context_store.rs

  1use std::path::{Path, PathBuf};
  2use std::sync::Arc;
  3
  4use anyhow::{anyhow, bail, Result};
  5use collections::{BTreeMap, HashMap, HashSet};
  6use futures::{self, future, Future, FutureExt};
  7use gpui::{AppContext, AsyncAppContext, Model, ModelContext, SharedString, Task, WeakView};
  8use language::Buffer;
  9use project::{ProjectPath, Worktree};
 10use rope::Rope;
 11use text::BufferId;
 12use workspace::Workspace;
 13
 14use crate::context::{
 15    Context, ContextBuffer, ContextId, ContextSnapshot, DirectoryContext, FetchedUrlContext,
 16    FileContext, ThreadContext,
 17};
 18use crate::thread::{Thread, ThreadId};
 19
 20pub struct ContextStore {
 21    workspace: WeakView<Workspace>,
 22    context: Vec<Context>,
 23    // TODO: If an EntityId is used for all context types (like BufferId), can remove ContextId.
 24    next_context_id: ContextId,
 25    files: BTreeMap<BufferId, ContextId>,
 26    directories: HashMap<PathBuf, ContextId>,
 27    threads: HashMap<ThreadId, ContextId>,
 28    fetched_urls: HashMap<String, ContextId>,
 29}
 30
 31impl ContextStore {
 32    pub fn new(workspace: WeakView<Workspace>) -> Self {
 33        Self {
 34            workspace,
 35            context: Vec::new(),
 36            next_context_id: ContextId(0),
 37            files: BTreeMap::default(),
 38            directories: HashMap::default(),
 39            threads: HashMap::default(),
 40            fetched_urls: HashMap::default(),
 41        }
 42    }
 43
 44    pub fn snapshot<'a>(
 45        &'a self,
 46        cx: &'a AppContext,
 47    ) -> impl Iterator<Item = ContextSnapshot> + 'a {
 48        self.context()
 49            .iter()
 50            .flat_map(|context| context.snapshot(cx))
 51    }
 52
 53    pub fn context(&self) -> &Vec<Context> {
 54        &self.context
 55    }
 56
 57    pub fn clear(&mut self) {
 58        self.context.clear();
 59        self.files.clear();
 60        self.directories.clear();
 61        self.threads.clear();
 62        self.fetched_urls.clear();
 63    }
 64
 65    pub fn add_file_from_path(
 66        &mut self,
 67        project_path: ProjectPath,
 68        cx: &mut ModelContext<Self>,
 69    ) -> Task<Result<()>> {
 70        let workspace = self.workspace.clone();
 71
 72        let Some(project) = workspace
 73            .upgrade()
 74            .map(|workspace| workspace.read(cx).project().clone())
 75        else {
 76            return Task::ready(Err(anyhow!("failed to read project")));
 77        };
 78
 79        cx.spawn(|this, mut cx| async move {
 80            let open_buffer_task = project.update(&mut cx, |project, cx| {
 81                project.open_buffer(project_path.clone(), cx)
 82            })?;
 83
 84            let buffer_model = open_buffer_task.await?;
 85            let buffer_id = this.update(&mut cx, |_, cx| buffer_model.read(cx).remote_id())?;
 86
 87            let already_included = this.update(&mut cx, |this, _cx| {
 88                match this.will_include_buffer(buffer_id, &project_path.path) {
 89                    Some(FileInclusion::Direct(context_id)) => {
 90                        this.remove_context(context_id);
 91                        true
 92                    }
 93                    Some(FileInclusion::InDirectory(_)) => true,
 94                    None => false,
 95                }
 96            })?;
 97
 98            if already_included {
 99                return anyhow::Ok(());
100            }
101
102            let (buffer_info, text_task) = this.update(&mut cx, |_, cx| {
103                let buffer = buffer_model.read(cx);
104                collect_buffer_info_and_text(
105                    project_path.path.clone(),
106                    buffer_model,
107                    buffer,
108                    cx.to_async(),
109                )
110            })?;
111
112            let text = text_task.await;
113
114            this.update(&mut cx, |this, _cx| {
115                this.insert_file(make_context_buffer(buffer_info, text));
116            })?;
117
118            anyhow::Ok(())
119        })
120    }
121
122    pub fn add_file_from_buffer(
123        &mut self,
124        buffer_model: Model<Buffer>,
125        cx: &mut ModelContext<Self>,
126    ) -> Task<Result<()>> {
127        cx.spawn(|this, mut cx| async move {
128            let (buffer_info, text_task) = this.update(&mut cx, |_, cx| {
129                let buffer = buffer_model.read(cx);
