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(
144 &repo_path,
145 &["rev-parse", &format!("{}^{{commit}}", self.base.revision)],
146 )
147 .await
148 .is_ok();
149
150 if !revision_exists {
151 println!(
152 "{}Fetching revision {}",
153 self.log_prefix, &self.base.revision
154 );
155 run_git(
156 &repo_path,
157 &["fetch", "--depth", "1", "origin", &self.base.revision],
158 )
159 .await?;
160 }
161
162 let worktree_path = self.worktree_path();
163
164 if worktree_path.is_dir() {
165 println!("{}Resetting existing worktree", self.log_prefix);
166
167 // TODO: consider including "-x" to remove ignored files. The downside of this is that
168 // it will also remove build artifacts, and so prevent incremental reuse there.
169 run_git(&worktree_path, &["clean", "--force", "-d"]).await?;
170 run_git(&worktree_path, &["reset", "--hard", "HEAD"]).await?;
171 run_git(&worktree_path, &["checkout", &self.base.revision]).await?;
172 } else {
173 println!("{}Creating worktree", self.log_prefix);
174
175 let worktree_path_string = worktree_path.to_string_lossy().to_string();
176
177 run_git(
178 &repo_path,
179 &[
180 "worktree",
181 "add",
182 "-f",
183 &worktree_path_string,
184 &self.base.revision,
185 ],
186 )
187 .await?;
188 }
189
190 // Create the output file
191 let output_file = Arc::new(Mutex::new(File::create(&self.output_file_path)?));
192 self.output_file = Some(output_file);
193
194 Ok(())
195 }
196
197 /// Returns the output file, panicking if it's not set
198 fn output_file(&self) -> Arc<Mutex<File>> {
199 self.output_file
200 .clone()
201 .expect("Output file not created. Call setup() first.")
202 }
203
204 pub fn run(
205 &self,
206 model: Arc<dyn LanguageModel>,
207 app_state: Arc<AgentAppState>,
208 cx: &mut App,
209 ) -> Task<Result<RunOutput>> {
210 let project = Project::local(
211 app_state.client.clone(),
212 app_state.node_runtime.clone(),
213 app_state.user_store.clone(),
214 app_state.languages.clone(),
215 Arc::new(DapRegistry::default()),
216 app_state.fs.clone(),
217 None,
218 cx,
219 );
220
221 let worktree_path = self.worktree_path();
222 let worktree = project.update(cx, |project, cx| {
223 project.create_worktree(&worktree_path, true, cx)
224 });
225
226 let tools = cx.new(|_| ToolWorkingSet::default());
227 let thread_store =
228 ThreadStore::load(project.clone(), tools, app_state.prompt_builder.clone(), cx);
229 let this = self.clone();
230
231 cx.spawn(async move |cx| {
232 let worktree = worktree.await?;
233
234 // Wait for worktree scan to finish before choosing a file to open.
235 worktree
236 .update(cx, |worktree, _cx| {
237 worktree.as_local().unwrap().scan_complete()
238 })?
239 .await;
240
241 let lsp_open_handle_and_store = if this.base.require_lsp {
242 let language_extension = this.base.language_extension.as_deref().context(
243 "language_extension field is required in base.toml when `require_lsp == true`",
244 )?;
245
246 // Open a file that matches the language to cause LSP to start.
