_buffer_for_full_scan_tx = bounded(1).0;
654 }
655
656 },
657 scan_path = scan_path.next() => {
658 if let Some(path_to_scan) = scan_path {
659 handler.handle_scan_path(path_to_scan).await;
660 } else {
661 // If we're the last worker to notice that this is not producing values, close the upstream.
662 confirm_contents_will_match_tx = bounded(1).0;
663 }
664
665 }
666 complete => {
667 break
668 },
669
670 }
671 }
672 }
673}
674
675struct RequestHandler<'worker> {
676 query: &'worker SearchQuery,
677 fs: Option<&'worker dyn Fs>,
678 open_entries: &'worker HashSet<ProjectEntryId>,
679 confirm_contents_will_match_tx: &'worker Sender<MatchingEntry>,
680 get_buffer_for_full_scan_tx: &'worker Sender<ProjectPath>,
681}
682
683impl RequestHandler<'_> {
684 async fn handle_find_all_matches(
685 &self,
686 (buffer, snapshot, mut report_matches): (
687 Entity<Buffer>,
688 BufferSnapshot,
689 oneshot::Sender<(Entity<Buffer>, Vec<Range<language::Anchor>>)>,
690 ),
691 ) {
692 let ranges = self
693 .query
694 .search(&snapshot, None)
695 .await
696 .iter()
697 .map(|range| snapshot.anchor_before(range.start)..snapshot.anchor_after(range.end))
698 .collect::<Vec<_>>();
699
700 _ = report_matches.send((buffer, ranges)).await;
701 }
702
703 async fn handle_find_first_match(&self, mut entry: MatchingEntry) {
704 _=maybe!(async move {
705 let abs_path = entry.worktree_root.join(entry.path.path.as_std_path());
706 let Some(file) = self.fs.context("Trying to query filesystem in remote project search")?.open_sync(&abs_path).await.log_err() else {
707 return anyhow::Ok(());
708 };
709
710 let mut file = BufReader::new(file);
711 let file_start = file.fill_buf()?;
712
713 if let Err(Some(starting_position)) =
714 std::str::from_utf8(file_start).map_err(|e| e.error_len())
715 {
716 // Before attempting to match the file content, throw away files that have invalid UTF-8 sequences early on;
717 // That way we can still match files in a streaming fashion without having look at "obviously binary" files.
718 log::debug!(
719 "Invalid UTF-8 sequence in file {abs_path:?} at byte position {starting_position}"
720 );
721 return Ok(());
722 }
723
724 if self.query.detect(file).unwrap_or(false) {
725 // Yes, we should scan the whole file.
726 entry.should_scan_tx.send(entry.path).await?;
727 }
728 Ok(())
729 }).await;
730 }
731
732 async fn handle_scan_path(&self, req: InputPath) {
733 _ = maybe!(async move {
734 let InputPath {
735 entry,
736
737 snapshot,
738 should_scan_tx,
739 } = req;
740
741 if entry.is_fifo || !entry.is_file() {
742 return Ok(());
743 }
744
745 if self.query.filters_path() {
746 let matched_path = if self.query.match_full_paths() {
747 let mut full_path = snapshot.root_name().as_std_path().to_owned();
748 full_path.push(entry.path.as_std_path());
749 self.query.match_path(&full_path)
750 } else {
751 self.query.match_path(entry.path.as_std_path())
752 };
753 if !matched_path {
754 return Ok(());
755 }
756 }
757
758 if self.open_entries.contains(&entry.id) {
759 // The buffer is already in memory and that's the version we want to scan;
760 // hence skip the dilly-dally and look for all matches straight away.
