1use crate::{
2 example::{Example, ExampleState},
3 git,
4 headless::EpAppState,
5 progress::{ExampleProgress, InfoStyle, Step, StepProgress},
6};
7use anyhow::{Context as _, Result};
8use edit_prediction::{
9 EditPredictionStore,
10 cursor_excerpt::compute_excerpt_ranges,
11 udiff::{OpenedBuffers, refresh_worktree_entries, strip_diff_path_prefix},
12};
13use futures::AsyncWriteExt as _;
14use gpui::{AsyncApp, Entity};
15use language::{Anchor, Buffer, LanguageNotFound, ToOffset, ToPoint};
16use project::{Project, ProjectPath, buffer_store::BufferStoreEvent};
17use std::{fs, path::PathBuf, sync::Arc};
18use zeta_prompt::ZetaPromptInput;
19
20pub async fn run_load_project(
21 example: &mut Example,
22 app_state: Arc<EpAppState>,
23 example_progress: &ExampleProgress,
24 mut cx: AsyncApp,
25) -> Result<()> {
26 if example.state.is_some() {
27 return Ok(());
28 }
29
30 let progress = example_progress.start(Step::LoadProject);
31
32 let project = setup_project(example, &app_state, &progress, &mut cx).await?;
33
34 progress.set_substatus("applying edit history");
35 let open_buffers = apply_edit_history(example, &project, &mut cx).await?;
36
37 let ep_store = cx
38 .update(|cx| EditPredictionStore::try_global(cx))
39 .context("EditPredictionStore not initialized")?;
40
41 let recent_paths: Vec<ProjectPath> = open_buffers
42 .buffers()
43 .filter_map(|buffer| {
44 buffer.read_with(&cx, |buffer, cx| {
45 buffer
46 .file()
47 .map(|file| ProjectPath::from_file(file.as_ref(), cx))
48 })
49 })
50 .collect();
51
52 ep_store.update(&mut cx, |store, cx| {
53 store.set_recent_paths_for_project(&project, recent_paths, cx);
54 });
55
56 progress.set_substatus("resolving cursor");
57 let (buffer, cursor_position) =
58 cursor_position(example, &project, &open_buffers, &mut cx).await?;
59 buffer
60 .read_with(&cx, |buffer, _| buffer.parsing_idle())
61 .await;
62
63 let events: Vec<Arc<zeta_prompt::Event>> = ep_store.update(&mut cx, |store, cx| {
64 store
65 .edit_history_for_project(&project, cx)
66 .into_iter()
67 .map(|e| e.event)
68 .collect()
69 });
70
71 let existing_related_files = example
72 .prompt_inputs
73 .take()
74 .map(|inputs| inputs.related_files)
75 .unwrap_or_default();
76
77 let (prompt_inputs, language_name) = buffer.read_with(&cx, |buffer, _cx| {
78 let snapshot = buffer.snapshot();
79 let cursor_point = cursor_position.to_point(&snapshot);
80 let cursor_offset = cursor_position.to_offset(&snapshot);
81 let language_name = buffer
82 .language()
83 .map(|l| l.name().to_string())
84 .unwrap_or_else(|| "Unknown".to_string());
85
86 let (full_context_point_range, full_context_offset_range, excerpt_ranges) =
87 compute_excerpt_ranges(cursor_point, &snapshot);
88
89 let cursor_excerpt: Arc<str> = buffer
90 .text_for_range(full_context_offset_range.clone())
91 .collect::<String>()
92 .into();
93 let cursor_offset_in_excerpt = cursor_offset - full_context_offset_range.start;
94 let excerpt_start_row = Some(full_context_point_range.start.row);
95
96 (
97 ZetaPromptInput {
98 cursor_path: example.spec.cursor_path.clone(),
99 cursor_excerpt,
100 cursor_offset_in_excerpt,
101 excerpt_start_row,
102 events,
103 related_files: existing_related_files,
104 excerpt_ranges,
105 in_open_source_repo: false,
106 can_collect_data: false,
107 experiment: None,
108 repo_url: None,
109 },
110 language_name,
111 )
112 });
113
114 progress.set_info(language_name, InfoStyle::Normal);
115
116 example.prompt_inputs = Some(prompt_inputs);
117 example.state = Some(ExampleState {
118 buffer,
119 project,
120 cursor_position,
121 _open_buffers: open_buffers,
122 });
123 Ok(())
124}
125
126async fn cursor_position(
127 example: &Example,
128 project: &Entity<Project>,
129 open_buffers: &OpenedBuffers,
130 cx: &mut AsyncApp,
131) -> Result<(Entity<Buffer>, Anchor)> {
132 let language_registry = project.read_with(cx, |project, _| project.languages().clone());
133 let result = language_registry
134 .load_language_for_file_path(&example.spec.cursor_path)
135 .await;
136
137 if let Err(error) = result
138 && !error.is::<LanguageNotFound>()
139 {
140 return Err(error);
141 }
142
143 let cursor_path_str = example.spec.cursor_path.to_string_lossy();
144 // Also try cursor path with first component stripped - old examples may have
145 // paths like "zed/crates/foo.rs" instead of "crates/foo.rs".
