1use crate::{AgentAppState, ToolMetrics};
2use agent::{ThreadEvent, ThreadStore};
3use anyhow::{Context as _, Result, anyhow};
4use assistant_tool::ToolWorkingSet;
5use client::proto::LspWorkProgress;
6use futures::channel::mpsc;
7use futures::{FutureExt, StreamExt as _, select_biased};
8use gpui::{App, AppContext as _, AsyncApp, Entity, Task};
9use handlebars::Handlebars;
10use language::{Buffer, DiagnosticSeverity, OffsetRangeExt};
11use language_model::{
12 LanguageModel, LanguageModelCompletionEvent, LanguageModelRequest, LanguageModelRequestMessage,
13 MessageContent, Role, StopReason, TokenUsage,
14};
15use project::{Project, ProjectPath};
16use serde::{Deserialize, Serialize};
17use std::cell::RefCell;
18use std::fmt::Write as _;
19use std::fs::File;
20use std::io::Write as _;
21use std::rc::Rc;
22use std::sync::{Arc, Mutex};
23use std::time::Duration;
24use std::{
25 fs,
26 path::{Path, PathBuf},
27};
28use unindent::Unindent as _;
29use util::ResultExt as _;
30use util::command::new_smol_command;
31use util::markdown::MarkdownString;
32use util::serde::default_true;
33
34const THREAD_EVENT_TIMEOUT: Duration = Duration::from_secs(60 * 2);
35
36const ZED_REPO_URL: &str = "https://github.com/zed-industries/zed.git";
37
38#[derive(Clone, Debug, Deserialize)]
39pub struct ExampleBase {
40 pub url: String,
41 pub revision: String,
42 pub language_extension: Option<String>,
43 pub insert_id: Option<String>,
44 #[serde(default = "default_true")]
45 pub require_lsp: bool,
46 #[serde(default)]
47 pub allow_preexisting_diagnostics: bool,
48}
49
50impl ExampleBase {
51 pub fn repo_name(&self) -> String {
52 self.url
53 .split('/')
54 .next_back()
55 .unwrap_or(&"")
56 .trim_end_matches(".git")
57 .into()
58 }
59}
60
61#[derive(Clone, Debug)]
62pub struct Example {
63 pub name: String,
64 /// Content of `base.toml`
65 pub base: ExampleBase,
66 /// Content of `prompt.md`
67 pub prompt: String,
68 /// Content of `diff_criteria.md`
69 pub diff_criteria: String,
70 /// Content of `thread_criteria.md`, if that file exists (it's optional)
71 pub thread_criteria: Option<String>,
72 /// Path to the directory containing the requests and responses for the agentic loop
73 pub run_directory_path: PathBuf,
74 /// Prefix used for logging that identifies this example
75 pub log_prefix: String,
76 pub worktree_path: PathBuf,
77 pub repo_path: PathBuf,
78}
79
80#[derive(Debug, Serialize, Deserialize, Clone)]
81pub struct RunOutput {
82 pub repository_diff: String,
83 pub ran_diagnostics_check: bool,
84 pub diagnostics_before: Option<String>,
85 pub diagnostics_after: Option<String>,
86 pub response_count: usize,
87 pub token_usage: TokenUsage,
88 pub tool_metrics: ToolMetrics,
89 pub last_request: LanguageModelRequest,
90}
91
92#[derive(Debug, Clone, Serialize, Deserialize)]
93pub struct JudgeDiffInput {
94 pub repository_diff: String,
95 pub ran_diagnostics_check: bool,
96 #[serde(skip_serializing_if = "Option::is_none")]
97 pub diagnostics_before: Option<String>,
98 #[serde(skip_serializing_if = "Option::is_none")]
99 pub diagnostics_after: Option<String>,
100 pub criteria: String,
101}
102
103#[derive(Debug, Clone, Serialize, Deserialize)]
104pub struct JudgeThreadInput {
105 pub messages: String,
106 pub criteria: String,
107}
108
109#[derive(Debug, Clone, Serialize, Deserialize)]
110pub struct JudgeResponse {
111 pub analysis: String,
112 pub score: u32,
113}
114
115#[derive(Debug, Clone, Serialize, Deserialize)]
116pub struct JudgeOutput {
117 pub thread: Option<JudgeResponse>,
118 pub diff: JudgeResponse,
119}
120
121impl Example {
122 /// Load an example from a directory containing base.toml, prompt.md, and criteria.md
123 pub fn load_from_directory(
124 dir_path: &Path,
125 run_dir: &Path,
126 worktrees_dir: &Path,
127 repos_dir: &Path,
128 ) -> Result<Self> {
129 let name = Self::name_from_path(dir_path);
130 let base_path = dir_path.join("base.toml");
131 let prompt_path = dir_path.join("prompt.md");
132 let diff_criteria_path = dir_path.join("diff_criteria.md");
133 let thread_criteria_path = dir_path.join("thread_criteria.md");