130                let Some(file) = buffer.file() else {
131                    return Err(anyhow!("Buffer has no path."));
132                };
133                Ok(collect_buffer_info_and_text(
134                    file.path().clone(),
135                    buffer_model,
136                    buffer,
137                    cx.to_async(),
138                ))
139            })??;
140
141            let text = text_task.await;
142
143            this.update(&mut cx, |this, _cx| {
144                this.insert_file(make_context_buffer(buffer_info, text))
145            })?;
146
147            anyhow::Ok(())
148        })
149    }
150
151    pub fn insert_file(&mut self, context_buffer: ContextBuffer) {
152        let id = self.next_context_id.post_inc();
153        self.files.insert(context_buffer.id, id);
154        self.context
155            .push(Context::File(FileContext { id, context_buffer }));
156    }
157
158    pub fn add_directory(
159        &mut self,
160        project_path: ProjectPath,
161        cx: &mut ModelContext<Self>,
162    ) -> Task<Result<()>> {
163        let workspace = self.workspace.clone();
164        let Some(project) = workspace
165            .upgrade()
166            .map(|workspace| workspace.read(cx).project().clone())
167        else {
168            return Task::ready(Err(anyhow!("failed to read project")));
169        };
170
171        let already_included = if let Some(context_id) = self.includes_directory(&project_path.path)
172        {
173            self.remove_context(context_id);
174            true
175        } else {
176            false
177        };
178        if already_included {
179            return Task::ready(Ok(()));
180        }
181
182        let worktree_id = project_path.worktree_id;
183        cx.spawn(|this, mut cx| async move {
184            let worktree = project.update(&mut cx, |project, cx| {
185                project
186                    .worktree_for_id(worktree_id, cx)
187                    .ok_or_else(|| anyhow!("no worktree found for {worktree_id:?}"))
188            })??;
189
190            let files = worktree.update(&mut cx, |worktree, _cx| {
191                collect_files_in_path(worktree, &project_path.path)
192            })?;
193
194            let open_buffers_task = project.update(&mut cx, |project, cx| {
195                let tasks = files.iter().map(|file_path| {
196                    project.open_buffer(
197                        ProjectPath {
198                            worktree_id,
199                            path: file_path.clone(),
200                        },
201                        cx,
202                    )
203                });
204                future::join_all(tasks)
205            })?;
206
207            let buffers = open_buffers_task.await;
208
209            let mut buffer_infos = Vec::new();
210            let mut text_tasks = Vec::new();
211            this.update(&mut cx, |_, cx| {
212                for (path, buffer_model) in files.into_iter().zip(buffers) {
213                    let buffer_model = buffer_model?;
214                    let buffer = buffer_model.read(cx);
215                    let (buffer_info, text_task) =
216                        collect_buffer_info_and_text(path, buffer_model, buffer, cx.to_async());
217                    buffer_infos.push(buffer_info);
218                    text_tasks.push(text_task);
219                }
220                anyhow::Ok(())
221            })??;
222
223            let buffer_texts = future::join_all(text_tasks).await;
224            let context_buffers = buffer_infos
225                .into_iter()
226                .zip(buffer_texts)
227                .map(|(info, text)| make_context_buffer(info, text))
228                .collect::<Vec<_>>();
229
230            if context_buffers.is_empty() {
231                bail!("No text files found in {}", &project_path.path.display());
232            }
233
234            this.update(&mut cx, |this, _| {
235                this.insert_directory(&project_path.path, context_buffers);
236            })?;
237
238            anyhow::Ok(())
239        })
240    }
241
242    pub fn insert_directory(&mut self, path: &Path, context_buffers: Vec<ContextBuffer>) {
243        let id = self.next_context_id.post_inc();
244        self.directories.insert(path.to_path_buf(), id);
245
246        self.context.push(Context::Directory(DirectoryContext::new(
247            id,
248            path,