146 let cursor_path_without_prefix: PathBuf =
147 example.spec.cursor_path.components().skip(1).collect();
148 let cursor_path_without_prefix_str = cursor_path_without_prefix.to_string_lossy();
149
150 // We try open_buffers first because the file might be new and not saved to disk
151 let cursor_buffer = if let Some(buffer) = open_buffers.get(cursor_path_str.as_ref()) {
152 buffer.clone()
153 } else if let Some(buffer) = open_buffers.get(cursor_path_without_prefix_str.as_ref()) {
154 buffer.clone()
155 } else {
156 // Since the worktree scanner is disabled, manually refresh entries for the cursor path.
157 if let Some(worktree) = project.read_with(cx, |project, cx| project.worktrees(cx).next()) {
158 refresh_worktree_entries(&worktree, [&*example.spec.cursor_path], cx).await?;
159 }
160
161 let cursor_path = project
162 .read_with(cx, |project, cx| {
163 project
164 .find_project_path(&example.spec.cursor_path, cx)
165 .or_else(|| project.find_project_path(&cursor_path_without_prefix, cx))
166 })
167 .with_context(|| {
168 format!(
169 "failed to find cursor path {}",
170 example.spec.cursor_path.display()
171 )
172 })?;
173
174 project
175 .update(cx, |project, cx| project.open_buffer(cursor_path, cx))
176 .await?
177 };
178
179 let (cursor_excerpt, cursor_offset_within_excerpt) = example.spec.cursor_excerpt()?;
180
181 let excerpt_offset = cursor_buffer.read_with(&*cx, |buffer, _cx| {
182 let text = buffer.text();
183
184 let mut matches = text.match_indices(&cursor_excerpt);
185 let (excerpt_offset, _) = matches.next().with_context(|| {
186 format!("Cursor excerpt did not exist in buffer:\n\n{cursor_excerpt}\n",)
187 })?;
188 anyhow::ensure!(
189 matches.next().is_none(),
190 "More than one cursor position match found",
191 );
192 Ok(excerpt_offset)
193 })?;
194
195 let cursor_offset = excerpt_offset + cursor_offset_within_excerpt;
196 let cursor_anchor =
197 cursor_buffer.read_with(&*cx, |buffer, _| buffer.anchor_after(cursor_offset));
198
199 Ok((cursor_buffer, cursor_anchor))
200}
201
202async fn setup_project(
203 example: &mut Example,
204 app_state: &Arc<EpAppState>,
205 step_progress: &StepProgress,
206 cx: &mut AsyncApp,
207) -> Result<Entity<Project>> {
208 let ep_store = cx
209 .update(|cx| EditPredictionStore::try_global(cx))
210 .context("Store should be initialized at init")?;
211
212 let worktree_path = setup_worktree(example, step_progress).await?;
213
214 let project = cx.update(|cx| {
215 Project::local(
216 app_state.client.clone(),
217 app_state.node_runtime.clone(),
218 app_state.user_store.clone(),
219 app_state.languages.clone(),
220 app_state.fs.clone(),
221 None,
222 project::LocalProjectFlags {
223 init_worktree_trust: false,
224 watch_global_configs: false,
225 },
226 cx,
227 )
228 });
229
230 project
231 .update(cx, |project, cx| {
232 project.disable_worktree_scanner(cx);
233 project.create_worktree(&worktree_path, true, cx)
234 })
235 .await?;
236
237 let buffer_store = project.read_with(cx, |project, _| project.buffer_store().clone());
238 cx.subscribe(&buffer_store, {
239 let project = project.downgrade();
240 let ep_store = ep_store.downgrade();
241 move |_, event, cx| match event {
242 BufferStoreEvent::BufferAdded(buffer) => {
243 let Some(project) = project.upgrade() else {
244 return;
245 };
246 ep_store
247 .update(cx, |store, cx| store.register_buffer(&buffer, &project, cx))
248 .ok();
249 }
250 _ => {}
251 }
252 })
253 .detach();
254
255 Ok(project)
256}
257
258async fn setup_worktree(example: &Example, step_progress: &StepProgress) -> Result<PathBuf> {
259 let repo_name = example.repo_name().context("failed to get repo name")?;
260 let repo_dir = git::repo_path_for_url(&example.spec.repository_url)?;
261 let worktree_path = repo_name.worktree_path();
262 let repo_lock = git::lock_repo(&repo_dir).await;
263
264 // Clean up any stale git lock files from previous crashed runs.