247 let language_file = worktree.read_with(cx, |worktree, _cx| {
248 worktree
249 .files(false, 0)
250 .find_map(|e| {
251 if e.path.clone().extension().and_then(|ext| ext.to_str())
252 == Some(language_extension)
253 {
254 Some(ProjectPath {
255 worktree_id: worktree.id(),
256 path: e.path.clone(),
257 })
258 } else {
259 None
260 }
261 })
262 .context("Failed to find a file for example language")
263 })??;
264
265 let open_language_file_buffer_task = project.update(cx, |project, cx| {
266 project.open_buffer(language_file.clone(), cx)
267 })?;
268
269 let language_file_buffer = open_language_file_buffer_task.await?;
270
271 let (lsp_open_handle, lsp_store) = project.update(cx, |project, cx| {
272 (
273 project.register_buffer_with_language_servers(&language_file_buffer, cx),
274 project.lsp_store().clone(),
275 )
276 })?;
277
278 // TODO: remove this once the diagnostics tool waits for new diagnostics
279 cx.background_executor().timer(Duration::new(5, 0)).await;
280 wait_for_lang_server(&lsp_store, this.log_prefix.clone(), cx).await?;
281
282 lsp_store.update(cx, |lsp_store, cx| {
283 lsp_open_handle.update(cx, |buffer, cx| {
284 buffer.update(cx, |buffer, cx| {
285 let has_language_server = lsp_store
286 .language_servers_for_local_buffer(buffer, cx)
287 .next()
288 .is_some();
289 if has_language_server {
290 Ok(())
291 } else {
292 Err(anyhow!(
293 "`{:?}` was opened to cause the language server to start, \
294 but no language servers are registered for its buffer. \
295 Set `require_lsp = false` in `base.toml` to skip this.",
296 language_file
297 ))
298 }
299 })
300 })
301 })??;
302
303 Some((lsp_open_handle, lsp_store))
304 } else {
305 None
306 };
307
308 if std::env::var("ZED_EVAL_SETUP_ONLY").is_ok() {
309 return Err(anyhow!("Setup only mode"));
310 }
311
312 let thread_store = thread_store.await;
313 let thread =
314 thread_store.update(cx, |thread_store, cx| thread_store.create_thread(cx))?;
315
316 {
317 let output_file_ref = this.output_file();
318 let mut output_file = output_file_ref.lock().unwrap();
319 writeln!(&mut output_file, "👤 USER:").log_err();
320 writeln!(&mut output_file, "{}", this.prompt).log_err();
321 writeln!(&mut output_file, "🤖 ASSISTANT:").log_err();
322 output_file.flush().log_err();
323 }
324
325 let tool_use_counts: Arc<Mutex<HashMap<Arc<str>, u32>>> =
326 Mutex::new(HashMap::default()).into();
327
328 let (thread_event_tx, mut thread_event_rx) = mpsc::unbounded();
329
330 let subscription = cx.subscribe(&thread, move |_thread, event: &ThreadEvent, _cx| {
331 thread_event_tx.unbounded_send(event.clone()).log_err();
332 });
333
334 let event_handler_task = cx.spawn({
335 // Need to clone the Arc here because the reference from output_file() won't live long enough
336 let output_file = this.output_file.clone().unwrap();
337 let log_prefix = this.log_prefix.clone();
338 let tool_use_counts = tool_use_counts.clone();
339 let thread = thread.downgrade();
340 async move |cx| {
341 loop {
342 let event = select_biased! {
343 event = thread_event_rx.next() => event,
344 _ = cx.background_executor().timer(THREAD_EVENT_TIMEOUT).fuse() => {
345 return Err(anyhow!("Agentic loop stalled - waited {:?} without any events", THREAD_EVENT_TIMEOUT));
346 }
347 };
348 let Some(event) = event else {
349 return Err(anyhow!("ThreadEvent channel ended early"));
350 };
351
352 let mut output_file = output_file.lock().unwrap();
353
354 match event {
355 ThreadEvent::Stopped(reason) => match reason {
356 Ok(StopReason::EndTurn) => {
357 return Ok(());
358 }
359 Ok(StopReason::MaxTokens) => {
360 return Err(anyhow!("Exceeded maximum tokens"));
361 }
362 Ok(StopReason::ToolUse) => {
363 if std::env::var("ZED_EVAL_DEBUG").is_ok() {
364 println!("{}StopReason: Tool use", log_prefix);
365 }
366 }
367 Err(error) => {
368 return Err(anyhow!(error.clone()));
369 }
370 },
371 ThreadEvent::ShowError(thread_error) => {
372 break Err(anyhow!(thread_error.clone()));
373 }
374 ThreadEvent::StreamedAssistantText(_, chunk) => {
375 write!(&mut output_file, "{}", chunk).log_err();
376 }
377 ThreadEvent::StreamedAssistantThinking(_, chunk) => {
378 write!(&mut output_file, "{}", chunk).log_err();
379 }
380 ThreadEvent::UsePendingTools { tool_uses } => {
381 writeln!(&mut output_file, "\n\nUSING TOOLS:").log_err();
382 for tool_use in tool_uses {
383 writeln!(&mut output_file, "{}: {}", tool_use.name, tool_use.input)
384 .log_err();
385 }
386 }
387 ThreadEvent::ToolFinished {
388 tool_use_id,
389 pending_tool_use,
390 ..