134 let thread_criteria = if thread_criteria_path.exists() {
135 Some(fs::read_to_string(thread_criteria_path.clone())?)
136 } else {
137 None
138 };
139
140 let base: ExampleBase = toml::from_str(&fs::read_to_string(&base_path)?)?;
141
142 let repo_path = repo_path_for_url(repos_dir, &base.url);
143
144 let worktree_path = worktrees_dir
145 .canonicalize()
146 .unwrap()
147 .join(&name)
148 .join(&base.repo_name());
149
150 Ok(Example {
151 name: name.clone(),
152 base,
153 prompt: fs::read_to_string(prompt_path.clone())?,
154 thread_criteria,
155 diff_criteria: fs::read_to_string(diff_criteria_path.clone())?,
156 run_directory_path: run_dir.to_path_buf(),
157 worktree_path,
158 repo_path,
159 log_prefix: name,
160 })
161 }
162
163 pub fn set_repetition_number(&mut self, repetition_number: u32) {
164 if repetition_number > 0 {
165 self.name = format!("{}-{}", self.name, repetition_number);
166 }
167 }
168
169 pub fn example_output_directory(&self) -> PathBuf {
170 self.run_directory_path.join(&self.name)
171 }
172
173 pub fn set_log_prefix_style(&mut self, color: &str, name_width: usize) {
174 self.log_prefix = format!(
175 "{}{:<width$}\x1b[0m | ",
176 color,
177 self.name,
178 width = name_width
179 );
180 }
181
182 pub fn name_from_path(path: &Path) -> String {
183 path.file_name().unwrap().to_string_lossy().to_string()
184 }
185
186 /// Set up the example by checking out the specified Git revision
187 pub async fn setup(&mut self) -> Result<()> {
188 let revision_exists = run_git(
189 &self.repo_path,
190 &["rev-parse", &format!("{}^{{commit}}", self.base.revision)],
191 )
192 .await
193 .is_ok();
194
195 if !revision_exists {
196 println!(
197 "{}Fetching revision {}",
198 self.log_prefix, &self.base.revision
199 );
200 run_git(
201 &self.repo_path,
202 &["fetch", "--depth", "1", "origin", &self.base.revision],
203 )
204 .await?;
205 }
206
207 if self.worktree_path.is_dir() {
208 println!("{}Resetting existing worktree", self.log_prefix);
209
210 // TODO: consider including "-x" to remove ignored files. The downside of this is that
211 // it will also remove build artifacts, and so prevent incremental reuse there.
212 run_git(&self.worktree_path, &["clean", "--force", "-d"]).await?;
213 run_git(&self.worktree_path, &["reset", "--hard", "HEAD"]).await?;
214 run_git(&self.worktree_path, &["checkout", &self.base.revision]).await?;
215 } else {
216 println!("{}Creating worktree", self.log_prefix);
217
218 let worktree_path_string = self.worktree_path.to_string_lossy().to_string();
219
220 run_git(
221 &self.repo_path,
222 &[
223 "worktree",
224 "add",
225 "-f",
226 &worktree_path_string,
227 &self.base.revision,
228 ],
229 )
230 .await?;
231 }
232
233 if self.base.url == ZED_REPO_URL {
234 std::fs::write(self.worktree_path.join(".rules"), std::fs::read(".rules")?)?;
235 }
236
237 std::fs::create_dir_all(self.example_output_directory())?;
238
239 Ok(())
240 }
241
242 pub fn run(
243 &self,
244 model: Arc<dyn LanguageModel>,
245 app_state: Arc<AgentAppState>,
246 cx: &mut App,
247 ) -> Task<Result<RunOutput>> {
248 let project = Project::local(
249 app_state.client.clone(),
250 app_state.node_runtime.clone(),
251 app_state.user_store.clone(),
252 app_state.languages.clone(),
253 app_state.fs.clone(),
254 None,
255 cx,
256 );
257
258 let worktree = project.update(cx, |project, cx| {
259 project.create_worktree(&self.worktree_path, true, cx)
260 });
261
262 let tools = cx.new(|_| ToolWorkingSet::default());
263 let thread_store =
264 ThreadStore::load(project.clone(), tools, app_state.prompt_builder.clone(), cx);
265 let this = self.clone();
266
267 cx.spawn(async move |cx| {
268 let worktree = worktree.await?;
269
270 // Wait for worktree scan to finish before choosing a file to open.
271 worktree
272 .update(cx, |worktree, _cx| {
273 worktree.as_local().unwrap().scan_complete()
274 })?
275 .await;
276
277 let lsp = if this.base.require_lsp {
278 let language_extension = this.base.language_extension.as_deref().context(
279 "language_extension field is required in base.toml when `require_lsp == true`",
280 )?;
281
282 // Open a file that matches the language to cause LSP to start.
283 let language_file = worktree.read_with(cx, |worktree, _cx| {
284 worktree
285 .files(false, 0)
286 .find_map(|e| {
287 if e.path.clone().extension().and_then(|ext| ext.to_str())
288 == Some(language_extension)
289 {
290 Some(ProjectPath {
291 worktree_id: worktree.id(),
292 path: e.path.clone(),
293 })
294 } else {
295 None
296 }
297 })
298 .context("Failed to find a file for example language")
299 })??;
300
301 let open_language_file_buffer_task = project.update(cx, |project, cx| {
302 project.open_buffer(language_file.clone(), cx)
303 })?;
304
305 let language_file_buffer = open_language_file_buffer_task.await?;
306
307 let lsp_open_handle = project.update(cx, |project, cx| {
308 project.register_buffer_with_language_servers(&language_file_buffer, cx)
309 })?;
310
311 wait_for_lang_server(&project, &language_file_buffer, this.log_prefix.clone(), cx).await?;
312
313 Some((lsp_open_handle, language_file_buffer))
314 } else {
315 None
316 };
317
318 let diagnostics_before = query_lsp_diagnostics(project.clone(), cx).await?;
319 if diagnostics_before.is_some() && !this.base.allow_preexisting_diagnostics {
320 return Err(anyhow!("Example has pre-existing diagnostics. If you want to run this example regardless, set `allow_preexisting_diagnostics` to `true` in `base.toml`"));
321 }
322
323 if std::env::var("ZED_EVAL_SETUP_ONLY").is_ok() {
324 return Err(anyhow!("Setup only mode"));
325 }
326
327 let example_output_dir = this.example_output_directory();
328 let last_diff_file_path = example_output_dir.join("last.diff");
329
330 // Write an empty "last.diff" so that it can be opened in Zed for convenient view of the
331 // history using undo/redo.