249            context_buffers,
250        )));
251    }
252
253    pub fn add_thread(&mut self, thread: Model<Thread>, cx: &mut ModelContext<Self>) {
254        if let Some(context_id) = self.includes_thread(&thread.read(cx).id()) {
255            self.remove_context(context_id);
256        } else {
257            self.insert_thread(thread, cx);
258        }
259    }
260
261    pub fn insert_thread(&mut self, thread: Model<Thread>, cx: &AppContext) {
262        let id = self.next_context_id.post_inc();
263        let text = thread.read(cx).text().into();
264
265        self.threads.insert(thread.read(cx).id().clone(), id);
266        self.context
267            .push(Context::Thread(ThreadContext { id, thread, text }));
268    }
269
270    pub fn insert_fetched_url(&mut self, url: String, text: impl Into<SharedString>) {
271        let id = self.next_context_id.post_inc();
272
273        self.fetched_urls.insert(url.clone(), id);
274        self.context.push(Context::FetchedUrl(FetchedUrlContext {
275            id,
276            url: url.into(),
277            text: text.into(),
278        }));
279    }
280
281    pub fn remove_context(&mut self, id: ContextId) {
282        let Some(ix) = self.context.iter().position(|context| context.id() == id) else {
283            return;
284        };
285
286        match self.context.remove(ix) {
287            Context::File(_) => {
288                self.files.retain(|_, context_id| *context_id != id);
289            }
290            Context::Directory(_) => {
291                self.directories.retain(|_, context_id| *context_id != id);
292            }
293            Context::FetchedUrl(_) => {
294                self.fetched_urls.retain(|_, context_id| *context_id != id);
295            }
296            Context::Thread(_) => {
297                self.threads.retain(|_, context_id| *context_id != id);
298            }
299        }
300    }
301
302    /// Returns whether the buffer is already included directly in the context, or if it will be
303    /// included in the context via a directory. Directory inclusion is based on paths rather than
304    /// buffer IDs as the directory will be re-scanned.
305    pub fn will_include_buffer(&self, buffer_id: BufferId, path: &Path) -> Option<FileInclusion> {
306        if let Some(context_id) = self.files.get(&buffer_id) {
307            return Some(FileInclusion::Direct(*context_id));
308        }
309
310        self.will_include_file_path_via_directory(path)
311    }
312
313    /// Returns whether this file path is already included directly in the context, or if it will be
314    /// included in the context via a directory.
315    pub fn will_include_file_path(&self, path: &Path, cx: &AppContext) -> Option<FileInclusion> {
316        if !self.files.is_empty() {
317            let found_file_context = self.context.iter().find(|context| match &context {
318                Context::File(file_context) => {
319                    let buffer = file_context.context_buffer.buffer.read(cx);
320                    if let Some(file_path) = buffer_path_log_err(buffer) {
321                        *file_path == *path
322                    } else {
323                        false
324                    }
325                }
326                _ => false,
327            });
328            if let Some(context) = found_file_context {
329                return Some(FileInclusion::Direct(context.id()));
330            }
331        }
332
333        self.will_include_file_path_via_directory(path)
334    }
335
336    fn will_include_file_path_via_directory(&self, path: &Path) -> Option<FileInclusion> {
337        if self.directories.is_empty() {
338            return None;
339        }
340
341        let mut buf = path.to_path_buf();
342
343        while buf.pop() {
344            if let Some(_) = self.directories.get(&buf) {
345                return Some(FileInclusion::InDirectory(buf));
346            }
347        }
348
349        None
350    }
351
352    pub fn includes_directory(&self, path: &Path) -> Option<ContextId> {
353        self.directories.get(path).copied()
354    }
355
356    pub fn includes_thread(&self, thread_id: &ThreadId) -> Option<ContextId> {
357        self.threads.get(thread_id).copied()
358    }
359
360    pub fn includes_url(&self, url: &str) -> Option<ContextId> {
361        self.fetched_urls.get(url).copied()
362    }
363
364    /// Replaces the context that matches the ID of the new context, if any match.