391 } => {
392 thread.update(cx, |thread, _cx| {
393 if let Some(tool_use) = pending_tool_use {
394 if let Some(tool_result) = thread.tool_result(&tool_use_id) {
395 let message = if tool_result.is_error {
396 format!("TOOL FAILED: {}", tool_use.name)
397 } else {
398 format!("TOOL FINISHED: {}", tool_use.name)
399 };
400 println!("{log_prefix}{message}");
401 writeln!(&mut output_file, "\n{}", message).log_err();
402 writeln!(&mut output_file, "\n{}\n", tool_result.content).log_err();
403 let mut tool_use_counts = tool_use_counts.lock().unwrap();
404 *tool_use_counts
405 .entry(tool_result.tool_name.clone())
406 .or_insert(0) += 1;
407 } else {
408 let message = format!("TOOL FINISHED WITHOUT RESULT: {}", tool_use.name);
409 println!("{log_prefix}{message}");
410 writeln!(&mut output_file, "\n{}", message).log_err();
411 }
412 }
413 })?;
414 }
415 ThreadEvent::ToolConfirmationNeeded => {
416 panic!("{}Bug: Tool confirmation should not be required in eval", log_prefix);
417 },
418 ThreadEvent::StreamedCompletion |
419 ThreadEvent::MessageAdded(_) |
420 ThreadEvent::MessageEdited(_) |
421 ThreadEvent::MessageDeleted(_) |
422 ThreadEvent::SummaryChanged |
423 ThreadEvent::SummaryGenerated |
424 ThreadEvent::CheckpointChanged |
425 ThreadEvent::UsageUpdated(_) => {
426 if std::env::var("ZED_EVAL_DEBUG").is_ok() {
427 println!("{}Event: {:#?}", log_prefix, event);
428 }
429 }
430 }
431
432 output_file.flush().log_err();
433 }
434 }
435 });
436
437 thread.update(cx, |thread, cx| {
438 let context = vec![];
439 thread.insert_user_message(this.prompt.clone(), context, None, cx);
440 thread.send_to_model(model, RequestKind::Chat, cx);
441 })?;
442
443 event_handler_task.await?;
444
445 println!("{}Stopped", this.log_prefix);
446
447 if let Some((_, lsp_store)) = lsp_open_handle_and_store.as_ref() {
448 wait_for_lang_server(lsp_store, this.log_prefix.clone(), cx).await?;
449 }
450
451 println!("{}Getting repository diff", this.log_prefix);
452 let repository_diff = this.repository_diff().await?;
453
454 let repository_diff_path = this.run_dir.join(format!("{}.diff", this.name));
455 let mut repository_diff_output_file = File::create(&repository_diff_path)?;
456 writeln!(&mut repository_diff_output_file, "{}", &repository_diff).log_err();
457
458 println!("{}Getting diagnostics", this.log_prefix);
459 let diagnostics = cx
460 .update(move |cx| {
461 cx.spawn(async move |cx| query_lsp_diagnostics(project, cx).await)
462 })?