332 std::fs::write(&last_diff_file_path, "")?;
333
334 let thread_store = thread_store.await?;
335 let thread =
336 thread_store.update(cx, |thread_store, cx| thread_store.create_thread(cx))?;
337 let last_request = Rc::new(RefCell::new(None));
338
339 thread.update(cx, |thread, _cx| {
340 let mut request_count = 0;
341 let last_request = Rc::clone(&last_request);
342 let previous_diff = Rc::new(RefCell::new("".to_string()));
343 let example_output_dir = example_output_dir.clone();
344 let last_diff_file_path = last_diff_file_path.clone();
345 let this = this.clone();
346 thread.set_request_callback(move |request, response_events| {
347 *last_request.borrow_mut() = Some(request.clone());
348
349 request_count += 1;
350 let messages_file_path = example_output_dir.join(format!("{request_count}.messages.md"));
351 let diff_file_path = example_output_dir.join(format!("{request_count}.diff"));
352 let last_messages_file_path = example_output_dir.join("last.messages.md");
353 let request_markdown = RequestMarkdown::new(request);
354 let response_events_markdown = response_events_to_markdown(response_events);
355
356 let messages = format!("{}\n\n{}", request_markdown.messages, response_events_markdown);
357 fs::write(&messages_file_path, messages.clone()).expect("failed to write messages file");
358 fs::write(&last_messages_file_path, messages).expect("failed to write last messages file");
359
360 let diff_result = smol::block_on(this.repository_diff());
361 match diff_result {
362 Ok(diff) => {
363 if diff != previous_diff.borrow().clone() {
364 fs::write(&diff_file_path, &diff).expect("failed to write diff file");
365 fs::write(&last_diff_file_path, &diff).expect("failed to write last diff file");
366 *previous_diff.borrow_mut() = diff;
367 }
368 }
369 Err(err) => {
370 let error_message = format!("{err:?}");
371 fs::write(&diff_file_path, &error_message).expect("failed to write diff error to file");
372 fs::write(&last_diff_file_path, &error_message).expect("failed to write last diff file");
373 }
374 }
375
376 if request_count == 1 {
377 let tools_file_path = example_output_dir.join("tools.md");
378 fs::write(tools_file_path, request_markdown.tools).expect("failed to write tools file");
379 }
380 });
381 })?;
382
383 let tool_metrics = Arc::new(Mutex::new(ToolMetrics::default()));
384
385 let (thread_event_tx, mut thread_event_rx) = mpsc::unbounded();
386
387 let subscription = cx.subscribe(&thread, move |_thread, event: &ThreadEvent, _cx| {
388 thread_event_tx.unbounded_send(event.clone()).log_err();
389 });
390
391 let event_handler_task = cx.spawn({
392 let log_prefix = this.log_prefix.clone();
393 let tool_metrics = tool_metrics.clone();
394 let thread = thread.downgrade();
395 async move |cx| {
396 loop {
397 let event = select_biased! {
398 event = thread_event_rx.next() => event,
399 _ = cx.background_executor().timer(THREAD_EVENT_TIMEOUT).fuse() => {
400 return Err(anyhow!("Agentic loop stalled - waited {:?} without any events", THREAD_EVENT_TIMEOUT));
401 }
402 };
403 let Some(event) = event else {
404 return Err(anyhow!("ThreadEvent channel ended early"));
405 };
406
407 match event {
408 ThreadEvent::Stopped(reason) => match reason {
409 Ok(StopReason::EndTurn) => {
410 return Ok(());
411 }
412 Ok(StopReason::MaxTokens) => {
413 return Err(anyhow!("Exceeded maximum tokens"));
414 }
415 Ok(StopReason::ToolUse) => {
416 if std::env::var("ZED_EVAL_DEBUG").is_ok() {
417 println!("{}StopReason: Tool use", log_prefix);
418 }
419 }
420 Err(error) => {
421 return Err(anyhow!(error.clone()));
422 }
423 },
424 ThreadEvent::ShowError(thread_error) => {
425 break Err(anyhow!(thread_error.clone()));
426 }
427 ThreadEvent::StreamedAssistantText(_, _)| ThreadEvent::StreamedAssistantThinking(_, _) | ThreadEvent::UsePendingTools { .. } => {
428 }
429 ThreadEvent::ToolFinished {
430 tool_use_id,
431 pending_tool_use,
432 ..
