1use agent::{RequestKind, ThreadEvent, ThreadStore};
2use anyhow::{Context as _, Result, anyhow};
3use assistant_tool::ToolWorkingSet;
4use client::proto::LspWorkProgress;
5use collections::HashMap;
6use dap::DapRegistry;
7use futures::channel::mpsc;
8use futures::{FutureExt, StreamExt as _, select_biased};
9use gpui::{App, AppContext as _, AsyncApp, Entity, Task};
10use handlebars::Handlebars;
11use language::{DiagnosticSeverity, OffsetRangeExt};
12use language_model::{
13 LanguageModel, LanguageModelRequest, LanguageModelRequestMessage, MessageContent, Role,
14 StopReason, TokenUsage,
15};
16use project::{LspStore, Project, ProjectPath};
17use serde::{Deserialize, Serialize};
18use std::fmt::Write as _;
19use std::fs::File;
20use std::io::Write as _;
21use std::sync::{Arc, Mutex};
22use std::time::Duration;
23use std::{
24 fs,
25 path::{Path, PathBuf},
26};
27use unindent::Unindent as _;
28use util::ResultExt as _;
29use util::command::new_smol_command;
30use util::serde::default_true;
31
32use crate::AgentAppState;
33
34pub const EXAMPLES_DIR: &str = "./crates/eval/examples";
35pub const REPOS_DIR: &str = "./crates/eval/repos";
36pub const WORKTREES_DIR: &str = "./crates/eval/worktrees";
37
38const THREAD_EVENT_TIMEOUT: Duration = Duration::from_secs(60 * 2);
39
40#[derive(Clone, Debug, Deserialize)]
41pub struct ExampleBase {
42 pub url: String,
43 pub revision: String,
44 pub language_extension: Option<String>,
45 pub insert_id: Option<String>,
46 #[serde(default = "default_true")]
47 pub require_lsp: bool,
48}
49
50#[derive(Clone, Debug)]
51pub struct Example {
52 pub name: String,
53 /// Content of `base.toml`
54 pub base: ExampleBase,
55 /// Content of `prompt.md`
56 pub prompt: String,
57 /// Content of `criteria.md`
58 pub criteria: String,
59 /// Markdown output file to append to
60 pub output_file: Option<Arc<Mutex<File>>>,
61 /// Path to the output run directory.
62 pub run_dir: PathBuf,
63 /// Path to markdown output file
64 pub output_file_path: PathBuf,
65 /// Prefix used for logging that identifies this example
66 pub log_prefix: String,
67}
68
69#[derive(Debug, Serialize, Deserialize, Clone)]
70pub struct RunOutput {
71 pub repository_diff: String,
72 pub diagnostics: String,
73 pub response_count: usize,
74 pub token_usage: TokenUsage,
75 pub tool_use_counts: HashMap<Arc<str>, u32>,
76}
77
78#[derive(Debug, Clone, Serialize, Deserialize)]
79pub struct JudgeInput {
80 pub repository_diff: String,
81 pub criteria: String,
82}
83
84#[derive(Debug, Clone, Serialize, Deserialize)]
85pub struct JudgeOutput {
86 pub analysis: String,
87 pub score: u32,
88}
89
90impl Example {
91 /// Load an example from a directory containing base.toml, prompt.md, and criteria.md
92 pub fn load_from_directory(dir_path: &Path, run_dir: &Path) -> Result<Self> {
93 let name = Self::name_from_path(dir_path);
94 let base_path = dir_path.join("base.toml");
95 let prompt_path = dir_path.join("prompt.md");
96 let criteria_path = dir_path.join("criteria.md");
97 let output_file_path = run_dir.join(format!("{}.md", name));
98
99 Ok(Example {
100 name: name.clone(),
101 base: toml::from_str(&fs::read_to_string(&base_path)?)?,
102 prompt: fs::read_to_string(prompt_path.clone())?,
103 criteria: fs::read_to_string(criteria_path.clone())?,
104 run_dir: run_dir.to_path_buf(),
105 output_file: None,
106 output_file_path,
107 log_prefix: name,
108 })
109 }
110
111 pub fn set_repetition_number(&mut self, repetition_number: u32) {
112 if repetition_number > 0 {
113 self.name = format!("{}-{}", self.name, repetition_number);
114 self.output_file_path = self.run_dir.join(format!("{}.md", self.name));
115 }