463 .await?;
464 println!("{}Got diagnostics", this.log_prefix);
465
466 drop(subscription);
467 drop(lsp_open_handle_and_store);
468
469 thread.update(cx, |thread, _cx| {
470 let response_count = thread
471 .messages()
472 .filter(|message| message.role == language_model::Role::Assistant)
473 .count();
474 RunOutput {
475 repository_diff,
476 diagnostics,
477 response_count,
478 token_usage: thread.cumulative_token_usage(),
479 tool_use_counts: tool_use_counts.lock().unwrap().clone(),
480 }
481 })
482 })
483 }
484
485 pub async fn judge(
486 &self,
487 model: Arc<dyn LanguageModel>,
488 repository_diff: String,
489 judge_repetitions: u32,
490 cx: &AsyncApp,
491 ) -> Result<JudgeOutput> {
492 let judge_prompt = include_str!("judge_prompt.hbs");
493 let judge_prompt_name = "judge_prompt";
494 let mut handlebars = Handlebars::new();
495 handlebars.register_template_string(judge_prompt_name, judge_prompt)?;
496 let prompt = handlebars.render(
497 judge_prompt_name,
498 &JudgeInput {
499 repository_diff,
500 criteria: self.criteria.clone(),
501 },
502 )?;
503
504 let request = LanguageModelRequest {
505 messages: vec![LanguageModelRequestMessage {
506 role: Role::User,
507 content: vec![MessageContent::Text(prompt)],
508 cache: false,
509 }],
510 temperature: None,
511 tools: Vec::new(),
512 stop: Vec::new(),
513 };
514
515 let response = send_language_model_request(model, request, cx).await?;
516
517 let judge_file_path = self.run_dir.join(format!(
518 "{}_judge_{}.md",
519 self.name, // This is the eval_name
520 judge_repetitions
521 ));
522
523 let mut judge_output_file = File::create(&judge_file_path)?;
524 writeln!(&mut judge_output_file, "{}", &response).log_err();
525
526 parse_judge_output(&response)
527 }
528
529 pub async fn repository_diff(&self) -> Result<String> {
530 let worktree_path = self.worktree_path();
531 run_git(&worktree_path, &["add", "-N"]).await?;
532 run_git(&worktree_path, &["diff"]).await
533 }
534}
535
536fn wait_for_lang_server(
537 lsp_store: &Entity<LspStore>,
538 log_prefix: String,
539 cx: &mut AsyncApp,
540) -> Task<Result<()>> {
541 if cx
542 .update(|cx| !has_pending_lang_server_work(lsp_store, cx))
543 .unwrap()
544 || std::env::var("ZED_EVAL_SKIP_LS_WAIT").is_ok()
545 {
546 return Task::ready(anyhow::Ok(()));
547 }
548
549 println!("{}⏵ Waiting for language server", log_prefix);
550
551 let (mut tx, mut rx) = mpsc::channel(1);
552
553 let subscription =
554 cx.subscribe(&lsp_store, {
555 let log_prefix = log_prefix.clone();
556 move |lsp_store, event, cx| {
557 match event {
558 project::LspStoreEvent::LanguageServerUpdate {
559 message:
560 client::proto::update_language_server::Variant::WorkProgress(
561 LspWorkProgress {
562 message: Some(message),
563 ..
564 },
565 ),
566 ..
567 } => println!("{}⟲ {message}", log_prefix),
568 _ => {}
569 }
570
571 if !has_pending_lang_server_work(&lsp_store, cx) {
572 tx.try_send(()).ok();
573 }
574 }
575 });
576
577 cx.spawn(async move |cx| {
578 let timeout = cx.background_executor().timer(Duration::new(60 * 5, 0));
579 let result = futures::select! {
580 _ = rx.next() => {
581 println!("{}⚑ Language server idle", log_prefix);
582 anyhow::Ok(())
583 },
584 _ = timeout.fuse() => {
585 Err(anyhow!("LSP wait timed out after 5 minutes"))
586 }
587 };
588 drop(subscription);
589 result
590 })
591}
592
593fn has_pending_lang_server_work(lsp_store: &Entity<LspStore>, cx: &App) -> bool {
594 lsp_store
595 .read(cx)
596 .language_server_statuses()
597 .any(|(_, status)| !status.pending_work.is_empty())
598}
599
600async fn query_lsp_diagnostics(project: Entity<Project>, cx: &mut AsyncApp) -> Result<String> {
601 let paths_with_diagnostics = project.update(cx, |project, cx| {
602 project
603 .diagnostic_summaries(true, cx)
604 .filter(|(_, _, summary)| summary.error_count > 0 || summary.warning_count > 0)
605 .map(|(project_path, _, _)| project_path)
606 .collect::<Vec<_>>()
607 })?;
608
609 let mut output = String::new();
610 for project_path in paths_with_diagnostics {
611 let buffer = project
612 .update(cx, |project, cx| project.open_buffer(project_path, cx))?