433 } => {
434 thread.update(cx, |thread, _cx| {
435 if let Some(tool_use) = pending_tool_use {
436 let mut tool_metrics = tool_metrics.lock().unwrap();
437 if let Some(tool_result) = thread.tool_result(&tool_use_id) {
438 let message = if tool_result.is_error {
439 format!("TOOL FAILED: {}", tool_use.name)
440 } else {
441 format!("TOOL FINISHED: {}", tool_use.name)
442 };
443 println!("{log_prefix}{message}");
444 tool_metrics.insert(tool_result.tool_name.clone(), !tool_result.is_error);
445 } else {
446 let message = format!("TOOL FINISHED WITHOUT RESULT: {}", tool_use.name);
447 println!("{log_prefix}{message}");
448 tool_metrics.insert(tool_use.name.clone(), true);
449 }
450 }
451 })?;
452 }
453 ThreadEvent::ToolConfirmationNeeded => {
454 panic!("{}Bug: Tool confirmation should not be required in eval", log_prefix);
455 },
456 ThreadEvent::StreamedToolUse { .. } |
457 ThreadEvent::StreamedCompletion |
458 ThreadEvent::MessageAdded(_) |
459 ThreadEvent::MessageEdited(_) |
460 ThreadEvent::MessageDeleted(_) |
461 ThreadEvent::SummaryChanged |
462 ThreadEvent::SummaryGenerated |
463 ThreadEvent::CheckpointChanged |
464 ThreadEvent::ReceivedTextChunk |
465 ThreadEvent::UsageUpdated(_) => {
466 if std::env::var("ZED_EVAL_DEBUG").is_ok() {
467 println!("{}Event: {:#?}", log_prefix, event);
468 }
469 }
470 }
471 }
472 }
473 });
474
475 thread.update(cx, |thread, cx| {
476 let context = vec![];
477 thread.insert_user_message(this.prompt.clone(), context, None, cx);
478 thread.send_to_model(model, cx);
479 })?;
480
481 event_handler_task.await?;
482
483 println!("{}Stopped", this.log_prefix);
484
485 if let Some((_, language_file_buffer)) = lsp.as_ref() {
486 wait_for_lang_server(&project, &language_file_buffer, this.log_prefix.clone(), cx).await?;
487 }
488
489 println!("{}Getting repository diff", this.log_prefix);
490 let repository_diff = this.repository_diff().await?;
491 std::fs::write(last_diff_file_path, &repository_diff)?;
492
493 println!("{}Getting diagnostics", this.log_prefix);
494 let diagnostics_after = cx
495 .update(move |cx| {
496 cx.spawn(async move |cx| query_lsp_diagnostics(project, cx).await)
497 })?
498 .await?;
499 println!("{}Got diagnostics", this.log_prefix);
500
501 let Some(last_request) = last_request.borrow_mut().take() else {
502 return Err(anyhow!("No requests ran."));
503 };
504
505 drop(subscription);
506 drop(lsp);
507
508 if let Some(diagnostics_before) = &diagnostics_before {
509 fs::write(example_output_dir.join("diagnostics_before.txt"), diagnostics_before)?;
510 }
511
512 if let Some(diagnostics_after) = &diagnostics_after {
513 fs::write(example_output_dir.join("diagnostics_after.txt"), diagnostics_after)?;
514 }
515
516
517 thread.update(cx, |thread, _cx| {
518 let response_count = thread
519 .messages()
520 .filter(|message| message.role == language_model::Role::Assistant)
521 .count();
522 RunOutput {
523 repository_diff,
524 ran_diagnostics_check: this.base.require_lsp,
525 diagnostics_before,
526 diagnostics_after,
527 response_count,
528 token_usage: thread.cumulative_token_usage(),
529 tool_metrics: tool_metrics.lock().unwrap().clone(),
530 last_request,
531 }
532 })
533 })
534 }
535
536 async fn judge_diff(
537 &self,
538 model: Arc<dyn LanguageModel>,
539 run_output: &RunOutput,
540 judge_number: u32,
541 cx: &AsyncApp,
542 ) -> Result<(String, JudgeResponse)> {
543 let judge_diff_prompt = include_str!("judge_diff_prompt.hbs");
544 let judge_diff_prompt_name = "judge_diff_prompt";
545 let mut hbs = Handlebars::new();
546 hbs.register_template_string(judge_diff_prompt_name, judge_diff_prompt)?;
547
548 let diff_prompt = hbs.render(
549 judge_diff_prompt_name,
550 &JudgeDiffInput {
551 repository_diff: run_output.repository_diff.clone(),
552 ran_diagnostics_check: run_output.ran_diagnostics_check,
553 diagnostics_before: run_output.diagnostics_before.clone(),
554 diagnostics_after: run_output.diagnostics_after.clone(),
555 criteria: self.diff_criteria.clone(),
556 },
557 )?;
558
559 let request = LanguageModelRequest {
560 thread_id: None,
561 prompt_id: None,
562 messages: vec![LanguageModelRequestMessage {
563 role: Role::User,
564 content: vec![MessageContent::Text(diff_prompt)],
565 cache: false,
566 }],
567 temperature: None,
568 tools: Vec::new(),
569 stop: Vec::new(),
570 };
571
572 let diff_response = send_language_model_request(model, request, cx).await?;
573 let diff_output = JudgeResponse::parse(&diff_response)?;
574
575 println!(
576 "{}Judge #{judge_number} - Diff score: {}",
577 self.log_prefix, diff_output.score
578 );
579
580 Ok((diff_response, diff_output))
581 }
582
583 async fn judge_thread(
584 &self,
585 model: Arc<dyn LanguageModel>,
586 run_output: &RunOutput,
587 judge_number: u32,
588 cx: &AsyncApp,
589 ) -> Result<(String, Option<JudgeResponse>)> {
590 if let Some(criteria) = self.thread_criteria.clone() {
591 let judge_thread_prompt = include_str!("judge_thread_prompt.hbs");
592 let judge_thread_prompt_name = "judge_thread_prompt";
593 let mut hbs = Handlebars::new();
594 hbs.register_template_string(judge_thread_prompt_name, judge_thread_prompt)?;