116 }
117
118 pub fn set_log_prefix_style(&mut self, color: &str, name_width: usize) {
119 self.log_prefix = format!(
120 "{}{:<width$}\x1b[0m | ",
121 color,
122 self.name,
123 width = name_width
124 );
125 }
126
127 pub fn name_from_path(path: &Path) -> String {
128 path.file_name().unwrap().to_string_lossy().to_string()
129 }
130
131 pub fn worktree_path(&self) -> PathBuf {
132 Path::new(WORKTREES_DIR)
133 .canonicalize()
134 .context(format!("No such directory {WORKTREES_DIR}"))
135 .unwrap()
136 .join(&self.name)
137 }
138
139 /// Set up the example by checking out the specified Git revision
140 pub async fn setup(&mut self) -> Result<()> {
141 let repo_path = repo_path_for_url(&self.base.url);
142
143 let revision_exists = run_git(&repo_path, &["rev-parse", "--verify", &self.base.revision])
144 .await
145 .is_ok();
146
147 if !revision_exists {
148 println!(
149 "{}Fetching revision {}",
150 self.log_prefix, &self.base.revision
151 );
152 run_git(
153 &repo_path,
154 &["fetch", "--depth", "1", "origin", &self.base.revision],
155 )
156 .await?;
157 }
158
159 let worktree_path = self.worktree_path();
160
161 if worktree_path.is_dir() {
162 println!("{}Resetting existing worktree", self.log_prefix);
163
164 // TODO: consider including "-x" to remove ignored files. The downside of this is that
165 // it will also remove build artifacts, and so prevent incremental reuse there.
166 run_git(&worktree_path, &["clean", "--force", "-d"]).await?;
167 run_git(&worktree_path, &["reset", "--hard", "HEAD"]).await?;
168 run_git(&worktree_path, &["checkout", &self.base.revision]).await?;
169 } else {
170 println!("{}Creating worktree", self.log_prefix);
171
172 let worktree_path_string = worktree_path.to_string_lossy().to_string();
173
174 run_git(
175 &repo_path,
176 &[
177 "worktree",
178 "add",
179 "-f",
180 &worktree_path_string,
181 &self.base.revision,
182 ],
183 )
184 .await?;
185 }
186
187 // Create the output file
188 let output_file = Arc::new(Mutex::new(File::create(&self.output_file_path)?));
189 self.output_file = Some(output_file);
190
191 Ok(())
192 }
193
194 /// Returns the output file, panicking if it's not set
195 fn output_file(&self) -> Arc<Mutex<File>> {
196 self.output_file
197 .clone()
198 .expect("Output file not created. Call setup() first.")
199 }
200
201 pub fn run(
202 &self,
203 model: Arc<dyn LanguageModel>,
204 app_state: Arc<AgentAppState>,
205 cx: &mut App,
206 ) -> Task<Result<RunOutput>> {
207 let project = Project::local(
208 app_state.client.clone(),
209 app_state.node_runtime.clone(),
210 app_state.user_store.clone(),
211 app_state.languages.clone(),
212 Arc::new(DapRegistry::default()),
213 app_state.fs.clone(),
214 None,
215 cx,
216 );
217
218 let worktree_path = self.worktree_path();
219 let worktree = project.update(cx, |project, cx| {
220 project.create_worktree(&worktree_path, true, cx)
221 });
222
223 let tools = cx.new(|_| ToolWorkingSet::default());
224 let thread_store =
225 ThreadStore::load(project.clone(), tools, app_state.prompt_builder.clone(), cx);
226 let this = self.clone();
227
228 cx.spawn(async move |cx| {
229 let worktree = worktree.await?;
230
231 // Wait for worktree scan to finish before choosing a file to open.
232 worktree
233 .update(cx, |worktree, _cx| {
234 worktree.as_local().unwrap().scan_complete()
235 })?
236 .await;
237
238 let lsp_open_handle_and_store = if this.base.require_lsp {
239 let language_extension = this.base.language_extension.as_deref().context(
240 "language_extension field is required in base.toml when `require_lsp == true`",
241 )?;
242
243 // Open a file that matches the language to cause LSP to start.