761 self.get_buffer_for_full_scan_tx
762 .send(ProjectPath {
763 worktree_id: snapshot.id(),
764 path: entry.path.clone(),
765 })
766 .await?;
767 } else {
768 self.confirm_contents_will_match_tx
769 .send(MatchingEntry {
770 should_scan_tx: should_scan_tx,
771 worktree_root: snapshot.abs_path().clone(),
772 path: ProjectPath {
773 worktree_id: snapshot.id(),
774 path: entry.path.clone(),
775 },
776 })
777 .await?;
778 }
779
780 anyhow::Ok(())
781 })
782 .await;
783 }
784}
785
786struct InputPath {
787 entry: Entry,
788 snapshot: Snapshot,
789 should_scan_tx: oneshot::Sender<ProjectPath>,
790}
791
792struct MatchingEntry {
793 worktree_root: Arc<Path>,
794 path: ProjectPath,
795 should_scan_tx: oneshot::Sender<ProjectPath>,
796}
797
798/// This struct encapsulates the logic to decide whether a given gitignored directory should be
799/// scanned based on include/exclude patterns of a search query (as include/exclude parameters may match paths inside it).
800/// It is kind-of doing an inverse of glob. Given a glob pattern like `src/**/` and a parent path like `src`, we need to decide whether the parent
801/// may contain glob hits.
802struct PathInclusionMatcher {
803 included: BTreeSet<PathBuf>,
804 query: Arc<SearchQuery>,
805}
806
807impl PathInclusionMatcher {
808 fn new(query: Arc<SearchQuery>) -> Self {
809 let mut included = BTreeSet::new();
810 // To do an inverse glob match, we split each glob into it's prefix and the glob part.
811 // For example, `src/**/*.rs` becomes `src/` and `**/*.rs`. The glob part gets dropped.
812 // Then, when checking whether a given directory should be scanned, we check whether it is a non-empty substring of any glob prefix.
813 if query.filters_path() {
814 included.extend(
815 query
816 .files_to_include()
817 .sources()
818 .iter()
819 .flat_map(|glob| Some(wax::Glob::new(glob).ok()?.partition().0)),
820 );
821 }
822 Self { included, query }
823 }
824
825 fn should_scan_gitignored_dir(
826 &self,
827 entry: &Entry,
828 snapshot: &Snapshot,
829 worktree_settings: &WorktreeSettings,
830 ) -> bool {
831 if !entry.is_ignored || !entry.kind.is_unloaded() {
832 return false;
833 }
834 if !self.query.include_ignored() {
835 return false;
836 }
837 if worktree_settings.is_path_excluded(&entry.path) {
838 return false;
839 }
840 if !self.query.filters_path() {
841 return true;
842 }
843
844 let as_abs_path = LazyCell::new(move || snapshot.absolutize(&entry.path));
845 let entry_path = entry.path.as_std_path();
846 // 3. Check Exclusions (Pruning)
847 // If the current path is a child of an excluded path, we stop.
848 let is_excluded = self.path_is_definitely_excluded(entry_path, snapshot);
849
850 if is_excluded {
851 return false;
852 }
853
854 // 4. Check Inclusions (Traversal)
855 if self.included.is_empty() {
856 return true;
857 }
858
859 // We scan if the current path is a descendant of an include prefix
860 // OR if the current path is an ancestor of an include prefix (we need to go deeper to find it).
861 let is_included = self.included.iter().any(|prefix| {
862 let (prefix_matches_entry, entry_matches_prefix) = if prefix.is_absolute() {
863 (
864 prefix.starts_with(&**as_abs_path),
865 as_abs_path.starts_with(prefix),
866 )
867 } else {
868 (
869 prefix.starts_with(entry_path),
870 entry_path.starts_with(prefix),
871 )
872 };
873
874 // Logic:
875 // 1. entry_matches_prefix: We are inside the target zone (e.g. glob: src/, current: src/lib/). Keep scanning.
876 // 2. prefix_matches_entry: We are above the target zone (e.g. glob: src/foo/, current: src/). Keep scanning to reach foo.