595
596 let request_markdown = RequestMarkdown::new(&run_output.last_request);
597 let thread_prompt = hbs.render(
598 judge_thread_prompt_name,
599 &JudgeThreadInput {
600 messages: request_markdown.messages,
601 criteria,
602 },
603 )?;
604
605 let request = LanguageModelRequest {
606 thread_id: None,
607 prompt_id: None,
608 messages: vec![LanguageModelRequestMessage {
609 role: Role::User,
610 content: vec![MessageContent::Text(thread_prompt)],
611 cache: false,
612 }],
613 temperature: None,
614 tools: Vec::new(),
615 stop: Vec::new(),
616 };
617
618 let thread_response = send_language_model_request(model, request, cx).await?;
619 let thread_output = JudgeResponse::parse(&thread_response)?;
620
621 println!(
622 "{}Judge #{judge_number} - Thread score: {}",
623 self.log_prefix, thread_output.score
624 );
625
626 Ok((thread_response, Some(thread_output)))
627 } else {
628 let msg = "There were no criteria specified for this thread, so this example was not judged on its thread.".to_string();
629 Ok((msg, None))
630 }
631 }
632
633 pub async fn judge(
634 &self,
635 model: Arc<dyn LanguageModel>,
636 run_output: &RunOutput,
637 judge_number: u32,
638 cx: &AsyncApp,
639 ) -> Result<JudgeOutput> {
640 let mut output_file = File::create(
641 self.example_output_directory()
642 .join(format!("judge_{}.md", judge_number)),
643 )
644 .expect("failed to create judge.md");
645
646 println!("{}Running judge #{judge_number}", self.log_prefix);
647
648 let diff_task = self.judge_diff(model.clone(), &run_output, judge_number, cx);
649 let thread_task = self.judge_thread(model.clone(), &run_output, judge_number, cx);
650
651 let (diff_result, thread_result) = futures::join!(diff_task, thread_task);
652
653 let (diff_response, diff_output) = diff_result?;
654 let (thread_response, thread_output) = thread_result?;
655
656 writeln!(
657 &mut output_file,
658 "# Judgment\n\n## Thread\n\n{thread_response}\n\n## Diff\n\n{diff_response}",
659 )
660 .log_err();
661
662 Ok(JudgeOutput {
663 thread: thread_output,
664 diff: diff_output,
665 })
666 }
667
668 async fn repository_diff(&self) -> Result<String> {
669 run_git(&self.worktree_path, &["add", "."]).await?;
670 let mut diff_args = vec!["diff", "--staged"];
671 if self.base.url == ZED_REPO_URL {
672 diff_args.push(":(exclude).rules");
673 }
674 run_git(&self.worktree_path, &diff_args).await
675 }
676}
677
678fn wait_for_lang_server(
679 project: &Entity<Project>,
680 buffer: &Entity<Buffer>,
681 log_prefix: String,
682 cx: &mut AsyncApp,
683) -> Task<Result<()>> {
684 println!("{}⏵ Waiting for language server", log_prefix);
685
686 let (mut tx, mut rx) = mpsc::channel(1);
687
688 let lsp_store = project
689 .update(cx, |project, _| project.lsp_store())
690 .unwrap();
691
692 let has_lang_server = buffer
693 .update(cx, |buffer, cx| {
694 lsp_store.update(cx, |lsp_store, cx| {
695 lsp_store
696 .language_servers_for_local_buffer(&buffer, cx)
697 .next()
698 .is_some()
699 })
700 })
701 .unwrap_or(false);
702
703 if has_lang_server {
704 project
705 .update(cx, |project, cx| project.save_buffer(buffer.clone(), cx))
706 .unwrap()
707 .detach();
708 }
709
710 let subscriptions =
711 [
712 cx.subscribe(&lsp_store, {
713 let log_prefix = log_prefix.clone();
714 move |_, event, _| match event {
715 project::LspStoreEvent::LanguageServerUpdate {
716 message:
717 client::proto::update_language_server::Variant::WorkProgress(
718 LspWorkProgress {
719 message: Some(message),
720 ..
721 },
722 ),
723 ..
724 } => println!("{}⟲ {message}", log_prefix),
725 _ => {}
726 }
727 }),
728 cx.subscribe(&project, {
729 let buffer = buffer.clone();
730 move |project, event, cx| match event {
731 project::Event::LanguageServerAdded(_, _, _) => {
732 let buffer = buffer.clone();
733 project
734 .update(cx, |project, cx| project.save_buffer(buffer, cx))
735 .detach();
736 }
737 project::Event::DiskBasedDiagnosticsFinished { .. } => {
738 tx.try_send(()).ok();
739 }
740 _ => {}
741 }
742 }),
743 ];
744
745 cx.spawn(async move |cx| {
746 let timeout = cx.background_executor().timer(Duration::new(60 * 5, 0));
747 let result = futures::select! {
748 _ = rx.next() => {
749 println!("{}⚑ Language server idle", log_prefix);
750 anyhow::Ok(())
751 },
752 _ = timeout.fuse() => {
753 Err(anyhow!("LSP wait timed out after 5 minutes"))
754 }
755 };
756 drop(subscriptions);
757 result
758 })
759}
760
761async fn query_lsp_diagnostics(
762 project: Entity<Project>,
763 cx: &mut AsyncApp,
764) -> Result<Option<String>> {
765 let paths_with_diagnostics = project.update(cx, |project, cx| {
766 project
767 .diagnostic_summaries(true, cx)
768 .filter(|(_, _, summary)| summary.error_count > 0 || summary.warning_count > 0)
769 .map(|(project_path, _, _)| project_path)
770 .collect::<Vec<_>>()
771 })?;
772
773 if paths_with_diagnostics.is_empty() {
774 return Ok(None);
775 }
776
777 let mut output = String::new();
778 for project_path in paths_with_diagnostics {
779 let buffer = project
780 .update(cx, |project, cx| project.open_buffer(project_path, cx))?