613 .await?;
614 let snapshot = buffer.read_with(cx, |buffer, _cx| buffer.snapshot())?;
615
616 for (_, group) in snapshot.diagnostic_groups(None) {
617 let entry = &group.entries[group.primary_ix];
618 let range = entry.range.to_point(&snapshot);
619 let severity = match entry.diagnostic.severity {
620 DiagnosticSeverity::ERROR => "error",
621 DiagnosticSeverity::WARNING => "warning",
622 _ => continue,
623 };
624
625 writeln!(
626 output,
627 "{} at line {}: {}",
628 severity,
629 range.start.row + 1,
630 entry.diagnostic.message
631 )?;
632 }
633 }
634 anyhow::Ok(output)
635}
636
637fn parse_judge_output(response: &str) -> Result<JudgeOutput> {
638 let analysis = get_tag("analysis", response)?.to_string();
639 let score = get_tag("score", response)?
640 .parse()
641 .context("error parsing score")?;
642
643 Ok(JudgeOutput { analysis, score })
644}
645
646fn get_tag(name: &'static str, response: &str) -> Result<String> {
647 let start_tag = format!("<{}>", name);
648 let end_tag = format!("</{}>", name);
649
650 let start_ix = response
651 .find(&start_tag)
652 .context(format!("{} start tag not found", name))?;
653 let content_start_ix = start_ix + start_tag.len();
654
655 let end_ix = content_start_ix
656 + response[content_start_ix..]
657 .find(&end_tag)
658 .context(format!("{} end tag not found", name))?;
659
660 let content = response[content_start_ix..end_ix].trim().unindent();
661
662 anyhow::Ok(content)
663}
664
665pub fn repo_path_for_url(repo_url: &str) -> PathBuf {
666 let repo_name = repo_url
667 .trim_start_matches("https://")
668 .replace(|c: char| !c.is_alphanumeric(), "-");
669 Path::new(REPOS_DIR)
670 .canonicalize()
671 .context(format!("No such directory {REPOS_DIR}"))
672 .unwrap()
673 .join(repo_name)
674}
675
676pub async fn run_git(repo_path: &Path, args: &[&str]) -> Result<String> {
677 let output = new_smol_command("git")
678 .current_dir(repo_path)
679 .args(args)
680 .output()
681 .await?;
682
683 if output.status.success() {
684 Ok(String::from_utf8(output.stdout)?.trim().to_string())
685 } else {
686 Err(anyhow!(
687 "`git {}` within `{}` failed with status: {}\nstderr:\n{}\nstdout:\n{}",
688 args.join(" "),
689 repo_path.display(),
690 output.status,
691 String::from_utf8_lossy(&output.stderr),
692 String::from_utf8_lossy(&output.stdout),
693 ))
694 }
695}
696
697pub async fn send_language_model_request(
698 model: Arc<dyn LanguageModel>,
699 request: LanguageModelRequest,
700 cx: &AsyncApp,
701) -> anyhow::Result<String> {
702 match model.stream_completion_text(request, &cx).await {
703 Ok(mut stream) => {
704 let mut full_response = String::new();
705 while let Some(chunk_result) = stream.stream.next().await {
706 match chunk_result {
707 Ok(chunk_str) => {
708 full_response.push_str(&chunk_str);
709 }
710 Err(err) => {
711 return Err(anyhow!(
712 "Error receiving response from language model: {err}"
713 ));
714 }
715 }
716 }
717 Ok(full_response)
718 }
719 Err(err) => Err(anyhow!(
720 "Failed to get response from language model. Error was: {err}"
721 )),
722 }
723}
724
725#[cfg(test)]
726mod test {
727 use super::*;
728
729 #[test]
730 fn test_parse_judge_output() {
731 let response = r#"
732 <analysis>The model did a good job but there were still compilations errors.</analysis>
733 <score>3</score>
734 "#
735 .unindent();
736
737 let output = parse_judge_output(&response).unwrap();
738 assert_eq!(
739 output.analysis,
740 "The model did a good job but there were still compilations errors."
741 );
742 assert_eq!(output.score, 3);
743
744 let response = r#"
745 Text around ignored
746
747 <analysis>
748 Failed to compile:
749 - Error 1
750 - Error 2
751 </analysis>
752
753 <score>1</score>
754 "#
755 .unindent();
756
757 let output = parse_judge_output(&response).unwrap();
758 assert_eq!(output.analysis, "Failed to compile:\n- Error 1\n- Error 2");
759 assert_eq!(output.score, 1);
760 }
761}