877 prefix_matches_entry || entry_matches_prefix
878 });
879
880 is_included
881 }
882 fn path_is_definitely_excluded(&self, path: &Path, snapshot: &Snapshot) -> bool {
883 if !self.query.files_to_exclude().sources().is_empty() {
884 let mut path = if self.query.match_full_paths() {
885 let mut full_path = snapshot.root_name().as_std_path().to_owned();
886 full_path.push(path);
887 full_path
888 } else {
889 path.to_owned()
890 };
891 loop {
892 if self.query.files_to_exclude().is_match(&path) {
893 return true;
894 } else if !path.pop() {
895 return false;
896 }
897 }
898 } else {
899 false
900 }
901 }
902}
903
904#[cfg(test)]
905mod tests {
906 use super::*;
907 use fs::FakeFs;
908 use serde_json::json;
909 use settings::Settings;
910 use util::{
911 path,
912 paths::{PathMatcher, PathStyle},
913 rel_path::RelPath,
914 };
915 use worktree::{Entry, EntryKind, WorktreeSettings};
916
917 use crate::{
918 Project, project_search::PathInclusionMatcher, project_tests::init_test,
919 search::SearchQuery,
920 };
921
922 #[gpui::test]
923 async fn test_path_inclusion_matcher(cx: &mut gpui::TestAppContext) {
924 init_test(cx);
925
926 let fs = FakeFs::new(cx.background_executor.clone());
927 fs.insert_tree(
928 "/root",
929 json!({
930 ".gitignore": "src/data/\n",
931 "src": {
932 "data": {
933 "main.csv": "field_1,field_2,field_3",
934 },
935 "lib": {
936 "main.txt": "Are you familiar with fields?",
937 },
938 },
939 }),
940 )
941 .await;
942
943 let project = Project::test(fs.clone(), [path!("/root").as_ref()], cx).await;
944 let worktree = project.update(cx, |project, cx| project.worktrees(cx).next().unwrap());
945 let (worktree_settings, worktree_snapshot) = worktree.update(cx, |worktree, cx| {
946 let settings_location = worktree.settings_location(cx);
947 return (
948 WorktreeSettings::get(Some(settings_location), cx).clone(),
949 worktree.snapshot(),
950 );
951 });
952
953 // Manually create a test entry for the gitignored directory since it won't
954 // be loaded by the worktree
955 let entry = Entry {
956 id: ProjectEntryId::from_proto(1),
957 kind: EntryKind::UnloadedDir,
958 path: Arc::from(RelPath::unix(Path::new("src/data")).unwrap()),
959 inode: 0,
960 mtime: None,
961 canonical_path: None,
962 is_ignored: true,
963 is_hidden: false,
964 is_always_included: false,
965 is_external: false,
966 is_private: false,
967 size: 0,
968 char_bag: Default::default(),
969 is_fifo: false,
970 };
971
972 // 1. Test searching for `field`, including ignored files without any
973 // inclusion and exclusion filters.
974 let include_ignored = true;
975 let files_to_include = PathMatcher::default();
976 let files_to_exclude = PathMatcher::default();
977 let match_full_paths = false;
978 let search_query = SearchQuery::text(
979 "field",
980 false,
981 false,
982 include_ignored,
983 files_to_include,
984 files_to_exclude,
985 match_full_paths,
986 None,
987 )
988 .unwrap();
989
990 let path_matcher = PathInclusionMatcher::new(Arc::new(search_query));
991 assert!(path_matcher.should_scan_gitignored_dir(
992 &entry,
993 &worktree_snapshot,
994 &worktree_settings
995 ));
996
997 // 2. Test searching for `field`, including ignored files but updating
998 // `files_to_include` to only include files under `src/lib`.
999 let include_ignored = true;
1000 let files_to_include = PathMatcher::new(vec!["src/lib"], PathStyle::Posix).unwrap();
1001 let files_to_exclude = PathMatcher::default();
1002 let match_full_paths = false;
1003 let search_query = SearchQuery::text(
1004 "field",
1005 false,
1006 false,
1007 include_ignored,
1008 files_to_include,
1009 files_to_exclude,
1010 match_full_paths,
1011 None,
1012 )
1013 .unwrap();
1014
1015 let path_matcher = PathInclusionMatcher::new(Arc::new(search_query));
1016 assert!(!path_matcher.should_scan_gitignored_dir(
1017 &entry,
1018 &worktree_snapshot,
1019 &worktree_settings
1020 ));
1021 }
1022}