265 // Safe-ish since we have our own lock.
266 // WARNING: Can corrupt worktrees if multiple processes of the CLI are running.
267 let worktree_git_dir = repo_dir
268 .join(".git/worktrees")
269 .join(repo_name.name.as_ref());
270 for lock_file in &["index.lock", "HEAD.lock", "config.lock"] {
271 let worktree_lock_path = worktree_git_dir.join(lock_file);
272 let repo_lock_path = repo_dir.join(".git").join(lock_file);
273 if worktree_lock_path.exists() {
274 fs::remove_file(&worktree_lock_path).ok();
275 }
276 if repo_lock_path.exists() {
277 fs::remove_file(&repo_lock_path).ok();
278 }
279 }
280
281 let mut git_repo_exists = false;
282 if repo_dir.is_dir() {
283 if git::run_git(&repo_dir, &["remote", "get-url", "origin"])
284 .await
285 .map_or(false, |origin| origin.trim() == example.spec.repository_url)
286 {
287 git_repo_exists = true;
288 } else {
289 fs::remove_dir_all(&repo_dir).ok();
290 }
291 }
292
293 if !git_repo_exists {
294 step_progress.set_substatus(format!("cloning {}", repo_name.name));
295 fs::create_dir_all(&repo_dir)?;
296 git::run_git(&repo_dir, &["init"]).await?;
297 git::run_git(
298 &repo_dir,
299 &["remote", "add", "origin", &example.spec.repository_url],
300 )
301 .await?;
302 }
303
304 // Resolve the example to a revision, fetching it if needed.
305 step_progress.set_substatus("fetching");
306 let revision = git::fetch_if_needed(&repo_dir, &example.spec.revision).await?;
307
308 // Clean up any stale worktree registrations from previous crashed runs.
309 git::run_git(&repo_dir, &["worktree", "prune"]).await.ok();
310
311 // Create the worktree for this example if needed.
312 step_progress.set_substatus("preparing worktree");
313
314 // Check if worktree exists and is valid (not just a directory from a crashed run).
315 let worktree_valid = worktree_path.is_dir()
316 && git::run_git(&worktree_path, &["rev-parse", "--git-dir"])
317 .await
318 .is_ok();
319
320 if worktree_valid {
321 git::run_git(&worktree_path, &["clean", "--force", "-d"]).await?;
322 git::run_git(&worktree_path, &["reset", "--hard", "HEAD"]).await?;
323 git::run_git(&worktree_path, &["checkout", revision.as_str()]).await?;
324 } else {
325 let worktree_path_string = worktree_path.to_string_lossy();
326
327 // Clean up invalid worktree directory and registration if they exist.
328 if worktree_path.exists() {
329 fs::remove_dir_all(&worktree_path).ok();
330 }
331 git::run_git(
332 &repo_dir,
333 &["worktree", "remove", "--force", &worktree_path_string],
334 )
335 .await
336 .ok();
337
338 let branch_name = example.spec.filename();
339 git::run_git(
340 &repo_dir,
341 &["branch", "-f", &branch_name, revision.as_str()],
342 )
343 .await?;
344 git::run_git(
345 &repo_dir,
346 &["worktree", "add", "-f", &worktree_path_string, &branch_name],
347 )
348 .await?;
349 }
350 drop(repo_lock);
351
352 if !example.spec.uncommitted_diff.is_empty() {
353 step_progress.set_substatus("applying diff");
354
355 // old examples had full paths in the uncommitted diff.
356 let uncommitted_diff =
357 strip_diff_path_prefix(&example.spec.uncommitted_diff, &repo_name.name);
358
359 let mut apply_process = smol::process::Command::new("git")
360 .current_dir(&worktree_path)
361 .args(&["apply", "-"])
362 .stdin(std::process::Stdio::piped())
363 .spawn()?;
364
365 let mut stdin = apply_process.stdin.take().context("Failed to get stdin")?;
366 stdin.write_all(uncommitted_diff.as_bytes()).await?;
367 stdin.close().await?;
368 drop(stdin);
369
370 let apply_result = apply_process.output().await?;
371 anyhow::ensure!(
372 apply_result.status.success(),
373 "Failed to apply uncommitted diff patch with status: {}\nstderr:\n{}\nstdout:\n{}",
374 apply_result.status,
375 String::from_utf8_lossy(&apply_result.stderr),
376 String::from_utf8_lossy(&apply_result.stdout),
377 );
378 }
379
380 step_progress.clear_substatus();
381 Ok(worktree_path)
382}
383
384async fn apply_edit_history(
385 example: &Example,
386 project: &Entity<Project>,
387 cx: &mut AsyncApp,
388) -> Result<OpenedBuffers> {
389 edit_prediction::udiff::apply_diff(&example.spec.edit_history, project, cx).await
390}