781 .await?;
782 let snapshot = buffer.read_with(cx, |buffer, _cx| buffer.snapshot())?;
783
784 for (_, group) in snapshot.diagnostic_groups(None) {
785 let entry = &group.entries[group.primary_ix];
786 let range = entry.range.to_point(&snapshot);
787 let severity = match entry.diagnostic.severity {
788 DiagnosticSeverity::ERROR => "error",
789 DiagnosticSeverity::WARNING => "warning",
790 _ => continue,
791 };
792
793 writeln!(
794 output,
795 "{} at line {}: {}",
796 severity,
797 range.start.row + 1,
798 entry.diagnostic.message
799 )?;
800 }
801 }
802 anyhow::Ok(Some(output))
803}
804
805impl JudgeResponse {
806 fn parse(response: &str) -> Result<Self> {
807 let analysis = get_tag("analysis", response)?.to_string();
808 let score = get_tag("score", response)?
809 .parse()
810 .context("error parsing score")?;
811
812 Ok(Self { analysis, score })
813 }
814}
815
816fn get_tag(name: &'static str, response: &str) -> Result<String> {
817 let start_tag = format!("<{}>", name);
818 let end_tag = format!("</{}>", name);
819
820 let start_ix = response
821 .find(&start_tag)
822 .context(format!("{} start tag not found", name))?;
823 let content_start_ix = start_ix + start_tag.len();
824
825 let end_ix = content_start_ix
826 + response[content_start_ix..]
827 .find(&end_tag)
828 .context(format!("{} end tag not found", name))?;
829
830 let content = response[content_start_ix..end_ix].trim().unindent();
831
832 anyhow::Ok(content)
833}
834
835pub fn repo_path_for_url(repos_dir: &Path, repo_url: &str) -> PathBuf {
836 let repo_name = repo_url
837 .trim_start_matches("https://")
838 .replace(|c: char| !c.is_alphanumeric(), "-");
839 Path::new(repos_dir)
840 .canonicalize()
841 .context(format!("No such directory {}", repos_dir.display()))
842 .unwrap()
843 .join(repo_name)
844}
845
846pub async fn run_git(repo_path: &Path, args: &[&str]) -> Result<String> {
847 let output = new_smol_command("git")
848 .current_dir(repo_path)
849 .args(args)
850 .output()
851 .await?;
852
853 if output.status.success() {
854 Ok(String::from_utf8(output.stdout)?.trim().to_string())
855 } else {
856 Err(anyhow!(
857 "`git {}` within `{}` failed with status: {}\nstderr:\n{}\nstdout:\n{}",
858 args.join(" "),
859 repo_path.display(),
860 output.status,
861 String::from_utf8_lossy(&output.stderr),
862 String::from_utf8_lossy(&output.stdout),
863 ))
864 }
865}
866
867pub async fn send_language_model_request(
868 model: Arc<dyn LanguageModel>,
869 request: LanguageModelRequest,
870 cx: &AsyncApp,
871) -> anyhow::Result<String> {
872 match model.stream_completion_text(request, &cx).await {
873 Ok(mut stream) => {
874 let mut full_response = String::new();
875 while let Some(chunk_result) = stream.stream.next().await {
876 match chunk_result {
877 Ok(chunk_str) => {
878 full_response.push_str(&chunk_str);
879 }
880 Err(err) => {
881 return Err(anyhow!(
882 "Error receiving response from language model: {err}"
883 ));
884 }
885 }
886 }
887 Ok(full_response)
888 }
889 Err(err) => Err(anyhow!(
890 "Failed to get response from language model. Error was: {err}"
891 )),
892 }
893}
894
895struct RequestMarkdown {
896 tools: String,
897 messages: String,
898}
899
900impl RequestMarkdown {
901 fn new(request: &LanguageModelRequest) -> Self {
902 let mut tools = String::new();
903 let mut messages = String::new();
904 let mut assistant_message_number: u32 = 1;
905
906 // Print the tools
907 if !request.tools.is_empty() {
908 for tool in &request.tools {
909 write!(&mut tools, "# {}\n\n", tool.name).unwrap();
910 write!(&mut tools, "{}\n\n", tool.description).unwrap();
911 write!(
912 &mut tools,
913 "{}\n",
914 MarkdownString::code_block("json", &format!("{:#}", tool.input_schema))
915 )
916 .unwrap();
917 }
918 }
919
920 // Print the messages
921 for message in &request.messages {
922 match message.role {
923 Role::System => messages.push_str("# ⚙️ SYSTEM\n\n"),
924 Role::User => messages.push_str("# 👤 USER\n\n"),
925 Role::Assistant => {