365    fn replace_context(&mut self, new_context: Context) {
366        let id = new_context.id();
367        for context in self.context.iter_mut() {
368            if context.id() == id {
369                *context = new_context;
370                break;
371            }
372        }
373    }
374
375    pub fn file_paths(&self, cx: &AppContext) -> HashSet<PathBuf> {
376        self.context
377            .iter()
378            .filter_map(|context| match context {
379                Context::File(file) => {
380                    let buffer = file.context_buffer.buffer.read(cx);
381                    buffer_path_log_err(buffer).map(|p| p.to_path_buf())
382                }
383                Context::Directory(_) | Context::FetchedUrl(_) | Context::Thread(_) => None,
384            })
385            .collect()
386    }
387
388    pub fn thread_ids(&self) -> HashSet<ThreadId> {
389        self.threads.keys().cloned().collect()
390    }
391}
392
393pub enum FileInclusion {
394    Direct(ContextId),
395    InDirectory(PathBuf),
396}
397
398// ContextBuffer without text.
399struct BufferInfo {
400    buffer_model: Model<Buffer>,
401    id: BufferId,
402    version: clock::Global,
403}
404
405fn make_context_buffer(info: BufferInfo, text: SharedString) -> ContextBuffer {
406    ContextBuffer {
407        id: info.id,
408        buffer: info.buffer_model,
409        version: info.version,
410        text,
411    }
412}
413
414fn collect_buffer_info_and_text(
415    path: Arc<Path>,
416    buffer_model: Model<Buffer>,
417    buffer: &Buffer,
418    cx: AsyncAppContext,
419) -> (BufferInfo, Task<SharedString>) {
420    let buffer_info = BufferInfo {
421        id: buffer.remote_id(),
422        buffer_model,
423        version: buffer.version(),
424    };
425    // Important to collect version at the same time as content so that staleness logic is correct.
426    let content = buffer.as_rope().clone();
427    let text_task = cx
428        .background_executor()
429        .spawn(async move { to_fenced_codeblock(&path, content) });
430    (buffer_info, text_task)
431}
432
433pub fn buffer_path_log_err(buffer: &Buffer) -> Option<Arc<Path>> {
434    if let Some(file) = buffer.file() {
435        Some(file.path().clone())
436    } else {
437        log::error!("Buffer that had a path unexpectedly no longer has a path.");
438        None
439    }
440}
441
442fn to_fenced_codeblock(path: &Path, content: Rope) -> SharedString {
443    let path_extension = path.extension().and_then(|ext| ext.to_str());
444    let path_string = path.to_string_lossy();
445    let capacity = 3
446        + path_extension.map_or(0, |extension| extension.len() + 1)
447        + path_string.len()
448        + 1
449        + content.len()
450        + 5;
451    let mut buffer = String::with_capacity(capacity);
452
453    buffer.push_str("```");
454
455    if let Some(extension) = path_extension {
456        buffer.push_str(extension);
457        buffer.push(' ');
458    }
459    buffer.push_str(&path_string);
460
461    buffer.push('\n');
462    for chunk in content.chunks() {
463        buffer.push_str(&chunk);
464    }
465
466    if !buffer.ends_with('\n') {
467        buffer.push('\n');
468    }
469
470    buffer.push_str("```\n");
471
472    debug_assert!(
473        buffer.len() == capacity - 1 || buffer.len() == capacity,
474        "to_fenced_codeblock calculated capacity of {}, but length was {}",
475        capacity,
476        buffer.len(),
477    );
478
479    buffer.into()
480}
481
482fn collect_files_in_path(worktree: &Worktree, path: &Path) -> Vec<Arc<Path>> {
483    let mut files = Vec::new();
484
485    for entry in worktree.child_entries(path) {
486        if entry.is_dir() {
487            files.extend(collect_files_in_path(worktree, &entry.path));
488        } else if entry.is_file() {
489            files.push(entry.path.clone());
490        }
491    }
492
493    files
494}
495
496pub fn refresh_context_store_text(
497    context_store: Model<ContextStore>,
498    cx: &AppContext,
499) -> impl Future<Output = ()> {
500    let mut tasks = Vec::new();
501    for context in &context_store.read(cx).context {
502        match context {
503            Context::File(file_context) => {
504                let context_store = context_store.clone();
505                if let Some(task) = refresh_file_text(context_store, file_context, cx) {
506                    tasks.push(task);
507                }
508            }
509            Context::Directory(directory_context) => {
510                let context_store = context_store.clone();
511                if let Some(task) = refresh_directory_text(context_store, directory_context, cx) {
512                    tasks.push(task);
513                }
514            }
515            Context::Thread(thread_context) => {
516                let context_store = context_store.clone();
517                tasks.push(refresh_thread_text(context_store, thread_context, cx));
518            }
519            // Intentionally omit refreshing fetched URLs as it doesn't seem all that useful,
520            // and doing the caching properly could be tricky (unless it's already handled by
521            // the HttpClient?).