244 let language_file = worktree.read_with(cx, |worktree, _cx| {
245 worktree
246 .files(false, 0)
247 .find_map(|e| {
248 if e.path.clone().extension().and_then(|ext| ext.to_str())
249 == Some(language_extension)
250 {
251 Some(ProjectPath {
252 worktree_id: worktree.id(),
253 path: e.path.clone(),
254 })
255 } else {
256 None
257 }
258 })
259 .context("Failed to find a file for example language")
260 })??;
261
262 let open_language_file_buffer_task = project.update(cx, |project, cx| {
263 project.open_buffer(language_file.clone(), cx)
264 })?;
265
266 let language_file_buffer = open_language_file_buffer_task.await?;
267
268 let (lsp_open_handle, lsp_store) = project.update(cx, |project, cx| {
269 (
270 project.register_buffer_with_language_servers(&language_file_buffer, cx),
271 project.lsp_store().clone(),
272 )
273 })?;
274
275 // TODO: remove this once the diagnostics tool waits for new diagnostics
276 cx.background_executor().timer(Duration::new(5, 0)).await;
277 wait_for_lang_server(&lsp_store, this.log_prefix.clone(), cx).await?;
278
279 lsp_store.update(cx, |lsp_store, cx| {
280 lsp_open_handle.update(cx, |buffer, cx| {
281 buffer.update(cx, |buffer, cx| {
282 let has_language_server = lsp_store
283 .language_servers_for_local_buffer(buffer, cx)
284 .next()
285 .is_some();
286 if has_language_server {
287 Ok(())
288 } else {
289 Err(anyhow!(
290 "`{:?}` was opened to cause the language server to start, \
291 but no language servers are registered for its buffer. \
292 Set `require_lsp = false` in `base.toml` to skip this.",
293 language_file
294 ))
295 }
296 })
297 })
298 })??;
299
300 Some((lsp_open_handle, lsp_store))
301 } else {
302 None
303 };
304
305 if std::env::var("ZED_EVAL_SETUP_ONLY").is_ok() {
306 return Err(anyhow!("Setup only mode"));
307 }
308
309 let thread_store = thread_store.await;
310 let thread =
311 thread_store.update(cx, |thread_store, cx| thread_store.create_thread(cx))?;
312
313 {
314 let output_file_ref = this.output_file();
315 let mut output_file = output_file_ref.lock().unwrap();
316 writeln!(&mut output_file, "👤 USER:").log_err();
317 writeln!(&mut output_file, "{}", this.prompt).log_err();
318 writeln!(&mut output_file, "🤖 ASSISTANT:").log_err();
319 output_file.flush().log_err();
320 }
321
322 let tool_use_counts: Arc<Mutex<HashMap<Arc<str>, u32>>> =
323 Mutex::new(HashMap::default()).into();
324
325 let (thread_event_tx, mut thread_event_rx) = mpsc::unbounded();
326
327 let subscription = cx.subscribe(&thread, move |_thread, event: &ThreadEvent, _cx| {
328 thread_event_tx.unbounded_send(event.clone()).log_err();
329 });
330
331 let event_handler_task = cx.spawn({
332 // Need to clone the Arc here because the reference from output_file() won't live long enough
333 let output_file = this.output_file.clone().unwrap();
334 let log_prefix = this.log_prefix.clone();
335 let tool_use_counts = tool_use_counts.clone();
336 let thread = thread.downgrade();
337 async move |cx| {
338 loop {
339 let event = select_biased! {
340 event = thread_event_rx.next() => event,
341 _ = cx.background_executor().timer(THREAD_EVENT_TIMEOUT).fuse() => {
342 return Err(anyhow!("Agentic loop stalled - waited {:?} without any events", THREAD_EVENT_TIMEOUT));
343 }
344 };
345 let Some(event) = event else {
346 return Err(anyhow!("ThreadEvent channel ended early"));
347 };
348
349 let mut output_file = output_file.lock().unwrap();
350
351 match event {
352 ThreadEvent::Stopped(reason) => match reason {
353 Ok(StopReason::EndTurn) => {
354 return Ok(());
355 }
356 Ok(StopReason::MaxTokens) => {
357 return Err(anyhow!("Exceeded maximum tokens"));
358 }
359 Ok(StopReason::ToolUse) => {
360 if std::env::var("ZED_EVAL_DEBUG").is_ok() {
361 println!("{}StopReason: Tool use", log_prefix);
362 }
363 }
364 Err(error) => {
365 return Err(anyhow!(error.clone()));
366 }
367 },
368 ThreadEvent::ShowError(thread_error) => {
369 break Err(anyhow!(thread_error.clone()));
370 }
371 ThreadEvent::StreamedAssistantText(_, chunk) => {
372 write!(&mut output_file, "{}", chunk).log_err();
373 }
374 ThreadEvent::StreamedAssistantThinking(_, chunk) => {
375 write!(&mut output_file, "{}", chunk).log_err();
376 }
377 ThreadEvent::UsePendingTools { tool_uses } => {
378 writeln!(&mut output_file, "\n\nUSING TOOLS:").log_err();
379 for tool_use in tool_uses {
380 writeln!(&mut output_file, "{}: {}", tool_use.name, tool_use.input)
381 .log_err();
382 }
383 }
384 ThreadEvent::ToolFinished {
385 tool_use_id,
386 pending_tool_use,
387 ..