926 messages.push_str(&format!("# 🤖 ASSISTANT {assistant_message_number}\n\n"));
927 assistant_message_number += 1;
928 }
929 };
930
931 for content in &message.content {
932 match content {
933 MessageContent::Text(text) => {
934 messages.push_str(text);
935 messages.push_str("\n\n");
936 }
937 MessageContent::Image(_) => {
938 messages.push_str("[IMAGE DATA]\n\n");
939 }
940 MessageContent::Thinking { text, signature } => {
941 messages.push_str("**Thinking**:\n\n");
942 if let Some(sig) = signature {
943 messages.push_str(&format!("Signature: {}\n\n", sig));
944 }
945 messages.push_str(text);
946 messages.push_str("\n");
947 }
948 MessageContent::RedactedThinking(items) => {
949 messages.push_str(&format!(
950 "**Redacted Thinking**: {} item(s)\n\n",
951 items.len()
952 ));
953 }
954 MessageContent::ToolUse(tool_use) => {
955 messages.push_str(&format!(
956 "**Tool Use**: {} (ID: {})\n",
957 tool_use.name, tool_use.id
958 ));
959 messages.push_str(&format!(
960 "{}\n",
961 MarkdownString::code_block("json", &format!("{:#}", tool_use.input))
962 ));
963 }
964 MessageContent::ToolResult(tool_result) => {
965 messages.push_str(&format!(
966 "**Tool Result**: {} (ID: {})\n\n",
967 tool_result.tool_name, tool_result.tool_use_id
968 ));
969 if tool_result.is_error {
970 messages.push_str("**ERROR:**\n");
971 }
972 messages.push_str(&format!("{}\n\n", tool_result.content));
973 }
974 }
975 }
976 }
977
978 Self { tools, messages }
979 }
980}
981
982fn response_events_to_markdown(
983 response_events: &[std::result::Result<LanguageModelCompletionEvent, String>],
984) -> String {
985 let mut response = String::new();
986 // Print the response events if any
987 response.push_str("# Response\n\n");
988 let mut text_buffer = String::new();
989 let mut thinking_buffer = String::new();
990
991 let flush_buffers =
992 |output: &mut String, text_buffer: &mut String, thinking_buffer: &mut String| {
993 if !text_buffer.is_empty() {
994 output.push_str(&format!("**Text**:\n{}\n\n", text_buffer));
995 text_buffer.clear();
996 }
997 if !thinking_buffer.is_empty() {
998 output.push_str(&format!("**Thinking**:\n{}\n\n", thinking_buffer));
999 thinking_buffer.clear();
1000 }
1001 };
1002
1003 for event in response_events {
1004 match event {
1005 Ok(LanguageModelCompletionEvent::Text(text)) => {
1006 text_buffer.push_str(text);
1007 }
1008 Ok(LanguageModelCompletionEvent::Thinking { text, .. }) => {
1009 thinking_buffer.push_str(text);
1010 }
1011 Ok(LanguageModelCompletionEvent::Stop(reason)) => {
1012 flush_buffers(&mut response, &mut text_buffer, &mut thinking_buffer);
1013 response.push_str(&format!("**Stop**: {:?}\n\n", reason));
1014 }
1015 Ok(LanguageModelCompletionEvent::ToolUse(tool_use)) => {
1016 flush_buffers(&mut response, &mut text_buffer, &mut thinking_buffer);
1017 response.push_str(&format!(
1018 "**Tool Use**: {} (ID: {})\n",
1019 tool_use.name, tool_use.id
1020 ));
1021 response.push_str(&format!(
1022 "{}\n",
1023 MarkdownString::code_block("json", &format!("{:#}", tool_use.input))
1024 ));
1025 }
1026 Ok(
1027 LanguageModelCompletionEvent::UsageUpdate(_)
1028 | LanguageModelCompletionEvent::StartMessage { .. },
1029 ) => {}
1030 Err(error) => {
1031 flush_buffers(&mut response, &mut text_buffer, &mut thinking_buffer);
1032 response.push_str(&format!("**Error**: {}\n\n", error));
1033 }
1034 }
1035 }
1036
1037 flush_buffers(&mut response, &mut text_buffer, &mut thinking_buffer);
1038
1039 response
1040}
1041
1042#[cfg(test)]
1043mod test {
1044 use super::*;
1045 use handlebars::Handlebars;
1046
1047 #[test]
1048 fn test_parse_judge_output() {
1049 let response = r#"
1050 <analysis>The model did a good job but there were still compilations errors.</analysis>
1051 <score>3</score>
1052 "#
1053 .unindent();
1054
1055 let output = JudgeResponse::parse(&response).unwrap();
1056 assert_eq!(
1057 output.analysis,
1058 "The model did a good job but there were still compilations errors."