522            Context::FetchedUrl(_) => {}
523        }
524    }
525
526    future::join_all(tasks).map(|_| ())
527}
528
529fn refresh_file_text(
530    context_store: Model<ContextStore>,
531    file_context: &FileContext,
532    cx: &AppContext,
533) -> Option<Task<()>> {
534    let id = file_context.id;
535    let task = refresh_context_buffer(&file_context.context_buffer, cx);
536    if let Some(task) = task {
537        Some(cx.spawn(|mut cx| async move {
538            let context_buffer = task.await;
539            context_store
540                .update(&mut cx, |context_store, _| {
541                    let new_file_context = FileContext { id, context_buffer };
542                    context_store.replace_context(Context::File(new_file_context));
543                })
544                .ok();
545        }))
546    } else {
547        None
548    }
549}
550
551fn refresh_directory_text(
552    context_store: Model<ContextStore>,
553    directory_context: &DirectoryContext,
554    cx: &AppContext,
555) -> Option<Task<()>> {
556    let mut stale = false;
557    let futures = directory_context
558        .context_buffers
559        .iter()
560        .map(|context_buffer| {
561            if let Some(refresh_task) = refresh_context_buffer(context_buffer, cx) {
562                stale = true;
563                future::Either::Left(refresh_task)
564            } else {
565                future::Either::Right(future::ready((*context_buffer).clone()))
566            }
567        })
568        .collect::<Vec<_>>();
569
570    if !stale {
571        return None;
572    }
573
574    let context_buffers = future::join_all(futures);
575
576    let id = directory_context.snapshot.id;
577    let path = directory_context.path.clone();
578    Some(cx.spawn(|mut cx| async move {
579        let context_buffers = context_buffers.await;
580        context_store
581            .update(&mut cx, |context_store, _| {
582                let new_directory_context = DirectoryContext::new(id, &path, context_buffers);
583                context_store.replace_context(Context::Directory(new_directory_context));
584            })
585            .ok();
586    }))
587}
588
589fn refresh_thread_text(
590    context_store: Model<ContextStore>,
591    thread_context: &ThreadContext,
592    cx: &AppContext,
593) -> Task<()> {
594    let id = thread_context.id;
595    let thread = thread_context.thread.clone();
596    cx.spawn(move |mut cx| async move {
597        context_store
598            .update(&mut cx, |context_store, cx| {
599                let text = thread.read(cx).text().into();
600                context_store.replace_context(Context::Thread(ThreadContext { id, thread, text }));
601            })
602            .ok();
603    })
604}
605
606fn refresh_context_buffer(
607    context_buffer: &ContextBuffer,
608    cx: &AppContext,
609) -> Option<impl Future<Output = ContextBuffer>> {
610    let buffer = context_buffer.buffer.read(cx);
611    let path = buffer_path_log_err(buffer)?;
612    if buffer.version.changed_since(&context_buffer.version) {
613        let (buffer_info, text_task) = collect_buffer_info_and_text(
614            path,
615            context_buffer.buffer.clone(),
616            buffer,
617            cx.to_async(),
618        );
619        Some(text_task.map(move |text| make_context_buffer(buffer_info, text)))
620    } else {
621        None
622    }
623}