388 } => {
389 thread.update(cx, |thread, _cx| {
390 if let Some(tool_use) = pending_tool_use {
391 if let Some(tool_result) = thread.tool_result(&tool_use_id) {
392 let message = if tool_result.is_error {
393 format!("TOOL FAILED: {}", tool_use.name)
394 } else {
395 format!("TOOL FINISHED: {}", tool_use.name)
396 };
397 println!("{log_prefix}{message}");
398 writeln!(&mut output_file, "\n{}", message).log_err();
399 writeln!(&mut output_file, "\n{}\n", tool_result.content).log_err();
400 let mut tool_use_counts = tool_use_counts.lock().unwrap();
401 *tool_use_counts
402 .entry(tool_result.tool_name.clone())
403 .or_insert(0) += 1;
404 } else {
405 let message = format!("TOOL FINISHED WITHOUT RESULT: {}", tool_use.name);
406 println!("{log_prefix}{message}");
407 writeln!(&mut output_file, "\n{}", message).log_err();
408 }
409 }
410 })?;
411 }
412 ThreadEvent::ToolConfirmationNeeded => {
413 panic!("{}Bug: Tool confirmation should not be required in eval", log_prefix);
414 },
415 ThreadEvent::StreamedCompletion |
416 ThreadEvent::MessageAdded(_) |
417 ThreadEvent::MessageEdited(_) |
418 ThreadEvent::MessageDeleted(_) |
419 ThreadEvent::SummaryChanged |
420 ThreadEvent::SummaryGenerated |
421 ThreadEvent::CheckpointChanged |
422 ThreadEvent::UsageUpdated(_) => {
423 if std::env::var("ZED_EVAL_DEBUG").is_ok() {
424 println!("{}Event: {:#?}", log_prefix, event);
425 }
426 }
427 }
428
429 output_file.flush().log_err();
430 }
431 }
432 });
433
434 thread.update(cx, |thread, cx| {
435 let context = vec![];
436 thread.insert_user_message(this.prompt.clone(), context, None, cx);
437 thread.send_to_model(model, RequestKind::Chat, cx);
438 })?;
439
440 event_handler_task.await?;
441
442 println!("{}Stopped", this.log_prefix);
443
444 if let Some((_, lsp_store)) = lsp_open_handle_and_store.as_ref() {
445 wait_for_lang_server(lsp_store, this.log_prefix.clone(), cx).await?;
446 }
447
448 println!("{}Getting repository diff", this.log_prefix);
449 let repository_diff = this.repository_diff().await?;
450
451 let repository_diff_path = this.run_dir.join(format!("{}.diff", this.name));
452 let mut repository_diff_output_file = File::create(&repository_diff_path)?;
453 writeln!(&mut repository_diff_output_file, "{}", &repository_diff).log_err();
454
455 println!("{}Getting diagnostics", this.log_prefix);
456 let diagnostics = cx
457 .update(move |cx| {
458 cx.spawn(async move |cx| query_lsp_diagnostics(project, cx).await)
459 })?