1059 );
1060 assert_eq!(output.score, 3);
1061
1062 let response = r#"
1063 Text around ignored
1064
1065 <analysis>
1066 Failed to compile:
1067 - Error 1
1068 - Error 2
1069 </analysis>
1070
1071 <score>1</score>
1072 "#
1073 .unindent();
1074
1075 let output = JudgeResponse::parse(&response).unwrap();
1076 assert_eq!(output.analysis, "Failed to compile:\n- Error 1\n- Error 2");
1077 assert_eq!(output.score, 1);
1078 }
1079
1080 #[test]
1081 fn test_judge_prompt_with_diagnostics() {
1082 // Case 1: Both diagnostics before and after are present
1083 let input = JudgeDiffInput {
1084 repository_diff: "diff content goes here".to_string(),
1085 ran_diagnostics_check: true,
1086 diagnostics_before: Some("Error at line 10: variable not found".to_string()),
1087 diagnostics_after: Some("Error at line 15: missing semicolon".to_string()),
1088 criteria: "Fix all bugs".to_string(),
1089 };
1090
1091 let rendered = templates().render(JUDGE_PROMPT_NAME, &input).unwrap();
1092
1093 let expected_diagnostics_section = r#"
1094 Take into account the diagnostics before and after applying the change:
1095
1096 <diagnostics_before>
1097 Error at line 10: variable not found
1098 </diagnostics_before>
1099
1100 <diagnostics_after>
1101 Error at line 15: missing semicolon
1102 </diagnostics_after>
1103 "#
1104 .unindent();
1105
1106 assert!(rendered.contains(&expected_diagnostics_section));
1107 }
1108
1109 #[test]
1110 fn test_judge_prompt_with_empty_diagnostics() {
1111 // Case 2: Diagnostics check run but no diagnostics found
1112 let input = JudgeDiffInput {
1113 repository_diff: "diff content goes here".to_string(),
1114 ran_diagnostics_check: true,
1115 diagnostics_before: None,
1116 diagnostics_after: None,
1117 criteria: "Fix all bugs".to_string(),
1118 };
1119
1120 let rendered = templates().render(JUDGE_PROMPT_NAME, &input).unwrap();
1121
1122 let expected_diagnostics_section = r#"
1123 Take into account the diagnostics before and after applying the change:
1124
1125 <diagnostics_before>
1126 No diagnostics before applying the edits.
1127 </diagnostics_before>
1128
1129 <diagnostics_after>
1130 No diagnostics after applying the edits.
1131 </diagnostics_after>
1132 "#
1133 .unindent();
1134
1135 assert!(rendered.contains(&expected_diagnostics_section));
1136 }
1137
1138 #[test]
1139 fn test_judge_prompt_with_mixed_diagnostics() {
1140 let templates = templates();
1141
1142 // Case 3: Before diagnostics present, after diagnostics absent
1143 let input = JudgeDiffInput {
1144 repository_diff: "diff content goes here".to_string(),
1145 ran_diagnostics_check: true,
1146 diagnostics_before: Some("Error at line 10: variable not found".to_string()),
1147 diagnostics_after: None,
1148 criteria: "Fix all bugs".to_string(),
1149 };
1150
1151 let rendered = templates.render(JUDGE_PROMPT_NAME, &input).unwrap();
1152
1153 let expected_diagnostics_section = r#"
1154 Take into account the diagnostics before and after applying the change:
1155
1156 <diagnostics_before>
1157 Error at line 10: variable not found
1158 </diagnostics_before>
1159
1160 <diagnostics_after>
1161 No diagnostics after applying the edits.
1162 </diagnostics_after>
1163 "#
1164 .unindent();
1165
1166 assert!(rendered.contains(&expected_diagnostics_section));
1167
1168 // Case 4: Before diagnostics absent, after diagnostics present
1169 let input = JudgeDiffInput {
1170 repository_diff: "diff content goes here".to_string(),
1171 ran_diagnostics_check: true,
1172 diagnostics_before: None,
1173 diagnostics_after: Some("Error at line 15: missing semicolon".to_string()),
1174 criteria: "Fix all bugs".to_string(),
1175 };
1176
1177 let rendered = templates.render(JUDGE_PROMPT_NAME, &input).unwrap();
1178
1179 let expected_diagnostics_section = r#"
1180 Take into account the diagnostics before and after applying the change:
1181
1182 <diagnostics_before>
1183 No diagnostics before applying the edits.
1184 </diagnostics_before>
1185
1186 <diagnostics_after>
1187 Error at line 15: missing semicolon
1188 </diagnostics_after>
1189 "#
1190 .unindent();
1191
1192 assert!(rendered.contains(&expected_diagnostics_section));
1193 }
1194
1195 #[test]
1196 fn test_judge_prompt_without_diagnostics() {
1197 let templates = templates();
1198
1199 // Case 5: No diagnostics check run
1200 let input = JudgeDiffInput {
1201 repository_diff: "diff content goes here".to_string(),
1202 ran_diagnostics_check: false,
1203 diagnostics_before: None,
1204 diagnostics_after: None,
1205 criteria: "Fix all bugs".to_string(),
1206 };
1207
1208 let rendered = templates.render(JUDGE_PROMPT_NAME, &input).unwrap();
1209
1210 // Check for the message when no diagnostics were performed
1211 let diagnostics_message = "No diagnostic checks were performed.";
1212
1213 assert!(rendered.contains(diagnostics_message));
1214 assert!(!rendered.contains("<diagnostics_before>"));
1215 assert!(!rendered.contains("<diagnostics_after>"));
1216 }
1217
1218 const JUDGE_PROMPT_NAME: &str = "judge_prompt";
1219
1220 fn templates() -> Handlebars<'static> {
1221 let mut judge_prompt = include_str!("judge_diff_prompt.hbs").to_string();
1222 language::LineEnding::normalize(&mut judge_prompt);
1223 let mut handlebars = Handlebars::new();
1224 handlebars
1225 .register_template_string(JUDGE_PROMPT_NAME, judge_prompt)
1226 .unwrap();
1227 handlebars
1228 }
1229}