460 .await?;
461 println!("{}Got diagnostics", this.log_prefix);
462
463 drop(subscription);
464 drop(lsp_open_handle_and_store);
465
466 thread.update(cx, |thread, _cx| {
467 let response_count = thread
468 .messages()
469 .filter(|message| message.role == language_model::Role::Assistant)
470 .count();
471 RunOutput {
472 repository_diff,
473 diagnostics,
474 response_count,
475 token_usage: thread.cumulative_token_usage(),
476 tool_use_counts: tool_use_counts.lock().unwrap().clone(),
477 }
478 })
479 })
480 }
481
482 pub async fn judge(
483 &self,
484 model: Arc<dyn LanguageModel>,
485 repository_diff: String,
486 judge_repetitions: u32,
487 cx: &AsyncApp,
488 ) -> Result<JudgeOutput> {
489 let judge_prompt = include_str!("judge_prompt.hbs");
490 let judge_prompt_name = "judge_prompt";
491 let mut handlebars = Handlebars::new();
492 handlebars.register_template_string(judge_prompt_name, judge_prompt)?;
493 let prompt = handlebars.render(
494 judge_prompt_name,
495 &JudgeInput {
496 repository_diff,
497 criteria: self.criteria.clone(),
498 },
499 )?;
500
501 let request = LanguageModelRequest {
502 messages: vec![LanguageModelRequestMessage {
503 role: Role::User,
504 content: vec![MessageContent::Text(prompt)],
505 cache: false,
506 }],
507 temperature: None,
508 tools: Vec::new(),
509 stop: Vec::new(),
510 };
511
512 let response = send_language_model_request(model, request, cx).await?;
513
514 let judge_file_path = self.run_dir.join(format!(
515 "{}_judge_{}.md",
516 self.name, // This is the eval_name
517 judge_repetitions
518 ));
519
520 let mut judge_output_file = File::create(&judge_file_path)?;
521 writeln!(&mut judge_output_file, "{}", &response).log_err();
522
523 parse_judge_output(&response)
524 }
525
526 pub async fn repository_diff(&self) -> Result<String> {
527 let worktree_path = self.worktree_path();
528 run_git(&worktree_path, &["add", "-N"]).await?;
529 run_git(&worktree_path, &["diff"]).await
530 }
531}
532
533fn wait_for_lang_server(
534 lsp_store: &Entity<LspStore>,
535 log_prefix: String,
536 cx: &mut AsyncApp,
537) -> Task<Result<()>> {
538 if cx
539 .update(|cx| !has_pending_lang_server_work(lsp_store, cx))
540 .unwrap()
541 || std::env::var("ZED_EVAL_SKIP_LS_WAIT").is_ok()
542 {
543 return Task::ready(anyhow::Ok(()));
544 }
545
546 println!("{}⏵ Waiting for language server", log_prefix);
547
548 let (mut tx, mut rx) = mpsc::channel(1);
549
550 let subscription =
551 cx.subscribe(&lsp_store, {
552 let log_prefix = log_prefix.clone();
553 move |lsp_store, event, cx| {
554 match event {
555 project::LspStoreEvent::LanguageServerUpdate {
556 message:
557 client::proto::update_language_server::Variant::WorkProgress(
558 LspWorkProgress {
559 message: Some(message),
560 ..
561 },
562 ),
563 ..
564 } => println!("{}⟲ {message}", log_prefix),
565 _ => {}
566 }
567
568 if !has_pending_lang_server_work(&lsp_store, cx) {
569 tx.try_send(()).ok();
570 }
571 }
572 });
573
574 cx.spawn(async move |cx| {
575 let timeout = cx.background_executor().timer(Duration::new(60 * 5, 0));
576 let result = futures::select! {
577 _ = rx.next() => {
578 println!("{}⚑ Language server idle", log_prefix);
579 anyhow::Ok(())
580 },
581 _ = timeout.fuse() => {
582 Err(anyhow!("LSP wait timed out after 5 minutes"))
583 }
584 };
585 drop(subscription);
586 result
587 })
588}
589
590fn has_pending_lang_server_work(lsp_store: &Entity<LspStore>, cx: &App) -> bool {
591 lsp_store
592 .read(cx)
593 .language_server_statuses()
594 .any(|(_, status)| !status.pending_work.is_empty())
595}
596
597async fn query_lsp_diagnostics(project: Entity<Project>, cx: &mut AsyncApp) -> Result<String> {
598 let paths_with_diagnostics = project.update(cx, |project, cx| {
599 project
600 .diagnostic_summaries(true, cx)
601 .filter(|(_, _, summary)| summary.error_count > 0 || summary.warning_count > 0)
602 .map(|(project_path, _, _)| project_path)
603 .collect::<Vec<_>>()
604 })?;
605
606 let mut output = String::new();
607 for project_path in paths_with_diagnostics {
608 let buffer = project
609 .update(cx, |project, cx| project.open_buffer(project_path, cx))?
610 .await?;
611 let snapshot = buffer.read_with(cx, |buffer, _cx| buffer.snapshot())?;
612
613 for (_, group) in snapshot.diagnostic_groups(None) {
614 let entry = &group.entries[group.primary_ix];
615 let range = entry.range.to_point(&snapshot);
616 let severity = match entry.diagnostic.severity {
617 DiagnosticSeverity::ERROR => "error",
618 DiagnosticSeverity::WARNING => "warning",
619 _ => continue,
620 };
621
622 writeln!(
623 output,
624 "{} at line {}: {}",
625 severity,
626 range.start.row + 1,
627 entry.diagnostic.message
628 )?;
629 }
630 }
631 anyhow::Ok(output)
632}
633
634fn parse_judge_output(response: &str) -> Result<JudgeOutput> {
635 let analysis = get_tag("analysis", response)?.to_string();
636 let score = get_tag("score", response)?
637 .parse()
638 .context("error parsing score")?;
639
640 Ok(JudgeOutput { analysis, score })
641}
642
643fn get_tag(name: &'static str, response: &str) -> Result<String> {
644 let start_tag = format!("<{}>", name);
645 let end_tag = format!("</{}>", name);
646
647 let start_ix = response
648 .find(&start_tag)
649 .context(format!("{} start tag not found", name))?;
650 let content_start_ix = start_ix + start_tag.len();
651
652 let end_ix = content_start_ix
653 + response[content_start_ix..]
654 .find(&end_tag)
655 .context(format!("{} end tag not found", name))?;
656
657 let content = response[content_start_ix..end_ix].trim().unindent();
658
659 anyhow::Ok(content)
660}
661
662pub fn repo_path_for_url(repo_url: &str) -> PathBuf {
663 let repo_name = repo_url
664 .trim_start_matches("https://")
665 .replace(|c: char| !c.is_alphanumeric(), "-");
666 Path::new(REPOS_DIR)
667 .canonicalize()
668 .context(format!("No such directory {REPOS_DIR}"))
669 .unwrap()
670 .join(repo_name)
671}
672
673pub async fn run_git(repo_path: &Path, args: &[&str]) -> Result<String> {
674 let output = new_smol_command("git")
675 .current_dir(repo_path)
676 .args(args)
677 .output()
678 .await?;
679
680 if output.status.success() {
681 Ok(String::from_utf8(output.stdout)?.trim().to_string())
682 } else {
683 Err(anyhow!(
684 "`git {}` within `{}` failed with status: {}\nstderr:\n{}\nstdout:\n{}",
685 args.join(" "),
686 repo_path.display(),
687 output.status,
688 String::from_utf8_lossy(&output.stderr),
689 String::from_utf8_lossy(&output.stdout),
690 ))
691 }
692}
693
694pub async fn send_language_model_request(
695 model: Arc<dyn LanguageModel>,
696 request: LanguageModelRequest,
697 cx: &AsyncApp,
698) -> anyhow::Result<String> {
699 match model.stream_completion_text(request, &cx).await {
700 Ok(mut stream) => {
701 let mut full_response = String::new();
702 while let Some(chunk_result) = stream.stream.next().await {
703 match chunk_result {
704 Ok(chunk_str) => {
705 full_response.push_str(&chunk_str);
706 }
707 Err(err) => {
708 return Err(anyhow!(
709 "Error receiving response from language model: {err}"
710 ));
711 }
712 }
713 }
714 Ok(full_response)
715 }
716 Err(err) => Err(anyhow!(
717 "Failed to get response from language model. Error was: {err}"
718 )),
719 }
720}
721
722#[cfg(test)]
723mod test {
724 use super::*;
725
726 #[test]
727 fn test_parse_judge_output() {
728 let response = r#"
729 <analysis>The model did a good job but there were still compilations errors.</analysis>
730 <score>3</score>
731 "#
732 .unindent();
733
734 let output = parse_judge_output(&response).unwrap();
735 assert_eq!(
736 output.analysis,
737 "The model did a good job but there were still compilations errors."
738 );
739 assert_eq!(output.score, 3);
740
741 let response = r#"
742 Text around ignored
743
744 <analysis>
745 Failed to compile:
746 - Error 1
747 - Error 2
748 </analysis>
749
750 <score>1</score>
751 "#
752 .unindent();
753
754 let output = parse_judge_output(&response).unwrap();
755 assert_eq!(output.analysis, "Failed to compile:\n- Error 1\n- Error 2");
756 assert_eq!(output.score, 1);
757 }
758}